Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Sagar <sa...@gmail.com>]

Hi Guozhang/David,

I created a confluence page to discuss how Connect would need to change
based on the new rebalance protocol. Here's the page:
https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol

It's also pretty longish and I have tried to keep a format similar to
KIP-848. Let me know what you think. Also, do you think this should be
moved to a separate discussion thread or is this one fine?

Thanks!
Sagar.

On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:

> Hello Guozhang,
>
> Thank you so much for the doc on Kafka Streams. Sure, I would do the
> analysis and come up with such a document.
>
> Thanks!
> Sagar.
>
> On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com> wrote:
>
>> Hello Sagar,
>>
>> It would be great if you could come back with some analysis on how to
>> implement the Connect side integration with the new protocol; so far
>> besides leveraging on the new "protocol type" we did not yet think through
>> the Connect side implementations. For Streams here's a draft of
>> integration
>> plan:
>>
>> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
>> just FYI for your analysis on Connect.
>>
>> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com> wrote:
>>
>> > Hi David,
>> >
>> > Thank you for your response. The reason I thought connect can also fit
>> into
>> > this new scheme is that even today the connect uses a WorkerCoordinator
>> > extending from AbstractCoordinator to empower rebalances of
>> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
>> connect
>> > and uses the metadata() method by plumbing into
>> JoinGroupRequestProtocol.
>> >
>> > I think the changes to support connect would be similar at a high level
>> to
>> > the changes in streams mainly because of the Client side assignors being
>> > used in both. At an implementation level, we might need to make a lot of
>> > changes to get onto this new assignment protocol like enhancing the
>> > JoinGroup request/response and SyncGroup and using
>> ConsumerGroupHeartbeat
>> > API etc again on similar lines to streams (or there might be
>> deviations). I
>> > would try to perform a detailed analysis of the same  and we can have a
>> > separate discussion thread for that as that would derail this discussion
>> > thread. Let me know if that sounds good to you.
>> >
>> > Thanks!
>> > Sagar.
>> >
>> >
>> >
>> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot <djacot@confluent.io.invalid
>> >
>> > wrote:
>> >
>> > > Hi Sagar,
>> > >
>> > > Thanks for your comments.
>> > >
>> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
>> > >
>> > > 2) The idea is to transition C from his current assignment to his
>> > > target assignment when he can move to epoch 3. When that happens, the
>> > > member assignment is updated and persisted with all its assigned
>> > > partitions even if they are not all revoked yet. In other words, the
>> > > member assignment becomes the target assignment. This is basically an
>> > > optimization to avoid having to write all the changes to the log. The
>> > > examples are based on the persisted state so I understand the
>> > > confusion. Let me see if I can improve this in the description.
>> > >
>> > > 3) Regarding Connect, it could reuse the protocol with a client side
>> > > assignor if it fits in the protocol. The assignment is about
>> > > topicid-partitions + metadata, could Connect fit into this?
>> > >
>> > > Best,
>> > > David
>> > >
>> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
>> wrote:
>> > > >
>> > > > Hi David,
>> > > >
>> > > > Thanks for the KIP. I just had minor observations:
>> > > >
>> > > > 1) In the Assignment Error section in Client Side mode Assignment
>> > > process,
>> > > > you mentioned => `In this case, the client side assignor can return
>> an
>> > > > error to the group coordinator`. In this case are you referring to
>> the
>> > > > Assignor returning an AssignmentError that's listed down towards the
>> > end?
>> > > > If yes, do you think it would make sense to mention this explicitly
>> > here?
>> > > >
>> > > > 2) In the Case Studies section, I have a slight confusion, not sure
>> if
>> > > > others have the same. Consider this step:
>> > > >
>> > > > When B heartbeats, the group coordinator transitions him to epoch 3
>> > > because
>> > > > B has no partitions to revoke. It persists the change and reply.
>> > > >
>> > > >    - Group (epoch=3)
>> > > >       - A
>> > > >       - B
>> > > >       - C
>> > > >    - Target Assignment (epoch=3)
>> > > >       - A - partitions=[foo-0]
>> > > >       - B - partitions=[foo-2]
>> > > >       - C - partitions=[foo-1]
>> > > >    - Member Assignment
>> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
>> > > >       - B - epoch=3, partitions=[foo-2]
>> > > >       - C - epoch=3, partitions=[foo-1]
>> > > >
>> > > > When C heartbeats, it transitions to epoch 3 but cannot get foo-1
>> yet.
>> > > >
>> > > > Here,it's mentioned that member C can't get the foo-1 partition yet,
>> > but
>> > > > based on the description above, it seems it already has it. Do you
>> > think
>> > > it
>> > > > would be better to remove it and populate it only when it actually
>> gets
>> > > it?
>> > > > I see this in a lot of other places, so have I understood it
>> > incorrectly
>> > > ?
>> > > >
>> > > >
>> > > > Regarding connect , it might be out of scope of this discussion, but
>> > from
>> > > > what I understood it would probably be running in client side
>> assignor
>> > > mode
>> > > > even on the new rebalance protocol as it has its own Custom
>> > > Assignors(Eager
>> > > > and IncrementalCooperative).
>> > > >
>> > > > Thanks!
>> > > >
>> > > > Sagar.
>> > > >
>> > > >
>> > > >
>> > > >
>> > > >
>> > > >
>> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
>> > <djacot@confluent.io.invalid
>> > > >
>> > > > wrote:
>> > > >
>> > > > > Thanks Hector! Our goal is to move forward with specialized API
>> > > > > instead of relying on one generic API. For Connect, we can apply
>> the
>> > > > > exact same pattern and reuse/share the core implementation on the
>> > > > > server side. For the schema registry, I think that we should
>> consider
>> > > > > having a tailored API to do simple membership/leader election.
>> > > > >
>> > > > > Best,
>> > > > > David
>> > > > >
>> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <is...@juma.me.uk>
>> > > wrote:
>> > > > > >
>> > > > > > Three quick comments:
>> > > > > >
>> > > > > > 1. Regarding java.util.regex.Pattern vs
>> com.google.re2j.Pattern, we
>> > > > > should
>> > > > > > document the differences in more detail before deciding one way
>> or
>> > > > > another.
>> > > > > > That said, if people pass java.util.regex.Pattern, they expect
>> > their
>> > > > > > semantics to be honored. If we are doing something different,
>> then
>> > we
>> > > > > > should consider adding an overload with our own Pattern class (I
>> > > don't
>> > > > > > think we'd want to expose re2j's at this point).
>> > > > > > 2. Regarding topic ids, any major new protocol should integrate
>> > fully
>> > > > > with
>> > > > > > it and should handle the topic recreation case correctly. That's
>> > the
>> > > main
>> > > > > > part we need to handle. I agree with David that we'd want to add
>> > > topic
>> > > > > ids
>> > > > > > to the relevant protocols that don't have it yet and that we can
>> > > probably
>> > > > > > focus on the internals versus adding new APIs to the Java
>> Consumer
>> > > > > (unless
>> > > > > > we find that adding new APIs is required for reasonable
>> semantics).
>> > > > > > 3. I am still not sure about the coordinator storing the
>> configs.
>> > > It's
>> > > > > > powerful for configs to be centralized in the metadata log for
>> > > various
>> > > > > > reasons (auditability, visibility, consistency, etc.).
>> Similarly, I
>> > > am
>> > > > > not
>> > > > > > sure about automatically deleting configs in a way that they
>> cannot
>> > > be
>> > > > > > recovered. A good property for modern systems is to minimize the
>> > > number
>> > > > > of
>> > > > > > unrecoverable data loss scenarios.
>> > > > > >
>> > > > > > Ismael
>> > > > > >
>> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
>> > > <djacot@confluent.io.invalid
>> > > > > >
>> > > > > > wrote:
>> > > > > >
>> > > > > > > Thanks Guozhang. My answers are below:
>> > > > > > >
>> > > > > > > > 1) the migration path, especially the last step when clients
>> > > flip the
>> > > > > > > flag
>> > > > > > > > to enable the new protocol, in which we would have a window
>> > where
>> > > > > both
>> > > > > > > new
>> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
>> members
>> > of
>> > > the
>> > > > > same
>> > > > > > > > group. How the coordinator could "mimic" the old behavior
>> while
>> > > > > using the
>> > > > > > > > new protocol is something we need to present about.
>> > > > > > >
>> > > > > > > Noted. I just published a new version of KIP which includes
>> more
>> > > > > > > details about this. See the "Supporting Online Consumer Group
>> > > Upgrade"
>> > > > > > > and the "Compatibility, Deprecation, and Migration Plan". I
>> think
>> > > that
>> > > > > > > I have to think through a few cases now but the overall idea
>> and
>> > > > > > > mechanism should be understandable.
>> > > > > > >
>> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
>> are
>> > > only
>> > > > > used
>> > > > > > > as
>> > > > > > > > part of RPCs and admin client, but they are not exposed via
>> any
>> > > > > public
>> > > > > > > APIs
>> > > > > > > > to consumers yet. I think the question is, first should we
>> let
>> > > the
>> > > > > > > consumer
>> > > > > > > > client to be maintaining the names -> ids mapping itself to
>> > fully
>> > > > > > > leverage
>> > > > > > > > on all the augmented existing RPCs and the new RPCs with the
>> > > topic
>> > > > > ids;
>> > > > > > > and
>> > > > > > > > secondly, should we ever consider exposing the topic ids in
>> the
>> > > > > consumer
>> > > > > > > > public APIs as well (both subscribe/assign, as well as in
>> the
>> > > > > rebalance
>> > > > > > > > listener for cases like topic deletion-and-recreation).
>> > > > > > >
>> > > > > > > a) Assuming that we would include converting all the offsets
>> > > related
>> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
>> able
>> > to
>> > > > > > > fully operate with topic ids. That being said, it still has to
>> > > provide
>> > > > > > > the topics names in various APIs so having a mapping in the
>> > > consumer
>> > > > > > > seems inevitable to me.
>> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
>> this
>> > > goes
>> > > > > > > beyond the scope of this KIP. I would rather focus on the
>> > internals
>> > > > > > > here and we can consider this separately if we see value in
>> doing
>> > > it.
>> > > > > > >
>> > > > > > > Coming back to Ismael's point about using topic ids in the
>> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
>> > advantage
>> > > in
>> > > > > > > favour of it. The consumer will have the opportunity to
>> validate
>> > > that
>> > > > > > > the topics exists before passing them into the group rebalance
>> > > > > > > protocol. Obviously, the coordinator will also notice it but
>> it
>> > > does
>> > > > > > > not really have a way to reject an invalid topic in the
>> response.
>> > > > > > >
>> > > > > > > > I'm agreeing with David on all other minor questions except
>> for
>> > > the
>> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
>> > > necessary
>> > > > > to
>> > > > > > > > deprecate the subscribe API with Pattern, but instead we
>> still
>> > > use
>> > > > > > > Pattern
>> > > > > > > > while just documenting that our subscription may be
>> rejected by
>> > > the
>> > > > > > > server.
>> > > > > > > > Since the incompatible case is a very rare scenario I felt
>> > using
>> > > an
>> > > > > > > > overloaded `String` based subscription may be more
>> vulnerable
>> > to
>> > > > > various
>> > > > > > > > invalid regexes.
>> > > > > > >
>> > > > > > > That could work. I have to look at the differences between the
>> > two
>> > > > > > > engines to better understand the potential issues. My
>> > > understanding is
>> > > > > > > that would work for all the basic regular expressions. The
>> > > differences
>> > > > > > > between the two are mainly about the various character
>> classes. I
>> > > > > > > wonder what other people think about this.
>> > > > > > >
>> > > > > > > Best,
>> > > > > > > David
>> > > > > > >
>> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
>> > wangguoz@gmail.com
>> > > >
>> > > > > wrote:
>> > > > > > > >
>> > > > > > > > Thanks David! I think on the high level there are two meta
>> > > points we
>> > > > > need
>> > > > > > > > to concretize a bit more:
>> > > > > > > >
>> > > > > > > > 1) the migration path, especially the last step when clients
>> > > flip the
>> > > > > > > flag
>> > > > > > > > to enable the new protocol, in which we would have a window
>> > where
>> > > > > both
>> > > > > > > new
>> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
>> members
>> > of
>> > > the
>> > > > > same
>> > > > > > > > group. How the coordinator could "mimic" the old behavior
>> while
>> > > > > using the
>> > > > > > > > new protocol is something we need to present about.
>> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
>> are
>> > > only
>> > > > > used
>> > > > > > > as
>> > > > > > > > part of RPCs and admin client, but they are not exposed via
>> any
>> > > > > public
>> > > > > > > APIs
>> > > > > > > > to consumers yet. I think the question is, first should we
>> let
>> > > the
>> > > > > > > consumer
>> > > > > > > > client to be maintaining the names -> ids mapping itself to
>> > fully
>> > > > > > > leverage
>> > > > > > > > on all the augmented existing RPCs and the new RPCs with the
>> > > topic
>> > > > > ids;
>> > > > > > > and
>> > > > > > > > secondly, should we ever consider exposing the topic ids in
>> the
>> > > > > consumer
>> > > > > > > > public APIs as well (both subscribe/assign, as well as in
>> the
>> > > > > rebalance
>> > > > > > > > listener for cases like topic deletion-and-recreation).
>> > > > > > > >
>> > > > > > > > I'm agreeing with David on all other minor questions except
>> for
>> > > the
>> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
>> > > necessary
>> > > > > to
>> > > > > > > > deprecate the subscribe API with Pattern, but instead we
>> still
>> > > use
>> > > > > > > Pattern
>> > > > > > > > while just documenting that our subscription may be
>> rejected by
>> > > the
>> > > > > > > server.
>> > > > > > > > Since the incompatible case is a very rare scenario I felt
>> > using
>> > > an
>> > > > > > > > overloaded `String` based subscription may be more
>> vulnerable
>> > to
>> > > > > various
>> > > > > > > > invalid regexes.
>> > > > > > > >
>> > > > > > > >
>> > > > > > > > Guozhang
>> > > > > > > >
>> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
>> > > > > <djacot@confluent.io.invalid
>> > > > > > > >
>> > > > > > > > wrote:
>> > > > > > > >
>> > > > > > > > > Hi Ismael,
>> > > > > > > > >
>> > > > > > > > > Thanks for your feedback. Let me answer your questions
>> > inline.
>> > > > > > > > >
>> > > > > > > > > > 1. I think it's premature to talk about target versions
>> for
>> > > > > > > deprecation
>> > > > > > > > > and
>> > > > > > > > > > removal of the existing group protocol. Unlike KRaft,
>> this
>> > > > > affects a
>> > > > > > > core
>> > > > > > > > > > client protocol and hence deprecation/removal will be
>> > heavily
>> > > > > > > dependent
>> > > > > > > > > on
>> > > > > > > > > > how quickly applications migrate to the new protocol.
>> > > > > > > > >
>> > > > > > > > > That makes sense. I will remove it.
>> > > > > > > > >
>> > > > > > > > > > 2. The KIP says we intend to release this in 4.x, but it
>> > > wasn't
>> > > > > made
>> > > > > > > > > clear
>> > > > > > > > > > why. If we added that as a way to estimate when we'd
>> > > deprecate
>> > > > > and
>> > > > > > > remove
>> > > > > > > > > > the group protocol, I also suggest removing this part.
>> > > > > > > > >
>> > > > > > > > > Let me explain my reasoning. As explained, I plan to
>> rewrite
>> > > the
>> > > > > group
>> > > > > > > > > coordinator in Java while we implement the new protocol.
>> This
>> > > means
>> > > > > > > > > that the internals will be slightly different (e.g.
>> threading
>> > > > > model).
>> > > > > > > > > Therefore, I wanted to tighten the switch from the old
>> group
>> > > > > > > > > coordinator to the new group coordinator to a major
>> release.
>> > > The
>> > > > > > > > > alternative would be to use a flag to do the switch
>> instead
>> > of
>> > > > > relying
>> > > > > > > > > on the software upgrade.
>> > > > > > > > >
>> > > > > > > > > > 3. We need to flesh out the details of the migration
>> story.
>> > > It
>> > > > > sounds
>> > > > > > > > > like
>> > > > > > > > > > we're saying we will support online migrations. Is that
>> > > correct?
>> > > > > We
>> > > > > > > > > should
>> > > > > > > > > > explain this in detail. It could also be done as a
>> separate
>> > > KIP,
>> > > > > if
>> > > > > > > it's
>> > > > > > > > > > easier.
>> > > > > > > > >
>> > > > > > > > > Yes, we will support online migrations for the group. That
>> > > means
>> > > > > that
>> > > > > > > > > a group using the old protocol will be able to switch to
>> the
>> > > new
>> > > > > > > > > protocol.
>> > > > > > > > >
>> > > > > > > > > Let me briefly explain how that will work though. It is
>> > > basically a
>> > > > > > > > > four step process:
>> > > > > > > > >
>> > > > > > > > > 1. The cluster must be upgraded or rolled to a software
>> > > supporting
>> > > > > the
>> > > > > > > > > new group coordinator. Both the old and the new
>> coordinator
>> > > will
>> > > > > > > > > support the old protocol and rely on the same persisted
>> > > metadata so
>> > > > > > > > > they can work together. This point is an offline
>> migration.
>> > We
>> > > > > cannot
>> > > > > > > > > do this one live because it would require shutting down
>> the
>> > > current
>> > > > > > > > > coordinator and starting up the new one and that would
>> cause
>> > > > > > > > > unavailabilities.
>> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded to
>> X
>> > in
>> > > > > order
>> > > > > > > > > to enable the new protocol. This cannot be done before 1)
>> is
>> > > > > > > > > terminated because the old coordinator doesn't support the
>> > new
>> > > > > > > > > protocol.
>> > > > > > > > > 3. The consumers must be upgraded to a version supporting
>> the
>> > > > > online
>> > > > > > > > > migration (must have KIP-792). If the consumer is already
>> > > there.
>> > > > > > > > > Nothing must be done at this point.
>> > > > > > > > > 4. The consumers must be rolled with the feature flag
>> turned
>> > > on.
>> > > > > The
>> > > > > > > > > consumer group is automatically converted when the first
>> > > consumer
>> > > > > > > > > using the new protocol joins the group. While the members
>> > > using the
>> > > > > > > > > old protocol are being upgraded, the old protocol is
>> proxied
>> > > into
>> > > > > the
>> > > > > > > > > new one.
>> > > > > > > > >
>> > > > > > > > > Let me clarify all of this in the KIP.
>> > > > > > > > >
>> > > > > > > > > > 4. I am happy that we are pushing the pattern
>> subscriptions
>> > > to
>> > > > > the
>> > > > > > > > > server,
>> > > > > > > > > > but it seems like there could be some tricky
>> compatibility
>> > > > > issues.
>> > > > > > > Will
>> > > > > > > > > we
>> > > > > > > > > > have a mechanism for users to detect that they need to
>> > update
>> > > > > their
>> > > > > > > regex
>> > > > > > > > > > before switching to the new protocol?
>> > > > > > > > >
>> > > > > > > > > I think that I am a bit more optimistic than you on this
>> > > point. I
>> > > > > > > > > believe that the majority of the cases are simple regexes
>> > which
>> > > > > should
>> > > > > > > > > work with the new engine. The coordinator will verify the
>> > regex
>> > > > > anyway
>> > > > > > > > > and reject the consumer if the regex is not valid. Coming
>> > back
>> > > to
>> > > > > the
>> > > > > > > > > migration path, in the worst case, the first upgraded
>> > consumer
>> > > > > joining
>> > > > > > > > > the group will be rejected. This should be used as the
>> last
>> > > > > defence, I
>> > > > > > > > > would say.
>> > > > > > > > >
>> > > > > > > > > One way for customers to validate their regex before
>> > upgrading
>> > > > > their
>> > > > > > > > > prod would be to test them with another group. For
>> instance,
>> > > that
>> > > > > > > > > could be done in a pre-prod environment. Another way
>> would be
>> > > to
>> > > > > > > > > extend the consumer-group tool to provide a regex
>> validation
>> > > > > > > > > mechanism. Would this be enough in your opinion?
>> > > > > > > > >
>> > > > > > > > > > 5. Related to the last question, will the Java client
>> allow
>> > > the
>> > > > > > > users to
>> > > > > > > > > > stick with the current regex engine for compatibility
>> > > reasons?
>> > > > > For
>> > > > > > > > > example,
>> > > > > > > > > > it may be handy to keep using client based regex at
>> first
>> > to
>> > > keep
>> > > > > > > > > > migrations simple and then migrate to server based
>> regexes
>> > > as a
>> > > > > > > second
>> > > > > > > > > step.
>> > > > > > > > >
>> > > > > > > > > I understand your point but I am concerned that this would
>> > > allow
>> > > > > users
>> > > > > > > > > to actually stay in this mode. That would go against our
>> goal
>> > > of
>> > > > > > > > > simplifying the client because we would have to continue
>> > > monitoring
>> > > > > > > > > the metadata on the client side. I would rather not do
>> this.
>> > > > > > > > >
>> > > > > > > > > > 6. When we say that the group coordinator will be
>> > > responsible for
>> > > > > > > storing
>> > > > > > > > > > the configurations and that the configurations will be
>> > > deleted
>> > > > > when
>> > > > > > > the
>> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
>> > deletion
>> > > of
>> > > > > > > > > > configurations?
>> > > > > > > > >
>> > > > > > > > > That's right. The configurations will be deleted when the
>> > > group is
>> > > > > > > > > deleted. They go together.
>> > > > > > > > >
>> > > > > > > > > > 7. Will the choice to store the configs in the group
>> > > coordinator
>> > > > > > > make it
>> > > > > > > > > > harder to list all cluster configs and their values?
>> > > > > > > > >
>> > > > > > > > > I don't think so. The group configurations are overrides
>> of
>> > > cluster
>> > > > > > > > > configs. If you want to know all the overrides though, you
>> > > would
>> > > > > have
>> > > > > > > > > to ask all the group coordinators. You cannot rely on the
>> > > metadata
>> > > > > log
>> > > > > > > > > for instance.
>> > > > > > > > >
>> > > > > > > > > > 8. How would someone configure a group before starting
>> the
>> > > > > consumers?
>> > > > > > > > > Have
>> > > > > > > > > > we considered allowing the explicit creation of groups?
>> > > > > > > Alternatively,
>> > > > > > > > > the
>> > > > > > > > > > configs could be decoupled from the group lifecycle.
>> > > > > > > > >
>> > > > > > > > > Yes. The group will be automatically created in this case.
>> > > However,
>> > > > > > > > > the configs will be lost after the retention period of the
>> > > group
>> > > > > > > > > passes.
>> > > > > > > > >
>> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
>> client
>> > > still
>> > > > > take
>> > > > > > > a
>> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce an
>> > > overload?
>> > > > > > > > >
>> > > > > > > > > That's a very group question. I forgot about that one. As
>> the
>> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with the
>> > > engine
>> > > > > that
>> > > > > > > > > we plan to use, it might be better to deprecate it and
>> use an
>> > > > > overload
>> > > > > > > > > which takes a string. We would rely on the server side
>> > > validation.
>> > > > > > > > > During the migration, I think that we could still try to
>> > > toString
>> > > > > the
>> > > > > > > > > regex and use it. That should work, I think, in the
>> majority
>> > > of the
>> > > > > > > > > cases.
>> > > > > > > > >
>> > > > > > > > > > 10. I agree with Justine that we should be clearer about
>> > the
>> > > > > reason
>> > > > > > > to
>> > > > > > > > > > switch to IBP/metadata.version from the feature flag.
>> Maybe
>> > > we
>> > > > > mean
>> > > > > > > that
>> > > > > > > > > we
>> > > > > > > > > > can switch the default for the feature flag to true
>> based
>> > on
>> > > the
>> > > > > > > > > > metadata.version once we want to make it the default.
>> > > > > > > > >
>> > > > > > > > > My plan was to use that feature flag mainly during the
>> > > development
>> > > > > > > > > phase. I should not have mentioned it, I think, because we
>> > > could
>> > > > > use
>> > > > > > > > > an internal config for it.
>> > > > > > > > >
>> > > > > > > > > > 11. Some of the protocol APIs don't mention the required
>> > > ACLs, it
>> > > > > > > would
>> > > > > > > > > be
>> > > > > > > > > > good to add that for consistency.
>> > > > > > > > >
>> > > > > > > > > Noted.
>> > > > > > > > >
>> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat requires
>> > > "Read
>> > > > > Group"
>> > > > > > > > > even
>> > > > > > > > > > though it seems to do more than reading.
>> > > > > > > > >
>> > > > > > > > > I agree. This is how the current protocol works though. We
>> > only
>> > > > > > > > > require "Read Group" to join a group. We could consider
>> > > changing
>> > > > > this
>> > > > > > > > > but I am not sure that it is worth it.
>> > > > > > > > >
>> > > > > > > > > > 13. How is topic recreation handled by the consumer with
>> > the
>> > > new
>> > > > > > > group
>> > > > > > > > > > protocol? It would be good to have a section on this.
>> > > > > > > > >
>> > > > > > > > > Noted. From a protocol perspective, the new topic will
>> have a
>> > > new
>> > > > > > > > > topic id so it will treat it like a topic with a different
>> > > name.
>> > > > > The
>> > > > > > > > > only issue is that the fetch/commit offsets APIs do not
>> > support
>> > > > > topic
>> > > > > > > > > IDs so the consumer would reuse the offsets based on the
>> > same.
>> > > I
>> > > > > think
>> > > > > > > > > that we should update those APIs as well in order to be
>> > > consistent
>> > > > > end
>> > > > > > > > > to end. That would strengthen the semantics of the
>> consumer.
>> > > > > > > > >
>> > > > > > > > > > 14. The KIP mentions we will write the new coordinator
>> in
>> > > Java.
>> > > > > Even
>> > > > > > > > > though
>> > > > > > > > > > this is an implementation detail, do we plan to have a
>> new
>> > > gradle
>> > > > > > > module
>> > > > > > > > > > for it?
>> > > > > > > > >
>> > > > > > > > > Yes.
>> > > > > > > > >
>> > > > > > > > > > 15. Do we have a scalability goal when it comes to how
>> many
>> > > > > members
>> > > > > > > the
>> > > > > > > > > new
>> > > > > > > > > > group protocol can support?
>> > > > > > > > >
>> > > > > > > > > We don't have numbers at the moment. The protocol should
>> > > support
>> > > > > 1000s
>> > > > > > > > > of members per group. We will measure this when we have a
>> > first
>> > > > > > > > > implementation. Note that we might have other bottlenecks
>> > down
>> > > the
>> > > > > > > > > road (e.g. offset commits).
>> > > > > > > > >
>> > > > > > > > > > 16. Did we consider having SubscribedTopidIds instead of
>> > > > > > > > > > SubscribedTopicNames in ConsumerGroupHeartbeatRequest?
>> Is
>> > the
>> > > > > idea
>> > > > > > > that
>> > > > > > > > > > since we have to resolve the regex on the server, we
>> can do
>> > > the
>> > > > > same
>> > > > > > > for
>> > > > > > > > > > the topic name? The difference is that sending the
>> regex is
>> > > more
>> > > > > > > > > efficient
>> > > > > > > > > > whereas sending the topic names is less efficient.
>> > > Furthermore,
>> > > > > > > delete
>> > > > > > > > > and
>> > > > > > > > > > recreation is easier to handle if we have topic ids.
>> > > > > > > > >
>> > > > > > > > > The idea was to consolidate the metadata lookup on the
>> server
>> > > for
>> > > > > both
>> > > > > > > > > paths but I do agree with your point. As a second though,
>> > using
>> > > > > topic
>> > > > > > > > > ids may be better here for the delete and recreation case.
>> > > Also, I
>> > > > > > > > > suppose that we may allow users to subscribe with topic
>> ids
>> > in
>> > > the
>> > > > > > > > > future because that is the only way to be really robust to
>> > > topic
>> > > > > > > > > re-creation.
>> > > > > > > > >
>> > > > > > > > > Best,
>> > > > > > > > > David
>> > > > > > > > >
>> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
>> > > djacot@confluent.io>
>> > > > > > > wrote:
>> > > > > > > > > >
>> > > > > > > > > > Hi Justine,
>> > > > > > > > > >
>> > > > > > > > > > Thanks for your comments. Please find my answers below.
>> > > > > > > > > >
>> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
>> > > exception
>> > > > > of the
>> > > > > > > > > > topic names based in the ConsumerGroupHeartbeatRequest.
>> I
>> > am
>> > > not
>> > > > > sure
>> > > > > > > > > > if using topic names is the right call here. I need to
>> > think
>> > > > > about it
>> > > > > > > > > > a little more. Obviously, the KIP does not change the
>> > > > > fetch/commit
>> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
>> that
>> > we
>> > > > > should
>> > > > > > > > > > include though as it would give better overall
>> guarantee in
>> > > the
>> > > > > > > > > > producer.
>> > > > > > > > > > - You're right. I think that I should not have mentioned
>> > this
>> > > > > flag at
>> > > > > > > > > > all. I will remove it. We can use an internal
>> configuration
>> > > while
>> > > > > > > > > > developing the feature.
>> > > > > > > > > > - Both cluster types will be supported. The change is
>> > > > > orthogonal. The
>> > > > > > > > > > only requirement is that the cluster uses topic IDs.
>> > > > > > > > > >
>> > > > > > > > > > Best,
>> > > > > > > > > > David
>> > > > > > > > > >
>> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
>> > > > > wangguoz@gmail.com>
>> > > > > > > > > wrote:
>> > > > > > > > > > >
>> > > > > > > > > > > Hi Ismael,
>> > > > > > > > > > >
>> > > > > > > > > > > Thanks for the feedback. Here are some replies inlined
>> > > below:
>> > > > > > > > > > >
>> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
>> > > ismael@juma.me.uk>
>> > > > > > > wrote:
>> > > > > > > > > > >
>> > > > > > > > > > > > Thanks for the KIP. This has the potential to be a
>> > great
>> > > > > > > > > improvement. A few
>> > > > > > > > > > > > initial questions/comments:
>> > > > > > > > > > > >
>> > > > > > > > > > > > 1. I think it's premature to talk about target
>> versions
>> > > for
>> > > > > > > > > deprecation and
>> > > > > > > > > > > > removal of the existing group protocol. Unlike
>> KRaft,
>> > > this
>> > > > > > > affects a
>> > > > > > > > > core
>> > > > > > > > > > > > client protocol and hence deprecation/removal will
>> be
>> > > heavily
>> > > > > > > > > dependent on
>> > > > > > > > > > > > how quickly applications migrate to the new
>> protocol.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Yeah I agree with you. I think we can remove the
>> proposed
>> > > > > timeline
>> > > > > > > in
>> > > > > > > > > the
>> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan` and
>> > > instead
>> > > > > just
>> > > > > > > state
>> > > > > > > > > > > that we will decide in the future about when we would
>> > > > > deprecate old
>> > > > > > > > > > > protocol and behaviors.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
>> but
>> > it
>> > > > > wasn't
>> > > > > > > made
>> > > > > > > > > clear
>> > > > > > > > > > > > why. If we added that as a way to estimate when we'd
>> > > > > deprecate
>> > > > > > > and
>> > > > > > > > > remove
>> > > > > > > > > > > > the group protocol, I also suggest removing this
>> part.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > I think that's not specifically related to the
>> > > > > deprecation/removal
>> > > > > > > > > timeline
>> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
>> > > broker-side
>> > > > > > > > > > > implementation may be done first, and then the client
>> > side,
>> > > > > and we
>> > > > > > > > > would
>> > > > > > > > > > > only mark it as "released" by the time clients
>> > > implementations
>> > > > > are
>> > > > > > > > > done. At
>> > > > > > > > > > > that time, to enable the feature the clients need to
>> > first
>> > > > > swap-in
>> > > > > > > the
>> > > > > > > > > > > bytecode with a rolling bounce and then set the flag
>> > with a
>> > > > > second
>> > > > > > > > > rolling
>> > > > > > > > > > > bounce, and hence we feel it's better to be released
>> in a
>> > > major
>> > > > > > > > > version.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 3. We need to flesh out the details of the migration
>> > > story.
>> > > > > It
>> > > > > > > > > sounds like
>> > > > > > > > > > > > we're saying we will support online migrations. Is
>> that
>> > > > > correct?
>> > > > > > > We
>> > > > > > > > > should
>> > > > > > > > > > > > explain this in detail. It could also be done as a
>> > > separate
>> > > > > KIP,
>> > > > > > > if
>> > > > > > > > > it's
>> > > > > > > > > > > > easier.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Yes I think that's the part we can be more concrete
>> about
>> > > for
>> > > > > sure
>> > > > > > > (and
>> > > > > > > > > > > this is related to your question 2) above). We will
>> work
>> > on
>> > > > > making
>> > > > > > > it
>> > > > > > > > > more
>> > > > > > > > > > > explicit in parallel as we solicit more feedback.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
>> > > subscriptions
>> > > > > to
>> > > > > > > the
>> > > > > > > > > server,
>> > > > > > > > > > > > but it seems like there could be some tricky
>> > > compatibility
>> > > > > > > issues.
>> > > > > > > > > Will we
>> > > > > > > > > > > > have a mechanism for users to detect that they need
>> to
>> > > update
>> > > > > > > their
>> > > > > > > > > regex
>> > > > > > > > > > > > before switching to the new protocol?
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Yes I think we need some tooling for non-java client
>> > users
>> > > to
>> > > > > sort
>> > > > > > > of
>> > > > > > > > > > > "dry-run" the client before switching to the new
>> > protocol.
>> > > I
>> > > > > do not
>> > > > > > > > > have a
>> > > > > > > > > > > specific idea on top of my head though, maybe others
>> like
>> > > @Matt
>> > > > > > > > > Howlett can
>> > > > > > > > > > > chime-in here?
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 5. Related to the last question, will the Java
>> client
>> > > allow
>> > > > > the
>> > > > > > > > > users to
>> > > > > > > > > > > > stick with the current regex engine for
>> compatibility
>> > > > > reasons?
>> > > > > > > For
>> > > > > > > > > example,
>> > > > > > > > > > > > it may be handy to keep using client based regex at
>> > > first to
>> > > > > keep
>> > > > > > > > > > > > migrations simple and then migrate to server based
>> > > regexes
>> > > > > as a
>> > > > > > > > > second
>> > > > > > > > > > > > step.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Honestly I have not thought about that for java
>> clients,
>> > > and
>> > > > > we can
>> > > > > > > > > discuss
>> > > > > > > > > > > that. What kind of compatibility issues do you have in
>> > > mind?
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 6. When we say that the group coordinator will be
>> > > > > responsible for
>> > > > > > > > > storing
>> > > > > > > > > > > > the configurations and that the configurations will
>> be
>> > > > > deleted
>> > > > > > > when
>> > > > > > > > > the
>> > > > > > > > > > > > group is deleted. Will a transition to DEAD trigger
>> > > deletion
>> > > > > of
>> > > > > > > > > > > > configurations?
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Yes, since the DEAD state is an ending state (we would
>> > only
>> > > > > > > transit to
>> > > > > > > > > that
>> > > > > > > > > > > state when the group is EMPTY and also all of its
>> > metadata
>> > > are
>> > > > > > > gone),
>> > > > > > > > > once
>> > > > > > > > > > > it's transited to DEAD this group would never be
>> revived.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 7. Will the choice to store the configs in the group
>> > > > > coordinator
>> > > > > > > > > make it
>> > > > > > > > > > > > harder to list all cluster configs and their values?
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > That's a good question, and our thoughts are that the
>> > > so-called
>> > > > > > > "group
>> > > > > > > > > > > configurations" are overrides of the cluster-level
>> > > > > configurations
>> > > > > > > > > > > customized per group so when an admin list cluster
>> > configs
>> > > it's
>> > > > > > > okay to
>> > > > > > > > > > > list just the cluster-level "defaults", not showing
>> any
>> > > > > per-group
>> > > > > > > > > > > customizations.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 8. How would someone configure a group before
>> starting
>> > > the
>> > > > > > > > > consumers? Have
>> > > > > > > > > > > > we considered allowing the explicit creation of
>> groups?
>> > > > > > > > > Alternatively, the
>> > > > > > > > > > > > configs could be decoupled from the group lifecycle.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > The configs can be created before the group itself as
>> an
>> > > > > > > independent
>> > > > > > > > > entity
>> > > > > > > > > > > --- of course, this requires the corresponding
>> request to
>> > > be
>> > > > > > > routed to
>> > > > > > > > > the
>> > > > > > > > > > > right coordinator based on the group id --- the only
>> > thing
>> > > that
>> > > > > > > > > differs is,
>> > > > > > > > > > > when the group itself is gone we also check if there
>> are
>> > > any
>> > > > > > > > > configuration
>> > > > > > > > > > > entities related to that group and delete as well.
>> > > > > > > > > > >
>> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on the
>> > > creation
>> > > > > /
>> > > > > > > > > deletion
>> > > > > > > > > > > lifecycles of the config entities, and we would like
>> to
>> > > hear
>> > > > > > > everyone's
>> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
>> totally
>> > > > > decouple
>> > > > > > > group
>> > > > > > > > > > > configs and hence not delete them at all when the
>> group
>> > is
>> > > > > gone,
>> > > > > > > but
>> > > > > > > > > always
>> > > > > > > > > > > require explicit deletion operations by themselves.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
>> > client
>> > > > > still
>> > > > > > > take
>> > > > > > > > > a
>> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
>> introduce an
>> > > > > overload?
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > I think we do not need to introduce an overload, but
>> I'm
>> > > all
>> > > > > ears
>> > > > > > > if
>> > > > > > > > > there
>> > > > > > > > > > > may be some compatibility issues that we may overlook.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 10. I agree with Justine that we should be clearer
>> > about
>> > > the
>> > > > > > > reason
>> > > > > > > > > to
>> > > > > > > > > > > > switch to IBP/metadata.version from the feature
>> flag.
>> > > Maybe
>> > > > > we
>> > > > > > > mean
>> > > > > > > > > that we
>> > > > > > > > > > > > can switch the default for the feature flag to true
>> > > based on
>> > > > > the
>> > > > > > > > > > > > metadata.version once we want to make it the
>> default.
>> > > > > > > > > > >
>> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
>> required
>> > > ACLs,
>> > > > > it
>> > > > > > > > > would be
>> > > > > > > > > > > > good to add that for consistency.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Ack, we can certainly do that.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
>> > requires
>> > > > > "Read
>> > > > > > > > > Group" even
>> > > > > > > > > > > > though it seems to do more than reading.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > I had that thought myself as well, but in the end we
>> > could
>> > > not
>> > > > > > > find a
>> > > > > > > > > > > better alternative: adding Write Group seems an
>> overkill
>> > > here
>> > > > > > > since we
>> > > > > > > > > do
>> > > > > > > > > > > not have it elsewhere (we only have Read / Delete and
>> > > Describe
>> > > > > on
>> > > > > > > > > groups so
>> > > > > > > > > > > far). Would like to hear others thoughts.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 13. How is topic recreation handled by the consumer
>> > with
>> > > the
>> > > > > new
>> > > > > > > > > group
>> > > > > > > > > > > > protocol? It would be good to have a section on
>> this.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > You mean with regex subscription right? Yes we can
>> add a
>> > > > > section
>> > > > > > > about
>> > > > > > > > > > > that, but basically the idea is that consumer would be
>> > > totally
>> > > > > > > > > agnostic in
>> > > > > > > > > > > the new protocol as it's handled all by the brokers.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 14. The KIP mentions we will write the new
>> coordinator
>> > in
>> > > > > Java.
>> > > > > > > Even
>> > > > > > > > > though
>> > > > > > > > > > > > this is an implementation detail, do we plan to
>> have a
>> > > new
>> > > > > gradle
>> > > > > > > > > module
>> > > > > > > > > > > > for it?
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > We have not thought about that. But I think the answer
>> > > should
>> > > > > be
>> > > > > > > yes.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 15. Do we have a scalability goal when it comes to
>> how
>> > > many
>> > > > > > > members
>> > > > > > > > > the new
>> > > > > > > > > > > > group protocol can support?
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Within a group, I think we should shoot for 1000s of
>> > > members.
>> > > > > But
>> > > > > > > that
>> > > > > > > > > > > scalability goals also depend on the offset management
>> > > (commit,
>> > > > > > > fetch)
>> > > > > > > > > > > capabilities of the coordinator which we did not
>> cover in
>> > > this
>> > > > > > > KIP, so
>> > > > > > > > > it's
>> > > > > > > > > > > hard to give a number that applies universally.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
>> instead
>> > of
>> > > > > > > > > > > > SubscribedTopicNames in
>> ConsumerGroupHeartbeatRequest?
>> > > Is the
>> > > > > > > idea
>> > > > > > > > > that
>> > > > > > > > > > > > since we have to resolve the regex on the server, we
>> > can
>> > > do
>> > > > > the
>> > > > > > > same
>> > > > > > > > > for
>> > > > > > > > > > > > the topic name? The difference is that sending the
>> > regex
>> > > is
>> > > > > more
>> > > > > > > > > efficient
>> > > > > > > > > > > > whereas sending the topic names is less efficient.
>> > > > > Furthermore,
>> > > > > > > > > delete and
>> > > > > > > > > > > > recreation is easier to handle if we have topic ids.
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > The main reason to still let the clients send names
>> is to
>> > > keep
>> > > > > the
>> > > > > > > > > > > reasoning of names -> ids on the broker / admin client
>> > > only.
>> > > > > Note
>> > > > > > > that
>> > > > > > > > > > > although we added topic id in KIP-516, we never
>> > > implemented the
>> > > > > > > logic
>> > > > > > > > > on
>> > > > > > > > > > > consumer/producers leveraging the related newer
>> versioned
>> > > RPCs,
>> > > > > > > > > instead we
>> > > > > > > > > > > just set the topic id as empty UUID. We want to keep
>> the
>> > > > > > > > > consumer/producer
>> > > > > > > > > > > to be thin and only delegate the reasoning on broker
>> and
>> > > > > > > potentially
>> > > > > > > > > admin
>> > > > > > > > > > > clients.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > > > > Thanks,
>> > > > > > > > > > > > Ismael
>> > > > > > > > > > > >
>> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
>> > > > > > > > > <dj...@confluent.io.invalid>
>> > > > > > > > > > > > wrote:
>> > > > > > > > > > > >
>> > > > > > > > > > > > > Hi all,
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > I would like to start a discussion thread on
>> KIP-848:
>> > > The
>> > > > > Next
>> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
>> With
>> > > this
>> > > > > KIP,
>> > > > > > > we
>> > > > > > > > > aim
>> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
>> more
>> > > > > reliable,
>> > > > > > > more
>> > > > > > > > > > > > > scalable, easier to implement for clients, and
>> easier
>> > > to
>> > > > > debug
>> > > > > > > for
>> > > > > > > > > > > > > operators.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > The KIP is here:
>> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > Please take a look and let me know what you think.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > Best,
>> > > > > > > > > > > > > David
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > PS: I will be away from July 18th to August 8th.
>> That
>> > > gives
>> > > > > > > you a
>> > > > > > > > > bit
>> > > > > > > > > > > > > of time to read and digest this long KIP.
>> > > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > --
>> > > > > > > > > > > -- Guozhang
>> > > > > > > > >
>> > > > > > > >
>> > > > > > > >
>> > > > > > > > --
>> > > > > > > > -- Guozhang
>> > > > > > >
>> > > > >
>> > >
>> > >
>> >
>>
>>
>> --
>> -- Guozhang
>>
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Thanks Sagar, I made a quick read on your doc, and am also leaning towards
having a separate API for connect. Would also love to hear @Randall Hauch
<rh...@gmail.com> @Konstantin 's opinion about that.

One thing is that for Kafka Connect, their compatibility story, especially
downgrading is more flexible than consumer (you can find that from
https://cwiki.apache.org/confluence/display/KAFKA/KIP-415%3A+Incremental+Cooperative+Rebalancing+in+Kafka+Connect)
so maybe we need to think a bit differently for the compatibility.


Guozhang

On Wed, Aug 3, 2022 at 5:39 AM Sagar <sa...@gmail.com> wrote:

> Hi Guozhang/David,
>
> I created a confluence page to discuss how Connect would need to change
> based on the new rebalance protocol. Here's the page:
>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>
> It's also pretty longish and I have tried to keep a format similar to
> KIP-848. Let me know what you think. Also, do you think this should be
> moved to a separate discussion thread or is this one fine?
>
> Thanks!
> Sagar.
>
> On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
>
> > Hello Guozhang,
> >
> > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > analysis and come up with such a document.
> >
> > Thanks!
> > Sagar.
> >
> > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> wrote:
> >
> >> Hello Sagar,
> >>
> >> It would be great if you could come back with some analysis on how to
> >> implement the Connect side integration with the new protocol; so far
> >> besides leveraging on the new "protocol type" we did not yet think
> through
> >> the Connect side implementations. For Streams here's a draft of
> >> integration
> >> plan:
> >>
> >>
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >> just FYI for your analysis on Connect.
> >>
> >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> wrote:
> >>
> >> > Hi David,
> >> >
> >> > Thank you for your response. The reason I thought connect can also fit
> >> into
> >> > this new scheme is that even today the connect uses a
> WorkerCoordinator
> >> > extending from AbstractCoordinator to empower rebalances of
> >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> >> connect
> >> > and uses the metadata() method by plumbing into
> >> JoinGroupRequestProtocol.
> >> >
> >> > I think the changes to support connect would be similar at a high
> level
> >> to
> >> > the changes in streams mainly because of the Client side assignors
> being
> >> > used in both. At an implementation level, we might need to make a lot
> of
> >> > changes to get onto this new assignment protocol like enhancing the
> >> > JoinGroup request/response and SyncGroup and using
> >> ConsumerGroupHeartbeat
> >> > API etc again on similar lines to streams (or there might be
> >> deviations). I
> >> > would try to perform a detailed analysis of the same  and we can have
> a
> >> > separate discussion thread for that as that would derail this
> discussion
> >> > thread. Let me know if that sounds good to you.
> >> >
> >> > Thanks!
> >> > Sagar.
> >> >
> >> >
> >> >
> >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> <djacot@confluent.io.invalid
> >> >
> >> > wrote:
> >> >
> >> > > Hi Sagar,
> >> > >
> >> > > Thanks for your comments.
> >> > >
> >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> >> > >
> >> > > 2) The idea is to transition C from his current assignment to his
> >> > > target assignment when he can move to epoch 3. When that happens,
> the
> >> > > member assignment is updated and persisted with all its assigned
> >> > > partitions even if they are not all revoked yet. In other words, the
> >> > > member assignment becomes the target assignment. This is basically
> an
> >> > > optimization to avoid having to write all the changes to the log.
> The
> >> > > examples are based on the persisted state so I understand the
> >> > > confusion. Let me see if I can improve this in the description.
> >> > >
> >> > > 3) Regarding Connect, it could reuse the protocol with a client side
> >> > > assignor if it fits in the protocol. The assignment is about
> >> > > topicid-partitions + metadata, could Connect fit into this?
> >> > >
> >> > > Best,
> >> > > David
> >> > >
> >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> >> wrote:
> >> > > >
> >> > > > Hi David,
> >> > > >
> >> > > > Thanks for the KIP. I just had minor observations:
> >> > > >
> >> > > > 1) In the Assignment Error section in Client Side mode Assignment
> >> > > process,
> >> > > > you mentioned => `In this case, the client side assignor can
> return
> >> an
> >> > > > error to the group coordinator`. In this case are you referring to
> >> the
> >> > > > Assignor returning an AssignmentError that's listed down towards
> the
> >> > end?
> >> > > > If yes, do you think it would make sense to mention this
> explicitly
> >> > here?
> >> > > >
> >> > > > 2) In the Case Studies section, I have a slight confusion, not
> sure
> >> if
> >> > > > others have the same. Consider this step:
> >> > > >
> >> > > > When B heartbeats, the group coordinator transitions him to epoch
> 3
> >> > > because
> >> > > > B has no partitions to revoke. It persists the change and reply.
> >> > > >
> >> > > >    - Group (epoch=3)
> >> > > >       - A
> >> > > >       - B
> >> > > >       - C
> >> > > >    - Target Assignment (epoch=3)
> >> > > >       - A - partitions=[foo-0]
> >> > > >       - B - partitions=[foo-2]
> >> > > >       - C - partitions=[foo-1]
> >> > > >    - Member Assignment
> >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> >> > > >       - B - epoch=3, partitions=[foo-2]
> >> > > >       - C - epoch=3, partitions=[foo-1]
> >> > > >
> >> > > > When C heartbeats, it transitions to epoch 3 but cannot get foo-1
> >> yet.
> >> > > >
> >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> yet,
> >> > but
> >> > > > based on the description above, it seems it already has it. Do you
> >> > think
> >> > > it
> >> > > > would be better to remove it and populate it only when it actually
> >> gets
> >> > > it?
> >> > > > I see this in a lot of other places, so have I understood it
> >> > incorrectly
> >> > > ?
> >> > > >
> >> > > >
> >> > > > Regarding connect , it might be out of scope of this discussion,
> but
> >> > from
> >> > > > what I understood it would probably be running in client side
> >> assignor
> >> > > mode
> >> > > > even on the new rebalance protocol as it has its own Custom
> >> > > Assignors(Eager
> >> > > > and IncrementalCooperative).
> >> > > >
> >> > > > Thanks!
> >> > > >
> >> > > > Sagar.
> >> > > >
> >> > > >
> >> > > >
> >> > > >
> >> > > >
> >> > > >
> >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> >> > <djacot@confluent.io.invalid
> >> > > >
> >> > > > wrote:
> >> > > >
> >> > > > > Thanks Hector! Our goal is to move forward with specialized API
> >> > > > > instead of relying on one generic API. For Connect, we can apply
> >> the
> >> > > > > exact same pattern and reuse/share the core implementation on
> the
> >> > > > > server side. For the schema registry, I think that we should
> >> consider
> >> > > > > having a tailored API to do simple membership/leader election.
> >> > > > >
> >> > > > > Best,
> >> > > > > David
> >> > > > >
> >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <ismael@juma.me.uk
> >
> >> > > wrote:
> >> > > > > >
> >> > > > > > Three quick comments:
> >> > > > > >
> >> > > > > > 1. Regarding java.util.regex.Pattern vs
> >> com.google.re2j.Pattern, we
> >> > > > > should
> >> > > > > > document the differences in more detail before deciding one
> way
> >> or
> >> > > > > another.
> >> > > > > > That said, if people pass java.util.regex.Pattern, they expect
> >> > their
> >> > > > > > semantics to be honored. If we are doing something different,
> >> then
> >> > we
> >> > > > > > should consider adding an overload with our own Pattern class
> (I
> >> > > don't
> >> > > > > > think we'd want to expose re2j's at this point).
> >> > > > > > 2. Regarding topic ids, any major new protocol should
> integrate
> >> > fully
> >> > > > > with
> >> > > > > > it and should handle the topic recreation case correctly.
> That's
> >> > the
> >> > > main
> >> > > > > > part we need to handle. I agree with David that we'd want to
> add
> >> > > topic
> >> > > > > ids
> >> > > > > > to the relevant protocols that don't have it yet and that we
> can
> >> > > probably
> >> > > > > > focus on the internals versus adding new APIs to the Java
> >> Consumer
> >> > > > > (unless
> >> > > > > > we find that adding new APIs is required for reasonable
> >> semantics).
> >> > > > > > 3. I am still not sure about the coordinator storing the
> >> configs.
> >> > > It's
> >> > > > > > powerful for configs to be centralized in the metadata log for
> >> > > various
> >> > > > > > reasons (auditability, visibility, consistency, etc.).
> >> Similarly, I
> >> > > am
> >> > > > > not
> >> > > > > > sure about automatically deleting configs in a way that they
> >> cannot
> >> > > be
> >> > > > > > recovered. A good property for modern systems is to minimize
> the
> >> > > number
> >> > > > > of
> >> > > > > > unrecoverable data loss scenarios.
> >> > > > > >
> >> > > > > > Ismael
> >> > > > > >
> >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> >> > > <djacot@confluent.io.invalid
> >> > > > > >
> >> > > > > > wrote:
> >> > > > > >
> >> > > > > > > Thanks Guozhang. My answers are below:
> >> > > > > > >
> >> > > > > > > > 1) the migration path, especially the last step when
> clients
> >> > > flip the
> >> > > > > > > flag
> >> > > > > > > > to enable the new protocol, in which we would have a
> window
> >> > where
> >> > > > > both
> >> > > > > > > new
> >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> >> members
> >> > of
> >> > > the
> >> > > > > same
> >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> >> while
> >> > > > > using the
> >> > > > > > > > new protocol is something we need to present about.
> >> > > > > > >
> >> > > > > > > Noted. I just published a new version of KIP which includes
> >> more
> >> > > > > > > details about this. See the "Supporting Online Consumer
> Group
> >> > > Upgrade"
> >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan". I
> >> think
> >> > > that
> >> > > > > > > I have to think through a few cases now but the overall idea
> >> and
> >> > > > > > > mechanism should be understandable.
> >> > > > > > >
> >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> >> are
> >> > > only
> >> > > > > used
> >> > > > > > > as
> >> > > > > > > > part of RPCs and admin client, but they are not exposed
> via
> >> any
> >> > > > > public
> >> > > > > > > APIs
> >> > > > > > > > to consumers yet. I think the question is, first should we
> >> let
> >> > > the
> >> > > > > > > consumer
> >> > > > > > > > client to be maintaining the names -> ids mapping itself
> to
> >> > fully
> >> > > > > > > leverage
> >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> the
> >> > > topic
> >> > > > > ids;
> >> > > > > > > and
> >> > > > > > > > secondly, should we ever consider exposing the topic ids
> in
> >> the
> >> > > > > consumer
> >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> >> the
> >> > > > > rebalance
> >> > > > > > > > listener for cases like topic deletion-and-recreation).
> >> > > > > > >
> >> > > > > > > a) Assuming that we would include converting all the offsets
> >> > > related
> >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> >> able
> >> > to
> >> > > > > > > fully operate with topic ids. That being said, it still has
> to
> >> > > provide
> >> > > > > > > the topics names in various APIs so having a mapping in the
> >> > > consumer
> >> > > > > > > seems inevitable to me.
> >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> >> this
> >> > > goes
> >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> >> > internals
> >> > > > > > > here and we can consider this separately if we see value in
> >> doing
> >> > > it.
> >> > > > > > >
> >> > > > > > > Coming back to Ismael's point about using topic ids in the
> >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> >> > advantage
> >> > > in
> >> > > > > > > favour of it. The consumer will have the opportunity to
> >> validate
> >> > > that
> >> > > > > > > the topics exists before passing them into the group
> rebalance
> >> > > > > > > protocol. Obviously, the coordinator will also notice it but
> >> it
> >> > > does
> >> > > > > > > not really have a way to reject an invalid topic in the
> >> response.
> >> > > > > > >
> >> > > > > > > > I'm agreeing with David on all other minor questions
> except
> >> for
> >> > > the
> >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> >> > > necessary
> >> > > > > to
> >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> >> still
> >> > > use
> >> > > > > > > Pattern
> >> > > > > > > > while just documenting that our subscription may be
> >> rejected by
> >> > > the
> >> > > > > > > server.
> >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> >> > using
> >> > > an
> >> > > > > > > > overloaded `String` based subscription may be more
> >> vulnerable
> >> > to
> >> > > > > various
> >> > > > > > > > invalid regexes.
> >> > > > > > >
> >> > > > > > > That could work. I have to look at the differences between
> the
> >> > two
> >> > > > > > > engines to better understand the potential issues. My
> >> > > understanding is
> >> > > > > > > that would work for all the basic regular expressions. The
> >> > > differences
> >> > > > > > > between the two are mainly about the various character
> >> classes. I
> >> > > > > > > wonder what other people think about this.
> >> > > > > > >
> >> > > > > > > Best,
> >> > > > > > > David
> >> > > > > > >
> >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> >> > wangguoz@gmail.com
> >> > > >
> >> > > > > wrote:
> >> > > > > > > >
> >> > > > > > > > Thanks David! I think on the high level there are two meta
> >> > > points we
> >> > > > > need
> >> > > > > > > > to concretize a bit more:
> >> > > > > > > >
> >> > > > > > > > 1) the migration path, especially the last step when
> clients
> >> > > flip the
> >> > > > > > > flag
> >> > > > > > > > to enable the new protocol, in which we would have a
> window
> >> > where
> >> > > > > both
> >> > > > > > > new
> >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> >> members
> >> > of
> >> > > the
> >> > > > > same
> >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> >> while
> >> > > > > using the
> >> > > > > > > > new protocol is something we need to present about.
> >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> >> are
> >> > > only
> >> > > > > used
> >> > > > > > > as
> >> > > > > > > > part of RPCs and admin client, but they are not exposed
> via
> >> any
> >> > > > > public
> >> > > > > > > APIs
> >> > > > > > > > to consumers yet. I think the question is, first should we
> >> let
> >> > > the
> >> > > > > > > consumer
> >> > > > > > > > client to be maintaining the names -> ids mapping itself
> to
> >> > fully
> >> > > > > > > leverage
> >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> the
> >> > > topic
> >> > > > > ids;
> >> > > > > > > and
> >> > > > > > > > secondly, should we ever consider exposing the topic ids
> in
> >> the
> >> > > > > consumer
> >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> >> the
> >> > > > > rebalance
> >> > > > > > > > listener for cases like topic deletion-and-recreation).
> >> > > > > > > >
> >> > > > > > > > I'm agreeing with David on all other minor questions
> except
> >> for
> >> > > the
> >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> >> > > necessary
> >> > > > > to
> >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> >> still
> >> > > use
> >> > > > > > > Pattern
> >> > > > > > > > while just documenting that our subscription may be
> >> rejected by
> >> > > the
> >> > > > > > > server.
> >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> >> > using
> >> > > an
> >> > > > > > > > overloaded `String` based subscription may be more
> >> vulnerable
> >> > to
> >> > > > > various
> >> > > > > > > > invalid regexes.
> >> > > > > > > >
> >> > > > > > > >
> >> > > > > > > > Guozhang
> >> > > > > > > >
> >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> >> > > > > <djacot@confluent.io.invalid
> >> > > > > > > >
> >> > > > > > > > wrote:
> >> > > > > > > >
> >> > > > > > > > > Hi Ismael,
> >> > > > > > > > >
> >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> >> > inline.
> >> > > > > > > > >
> >> > > > > > > > > > 1. I think it's premature to talk about target
> versions
> >> for
> >> > > > > > > deprecation
> >> > > > > > > > > and
> >> > > > > > > > > > removal of the existing group protocol. Unlike KRaft,
> >> this
> >> > > > > affects a
> >> > > > > > > core
> >> > > > > > > > > > client protocol and hence deprecation/removal will be
> >> > heavily
> >> > > > > > > dependent
> >> > > > > > > > > on
> >> > > > > > > > > > how quickly applications migrate to the new protocol.
> >> > > > > > > > >
> >> > > > > > > > > That makes sense. I will remove it.
> >> > > > > > > > >
> >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x, but
> it
> >> > > wasn't
> >> > > > > made
> >> > > > > > > > > clear
> >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> >> > > deprecate
> >> > > > > and
> >> > > > > > > remove
> >> > > > > > > > > > the group protocol, I also suggest removing this part.
> >> > > > > > > > >
> >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> >> rewrite
> >> > > the
> >> > > > > group
> >> > > > > > > > > coordinator in Java while we implement the new protocol.
> >> This
> >> > > means
> >> > > > > > > > > that the internals will be slightly different (e.g.
> >> threading
> >> > > > > model).
> >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> >> group
> >> > > > > > > > > coordinator to the new group coordinator to a major
> >> release.
> >> > > The
> >> > > > > > > > > alternative would be to use a flag to do the switch
> >> instead
> >> > of
> >> > > > > relying
> >> > > > > > > > > on the software upgrade.
> >> > > > > > > > >
> >> > > > > > > > > > 3. We need to flesh out the details of the migration
> >> story.
> >> > > It
> >> > > > > sounds
> >> > > > > > > > > like
> >> > > > > > > > > > we're saying we will support online migrations. Is
> that
> >> > > correct?
> >> > > > > We
> >> > > > > > > > > should
> >> > > > > > > > > > explain this in detail. It could also be done as a
> >> separate
> >> > > KIP,
> >> > > > > if
> >> > > > > > > it's
> >> > > > > > > > > > easier.
> >> > > > > > > > >
> >> > > > > > > > > Yes, we will support online migrations for the group.
> That
> >> > > means
> >> > > > > that
> >> > > > > > > > > a group using the old protocol will be able to switch to
> >> the
> >> > > new
> >> > > > > > > > > protocol.
> >> > > > > > > > >
> >> > > > > > > > > Let me briefly explain how that will work though. It is
> >> > > basically a
> >> > > > > > > > > four step process:
> >> > > > > > > > >
> >> > > > > > > > > 1. The cluster must be upgraded or rolled to a software
> >> > > supporting
> >> > > > > the
> >> > > > > > > > > new group coordinator. Both the old and the new
> >> coordinator
> >> > > will
> >> > > > > > > > > support the old protocol and rely on the same persisted
> >> > > metadata so
> >> > > > > > > > > they can work together. This point is an offline
> >> migration.
> >> > We
> >> > > > > cannot
> >> > > > > > > > > do this one live because it would require shutting down
> >> the
> >> > > current
> >> > > > > > > > > coordinator and starting up the new one and that would
> >> cause
> >> > > > > > > > > unavailabilities.
> >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> to
> >> X
> >> > in
> >> > > > > order
> >> > > > > > > > > to enable the new protocol. This cannot be done before
> 1)
> >> is
> >> > > > > > > > > terminated because the old coordinator doesn't support
> the
> >> > new
> >> > > > > > > > > protocol.
> >> > > > > > > > > 3. The consumers must be upgraded to a version
> supporting
> >> the
> >> > > > > online
> >> > > > > > > > > migration (must have KIP-792). If the consumer is
> already
> >> > > there.
> >> > > > > > > > > Nothing must be done at this point.
> >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> >> turned
> >> > > on.
> >> > > > > The
> >> > > > > > > > > consumer group is automatically converted when the first
> >> > > consumer
> >> > > > > > > > > using the new protocol joins the group. While the
> members
> >> > > using the
> >> > > > > > > > > old protocol are being upgraded, the old protocol is
> >> proxied
> >> > > into
> >> > > > > the
> >> > > > > > > > > new one.
> >> > > > > > > > >
> >> > > > > > > > > Let me clarify all of this in the KIP.
> >> > > > > > > > >
> >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> >> subscriptions
> >> > > to
> >> > > > > the
> >> > > > > > > > > server,
> >> > > > > > > > > > but it seems like there could be some tricky
> >> compatibility
> >> > > > > issues.
> >> > > > > > > Will
> >> > > > > > > > > we
> >> > > > > > > > > > have a mechanism for users to detect that they need to
> >> > update
> >> > > > > their
> >> > > > > > > regex
> >> > > > > > > > > > before switching to the new protocol?
> >> > > > > > > > >
> >> > > > > > > > > I think that I am a bit more optimistic than you on this
> >> > > point. I
> >> > > > > > > > > believe that the majority of the cases are simple
> regexes
> >> > which
> >> > > > > should
> >> > > > > > > > > work with the new engine. The coordinator will verify
> the
> >> > regex
> >> > > > > anyway
> >> > > > > > > > > and reject the consumer if the regex is not valid.
> Coming
> >> > back
> >> > > to
> >> > > > > the
> >> > > > > > > > > migration path, in the worst case, the first upgraded
> >> > consumer
> >> > > > > joining
> >> > > > > > > > > the group will be rejected. This should be used as the
> >> last
> >> > > > > defence, I
> >> > > > > > > > > would say.
> >> > > > > > > > >
> >> > > > > > > > > One way for customers to validate their regex before
> >> > upgrading
> >> > > > > their
> >> > > > > > > > > prod would be to test them with another group. For
> >> instance,
> >> > > that
> >> > > > > > > > > could be done in a pre-prod environment. Another way
> >> would be
> >> > > to
> >> > > > > > > > > extend the consumer-group tool to provide a regex
> >> validation
> >> > > > > > > > > mechanism. Would this be enough in your opinion?
> >> > > > > > > > >
> >> > > > > > > > > > 5. Related to the last question, will the Java client
> >> allow
> >> > > the
> >> > > > > > > users to
> >> > > > > > > > > > stick with the current regex engine for compatibility
> >> > > reasons?
> >> > > > > For
> >> > > > > > > > > example,
> >> > > > > > > > > > it may be handy to keep using client based regex at
> >> first
> >> > to
> >> > > keep
> >> > > > > > > > > > migrations simple and then migrate to server based
> >> regexes
> >> > > as a
> >> > > > > > > second
> >> > > > > > > > > step.
> >> > > > > > > > >
> >> > > > > > > > > I understand your point but I am concerned that this
> would
> >> > > allow
> >> > > > > users
> >> > > > > > > > > to actually stay in this mode. That would go against our
> >> goal
> >> > > of
> >> > > > > > > > > simplifying the client because we would have to continue
> >> > > monitoring
> >> > > > > > > > > the metadata on the client side. I would rather not do
> >> this.
> >> > > > > > > > >
> >> > > > > > > > > > 6. When we say that the group coordinator will be
> >> > > responsible for
> >> > > > > > > storing
> >> > > > > > > > > > the configurations and that the configurations will be
> >> > > deleted
> >> > > > > when
> >> > > > > > > the
> >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> >> > deletion
> >> > > of
> >> > > > > > > > > > configurations?
> >> > > > > > > > >
> >> > > > > > > > > That's right. The configurations will be deleted when
> the
> >> > > group is
> >> > > > > > > > > deleted. They go together.
> >> > > > > > > > >
> >> > > > > > > > > > 7. Will the choice to store the configs in the group
> >> > > coordinator
> >> > > > > > > make it
> >> > > > > > > > > > harder to list all cluster configs and their values?
> >> > > > > > > > >
> >> > > > > > > > > I don't think so. The group configurations are overrides
> >> of
> >> > > cluster
> >> > > > > > > > > configs. If you want to know all the overrides though,
> you
> >> > > would
> >> > > > > have
> >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> the
> >> > > metadata
> >> > > > > log
> >> > > > > > > > > for instance.
> >> > > > > > > > >
> >> > > > > > > > > > 8. How would someone configure a group before starting
> >> the
> >> > > > > consumers?
> >> > > > > > > > > Have
> >> > > > > > > > > > we considered allowing the explicit creation of
> groups?
> >> > > > > > > Alternatively,
> >> > > > > > > > > the
> >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> >> > > > > > > > >
> >> > > > > > > > > Yes. The group will be automatically created in this
> case.
> >> > > However,
> >> > > > > > > > > the configs will be lost after the retention period of
> the
> >> > > group
> >> > > > > > > > > passes.
> >> > > > > > > > >
> >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> >> client
> >> > > still
> >> > > > > take
> >> > > > > > > a
> >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> an
> >> > > overload?
> >> > > > > > > > >
> >> > > > > > > > > That's a very group question. I forgot about that one.
> As
> >> the
> >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> the
> >> > > engine
> >> > > > > that
> >> > > > > > > > > we plan to use, it might be better to deprecate it and
> >> use an
> >> > > > > overload
> >> > > > > > > > > which takes a string. We would rely on the server side
> >> > > validation.
> >> > > > > > > > > During the migration, I think that we could still try to
> >> > > toString
> >> > > > > the
> >> > > > > > > > > regex and use it. That should work, I think, in the
> >> majority
> >> > > of the
> >> > > > > > > > > cases.
> >> > > > > > > > >
> >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> about
> >> > the
> >> > > > > reason
> >> > > > > > > to
> >> > > > > > > > > > switch to IBP/metadata.version from the feature flag.
> >> Maybe
> >> > > we
> >> > > > > mean
> >> > > > > > > that
> >> > > > > > > > > we
> >> > > > > > > > > > can switch the default for the feature flag to true
> >> based
> >> > on
> >> > > the
> >> > > > > > > > > > metadata.version once we want to make it the default.
> >> > > > > > > > >
> >> > > > > > > > > My plan was to use that feature flag mainly during the
> >> > > development
> >> > > > > > > > > phase. I should not have mentioned it, I think, because
> we
> >> > > could
> >> > > > > use
> >> > > > > > > > > an internal config for it.
> >> > > > > > > > >
> >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> required
> >> > > ACLs, it
> >> > > > > > > would
> >> > > > > > > > > be
> >> > > > > > > > > > good to add that for consistency.
> >> > > > > > > > >
> >> > > > > > > > > Noted.
> >> > > > > > > > >
> >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> requires
> >> > > "Read
> >> > > > > Group"
> >> > > > > > > > > even
> >> > > > > > > > > > though it seems to do more than reading.
> >> > > > > > > > >
> >> > > > > > > > > I agree. This is how the current protocol works though.
> We
> >> > only
> >> > > > > > > > > require "Read Group" to join a group. We could consider
> >> > > changing
> >> > > > > this
> >> > > > > > > > > but I am not sure that it is worth it.
> >> > > > > > > > >
> >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> with
> >> > the
> >> > > new
> >> > > > > > > group
> >> > > > > > > > > > protocol? It would be good to have a section on this.
> >> > > > > > > > >
> >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> >> have a
> >> > > new
> >> > > > > > > > > topic id so it will treat it like a topic with a
> different
> >> > > name.
> >> > > > > The
> >> > > > > > > > > only issue is that the fetch/commit offsets APIs do not
> >> > support
> >> > > > > topic
> >> > > > > > > > > IDs so the consumer would reuse the offsets based on the
> >> > same.
> >> > > I
> >> > > > > think
> >> > > > > > > > > that we should update those APIs as well in order to be
> >> > > consistent
> >> > > > > end
> >> > > > > > > > > to end. That would strengthen the semantics of the
> >> consumer.
> >> > > > > > > > >
> >> > > > > > > > > > 14. The KIP mentions we will write the new coordinator
> >> in
> >> > > Java.
> >> > > > > Even
> >> > > > > > > > > though
> >> > > > > > > > > > this is an implementation detail, do we plan to have a
> >> new
> >> > > gradle
> >> > > > > > > module
> >> > > > > > > > > > for it?
> >> > > > > > > > >
> >> > > > > > > > > Yes.
> >> > > > > > > > >
> >> > > > > > > > > > 15. Do we have a scalability goal when it comes to how
> >> many
> >> > > > > members
> >> > > > > > > the
> >> > > > > > > > > new
> >> > > > > > > > > > group protocol can support?
> >> > > > > > > > >
> >> > > > > > > > > We don't have numbers at the moment. The protocol should
> >> > > support
> >> > > > > 1000s
> >> > > > > > > > > of members per group. We will measure this when we have
> a
> >> > first
> >> > > > > > > > > implementation. Note that we might have other
> bottlenecks
> >> > down
> >> > > the
> >> > > > > > > > > road (e.g. offset commits).
> >> > > > > > > > >
> >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds instead
> of
> >> > > > > > > > > > SubscribedTopicNames in ConsumerGroupHeartbeatRequest?
> >> Is
> >> > the
> >> > > > > idea
> >> > > > > > > that
> >> > > > > > > > > > since we have to resolve the regex on the server, we
> >> can do
> >> > > the
> >> > > > > same
> >> > > > > > > for
> >> > > > > > > > > > the topic name? The difference is that sending the
> >> regex is
> >> > > more
> >> > > > > > > > > efficient
> >> > > > > > > > > > whereas sending the topic names is less efficient.
> >> > > Furthermore,
> >> > > > > > > delete
> >> > > > > > > > > and
> >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> >> > > > > > > > >
> >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> >> server
> >> > > for
> >> > > > > both
> >> > > > > > > > > paths but I do agree with your point. As a second
> though,
> >> > using
> >> > > > > topic
> >> > > > > > > > > ids may be better here for the delete and recreation
> case.
> >> > > Also, I
> >> > > > > > > > > suppose that we may allow users to subscribe with topic
> >> ids
> >> > in
> >> > > the
> >> > > > > > > > > future because that is the only way to be really robust
> to
> >> > > topic
> >> > > > > > > > > re-creation.
> >> > > > > > > > >
> >> > > > > > > > > Best,
> >> > > > > > > > > David
> >> > > > > > > > >
> >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> >> > > djacot@confluent.io>
> >> > > > > > > wrote:
> >> > > > > > > > > >
> >> > > > > > > > > > Hi Justine,
> >> > > > > > > > > >
> >> > > > > > > > > > Thanks for your comments. Please find my answers
> below.
> >> > > > > > > > > >
> >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> >> > > exception
> >> > > > > of the
> >> > > > > > > > > > topic names based in the
> ConsumerGroupHeartbeatRequest.
> >> I
> >> > am
> >> > > not
> >> > > > > sure
> >> > > > > > > > > > if using topic names is the right call here. I need to
> >> > think
> >> > > > > about it
> >> > > > > > > > > > a little more. Obviously, the KIP does not change the
> >> > > > > fetch/commit
> >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> >> that
> >> > we
> >> > > > > should
> >> > > > > > > > > > include though as it would give better overall
> >> guarantee in
> >> > > the
> >> > > > > > > > > > producer.
> >> > > > > > > > > > - You're right. I think that I should not have
> mentioned
> >> > this
> >> > > > > flag at
> >> > > > > > > > > > all. I will remove it. We can use an internal
> >> configuration
> >> > > while
> >> > > > > > > > > > developing the feature.
> >> > > > > > > > > > - Both cluster types will be supported. The change is
> >> > > > > orthogonal. The
> >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> >> > > > > > > > > >
> >> > > > > > > > > > Best,
> >> > > > > > > > > > David
> >> > > > > > > > > >
> >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> >> > > > > wangguoz@gmail.com>
> >> > > > > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > Hi Ismael,
> >> > > > > > > > > > >
> >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> inlined
> >> > > below:
> >> > > > > > > > > > >
> >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> >> > > ismael@juma.me.uk>
> >> > > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > > Thanks for the KIP. This has the potential to be a
> >> > great
> >> > > > > > > > > improvement. A few
> >> > > > > > > > > > > > initial questions/comments:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > 1. I think it's premature to talk about target
> >> versions
> >> > > for
> >> > > > > > > > > deprecation and
> >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> >> KRaft,
> >> > > this
> >> > > > > > > affects a
> >> > > > > > > > > core
> >> > > > > > > > > > > > client protocol and hence deprecation/removal will
> >> be
> >> > > heavily
> >> > > > > > > > > dependent on
> >> > > > > > > > > > > > how quickly applications migrate to the new
> >> protocol.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> >> proposed
> >> > > > > timeline
> >> > > > > > > in
> >> > > > > > > > > the
> >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan` and
> >> > > instead
> >> > > > > just
> >> > > > > > > state
> >> > > > > > > > > > > that we will decide in the future about when we
> would
> >> > > > > deprecate old
> >> > > > > > > > > > > protocol and behaviors.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> >> but
> >> > it
> >> > > > > wasn't
> >> > > > > > > made
> >> > > > > > > > > clear
> >> > > > > > > > > > > > why. If we added that as a way to estimate when
> we'd
> >> > > > > deprecate
> >> > > > > > > and
> >> > > > > > > > > remove
> >> > > > > > > > > > > > the group protocol, I also suggest removing this
> >> part.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > I think that's not specifically related to the
> >> > > > > deprecation/removal
> >> > > > > > > > > timeline
> >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> >> > > broker-side
> >> > > > > > > > > > > implementation may be done first, and then the
> client
> >> > side,
> >> > > > > and we
> >> > > > > > > > > would
> >> > > > > > > > > > > only mark it as "released" by the time clients
> >> > > implementations
> >> > > > > are
> >> > > > > > > > > done. At
> >> > > > > > > > > > > that time, to enable the feature the clients need to
> >> > first
> >> > > > > swap-in
> >> > > > > > > the
> >> > > > > > > > > > > bytecode with a rolling bounce and then set the flag
> >> > with a
> >> > > > > second
> >> > > > > > > > > rolling
> >> > > > > > > > > > > bounce, and hence we feel it's better to be released
> >> in a
> >> > > major
> >> > > > > > > > > version.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 3. We need to flesh out the details of the
> migration
> >> > > story.
> >> > > > > It
> >> > > > > > > > > sounds like
> >> > > > > > > > > > > > we're saying we will support online migrations. Is
> >> that
> >> > > > > correct?
> >> > > > > > > We
> >> > > > > > > > > should
> >> > > > > > > > > > > > explain this in detail. It could also be done as a
> >> > > separate
> >> > > > > KIP,
> >> > > > > > > if
> >> > > > > > > > > it's
> >> > > > > > > > > > > > easier.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yes I think that's the part we can be more concrete
> >> about
> >> > > for
> >> > > > > sure
> >> > > > > > > (and
> >> > > > > > > > > > > this is related to your question 2) above). We will
> >> work
> >> > on
> >> > > > > making
> >> > > > > > > it
> >> > > > > > > > > more
> >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> >> > > subscriptions
> >> > > > > to
> >> > > > > > > the
> >> > > > > > > > > server,
> >> > > > > > > > > > > > but it seems like there could be some tricky
> >> > > compatibility
> >> > > > > > > issues.
> >> > > > > > > > > Will we
> >> > > > > > > > > > > > have a mechanism for users to detect that they
> need
> >> to
> >> > > update
> >> > > > > > > their
> >> > > > > > > > > regex
> >> > > > > > > > > > > > before switching to the new protocol?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yes I think we need some tooling for non-java client
> >> > users
> >> > > to
> >> > > > > sort
> >> > > > > > > of
> >> > > > > > > > > > > "dry-run" the client before switching to the new
> >> > protocol.
> >> > > I
> >> > > > > do not
> >> > > > > > > > > have a
> >> > > > > > > > > > > specific idea on top of my head though, maybe others
> >> like
> >> > > @Matt
> >> > > > > > > > > Howlett can
> >> > > > > > > > > > > chime-in here?
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 5. Related to the last question, will the Java
> >> client
> >> > > allow
> >> > > > > the
> >> > > > > > > > > users to
> >> > > > > > > > > > > > stick with the current regex engine for
> >> compatibility
> >> > > > > reasons?
> >> > > > > > > For
> >> > > > > > > > > example,
> >> > > > > > > > > > > > it may be handy to keep using client based regex
> at
> >> > > first to
> >> > > > > keep
> >> > > > > > > > > > > > migrations simple and then migrate to server based
> >> > > regexes
> >> > > > > as a
> >> > > > > > > > > second
> >> > > > > > > > > > > > step.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Honestly I have not thought about that for java
> >> clients,
> >> > > and
> >> > > > > we can
> >> > > > > > > > > discuss
> >> > > > > > > > > > > that. What kind of compatibility issues do you have
> in
> >> > > mind?
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 6. When we say that the group coordinator will be
> >> > > > > responsible for
> >> > > > > > > > > storing
> >> > > > > > > > > > > > the configurations and that the configurations
> will
> >> be
> >> > > > > deleted
> >> > > > > > > when
> >> > > > > > > > > the
> >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> trigger
> >> > > deletion
> >> > > > > of
> >> > > > > > > > > > > > configurations?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> would
> >> > only
> >> > > > > > > transit to
> >> > > > > > > > > that
> >> > > > > > > > > > > state when the group is EMPTY and also all of its
> >> > metadata
> >> > > are
> >> > > > > > > gone),
> >> > > > > > > > > once
> >> > > > > > > > > > > it's transited to DEAD this group would never be
> >> revived.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> group
> >> > > > > coordinator
> >> > > > > > > > > make it
> >> > > > > > > > > > > > harder to list all cluster configs and their
> values?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > That's a good question, and our thoughts are that
> the
> >> > > so-called
> >> > > > > > > "group
> >> > > > > > > > > > > configurations" are overrides of the cluster-level
> >> > > > > configurations
> >> > > > > > > > > > > customized per group so when an admin list cluster
> >> > configs
> >> > > it's
> >> > > > > > > okay to
> >> > > > > > > > > > > list just the cluster-level "defaults", not showing
> >> any
> >> > > > > per-group
> >> > > > > > > > > > > customizations.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 8. How would someone configure a group before
> >> starting
> >> > > the
> >> > > > > > > > > consumers? Have
> >> > > > > > > > > > > > we considered allowing the explicit creation of
> >> groups?
> >> > > > > > > > > Alternatively, the
> >> > > > > > > > > > > > configs could be decoupled from the group
> lifecycle.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > The configs can be created before the group itself
> as
> >> an
> >> > > > > > > independent
> >> > > > > > > > > entity
> >> > > > > > > > > > > --- of course, this requires the corresponding
> >> request to
> >> > > be
> >> > > > > > > routed to
> >> > > > > > > > > the
> >> > > > > > > > > > > right coordinator based on the group id --- the only
> >> > thing
> >> > > that
> >> > > > > > > > > differs is,
> >> > > > > > > > > > > when the group itself is gone we also check if there
> >> are
> >> > > any
> >> > > > > > > > > configuration
> >> > > > > > > > > > > entities related to that group and delete as well.
> >> > > > > > > > > > >
> >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> the
> >> > > creation
> >> > > > > /
> >> > > > > > > > > deletion
> >> > > > > > > > > > > lifecycles of the config entities, and we would like
> >> to
> >> > > hear
> >> > > > > > > everyone's
> >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> >> totally
> >> > > > > decouple
> >> > > > > > > group
> >> > > > > > > > > > > configs and hence not delete them at all when the
> >> group
> >> > is
> >> > > > > gone,
> >> > > > > > > but
> >> > > > > > > > > always
> >> > > > > > > > > > > require explicit deletion operations by themselves.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> >> > client
> >> > > > > still
> >> > > > > > > take
> >> > > > > > > > > a
> >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> >> introduce an
> >> > > > > overload?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > I think we do not need to introduce an overload, but
> >> I'm
> >> > > all
> >> > > > > ears
> >> > > > > > > if
> >> > > > > > > > > there
> >> > > > > > > > > > > may be some compatibility issues that we may
> overlook.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 10. I agree with Justine that we should be clearer
> >> > about
> >> > > the
> >> > > > > > > reason
> >> > > > > > > > > to
> >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> >> flag.
> >> > > Maybe
> >> > > > > we
> >> > > > > > > mean
> >> > > > > > > > > that we
> >> > > > > > > > > > > > can switch the default for the feature flag to
> true
> >> > > based on
> >> > > > > the
> >> > > > > > > > > > > > metadata.version once we want to make it the
> >> default.
> >> > > > > > > > > > >
> >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> >> required
> >> > > ACLs,
> >> > > > > it
> >> > > > > > > > > would be
> >> > > > > > > > > > > > good to add that for consistency.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Ack, we can certainly do that.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> >> > requires
> >> > > > > "Read
> >> > > > > > > > > Group" even
> >> > > > > > > > > > > > though it seems to do more than reading.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > I had that thought myself as well, but in the end we
> >> > could
> >> > > not
> >> > > > > > > find a
> >> > > > > > > > > > > better alternative: adding Write Group seems an
> >> overkill
> >> > > here
> >> > > > > > > since we
> >> > > > > > > > > do
> >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> and
> >> > > Describe
> >> > > > > on
> >> > > > > > > > > groups so
> >> > > > > > > > > > > far). Would like to hear others thoughts.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 13. How is topic recreation handled by the
> consumer
> >> > with
> >> > > the
> >> > > > > new
> >> > > > > > > > > group
> >> > > > > > > > > > > > protocol? It would be good to have a section on
> >> this.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> >> add a
> >> > > > > section
> >> > > > > > > about
> >> > > > > > > > > > > that, but basically the idea is that consumer would
> be
> >> > > totally
> >> > > > > > > > > agnostic in
> >> > > > > > > > > > > the new protocol as it's handled all by the brokers.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> >> coordinator
> >> > in
> >> > > > > Java.
> >> > > > > > > Even
> >> > > > > > > > > though
> >> > > > > > > > > > > > this is an implementation detail, do we plan to
> >> have a
> >> > > new
> >> > > > > gradle
> >> > > > > > > > > module
> >> > > > > > > > > > > > for it?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > We have not thought about that. But I think the
> answer
> >> > > should
> >> > > > > be
> >> > > > > > > yes.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes to
> >> how
> >> > > many
> >> > > > > > > members
> >> > > > > > > > > the new
> >> > > > > > > > > > > > group protocol can support?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Within a group, I think we should shoot for 1000s of
> >> > > members.
> >> > > > > But
> >> > > > > > > that
> >> > > > > > > > > > > scalability goals also depend on the offset
> management
> >> > > (commit,
> >> > > > > > > fetch)
> >> > > > > > > > > > > capabilities of the coordinator which we did not
> >> cover in
> >> > > this
> >> > > > > > > KIP, so
> >> > > > > > > > > it's
> >> > > > > > > > > > > hard to give a number that applies universally.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> >> instead
> >> > of
> >> > > > > > > > > > > > SubscribedTopicNames in
> >> ConsumerGroupHeartbeatRequest?
> >> > > Is the
> >> > > > > > > idea
> >> > > > > > > > > that
> >> > > > > > > > > > > > since we have to resolve the regex on the server,
> we
> >> > can
> >> > > do
> >> > > > > the
> >> > > > > > > same
> >> > > > > > > > > for
> >> > > > > > > > > > > > the topic name? The difference is that sending the
> >> > regex
> >> > > is
> >> > > > > more
> >> > > > > > > > > efficient
> >> > > > > > > > > > > > whereas sending the topic names is less efficient.
> >> > > > > Furthermore,
> >> > > > > > > > > delete and
> >> > > > > > > > > > > > recreation is easier to handle if we have topic
> ids.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > The main reason to still let the clients send names
> >> is to
> >> > > keep
> >> > > > > the
> >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> client
> >> > > only.
> >> > > > > Note
> >> > > > > > > that
> >> > > > > > > > > > > although we added topic id in KIP-516, we never
> >> > > implemented the
> >> > > > > > > logic
> >> > > > > > > > > on
> >> > > > > > > > > > > consumer/producers leveraging the related newer
> >> versioned
> >> > > RPCs,
> >> > > > > > > > > instead we
> >> > > > > > > > > > > just set the topic id as empty UUID. We want to keep
> >> the
> >> > > > > > > > > consumer/producer
> >> > > > > > > > > > > to be thin and only delegate the reasoning on broker
> >> and
> >> > > > > > > potentially
> >> > > > > > > > > admin
> >> > > > > > > > > > > clients.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Thanks,
> >> > > > > > > > > > > > Ismael
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> >> > > > > > > > > <dj...@confluent.io.invalid>
> >> > > > > > > > > > > > wrote:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > > Hi all,
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > I would like to start a discussion thread on
> >> KIP-848:
> >> > > The
> >> > > > > Next
> >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> >> With
> >> > > this
> >> > > > > KIP,
> >> > > > > > > we
> >> > > > > > > > > aim
> >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> >> more
> >> > > > > reliable,
> >> > > > > > > more
> >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> >> easier
> >> > > to
> >> > > > > debug
> >> > > > > > > for
> >> > > > > > > > > > > > > operators.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > The KIP is here:
> >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Please take a look and let me know what you
> think.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Best,
> >> > > > > > > > > > > > > David
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > PS: I will be away from July 18th to August 8th.
> >> That
> >> > > gives
> >> > > > > > > you a
> >> > > > > > > > > bit
> >> > > > > > > > > > > > > of time to read and digest this long KIP.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > --
> >> > > > > > > > > > > -- Guozhang
> >> > > > > > > > >
> >> > > > > > > >
> >> > > > > > > >
> >> > > > > > > > --
> >> > > > > > > > -- Guozhang
> >> > > > > > >
> >> > > > >
> >> > >
> >> > >
> >> >
> >>
> >>
> >> --
> >> -- Guozhang
> >>
> >
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi all,

Thanks again for all the very good feedback so far. I have addressed,
I think, all the outstanding points. Here are the main updates:

* I have introduced STALE_MEMBER_EPOCH to differentiate it from the
fatal FENCED_MEMBER_EPOCH error.

* I have introduced topic ids in the OffsetCommit/OffsetFetch APIs.
This allows the Consumer to use Topic IDs everywhere and thus gives it
stronger guarantees. I have also added a few words about topic
recreation in this context.

* After discussing with Jason and Guozhang offline, I have decided to
follow Ismael's advice to store dynamic group configuration in the
controller. This is simpler with the IncrementalAlterConfigs API and
decouple the life cycle of the configs from the life cycle of the
groups.

* I have introduced a new Pattern class to model the new regular
expressions syntax (Google RE2/J) and differentiate them from the
java.util.regex.Pattern ones. Simple regular expressions should work
in both cases but more advances won't so it seems preferable, from a
user point of view, to be clear and to differentiate them.

Please take another look and let me know what you think.

Best,
David

On Wed, Aug 10, 2022 at 2:36 AM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hello Greg,
>
> Thanks for reviewing the KIP! I hope David has addressed your questions
> above, and I'd like to just emphasize one thing regarding your question 1):
> one principle of the reconciliation is that, for a single client, before it
> has completely revoked all the partitions that it should revoke, the
> coordinator would not assign new partitions to it even if those partitions
> have already been revoked by the old host and hence are available. And when
> it has revoked all its partitions, the coordinator would start giving it
> new partitions while at the same time bumping its epoch. Of course in
> practice, a client should try to always revoke all its should-be-revoked
> partitions together in a single shot which is more efficient, but from the
> point view of the coordinator it's always "first revoke all you should
> revoke, and then I will bump your epoch up and start giving you more
> partitions". Though this may not be most efficient since coordinators maybe
> holding on some available-partitions-to-give to the client before the
> client has complete revoking, it gives us a good barrier (hence the epoch
> bumping) to determine if the client can continue getting new partitions
> even after soft failures or network partitions.
>
>
> Guozhang
>
> On Wed, Aug 3, 2022 at 1:05 PM Gregory Harris <gh...@gmail.com> wrote:
>
> > Hey All!
> >
> > Thanks for the KIP, it's wonderful to see cooperative rebalancing making it
> > down the stack!
> >
> > I had a few questions:
> >
> > 1. The 'Rejected Alternatives' section describes how member epoch should
> > advance in step with the group epoch and assignment epoch values. I think
> > that this is a good idea for the reasons described in the KIP. When the
> > protocol is incrementally assigning partitions to a worker, what member
> > epoch does each incremental assignment use? Are member epochs re-used, and
> > a single member epoch can correspond to multiple different (monotonically
> > larger) assignments?
> >
> > 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> > not, should custom client-side assignor implementations interact with the
> > Reason field, and how is its common meaning agreed upon? If so, what is the
> > benefit of a distinct Reason field over including such functionality in the
> > opaque metadata?
> >
> > 3. The following is included in the KIP: "Thanks to this, the input of the
> > client side assignor is entirely driven by the group coordinator. The
> > consumer is no longer responsible for maintaining any state besides its
> > assigned partitions." Does this mean that the client-side assignor MAY
> > incorporate additional non-Metadata state (such as partition throughput,
> > cpu/memory metrics, config topics, etc), or that additional non-Metadata
> > state SHOULD NOT be used?
> >
> > 4. I see that there are separate classes
> > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> > overlap significantly. Is it possible for these two implementations to be
> > unified? This would serve to promote feature parity of server-side and
> > client-side assignors, and would also facilitate operational flexibility in
> > certain situations. For example, if a server-side assignor has some poor
> > behavior and needs a patch, deploying the patched assignor to the client
> > and switching one consumer group to a client-side assignor may be faster
> > and less risky than patching all of the brokers. With the currently
> > proposed distinct APIs, a non-trivial reimplementation would have to be
> > assembled, and if the two APIs have diverged significantly, then it is
> > possible that a reimplementation would not be possible.
> >
> > --
> > Greg Harris
> > gharris1727@gmail.com
> > github.com/gharris1727
> >
> > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com> wrote:
> >
> > > Hi Guozhang/David,
> > >
> > > I created a confluence page to discuss how Connect would need to change
> > > based on the new rebalance protocol. Here's the page:
> > >
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >
> > > It's also pretty longish and I have tried to keep a format similar to
> > > KIP-848. Let me know what you think. Also, do you think this should be
> > > moved to a separate discussion thread or is this one fine?
> > >
> > > Thanks!
> > > Sagar.
> > >
> > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
> > >
> > > > Hello Guozhang,
> > > >
> > > > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > > > analysis and come up with such a document.
> > > >
> > > > Thanks!
> > > > Sagar.
> > > >
> > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > >
> > > >> Hello Sagar,
> > > >>
> > > >> It would be great if you could come back with some analysis on how to
> > > >> implement the Connect side integration with the new protocol; so far
> > > >> besides leveraging on the new "protocol type" we did not yet think
> > > through
> > > >> the Connect side implementations. For Streams here's a draft of
> > > >> integration
> > > >> plan:
> > > >>
> > > >>
> > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > >> just FYI for your analysis on Connect.
> > > >>
> > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> > > wrote:
> > > >>
> > > >> > Hi David,
> > > >> >
> > > >> > Thank you for your response. The reason I thought connect can also
> > fit
> > > >> into
> > > >> > this new scheme is that even today the connect uses a
> > > WorkerCoordinator
> > > >> > extending from AbstractCoordinator to empower rebalances of
> > > >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> > > >> connect
> > > >> > and uses the metadata() method by plumbing into
> > > >> JoinGroupRequestProtocol.
> > > >> >
> > > >> > I think the changes to support connect would be similar at a high
> > > level
> > > >> to
> > > >> > the changes in streams mainly because of the Client side assignors
> > > being
> > > >> > used in both. At an implementation level, we might need to make a
> > lot
> > > of
> > > >> > changes to get onto this new assignment protocol like enhancing the
> > > >> > JoinGroup request/response and SyncGroup and using
> > > >> ConsumerGroupHeartbeat
> > > >> > API etc again on similar lines to streams (or there might be
> > > >> deviations). I
> > > >> > would try to perform a detailed analysis of the same  and we can
> > have
> > > a
> > > >> > separate discussion thread for that as that would derail this
> > > discussion
> > > >> > thread. Let me know if that sounds good to you.
> > > >> >
> > > >> > Thanks!
> > > >> > Sagar.
> > > >> >
> > > >> >
> > > >> >
> > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > <djacot@confluent.io.invalid
> > > >> >
> > > >> > wrote:
> > > >> >
> > > >> > > Hi Sagar,
> > > >> > >
> > > >> > > Thanks for your comments.
> > > >> > >
> > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> > > >> > >
> > > >> > > 2) The idea is to transition C from his current assignment to his
> > > >> > > target assignment when he can move to epoch 3. When that happens,
> > > the
> > > >> > > member assignment is updated and persisted with all its assigned
> > > >> > > partitions even if they are not all revoked yet. In other words,
> > the
> > > >> > > member assignment becomes the target assignment. This is basically
> > > an
> > > >> > > optimization to avoid having to write all the changes to the log.
> > > The
> > > >> > > examples are based on the persisted state so I understand the
> > > >> > > confusion. Let me see if I can improve this in the description.
> > > >> > >
> > > >> > > 3) Regarding Connect, it could reuse the protocol with a client
> > side
> > > >> > > assignor if it fits in the protocol. The assignment is about
> > > >> > > topicid-partitions + metadata, could Connect fit into this?
> > > >> > >
> > > >> > > Best,
> > > >> > > David
> > > >> > >
> > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> > > >> wrote:
> > > >> > > >
> > > >> > > > Hi David,
> > > >> > > >
> > > >> > > > Thanks for the KIP. I just had minor observations:
> > > >> > > >
> > > >> > > > 1) In the Assignment Error section in Client Side mode
> > Assignment
> > > >> > > process,
> > > >> > > > you mentioned => `In this case, the client side assignor can
> > > return
> > > >> an
> > > >> > > > error to the group coordinator`. In this case are you referring
> > to
> > > >> the
> > > >> > > > Assignor returning an AssignmentError that's listed down towards
> > > the
> > > >> > end?
> > > >> > > > If yes, do you think it would make sense to mention this
> > > explicitly
> > > >> > here?
> > > >> > > >
> > > >> > > > 2) In the Case Studies section, I have a slight confusion, not
> > > sure
> > > >> if
> > > >> > > > others have the same. Consider this step:
> > > >> > > >
> > > >> > > > When B heartbeats, the group coordinator transitions him to
> > epoch
> > > 3
> > > >> > > because
> > > >> > > > B has no partitions to revoke. It persists the change and reply.
> > > >> > > >
> > > >> > > >    - Group (epoch=3)
> > > >> > > >       - A
> > > >> > > >       - B
> > > >> > > >       - C
> > > >> > > >    - Target Assignment (epoch=3)
> > > >> > > >       - A - partitions=[foo-0]
> > > >> > > >       - B - partitions=[foo-2]
> > > >> > > >       - C - partitions=[foo-1]
> > > >> > > >    - Member Assignment
> > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > >> > > >
> > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot get
> > foo-1
> > > >> yet.
> > > >> > > >
> > > >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> > > yet,
> > > >> > but
> > > >> > > > based on the description above, it seems it already has it. Do
> > you
> > > >> > think
> > > >> > > it
> > > >> > > > would be better to remove it and populate it only when it
> > actually
> > > >> gets
> > > >> > > it?
> > > >> > > > I see this in a lot of other places, so have I understood it
> > > >> > incorrectly
> > > >> > > ?
> > > >> > > >
> > > >> > > >
> > > >> > > > Regarding connect , it might be out of scope of this discussion,
> > > but
> > > >> > from
> > > >> > > > what I understood it would probably be running in client side
> > > >> assignor
> > > >> > > mode
> > > >> > > > even on the new rebalance protocol as it has its own Custom
> > > >> > > Assignors(Eager
> > > >> > > > and IncrementalCooperative).
> > > >> > > >
> > > >> > > > Thanks!
> > > >> > > >
> > > >> > > > Sagar.
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > >> > <djacot@confluent.io.invalid
> > > >> > > >
> > > >> > > > wrote:
> > > >> > > >
> > > >> > > > > Thanks Hector! Our goal is to move forward with specialized
> > API
> > > >> > > > > instead of relying on one generic API. For Connect, we can
> > apply
> > > >> the
> > > >> > > > > exact same pattern and reuse/share the core implementation on
> > > the
> > > >> > > > > server side. For the schema registry, I think that we should
> > > >> consider
> > > >> > > > > having a tailored API to do simple membership/leader election.
> > > >> > > > >
> > > >> > > > > Best,
> > > >> > > > > David
> > > >> > > > >
> > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> > ismael@juma.me.uk
> > > >
> > > >> > > wrote:
> > > >> > > > > >
> > > >> > > > > > Three quick comments:
> > > >> > > > > >
> > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > >> com.google.re2j.Pattern, we
> > > >> > > > > should
> > > >> > > > > > document the differences in more detail before deciding one
> > > way
> > > >> or
> > > >> > > > > another.
> > > >> > > > > > That said, if people pass java.util.regex.Pattern, they
> > expect
> > > >> > their
> > > >> > > > > > semantics to be honored. If we are doing something
> > different,
> > > >> then
> > > >> > we
> > > >> > > > > > should consider adding an overload with our own Pattern
> > class
> > > (I
> > > >> > > don't
> > > >> > > > > > think we'd want to expose re2j's at this point).
> > > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > > integrate
> > > >> > fully
> > > >> > > > > with
> > > >> > > > > > it and should handle the topic recreation case correctly.
> > > That's
> > > >> > the
> > > >> > > main
> > > >> > > > > > part we need to handle. I agree with David that we'd want to
> > > add
> > > >> > > topic
> > > >> > > > > ids
> > > >> > > > > > to the relevant protocols that don't have it yet and that we
> > > can
> > > >> > > probably
> > > >> > > > > > focus on the internals versus adding new APIs to the Java
> > > >> Consumer
> > > >> > > > > (unless
> > > >> > > > > > we find that adding new APIs is required for reasonable
> > > >> semantics).
> > > >> > > > > > 3. I am still not sure about the coordinator storing the
> > > >> configs.
> > > >> > > It's
> > > >> > > > > > powerful for configs to be centralized in the metadata log
> > for
> > > >> > > various
> > > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > > >> Similarly, I
> > > >> > > am
> > > >> > > > > not
> > > >> > > > > > sure about automatically deleting configs in a way that they
> > > >> cannot
> > > >> > > be
> > > >> > > > > > recovered. A good property for modern systems is to minimize
> > > the
> > > >> > > number
> > > >> > > > > of
> > > >> > > > > > unrecoverable data loss scenarios.
> > > >> > > > > >
> > > >> > > > > > Ismael
> > > >> > > > > >
> > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > >> > > <djacot@confluent.io.invalid
> > > >> > > > > >
> > > >> > > > > > wrote:
> > > >> > > > > >
> > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > >> > > > > > >
> > > >> > > > > > > > 1) the migration path, especially the last step when
> > > clients
> > > >> > > flip the
> > > >> > > > > > > flag
> > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > window
> > > >> > where
> > > >> > > > > both
> > > >> > > > > > > new
> > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > > >> members
> > > >> > of
> > > >> > > the
> > > >> > > > > same
> > > >> > > > > > > > group. How the coordinator could "mimic" the old
> > behavior
> > > >> while
> > > >> > > > > using the
> > > >> > > > > > > > new protocol is something we need to present about.
> > > >> > > > > > >
> > > >> > > > > > > Noted. I just published a new version of KIP which
> > includes
> > > >> more
> > > >> > > > > > > details about this. See the "Supporting Online Consumer
> > > Group
> > > >> > > Upgrade"
> > > >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan".
> > I
> > > >> think
> > > >> > > that
> > > >> > > > > > > I have to think through a few cases now but the overall
> > idea
> > > >> and
> > > >> > > > > > > mechanism should be understandable.
> > > >> > > > > > >
> > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic
> > ids
> > > >> are
> > > >> > > only
> > > >> > > > > used
> > > >> > > > > > > as
> > > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > > via
> > > >> any
> > > >> > > > > public
> > > >> > > > > > > APIs
> > > >> > > > > > > > to consumers yet. I think the question is, first should
> > we
> > > >> let
> > > >> > > the
> > > >> > > > > > > consumer
> > > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > > to
> > > >> > fully
> > > >> > > > > > > leverage
> > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > > the
> > > >> > > topic
> > > >> > > > > ids;
> > > >> > > > > > > and
> > > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > > in
> > > >> the
> > > >> > > > > consumer
> > > >> > > > > > > > public APIs as well (both subscribe/assign, as well as
> > in
> > > >> the
> > > >> > > > > rebalance
> > > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > > >> > > > > > >
> > > >> > > > > > > a) Assuming that we would include converting all the
> > offsets
> > > >> > > related
> > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> > > >> able
> > > >> > to
> > > >> > > > > > > fully operate with topic ids. That being said, it still
> > has
> > > to
> > > >> > > provide
> > > >> > > > > > > the topics names in various APIs so having a mapping in
> > the
> > > >> > > consumer
> > > >> > > > > > > seems inevitable to me.
> > > >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> > > >> this
> > > >> > > goes
> > > >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> > > >> > internals
> > > >> > > > > > > here and we can consider this separately if we see value
> > in
> > > >> doing
> > > >> > > it.
> > > >> > > > > > >
> > > >> > > > > > > Coming back to Ismael's point about using topic ids in the
> > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> > > >> > advantage
> > > >> > > in
> > > >> > > > > > > favour of it. The consumer will have the opportunity to
> > > >> validate
> > > >> > > that
> > > >> > > > > > > the topics exists before passing them into the group
> > > rebalance
> > > >> > > > > > > protocol. Obviously, the coordinator will also notice it
> > but
> > > >> it
> > > >> > > does
> > > >> > > > > > > not really have a way to reject an invalid topic in the
> > > >> response.
> > > >> > > > > > >
> > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > except
> > > >> for
> > > >> > > the
> > > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's
> > not
> > > >> > > necessary
> > > >> > > > > to
> > > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > > >> still
> > > >> > > use
> > > >> > > > > > > Pattern
> > > >> > > > > > > > while just documenting that our subscription may be
> > > >> rejected by
> > > >> > > the
> > > >> > > > > > > server.
> > > >> > > > > > > > Since the incompatible case is a very rare scenario I
> > felt
> > > >> > using
> > > >> > > an
> > > >> > > > > > > > overloaded `String` based subscription may be more
> > > >> vulnerable
> > > >> > to
> > > >> > > > > various
> > > >> > > > > > > > invalid regexes.
> > > >> > > > > > >
> > > >> > > > > > > That could work. I have to look at the differences between
> > > the
> > > >> > two
> > > >> > > > > > > engines to better understand the potential issues. My
> > > >> > > understanding is
> > > >> > > > > > > that would work for all the basic regular expressions. The
> > > >> > > differences
> > > >> > > > > > > between the two are mainly about the various character
> > > >> classes. I
> > > >> > > > > > > wonder what other people think about this.
> > > >> > > > > > >
> > > >> > > > > > > Best,
> > > >> > > > > > > David
> > > >> > > > > > >
> > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > > >> > wangguoz@gmail.com
> > > >> > > >
> > > >> > > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > Thanks David! I think on the high level there are two
> > meta
> > > >> > > points we
> > > >> > > > > need
> > > >> > > > > > > > to concretize a bit more:
> > > >> > > > > > > >
> > > >> > > > > > > > 1) the migration path, especially the last step when
> > > clients
> > > >> > > flip the
> > > >> > > > > > > flag
> > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > window
> > > >> > where
> > > >> > > > > both
> > > >> > > > > > > new
> > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > > >> members
> > > >> > of
> > > >> > > the
> > > >> > > > > same
> > > >> > > > > > > > group. How the coordinator could "mimic" the old
> > behavior
> > > >> while
> > > >> > > > > using the
> > > >> > > > > > > > new protocol is something we need to present about.
> > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic
> > ids
> > > >> are
> > > >> > > only
> > > >> > > > > used
> > > >> > > > > > > as
> > > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > > via
> > > >> any
> > > >> > > > > public
> > > >> > > > > > > APIs
> > > >> > > > > > > > to consumers yet. I think the question is, first should
> > we
> > > >> let
> > > >> > > the
> > > >> > > > > > > consumer
> > > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > > to
> > > >> > fully
> > > >> > > > > > > leverage
> > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > > the
> > > >> > > topic
> > > >> > > > > ids;
> > > >> > > > > > > and
> > > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > > in
> > > >> the
> > > >> > > > > consumer
> > > >> > > > > > > > public APIs as well (both subscribe/assign, as well as
> > in
> > > >> the
> > > >> > > > > rebalance
> > > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > > >> > > > > > > >
> > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > except
> > > >> for
> > > >> > > the
> > > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's
> > not
> > > >> > > necessary
> > > >> > > > > to
> > > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > > >> still
> > > >> > > use
> > > >> > > > > > > Pattern
> > > >> > > > > > > > while just documenting that our subscription may be
> > > >> rejected by
> > > >> > > the
> > > >> > > > > > > server.
> > > >> > > > > > > > Since the incompatible case is a very rare scenario I
> > felt
> > > >> > using
> > > >> > > an
> > > >> > > > > > > > overloaded `String` based subscription may be more
> > > >> vulnerable
> > > >> > to
> > > >> > > > > various
> > > >> > > > > > > > invalid regexes.
> > > >> > > > > > > >
> > > >> > > > > > > >
> > > >> > > > > > > > Guozhang
> > > >> > > > > > > >
> > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > >> > > > > <djacot@confluent.io.invalid
> > > >> > > > > > > >
> > > >> > > > > > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > > Hi Ismael,
> > > >> > > > > > > > >
> > > >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> > > >> > inline.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 1. I think it's premature to talk about target
> > > versions
> > > >> for
> > > >> > > > > > > deprecation
> > > >> > > > > > > > > and
> > > >> > > > > > > > > > removal of the existing group protocol. Unlike
> > KRaft,
> > > >> this
> > > >> > > > > affects a
> > > >> > > > > > > core
> > > >> > > > > > > > > > client protocol and hence deprecation/removal will
> > be
> > > >> > heavily
> > > >> > > > > > > dependent
> > > >> > > > > > > > > on
> > > >> > > > > > > > > > how quickly applications migrate to the new
> > protocol.
> > > >> > > > > > > > >
> > > >> > > > > > > > > That makes sense. I will remove it.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> > but
> > > it
> > > >> > > wasn't
> > > >> > > > > made
> > > >> > > > > > > > > clear
> > > >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> > > >> > > deprecate
> > > >> > > > > and
> > > >> > > > > > > remove
> > > >> > > > > > > > > > the group protocol, I also suggest removing this
> > part.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> > > >> rewrite
> > > >> > > the
> > > >> > > > > group
> > > >> > > > > > > > > coordinator in Java while we implement the new
> > protocol.
> > > >> This
> > > >> > > means
> > > >> > > > > > > > > that the internals will be slightly different (e.g.
> > > >> threading
> > > >> > > > > model).
> > > >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> > > >> group
> > > >> > > > > > > > > coordinator to the new group coordinator to a major
> > > >> release.
> > > >> > > The
> > > >> > > > > > > > > alternative would be to use a flag to do the switch
> > > >> instead
> > > >> > of
> > > >> > > > > relying
> > > >> > > > > > > > > on the software upgrade.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 3. We need to flesh out the details of the migration
> > > >> story.
> > > >> > > It
> > > >> > > > > sounds
> > > >> > > > > > > > > like
> > > >> > > > > > > > > > we're saying we will support online migrations. Is
> > > that
> > > >> > > correct?
> > > >> > > > > We
> > > >> > > > > > > > > should
> > > >> > > > > > > > > > explain this in detail. It could also be done as a
> > > >> separate
> > > >> > > KIP,
> > > >> > > > > if
> > > >> > > > > > > it's
> > > >> > > > > > > > > > easier.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Yes, we will support online migrations for the group.
> > > That
> > > >> > > means
> > > >> > > > > that
> > > >> > > > > > > > > a group using the old protocol will be able to switch
> > to
> > > >> the
> > > >> > > new
> > > >> > > > > > > > > protocol.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Let me briefly explain how that will work though. It
> > is
> > > >> > > basically a
> > > >> > > > > > > > > four step process:
> > > >> > > > > > > > >
> > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a
> > software
> > > >> > > supporting
> > > >> > > > > the
> > > >> > > > > > > > > new group coordinator. Both the old and the new
> > > >> coordinator
> > > >> > > will
> > > >> > > > > > > > > support the old protocol and rely on the same
> > persisted
> > > >> > > metadata so
> > > >> > > > > > > > > they can work together. This point is an offline
> > > >> migration.
> > > >> > We
> > > >> > > > > cannot
> > > >> > > > > > > > > do this one live because it would require shutting
> > down
> > > >> the
> > > >> > > current
> > > >> > > > > > > > > coordinator and starting up the new one and that would
> > > >> cause
> > > >> > > > > > > > > unavailabilities.
> > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> > > to
> > > >> X
> > > >> > in
> > > >> > > > > order
> > > >> > > > > > > > > to enable the new protocol. This cannot be done before
> > > 1)
> > > >> is
> > > >> > > > > > > > > terminated because the old coordinator doesn't support
> > > the
> > > >> > new
> > > >> > > > > > > > > protocol.
> > > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > > supporting
> > > >> the
> > > >> > > > > online
> > > >> > > > > > > > > migration (must have KIP-792). If the consumer is
> > > already
> > > >> > > there.
> > > >> > > > > > > > > Nothing must be done at this point.
> > > >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> > > >> turned
> > > >> > > on.
> > > >> > > > > The
> > > >> > > > > > > > > consumer group is automatically converted when the
> > first
> > > >> > > consumer
> > > >> > > > > > > > > using the new protocol joins the group. While the
> > > members
> > > >> > > using the
> > > >> > > > > > > > > old protocol are being upgraded, the old protocol is
> > > >> proxied
> > > >> > > into
> > > >> > > > > the
> > > >> > > > > > > > > new one.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > >> subscriptions
> > > >> > > to
> > > >> > > > > the
> > > >> > > > > > > > > server,
> > > >> > > > > > > > > > but it seems like there could be some tricky
> > > >> compatibility
> > > >> > > > > issues.
> > > >> > > > > > > Will
> > > >> > > > > > > > > we
> > > >> > > > > > > > > > have a mechanism for users to detect that they need
> > to
> > > >> > update
> > > >> > > > > their
> > > >> > > > > > > regex
> > > >> > > > > > > > > > before switching to the new protocol?
> > > >> > > > > > > > >
> > > >> > > > > > > > > I think that I am a bit more optimistic than you on
> > this
> > > >> > > point. I
> > > >> > > > > > > > > believe that the majority of the cases are simple
> > > regexes
> > > >> > which
> > > >> > > > > should
> > > >> > > > > > > > > work with the new engine. The coordinator will verify
> > > the
> > > >> > regex
> > > >> > > > > anyway
> > > >> > > > > > > > > and reject the consumer if the regex is not valid.
> > > Coming
> > > >> > back
> > > >> > > to
> > > >> > > > > the
> > > >> > > > > > > > > migration path, in the worst case, the first upgraded
> > > >> > consumer
> > > >> > > > > joining
> > > >> > > > > > > > > the group will be rejected. This should be used as the
> > > >> last
> > > >> > > > > defence, I
> > > >> > > > > > > > > would say.
> > > >> > > > > > > > >
> > > >> > > > > > > > > One way for customers to validate their regex before
> > > >> > upgrading
> > > >> > > > > their
> > > >> > > > > > > > > prod would be to test them with another group. For
> > > >> instance,
> > > >> > > that
> > > >> > > > > > > > > could be done in a pre-prod environment. Another way
> > > >> would be
> > > >> > > to
> > > >> > > > > > > > > extend the consumer-group tool to provide a regex
> > > >> validation
> > > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 5. Related to the last question, will the Java
> > client
> > > >> allow
> > > >> > > the
> > > >> > > > > > > users to
> > > >> > > > > > > > > > stick with the current regex engine for
> > compatibility
> > > >> > > reasons?
> > > >> > > > > For
> > > >> > > > > > > > > example,
> > > >> > > > > > > > > > it may be handy to keep using client based regex at
> > > >> first
> > > >> > to
> > > >> > > keep
> > > >> > > > > > > > > > migrations simple and then migrate to server based
> > > >> regexes
> > > >> > > as a
> > > >> > > > > > > second
> > > >> > > > > > > > > step.
> > > >> > > > > > > > >
> > > >> > > > > > > > > I understand your point but I am concerned that this
> > > would
> > > >> > > allow
> > > >> > > > > users
> > > >> > > > > > > > > to actually stay in this mode. That would go against
> > our
> > > >> goal
> > > >> > > of
> > > >> > > > > > > > > simplifying the client because we would have to
> > continue
> > > >> > > monitoring
> > > >> > > > > > > > > the metadata on the client side. I would rather not do
> > > >> this.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 6. When we say that the group coordinator will be
> > > >> > > responsible for
> > > >> > > > > > > storing
> > > >> > > > > > > > > > the configurations and that the configurations will
> > be
> > > >> > > deleted
> > > >> > > > > when
> > > >> > > > > > > the
> > > >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> > > >> > deletion
> > > >> > > of
> > > >> > > > > > > > > > configurations?
> > > >> > > > > > > > >
> > > >> > > > > > > > > That's right. The configurations will be deleted when
> > > the
> > > >> > > group is
> > > >> > > > > > > > > deleted. They go together.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 7. Will the choice to store the configs in the group
> > > >> > > coordinator
> > > >> > > > > > > make it
> > > >> > > > > > > > > > harder to list all cluster configs and their values?
> > > >> > > > > > > > >
> > > >> > > > > > > > > I don't think so. The group configurations are
> > overrides
> > > >> of
> > > >> > > cluster
> > > >> > > > > > > > > configs. If you want to know all the overrides though,
> > > you
> > > >> > > would
> > > >> > > > > have
> > > >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> > > the
> > > >> > > metadata
> > > >> > > > > log
> > > >> > > > > > > > > for instance.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 8. How would someone configure a group before
> > starting
> > > >> the
> > > >> > > > > consumers?
> > > >> > > > > > > > > Have
> > > >> > > > > > > > > > we considered allowing the explicit creation of
> > > groups?
> > > >> > > > > > > Alternatively,
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Yes. The group will be automatically created in this
> > > case.
> > > >> > > However,
> > > >> > > > > > > > > the configs will be lost after the retention period of
> > > the
> > > >> > > group
> > > >> > > > > > > > > passes.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > > >> client
> > > >> > > still
> > > >> > > > > take
> > > >> > > > > > > a
> > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> > > an
> > > >> > > overload?
> > > >> > > > > > > > >
> > > >> > > > > > > > > That's a very group question. I forgot about that one.
> > > As
> > > >> the
> > > >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> > > the
> > > >> > > engine
> > > >> > > > > that
> > > >> > > > > > > > > we plan to use, it might be better to deprecate it and
> > > >> use an
> > > >> > > > > overload
> > > >> > > > > > > > > which takes a string. We would rely on the server side
> > > >> > > validation.
> > > >> > > > > > > > > During the migration, I think that we could still try
> > to
> > > >> > > toString
> > > >> > > > > the
> > > >> > > > > > > > > regex and use it. That should work, I think, in the
> > > >> majority
> > > >> > > of the
> > > >> > > > > > > > > cases.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> > > about
> > > >> > the
> > > >> > > > > reason
> > > >> > > > > > > to
> > > >> > > > > > > > > > switch to IBP/metadata.version from the feature
> > flag.
> > > >> Maybe
> > > >> > > we
> > > >> > > > > mean
> > > >> > > > > > > that
> > > >> > > > > > > > > we
> > > >> > > > > > > > > > can switch the default for the feature flag to true
> > > >> based
> > > >> > on
> > > >> > > the
> > > >> > > > > > > > > > metadata.version once we want to make it the
> > default.
> > > >> > > > > > > > >
> > > >> > > > > > > > > My plan was to use that feature flag mainly during the
> > > >> > > development
> > > >> > > > > > > > > phase. I should not have mentioned it, I think,
> > because
> > > we
> > > >> > > could
> > > >> > > > > use
> > > >> > > > > > > > > an internal config for it.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > > required
> > > >> > > ACLs, it
> > > >> > > > > > > would
> > > >> > > > > > > > > be
> > > >> > > > > > > > > > good to add that for consistency.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Noted.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > > requires
> > > >> > > "Read
> > > >> > > > > Group"
> > > >> > > > > > > > > even
> > > >> > > > > > > > > > though it seems to do more than reading.
> > > >> > > > > > > > >
> > > >> > > > > > > > > I agree. This is how the current protocol works
> > though.
> > > We
> > > >> > only
> > > >> > > > > > > > > require "Read Group" to join a group. We could
> > consider
> > > >> > > changing
> > > >> > > > > this
> > > >> > > > > > > > > but I am not sure that it is worth it.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> > > with
> > > >> > the
> > > >> > > new
> > > >> > > > > > > group
> > > >> > > > > > > > > > protocol? It would be good to have a section on
> > this.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> > > >> have a
> > > >> > > new
> > > >> > > > > > > > > topic id so it will treat it like a topic with a
> > > different
> > > >> > > name.
> > > >> > > > > The
> > > >> > > > > > > > > only issue is that the fetch/commit offsets APIs do
> > not
> > > >> > support
> > > >> > > > > topic
> > > >> > > > > > > > > IDs so the consumer would reuse the offsets based on
> > the
> > > >> > same.
> > > >> > > I
> > > >> > > > > think
> > > >> > > > > > > > > that we should update those APIs as well in order to
> > be
> > > >> > > consistent
> > > >> > > > > end
> > > >> > > > > > > > > to end. That would strengthen the semantics of the
> > > >> consumer.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 14. The KIP mentions we will write the new
> > coordinator
> > > >> in
> > > >> > > Java.
> > > >> > > > > Even
> > > >> > > > > > > > > though
> > > >> > > > > > > > > > this is an implementation detail, do we plan to
> > have a
> > > >> new
> > > >> > > gradle
> > > >> > > > > > > module
> > > >> > > > > > > > > > for it?
> > > >> > > > > > > > >
> > > >> > > > > > > > > Yes.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 15. Do we have a scalability goal when it comes to
> > how
> > > >> many
> > > >> > > > > members
> > > >> > > > > > > the
> > > >> > > > > > > > > new
> > > >> > > > > > > > > > group protocol can support?
> > > >> > > > > > > > >
> > > >> > > > > > > > > We don't have numbers at the moment. The protocol
> > should
> > > >> > > support
> > > >> > > > > 1000s
> > > >> > > > > > > > > of members per group. We will measure this when we
> > have
> > > a
> > > >> > first
> > > >> > > > > > > > > implementation. Note that we might have other
> > > bottlenecks
> > > >> > down
> > > >> > > the
> > > >> > > > > > > > > road (e.g. offset commits).
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > instead
> > > of
> > > >> > > > > > > > > > SubscribedTopicNames in
> > ConsumerGroupHeartbeatRequest?
> > > >> Is
> > > >> > the
> > > >> > > > > idea
> > > >> > > > > > > that
> > > >> > > > > > > > > > since we have to resolve the regex on the server, we
> > > >> can do
> > > >> > > the
> > > >> > > > > same
> > > >> > > > > > > for
> > > >> > > > > > > > > > the topic name? The difference is that sending the
> > > >> regex is
> > > >> > > more
> > > >> > > > > > > > > efficient
> > > >> > > > > > > > > > whereas sending the topic names is less efficient.
> > > >> > > Furthermore,
> > > >> > > > > > > delete
> > > >> > > > > > > > > and
> > > >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> > > >> > > > > > > > >
> > > >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> > > >> server
> > > >> > > for
> > > >> > > > > both
> > > >> > > > > > > > > paths but I do agree with your point. As a second
> > > though,
> > > >> > using
> > > >> > > > > topic
> > > >> > > > > > > > > ids may be better here for the delete and recreation
> > > case.
> > > >> > > Also, I
> > > >> > > > > > > > > suppose that we may allow users to subscribe with
> > topic
> > > >> ids
> > > >> > in
> > > >> > > the
> > > >> > > > > > > > > future because that is the only way to be really
> > robust
> > > to
> > > >> > > topic
> > > >> > > > > > > > > re-creation.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Best,
> > > >> > > > > > > > > David
> > > >> > > > > > > > >
> > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > > >> > > djacot@confluent.io>
> > > >> > > > > > > wrote:
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Hi Justine,
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Thanks for your comments. Please find my answers
> > > below.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> > > >> > > exception
> > > >> > > > > of the
> > > >> > > > > > > > > > topic names based in the
> > > ConsumerGroupHeartbeatRequest.
> > > >> I
> > > >> > am
> > > >> > > not
> > > >> > > > > sure
> > > >> > > > > > > > > > if using topic names is the right call here. I need
> > to
> > > >> > think
> > > >> > > > > about it
> > > >> > > > > > > > > > a little more. Obviously, the KIP does not change
> > the
> > > >> > > > > fetch/commit
> > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> > > >> that
> > > >> > we
> > > >> > > > > should
> > > >> > > > > > > > > > include though as it would give better overall
> > > >> guarantee in
> > > >> > > the
> > > >> > > > > > > > > > producer.
> > > >> > > > > > > > > > - You're right. I think that I should not have
> > > mentioned
> > > >> > this
> > > >> > > > > flag at
> > > >> > > > > > > > > > all. I will remove it. We can use an internal
> > > >> configuration
> > > >> > > while
> > > >> > > > > > > > > > developing the feature.
> > > >> > > > > > > > > > - Both cluster types will be supported. The change
> > is
> > > >> > > > > orthogonal. The
> > > >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Best,
> > > >> > > > > > > > > > David
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> > > >> > > > > wangguoz@gmail.com>
> > > >> > > > > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Hi Ismael,
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> > > inlined
> > > >> > > below:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > > >> > > ismael@juma.me.uk>
> > > >> > > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > Thanks for the KIP. This has the potential to
> > be a
> > > >> > great
> > > >> > > > > > > > > improvement. A few
> > > >> > > > > > > > > > > > initial questions/comments:
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > 1. I think it's premature to talk about target
> > > >> versions
> > > >> > > for
> > > >> > > > > > > > > deprecation and
> > > >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> > > >> KRaft,
> > > >> > > this
> > > >> > > > > > > affects a
> > > >> > > > > > > > > core
> > > >> > > > > > > > > > > > client protocol and hence deprecation/removal
> > will
> > > >> be
> > > >> > > heavily
> > > >> > > > > > > > > dependent on
> > > >> > > > > > > > > > > > how quickly applications migrate to the new
> > > >> protocol.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> > > >> proposed
> > > >> > > > > timeline
> > > >> > > > > > > in
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan`
> > and
> > > >> > > instead
> > > >> > > > > just
> > > >> > > > > > > state
> > > >> > > > > > > > > > > that we will decide in the future about when we
> > > would
> > > >> > > > > deprecate old
> > > >> > > > > > > > > > > protocol and behaviors.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 2. The KIP says we intend to release this in
> > 4.x,
> > > >> but
> > > >> > it
> > > >> > > > > wasn't
> > > >> > > > > > > made
> > > >> > > > > > > > > clear
> > > >> > > > > > > > > > > > why. If we added that as a way to estimate when
> > > we'd
> > > >> > > > > deprecate
> > > >> > > > > > > and
> > > >> > > > > > > > > remove
> > > >> > > > > > > > > > > > the group protocol, I also suggest removing this
> > > >> part.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > I think that's not specifically related to the
> > > >> > > > > deprecation/removal
> > > >> > > > > > > > > timeline
> > > >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> > > >> > > broker-side
> > > >> > > > > > > > > > > implementation may be done first, and then the
> > > client
> > > >> > side,
> > > >> > > > > and we
> > > >> > > > > > > > > would
> > > >> > > > > > > > > > > only mark it as "released" by the time clients
> > > >> > > implementations
> > > >> > > > > are
> > > >> > > > > > > > > done. At
> > > >> > > > > > > > > > > that time, to enable the feature the clients need
> > to
> > > >> > first
> > > >> > > > > swap-in
> > > >> > > > > > > the
> > > >> > > > > > > > > > > bytecode with a rolling bounce and then set the
> > flag
> > > >> > with a
> > > >> > > > > second
> > > >> > > > > > > > > rolling
> > > >> > > > > > > > > > > bounce, and hence we feel it's better to be
> > released
> > > >> in a
> > > >> > > major
> > > >> > > > > > > > > version.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > > migration
> > > >> > > story.
> > > >> > > > > It
> > > >> > > > > > > > > sounds like
> > > >> > > > > > > > > > > > we're saying we will support online migrations.
> > Is
> > > >> that
> > > >> > > > > correct?
> > > >> > > > > > > We
> > > >> > > > > > > > > should
> > > >> > > > > > > > > > > > explain this in detail. It could also be done
> > as a
> > > >> > > separate
> > > >> > > > > KIP,
> > > >> > > > > > > if
> > > >> > > > > > > > > it's
> > > >> > > > > > > > > > > > easier.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yes I think that's the part we can be more
> > concrete
> > > >> about
> > > >> > > for
> > > >> > > > > sure
> > > >> > > > > > > (and
> > > >> > > > > > > > > > > this is related to your question 2) above). We
> > will
> > > >> work
> > > >> > on
> > > >> > > > > making
> > > >> > > > > > > it
> > > >> > > > > > > > > more
> > > >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > >> > > subscriptions
> > > >> > > > > to
> > > >> > > > > > > the
> > > >> > > > > > > > > server,
> > > >> > > > > > > > > > > > but it seems like there could be some tricky
> > > >> > > compatibility
> > > >> > > > > > > issues.
> > > >> > > > > > > > > Will we
> > > >> > > > > > > > > > > > have a mechanism for users to detect that they
> > > need
> > > >> to
> > > >> > > update
> > > >> > > > > > > their
> > > >> > > > > > > > > regex
> > > >> > > > > > > > > > > > before switching to the new protocol?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yes I think we need some tooling for non-java
> > client
> > > >> > users
> > > >> > > to
> > > >> > > > > sort
> > > >> > > > > > > of
> > > >> > > > > > > > > > > "dry-run" the client before switching to the new
> > > >> > protocol.
> > > >> > > I
> > > >> > > > > do not
> > > >> > > > > > > > > have a
> > > >> > > > > > > > > > > specific idea on top of my head though, maybe
> > others
> > > >> like
> > > >> > > @Matt
> > > >> > > > > > > > > Howlett can
> > > >> > > > > > > > > > > chime-in here?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 5. Related to the last question, will the Java
> > > >> client
> > > >> > > allow
> > > >> > > > > the
> > > >> > > > > > > > > users to
> > > >> > > > > > > > > > > > stick with the current regex engine for
> > > >> compatibility
> > > >> > > > > reasons?
> > > >> > > > > > > For
> > > >> > > > > > > > > example,
> > > >> > > > > > > > > > > > it may be handy to keep using client based regex
> > > at
> > > >> > > first to
> > > >> > > > > keep
> > > >> > > > > > > > > > > > migrations simple and then migrate to server
> > based
> > > >> > > regexes
> > > >> > > > > as a
> > > >> > > > > > > > > second
> > > >> > > > > > > > > > > > step.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Honestly I have not thought about that for java
> > > >> clients,
> > > >> > > and
> > > >> > > > > we can
> > > >> > > > > > > > > discuss
> > > >> > > > > > > > > > > that. What kind of compatibility issues do you
> > have
> > > in
> > > >> > > mind?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 6. When we say that the group coordinator will
> > be
> > > >> > > > > responsible for
> > > >> > > > > > > > > storing
> > > >> > > > > > > > > > > > the configurations and that the configurations
> > > will
> > > >> be
> > > >> > > > > deleted
> > > >> > > > > > > when
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> > > trigger
> > > >> > > deletion
> > > >> > > > > of
> > > >> > > > > > > > > > > > configurations?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> > > would
> > > >> > only
> > > >> > > > > > > transit to
> > > >> > > > > > > > > that
> > > >> > > > > > > > > > > state when the group is EMPTY and also all of its
> > > >> > metadata
> > > >> > > are
> > > >> > > > > > > gone),
> > > >> > > > > > > > > once
> > > >> > > > > > > > > > > it's transited to DEAD this group would never be
> > > >> revived.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> > > group
> > > >> > > > > coordinator
> > > >> > > > > > > > > make it
> > > >> > > > > > > > > > > > harder to list all cluster configs and their
> > > values?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > That's a good question, and our thoughts are that
> > > the
> > > >> > > so-called
> > > >> > > > > > > "group
> > > >> > > > > > > > > > > configurations" are overrides of the cluster-level
> > > >> > > > > configurations
> > > >> > > > > > > > > > > customized per group so when an admin list cluster
> > > >> > configs
> > > >> > > it's
> > > >> > > > > > > okay to
> > > >> > > > > > > > > > > list just the cluster-level "defaults", not
> > showing
> > > >> any
> > > >> > > > > per-group
> > > >> > > > > > > > > > > customizations.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 8. How would someone configure a group before
> > > >> starting
> > > >> > > the
> > > >> > > > > > > > > consumers? Have
> > > >> > > > > > > > > > > > we considered allowing the explicit creation of
> > > >> groups?
> > > >> > > > > > > > > Alternatively, the
> > > >> > > > > > > > > > > > configs could be decoupled from the group
> > > lifecycle.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > The configs can be created before the group itself
> > > as
> > > >> an
> > > >> > > > > > > independent
> > > >> > > > > > > > > entity
> > > >> > > > > > > > > > > --- of course, this requires the corresponding
> > > >> request to
> > > >> > > be
> > > >> > > > > > > routed to
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > right coordinator based on the group id --- the
> > only
> > > >> > thing
> > > >> > > that
> > > >> > > > > > > > > differs is,
> > > >> > > > > > > > > > > when the group itself is gone we also check if
> > there
> > > >> are
> > > >> > > any
> > > >> > > > > > > > > configuration
> > > >> > > > > > > > > > > entities related to that group and delete as well.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> > > the
> > > >> > > creation
> > > >> > > > > /
> > > >> > > > > > > > > deletion
> > > >> > > > > > > > > > > lifecycles of the config entities, and we would
> > like
> > > >> to
> > > >> > > hear
> > > >> > > > > > > everyone's
> > > >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> > > >> totally
> > > >> > > > > decouple
> > > >> > > > > > > group
> > > >> > > > > > > > > > > configs and hence not delete them at all when the
> > > >> group
> > > >> > is
> > > >> > > > > gone,
> > > >> > > > > > > but
> > > >> > > > > > > > > always
> > > >> > > > > > > > > > > require explicit deletion operations by
> > themselves.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the
> > Java
> > > >> > client
> > > >> > > > > still
> > > >> > > > > > > take
> > > >> > > > > > > > > a
> > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > > >> introduce an
> > > >> > > > > overload?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > I think we do not need to introduce an overload,
> > but
> > > >> I'm
> > > >> > > all
> > > >> > > > > ears
> > > >> > > > > > > if
> > > >> > > > > > > > > there
> > > >> > > > > > > > > > > may be some compatibility issues that we may
> > > overlook.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 10. I agree with Justine that we should be
> > clearer
> > > >> > about
> > > >> > > the
> > > >> > > > > > > reason
> > > >> > > > > > > > > to
> > > >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> > > >> flag.
> > > >> > > Maybe
> > > >> > > > > we
> > > >> > > > > > > mean
> > > >> > > > > > > > > that we
> > > >> > > > > > > > > > > > can switch the default for the feature flag to
> > > true
> > > >> > > based on
> > > >> > > > > the
> > > >> > > > > > > > > > > > metadata.version once we want to make it the
> > > >> default.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > > >> required
> > > >> > > ACLs,
> > > >> > > > > it
> > > >> > > > > > > > > would be
> > > >> > > > > > > > > > > > good to add that for consistency.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > > >> > requires
> > > >> > > > > "Read
> > > >> > > > > > > > > Group" even
> > > >> > > > > > > > > > > > though it seems to do more than reading.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > I had that thought myself as well, but in the end
> > we
> > > >> > could
> > > >> > > not
> > > >> > > > > > > find a
> > > >> > > > > > > > > > > better alternative: adding Write Group seems an
> > > >> overkill
> > > >> > > here
> > > >> > > > > > > since we
> > > >> > > > > > > > > do
> > > >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> > > and
> > > >> > > Describe
> > > >> > > > > on
> > > >> > > > > > > > > groups so
> > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > > consumer
> > > >> > with
> > > >> > > the
> > > >> > > > > new
> > > >> > > > > > > > > group
> > > >> > > > > > > > > > > > protocol? It would be good to have a section on
> > > >> this.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> > > >> add a
> > > >> > > > > section
> > > >> > > > > > > about
> > > >> > > > > > > > > > > that, but basically the idea is that consumer
> > would
> > > be
> > > >> > > totally
> > > >> > > > > > > > > agnostic in
> > > >> > > > > > > > > > > the new protocol as it's handled all by the
> > brokers.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > > >> coordinator
> > > >> > in
> > > >> > > > > Java.
> > > >> > > > > > > Even
> > > >> > > > > > > > > though
> > > >> > > > > > > > > > > > this is an implementation detail, do we plan to
> > > >> have a
> > > >> > > new
> > > >> > > > > gradle
> > > >> > > > > > > > > module
> > > >> > > > > > > > > > > > for it?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > We have not thought about that. But I think the
> > > answer
> > > >> > > should
> > > >> > > > > be
> > > >> > > > > > > yes.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes
> > to
> > > >> how
> > > >> > > many
> > > >> > > > > > > members
> > > >> > > > > > > > > the new
> > > >> > > > > > > > > > > > group protocol can support?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Within a group, I think we should shoot for 1000s
> > of
> > > >> > > members.
> > > >> > > > > But
> > > >> > > > > > > that
> > > >> > > > > > > > > > > scalability goals also depend on the offset
> > > management
> > > >> > > (commit,
> > > >> > > > > > > fetch)
> > > >> > > > > > > > > > > capabilities of the coordinator which we did not
> > > >> cover in
> > > >> > > this
> > > >> > > > > > > KIP, so
> > > >> > > > > > > > > it's
> > > >> > > > > > > > > > > hard to give a number that applies universally.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > > >> instead
> > > >> > of
> > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > >> ConsumerGroupHeartbeatRequest?
> > > >> > > Is the
> > > >> > > > > > > idea
> > > >> > > > > > > > > that
> > > >> > > > > > > > > > > > since we have to resolve the regex on the
> > server,
> > > we
> > > >> > can
> > > >> > > do
> > > >> > > > > the
> > > >> > > > > > > same
> > > >> > > > > > > > > for
> > > >> > > > > > > > > > > > the topic name? The difference is that sending
> > the
> > > >> > regex
> > > >> > > is
> > > >> > > > > more
> > > >> > > > > > > > > efficient
> > > >> > > > > > > > > > > > whereas sending the topic names is less
> > efficient.
> > > >> > > > > Furthermore,
> > > >> > > > > > > > > delete and
> > > >> > > > > > > > > > > > recreation is easier to handle if we have topic
> > > ids.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > The main reason to still let the clients send
> > names
> > > >> is to
> > > >> > > keep
> > > >> > > > > the
> > > >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> > > client
> > > >> > > only.
> > > >> > > > > Note
> > > >> > > > > > > that
> > > >> > > > > > > > > > > although we added topic id in KIP-516, we never
> > > >> > > implemented the
> > > >> > > > > > > logic
> > > >> > > > > > > > > on
> > > >> > > > > > > > > > > consumer/producers leveraging the related newer
> > > >> versioned
> > > >> > > RPCs,
> > > >> > > > > > > > > instead we
> > > >> > > > > > > > > > > just set the topic id as empty UUID. We want to
> > keep
> > > >> the
> > > >> > > > > > > > > consumer/producer
> > > >> > > > > > > > > > > to be thin and only delegate the reasoning on
> > broker
> > > >> and
> > > >> > > > > > > potentially
> > > >> > > > > > > > > admin
> > > >> > > > > > > > > > > clients.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > Thanks,
> > > >> > > > > > > > > > > > Ismael
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > >> > > > > > > > > > > > wrote:
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > > Hi all,
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > I would like to start a discussion thread on
> > > >> KIP-848:
> > > >> > > The
> > > >> > > > > Next
> > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> > > >> With
> > > >> > > this
> > > >> > > > > KIP,
> > > >> > > > > > > we
> > > >> > > > > > > > > aim
> > > >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> > > >> more
> > > >> > > > > reliable,
> > > >> > > > > > > more
> > > >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> > > >> easier
> > > >> > > to
> > > >> > > > > debug
> > > >> > > > > > > for
> > > >> > > > > > > > > > > > > operators.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > The KIP is here:
> > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > Please take a look and let me know what you
> > > think.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > Best,
> > > >> > > > > > > > > > > > > David
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > PS: I will be away from July 18th to August
> > 8th.
> > > >> That
> > > >> > > gives
> > > >> > > > > > > you a
> > > >> > > > > > > > > bit
> > > >> > > > > > > > > > > > > of time to read and digest this long KIP.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > --
> > > >> > > > > > > > > > > -- Guozhang
> > > >> > > > > > > > >
> > > >> > > > > > > >
> > > >> > > > > > > >
> > > >> > > > > > > > --
> > > >> > > > > > > > -- Guozhang
> > > >> > > > > > >
> > > >> > > > >
> > > >> > >
> > > >> > >
> > > >> >
> > > >>
> > > >>
> > > >> --
> > > >> -- Guozhang
> > > >>
> > > >
> > >
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hello Greg,

Thanks for reviewing the KIP! I hope David has addressed your questions
above, and I'd like to just emphasize one thing regarding your question 1):
one principle of the reconciliation is that, for a single client, before it
has completely revoked all the partitions that it should revoke, the
coordinator would not assign new partitions to it even if those partitions
have already been revoked by the old host and hence are available. And when
it has revoked all its partitions, the coordinator would start giving it
new partitions while at the same time bumping its epoch. Of course in
practice, a client should try to always revoke all its should-be-revoked
partitions together in a single shot which is more efficient, but from the
point view of the coordinator it's always "first revoke all you should
revoke, and then I will bump your epoch up and start giving you more
partitions". Though this may not be most efficient since coordinators maybe
holding on some available-partitions-to-give to the client before the
client has complete revoking, it gives us a good barrier (hence the epoch
bumping) to determine if the client can continue getting new partitions
even after soft failures or network partitions.


Guozhang

On Wed, Aug 3, 2022 at 1:05 PM Gregory Harris <gh...@gmail.com> wrote:

> Hey All!
>
> Thanks for the KIP, it's wonderful to see cooperative rebalancing making it
> down the stack!
>
> I had a few questions:
>
> 1. The 'Rejected Alternatives' section describes how member epoch should
> advance in step with the group epoch and assignment epoch values. I think
> that this is a good idea for the reasons described in the KIP. When the
> protocol is incrementally assigning partitions to a worker, what member
> epoch does each incremental assignment use? Are member epochs re-used, and
> a single member epoch can correspond to multiple different (monotonically
> larger) assignments?
>
> 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> not, should custom client-side assignor implementations interact with the
> Reason field, and how is its common meaning agreed upon? If so, what is the
> benefit of a distinct Reason field over including such functionality in the
> opaque metadata?
>
> 3. The following is included in the KIP: "Thanks to this, the input of the
> client side assignor is entirely driven by the group coordinator. The
> consumer is no longer responsible for maintaining any state besides its
> assigned partitions." Does this mean that the client-side assignor MAY
> incorporate additional non-Metadata state (such as partition throughput,
> cpu/memory metrics, config topics, etc), or that additional non-Metadata
> state SHOULD NOT be used?
>
> 4. I see that there are separate classes
> for org.apache.kafka.server.group.consumer.PartitionAssignor
> and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> overlap significantly. Is it possible for these two implementations to be
> unified? This would serve to promote feature parity of server-side and
> client-side assignors, and would also facilitate operational flexibility in
> certain situations. For example, if a server-side assignor has some poor
> behavior and needs a patch, deploying the patched assignor to the client
> and switching one consumer group to a client-side assignor may be faster
> and less risky than patching all of the brokers. With the currently
> proposed distinct APIs, a non-trivial reimplementation would have to be
> assembled, and if the two APIs have diverged significantly, then it is
> possible that a reimplementation would not be possible.
>
> --
> Greg Harris
> gharris1727@gmail.com
> github.com/gharris1727
>
> On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com> wrote:
>
> > Hi Guozhang/David,
> >
> > I created a confluence page to discuss how Connect would need to change
> > based on the new rebalance protocol. Here's the page:
> >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >
> > It's also pretty longish and I have tried to keep a format similar to
> > KIP-848. Let me know what you think. Also, do you think this should be
> > moved to a separate discussion thread or is this one fine?
> >
> > Thanks!
> > Sagar.
> >
> > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
> >
> > > Hello Guozhang,
> > >
> > > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > > analysis and come up with such a document.
> > >
> > > Thanks!
> > > Sagar.
> > >
> > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > >> Hello Sagar,
> > >>
> > >> It would be great if you could come back with some analysis on how to
> > >> implement the Connect side integration with the new protocol; so far
> > >> besides leveraging on the new "protocol type" we did not yet think
> > through
> > >> the Connect side implementations. For Streams here's a draft of
> > >> integration
> > >> plan:
> > >>
> > >>
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > >> just FYI for your analysis on Connect.
> > >>
> > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> > wrote:
> > >>
> > >> > Hi David,
> > >> >
> > >> > Thank you for your response. The reason I thought connect can also
> fit
> > >> into
> > >> > this new scheme is that even today the connect uses a
> > WorkerCoordinator
> > >> > extending from AbstractCoordinator to empower rebalances of
> > >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> > >> connect
> > >> > and uses the metadata() method by plumbing into
> > >> JoinGroupRequestProtocol.
> > >> >
> > >> > I think the changes to support connect would be similar at a high
> > level
> > >> to
> > >> > the changes in streams mainly because of the Client side assignors
> > being
> > >> > used in both. At an implementation level, we might need to make a
> lot
> > of
> > >> > changes to get onto this new assignment protocol like enhancing the
> > >> > JoinGroup request/response and SyncGroup and using
> > >> ConsumerGroupHeartbeat
> > >> > API etc again on similar lines to streams (or there might be
> > >> deviations). I
> > >> > would try to perform a detailed analysis of the same  and we can
> have
> > a
> > >> > separate discussion thread for that as that would derail this
> > discussion
> > >> > thread. Let me know if that sounds good to you.
> > >> >
> > >> > Thanks!
> > >> > Sagar.
> > >> >
> > >> >
> > >> >
> > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > <djacot@confluent.io.invalid
> > >> >
> > >> > wrote:
> > >> >
> > >> > > Hi Sagar,
> > >> > >
> > >> > > Thanks for your comments.
> > >> > >
> > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> > >> > >
> > >> > > 2) The idea is to transition C from his current assignment to his
> > >> > > target assignment when he can move to epoch 3. When that happens,
> > the
> > >> > > member assignment is updated and persisted with all its assigned
> > >> > > partitions even if they are not all revoked yet. In other words,
> the
> > >> > > member assignment becomes the target assignment. This is basically
> > an
> > >> > > optimization to avoid having to write all the changes to the log.
> > The
> > >> > > examples are based on the persisted state so I understand the
> > >> > > confusion. Let me see if I can improve this in the description.
> > >> > >
> > >> > > 3) Regarding Connect, it could reuse the protocol with a client
> side
> > >> > > assignor if it fits in the protocol. The assignment is about
> > >> > > topicid-partitions + metadata, could Connect fit into this?
> > >> > >
> > >> > > Best,
> > >> > > David
> > >> > >
> > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> > >> wrote:
> > >> > > >
> > >> > > > Hi David,
> > >> > > >
> > >> > > > Thanks for the KIP. I just had minor observations:
> > >> > > >
> > >> > > > 1) In the Assignment Error section in Client Side mode
> Assignment
> > >> > > process,
> > >> > > > you mentioned => `In this case, the client side assignor can
> > return
> > >> an
> > >> > > > error to the group coordinator`. In this case are you referring
> to
> > >> the
> > >> > > > Assignor returning an AssignmentError that's listed down towards
> > the
> > >> > end?
> > >> > > > If yes, do you think it would make sense to mention this
> > explicitly
> > >> > here?
> > >> > > >
> > >> > > > 2) In the Case Studies section, I have a slight confusion, not
> > sure
> > >> if
> > >> > > > others have the same. Consider this step:
> > >> > > >
> > >> > > > When B heartbeats, the group coordinator transitions him to
> epoch
> > 3
> > >> > > because
> > >> > > > B has no partitions to revoke. It persists the change and reply.
> > >> > > >
> > >> > > >    - Group (epoch=3)
> > >> > > >       - A
> > >> > > >       - B
> > >> > > >       - C
> > >> > > >    - Target Assignment (epoch=3)
> > >> > > >       - A - partitions=[foo-0]
> > >> > > >       - B - partitions=[foo-2]
> > >> > > >       - C - partitions=[foo-1]
> > >> > > >    - Member Assignment
> > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > >> > > >       - B - epoch=3, partitions=[foo-2]
> > >> > > >       - C - epoch=3, partitions=[foo-1]
> > >> > > >
> > >> > > > When C heartbeats, it transitions to epoch 3 but cannot get
> foo-1
> > >> yet.
> > >> > > >
> > >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> > yet,
> > >> > but
> > >> > > > based on the description above, it seems it already has it. Do
> you
> > >> > think
> > >> > > it
> > >> > > > would be better to remove it and populate it only when it
> actually
> > >> gets
> > >> > > it?
> > >> > > > I see this in a lot of other places, so have I understood it
> > >> > incorrectly
> > >> > > ?
> > >> > > >
> > >> > > >
> > >> > > > Regarding connect , it might be out of scope of this discussion,
> > but
> > >> > from
> > >> > > > what I understood it would probably be running in client side
> > >> assignor
> > >> > > mode
> > >> > > > even on the new rebalance protocol as it has its own Custom
> > >> > > Assignors(Eager
> > >> > > > and IncrementalCooperative).
> > >> > > >
> > >> > > > Thanks!
> > >> > > >
> > >> > > > Sagar.
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > >> > <djacot@confluent.io.invalid
> > >> > > >
> > >> > > > wrote:
> > >> > > >
> > >> > > > > Thanks Hector! Our goal is to move forward with specialized
> API
> > >> > > > > instead of relying on one generic API. For Connect, we can
> apply
> > >> the
> > >> > > > > exact same pattern and reuse/share the core implementation on
> > the
> > >> > > > > server side. For the schema registry, I think that we should
> > >> consider
> > >> > > > > having a tailored API to do simple membership/leader election.
> > >> > > > >
> > >> > > > > Best,
> > >> > > > > David
> > >> > > > >
> > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> ismael@juma.me.uk
> > >
> > >> > > wrote:
> > >> > > > > >
> > >> > > > > > Three quick comments:
> > >> > > > > >
> > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > >> com.google.re2j.Pattern, we
> > >> > > > > should
> > >> > > > > > document the differences in more detail before deciding one
> > way
> > >> or
> > >> > > > > another.
> > >> > > > > > That said, if people pass java.util.regex.Pattern, they
> expect
> > >> > their
> > >> > > > > > semantics to be honored. If we are doing something
> different,
> > >> then
> > >> > we
> > >> > > > > > should consider adding an overload with our own Pattern
> class
> > (I
> > >> > > don't
> > >> > > > > > think we'd want to expose re2j's at this point).
> > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > integrate
> > >> > fully
> > >> > > > > with
> > >> > > > > > it and should handle the topic recreation case correctly.
> > That's
> > >> > the
> > >> > > main
> > >> > > > > > part we need to handle. I agree with David that we'd want to
> > add
> > >> > > topic
> > >> > > > > ids
> > >> > > > > > to the relevant protocols that don't have it yet and that we
> > can
> > >> > > probably
> > >> > > > > > focus on the internals versus adding new APIs to the Java
> > >> Consumer
> > >> > > > > (unless
> > >> > > > > > we find that adding new APIs is required for reasonable
> > >> semantics).
> > >> > > > > > 3. I am still not sure about the coordinator storing the
> > >> configs.
> > >> > > It's
> > >> > > > > > powerful for configs to be centralized in the metadata log
> for
> > >> > > various
> > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > >> Similarly, I
> > >> > > am
> > >> > > > > not
> > >> > > > > > sure about automatically deleting configs in a way that they
> > >> cannot
> > >> > > be
> > >> > > > > > recovered. A good property for modern systems is to minimize
> > the
> > >> > > number
> > >> > > > > of
> > >> > > > > > unrecoverable data loss scenarios.
> > >> > > > > >
> > >> > > > > > Ismael
> > >> > > > > >
> > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > >> > > <djacot@confluent.io.invalid
> > >> > > > > >
> > >> > > > > > wrote:
> > >> > > > > >
> > >> > > > > > > Thanks Guozhang. My answers are below:
> > >> > > > > > >
> > >> > > > > > > > 1) the migration path, especially the last step when
> > clients
> > >> > > flip the
> > >> > > > > > > flag
> > >> > > > > > > > to enable the new protocol, in which we would have a
> > window
> > >> > where
> > >> > > > > both
> > >> > > > > > > new
> > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > >> members
> > >> > of
> > >> > > the
> > >> > > > > same
> > >> > > > > > > > group. How the coordinator could "mimic" the old
> behavior
> > >> while
> > >> > > > > using the
> > >> > > > > > > > new protocol is something we need to present about.
> > >> > > > > > >
> > >> > > > > > > Noted. I just published a new version of KIP which
> includes
> > >> more
> > >> > > > > > > details about this. See the "Supporting Online Consumer
> > Group
> > >> > > Upgrade"
> > >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan".
> I
> > >> think
> > >> > > that
> > >> > > > > > > I have to think through a few cases now but the overall
> idea
> > >> and
> > >> > > > > > > mechanism should be understandable.
> > >> > > > > > >
> > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic
> ids
> > >> are
> > >> > > only
> > >> > > > > used
> > >> > > > > > > as
> > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > via
> > >> any
> > >> > > > > public
> > >> > > > > > > APIs
> > >> > > > > > > > to consumers yet. I think the question is, first should
> we
> > >> let
> > >> > > the
> > >> > > > > > > consumer
> > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > to
> > >> > fully
> > >> > > > > > > leverage
> > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > the
> > >> > > topic
> > >> > > > > ids;
> > >> > > > > > > and
> > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > in
> > >> the
> > >> > > > > consumer
> > >> > > > > > > > public APIs as well (both subscribe/assign, as well as
> in
> > >> the
> > >> > > > > rebalance
> > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > >> > > > > > >
> > >> > > > > > > a) Assuming that we would include converting all the
> offsets
> > >> > > related
> > >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> > >> able
> > >> > to
> > >> > > > > > > fully operate with topic ids. That being said, it still
> has
> > to
> > >> > > provide
> > >> > > > > > > the topics names in various APIs so having a mapping in
> the
> > >> > > consumer
> > >> > > > > > > seems inevitable to me.
> > >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> > >> this
> > >> > > goes
> > >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> > >> > internals
> > >> > > > > > > here and we can consider this separately if we see value
> in
> > >> doing
> > >> > > it.
> > >> > > > > > >
> > >> > > > > > > Coming back to Ismael's point about using topic ids in the
> > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> > >> > advantage
> > >> > > in
> > >> > > > > > > favour of it. The consumer will have the opportunity to
> > >> validate
> > >> > > that
> > >> > > > > > > the topics exists before passing them into the group
> > rebalance
> > >> > > > > > > protocol. Obviously, the coordinator will also notice it
> but
> > >> it
> > >> > > does
> > >> > > > > > > not really have a way to reject an invalid topic in the
> > >> response.
> > >> > > > > > >
> > >> > > > > > > > I'm agreeing with David on all other minor questions
> > except
> > >> for
> > >> > > the
> > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's
> not
> > >> > > necessary
> > >> > > > > to
> > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > >> still
> > >> > > use
> > >> > > > > > > Pattern
> > >> > > > > > > > while just documenting that our subscription may be
> > >> rejected by
> > >> > > the
> > >> > > > > > > server.
> > >> > > > > > > > Since the incompatible case is a very rare scenario I
> felt
> > >> > using
> > >> > > an
> > >> > > > > > > > overloaded `String` based subscription may be more
> > >> vulnerable
> > >> > to
> > >> > > > > various
> > >> > > > > > > > invalid regexes.
> > >> > > > > > >
> > >> > > > > > > That could work. I have to look at the differences between
> > the
> > >> > two
> > >> > > > > > > engines to better understand the potential issues. My
> > >> > > understanding is
> > >> > > > > > > that would work for all the basic regular expressions. The
> > >> > > differences
> > >> > > > > > > between the two are mainly about the various character
> > >> classes. I
> > >> > > > > > > wonder what other people think about this.
> > >> > > > > > >
> > >> > > > > > > Best,
> > >> > > > > > > David
> > >> > > > > > >
> > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > >> > wangguoz@gmail.com
> > >> > > >
> > >> > > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > Thanks David! I think on the high level there are two
> meta
> > >> > > points we
> > >> > > > > need
> > >> > > > > > > > to concretize a bit more:
> > >> > > > > > > >
> > >> > > > > > > > 1) the migration path, especially the last step when
> > clients
> > >> > > flip the
> > >> > > > > > > flag
> > >> > > > > > > > to enable the new protocol, in which we would have a
> > window
> > >> > where
> > >> > > > > both
> > >> > > > > > > new
> > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > >> members
> > >> > of
> > >> > > the
> > >> > > > > same
> > >> > > > > > > > group. How the coordinator could "mimic" the old
> behavior
> > >> while
> > >> > > > > using the
> > >> > > > > > > > new protocol is something we need to present about.
> > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic
> ids
> > >> are
> > >> > > only
> > >> > > > > used
> > >> > > > > > > as
> > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > via
> > >> any
> > >> > > > > public
> > >> > > > > > > APIs
> > >> > > > > > > > to consumers yet. I think the question is, first should
> we
> > >> let
> > >> > > the
> > >> > > > > > > consumer
> > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > to
> > >> > fully
> > >> > > > > > > leverage
> > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > the
> > >> > > topic
> > >> > > > > ids;
> > >> > > > > > > and
> > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > in
> > >> the
> > >> > > > > consumer
> > >> > > > > > > > public APIs as well (both subscribe/assign, as well as
> in
> > >> the
> > >> > > > > rebalance
> > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > >> > > > > > > >
> > >> > > > > > > > I'm agreeing with David on all other minor questions
> > except
> > >> for
> > >> > > the
> > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's
> not
> > >> > > necessary
> > >> > > > > to
> > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > >> still
> > >> > > use
> > >> > > > > > > Pattern
> > >> > > > > > > > while just documenting that our subscription may be
> > >> rejected by
> > >> > > the
> > >> > > > > > > server.
> > >> > > > > > > > Since the incompatible case is a very rare scenario I
> felt
> > >> > using
> > >> > > an
> > >> > > > > > > > overloaded `String` based subscription may be more
> > >> vulnerable
> > >> > to
> > >> > > > > various
> > >> > > > > > > > invalid regexes.
> > >> > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > > Guozhang
> > >> > > > > > > >
> > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > >> > > > > <djacot@confluent.io.invalid
> > >> > > > > > > >
> > >> > > > > > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > > Hi Ismael,
> > >> > > > > > > > >
> > >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> > >> > inline.
> > >> > > > > > > > >
> > >> > > > > > > > > > 1. I think it's premature to talk about target
> > versions
> > >> for
> > >> > > > > > > deprecation
> > >> > > > > > > > > and
> > >> > > > > > > > > > removal of the existing group protocol. Unlike
> KRaft,
> > >> this
> > >> > > > > affects a
> > >> > > > > > > core
> > >> > > > > > > > > > client protocol and hence deprecation/removal will
> be
> > >> > heavily
> > >> > > > > > > dependent
> > >> > > > > > > > > on
> > >> > > > > > > > > > how quickly applications migrate to the new
> protocol.
> > >> > > > > > > > >
> > >> > > > > > > > > That makes sense. I will remove it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> but
> > it
> > >> > > wasn't
> > >> > > > > made
> > >> > > > > > > > > clear
> > >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> > >> > > deprecate
> > >> > > > > and
> > >> > > > > > > remove
> > >> > > > > > > > > > the group protocol, I also suggest removing this
> part.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> > >> rewrite
> > >> > > the
> > >> > > > > group
> > >> > > > > > > > > coordinator in Java while we implement the new
> protocol.
> > >> This
> > >> > > means
> > >> > > > > > > > > that the internals will be slightly different (e.g.
> > >> threading
> > >> > > > > model).
> > >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> > >> group
> > >> > > > > > > > > coordinator to the new group coordinator to a major
> > >> release.
> > >> > > The
> > >> > > > > > > > > alternative would be to use a flag to do the switch
> > >> instead
> > >> > of
> > >> > > > > relying
> > >> > > > > > > > > on the software upgrade.
> > >> > > > > > > > >
> > >> > > > > > > > > > 3. We need to flesh out the details of the migration
> > >> story.
> > >> > > It
> > >> > > > > sounds
> > >> > > > > > > > > like
> > >> > > > > > > > > > we're saying we will support online migrations. Is
> > that
> > >> > > correct?
> > >> > > > > We
> > >> > > > > > > > > should
> > >> > > > > > > > > > explain this in detail. It could also be done as a
> > >> separate
> > >> > > KIP,
> > >> > > > > if
> > >> > > > > > > it's
> > >> > > > > > > > > > easier.
> > >> > > > > > > > >
> > >> > > > > > > > > Yes, we will support online migrations for the group.
> > That
> > >> > > means
> > >> > > > > that
> > >> > > > > > > > > a group using the old protocol will be able to switch
> to
> > >> the
> > >> > > new
> > >> > > > > > > > > protocol.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me briefly explain how that will work though. It
> is
> > >> > > basically a
> > >> > > > > > > > > four step process:
> > >> > > > > > > > >
> > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a
> software
> > >> > > supporting
> > >> > > > > the
> > >> > > > > > > > > new group coordinator. Both the old and the new
> > >> coordinator
> > >> > > will
> > >> > > > > > > > > support the old protocol and rely on the same
> persisted
> > >> > > metadata so
> > >> > > > > > > > > they can work together. This point is an offline
> > >> migration.
> > >> > We
> > >> > > > > cannot
> > >> > > > > > > > > do this one live because it would require shutting
> down
> > >> the
> > >> > > current
> > >> > > > > > > > > coordinator and starting up the new one and that would
> > >> cause
> > >> > > > > > > > > unavailabilities.
> > >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> > to
> > >> X
> > >> > in
> > >> > > > > order
> > >> > > > > > > > > to enable the new protocol. This cannot be done before
> > 1)
> > >> is
> > >> > > > > > > > > terminated because the old coordinator doesn't support
> > the
> > >> > new
> > >> > > > > > > > > protocol.
> > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > supporting
> > >> the
> > >> > > > > online
> > >> > > > > > > > > migration (must have KIP-792). If the consumer is
> > already
> > >> > > there.
> > >> > > > > > > > > Nothing must be done at this point.
> > >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> > >> turned
> > >> > > on.
> > >> > > > > The
> > >> > > > > > > > > consumer group is automatically converted when the
> first
> > >> > > consumer
> > >> > > > > > > > > using the new protocol joins the group. While the
> > members
> > >> > > using the
> > >> > > > > > > > > old protocol are being upgraded, the old protocol is
> > >> proxied
> > >> > > into
> > >> > > > > the
> > >> > > > > > > > > new one.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me clarify all of this in the KIP.
> > >> > > > > > > > >
> > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > >> subscriptions
> > >> > > to
> > >> > > > > the
> > >> > > > > > > > > server,
> > >> > > > > > > > > > but it seems like there could be some tricky
> > >> compatibility
> > >> > > > > issues.
> > >> > > > > > > Will
> > >> > > > > > > > > we
> > >> > > > > > > > > > have a mechanism for users to detect that they need
> to
> > >> > update
> > >> > > > > their
> > >> > > > > > > regex
> > >> > > > > > > > > > before switching to the new protocol?
> > >> > > > > > > > >
> > >> > > > > > > > > I think that I am a bit more optimistic than you on
> this
> > >> > > point. I
> > >> > > > > > > > > believe that the majority of the cases are simple
> > regexes
> > >> > which
> > >> > > > > should
> > >> > > > > > > > > work with the new engine. The coordinator will verify
> > the
> > >> > regex
> > >> > > > > anyway
> > >> > > > > > > > > and reject the consumer if the regex is not valid.
> > Coming
> > >> > back
> > >> > > to
> > >> > > > > the
> > >> > > > > > > > > migration path, in the worst case, the first upgraded
> > >> > consumer
> > >> > > > > joining
> > >> > > > > > > > > the group will be rejected. This should be used as the
> > >> last
> > >> > > > > defence, I
> > >> > > > > > > > > would say.
> > >> > > > > > > > >
> > >> > > > > > > > > One way for customers to validate their regex before
> > >> > upgrading
> > >> > > > > their
> > >> > > > > > > > > prod would be to test them with another group. For
> > >> instance,
> > >> > > that
> > >> > > > > > > > > could be done in a pre-prod environment. Another way
> > >> would be
> > >> > > to
> > >> > > > > > > > > extend the consumer-group tool to provide a regex
> > >> validation
> > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > >> > > > > > > > >
> > >> > > > > > > > > > 5. Related to the last question, will the Java
> client
> > >> allow
> > >> > > the
> > >> > > > > > > users to
> > >> > > > > > > > > > stick with the current regex engine for
> compatibility
> > >> > > reasons?
> > >> > > > > For
> > >> > > > > > > > > example,
> > >> > > > > > > > > > it may be handy to keep using client based regex at
> > >> first
> > >> > to
> > >> > > keep
> > >> > > > > > > > > > migrations simple and then migrate to server based
> > >> regexes
> > >> > > as a
> > >> > > > > > > second
> > >> > > > > > > > > step.
> > >> > > > > > > > >
> > >> > > > > > > > > I understand your point but I am concerned that this
> > would
> > >> > > allow
> > >> > > > > users
> > >> > > > > > > > > to actually stay in this mode. That would go against
> our
> > >> goal
> > >> > > of
> > >> > > > > > > > > simplifying the client because we would have to
> continue
> > >> > > monitoring
> > >> > > > > > > > > the metadata on the client side. I would rather not do
> > >> this.
> > >> > > > > > > > >
> > >> > > > > > > > > > 6. When we say that the group coordinator will be
> > >> > > responsible for
> > >> > > > > > > storing
> > >> > > > > > > > > > the configurations and that the configurations will
> be
> > >> > > deleted
> > >> > > > > when
> > >> > > > > > > the
> > >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> > >> > deletion
> > >> > > of
> > >> > > > > > > > > > configurations?
> > >> > > > > > > > >
> > >> > > > > > > > > That's right. The configurations will be deleted when
> > the
> > >> > > group is
> > >> > > > > > > > > deleted. They go together.
> > >> > > > > > > > >
> > >> > > > > > > > > > 7. Will the choice to store the configs in the group
> > >> > > coordinator
> > >> > > > > > > make it
> > >> > > > > > > > > > harder to list all cluster configs and their values?
> > >> > > > > > > > >
> > >> > > > > > > > > I don't think so. The group configurations are
> overrides
> > >> of
> > >> > > cluster
> > >> > > > > > > > > configs. If you want to know all the overrides though,
> > you
> > >> > > would
> > >> > > > > have
> > >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> > the
> > >> > > metadata
> > >> > > > > log
> > >> > > > > > > > > for instance.
> > >> > > > > > > > >
> > >> > > > > > > > > > 8. How would someone configure a group before
> starting
> > >> the
> > >> > > > > consumers?
> > >> > > > > > > > > Have
> > >> > > > > > > > > > we considered allowing the explicit creation of
> > groups?
> > >> > > > > > > Alternatively,
> > >> > > > > > > > > the
> > >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> > >> > > > > > > > >
> > >> > > > > > > > > Yes. The group will be automatically created in this
> > case.
> > >> > > However,
> > >> > > > > > > > > the configs will be lost after the retention period of
> > the
> > >> > > group
> > >> > > > > > > > > passes.
> > >> > > > > > > > >
> > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > >> client
> > >> > > still
> > >> > > > > take
> > >> > > > > > > a
> > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> > an
> > >> > > overload?
> > >> > > > > > > > >
> > >> > > > > > > > > That's a very group question. I forgot about that one.
> > As
> > >> the
> > >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> > the
> > >> > > engine
> > >> > > > > that
> > >> > > > > > > > > we plan to use, it might be better to deprecate it and
> > >> use an
> > >> > > > > overload
> > >> > > > > > > > > which takes a string. We would rely on the server side
> > >> > > validation.
> > >> > > > > > > > > During the migration, I think that we could still try
> to
> > >> > > toString
> > >> > > > > the
> > >> > > > > > > > > regex and use it. That should work, I think, in the
> > >> majority
> > >> > > of the
> > >> > > > > > > > > cases.
> > >> > > > > > > > >
> > >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> > about
> > >> > the
> > >> > > > > reason
> > >> > > > > > > to
> > >> > > > > > > > > > switch to IBP/metadata.version from the feature
> flag.
> > >> Maybe
> > >> > > we
> > >> > > > > mean
> > >> > > > > > > that
> > >> > > > > > > > > we
> > >> > > > > > > > > > can switch the default for the feature flag to true
> > >> based
> > >> > on
> > >> > > the
> > >> > > > > > > > > > metadata.version once we want to make it the
> default.
> > >> > > > > > > > >
> > >> > > > > > > > > My plan was to use that feature flag mainly during the
> > >> > > development
> > >> > > > > > > > > phase. I should not have mentioned it, I think,
> because
> > we
> > >> > > could
> > >> > > > > use
> > >> > > > > > > > > an internal config for it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > required
> > >> > > ACLs, it
> > >> > > > > > > would
> > >> > > > > > > > > be
> > >> > > > > > > > > > good to add that for consistency.
> > >> > > > > > > > >
> > >> > > > > > > > > Noted.
> > >> > > > > > > > >
> > >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > requires
> > >> > > "Read
> > >> > > > > Group"
> > >> > > > > > > > > even
> > >> > > > > > > > > > though it seems to do more than reading.
> > >> > > > > > > > >
> > >> > > > > > > > > I agree. This is how the current protocol works
> though.
> > We
> > >> > only
> > >> > > > > > > > > require "Read Group" to join a group. We could
> consider
> > >> > > changing
> > >> > > > > this
> > >> > > > > > > > > but I am not sure that it is worth it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> > with
> > >> > the
> > >> > > new
> > >> > > > > > > group
> > >> > > > > > > > > > protocol? It would be good to have a section on
> this.
> > >> > > > > > > > >
> > >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> > >> have a
> > >> > > new
> > >> > > > > > > > > topic id so it will treat it like a topic with a
> > different
> > >> > > name.
> > >> > > > > The
> > >> > > > > > > > > only issue is that the fetch/commit offsets APIs do
> not
> > >> > support
> > >> > > > > topic
> > >> > > > > > > > > IDs so the consumer would reuse the offsets based on
> the
> > >> > same.
> > >> > > I
> > >> > > > > think
> > >> > > > > > > > > that we should update those APIs as well in order to
> be
> > >> > > consistent
> > >> > > > > end
> > >> > > > > > > > > to end. That would strengthen the semantics of the
> > >> consumer.
> > >> > > > > > > > >
> > >> > > > > > > > > > 14. The KIP mentions we will write the new
> coordinator
> > >> in
> > >> > > Java.
> > >> > > > > Even
> > >> > > > > > > > > though
> > >> > > > > > > > > > this is an implementation detail, do we plan to
> have a
> > >> new
> > >> > > gradle
> > >> > > > > > > module
> > >> > > > > > > > > > for it?
> > >> > > > > > > > >
> > >> > > > > > > > > Yes.
> > >> > > > > > > > >
> > >> > > > > > > > > > 15. Do we have a scalability goal when it comes to
> how
> > >> many
> > >> > > > > members
> > >> > > > > > > the
> > >> > > > > > > > > new
> > >> > > > > > > > > > group protocol can support?
> > >> > > > > > > > >
> > >> > > > > > > > > We don't have numbers at the moment. The protocol
> should
> > >> > > support
> > >> > > > > 1000s
> > >> > > > > > > > > of members per group. We will measure this when we
> have
> > a
> > >> > first
> > >> > > > > > > > > implementation. Note that we might have other
> > bottlenecks
> > >> > down
> > >> > > the
> > >> > > > > > > > > road (e.g. offset commits).
> > >> > > > > > > > >
> > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> instead
> > of
> > >> > > > > > > > > > SubscribedTopicNames in
> ConsumerGroupHeartbeatRequest?
> > >> Is
> > >> > the
> > >> > > > > idea
> > >> > > > > > > that
> > >> > > > > > > > > > since we have to resolve the regex on the server, we
> > >> can do
> > >> > > the
> > >> > > > > same
> > >> > > > > > > for
> > >> > > > > > > > > > the topic name? The difference is that sending the
> > >> regex is
> > >> > > more
> > >> > > > > > > > > efficient
> > >> > > > > > > > > > whereas sending the topic names is less efficient.
> > >> > > Furthermore,
> > >> > > > > > > delete
> > >> > > > > > > > > and
> > >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> > >> > > > > > > > >
> > >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> > >> server
> > >> > > for
> > >> > > > > both
> > >> > > > > > > > > paths but I do agree with your point. As a second
> > though,
> > >> > using
> > >> > > > > topic
> > >> > > > > > > > > ids may be better here for the delete and recreation
> > case.
> > >> > > Also, I
> > >> > > > > > > > > suppose that we may allow users to subscribe with
> topic
> > >> ids
> > >> > in
> > >> > > the
> > >> > > > > > > > > future because that is the only way to be really
> robust
> > to
> > >> > > topic
> > >> > > > > > > > > re-creation.
> > >> > > > > > > > >
> > >> > > > > > > > > Best,
> > >> > > > > > > > > David
> > >> > > > > > > > >
> > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > >> > > djacot@confluent.io>
> > >> > > > > > > wrote:
> > >> > > > > > > > > >
> > >> > > > > > > > > > Hi Justine,
> > >> > > > > > > > > >
> > >> > > > > > > > > > Thanks for your comments. Please find my answers
> > below.
> > >> > > > > > > > > >
> > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> > >> > > exception
> > >> > > > > of the
> > >> > > > > > > > > > topic names based in the
> > ConsumerGroupHeartbeatRequest.
> > >> I
> > >> > am
> > >> > > not
> > >> > > > > sure
> > >> > > > > > > > > > if using topic names is the right call here. I need
> to
> > >> > think
> > >> > > > > about it
> > >> > > > > > > > > > a little more. Obviously, the KIP does not change
> the
> > >> > > > > fetch/commit
> > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> > >> that
> > >> > we
> > >> > > > > should
> > >> > > > > > > > > > include though as it would give better overall
> > >> guarantee in
> > >> > > the
> > >> > > > > > > > > > producer.
> > >> > > > > > > > > > - You're right. I think that I should not have
> > mentioned
> > >> > this
> > >> > > > > flag at
> > >> > > > > > > > > > all. I will remove it. We can use an internal
> > >> configuration
> > >> > > while
> > >> > > > > > > > > > developing the feature.
> > >> > > > > > > > > > - Both cluster types will be supported. The change
> is
> > >> > > > > orthogonal. The
> > >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> > >> > > > > > > > > >
> > >> > > > > > > > > > Best,
> > >> > > > > > > > > > David
> > >> > > > > > > > > >
> > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> > >> > > > > wangguoz@gmail.com>
> > >> > > > > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Hi Ismael,
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> > inlined
> > >> > > below:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > >> > > ismael@juma.me.uk>
> > >> > > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > Thanks for the KIP. This has the potential to
> be a
> > >> > great
> > >> > > > > > > > > improvement. A few
> > >> > > > > > > > > > > > initial questions/comments:
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > 1. I think it's premature to talk about target
> > >> versions
> > >> > > for
> > >> > > > > > > > > deprecation and
> > >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> > >> KRaft,
> > >> > > this
> > >> > > > > > > affects a
> > >> > > > > > > > > core
> > >> > > > > > > > > > > > client protocol and hence deprecation/removal
> will
> > >> be
> > >> > > heavily
> > >> > > > > > > > > dependent on
> > >> > > > > > > > > > > > how quickly applications migrate to the new
> > >> protocol.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> > >> proposed
> > >> > > > > timeline
> > >> > > > > > > in
> > >> > > > > > > > > the
> > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan`
> and
> > >> > > instead
> > >> > > > > just
> > >> > > > > > > state
> > >> > > > > > > > > > > that we will decide in the future about when we
> > would
> > >> > > > > deprecate old
> > >> > > > > > > > > > > protocol and behaviors.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 2. The KIP says we intend to release this in
> 4.x,
> > >> but
> > >> > it
> > >> > > > > wasn't
> > >> > > > > > > made
> > >> > > > > > > > > clear
> > >> > > > > > > > > > > > why. If we added that as a way to estimate when
> > we'd
> > >> > > > > deprecate
> > >> > > > > > > and
> > >> > > > > > > > > remove
> > >> > > > > > > > > > > > the group protocol, I also suggest removing this
> > >> part.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I think that's not specifically related to the
> > >> > > > > deprecation/removal
> > >> > > > > > > > > timeline
> > >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> > >> > > broker-side
> > >> > > > > > > > > > > implementation may be done first, and then the
> > client
> > >> > side,
> > >> > > > > and we
> > >> > > > > > > > > would
> > >> > > > > > > > > > > only mark it as "released" by the time clients
> > >> > > implementations
> > >> > > > > are
> > >> > > > > > > > > done. At
> > >> > > > > > > > > > > that time, to enable the feature the clients need
> to
> > >> > first
> > >> > > > > swap-in
> > >> > > > > > > the
> > >> > > > > > > > > > > bytecode with a rolling bounce and then set the
> flag
> > >> > with a
> > >> > > > > second
> > >> > > > > > > > > rolling
> > >> > > > > > > > > > > bounce, and hence we feel it's better to be
> released
> > >> in a
> > >> > > major
> > >> > > > > > > > > version.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > migration
> > >> > > story.
> > >> > > > > It
> > >> > > > > > > > > sounds like
> > >> > > > > > > > > > > > we're saying we will support online migrations.
> Is
> > >> that
> > >> > > > > correct?
> > >> > > > > > > We
> > >> > > > > > > > > should
> > >> > > > > > > > > > > > explain this in detail. It could also be done
> as a
> > >> > > separate
> > >> > > > > KIP,
> > >> > > > > > > if
> > >> > > > > > > > > it's
> > >> > > > > > > > > > > > easier.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes I think that's the part we can be more
> concrete
> > >> about
> > >> > > for
> > >> > > > > sure
> > >> > > > > > > (and
> > >> > > > > > > > > > > this is related to your question 2) above). We
> will
> > >> work
> > >> > on
> > >> > > > > making
> > >> > > > > > > it
> > >> > > > > > > > > more
> > >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> > >> > > subscriptions
> > >> > > > > to
> > >> > > > > > > the
> > >> > > > > > > > > server,
> > >> > > > > > > > > > > > but it seems like there could be some tricky
> > >> > > compatibility
> > >> > > > > > > issues.
> > >> > > > > > > > > Will we
> > >> > > > > > > > > > > > have a mechanism for users to detect that they
> > need
> > >> to
> > >> > > update
> > >> > > > > > > their
> > >> > > > > > > > > regex
> > >> > > > > > > > > > > > before switching to the new protocol?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes I think we need some tooling for non-java
> client
> > >> > users
> > >> > > to
> > >> > > > > sort
> > >> > > > > > > of
> > >> > > > > > > > > > > "dry-run" the client before switching to the new
> > >> > protocol.
> > >> > > I
> > >> > > > > do not
> > >> > > > > > > > > have a
> > >> > > > > > > > > > > specific idea on top of my head though, maybe
> others
> > >> like
> > >> > > @Matt
> > >> > > > > > > > > Howlett can
> > >> > > > > > > > > > > chime-in here?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 5. Related to the last question, will the Java
> > >> client
> > >> > > allow
> > >> > > > > the
> > >> > > > > > > > > users to
> > >> > > > > > > > > > > > stick with the current regex engine for
> > >> compatibility
> > >> > > > > reasons?
> > >> > > > > > > For
> > >> > > > > > > > > example,
> > >> > > > > > > > > > > > it may be handy to keep using client based regex
> > at
> > >> > > first to
> > >> > > > > keep
> > >> > > > > > > > > > > > migrations simple and then migrate to server
> based
> > >> > > regexes
> > >> > > > > as a
> > >> > > > > > > > > second
> > >> > > > > > > > > > > > step.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Honestly I have not thought about that for java
> > >> clients,
> > >> > > and
> > >> > > > > we can
> > >> > > > > > > > > discuss
> > >> > > > > > > > > > > that. What kind of compatibility issues do you
> have
> > in
> > >> > > mind?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 6. When we say that the group coordinator will
> be
> > >> > > > > responsible for
> > >> > > > > > > > > storing
> > >> > > > > > > > > > > > the configurations and that the configurations
> > will
> > >> be
> > >> > > > > deleted
> > >> > > > > > > when
> > >> > > > > > > > > the
> > >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> > trigger
> > >> > > deletion
> > >> > > > > of
> > >> > > > > > > > > > > > configurations?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> > would
> > >> > only
> > >> > > > > > > transit to
> > >> > > > > > > > > that
> > >> > > > > > > > > > > state when the group is EMPTY and also all of its
> > >> > metadata
> > >> > > are
> > >> > > > > > > gone),
> > >> > > > > > > > > once
> > >> > > > > > > > > > > it's transited to DEAD this group would never be
> > >> revived.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> > group
> > >> > > > > coordinator
> > >> > > > > > > > > make it
> > >> > > > > > > > > > > > harder to list all cluster configs and their
> > values?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > That's a good question, and our thoughts are that
> > the
> > >> > > so-called
> > >> > > > > > > "group
> > >> > > > > > > > > > > configurations" are overrides of the cluster-level
> > >> > > > > configurations
> > >> > > > > > > > > > > customized per group so when an admin list cluster
> > >> > configs
> > >> > > it's
> > >> > > > > > > okay to
> > >> > > > > > > > > > > list just the cluster-level "defaults", not
> showing
> > >> any
> > >> > > > > per-group
> > >> > > > > > > > > > > customizations.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 8. How would someone configure a group before
> > >> starting
> > >> > > the
> > >> > > > > > > > > consumers? Have
> > >> > > > > > > > > > > > we considered allowing the explicit creation of
> > >> groups?
> > >> > > > > > > > > Alternatively, the
> > >> > > > > > > > > > > > configs could be decoupled from the group
> > lifecycle.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > The configs can be created before the group itself
> > as
> > >> an
> > >> > > > > > > independent
> > >> > > > > > > > > entity
> > >> > > > > > > > > > > --- of course, this requires the corresponding
> > >> request to
> > >> > > be
> > >> > > > > > > routed to
> > >> > > > > > > > > the
> > >> > > > > > > > > > > right coordinator based on the group id --- the
> only
> > >> > thing
> > >> > > that
> > >> > > > > > > > > differs is,
> > >> > > > > > > > > > > when the group itself is gone we also check if
> there
> > >> are
> > >> > > any
> > >> > > > > > > > > configuration
> > >> > > > > > > > > > > entities related to that group and delete as well.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> > the
> > >> > > creation
> > >> > > > > /
> > >> > > > > > > > > deletion
> > >> > > > > > > > > > > lifecycles of the config entities, and we would
> like
> > >> to
> > >> > > hear
> > >> > > > > > > everyone's
> > >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> > >> totally
> > >> > > > > decouple
> > >> > > > > > > group
> > >> > > > > > > > > > > configs and hence not delete them at all when the
> > >> group
> > >> > is
> > >> > > > > gone,
> > >> > > > > > > but
> > >> > > > > > > > > always
> > >> > > > > > > > > > > require explicit deletion operations by
> themselves.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the
> Java
> > >> > client
> > >> > > > > still
> > >> > > > > > > take
> > >> > > > > > > > > a
> > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > >> introduce an
> > >> > > > > overload?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I think we do not need to introduce an overload,
> but
> > >> I'm
> > >> > > all
> > >> > > > > ears
> > >> > > > > > > if
> > >> > > > > > > > > there
> > >> > > > > > > > > > > may be some compatibility issues that we may
> > overlook.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 10. I agree with Justine that we should be
> clearer
> > >> > about
> > >> > > the
> > >> > > > > > > reason
> > >> > > > > > > > > to
> > >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> > >> flag.
> > >> > > Maybe
> > >> > > > > we
> > >> > > > > > > mean
> > >> > > > > > > > > that we
> > >> > > > > > > > > > > > can switch the default for the feature flag to
> > true
> > >> > > based on
> > >> > > > > the
> > >> > > > > > > > > > > > metadata.version once we want to make it the
> > >> default.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > >> required
> > >> > > ACLs,
> > >> > > > > it
> > >> > > > > > > > > would be
> > >> > > > > > > > > > > > good to add that for consistency.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Ack, we can certainly do that.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > >> > requires
> > >> > > > > "Read
> > >> > > > > > > > > Group" even
> > >> > > > > > > > > > > > though it seems to do more than reading.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I had that thought myself as well, but in the end
> we
> > >> > could
> > >> > > not
> > >> > > > > > > find a
> > >> > > > > > > > > > > better alternative: adding Write Group seems an
> > >> overkill
> > >> > > here
> > >> > > > > > > since we
> > >> > > > > > > > > do
> > >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> > and
> > >> > > Describe
> > >> > > > > on
> > >> > > > > > > > > groups so
> > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > consumer
> > >> > with
> > >> > > the
> > >> > > > > new
> > >> > > > > > > > > group
> > >> > > > > > > > > > > > protocol? It would be good to have a section on
> > >> this.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> > >> add a
> > >> > > > > section
> > >> > > > > > > about
> > >> > > > > > > > > > > that, but basically the idea is that consumer
> would
> > be
> > >> > > totally
> > >> > > > > > > > > agnostic in
> > >> > > > > > > > > > > the new protocol as it's handled all by the
> brokers.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > >> coordinator
> > >> > in
> > >> > > > > Java.
> > >> > > > > > > Even
> > >> > > > > > > > > though
> > >> > > > > > > > > > > > this is an implementation detail, do we plan to
> > >> have a
> > >> > > new
> > >> > > > > gradle
> > >> > > > > > > > > module
> > >> > > > > > > > > > > > for it?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > We have not thought about that. But I think the
> > answer
> > >> > > should
> > >> > > > > be
> > >> > > > > > > yes.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes
> to
> > >> how
> > >> > > many
> > >> > > > > > > members
> > >> > > > > > > > > the new
> > >> > > > > > > > > > > > group protocol can support?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Within a group, I think we should shoot for 1000s
> of
> > >> > > members.
> > >> > > > > But
> > >> > > > > > > that
> > >> > > > > > > > > > > scalability goals also depend on the offset
> > management
> > >> > > (commit,
> > >> > > > > > > fetch)
> > >> > > > > > > > > > > capabilities of the coordinator which we did not
> > >> cover in
> > >> > > this
> > >> > > > > > > KIP, so
> > >> > > > > > > > > it's
> > >> > > > > > > > > > > hard to give a number that applies universally.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > >> instead
> > >> > of
> > >> > > > > > > > > > > > SubscribedTopicNames in
> > >> ConsumerGroupHeartbeatRequest?
> > >> > > Is the
> > >> > > > > > > idea
> > >> > > > > > > > > that
> > >> > > > > > > > > > > > since we have to resolve the regex on the
> server,
> > we
> > >> > can
> > >> > > do
> > >> > > > > the
> > >> > > > > > > same
> > >> > > > > > > > > for
> > >> > > > > > > > > > > > the topic name? The difference is that sending
> the
> > >> > regex
> > >> > > is
> > >> > > > > more
> > >> > > > > > > > > efficient
> > >> > > > > > > > > > > > whereas sending the topic names is less
> efficient.
> > >> > > > > Furthermore,
> > >> > > > > > > > > delete and
> > >> > > > > > > > > > > > recreation is easier to handle if we have topic
> > ids.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > The main reason to still let the clients send
> names
> > >> is to
> > >> > > keep
> > >> > > > > the
> > >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> > client
> > >> > > only.
> > >> > > > > Note
> > >> > > > > > > that
> > >> > > > > > > > > > > although we added topic id in KIP-516, we never
> > >> > > implemented the
> > >> > > > > > > logic
> > >> > > > > > > > > on
> > >> > > > > > > > > > > consumer/producers leveraging the related newer
> > >> versioned
> > >> > > RPCs,
> > >> > > > > > > > > instead we
> > >> > > > > > > > > > > just set the topic id as empty UUID. We want to
> keep
> > >> the
> > >> > > > > > > > > consumer/producer
> > >> > > > > > > > > > > to be thin and only delegate the reasoning on
> broker
> > >> and
> > >> > > > > > > potentially
> > >> > > > > > > > > admin
> > >> > > > > > > > > > > clients.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > Thanks,
> > >> > > > > > > > > > > > Ismael
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> > >> > > > > > > > > <dj...@confluent.io.invalid>
> > >> > > > > > > > > > > > wrote:
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > > Hi all,
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > I would like to start a discussion thread on
> > >> KIP-848:
> > >> > > The
> > >> > > > > Next
> > >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> > >> With
> > >> > > this
> > >> > > > > KIP,
> > >> > > > > > > we
> > >> > > > > > > > > aim
> > >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> > >> more
> > >> > > > > reliable,
> > >> > > > > > > more
> > >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> > >> easier
> > >> > > to
> > >> > > > > debug
> > >> > > > > > > for
> > >> > > > > > > > > > > > > operators.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > The KIP is here:
> > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > Please take a look and let me know what you
> > think.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > Best,
> > >> > > > > > > > > > > > > David
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > PS: I will be away from July 18th to August
> 8th.
> > >> That
> > >> > > gives
> > >> > > > > > > you a
> > >> > > > > > > > > bit
> > >> > > > > > > > > > > > > of time to read and digest this long KIP.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > --
> > >> > > > > > > > > > > -- Guozhang
> > >> > > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > > --
> > >> > > > > > > > -- Guozhang
> > >> > > > > > >
> > >> > > > >
> > >> > >
> > >> > >
> > >> >
> > >>
> > >>
> > >> --
> > >> -- Guozhang
> > >>
> > >
> >
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Guozhang,

> So to complete the full picture here, we'd need to modify both HB request
> and response so that the client can also indicate a new rebalance via the
> HB request as well, right?

This is already the case. A member can only trigger a rebalance by
updating its state (subscriptions, metadata, reason, etc.) via the
heartbeat request. GroupPrepareAssignment is only used by the member
to get the group state when it has to run a local assignor.

Executing a client side assignment works as follows:
1) The coordinator selects a member and notifies him in his next heartbeat.
2) The member calls PrepareAssignment to get the group state.
3) The member computes the assignment.
4) The member calls InstallAssignment to install it.

If we remove PrepareAssignment, we would get the group state in 1)
directly so we would save the round trip to get the group state.
Whether the rebalance is triggered by a member or by the coordinator
does not change the flow.

Cheers,
David

On Tue, Sep 13, 2022 at 9:08 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hello David,
>
> Thanks for bringing this question up. I think the main benefits as you
> listed is 2) above if it stays; just to clarify, we would only be able to
> save one round trip if the rebalance is still triggered by the broker; if
> the rebalance is triggered by the client then the num.round.trips are the
> same:
>
> 1) With GroupPrepareAssignment:
>
> T0: client decides to do a new assignment, suppose it has already sent a HB
> and hence has to wait for it to return first since only one request /
> response can be inflight with the coordinator's socket.
> T1: client receives the HB response, and then sends the
> GroupPrepareAssignment request.
> T2: the GroupPrepareAssignment response is returned.
> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> request.
>
> In total, two round trips.
>
> 2) Without GroupPrepareAssignment:
>
> T0: client decides to do a new assignment, suppose it has already sent a HB
> and hence has to wait for it to return first since only one request /
> response can be inflight with the coordinator's socket.
> T1: client receives the HB response, and then sends the new HB request with
> the flag indicating a new rebalance needed..
> T2: the HB response with the optional member metadata map is returned.
> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> request.
>
> In total, two round trips as well.
>
> -----------------------------
>
> So to complete the full picture here, we'd need to modify both HB request
> and response so that the client can also indicate a new rebalance via the
> HB request as well, right?
>
> Assuming all above is true, I think it's okay to merge the
> GroupPrepareAssignment into HB given that we can make the additional fields
> encoding the full member (subscription) metadata and topic metadata as
> optional fields.
>
>
> Guozhang
>
>
>
>
>
>
>
>
>
>
> On Mon, Sep 12, 2022 at 5:22 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi all,
> >
> > During an offline conversation, someone asked why we need the
> > ConsumerGroupPrepareAssignment API and suggested that we could instead
> > provide the group state in the heartbeat response. This has a few
> > advantages: 1) it does not require using a special error code to
> > signal that a new assignment is required as the signal would be the
> > provided group state; 2) it removes one round trip when a client side
> > assignor is used. The downside is that it makes the heartbeat
> > response's definition quite large. I recall that I went with the
> > current approach due to this.
> >
> > Providing the group state in the heartbeat response is appealing. What
> > do you guys think?
> >
> > Best,
> > David
> >
> > On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io> wrote:
> > >
> > > Hi Guozhang,
> > >
> > > 1. I have added a reference to the relevant chapter instead of
> > > repeating the whole thing. Does that work for you?
> > >
> > > 2. The "Rebalance Triggers" section you are referring to is about when
> > > a rebalance should be triggered for the non-upgraded members using the
> > > old protocol. The section mentions that a rebalance must be triggered
> > > when a new assignment is installed. This implies that the group epoch
> > > was updated either by a native member or a non-upgraded member. For
> > > the latter, the JoinGroup request would be the trigger. I have added a
> > > reference to the relevant chapter in the "JoinGroup Handling" section
> > > as well. Does that make sense?
> > >
> > > Thanks,
> > > David
> > >
> > > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > >
> > > > Hello David,
> > > >
> > > > Alright I think that's sufficient. Just to make that clear in the doc,
> > > > could we update:
> > > >
> > > > 1) the heartbeat request handling section, stating when coordinator
> > will
> > > > trigger rebalance based on the HB's member metadata / reason?
> > > > 2) the "Rebalance Triggers" section to include what we described in
> > "Group
> > > > Epoch - Trigger a rebalance" section as well?
> > > >
> > > >
> > > > Guozhang
> > > >
> > > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot <djacot@confluent.io.invalid
> > >
> > > > wrote:
> > > >
> > > > > Hi Guozhang,
> > > > >
> > > > > I thought that the assignor will always be consulted when the next
> > > > > heartbeat request is constructed. In other words,
> > > > > `PartitionAssignor#metadata` will be called for every heartbeat. This
> > > > > gives the opportunity for the assignor to enforce a rebalance by
> > > > > setting the reason to a non-zero value or by changing the bytes. Do
> > > > > you think that this is not sufficient? Are you concerned by the
> > delay?
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > > > >
> > > > > > Hello David,
> > > > > >
> > > > > > One of Jun's comments make me thinking:
> > > > > >
> > > > > > ```
> > > > > > In this case, a new assignment is triggered by the client side
> > > > > > assignor. When constructing the HB, the consumer will always
> > consult
> > > > > > the client side assignor and propagate the information to the group
> > > > > > coordinator. In other words, we don't expect users to call
> > > > > > Consumer#enforceRebalance anymore.
> > > > > > ```
> > > > > >
> > > > > > As I looked at the current PartitionAssignor's interface, we
> > actually do
> > > > > > not have a way yet to instruct how to construct the next HB
> > request, e.g.
> > > > > > when the assignor wants to enforce a new rebalance with a new
> > assignment,
> > > > > > we'd need some customizable APIs inside the PartitionAssignor to
> > indicate
> > > > > > the next HB telling broker about so. WDYT about adding such an API
> > on the
> > > > > > PartitionAssignor?
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > <dj...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > > > Hi Jun,
> > > > > > >
> > > > > > > I have updated the KIP to include your feedback. I have also
> > tried to
> > > > > > > clarify the parts which were not cleared.
> > > > > > >
> > > > > > > Best,
> > > > > > > David
> > > > > > >
> > > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io>
> > > > > wrote:
> > > > > > > >
> > > > > > > > Hi Jun,
> > > > > > > >
> > > > > > > > Thanks for your feedback. Let me start by answering your
> > questions
> > > > > > > > inline and I will update the KIP next week.
> > > > > > > >
> > > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> > seem to
> > > > > be
> > > > > > > fewer
> > > > > > > > > RPCs during rebalance and more efficient support of
> > wildcard. A few
> > > > > > > > > comments below.
> > > > > > > >
> > > > > > > > I would also add that the KIP removes the global sync barrier
> > in the
> > > > > > > > protocol which is essential to improve group stability and
> > > > > > > > scalability, and the KIP also simplifies the client by moving
> > most of
> > > > > > > > the logic to the server side.
> > > > > > > >
> > > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> > rolling
> > > > > > > changing
> > > > > > > > > of the partition assignor in the consumers?
> > > > > > > >
> > > > > > > > Definitely. The group coordinator will use the assignor used
> > by a
> > > > > > > > majority of the members. This allows the group to move from one
> > > > > > > > assignor to another by a roll. This is explained in the
> > Assignor
> > > > > > > > Selection chapter.
> > > > > > > >
> > > > > > > > > 30.2 For each field, could you explain whether it's required
> > in
> > > > > every
> > > > > > > > > request or the scenarios when it needs to be filled? For
> > example,
> > > > > it's
> > > > > > > not
> > > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > > > >
> > > > > > > > The client is expected to set those fields in case of a
> > connection
> > > > > > > > issue (e.g. timeout) or when the fields have changed since the
> > last
> > > > > > > > HB. The server populates those fields as long as the member is
> > not
> > > > > > > > fully reconciled - the member should acknowledge that it has
> > the
> > > > > > > > expected epoch and assignment. I will clarify this in the KIP.
> > > > > > > >
> > > > > > > > > 31. In the current consumer protocol, the rack affinity
> > between the
> > > > > > > client
> > > > > > > > > and the broker is only considered during fetching, but not
> > during
> > > > > > > assigning
> > > > > > > > > partitions to consumers. Sometimes, once the assignment is
> > made,
> > > > > there
> > > > > > > is
> > > > > > > > > no opportunity for read affinity because no replicas of
> > assigned
> > > > > > > partitions
> > > > > > > > > are close to the member. I am wondering if we should use this
> > > > > > > opportunity
> > > > > > > > > to address this by including rack in GroupMember.
> > > > > > > >
> > > > > > > > That's an interesting idea. I don't see any issue with adding
> > the
> > > > > rack
> > > > > > > > to the members. I will do so.
> > > > > > > >
> > > > > > > > > 32. On the metric side, often, it's useful to know how busy
> > a group
> > > > > > > > > coordinator is. By moving the event loop model, it seems
> > that we
> > > > > could
> > > > > > > add
> > > > > > > > > a metric that tracks the fraction of the time the event loop
> > is
> > > > > doing
> > > > > > > the
> > > > > > > > > actual work.
> > > > > > > >
> > > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > >
> > > > > > > > > 33. Could we add a section on coordinator failover handling?
> > For
> > > > > > > example,
> > > > > > > > > does it need to trigger the check if any group with the
> > wildcard
> > > > > > > > > subscription now has a new matching topic?
> > > > > > > >
> > > > > > > > Sure. When the new group coordinator takes over, it has to:
> > > > > > > > * Setup the session timeouts.
> > > > > > > > * Trigger a new assignment if a client side assignor is used.
> > We
> > > > > don't
> > > > > > > > store the information about the member selected to run the
> > assignment
> > > > > > > > so we have to start a new one.
> > > > > > > > * Update the topics metadata, verify the wildcard
> > subscriptions, and
> > > > > > > > trigger a rebalance if needed.
> > > > > > > >
> > > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> > epoch
> > > > > > > field
> > > > > > > > > reflects? For example, does the epoch in
> > ConsumerGroupMetadataValue
> > > > > > > reflect
> > > > > > > > > the latest group epoch? What about the one in
> > > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >
> > > > > > > > Sure. I will clarify that but it is always the latest group
> > epoch.
> > > > > > > > When the group state is updated, the group epoch is bumped so
> > we use
> > > > > > > > that one for all the change records related to the update.
> > > > > > > >
> > > > > > > > > 35. "the group coordinator will ensure that the following
> > > > > invariants
> > > > > > > are
> > > > > > > > > met: ... All members exists." It's possible for a member not
> > to
> > > > > get any
> > > > > > > > > assigned partitions, right?
> > > > > > > >
> > > > > > > > That's right. Here I meant that the members provided by the
> > assignor
> > > > > > > > in the assignment must exist in the group. The assignor can
> > not make
> > > > > > > > up new member ids.
> > > > > > > >
> > > > > > > > > 36. "He can rejoins the group with a member epoch equals to
> > 0":
> > > > > When
> > > > > > > would
> > > > > > > > > a consumer rejoin and what member id would be used?
> > > > > > > >
> > > > > > > > A member is expected to abandon all its partitions and rejoins
> > when
> > > > > it
> > > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > > > > > > coordinator will have removed the member from the group. The
> > member
> > > > > > > > can rejoin the group with the same member id but with 0 as
> > epoch. Let
> > > > > > > > me see if I can clarify this in the KIP.
> > > > > > > >
> > > > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > > > reassignment
> > > > > > > > > by either providing a non-zero reason or by updating the
> > assignor
> > > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > deprecation
> > > > > of
> > > > > > > > > Consumer#enforeRebalance.
> > > > > > > >
> > > > > > > > In this case, a new assignment is triggered by the client side
> > > > > > > > assignor. When constructing the HB, the consumer will always
> > consult
> > > > > > > > the client side assignor and propagate the information to the
> > group
> > > > > > > > coordinator. In other words, we don't expect users to call
> > > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > >
> > > > > > > > > 38. The reassignment examples are nice. But the section
> > seems to
> > > > > have
> > > > > > > > > multiple typos.
> > > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> > gets into
> > > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> > into
> > > > > > > epoch=3,
> > > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > > > partitions=[foo-0].
> > > > > > > >
> > > > > > > > Sorry for that! I will revise them.
> > > > > > > >
> > > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> > from
> > > > > any
> > > > > > > old
> > > > > > > > > version to new one?
> > > > > > > >
> > > > > > > > We will support upgrading from the consumer protocol version 3,
> > > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
> > earliest
> > > > > > > > version is unknown at the moment. This is explained in the
> > migration
> > > > > > > > plan chapter.
> > > > > > > >
> > > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> > based on
> > > > > it
> > > > > > > > next week.
> > > > > > > >
> > > > > > > > Best,
> > > > > > > > David
> > > > > > > >
> > > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > <ju...@confluent.io.invalid>
> > > > > wrote:
> > > > > > > > >
> > > > > > > > > Hi, David,
> > > > > > > > >
> > > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> > seem to
> > > > > be
> > > > > > > fewer
> > > > > > > > > RPCs during rebalance and more efficient support of
> > wildcard. A few
> > > > > > > > > comments below.
> > > > > > > > >
> > > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> > rolling
> > > > > > > changing
> > > > > > > > > of the partition assignor in the consumers?
> > > > > > > > > 30.2 For each field, could you explain whether it's required
> > in
> > > > > every
> > > > > > > > > request or the scenarios when it needs to be filled? For
> > example,
> > > > > it's
> > > > > > > not
> > > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > > > > >
> > > > > > > > > 31. In the current consumer protocol, the rack affinity
> > between the
> > > > > > > client
> > > > > > > > > and the broker is only considered during fetching, but not
> > during
> > > > > > > assigning
> > > > > > > > > partitions to consumers. Sometimes, once the assignment is
> > made,
> > > > > there
> > > > > > > is
> > > > > > > > > no opportunity for read affinity because no replicas of
> > assigned
> > > > > > > partitions
> > > > > > > > > are close to the member. I am wondering if we should use this
> > > > > > > opportunity
> > > > > > > > > to address this by including rack in GroupMember.
> > > > > > > > >
> > > > > > > > > 32. On the metric side, often, it's useful to know how busy
> > a group
> > > > > > > > > coordinator is. By moving the event loop model, it seems
> > that we
> > > > > could
> > > > > > > add
> > > > > > > > > a metric that tracks the fraction of the time the event loop
> > is
> > > > > doing
> > > > > > > the
> > > > > > > > > actual work.
> > > > > > > > >
> > > > > > > > > 33. Could we add a section on coordinator failover handling?
> > For
> > > > > > > example,
> > > > > > > > > does it need to trigger the check if any group with the
> > wildcard
> > > > > > > > > subscription now has a new matching topic?
> > > > > > > > >
> > > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> > epoch
> > > > > > > field
> > > > > > > > > reflects? For example, does the epoch in
> > ConsumerGroupMetadataValue
> > > > > > > reflect
> > > > > > > > > the latest group epoch? What about the one in
> > > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > >
> > > > > > > > > 35. "the group coordinator will ensure that the following
> > > > > invariants
> > > > > > > are
> > > > > > > > > met: ... All members exists." It's possible for a member not
> > to
> > > > > get any
> > > > > > > > > assigned partitions, right?
> > > > > > > > >
> > > > > > > > > 36. "He can rejoins the group with a member epoch equals to
> > 0":
> > > > > When
> > > > > > > would
> > > > > > > > > a consumer rejoin and what member id would be used?
> > > > > > > > >
> > > > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > > > reassignment
> > > > > > > > > by either providing a non-zero reason or by updating the
> > assignor
> > > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > deprecation
> > > > > of
> > > > > > > > > Consumer#enforeRebalance.
> > > > > > > > >
> > > > > > > > > 38. The reassignment examples are nice. But the section
> > seems to
> > > > > have
> > > > > > > > > multiple typos.
> > > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> > gets into
> > > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> > into
> > > > > > > epoch=3,
> > > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > > > partitions=[foo-0].
> > > > > > > > >
> > > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> > from
> > > > > any
> > > > > > > old
> > > > > > > > > version to new one?
> > > > > > > > >
> > > > > > > > > Thanks,
> > > > > > > > >
> > > > > > > > > Jun
> > > > > > > > >
> > > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hi all,
> > > > > > > > > >
> > > > > > > > > > The KIP states that we will re-implement the coordinator in
> > > > > Java. I
> > > > > > > > > > discussed this offline with a few folks and folks are
> > concerned
> > > > > that
> > > > > > > > > > we could introduce many regressions in the old protocol if
> > we do
> > > > > so.
> > > > > > > > > > Therefore, I am going to remove this statement from the
> > KIP. It
> > > > > is an
> > > > > > > > > > implementation detail after all so it does not have to be
> > > > > decided at
> > > > > > > > > > this stage. We will likely start by trying to refactor the
> > > > > current
> > > > > > > > > > implementation as a first step.
> > > > > > > > > >
> > > > > > > > > > Cheers,
> > > > > > > > > > David
> > > > > > > > > >
> > > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > djacot@confluent.io
> > > > > >
> > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > Hi Luke,
> > > > > > > > > > >
> > > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > > > reconciling,
> > > > > > > > > > stable,
> > > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > right?
> > > > > > > > > > >
> > > > > > > > > > > This sentence does not refer to those group states but
> > rather
> > > > > to a
> > > > > > > > > > > state machine replication (SMR). This refers to the
> > entire
> > > > > state of
> > > > > > > > > > > group coordinator which is replicated via the log layer.
> > I will
> > > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > >
> > > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
> > as an
> > > > > event
> > > > > > > > > > loop"?
> > > > > > > > > > >
> > > > > > > > > > > The idea is to follow a model similar to the new quorum
> > > > > > > controller. We
> > > > > > > > > > > will have N threads to process events. Each
> > __consumer_offsets
> > > > > > > > > > > partition is assigned to a unique thread and all the
> > events
> > > > > (e.g.
> > > > > > > > > > > requests, callbacks, etc.) are processed by this thread.
> > This
> > > > > > > simplify
> > > > > > > > > > > concurrency and will enable us to do simulation testing
> > for the
> > > > > > > group
> > > > > > > > > > > coordinator.
> > > > > > > > > > >
> > > > > > > > > > > > 1.3. Why do we need a state machine per
> > *__consumer_offsets*
> > > > > > > > > > partitions?
> > > > > > > > > > > > Not a state machine "per consumer group" owned by a
> > group
> > > > > > > coordinator?
> > > > > > > > > > For
> > > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > groups, and
> > > > > > > both
> > > > > > > > > > exist in
> > > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> > for it,
> > > > > or
> > > > > > > 2?
> > > > > > > > > > >
> > > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > > > > > > > > >
> > > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> > number
> > > > > of
> > > > > > > threads
> > > > > > > > > > used
> > > > > > > > > > > > to run the state machines. But I'm wondering if the
> > purpose
> > > > > of
> > > > > > > the
> > > > > > > > > > threads
> > > > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > partitions?), and no heavy computation, why should we
> > need
> > > > > > > > > > multi-threads
> > > > > > > > > > > > here?
> > > > > > > > > > >
> > > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > > > > processing as
> > > > > > > the
> > > > > > > > > > > computation could be heavy.
> > > > > > > > > > >
> > > > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > > > session
> > > > > > > > > > timeout not
> > > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> > min/max
> > > > > > > session
> > > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > > > >
> > > > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout
> > to
> > > > > detect
> > > > > > > client
> > > > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > minimum
> > > > > > > session
> > > > > > > > > > timeout.
> > > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > maximum
> > > > > > > session
> > > > > > > > > > timeout.
> > > > > > > > > > >
> > > > > > > > > > > This is indeed a mistake. The default session timeout
> > should
> > > > > be 45s
> > > > > > > > > > > (the current default).
> > > > > > > > > > >
> > > > > > > > > > > > 2.2. The default server side assignor are [range,
> > uniform],
> > > > > > > which means
> > > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> > why not
> > > > > > > uniform
> > > > > > > > > > one? I
> > > > > > > > > > > > thought usually users will choose uniform assignor
> > (former
> > > > > sticky
> > > > > > > > > > assinor)
> > > > > > > > > > > > for better evenly distribution. Any other reason we
> > choose
> > > > > range
> > > > > > > > > > assignor
> > > > > > > > > > > > as default?
> > > > > > > > > > > > group.consumer.assignors List range, uniform The
> > server side
> > > > > > > assignors.
> > > > > > > > > > >
> > > > > > > > > > > The order on the server side has no influence because the
> > > > > client
> > > > > > > must
> > > > > > > > > > > chose the selector that he wants to use. There is no
> > default
> > > > > in the
> > > > > > > > > > > current proposal. If the assignor is not specified by the
> > > > > client,
> > > > > > > the
> > > > > > > > > > > request is rejected. The default client value for
> > > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > > > > >
> > > > > > > > > > > Thanks for your very good comments, Luke. I hope that my
> > > > > answers
> > > > > > > help
> > > > > > > > > > > to clarify things. I will update the KIP as well based
> > on your
> > > > > > > > > > > feedback.
> > > > > > > > > > >
> > > > > > > > > > > Cheers,
> > > > > > > > > > > David
> > > > > > > > > > >
> > > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > showuon@gmail.com>
> > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > Hi David,
> > > > > > > > > > > >
> > > > > > > > > > > > Thanks for the update.
> > > > > > > > > > > >
> > > > > > > > > > > > Some more questions:
> > > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > > > > > > The new group coordinator will have a state machine
> > per
> > > > > > > > > > > > *__consumer_offsets* partitions, where each state
> > machine is
> > > > > > > modelled
> > > > > > > > > > as an
> > > > > > > > > > > > event loop. Those state machines will be executed in
> > > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > > > >
> > > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > > > reconciling,
> > > > > > > > > > stable,
> > > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > right?
> > > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
> > as an
> > > > > event
> > > > > > > > > > loop"?
> > > > > > > > > > > > 1.3. Why do we need a state machine per
> > *__consumer_offsets*
> > > > > > > > > > partitions?
> > > > > > > > > > > > Not a state machine "per consumer group" owned by a
> > group
> > > > > > > coordinator?
> > > > > > > > > > For
> > > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > groups, and
> > > > > > > both
> > > > > > > > > > exist in
> > > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> > for it,
> > > > > or
> > > > > > > 2?
> > > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> > number
> > > > > of
> > > > > > > threads
> > > > > > > > > > used
> > > > > > > > > > > > to run the state machines. But I'm wondering if the
> > purpose
> > > > > of
> > > > > > > the
> > > > > > > > > > threads
> > > > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > partitions?), and no heavy computation, why should we
> > need
> > > > > > > > > > multi-threads
> > > > > > > > > > > > here?
> > > > > > > > > > > >
> > > > > > > > > > > > 2. For the default value in the new configs:
> > > > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > > > session
> > > > > > > > > > timeout not
> > > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> > min/max
> > > > > > > session
> > > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > > > >
> > > > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout
> > to
> > > > > detect
> > > > > > > client
> > > > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > minimum
> > > > > > > session
> > > > > > > > > > timeout.
> > > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > maximum
> > > > > > > session
> > > > > > > > > > timeout.
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > 2.2. The default server side assignor are [range,
> > uniform],
> > > > > > > which means
> > > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> > why not
> > > > > > > uniform
> > > > > > > > > > one? I
> > > > > > > > > > > > thought usually users will choose uniform assignor
> > (former
> > > > > sticky
> > > > > > > > > > assinor)
> > > > > > > > > > > > for better evenly distribution. Any other reason we
> > choose
> > > > > range
> > > > > > > > > > assignor
> > > > > > > > > > > > as default?
> > > > > > > > > > > > group.consumer.assignors List range, uniform The
> > server side
> > > > > > > assignors.
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > Thank you.
> > > > > > > > > > > > Luke
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > showuon@gmail.com
> > > > > >
> > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > > >
> > > > > > > > > > > > > I have some thoughts about Kafka Connect integrating
> > with
> > > > > > > KIP-848,
> > > > > > > > > > but I
> > > > > > > > > > > > > think we should have a separate discussion thread
> > for the
> > > > > Kafka
> > > > > > > > > > Connect
> > > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> > Rebalance
> > > > > > > Protocol
> > > > > > > > > > [1],
> > > > > > > > > > > > > and let this discussion thread focus on consumer
> > rebalance
> > > > > > > protocol,
> > > > > > > > > > WDYT?
> > > > > > > > > > > > >
> > > > > > > > > > > > > [1]
> > > > > > > > > > > > >
> > > > > > > > > >
> > > > > > >
> > > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > >
> > > > > > > > > > > > > Thank you.
> > > > > > > > > > > > > Luke
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > wrote:
> > > > > > > > > > > > >
> > > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
> > like
> > > > > there's
> > > > > > > > > > agreement on
> > > > > > > > > > > > >> using separate APIs for Connect. I would revisit
> > the doc
> > > > > and
> > > > > > > see
> > > > > > > > > > what
> > > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > > > >>
> > > > > > > > > > > > >> Thanks!
> > > > > > > > > > > > >> Sagar.
> > > > > > > > > > > > >>
> > > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > >>
> > > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > Thanks for the feedback and the document. That's
> > really
> > > > > > > helpful. I
> > > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > Overall, it seems to me that both Connect and the
> > > > > Consumer
> > > > > > > could
> > > > > > > > > > share
> > > > > > > > > > > > >> > the same underlying "engine". The main difference
> > is
> > > > > that
> > > > > > > the
> > > > > > > > > > Consumer
> > > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > Connect
> > > > > assigns
> > > > > > > tasks
> > > > > > > > > > to
> > > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> > different
> > > > > > > resource
> > > > > > > > > > types
> > > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> > dedicated
> > > > > APIs
> > > > > > > > > > would be
> > > > > > > > > > > > >> > similar to the new ones but different on the
> > > > > > > content/resources and
> > > > > > > > > > > > >> > they would rely on the same engine on the
> > coordinator
> > > > > side.
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > I personally lean towards 2) because I am not a
> > fan of
> > > > > > > > > > overcharging
> > > > > > > > > > > > >> > APIs to serve different purposes. That being
> > said, I am
> > > > > not
> > > > > > > > > > opposed to
> > > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > I think that we can continue to discuss it here
> > for now
> > > > > in
> > > > > > > order
> > > > > > > > > > to
> > > > > > > > > > > > >> > ensure that this KIP is compatible with what we
> > will do
> > > > > for
> > > > > > > > > > Connect in
> > > > > > > > > > > > >> > the future.
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > Best,
> > > > > > > > > > > > >> > David
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > > > > > djacot@confluent.io>
> > > > > > > > > > wrote:
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > I am back from vacation. I will go through and
> > address
> > > > > > > your
> > > > > > > > > > comments
> > > > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > > >> > > David
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > > > > cooperative
> > > > > > > > > > rebalancing
> > > > > > > > > > > > >> > making it
> > > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > describes how
> > > > > > > member
> > > > > > > > > > epoch
> > > > > > > > > > > > >> > should
> > > > > > > > > > > > >> > > > advance in step with the group epoch and
> > assignment
> > > > > > > epoch
> > > > > > > > > > values. I
> > > > > > > > > > > > >> > think
> > > > > > > > > > > > >> > > > that this is a good idea for the reasons
> > described
> > > > > in
> > > > > > > the
> > > > > > > > > > KIP. When
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > protocol is incrementally assigning
> > partitions to a
> > > > > > > worker,
> > > > > > > > > > what
> > > > > > > > > > > > >> member
> > > > > > > > > > > > >> > > > epoch does each incremental assignment use?
> > Are
> > > > > member
> > > > > > > epochs
> > > > > > > > > > > > >> re-used,
> > > > > > > > > > > > >> > and
> > > > > > > > > > > > >> > > > a single member epoch can correspond to
> > multiple
> > > > > > > different
> > > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to
> > the
> > > > > group
> > > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > > >> > > > not, should custom client-side assignor
> > > > > implementations
> > > > > > > > > > interact
> > > > > > > > > > > > >> with
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > Reason field, and how is its common meaning
> > agreed
> > > > > > > upon? If
> > > > > > > > > > so, what
> > > > > > > > > > > > >> > is the
> > > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > including
> > > > > such
> > > > > > > > > > functionality
> > > > > > > > > > > > >> > in the
> > > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> > "Thanks to
> > > > > > > this, the
> > > > > > > > > > input
> > > > > > > > > > > > >> of
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > client side assignor is entirely driven by
> > the group
> > > > > > > > > > coordinator.
> > > > > > > > > > > > >> The
> > > > > > > > > > > > >> > > > consumer is no longer responsible for
> > maintaining
> > > > > any
> > > > > > > state
> > > > > > > > > > besides
> > > > > > > > > > > > >> its
> > > > > > > > > > > > >> > > > assigned partitions." Does this mean that the
> > > > > > > client-side
> > > > > > > > > > assignor
> > > > > > > > > > > > >> MAY
> > > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> > (such as
> > > > > > > partition
> > > > > > > > > > > > >> > throughput,
> > > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or
> > that
> > > > > > > additional
> > > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > > > > > > >> > > > for
> > > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > > >> > > > and
> > > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > that
> > > > > > > > > > seem to
> > > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> > these two
> > > > > > > > > > implementations
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > be
> > > > > > > > > > > > >> > > > unified? This would serve to promote feature
> > parity
> > > > > of
> > > > > > > > > > server-side
> > > > > > > > > > > > >> and
> > > > > > > > > > > > >> > > > client-side assignors, and would also
> > facilitate
> > > > > > > operational
> > > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > > >> > > > certain situations. For example, if a
> > server-side
> > > > > > > assignor
> > > > > > > > > > has some
> > > > > > > > > > > > >> > poor
> > > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> > patched
> > > > > > > assignor to
> > > > > > > > > > the
> > > > > > > > > > > > >> > client
> > > > > > > > > > > > >> > > > and switching one consumer group to a
> > client-side
> > > > > > > assignor
> > > > > > > > > > may be
> > > > > > > > > > > > >> > faster
> > > > > > > > > > > > >> > > > and less risky than patching all of the
> > brokers.
> > > > > With
> > > > > > > the
> > > > > > > > > > currently
> > > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > > > reimplementation
> > > > > > > would
> > > > > > > > > > have
> > > > > > > > > > > > >> to be
> > > > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> > > > > > > significantly,
> > > > > > > > > > then it
> > > > > > > > > > > > >> is
> > > > > > > > > > > > >> > > > possible that a reimplementation would not be
> > > > > possible.
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > --
> > > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > I created a confluence page to discuss how
> > Connect
> > > > > > > would
> > > > > > > > > > need to
> > > > > > > > > > > > >> > change
> > > > > > > > > > > > >> > > > > based on the new rebalance protocol. Here's
> > the
> > > > > page:
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > >
> > > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > It's also pretty longish and I have tried
> > to keep
> > > > > a
> > > > > > > format
> > > > > > > > > > > > >> similar to
> > > > > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also,
> > do you
> > > > > > > think this
> > > > > > > > > > > > >> should
> > > > > > > > > > > > >> > be
> > > > > > > > > > > > >> > > > > moved to a separate discussion thread or is
> > this
> > > > > one
> > > > > > > fine?
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> > Streams.
> > > > > > > Sure, I
> > > > > > > > > > would do
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang
> > Wang <
> > > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > >> It would be great if you could come back
> > with
> > > > > some
> > > > > > > > > > analysis on
> > > > > > > > > > > > >> > how to
> > > > > > > > > > > > >> > > > > >> implement the Connect side integration
> > with
> > > > > the new
> > > > > > > > > > protocol;
> > > > > > > > > > > > >> so
> > > > > > > > > > > > >> > far
> > > > > > > > > > > > >> > > > > >> besides leveraging on the new "protocol
> > type"
> > > > > we
> > > > > > > did not
> > > > > > > > > > yet
> > > > > > > > > > > > >> think
> > > > > > > > > > > > >> > > > > through
> > > > > > > > > > > > >> > > > > >> the Connect side implementations. For
> > Streams
> > > > > > > here's a
> > > > > > > > > > draft of
> > > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > >
> > > > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> > reason I
> > > > > thought
> > > > > > > > > > connect can
> > > > > > > > > > > > >> > also fit
> > > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > > >> > > > > >> > this new scheme is that even today the
> > > > > connect
> > > > > > > uses a
> > > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to
> > empower
> > > > > > > > > > rebalances of
> > > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > WorkerCoordinator sets
> > > > > the
> > > > > > > > > > > > >> protocolType()
> > > > > > > > > > > > >> > to
> > > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> > plumbing
> > > > > into
> > > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> > I think the changes to support connect
> > would
> > > > > be
> > > > > > > > > > similar at a
> > > > > > > > > > > > >> > high
> > > > > > > > > > > > >> > > > > level
> > > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > > >> > > > > >> > the changes in streams mainly because
> > of the
> > > > > > > Client
> > > > > > > > > > side
> > > > > > > > > > > > >> > assignors
> > > > > > > > > > > > >> > > > > being
> > > > > > > > > > > > >> > > > > >> > used in both. At an implementation
> > level, we
> > > > > > > might
> > > > > > > > > > need to
> > > > > > > > > > > > >> make
> > > > > > > > > > > > >> > a lot
> > > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > >> > > > > >> > changes to get onto this new assignment
> > > > > protocol
> > > > > > > like
> > > > > > > > > > > > >> enhancing
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > SyncGroup and
> > > > > > > using
> > > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> > streams (or
> > > > > > > there
> > > > > > > > > > might be
> > > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> > analysis of
> > > > > the
> > > > > > > same
> > > > > > > > > > and we
> > > > > > > > > > > > >> > can have
> > > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > >> > > > > >> > separate discussion thread for that as
> > that
> > > > > would
> > > > > > > > > > derail this
> > > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
> > good to
> > > > > you.
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David
> > Jacot
> > > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > `Assignment#error`.
> > > > > > > Sure, I
> > > > > > > > > > can
> > > > > > > > > > > > >> > mention it.
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from
> > his
> > > > > current
> > > > > > > > > > assignment
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > his
> > > > > > > > > > > > >> > > > > >> > > target assignment when he can move to
> > > > > epoch 3.
> > > > > > > When
> > > > > > > > > > that
> > > > > > > > > > > > >> > happens,
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > member assignment is updated and
> > persisted
> > > > > > > with all
> > > > > > > > > > its
> > > > > > > > > > > > >> > assigned
> > > > > > > > > > > > >> > > > > >> > > partitions even if they are not all
> > revoked
> > > > > > > yet. In
> > > > > > > > > > other
> > > > > > > > > > > > >> > words, the
> > > > > > > > > > > > >> > > > > >> > > member assignment becomes the target
> > > > > > > assignment.
> > > > > > > > > > This is
> > > > > > > > > > > > >> > basically
> > > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> > write all
> > > > > the
> > > > > > > > > > changes to
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > log.
> > > > > > > > > > > > >> > > > > The
> > > > > > > > > > > > >> > > > > >> > > examples are based on the persisted
> > state
> > > > > so I
> > > > > > > > > > understand
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> > improve
> > > > > this in
> > > > > > > the
> > > > > > > > > > > > >> > description.
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse
> > the
> > > > > > > protocol
> > > > > > > > > > with a
> > > > > > > > > > > > >> > client side
> > > > > > > > > > > > >> > > > > >> > > assignor if it fits in the protocol.
> > The
> > > > > > > assignment
> > > > > > > > > > is
> > > > > > > > > > > > >> about
> > > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> > > > > Connect
> > > > > > > fit
> > > > > > > > > > into this?
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> > Sagar <
> > > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
> > minor
> > > > > > > observations:
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error section
> > in
> > > > > Client
> > > > > > > Side
> > > > > > > > > > mode
> > > > > > > > > > > > >> > Assignment
> > > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case, the
> > > > > client
> > > > > > > side
> > > > > > > > > > assignor
> > > > > > > > > > > > >> can
> > > > > > > > > > > > >> > > > > return
> > > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > > >> > > > > >> > > > error to the group coordinator`.
> > In this
> > > > > > > case are
> > > > > > > > > > you
> > > > > > > > > > > > >> > referring to
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > AssignmentError
> > > > > that's
> > > > > > > > > > listed down
> > > > > > > > > > > > >> > towards
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would make
> > sense
> > > > > to
> > > > > > > > > > mention this
> > > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I
> > have a
> > > > > > > slight
> > > > > > > > > > > > >> confusion,
> > > > > > > > > > > > >> > not
> > > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
> > this step:
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> > coordinator
> > > > > > > > > > transitions him
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > epoch
> > > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> > > > > persists
> > > > > > > the
> > > > > > > > > > change and
> > > > > > > > > > > > >> > reply.
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > partitions=[foo-0,
> > > > > > > foo-1]
> > > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > partitions=[foo-2]
> > > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > partitions=[foo-1]
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions
> > to
> > > > > epoch 3
> > > > > > > but
> > > > > > > > > > cannot
> > > > > > > > > > > > >> get
> > > > > > > > > > > > >> > foo-1
> > > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C
> > can't
> > > > > get
> > > > > > > the
> > > > > > > > > > foo-1
> > > > > > > > > > > > >> > partition
> > > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > > >> > > > > >> > > > based on the description above, it
> > seems
> > > > > it
> > > > > > > > > > already has
> > > > > > > > > > > > >> it.
> > > > > > > > > > > > >> > Do you
> > > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > > >> > > > > >> > > > would be better to remove it and
> > > > > populate it
> > > > > > > only
> > > > > > > > > > when it
> > > > > > > > > > > > >> > actually
> > > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> > places, so
> > > > > have
> > > > > > > I
> > > > > > > > > > > > >> understood it
> > > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
> > out of
> > > > > scope
> > > > > > > of
> > > > > > > > > > this
> > > > > > > > > > > > >> > discussion,
> > > > > > > > > > > > >> > > > > but
> > > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > probably be
> > > > > > > running in
> > > > > > > > > > client
> > > > > > > > > > > > >> > side
> > > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > > >> > > > > >> > > > even on the new rebalance protocol
> > as it
> > > > > has
> > > > > > > its
> > > > > > > > > > own
> > > > > > > > > > > > >> Custom
> > > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
> > David
> > > > > Jacot
> > > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
> > move
> > > > > forward
> > > > > > > with
> > > > > > > > > > > > >> > specialized API
> > > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> > generic API.
> > > > > For
> > > > > > > > > > Connect, we
> > > > > > > > > > > > >> > can apply
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > reuse/share the
> > > > > core
> > > > > > > > > > > > >> > implementation on
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> > registry, I
> > > > > > > think
> > > > > > > > > > that we
> > > > > > > > > > > > >> > should
> > > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> > simple
> > > > > > > > > > membership/leader
> > > > > > > > > > > > >> > election.
> > > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM
> > Ismael
> > > > > > > Juma <
> > > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > java.util.regex.Pattern
> > > > > vs
> > > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > >> > > > > >> > > > > > document the differences in
> > more
> > > > > detail
> > > > > > > before
> > > > > > > > > > > > >> deciding
> > > > > > > > > > > > >> > one
> > > > > > > > > > > > >> > > > > way
> > > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > > >> they
> > > > > > > > > > > > >> > expect
> > > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If we
> > are
> > > > > doing
> > > > > > > > > > something
> > > > > > > > > > > > >> > different,
> > > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> > overload
> > > > > with
> > > > > > > our own
> > > > > > > > > > > > >> Pattern
> > > > > > > > > > > > >> > class
> > > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> > re2j's at
> > > > > this
> > > > > > > > > > point).
> > > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> > major new
> > > > > > > protocol
> > > > > > > > > > should
> > > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > > > > > > recreation case
> > > > > > > > > > > > >> > correctly.
> > > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> > agree with
> > > > > > > David
> > > > > > > > > > that we'd
> > > > > > > > > > > > >> > want to
> > > > > > > > > > > > >> > > > > add
> > > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols that
> > don't
> > > > > > > have it
> > > > > > > > > > yet and
> > > > > > > > > > > > >> > that we
> > > > > > > > > > > > >> > > > > can
> > > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > > > >> > > > > >> > > > > > focus on the internals versus
> > adding
> > > > > new
> > > > > > > APIs
> > > > > > > > > > to the
> > > > > > > > > > > > >> > Java
> > > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs is
> > > > > required
> > > > > > > for
> > > > > > > > > > > > >> reasonable
> > > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about
> > the
> > > > > > > coordinator
> > > > > > > > > > storing
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > > > > centralized
> > > > > > > in the
> > > > > > > > > > > > >> metadata
> > > > > > > > > > > > >> > log for
> > > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > visibility,
> > > > > > > consistency,
> > > > > > > > > > > > >> etc.).
> > > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> > deleting
> > > > > > > configs in a
> > > > > > > > > > way
> > > > > > > > > > > > >> that
> > > > > > > > > > > > >> > they
> > > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property for
> > modern
> > > > > > > systems
> > > > > > > > > > is to
> > > > > > > > > > > > >> > minimize
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> > scenarios.
> > > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47
> > PM David
> > > > > > > Jacot
> > > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers
> > are
> > > > > below:
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> > especially
> > > > > > > the last
> > > > > > > > > > step
> > > > > > > > > > > > >> when
> > > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > protocol, in
> > > > > which
> > > > > > > we
> > > > > > > > > > would
> > > > > > > > > > > > >> have a
> > > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > > > protocols /
> > > > > > > rpcs
> > > > > > > > > > are
> > > > > > > > > > > > >> used
> > > > > > > > > > > > >> > by
> > > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> > could
> > > > > > > "mimic"
> > > > > > > > > > the old
> > > > > > > > > > > > >> > behavior
> > > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
> > we
> > > > > need to
> > > > > > > > > > present
> > > > > > > > > > > > >> about.
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a new
> > > > > version
> > > > > > > of KIP
> > > > > > > > > > which
> > > > > > > > > > > > >> > includes
> > > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> > > > > "Supporting
> > > > > > > > > > Online
> > > > > > > > > > > > >> > Consumer
> > > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > > > > Deprecation,
> > > > > > > and
> > > > > > > > > > Migration
> > > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
> > few cases
> > > > > > > now but
> > > > > > > > > > the
> > > > > > > > > > > > >> > overall idea
> > > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > understandable.
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids.
> > So
> > > > > far as
> > > > > > > > > > KIP-516 the
> > > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> > client,
> > > > > but
> > > > > > > they
> > > > > > > > > > are not
> > > > > > > > > > > > >> > exposed
> > > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
> > the
> > > > > > > question is,
> > > > > > > > > > first
> > > > > > > > > > > > >> > should we
> > > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> > the
> > > > > names
> > > > > > > -> ids
> > > > > > > > > > mapping
> > > > > > > > > > > > >> > itself
> > > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > existing
> > > > > RPCs
> > > > > > > and the
> > > > > > > > > > new
> > > > > > > > > > > > >> RPCs
> > > > > > > > > > > > >> > with
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> > consider
> > > > > > > exposing
> > > > > > > > > > the
> > > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > >> well
> > > > > > > > > > > > >> > as in
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> > topic
> > > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> > include
> > > > > > > converting
> > > > > > > > > > all
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > offsets
> > > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in
> > this
> > > > > KIP,
> > > > > > > the
> > > > > > > > > > consumer
> > > > > > > > > > > > >> > would be
> > > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
> > ids. That
> > > > > > > being
> > > > > > > > > > said, it
> > > > > > > > > > > > >> > still has
> > > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > > > >> > > > > >> > > > > > > the topics names in various
> > APIs so
> > > > > > > having a
> > > > > > > > > > > > >> mapping
> > > > > > > > > > > > >> > in the
> > > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> > opinion on
> > > > > > > this.
> > > > > > > > > > Here I
> > > > > > > > > > > > >> > wonder if
> > > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
> > KIP. I
> > > > > would
> > > > > > > rather
> > > > > > > > > > focus
> > > > > > > > > > > > >> on
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > > > > > > separately if
> > > > > > > > > > we see
> > > > > > > > > > > > >> > value in
> > > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> > point about
> > > > > > > using
> > > > > > > > > > topic ids
> > > > > > > > > > > > >> > in the
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > ConsumerGroupHeartbeatRequest, I
> > > > > think
> > > > > > > that
> > > > > > > > > > there
> > > > > > > > > > > > >> is
> > > > > > > > > > > > >> > one
> > > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer
> > will
> > > > > have
> > > > > > > the
> > > > > > > > > > > > >> opportunity
> > > > > > > > > > > > >> > to
> > > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> > passing
> > > > > them
> > > > > > > into
> > > > > > > > > > the
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > > > > coordinator
> > > > > > > will
> > > > > > > > > > also
> > > > > > > > > > > > >> notice
> > > > > > > > > > > > >> > it but
> > > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> > reject an
> > > > > > > invalid
> > > > > > > > > > topic in
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> > all
> > > > > other
> > > > > > > minor
> > > > > > > > > > > > >> questions
> > > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > question:
> > > > > > > personally
> > > > > > > > > > I think
> > > > > > > > > > > > >> > it's not
> > > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> > API with
> > > > > > > Pattern,
> > > > > > > > > > but
> > > > > > > > > > > > >> > instead we
> > > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> > that our
> > > > > > > > > > subscription may
> > > > > > > > > > > > >> be
> > > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> > case is a
> > > > > > > very rare
> > > > > > > > > > > > >> scenario
> > > > > > > > > > > > >> > I felt
> > > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > > > subscription may
> > > > > > > > > > be
> > > > > > > > > > > > >> more
> > > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
> > look at
> > > > > the
> > > > > > > > > > differences
> > > > > > > > > > > > >> > between
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > > >> > > > > >> > > > > > > engines to better understand
> > the
> > > > > > > potential
> > > > > > > > > > issues.
> > > > > > > > > > > > >> My
> > > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > > > >> > > > > >> > > > > > > that would work for all the
> > basic
> > > > > > > regular
> > > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
> > about
> > > > > the
> > > > > > > various
> > > > > > > > > > > > >> character
> > > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
> > think
> > > > > about
> > > > > > > this.
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> > 11:28 PM
> > > > > > > Guozhang
> > > > > > > > > > Wang <
> > > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on
> > the high
> > > > > > > level
> > > > > > > > > > there are
> > > > > > > > > > > > >> > two meta
> > > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> > especially
> > > > > > > the last
> > > > > > > > > > step
> > > > > > > > > > > > >> when
> > > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > protocol, in
> > > > > which
> > > > > > > we
> > > > > > > > > > would
> > > > > > > > > > > > >> have a
> > > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > > > protocols /
> > > > > > > rpcs
> > > > > > > > > > are
> > > > > > > > > > > > >> used
> > > > > > > > > > > > >> > by
> > > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> > could
> > > > > > > "mimic"
> > > > > > > > > > the old
> > > > > > > > > > > > >> > behavior
> > > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
> > we
> > > > > need to
> > > > > > > > > > present
> > > > > > > > > > > > >> about.
> > > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids.
> > So
> > > > > far as
> > > > > > > > > > KIP-516 the
> > > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> > client,
> > > > > but
> > > > > > > they
> > > > > > > > > > are not
> > > > > > > > > > > > >> > exposed
> > > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
> > the
> > > > > > > question is,
> > > > > > > > > > first
> > > > > > > > > > > > >> > should we
> > > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> > the
> > > > > names
> > > > > > > -> ids
> > > > > > > > > > mapping
> > > > > > > > > > > > >> > itself
> > > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > existing
> > > > > RPCs
> > > > > > > and the
> > > > > > > > > > new
> > > > > > > > > > > > >> RPCs
> > > > > > > > > > > > >> > with
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> > consider
> > > > > > > exposing
> > > > > > > > > > the
> > > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > >> well
> > > > > > > > > > > > >> > as in
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> > topic
> > > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> > all
> > > > > other
> > > > > > > minor
> > > > > > > > > > > > >> questions
> > > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > question:
> > > > > > > personally
> > > > > > > > > > I think
> > > > > > > > > > > > >> > it's not
> > > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> > API with
> > > > > > > Pattern,
> > > > > > > > > > but
> > > > > > > > > > > > >> > instead we
> > > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> > that our
> > > > > > > > > > subscription may
> > > > > > > > > > > > >> be
> > > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> > case is a
> > > > > > > very rare
> > > > > > > > > > > > >> scenario
> > > > > > > > > > > > >> > I felt
> > > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > > > subscription may
> > > > > > > > > > be
> > > > > > > > > > > > >> more
> > > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
> > 5:23 AM
> > > > > > > David Jacot
> > > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > feedback. Let
> > > > > me
> > > > > > > answer
> > > > > > > > > > your
> > > > > > > > > > > > >> > questions
> > > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> > premature to
> > > > > > > talk about
> > > > > > > > > > > > >> target
> > > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > existing group
> > > > > > > protocol.
> > > > > > > > > > > > >> Unlike
> > > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
> > hence
> > > > > > > > > > deprecation/removal
> > > > > > > > > > > > >> > will be
> > > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > applications
> > > > > migrate
> > > > > > > to
> > > > > > > > > > the new
> > > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> > > > > remove it.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> > intend to
> > > > > > > release
> > > > > > > > > > this in
> > > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > > >> > > > > it
> > > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that
> > as a
> > > > > way to
> > > > > > > > > > estimate
> > > > > > > > > > > > >> when
> > > > > > > > > > > > >> > we'd
> > > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I
> > also
> > > > > > > suggest
> > > > > > > > > > removing
> > > > > > > > > > > > >> > this part.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> > reasoning. As
> > > > > > > > > > explained, I
> > > > > > > > > > > > >> plan
> > > > > > > > > > > > >> > to
> > > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> > while we
> > > > > > > implement
> > > > > > > > > > the new
> > > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals will
> > be
> > > > > slightly
> > > > > > > > > > different
> > > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
> > tighten
> > > > > the
> > > > > > > > > > switch from
> > > > > > > > > > > > >> > the old
> > > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
> > group
> > > > > > > coordinator
> > > > > > > > > > to a
> > > > > > > > > > > > >> > major
> > > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to
> > use a
> > > > > flag
> > > > > > > to do
> > > > > > > > > > the
> > > > > > > > > > > > >> > switch
> > > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
> > out the
> > > > > > > details of
> > > > > > > > > > the
> > > > > > > > > > > > >> > migration
> > > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
> > support
> > > > > > > online
> > > > > > > > > > > > >> migrations.
> > > > > > > > > > > > >> > Is
> > > > > > > > > > > > >> > > > > that
> > > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> > detail. It
> > > > > could
> > > > > > > also
> > > > > > > > > > be done
> > > > > > > > > > > > >> > as a
> > > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
> > online
> > > > > > > migrations
> > > > > > > > > > for the
> > > > > > > > > > > > >> > group.
> > > > > > > > > > > > >> > > > > That
> > > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> > protocol
> > > > > > > will be
> > > > > > > > > > able to
> > > > > > > > > > > > >> > switch to
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain
> > how that
> > > > > > > will work
> > > > > > > > > > > > >> though.
> > > > > > > > > > > > >> > It is
> > > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> > > > > upgraded or
> > > > > > > > > > rolled to a
> > > > > > > > > > > > >> > software
> > > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
> > Both the
> > > > > > > old and
> > > > > > > > > > the new
> > > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > > >> > > > > >> > > > > > > > > support the old protocol
> > and
> > > > > rely
> > > > > > > on the
> > > > > > > > > > same
> > > > > > > > > > > > >> > persisted
> > > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > > > >> > > > > >> > > > > > > > > they can work together.
> > This
> > > > > point
> > > > > > > is an
> > > > > > > > > > > > >> offline
> > > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live because
> > it
> > > > > would
> > > > > > > require
> > > > > > > > > > > > >> > shutting down
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and starting
> > up the
> > > > > > > new one
> > > > > > > > > > and
> > > > > > > > > > > > >> that
> > > > > > > > > > > > >> > would
> > > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > > > > > > version/IBP
> > > > > > > > > > must be
> > > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> > protocol.
> > > > > This
> > > > > > > cannot
> > > > > > > > > > be done
> > > > > > > > > > > > >> > before
> > > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the
> > old
> > > > > > > coordinator
> > > > > > > > > > doesn't
> > > > > > > > > > > > >> > support
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> > > > > upgraded
> > > > > > > to a
> > > > > > > > > > version
> > > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> > KIP-792).
> > > > > If
> > > > > > > the
> > > > > > > > > > consumer
> > > > > > > > > > > > >> is
> > > > > > > > > > > > >> > > > > already
> > > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at
> > this
> > > > > point.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be
> > rolled
> > > > > > > with the
> > > > > > > > > > > > >> feature
> > > > > > > > > > > > >> > flag
> > > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> > automatically
> > > > > > > > > > converted when
> > > > > > > > > > > > >> > the first
> > > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
> > joins
> > > > > the
> > > > > > > group.
> > > > > > > > > > While
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > members
> > > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> > > > > upgraded,
> > > > > > > the old
> > > > > > > > > > > > >> protocol
> > > > > > > > > > > > >> > is
> > > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
> > this in
> > > > > the
> > > > > > > KIP.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we
> > are
> > > > > > > pushing the
> > > > > > > > > > pattern
> > > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like there
> > > > > could be
> > > > > > > some
> > > > > > > > > > tricky
> > > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
> > users to
> > > > > > > detect
> > > > > > > > > > that
> > > > > > > > > > > > >> they
> > > > > > > > > > > > >> > need to
> > > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
> > the new
> > > > > > > protocol?
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit
> > more
> > > > > > > optimistic
> > > > > > > > > > than
> > > > > > > > > > > > >> you
> > > > > > > > > > > > >> > on this
> > > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > majority of
> > > > > the
> > > > > > > cases
> > > > > > > > > > are
> > > > > > > > > > > > >> simple
> > > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> > engine. The
> > > > > > > > > > coordinator will
> > > > > > > > > > > > >> > verify
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer
> > if the
> > > > > > > regex is
> > > > > > > > > > not
> > > > > > > > > > > > >> valid.
> > > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
> > worst
> > > > > case,
> > > > > > > the
> > > > > > > > > > first
> > > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> > rejected.
> > > > > This
> > > > > > > should
> > > > > > > > > > be used
> > > > > > > > > > > > >> > as the
> > > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> > > > > validate
> > > > > > > their
> > > > > > > > > > regex
> > > > > > > > > > > > >> > before
> > > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
> > them with
> > > > > > > another
> > > > > > > > > > group.
> > > > > > > > > > > > >> For
> > > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> > pre-prod
> > > > > > > environment.
> > > > > > > > > > > > >> Another
> > > > > > > > > > > > >> > way
> > > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > consumer-group tool
> > > > > to
> > > > > > > > > > provide a
> > > > > > > > > > > > >> regex
> > > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> > > > > enough in
> > > > > > > your
> > > > > > > > > > > > >> opinion?
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> > > > > question,
> > > > > > > will
> > > > > > > > > > the
> > > > > > > > > > > > >> Java
> > > > > > > > > > > > >> > client
> > > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the current
> > regex
> > > > > > > engine for
> > > > > > > > > > > > >> > compatibility
> > > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
> > keep using
> > > > > > > client
> > > > > > > > > > based
> > > > > > > > > > > > >> > regex at
> > > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and
> > then
> > > > > > > migrate to
> > > > > > > > > > server
> > > > > > > > > > > > >> > based
> > > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your point
> > but I
> > > > > am
> > > > > > > > > > concerned that
> > > > > > > > > > > > >> > this
> > > > > > > > > > > > >> > > > > would
> > > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in this
> > mode.
> > > > > That
> > > > > > > > > > would go
> > > > > > > > > > > > >> > against our
> > > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
> > because
> > > > > we
> > > > > > > would
> > > > > > > > > > have to
> > > > > > > > > > > > >> > continue
> > > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> > client
> > > > > side. I
> > > > > > > would
> > > > > > > > > > rather
> > > > > > > > > > > > >> > not do
> > > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
> > the group
> > > > > > > > > > coordinator
> > > > > > > > > > > > >> will
> > > > > > > > > > > > >> > be
> > > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations and
> > that
> > > > > the
> > > > > > > > > > > > >> configurations
> > > > > > > > > > > > >> > will be
> > > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will
> > a
> > > > > > > transition to
> > > > > > > > > > DEAD
> > > > > > > > > > > > >> > trigger
> > > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > > > configurations
> > > > > > > will be
> > > > > > > > > > > > >> deleted
> > > > > > > > > > > > >> > when
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > together.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to
> > store
> > > > > the
> > > > > > > > > > configs in
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > group
> > > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> > cluster
> > > > > > > configs and
> > > > > > > > > > their
> > > > > > > > > > > > >> > values?
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
> > group
> > > > > > > > > > configurations are
> > > > > > > > > > > > >> > overrides
> > > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to
> > know
> > > > > all
> > > > > > > the
> > > > > > > > > > overrides
> > > > > > > > > > > > >> > though,
> > > > > > > > > > > > >> > > > > you
> > > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > > > > coordinators.
> > > > > > > You
> > > > > > > > > > cannot
> > > > > > > > > > > > >> > rely on
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> > > > > configure a
> > > > > > > group
> > > > > > > > > > before
> > > > > > > > > > > > >> > starting
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered allowing
> > the
> > > > > > > explicit
> > > > > > > > > > creation
> > > > > > > > > > > > >> of
> > > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> > decoupled
> > > > > from
> > > > > > > the
> > > > > > > > > > group
> > > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > > > > > > automatically
> > > > > > > > > > created in
> > > > > > > > > > > > >> > this
> > > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be lost
> > after
> > > > > the
> > > > > > > > > > retention
> > > > > > > > > > > > >> > period of
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > > > Consumer.subscribe
> > > > > > > method
> > > > > > > > > > for the
> > > > > > > > > > > > >> > Java
> > > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > `java.util.regex.Pattern` of
> > > > > do
> > > > > > > we
> > > > > > > > > > have to
> > > > > > > > > > > > >> > introduce
> > > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> > question. I
> > > > > > > forgot
> > > > > > > > > > about
> > > > > > > > > > > > >> that
> > > > > > > > > > > > >> > one.
> > > > > > > > > > > > >> > > > > As
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > `java.util.regex.Pattern` is
> > > > > not
> > > > > > > fully
> > > > > > > > > > > > >> compatible
> > > > > > > > > > > > >> > with
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might
> > be
> > > > > better
> > > > > > > to
> > > > > > > > > > deprecate
> > > > > > > > > > > > >> > it and
> > > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string. We
> > would
> > > > > > > rely on
> > > > > > > > > > the
> > > > > > > > > > > > >> server
> > > > > > > > > > > > >> > side
> > > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I
> > think
> > > > > that
> > > > > > > we
> > > > > > > > > > could
> > > > > > > > > > > > >> still
> > > > > > > > > > > > >> > try to
> > > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
> > should
> > > > > > > work, I
> > > > > > > > > > think, in
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> > Justine
> > > > > that we
> > > > > > > > > > should be
> > > > > > > > > > > > >> > clearer
> > > > > > > > > > > > >> > > > > about
> > > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > > IBP/metadata.version
> > > > > > > from the
> > > > > > > > > > > > >> feature
> > > > > > > > > > > > >> > flag.
> > > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the default
> > for
> > > > > the
> > > > > > > feature
> > > > > > > > > > flag
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > true
> > > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once
> > we
> > > > > want to
> > > > > > > make
> > > > > > > > > > it the
> > > > > > > > > > > > >> > default.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that
> > feature
> > > > > > > flag
> > > > > > > > > > mainly
> > > > > > > > > > > > >> > during the
> > > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> > > > > mentioned
> > > > > > > it, I
> > > > > > > > > > think,
> > > > > > > > > > > > >> > because
> > > > > > > > > > > > >> > > > > we
> > > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for
> > it.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> > protocol APIs
> > > > > > > don't
> > > > > > > > > > mention
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > required
> > > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> > > > > consistency.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
> > that
> > > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do
> > more
> > > > > than
> > > > > > > > > > reading.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> > > > > current
> > > > > > > > > > protocol works
> > > > > > > > > > > > >> > though.
> > > > > > > > > > > > >> > > > > We
> > > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to
> > join a
> > > > > > > group. We
> > > > > > > > > > could
> > > > > > > > > > > > >> > consider
> > > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that
> > it is
> > > > > worth
> > > > > > > it.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> > recreation
> > > > > > > handled by
> > > > > > > > > > the
> > > > > > > > > > > > >> > consumer
> > > > > > > > > > > > >> > > > > with
> > > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be
> > good to
> > > > > > > have a
> > > > > > > > > > section
> > > > > > > > > > > > >> on
> > > > > > > > > > > > >> > this.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > > > > > > perspective, the
> > > > > > > > > > new
> > > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > will
> > > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> > treat it
> > > > > like a
> > > > > > > > > > topic with
> > > > > > > > > > > > >> a
> > > > > > > > > > > > >> > > > > different
> > > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> > > > > fetch/commit
> > > > > > > > > > offsets
> > > > > > > > > > > > >> APIs
> > > > > > > > > > > > >> > do not
> > > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would
> > > > > reuse the
> > > > > > > > > > offsets
> > > > > > > > > > > > >> based
> > > > > > > > > > > > >> > on the
> > > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
> > those
> > > > > APIs
> > > > > > > as well
> > > > > > > > > > in
> > > > > > > > > > > > >> order
> > > > > > > > > > > > >> > to be
> > > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> > strengthen
> > > > > the
> > > > > > > > > > semantics of
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions
> > we will
> > > > > > > write the
> > > > > > > > > > new
> > > > > > > > > > > > >> > coordinator
> > > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > implementation
> > > > > > > detail, do
> > > > > > > > > > we plan
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > have a
> > > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> > scalability
> > > > > > > goal when
> > > > > > > > > > it
> > > > > > > > > > > > >> comes
> > > > > > > > > > > > >> > to how
> > > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> > support?
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at
> > the
> > > > > > > moment. The
> > > > > > > > > > > > >> protocol
> > > > > > > > > > > > >> > should
> > > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group. We
> > will
> > > > > > > measure
> > > > > > > > > > this when
> > > > > > > > > > > > >> > we have
> > > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
> > that we
> > > > > might
> > > > > > > have
> > > > > > > > > > other
> > > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> > commits).
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> > having
> > > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > > >> > instead
> > > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> > resolve the
> > > > > > > regex on
> > > > > > > > > > the
> > > > > > > > > > > > >> > server, we
> > > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> > > > > difference
> > > > > > > is that
> > > > > > > > > > > > >> sending
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
> > topic
> > > > > names
> > > > > > > is less
> > > > > > > > > > > > >> > efficient.
> > > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to
> > > > > handle
> > > > > > > if we
> > > > > > > > > > have
> > > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > ids.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > consolidate the
> > > > > > > metadata
> > > > > > > > > > lookup
> > > > > > > > > > > > >> > on the
> > > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
> > with your
> > > > > > > point. As
> > > > > > > > > > a
> > > > > > > > > > > > >> second
> > > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here
> > for the
> > > > > > > delete and
> > > > > > > > > > > > >> > recreation
> > > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow
> > > > > users to
> > > > > > > > > > subscribe
> > > > > > > > > > > > >> with
> > > > > > > > > > > > >> > topic
> > > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > future because that is
> > the only
> > > > > > > way to be
> > > > > > > > > > > > >> really
> > > > > > > > > > > > >> > robust
> > > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at
> > 1:38 PM
> > > > > > > David
> > > > > > > > > > Jacot <
> > > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> > comments.
> > > > > Please
> > > > > > > find
> > > > > > > > > > my
> > > > > > > > > > > > >> answers
> > > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol
> > > > > relies on
> > > > > > > > > > topic IDs
> > > > > > > > > > > > >> > with the
> > > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in
> > the
> > > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names
> > is the
> > > > > > > right call
> > > > > > > > > > here.
> > > > > > > > > > > > >> I
> > > > > > > > > > > > >> > need to
> > > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > Obviously, the
> > > > > > > KIP does
> > > > > > > > > > not
> > > > > > > > > > > > >> > change the
> > > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
> > topic
> > > > > IDs.
> > > > > > > This
> > > > > > > > > > may be
> > > > > > > > > > > > >> > something
> > > > > > > > > > > > >> > > > > >> that
> > > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
> > would
> > > > > give
> > > > > > > better
> > > > > > > > > > > > >> overall
> > > > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> > think that
> > > > > I
> > > > > > > should
> > > > > > > > > > not
> > > > > > > > > > > > >> have
> > > > > > > > > > > > >> > > > > mentioned
> > > > > > > > > > > > >> > > > > >> > this
> > > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it.
> > We can
> > > > > > > use an
> > > > > > > > > > internal
> > > > > > > > > > > > >> > > > > >> configuration
> > > > > > > > > > > > >> > > > > >> > > while
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
> > will be
> > > > > > > > > > supported. The
> > > > > > > > > > > > >> > change is
> > > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
> > that the
> > > > > > > cluster
> > > > > > > > > > uses
> > > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > IDs.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022
> > at 9:53
> > > > > PM
> > > > > > > > > > Guozhang
> > > > > > > > > > > > >> Wang <
> > > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> > feedback.
> > > > > Here
> > > > > > > are
> > > > > > > > > > some
> > > > > > > > > > > > >> replies
> > > > > > > > > > > > >> > > > > inlined
> > > > > > > > > > > > >> > > > > >> > > below:
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022
> > at
> > > > > 2:53 AM
> > > > > > > > > > Ismael Juma
> > > > > > > > > > > > >> <
> > > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
> > KIP. This
> > > > > has
> > > > > > > the
> > > > > > > > > > > > >> potential
> > > > > > > > > > > > >> > to be a
> > > > > > > > > > > > >> > > > > >> > great
> > > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > > > questions/comments:
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> > > > > premature to
> > > > > > > talk
> > > > > > > > > > about
> > > > > > > > > > > > >> > target
> > > > > > > > > > > > >> > > > > >> versions
> > > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> > existing
> > > > > group
> > > > > > > > > > protocol.
> > > > > > > > > > > > >> > Unlike
> > > > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
> > and hence
> > > > > > > > > > > > >> > deprecation/removal will
> > > > > > > > > > > > >> > > > > >> be
> > > > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> > applications
> > > > > > > migrate
> > > > > > > > > > to the
> > > > > > > > > > > > >> new
> > > > > > > > > > > > >> > > > > >> protocol.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
> > you. I
> > > > > think
> > > > > > > we can
> > > > > > > > > > > > >> remove
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> proposed
> > > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > > > > Deprecation,
> > > > > > > and
> > > > > > > > > > Migration
> > > > > > > > > > > > >> > Plan` and
> > > > > > > > > > > > >> > > > > >> > > instead
> > > > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide
> > in the
> > > > > > > future
> > > > > > > > > > about
> > > > > > > > > > > > >> when
> > > > > > > > > > > > >> > we
> > > > > > > > > > > > >> > > > > would
> > > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > behaviors.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> > > > > intend to
> > > > > > > > > > release this
> > > > > > > > > > > > >> > in 4.x,
> > > > > > > > > > > > >> > > > > >> but
> > > > > > > > > > > > >> > > > > >> > it
> > > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
> > that as
> > > > > a
> > > > > > > way to
> > > > > > > > > > > > >> estimate
> > > > > > > > > > > > >> > when
> > > > > > > > > > > > >> > > > > we'd
> > > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > protocol, I
> > > > > also
> > > > > > > suggest
> > > > > > > > > > > > >> removing
> > > > > > > > > > > > >> > this
> > > > > > > > > > > > >> > > > > >> part.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> > > > > specifically
> > > > > > > > > > related to
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more
> > for
> > > > > client
> > > > > > > > > > upgrades.
> > > > > > > > > > > > >> I.e.
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may
> > be done
> > > > > > > first,
> > > > > > > > > > and then
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > client
> > > > > > > > > > > > >> > > > > >> > side,
> > > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> > "released"
> > > > > by
> > > > > > > the
> > > > > > > > > > time
> > > > > > > > > > > > >> clients
> > > > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable
> > the
> > > > > > > feature the
> > > > > > > > > > > > >> clients
> > > > > > > > > > > > >> > need to
> > > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> > rolling
> > > > > bounce
> > > > > > > and
> > > > > > > > > > then set
> > > > > > > > > > > > >> > the flag
> > > > > > > > > > > > >> > > > > >> > with a
> > > > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we
> > feel
> > > > > it's
> > > > > > > > > > better to be
> > > > > > > > > > > > >> > released
> > > > > > > > > > > > >> > > > > >> in a
> > > > > > > > > > > > >> > > > > >> > > major
> > > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> > flesh out
> > > > > the
> > > > > > > > > > details of
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > migration
> > > > > > > > > > > > >> > > > > >> > > story.
> > > > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
> > will
> > > > > support
> > > > > > > online
> > > > > > > > > > > > >> > migrations. Is
> > > > > > > > > > > > >> > > > > >> that
> > > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> > detail.
> > > > > It
> > > > > > > could
> > > > > > > > > > also be
> > > > > > > > > > > > >> > done as a
> > > > > > > > > > > > >> > > > > >> > > separate
> > > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's
> > the
> > > > > part we
> > > > > > > can
> > > > > > > > > > be more
> > > > > > > > > > > > >> > concrete
> > > > > > > > > > > > >> > > > > >> about
> > > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to
> > your
> > > > > > > question 2)
> > > > > > > > > > above).
> > > > > > > > > > > > >> > We will
> > > > > > > > > > > > >> > > > > >> work
> > > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel
> > as we
> > > > > > > solicit
> > > > > > > > > > more
> > > > > > > > > > > > >> > feedback.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that
> > we are
> > > > > > > pushing
> > > > > > > > > > the
> > > > > > > > > > > > >> > pattern
> > > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like
> > there
> > > > > > > could be
> > > > > > > > > > some
> > > > > > > > > > > > >> tricky
> > > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism
> > for
> > > > > users to
> > > > > > > > > > detect that
> > > > > > > > > > > > >> > they
> > > > > > > > > > > > >> > > > > need
> > > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > > >> > > > > >> > > update
> > > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching
> > to the
> > > > > new
> > > > > > > > > > protocol?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need
> > some
> > > > > > > tooling for
> > > > > > > > > > > > >> non-java
> > > > > > > > > > > > >> > client
> > > > > > > > > > > > >> > > > > >> > users
> > > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client
> > before
> > > > > > > > > > switching to
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > new
> > > > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top
> > of my
> > > > > head
> > > > > > > > > > though,
> > > > > > > > > > > > >> maybe
> > > > > > > > > > > > >> > others
> > > > > > > > > > > > >> > > > > >> like
> > > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the
> > last
> > > > > > > question,
> > > > > > > > > > will the
> > > > > > > > > > > > >> > Java
> > > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> > current
> > > > > regex
> > > > > > > > > > engine for
> > > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to
> > keep
> > > > > using
> > > > > > > > > > client
> > > > > > > > > > > > >> based
> > > > > > > > > > > > >> > regex
> > > > > > > > > > > > >> > > > > at
> > > > > > > > > > > > >> > > > > >> > > first to
> > > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple
> > and
> > > > > then
> > > > > > > migrate
> > > > > > > > > > to
> > > > > > > > > > > > >> > server based
> > > > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not
> > thought
> > > > > > > about
> > > > > > > > > > that for
> > > > > > > > > > > > >> > java
> > > > > > > > > > > > >> > > > > >> clients,
> > > > > > > > > > > > >> > > > > >> > > and
> > > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > > > > > > compatibility
> > > > > > > > > > issues do
> > > > > > > > > > > > >> > you have
> > > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
> > that the
> > > > > group
> > > > > > > > > > coordinator
> > > > > > > > > > > > >> > will be
> > > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the configurations
> > and
> > > > > that
> > > > > > > the
> > > > > > > > > > > > >> > configurations
> > > > > > > > > > > > >> > > > > will
> > > > > > > > > > > > >> > > > > >> be
> > > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted.
> > Will a
> > > > > > > > > > transition to
> > > > > > > > > > > > >> DEAD
> > > > > > > > > > > > >> > > > > trigger
> > > > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD
> > state
> > > > > is an
> > > > > > > > > > ending
> > > > > > > > > > > > >> state
> > > > > > > > > > > > >> > (we
> > > > > > > > > > > > >> > > > > would
> > > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the group
> > is
> > > > > EMPTY
> > > > > > > and
> > > > > > > > > > also all
> > > > > > > > > > > > >> of
> > > > > > > > > > > > >> > its
> > > > > > > > > > > > >> > > > > >> > metadata
> > > > > > > > > > > > >> > > > > >> > > are
> > > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
> > DEAD this
> > > > > > > group
> > > > > > > > > > would
> > > > > > > > > > > > >> never
> > > > > > > > > > > > >> > be
> > > > > > > > > > > > >> > > > > >> revived.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice
> > to
> > > > > store
> > > > > > > the
> > > > > > > > > > configs
> > > > > > > > > > > > >> in
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > group
> > > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> > > > > cluster
> > > > > > > configs
> > > > > > > > > > and
> > > > > > > > > > > > >> their
> > > > > > > > > > > > >> > > > > values?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> > question,
> > > > > and our
> > > > > > > > > > thoughts
> > > > > > > > > > > > >> are
> > > > > > > > > > > > >> > that
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> > > > > overrides
> > > > > > > of the
> > > > > > > > > > > > >> > cluster-level
> > > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per group
> > so
> > > > > when an
> > > > > > > > > > admin list
> > > > > > > > > > > > >> > cluster
> > > > > > > > > > > > >> > > > > >> > configs
> > > > > > > > > > > > >> > > > > >> > > it's
> > > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> > cluster-level
> > > > > > > > > > "defaults", not
> > > > > > > > > > > > >> > showing
> > > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> > someone
> > > > > > > configure a
> > > > > > > > > > group
> > > > > > > > > > > > >> > before
> > > > > > > > > > > > >> > > > > >> starting
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> > allowing
> > > > > the
> > > > > > > explicit
> > > > > > > > > > > > >> > creation of
> > > > > > > > > > > > >> > > > > >> groups?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> > > > > decoupled
> > > > > > > from
> > > > > > > > > > the group
> > > > > > > > > > > > >> > > > > lifecycle.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
> > created
> > > > > > > before
> > > > > > > > > > the group
> > > > > > > > > > > > >> > itself
> > > > > > > > > > > > >> > > > > as
> > > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> > > > > requires
> > > > > > > the
> > > > > > > > > > > > >> corresponding
> > > > > > > > > > > > >> > > > > >> request to
> > > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> > based on
> > > > > the
> > > > > > > group
> > > > > > > > > > id ---
> > > > > > > > > > > > >> > the only
> > > > > > > > > > > > >> > > > > >> > thing
> > > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> > itself is
> > > > > gone
> > > > > > > we
> > > > > > > > > > also check
> > > > > > > > > > > > >> > if there
> > > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > >> > > > > >> > > any
> > > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to
> > that
> > > > > group
> > > > > > > and
> > > > > > > > > > delete
> > > > > > > > > > > > >> as
> > > > > > > > > > > > >> > well.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> > indeed
> > > > > > > introduces an
> > > > > > > > > > > > >> > asymmetry on
> > > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > creation
> > > > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> > config
> > > > > > > entities,
> > > > > > > > > > and we
> > > > > > > > > > > > >> > would like
> > > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > > >> > > > > >> > > hear
> > > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we
> > should
> > > > > aim
> > > > > > > for
> > > > > > > > > > symmetry
> > > > > > > > > > > > >> > i.e.
> > > > > > > > > > > > >> > > > > >> totally
> > > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not
> > > > > delete
> > > > > > > them at
> > > > > > > > > > all
> > > > > > > > > > > > >> when
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > > >> > > > > >> > is
> > > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> > deletion
> > > > > > > operations
> > > > > > > > > > by
> > > > > > > > > > > > >> > themselves.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > > > > > Consumer.subscribe
> > > > > > > > > > method for
> > > > > > > > > > > > >> > the Java
> > > > > > > > > > > > >> > > > > >> > client
> > > > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > `java.util.regex.Pattern` of
> > > > > > > do we
> > > > > > > > > > have
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > > >> introduce an
> > > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
> > need to
> > > > > > > introduce
> > > > > > > > > > an
> > > > > > > > > > > > >> > overload, but
> > > > > > > > > > > > >> > > > > >> I'm
> > > > > > > > > > > > >> > > > > >> > > all
> > > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > compatibility
> > > > > > > issues
> > > > > > > > > > that we
> > > > > > > > > > > > >> may
> > > > > > > > > > > > >> > > > > overlook.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
> > Justine
> > > > > > > that we
> > > > > > > > > > should
> > > > > > > > > > > > >> be
> > > > > > > > > > > > >> > clearer
> > > > > > > > > > > > >> > > > > >> > about
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > > > > IBP/metadata.version
> > > > > > > > > > from the
> > > > > > > > > > > > >> > feature
> > > > > > > > > > > > >> > > > > >> flag.
> > > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> > default
> > > > > for
> > > > > > > the
> > > > > > > > > > feature
> > > > > > > > > > > > >> flag
> > > > > > > > > > > > >> > to
> > > > > > > > > > > > >> > > > > true
> > > > > > > > > > > > >> > > > > >> > > based on
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
> > once we
> > > > > > > want to
> > > > > > > > > > make it
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > >> default.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> > protocol
> > > > > APIs
> > > > > > > don't
> > > > > > > > > > mention
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > >> required
> > > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that
> > for
> > > > > > > consistency.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > certainly do
> > > > > that.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
> > odd that
> > > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > > > > > >> > > > > >> > requires
> > > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to
> > do
> > > > > more
> > > > > > > than
> > > > > > > > > > reading.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
> > myself
> > > > > as
> > > > > > > well,
> > > > > > > > > > but in
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > end we
> > > > > > > > > > > > >> > > > > >> > could
> > > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative:
> > adding
> > > > > > > Write
> > > > > > > > > > Group
> > > > > > > > > > > > >> seems
> > > > > > > > > > > > >> > an
> > > > > > > > > > > > >> > > > > >> overkill
> > > > > > > > > > > > >> > > > > >> > > here
> > > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > elsewhere (we
> > > > > only
> > > > > > > have
> > > > > > > > > > Read /
> > > > > > > > > > > > >> > Delete
> > > > > > > > > > > > >> > > > > and
> > > > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to
> > hear
> > > > > others
> > > > > > > > > > thoughts.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> > > > > recreation
> > > > > > > > > > handled by
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > consumer
> > > > > > > > > > > > >> > > > > >> > with
> > > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would
> > be
> > > > > good
> > > > > > > to have
> > > > > > > > > > a
> > > > > > > > > > > > >> > section on
> > > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > > > > > > subscription
> > > > > > > > > > right? Yes
> > > > > > > > > > > > >> > we can
> > > > > > > > > > > > >> > > > > >> add a
> > > > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically
> > the
> > > > > idea
> > > > > > > is that
> > > > > > > > > > > > >> consumer
> > > > > > > > > > > > >> > would
> > > > > > > > > > > > >> > > > > be
> > > > > > > > > > > > >> > > > > >> > > totally
> > > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as
> > it's
> > > > > > > handled all
> > > > > > > > > > by the
> > > > > > > > > > > > >> > brokers.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> > mentions we
> > > > > will
> > > > > > > write
> > > > > > > > > > the
> > > > > > > > > > > > >> new
> > > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > implementation
> > > > > > > detail,
> > > > > > > > > > do we
> > > > > > > > > > > > >> > plan to
> > > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought
> > about
> > > > > > > that. But
> > > > > > > > > > I think
> > > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > answer
> > > > > > > > > > > > >> > > > > >> > > should
> > > > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> > > > > scalability
> > > > > > > goal
> > > > > > > > > > when it
> > > > > > > > > > > > >> > comes to
> > > > > > > > > > > > >> > > > > >> how
> > > > > > > > > > > > >> > > > > >> > > many
> > > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> > > > > support?
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> > think we
> > > > > > > should
> > > > > > > > > > shoot for
> > > > > > > > > > > > >> > 1000s of
> > > > > > > > > > > > >> > > > > >> > > members.
> > > > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
> > also
> > > > > depend
> > > > > > > on the
> > > > > > > > > > offset
> > > > > > > > > > > > >> > > > > management
> > > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> > > > > coordinator
> > > > > > > > > > which we
> > > > > > > > > > > > >> did
> > > > > > > > > > > > >> > not
> > > > > > > > > > > > >> > > > > >> cover in
> > > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> > number that
> > > > > > > applies
> > > > > > > > > > > > >> > universally.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider
> > > > > having
> > > > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > SubscribedTopicNames in
> > > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
> > resolve
> > > > > the
> > > > > > > regex
> > > > > > > > > > on the
> > > > > > > > > > > > >> > server,
> > > > > > > > > > > > >> > > > > we
> > > > > > > > > > > > >> > > > > >> > can
> > > > > > > > > > > > >> > > > > >> > > do
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > > > > > > difference is
> > > > > > > > > > that
> > > > > > > > > > > > >> > sending the
> > > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > > >> > > > > >> > > is
> > > > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
> > the topic
> > > > > > > names is
> > > > > > > > > > less
> > > > > > > > > > > > >> > efficient.
> > > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> > easier to
> > > > > > > handle if
> > > > > > > > > > we have
> > > > > > > > > > > > >> > topic
> > > > > > > > > > > > >> > > > > ids.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
> > still
> > > > > let
> > > > > > > the
> > > > > > > > > > clients
> > > > > > > > > > > > >> send
> > > > > > > > > > > > >> > names
> > > > > > > > > > > > >> > > > > >> is to
> > > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names
> > -> ids
> > > > > on
> > > > > > > the
> > > > > > > > > > broker /
> > > > > > > > > > > > >> > admin
> > > > > > > > > > > > >> > > > > client
> > > > > > > > > > > > >> > > > > >> > > only.
> > > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> > topic id
> > > > > in
> > > > > > > > > > KIP-516, we
> > > > > > > > > > > > >> > never
> > > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> > > > > leveraging
> > > > > > > the
> > > > > > > > > > related
> > > > > > > > > > > > >> > newer
> > > > > > > > > > > > >> > > > > >> versioned
> > > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic
> > id as
> > > > > empty
> > > > > > > UUID.
> > > > > > > > > > We
> > > > > > > > > > > > >> want
> > > > > > > > > > > > >> > to keep
> > > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> > > > > delegate
> > > > > > > the
> > > > > > > > > > reasoning
> > > > > > > > > > > > >> on
> > > > > > > > > > > > >> > broker
> > > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
> > 2022 at
> > > > > 10:45
> > > > > > > AM
> > > > > > > > > > David
> > > > > > > > > > > > >> Jacot
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > <dj...@confluent.io.invalid>
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to
> > start a
> > > > > > > > > > discussion
> > > > > > > > > > > > >> thread
> > > > > > > > > > > > >> > on
> > > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the
> > > > > Consumer
> > > > > > > > > > Rebalance
> > > > > > > > > > > > >> > Protocol.
> > > > > > > > > > > > >> > > > > >> With
> > > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> > rebalance
> > > > > > > protocol
> > > > > > > > > > (for
> > > > > > > > > > > > >> > consumers)
> > > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier
> > to
> > > > > > > implement for
> > > > > > > > > > > > >> > clients, and
> > > > > > > > > > > > >> > > > > >> easier
> > > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
> > look and
> > > > > let
> > > > > > > me
> > > > > > > > > > know what
> > > > > > > > > > > > >> > you
> > > > > > > > > > > > >> > > > > think.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
> > away from
> > > > > > > July
> > > > > > > > > > 18th to
> > > > > > > > > > > > >> > August 8th.
> > > > > > > > > > > > >> > > > > >> That
> > > > > > > > > > > > >> > > > > >> > > gives
> > > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read
> > and
> > > > > digest
> > > > > > > this
> > > > > > > > > > long
> > > > > > > > > > > > >> KIP.
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > >> --
> > > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >>
> > > > > > > > > > > > >
> > > > > > > > > >
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > -- Guozhang
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hey Jason,

Thanks for your comments.

> In regard to the RE2 library. I understand the intent to pass this through
> as a string without any validation so that we don't need the dependence on
> the client. I do wonder if that will be restrictive in practice. For
> example, if a regex changes on the client side, would we need to validate
> against the current assignment to see which partitions to continue
> fetching? Or would we instead invalidate the complete current assignment?
> Perhaps this is a rare case and we can do the latter. There could be more
> situations though.

I thought that the member should just pass the new regex to the group
coordinator and the group coordinator will take care of
adding/revoking partitions. The downside is that it is not immediately
enforced locally but that seems acceptable to me given that this is a
rare case. It should be reasonably quick though. The advantage is that
client does not have to think about this at all. It is all handler on
the server side.

> As an alternative, I wonder if we could define a minimalistic regex syntax
> (a subset of RE2 say) that could map more easily to regex support that
> already exists in clients? For example, in this scenario, we are matching
> topics, so we only need to be able to represent topic characters and not
> the whole smorgasbord of whitespace and unicode options. Would it make
> sense to have a reduced syntax which includes only standard topic
> characters, wildcards, character classes, and repetitions?

Yeah, I thought about this as well but I was a bit concerned by the
backward compatibility/migration path. I suppose that such a subset
would work for most of our users but does it work for all? For
instance, I can already think of | and possibly () that should be
included as well.

I suppose that we could use something like this:
- a-zA-Z0-9._-
- . (any chars)
- [] (char class) and [^] (negated char class)
- \d \D \w \W (common char classes)
- * + ? (do we need all the others?)
- |
- () (not sure about this one)

I don't have a strong opinion on this one. However, it seems that this
could be a pandora box if our subset turns out to not be sufficient
for some existing users. We will have to add more to it and the
compatibility story between versions could be a real nightmare. One
thing that we could do for sure is disabling all unicode groups in the
library. That seems possible and would already reduce the scope.

Best,
David

On Tue, Sep 27, 2022 at 3:42 AM Jason Gustafson
<ja...@confluent.io.invalid> wrote:
>
> Hey David,
>
> Thanks for the response.
>
> > Do you think that we could have a case where two members with the same
> static id are active?
>
> I was just thinking about the handoff when we spin up a new node before the
> old one completes shutdown. It's probably ok to fence the old member
> abruptly. I do recall we had a few issues with this handoff in the past. It
> was due to the fact that the replaced node might have been the leader.
> Since we've removed the leader notion, perhaps this kind of issue won't
> resurface.
>
> In regard to the RE2 library. I understand the intent to pass this through
> as a string without any validation so that we don't need the dependence on
> the client. I do wonder if that will be restrictive in practice. For
> example, if a regex changes on the client side, would we need to validate
> against the current assignment to see which partitions to continue
> fetching? Or would we instead invalidate the complete current assignment?
> Perhaps this is a rare case and we can do the latter. There could be more
> situations though.
>
> As an alternative, I wonder if we could define a minimalistic regex syntax
> (a subset of RE2 say) that could map more easily to regex support that
> already exists in clients? For example, in this scenario, we are matching
> topics, so we only need to be able to represent topic characters and not
> the whole smorgasbord of whitespace and unicode options. Would it make
> sense to have a reduced syntax which includes only standard topic
> characters, wildcards, character classes, and repetitions?
>
> Best,
> Jason
>
>
>
>
> On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com> wrote:
>
> > Hi David, thanks for all the detailed explanations. I think they all make
> > sense. Just want to have a couple follow-ups here:
> >
> > > I don't really see the benefits here because server side assignors
> > don't have metadata at all. They only assign topic-partitions. They
> > are not supposed to generate metadata nor to receive metadata from the
> > members.
> >
> > Just to clarify I was asking about the `version` of the assignor (i.e. up
> > to what version that the client would support), and I do agree we would not
> > need metadata. What I have in mind is that, for some specific built-in
> > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > newer version we would have a hierarchical rack ID string format, like
> > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > would still be in old format. In this case, the broker then needs to choose
> > the old versioned assignor. I'm probably making something up here for rack
> > aware assignors, but I'm wondering if in general such an "auto-downgrade"
> > behavior would be needed still for broker-side assignor, and if yes would
> > "version" still be useful.
> >
> > > Yeah, that's right. Within a rebalance, `onAssignment` is called once
> > when the member transitions to a new epoch. This one contains the full
> > metadata provided by the client side assignor. Then, `onAssignment`
> > can be called max N times where N is the number of partitions pending
> > revocation by other members. Let me try to clarify this in the KIP.
> >
> > Okay, my understanding is that the calling ordering of these callbacks
> > would be like the following:
> >
> > ----------------------------------------
> > onPartitionsRevoked();   // just once, since we do not really need
> > to revoke incrementally.
> >
> > onAssignment();    // the first call, with epoch incremented
> > onPartitionsAssigned();   // paired with the onAssignment
> >
> > onAssignment();              // the first onAssignment would bump up the
> > epoch, and the metadata reflected.
> > onPartitionsAssigned();   // each time we get an additional assignment, we
> > call onAssignment and then paired with an onPartitionsAssigned
> > ...
> > onAssignment();
> > onPartitionsAssigned();   // on each of the onAssignment calls, the encoded
> > metadata would not change, only the incrementally added partitions be
> > reflected
> >
> > Is that the case?
> >
> > I'm wondering if we would still call onAssignment just once, that encodes
> > all the assignment for this rebalance, including all the partitions that
> > should be assigned to the member but not yet assigned since they have not
> > been revoked by others. In that case the call ordering would be:
> >
> > ----------------------------------------
> > onPartitionsRevoked();   // just once
> > onAssignment();    // just once, with epoch incremented, and metadata
> > encoded changed, the "assignment" field also reflect the final target
> > assignment
> > onPartitionsAssigned();   // multiple times, which represent incrementally
> > added partitions
> > ...
> > onPartitionsAssigned();
> >
> > The motivation from this is that, most users would only implement the
> > rebalance callback listeners and hence we'd definitely need to make sure
> > the semantics of that does not change much, and the time
> > `onPartitionsAssigned` indicate the time when the partitions are actually
> > assigned to it; while for assignors, the `onAssignment` is used to indicate
> > what decision is made regarding for this member, i.e. when the partitions
> > are decided to be given to it, but not necessarily meant that it has been
> > given, since that time should be determined by the time of
> > `onPartitionsAssigned`. The benefits though, would be that assignor
> > implementers would not need to reason which `onAssignment` would be the
> > last one for this epoch.
> >
> >
> > Guozhang
> >
> >
> >
> >
> > Guozhang
> >
> > On Thu, Sep 22, 2022 at 2:20 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > field,
> > > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > > rationale seems to be assuming that we should always be able to do the
> > > > assignment at the broker-side assignor without errors. Personally I
> > think
> > > > it's still potentially beneficial to add the Error field even for
> > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > topics are not recognized with the current broker's metadata. What do
> > you
> > > > think?
> > >
> > > Yeah, that seems reasonable. However, I wonder if it would be better
> > > to use an exception on the server side. This is what we usually do for
> > > server side plugins. On the client side, we use a field because the
> > > errors are not defined in advance.
> > >
> > > Your comment also makes me think about what we should do when the
> > > server side assignor fails. I suppose that we have to keep the current
> > > assignment until a new event occurs. For instance, in your example,
> > > the coordinator would have to trigger a rebalance when unrecognized
> > > topics are available. This would be part of the metadata monitoring.
> > >
> > > > 2) The client-side "GroupMember" has three additional fields
> > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > agree
> > > > that broker-side assignor would not need reason/metadata since they are
> > > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > > it
> > > > possible that we evolve our broker-side built-in assignor but the old
> > > > versioned clients would not be able to work with the new version, in
> > > which
> > > > case we need to let the broker being aware of this and upgrade its
> > > behavior
> > > > to cooperate with the clients?
> > >
> > > I don't really see the benefits here because server side assignors
> > > don't have metadata at all. They only assign topic-partitions. They
> > > are not supposed to generate metadata nor to receive metadata from the
> > > members.
> > >
> > > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > > of
> > > > including these three fields, what about just adding the "Metadata"
> > field
> > > > class which has these three fields? Also, there are two "Metadata"
> > > > currently in the APIs, the first is a class that encodes
> > > > reason/version/metadata, and the second is just the encoded metadata
> > > bytes.
> > > > I'm wondering what about just naming the first as memberMetadata, which
> > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > bytebuffer field as metadataBytes?
> > >
> > > That's a good point. Let me try to rationalize this interface based on
> > > your suggestions.
> > >
> > > Best,
> > > David
> > >
> > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > >
> > > > Hello David,
> > > >
> > > > Just had a few more nit questions about the public APIs:
> > > >
> > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > field,
> > > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > > rationale seems to be assuming that we should always be able to do the
> > > > assignment at the broker-side assignor without errors. Personally I
> > think
> > > > it's still potentially beneficial to add the Error field even for
> > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > topics are not recognized with the current broker's metadata. What do
> > you
> > > > think?
> > > >
> > > > 2) The client-side "GroupMember" has three additional fields
> > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > agree
> > > > that broker-side assignor would not need reason/metadata since they are
> > > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > > it
> > > > possible that we evolve our broker-side built-in assignor but the old
> > > > versioned clients would not be able to work with the new version, in
> > > which
> > > > case we need to let the broker being aware of this and upgrade its
> > > behavior
> > > > to cooperate with the clients?
> > > >
> > > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > > of
> > > > including these three fields, what about just adding the "Metadata"
> > field
> > > > class which has these three fields? Also, there are two "Metadata"
> > > > currently in the APIs, the first is a class that encodes
> > > > reason/version/metadata, and the second is just the encoded metadata
> > > bytes.
> > > > I'm wondering what about just naming the first as memberMetadata, which
> > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > bytebuffer field as metadataBytes?
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > >
> > > > > Hello David,
> > > > >
> > > > > Thanks for bringing this question up. I think the main benefits as
> > you
> > > > > listed is 2) above if it stays; just to clarify, we would only be
> > able
> > > to
> > > > > save one round trip if the rebalance is still triggered by the
> > broker;
> > > if
> > > > > the rebalance is triggered by the client then the num.round.trips are
> > > the
> > > > > same:
> > > > >
> > > > > 1) With GroupPrepareAssignment:
> > > > >
> > > > > T0: client decides to do a new assignment, suppose it has already
> > sent
> > > a
> > > > > HB and hence has to wait for it to return first since only one
> > request
> > > /
> > > > > response can be inflight with the coordinator's socket.
> > > > > T1: client receives the HB response, and then sends the
> > > > > GroupPrepareAssignment request.
> > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > T3: it calculates the new assignment, and sends a
> > > GroupInstallAssignment
> > > > > request.
> > > > >
> > > > > In total, two round trips.
> > > > >
> > > > > 2) Without GroupPrepareAssignment:
> > > > >
> > > > > T0: client decides to do a new assignment, suppose it has already
> > sent
> > > a
> > > > > HB and hence has to wait for it to return first since only one
> > request
> > > /
> > > > > response can be inflight with the coordinator's socket.
> > > > > T1: client receives the HB response, and then sends the new HB
> > request
> > > > > with the flag indicating a new rebalance needed..
> > > > > T2: the HB response with the optional member metadata map is
> > returned.
> > > > > T3: it calculates the new assignment, and sends a
> > > GroupInstallAssignment
> > > > > request.
> > > > >
> > > > > In total, two round trips as well.
> > > > >
> > > > > -----------------------------
> > > > >
> > > > > So to complete the full picture here, we'd need to modify both HB
> > > request
> > > > > and response so that the client can also indicate a new rebalance via
> > > the
> > > > > HB request as well, right?
> > > > >
> > > > > Assuming all above is true, I think it's okay to merge the
> > > > > GroupPrepareAssignment into HB given that we can make the additional
> > > fields
> > > > > encoding the full member (subscription) metadata and topic metadata
> > as
> > > > > optional fields.
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > <dj...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > >> Hi all,
> > > > >>
> > > > >> During an offline conversation, someone asked why we need the
> > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > instead
> > > > >> provide the group state in the heartbeat response. This has a few
> > > > >> advantages: 1) it does not require using a special error code to
> > > > >> signal that a new assignment is required as the signal would be the
> > > > >> provided group state; 2) it removes one round trip when a client
> > side
> > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > >> response's definition quite large. I recall that I went with the
> > > > >> current approach due to this.
> > > > >>
> > > > >> Providing the group state in the heartbeat response is appealing.
> > What
> > > > >> do you guys think?
> > > > >>
> > > > >> Best,
> > > > >> David
> > > > >>
> > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > > >> >
> > > > >> > Hi Guozhang,
> > > > >> >
> > > > >> > 1. I have added a reference to the relevant chapter instead of
> > > > >> > repeating the whole thing. Does that work for you?
> > > > >> >
> > > > >> > 2. The "Rebalance Triggers" section you are referring to is about
> > > when
> > > > >> > a rebalance should be triggered for the non-upgraded members using
> > > the
> > > > >> > old protocol. The section mentions that a rebalance must be
> > > triggered
> > > > >> > when a new assignment is installed. This implies that the group
> > > epoch
> > > > >> > was updated either by a native member or a non-upgraded member.
> > For
> > > > >> > the latter, the JoinGroup request would be the trigger. I have
> > > added a
> > > > >> > reference to the relevant chapter in the "JoinGroup Handling"
> > > section
> > > > >> > as well. Does that make sense?
> > > > >> >
> > > > >> > Thanks,
> > > > >> > David
> > > > >> >
> > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wangguoz@gmail.com
> > >
> > > > >> wrote:
> > > > >> > >
> > > > >> > > Hello David,
> > > > >> > >
> > > > >> > > Alright I think that's sufficient. Just to make that clear in
> > the
> > > doc,
> > > > >> > > could we update:
> > > > >> > >
> > > > >> > > 1) the heartbeat request handling section, stating when
> > > coordinator
> > > > >> will
> > > > >> > > trigger rebalance based on the HB's member metadata / reason?
> > > > >> > > 2) the "Rebalance Triggers" section to include what we described
> > > in
> > > > >> "Group
> > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > >> > >
> > > > >> > >
> > > > >> > > Guozhang
> > > > >> > >
> > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > wrote:
> > > > >> > >
> > > > >> > > > Hi Guozhang,
> > > > >> > > >
> > > > >> > > > I thought that the assignor will always be consulted when the
> > > next
> > > > >> > > > heartbeat request is constructed. In other words,
> > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > heartbeat.
> > > > >> This
> > > > >> > > > gives the opportunity for the assignor to enforce a rebalance
> > by
> > > > >> > > > setting the reason to a non-zero value or by changing the
> > > bytes. Do
> > > > >> > > > you think that this is not sufficient? Are you concerned by
> > the
> > > > >> delay?
> > > > >> > > >
> > > > >> > > > Best,
> > > > >> > > > David
> > > > >> > > >
> > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > >> wrote:
> > > > >> > > > >
> > > > >> > > > > Hello David,
> > > > >> > > > >
> > > > >> > > > > One of Jun's comments make me thinking:
> > > > >> > > > >
> > > > >> > > > > ```
> > > > >> > > > > In this case, a new assignment is triggered by the client
> > side
> > > > >> > > > > assignor. When constructing the HB, the consumer will always
> > > > >> consult
> > > > >> > > > > the client side assignor and propagate the information to
> > the
> > > > >> group
> > > > >> > > > > coordinator. In other words, we don't expect users to call
> > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > >> > > > > ```
> > > > >> > > > >
> > > > >> > > > > As I looked at the current PartitionAssignor's interface, we
> > > > >> actually do
> > > > >> > > > > not have a way yet to instruct how to construct the next HB
> > > > >> request, e.g.
> > > > >> > > > > when the assignor wants to enforce a new rebalance with a
> > new
> > > > >> assignment,
> > > > >> > > > > we'd need some customizable APIs inside the
> > PartitionAssignor
> > > to
> > > > >> indicate
> > > > >> > > > > the next HB telling broker about so. WDYT about adding such
> > an
> > > > >> API on the
> > > > >> > > > > PartitionAssignor?
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > Guozhang
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > > > wrote:
> > > > >> > > > >
> > > > >> > > > > > Hi Jun,
> > > > >> > > > > >
> > > > >> > > > > > I have updated the KIP to include your feedback. I have
> > also
> > > > >> tried to
> > > > >> > > > > > clarify the parts which were not cleared.
> > > > >> > > > > >
> > > > >> > > > > > Best,
> > > > >> > > > > > David
> > > > >> > > > > >
> > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > djacot@confluent.io
> > > > >> >
> > > > >> > > > wrote:
> > > > >> > > > > > >
> > > > >> > > > > > > Hi Jun,
> > > > >> > > > > > >
> > > > >> > > > > > > Thanks for your feedback. Let me start by answering your
> > > > >> questions
> > > > >> > > > > > > inline and I will update the KIP next week.
> > > > >> > > > > > >
> > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > > KIP
> > > > >> seem to
> > > > >> > > > be
> > > > >> > > > > > fewer
> > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > >> wildcard. A few
> > > > >> > > > > > > > comments below.
> > > > >> > > > > > >
> > > > >> > > > > > > I would also add that the KIP removes the global sync
> > > barrier
> > > > >> in the
> > > > >> > > > > > > protocol which is essential to improve group stability
> > and
> > > > >> > > > > > > scalability, and the KIP also simplifies the client by
> > > moving
> > > > >> most of
> > > > >> > > > > > > the logic to the server side.
> > > > >> > > > > > >
> > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > support
> > > > >> rolling
> > > > >> > > > > > changing
> > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > >> > > > > > >
> > > > >> > > > > > > Definitely. The group coordinator will use the assignor
> > > used
> > > > >> by a
> > > > >> > > > > > > majority of the members. This allows the group to move
> > > from
> > > > >> one
> > > > >> > > > > > > assignor to another by a roll. This is explained in the
> > > > >> Assignor
> > > > >> > > > > > > Selection chapter.
> > > > >> > > > > > >
> > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > >> required in
> > > > >> > > > every
> > > > >> > > > > > > > request or the scenarios when it needs to be filled?
> > For
> > > > >> example,
> > > > >> > > > it's
> > > > >> > > > > > not
> > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > >> > > > > > >
> > > > >> > > > > > > The client is expected to set those fields in case of a
> > > > >> connection
> > > > >> > > > > > > issue (e.g. timeout) or when the fields have changed
> > since
> > > > >> the last
> > > > >> > > > > > > HB. The server populates those fields as long as the
> > > member
> > > > >> is not
> > > > >> > > > > > > fully reconciled - the member should acknowledge that it
> > > has
> > > > >> the
> > > > >> > > > > > > expected epoch and assignment. I will clarify this in
> > the
> > > KIP.
> > > > >> > > > > > >
> > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > affinity
> > > > >> between the
> > > > >> > > > > > client
> > > > >> > > > > > > > and the broker is only considered during fetching, but
> > > not
> > > > >> during
> > > > >> > > > > > assigning
> > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > assignment
> > > is
> > > > >> made,
> > > > >> > > > there
> > > > >> > > > > > is
> > > > >> > > > > > > > no opportunity for read affinity because no replicas
> > of
> > > > >> assigned
> > > > >> > > > > > partitions
> > > > >> > > > > > > > are close to the member. I am wondering if we should
> > use
> > > > >> this
> > > > >> > > > > > opportunity
> > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > >> > > > > > >
> > > > >> > > > > > > That's an interesting idea. I don't see any issue with
> > > adding
> > > > >> the
> > > > >> > > > rack
> > > > >> > > > > > > to the members. I will do so.
> > > > >> > > > > > >
> > > > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> > > busy
> > > > >> a group
> > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > > >> that we
> > > > >> > > > could
> > > > >> > > > > > add
> > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > event
> > > > >> loop is
> > > > >> > > > doing
> > > > >> > > > > > the
> > > > >> > > > > > > > actual work.
> > > > >> > > > > > >
> > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > >> > > > > > >
> > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > >> handling? For
> > > > >> > > > > > example,
> > > > >> > > > > > > > does it need to trigger the check if any group with
> > the
> > > > >> wildcard
> > > > >> > > > > > > > subscription now has a new matching topic?
> > > > >> > > > > > >
> > > > >> > > > > > > Sure. When the new group coordinator takes over, it has
> > > to:
> > > > >> > > > > > > * Setup the session timeouts.
> > > > >> > > > > > > * Trigger a new assignment if a client side assignor is
> > > used.
> > > > >> We
> > > > >> > > > don't
> > > > >> > > > > > > store the information about the member selected to run
> > the
> > > > >> assignment
> > > > >> > > > > > > so we have to start a new one.
> > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > >> subscriptions, and
> > > > >> > > > > > > trigger a rebalance if needed.
> > > > >> > > > > > >
> > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > > >> the epoch
> > > > >> > > > > > field
> > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > >> ConsumerGroupMetadataValue
> > > > >> > > > > > reflect
> > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > >> > > > > > >
> > > > >> > > > > > > Sure. I will clarify that but it is always the latest
> > > group
> > > > >> epoch.
> > > > >> > > > > > > When the group state is updated, the group epoch is
> > > bumped so
> > > > >> we use
> > > > >> > > > > > > that one for all the change records related to the
> > update.
> > > > >> > > > > > >
> > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > following
> > > > >> > > > invariants
> > > > >> > > > > > are
> > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > > >> not to
> > > > >> > > > get any
> > > > >> > > > > > > > assigned partitions, right?
> > > > >> > > > > > >
> > > > >> > > > > > > That's right. Here I meant that the members provided by
> > > the
> > > > >> assignor
> > > > >> > > > > > > in the assignment must exist in the group. The assignor
> > > can
> > > > >> not make
> > > > >> > > > > > > up new member ids.
> > > > >> > > > > > >
> > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > equals to
> > > > >> 0":
> > > > >> > > > When
> > > > >> > > > > > would
> > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > >> > > > > > >
> > > > >> > > > > > > A member is expected to abandon all its partitions and
> > > > >> rejoins when
> > > > >> > > > it
> > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case,
> > the
> > > > >> group
> > > > >> > > > > > > coordinator will have removed the member from the group.
> > > The
> > > > >> member
> > > > >> > > > > > > can rejoin the group with the same member id but with 0
> > as
> > > > >> epoch. Let
> > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > >> > > > > > >
> > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger a
> > > > >> > > > > > reassignment
> > > > >> > > > > > > > by either providing a non-zero reason or by updating
> > the
> > > > >> assignor
> > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > > >> deprecation
> > > > >> > > > of
> > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > >> > > > > > >
> > > > >> > > > > > > In this case, a new assignment is triggered by the
> > client
> > > side
> > > > >> > > > > > > assignor. When constructing the HB, the consumer will
> > > always
> > > > >> consult
> > > > >> > > > > > > the client side assignor and propagate the information
> > to
> > > the
> > > > >> group
> > > > >> > > > > > > coordinator. In other words, we don't expect users to
> > call
> > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > >> > > > > > >
> > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > section
> > > > >> seems to
> > > > >> > > > have
> > > > >> > > > > > > > multiple typos.
> > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > immediately
> > > > >> gets into
> > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > > get
> > > > >> into
> > > > >> > > > > > epoch=3,
> > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > >> epoch=2,
> > > > >> > > > > > > > partitions=[foo-0].
> > > > >> > > > > > >
> > > > >> > > > > > > Sorry for that! I will revise them.
> > > > >> > > > > > >
> > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > upgrade
> > > > >> from
> > > > >> > > > any
> > > > >> > > > > > old
> > > > >> > > > > > > > version to new one?
> > > > >> > > > > > >
> > > > >> > > > > > > We will support upgrading from the consumer protocol
> > > version
> > > > >> 3,
> > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> > > the
> > > > >> earliest
> > > > >> > > > > > > version is unknown at the moment. This is explained in
> > the
> > > > >> migration
> > > > >> > > > > > > plan chapter.
> > > > >> > > > > > >
> > > > >> > > > > > > Thanks again for your feedback, Jun. I will update the
> > KIP
> > > > >> based on
> > > > >> > > > it
> > > > >> > > > > > > next week.
> > > > >> > > > > > >
> > > > >> > > > > > > Best,
> > > > >> > > > > > > David
> > > > >> > > > > > >
> > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > >> <ju...@confluent.io.invalid>
> > > > >> > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > Hi, David,
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > > KIP
> > > > >> seem to
> > > > >> > > > be
> > > > >> > > > > > fewer
> > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > >> wildcard. A few
> > > > >> > > > > > > > comments below.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > support
> > > > >> rolling
> > > > >> > > > > > changing
> > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > >> required in
> > > > >> > > > every
> > > > >> > > > > > > > request or the scenarios when it needs to be filled?
> > For
> > > > >> example,
> > > > >> > > > it's
> > > > >> > > > > > not
> > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > affinity
> > > > >> between the
> > > > >> > > > > > client
> > > > >> > > > > > > > and the broker is only considered during fetching, but
> > > not
> > > > >> during
> > > > >> > > > > > assigning
> > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > assignment
> > > is
> > > > >> made,
> > > > >> > > > there
> > > > >> > > > > > is
> > > > >> > > > > > > > no opportunity for read affinity because no replicas
> > of
> > > > >> assigned
> > > > >> > > > > > partitions
> > > > >> > > > > > > > are close to the member. I am wondering if we should
> > use
> > > > >> this
> > > > >> > > > > > opportunity
> > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> > > busy
> > > > >> a group
> > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > > >> that we
> > > > >> > > > could
> > > > >> > > > > > add
> > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > event
> > > > >> loop is
> > > > >> > > > doing
> > > > >> > > > > > the
> > > > >> > > > > > > > actual work.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > >> handling? For
> > > > >> > > > > > example,
> > > > >> > > > > > > > does it need to trigger the check if any group with
> > the
> > > > >> wildcard
> > > > >> > > > > > > > subscription now has a new matching topic?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > > >> the epoch
> > > > >> > > > > > field
> > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > >> ConsumerGroupMetadataValue
> > > > >> > > > > > reflect
> > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > following
> > > > >> > > > invariants
> > > > >> > > > > > are
> > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > > >> not to
> > > > >> > > > get any
> > > > >> > > > > > > > assigned partitions, right?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > equals to
> > > > >> 0":
> > > > >> > > > When
> > > > >> > > > > > would
> > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger a
> > > > >> > > > > > reassignment
> > > > >> > > > > > > > by either providing a non-zero reason or by updating
> > the
> > > > >> assignor
> > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > > >> deprecation
> > > > >> > > > of
> > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > section
> > > > >> seems to
> > > > >> > > > have
> > > > >> > > > > > > > multiple typos.
> > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > immediately
> > > > >> gets into
> > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > > get
> > > > >> into
> > > > >> > > > > > epoch=3,
> > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > >> epoch=2,
> > > > >> > > > > > > > partitions=[foo-0].
> > > > >> > > > > > > >
> > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > upgrade
> > > > >> from
> > > > >> > > > any
> > > > >> > > > > > old
> > > > >> > > > > > > > version to new one?
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks,
> > > > >> > > > > > > >
> > > > >> > > > > > > > Jun
> > > > >> > > > > > > >
> > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > > Hi all,
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > coordinator
> > > > >> in
> > > > >> > > > Java. I
> > > > >> > > > > > > > > discussed this offline with a few folks and folks
> > are
> > > > >> concerned
> > > > >> > > > that
> > > > >> > > > > > > > > we could introduce many regressions in the old
> > > protocol
> > > > >> if we do
> > > > >> > > > so.
> > > > >> > > > > > > > > Therefore, I am going to remove this statement from
> > > the
> > > > >> KIP. It
> > > > >> > > > is an
> > > > >> > > > > > > > > implementation detail after all so it does not have
> > > to be
> > > > >> > > > decided at
> > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > refactor the
> > > > >> > > > current
> > > > >> > > > > > > > > implementation as a first step.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Cheers,
> > > > >> > > > > > > > > David
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > >> djacot@confluent.io
> > > > >> > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Hi Luke,
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > assigning,
> > > > >> > > > > > reconciling,
> > > > >> > > > > > > > > stable,
> > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > section,
> > > > >> right?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > This sentence does not refer to those group states
> > > but
> > > > >> rather
> > > > >> > > > to a
> > > > >> > > > > > > > > > state machine replication (SMR). This refers to
> > the
> > > > >> entire
> > > > >> > > > state of
> > > > >> > > > > > > > > > group coordinator which is replicated via the log
> > > > >> layer. I will
> > > > >> > > > > > > > > > clarify this in the KIP.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > modelled
> > > > >> as an
> > > > >> > > > event
> > > > >> > > > > > > > > loop"?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > The idea is to follow a model similar to the new
> > > quorum
> > > > >> > > > > > controller. We
> > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > >> __consumer_offsets
> > > > >> > > > > > > > > > partition is assigned to a unique thread and all
> > the
> > > > >> events
> > > > >> > > > (e.g.
> > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > > >> thread. This
> > > > >> > > > > > simplify
> > > > >> > > > > > > > > > concurrency and will enable us to do simulation
> > > testing
> > > > >> for the
> > > > >> > > > > > group
> > > > >> > > > > > > > > > coordinator.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > >> *__consumer_offsets*
> > > > >> > > > > > > > > partitions?
> > > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> > by
> > > a
> > > > >> group
> > > > >> > > > > > coordinator?
> > > > >> > > > > > > > > For
> > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > consumer
> > > > >> groups, and
> > > > >> > > > > > both
> > > > >> > > > > > > > > exist in
> > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > > >> for it,
> > > > >> > > > or
> > > > >> > > > > > 2?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > > the
> > > > >> number
> > > > >> > > > of
> > > > >> > > > > > threads
> > > > >> > > > > > > > > used
> > > > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> > > the
> > > > >> purpose
> > > > >> > > > of
> > > > >> > > > > > the
> > > > >> > > > > > > > > threads
> > > > >> > > > > > > > > > > is only to keep the state of each consumer group
> > > (or
> > > > >> > > > > > > > > *__consumer_offsets*
> > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > should we
> > > > >> need
> > > > >> > > > > > > > > multi-threads
> > > > >> > > > > > > > > > > here?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to shard
> > the
> > > > >> > > > processing as
> > > > >> > > > > > the
> > > > >> > > > > > > > > > computation could be heavy.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > > >> default
> > > > >> > > > session
> > > > >> > > > > > > > > timeout not
> > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > > the
> > > > >> min/max
> > > > >> > > > > > session
> > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > no?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > >> timeout to
> > > > >> > > > detect
> > > > >> > > > > > client
> > > > >> > > > > > > > > > > failures when using the consumer group protocol.
> > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > The
> > > > >> minimum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > The
> > > > >> maximum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > This is indeed a mistake. The default session
> > > timeout
> > > > >> should
> > > > >> > > > be 45s
> > > > >> > > > > > > > > > (the current default).
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > [range,
> > > > >> uniform],
> > > > >> > > > > > which means
> > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > know
> > > > >> why not
> > > > >> > > > > > uniform
> > > > >> > > > > > > > > one? I
> > > > >> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > > >> (former
> > > > >> > > > sticky
> > > > >> > > > > > > > > assinor)
> > > > >> > > > > > > > > > > for better evenly distribution. Any other reason
> > > we
> > > > >> choose
> > > > >> > > > range
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > as default?
> > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > > >> server side
> > > > >> > > > > > assignors.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > The order on the server side has no influence
> > > because
> > > > >> the
> > > > >> > > > client
> > > > >> > > > > > must
> > > > >> > > > > > > > > > chose the selector that he wants to use. There is
> > no
> > > > >> default
> > > > >> > > > in the
> > > > >> > > > > > > > > > current proposal. If the assignor is not specified
> > > by
> > > > >> the
> > > > >> > > > client,
> > > > >> > > > > > the
> > > > >> > > > > > > > > > request is rejected. The default client value for
> > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > > that my
> > > > >> > > > answers
> > > > >> > > > > > help
> > > > >> > > > > > > > > > to clarify things. I will update the KIP as well
> > > based
> > > > >> on your
> > > > >> > > > > > > > > > feedback.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Cheers,
> > > > >> > > > > > > > > > David
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > >> showuon@gmail.com>
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Hi David,
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thanks for the update.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Some more questions:
> > > > >> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > >> > > > > > > > > > > > The new group coordinator will have a state
> > > machine
> > > > >> per
> > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where each
> > state
> > > > >> machine is
> > > > >> > > > > > modelled
> > > > >> > > > > > > > > as an
> > > > >> > > > > > > > > > > event loop. Those state machines will be
> > executed
> > > in
> > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > assigning,
> > > > >> > > > > > reconciling,
> > > > >> > > > > > > > > stable,
> > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > section,
> > > > >> right?
> > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > modelled
> > > > >> as an
> > > > >> > > > event
> > > > >> > > > > > > > > loop"?
> > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > >> *__consumer_offsets*
> > > > >> > > > > > > > > partitions?
> > > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> > by
> > > a
> > > > >> group
> > > > >> > > > > > coordinator?
> > > > >> > > > > > > > > For
> > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > consumer
> > > > >> groups, and
> > > > >> > > > > > both
> > > > >> > > > > > > > > exist in
> > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > > >> for it,
> > > > >> > > > or
> > > > >> > > > > > 2?
> > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > > the
> > > > >> number
> > > > >> > > > of
> > > > >> > > > > > threads
> > > > >> > > > > > > > > used
> > > > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> > > the
> > > > >> purpose
> > > > >> > > > of
> > > > >> > > > > > the
> > > > >> > > > > > > > > threads
> > > > >> > > > > > > > > > > is only to keep the state of each consumer group
> > > (or
> > > > >> > > > > > > > > *__consumer_offsets*
> > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > should we
> > > > >> need
> > > > >> > > > > > > > > multi-threads
> > > > >> > > > > > > > > > > here?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 2. For the default value in the new configs:
> > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > > >> default
> > > > >> > > > session
> > > > >> > > > > > > > > timeout not
> > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > > the
> > > > >> min/max
> > > > >> > > > > > session
> > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > no?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > >> timeout to
> > > > >> > > > detect
> > > > >> > > > > > client
> > > > >> > > > > > > > > > > failures when using the consumer group protocol.
> > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > The
> > > > >> minimum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > The
> > > > >> maximum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > [range,
> > > > >> uniform],
> > > > >> > > > > > which means
> > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > know
> > > > >> why not
> > > > >> > > > > > uniform
> > > > >> > > > > > > > > one? I
> > > > >> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > > >> (former
> > > > >> > > > sticky
> > > > >> > > > > > > > > assinor)
> > > > >> > > > > > > > > > > for better evenly distribution. Any other reason
> > > we
> > > > >> choose
> > > > >> > > > range
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > as default?
> > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > > >> server side
> > > > >> > > > > > assignors.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thank you.
> > > > >> > > > > > > > > > > Luke
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > > >> showuon@gmail.com
> > > > >> > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > Hi Sagar,
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > >> integrating with
> > > > >> > > > > > KIP-848,
> > > > >> > > > > > > > > but I
> > > > >> > > > > > > > > > > > think we should have a separate discussion
> > > thread
> > > > >> for the
> > > > >> > > > Kafka
> > > > >> > > > > > > > > Connect
> > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > Consumer
> > > > >> Rebalance
> > > > >> > > > > > Protocol
> > > > >> > > > > > > > > [1],
> > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > consumer
> > > > >> rebalance
> > > > >> > > > > > protocol,
> > > > >> > > > > > > > > WDYT?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > [1]
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > Thank you.
> > > > >> > > > > > > > > > > > Luke
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > > Looks
> > > > >> like
> > > > >> > > > there's
> > > > >> > > > > > > > > agreement on
> > > > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> > > revisit
> > > > >> the doc
> > > > >> > > > and
> > > > >> > > > > > see
> > > > >> > > > > > > > > what
> > > > >> > > > > > > > > > > >> changes are to be made.
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> Thanks!
> > > > >> > > > > > > > > > > >> Sagar.
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > > > > >> wrote:
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > > That's
> > > > >> really
> > > > >> > > > > > helpful. I
> > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > > and the
> > > > >> > > > Consumer
> > > > >> > > > > > could
> > > > >> > > > > > > > > share
> > > > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > > > >> difference is
> > > > >> > > > that
> > > > >> > > > > > the
> > > > >> > > > > > > > > Consumer
> > > > >> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > > > >> Connect
> > > > >> > > > assigns
> > > > >> > > > > > tasks
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > support
> > > > >> different
> > > > >> > > > > > resource
> > > > >> > > > > > > > > types
> > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect.
> > The
> > > > >> dedicated
> > > > >> > > > APIs
> > > > >> > > > > > > > > would be
> > > > >> > > > > > > > > > > >> > similar to the new ones but different on
> > the
> > > > >> > > > > > content/resources and
> > > > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > > > >> coordinator
> > > > >> > > > side.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > > not a
> > > > >> fan of
> > > > >> > > > > > > > > overcharging
> > > > >> > > > > > > > > > > >> > APIs to serve different purposes. That
> > being
> > > > >> said, I am
> > > > >> > > > not
> > > > >> > > > > > > > > opposed to
> > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > I think that we can continue to discuss it
> > > here
> > > > >> for now
> > > > >> > > > in
> > > > >> > > > > > order
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible with
> > what
> > > we
> > > > >> will do
> > > > >> > > > for
> > > > >> > > > > > > > > Connect in
> > > > >> > > > > > > > > > > >> > the future.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Best,
> > > > >> > > > > > > > > > > >> > David
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot
> > <
> > > > >> > > > > > djacot@confluent.io>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > Hi all,
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > through
> > > and
> > > > >> address
> > > > >> > > > > > your
> > > > >> > > > > > > > > comments
> > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > feedback.
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > Cheers,
> > > > >> > > > > > > > > > > >> > > David
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > > Harris
> > > > >> <
> > > > >> > > > > > > > > gharris1727@gmail.com
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to
> > see
> > > > >> > > > cooperative
> > > > >> > > > > > > > > rebalancing
> > > > >> > > > > > > > > > > >> > making it
> > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > > > >> describes how
> > > > >> > > > > > member
> > > > >> > > > > > > > > epoch
> > > > >> > > > > > > > > > > >> > should
> > > > >> > > > > > > > > > > >> > > > advance in step with the group epoch
> > and
> > > > >> assignment
> > > > >> > > > > > epoch
> > > > >> > > > > > > > > values. I
> > > > >> > > > > > > > > > > >> > think
> > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > reasons
> > > > >> described
> > > > >> > > > in
> > > > >> > > > > > the
> > > > >> > > > > > > > > KIP. When
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > > >> partitions to a
> > > > >> > > > > > worker,
> > > > >> > > > > > > > > what
> > > > >> > > > > > > > > > > >> member
> > > > >> > > > > > > > > > > >> > > > epoch does each incremental assignment
> > > use?
> > > > >> Are
> > > > >> > > > member
> > > > >> > > > > > epochs
> > > > >> > > > > > > > > > > >> re-used,
> > > > >> > > > > > > > > > > >> > and
> > > > >> > > > > > > > > > > >> > > > a single member epoch can correspond to
> > > > >> multiple
> > > > >> > > > > > different
> > > > >> > > > > > > > > > > >> > (monotonically
> > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > > opaque
> > > > >> to the
> > > > >> > > > group
> > > > >> > > > > > > > > > > >> coordinator? If
> > > > >> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > > > >> > > > implementations
> > > > >> > > > > > > > > interact
> > > > >> > > > > > > > > > > >> with
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> > > meaning
> > > > >> agreed
> > > > >> > > > > > upon? If
> > > > >> > > > > > > > > so, what
> > > > >> > > > > > > > > > > >> > is the
> > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > > > >> including
> > > > >> > > > such
> > > > >> > > > > > > > > functionality
> > > > >> > > > > > > > > > > >> > in the
> > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 3. The following is included in the
> > KIP:
> > > > >> "Thanks to
> > > > >> > > > > > this, the
> > > > >> > > > > > > > > input
> > > > >> > > > > > > > > > > >> of
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > client side assignor is entirely driven
> > > by
> > > > >> the group
> > > > >> > > > > > > > > coordinator.
> > > > >> > > > > > > > > > > >> The
> > > > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > > >> maintaining
> > > > >> > > > any
> > > > >> > > > > > state
> > > > >> > > > > > > > > besides
> > > > >> > > > > > > > > > > >> its
> > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > > that the
> > > > >> > > > > > client-side
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > >> MAY
> > > > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> > state
> > > > >> (such as
> > > > >> > > > > > partition
> > > > >> > > > > > > > > > > >> > throughput,
> > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics,
> > etc),
> > > or
> > > > >> that
> > > > >> > > > > > additional
> > > > >> > > > > > > > > > > >> > non-Metadata
> > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > classes
> > > > >> > > > > > > > > > > >> > > > for
> > > > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > >> > > > > > > > > > > >> > > > and
> > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > >> > > > > > that
> > > > >> > > > > > > > > seem to
> > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it possible
> > for
> > > > >> these two
> > > > >> > > > > > > > > implementations
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> > > feature
> > > > >> parity
> > > > >> > > > of
> > > > >> > > > > > > > > server-side
> > > > >> > > > > > > > > > > >> and
> > > > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > > >> facilitate
> > > > >> > > > > > operational
> > > > >> > > > > > > > > > > >> > flexibility in
> > > > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > > >> server-side
> > > > >> > > > > > assignor
> > > > >> > > > > > > > > has some
> > > > >> > > > > > > > > > > >> > poor
> > > > >> > > > > > > > > > > >> > > > behavior and needs a patch, deploying
> > the
> > > > >> patched
> > > > >> > > > > > assignor to
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > > >> client-side
> > > > >> > > > > > assignor
> > > > >> > > > > > > > > may be
> > > > >> > > > > > > > > > > >> > faster
> > > > >> > > > > > > > > > > >> > > > and less risky than patching all of the
> > > > >> brokers.
> > > > >> > > > With
> > > > >> > > > > > the
> > > > >> > > > > > > > > currently
> > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > > >> > > > reimplementation
> > > > >> > > > > > would
> > > > >> > > > > > > > > have
> > > > >> > > > > > > > > > > >> to be
> > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > > diverged
> > > > >> > > > > > significantly,
> > > > >> > > > > > > > > then it
> > > > >> > > > > > > > > > > >> is
> > > > >> > > > > > > > > > > >> > > > possible that a reimplementation would
> > > not be
> > > > >> > > > possible.
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > --
> > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > discuss
> > > how
> > > > >> Connect
> > > > >> > > > > > would
> > > > >> > > > > > > > > need to
> > > > >> > > > > > > > > > > >> > change
> > > > >> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> > > > >> Here's the
> > > > >> > > > page:
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > > tried
> > > > >> to keep
> > > > >> > > > a
> > > > >> > > > > > format
> > > > >> > > > > > > > > > > >> similar to
> > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> > > Also,
> > > > >> do you
> > > > >> > > > > > think this
> > > > >> > > > > > > > > > > >> should
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion thread
> > > or
> > > > >> is this
> > > > >> > > > one
> > > > >> > > > > > fine?
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > Sagar <
> > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > > Kafka
> > > > >> Streams.
> > > > >> > > > > > Sure, I
> > > > >> > > > > > > > > would do
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > > >> document.
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > > Guozhang
> > > > >> Wang <
> > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> It would be great if you could
> > come
> > > > >> back with
> > > > >> > > > some
> > > > >> > > > > > > > > analysis on
> > > > >> > > > > > > > > > > >> > how to
> > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > integration
> > > > >> with
> > > > >> > > > the new
> > > > >> > > > > > > > > protocol;
> > > > >> > > > > > > > > > > >> so
> > > > >> > > > > > > > > > > >> > far
> > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > > "protocol
> > > > >> type"
> > > > >> > > > we
> > > > >> > > > > > did not
> > > > >> > > > > > > > > yet
> > > > >> > > > > > > > > > > >> think
> > > > >> > > > > > > > > > > >> > > > > through
> > > > >> > > > > > > > > > > >> > > > > >> the Connect side implementations.
> > > For
> > > > >> Streams
> > > > >> > > > > > here's a
> > > > >> > > > > > > > > draft of
> > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> > > Connect.
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> > > Sagar <
> > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> > > > >> reason I
> > > > >> > > > thought
> > > > >> > > > > > > > > connect can
> > > > >> > > > > > > > > > > >> > also fit
> > > > >> > > > > > > > > > > >> > > > > >> into
> > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that even
> > > today the
> > > > >> > > > connect
> > > > >> > > > > > uses a
> > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > AbstractCoordinator to
> > > > >> empower
> > > > >> > > > > > > > > rebalances of
> > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > >> WorkerCoordinator sets
> > > > >> > > > the
> > > > >> > > > > > > > > > > >> protocolType()
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata() method
> > by
> > > > >> plumbing
> > > > >> > > > into
> > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > > > >> connect would
> > > > >> > > > be
> > > > >> > > > > > > > > similar at a
> > > > >> > > > > > > > > > > >> > high
> > > > >> > > > > > > > > > > >> > > > > level
> > > > >> > > > > > > > > > > >> > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> > > because
> > > > >> of the
> > > > >> > > > > > Client
> > > > >> > > > > > > > > side
> > > > >> > > > > > > > > > > >> > assignors
> > > > >> > > > > > > > > > > >> > > > > being
> > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > implementation
> > > > >> level, we
> > > > >> > > > > > might
> > > > >> > > > > > > > > need to
> > > > >> > > > > > > > > > > >> make
> > > > >> > > > > > > > > > > >> > a lot
> > > > >> > > > > > > > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > > > >> assignment
> > > > >> > > > protocol
> > > > >> > > > > > like
> > > > >> > > > > > > > > > > >> enhancing
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > > > >> SyncGroup and
> > > > >> > > > > > using
> > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar lines
> > to
> > > > >> streams (or
> > > > >> > > > > > there
> > > > >> > > > > > > > > might be
> > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> > > > >> analysis of
> > > > >> > > > the
> > > > >> > > > > > same
> > > > >> > > > > > > > > and we
> > > > >> > > > > > > > > > > >> > can have
> > > > >> > > > > > > > > > > >> > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > separate discussion thread for
> > > that
> > > > >> as that
> > > > >> > > > would
> > > > >> > > > > > > > > derail this
> > > > >> > > > > > > > > > > >> > > > > discussion
> > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if that
> > sounds
> > > > >> good to
> > > > >> > > > you.
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> > > David
> > > > >> Jacot
> > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > >> `Assignment#error`.
> > > > >> > > > > > Sure, I
> > > > >> > > > > > > > > can
> > > > >> > > > > > > > > > > >> > mention it.
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> > > from
> > > > >> his
> > > > >> > > > current
> > > > >> > > > > > > > > assignment
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > his
> > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when he can
> > > move
> > > > >> to
> > > > >> > > > epoch 3.
> > > > >> > > > > > When
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > happens,
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is updated
> > and
> > > > >> persisted
> > > > >> > > > > > with all
> > > > >> > > > > > > > > its
> > > > >> > > > > > > > > > > >> > assigned
> > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if they are
> > not
> > > all
> > > > >> revoked
> > > > >> > > > > > yet. In
> > > > >> > > > > > > > > other
> > > > >> > > > > > > > > > > >> > words, the
> > > > >> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> > > target
> > > > >> > > > > > assignment.
> > > > >> > > > > > > > > This is
> > > > >> > > > > > > > > > > >> > basically
> > > > >> > > > > > > > > > > >> > > > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid having
> > to
> > > > >> write all
> > > > >> > > > the
> > > > >> > > > > > > > > changes to
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > log.
> > > > >> > > > > > > > > > > >> > > > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on the
> > > persisted
> > > > >> state
> > > > >> > > > so I
> > > > >> > > > > > > > > understand
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> > > > >> improve
> > > > >> > > > this in
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> > description.
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> > > > >> reuse the
> > > > >> > > > > > protocol
> > > > >> > > > > > > > > with a
> > > > >> > > > > > > > > > > >> > client side
> > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > > > >> protocol. The
> > > > >> > > > > > assignment
> > > > >> > > > > > > > > is
> > > > >> > > > > > > > > > > >> about
> > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> > > could
> > > > >> > > > Connect
> > > > >> > > > > > fit
> > > > >> > > > > > > > > into this?
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55
> > PM
> > > > >> Sagar <
> > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just
> > had
> > > > >> minor
> > > > >> > > > > > observations:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > > > >> section in
> > > > >> > > > Client
> > > > >> > > > > > Side
> > > > >> > > > > > > > > mode
> > > > >> > > > > > > > > > > >> > Assignment
> > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this
> > > case,
> > > > >> the
> > > > >> > > > client
> > > > >> > > > > > side
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > >> can
> > > > >> > > > > > > > > > > >> > > > > return
> > > > >> > > > > > > > > > > >> > > > > >> an
> > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > coordinator`.
> > > > >> In this
> > > > >> > > > > > case are
> > > > >> > > > > > > > > you
> > > > >> > > > > > > > > > > >> > referring to
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > > >> AssignmentError
> > > > >> > > > that's
> > > > >> > > > > > > > > listed down
> > > > >> > > > > > > > > > > >> > towards
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it
> > would
> > > > >> make sense
> > > > >> > > > to
> > > > >> > > > > > > > > mention this
> > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > > section, I
> > > > >> have a
> > > > >> > > > > > slight
> > > > >> > > > > > > > > > > >> confusion,
> > > > >> > > > > > > > > > > >> > not
> > > > >> > > > > > > > > > > >> > > > > sure
> > > > >> > > > > > > > > > > >> > > > > >> if
> > > > >> > > > > > > > > > > >> > > > > >> > > > others have the same.
> > Consider
> > > > >> this step:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> > > > >> coordinator
> > > > >> > > > > > > > > transitions him
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > epoch
> > > > >> > > > > > > > > > > >> > > > > 3
> > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > > revoke. It
> > > > >> > > > persists
> > > > >> > > > > > the
> > > > >> > > > > > > > > change and
> > > > >> > > > > > > > > > > >> > reply.
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > > (epoch=3)
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > partitions=[foo-0]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > partitions=[foo-2]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > partitions=[foo-1]
> > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > >> partitions=[foo-0,
> > > > >> > > > > > foo-1]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > >> partitions=[foo-2]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > >> partitions=[foo-1]
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > > transitions
> > > > >> to
> > > > >> > > > epoch 3
> > > > >> > > > > > but
> > > > >> > > > > > > > > cannot
> > > > >> > > > > > > > > > > >> get
> > > > >> > > > > > > > > > > >> > foo-1
> > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that
> > > member C
> > > > >> can't
> > > > >> > > > get
> > > > >> > > > > > the
> > > > >> > > > > > > > > foo-1
> > > > >> > > > > > > > > > > >> > partition
> > > > >> > > > > > > > > > > >> > > > > yet,
> > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > >> > > > > > > > > > > >> > > > > >> > > > based on the description
> > > above,
> > > > >> it seems
> > > > >> > > > it
> > > > >> > > > > > > > > already has
> > > > >> > > > > > > > > > > >> it.
> > > > >> > > > > > > > > > > >> > Do you
> > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to remove it
> > > and
> > > > >> > > > populate it
> > > > >> > > > > > only
> > > > >> > > > > > > > > when it
> > > > >> > > > > > > > > > > >> > actually
> > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> > > > >> places, so
> > > > >> > > > have
> > > > >> > > > > > I
> > > > >> > > > > > > > > > > >> understood it
> > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might
> > > be
> > > > >> out of
> > > > >> > > > scope
> > > > >> > > > > > of
> > > > >> > > > > > > > > this
> > > > >> > > > > > > > > > > >> > discussion,
> > > > >> > > > > > > > > > > >> > > > > but
> > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > > > >> probably be
> > > > >> > > > > > running in
> > > > >> > > > > > > > > client
> > > > >> > > > > > > > > > > >> > side
> > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > > > >> protocol as it
> > > > >> > > > has
> > > > >> > > > > > its
> > > > >> > > > > > > > > own
> > > > >> > > > > > > > > > > >> Custom
> > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > >> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00
> > > PM
> > > > >> David
> > > > >> > > > Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is
> > > to
> > > > >> move
> > > > >> > > > forward
> > > > >> > > > > > with
> > > > >> > > > > > > > > > > >> > specialized API
> > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> > > > >> generic API.
> > > > >> > > > For
> > > > >> > > > > > > > > Connect, we
> > > > >> > > > > > > > > > > >> > can apply
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > > > >> reuse/share the
> > > > >> > > > core
> > > > >> > > > > > > > > > > >> > implementation on
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the
> > schema
> > > > >> registry, I
> > > > >> > > > > > think
> > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > >> > should
> > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to
> > do
> > > > >> simple
> > > > >> > > > > > > > > membership/leader
> > > > >> > > > > > > > > > > >> > election.
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at
> > > 10:22
> > > > >> AM Ismael
> > > > >> > > > > > Juma <
> > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > >> java.util.regex.Pattern
> > > > >> > > > vs
> > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the differences
> > > in
> > > > >> more
> > > > >> > > > detail
> > > > >> > > > > > before
> > > > >> > > > > > > > > > > >> deciding
> > > > >> > > > > > > > > > > >> > one
> > > > >> > > > > > > > > > > >> > > > > way
> > > > >> > > > > > > > > > > >> > > > > >> or
> > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if people
> > pass
> > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > >> > > > > > > > > > > >> they
> > > > >> > > > > > > > > > > >> > expect
> > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored.
> > > If
> > > > >> we are
> > > > >> > > > doing
> > > > >> > > > > > > > > something
> > > > >> > > > > > > > > > > >> > different,
> > > > >> > > > > > > > > > > >> > > > > >> then
> > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider adding
> > an
> > > > >> overload
> > > > >> > > > with
> > > > >> > > > > > our own
> > > > >> > > > > > > > > > > >> Pattern
> > > > >> > > > > > > > > > > >> > class
> > > > >> > > > > > > > > > > >> > > > > (I
> > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to
> > expose
> > > > >> re2j's at
> > > > >> > > > this
> > > > >> > > > > > > > > point).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids,
> > > any
> > > > >> major new
> > > > >> > > > > > protocol
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> > > > > integrate
> > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> > > topic
> > > > >> > > > > > recreation case
> > > > >> > > > > > > > > > > >> > correctly.
> > > > >> > > > > > > > > > > >> > > > > That's
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle.
> > I
> > > > >> agree with
> > > > >> > > > > > David
> > > > >> > > > > > > > > that we'd
> > > > >> > > > > > > > > > > >> > want to
> > > > >> > > > > > > > > > > >> > > > > add
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > protocols
> > > > >> that don't
> > > > >> > > > > > have it
> > > > >> > > > > > > > > yet and
> > > > >> > > > > > > > > > > >> > that we
> > > > >> > > > > > > > > > > >> > > > > can
> > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> > > versus
> > > > >> adding
> > > > >> > > > new
> > > > >> > > > > > APIs
> > > > >> > > > > > > > > to the
> > > > >> > > > > > > > > > > >> > Java
> > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new
> > > APIs
> > > > >> is
> > > > >> > > > required
> > > > >> > > > > > for
> > > > >> > > > > > > > > > > >> reasonable
> > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> > > about
> > > > >> the
> > > > >> > > > > > coordinator
> > > > >> > > > > > > > > storing
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to
> > be
> > > > >> > > > centralized
> > > > >> > > > > > in the
> > > > >> > > > > > > > > > > >> metadata
> > > > >> > > > > > > > > > > >> > log for
> > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > > > >> visibility,
> > > > >> > > > > > consistency,
> > > > >> > > > > > > > > > > >> etc.).
> > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> > > > >> deleting
> > > > >> > > > > > configs in a
> > > > >> > > > > > > > > way
> > > > >> > > > > > > > > > > >> that
> > > > >> > > > > > > > > > > >> > they
> > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> > property
> > > > >> for modern
> > > > >> > > > > > systems
> > > > >> > > > > > > > > is to
> > > > >> > > > > > > > > > > >> > minimize
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> > > > >> scenarios.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at
> > > 3:47
> > > > >> PM David
> > > > >> > > > > > Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> > > answers
> > > > >> are
> > > > >> > > > below:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > path,
> > > > >> especially
> > > > >> > > > > > the last
> > > > >> > > > > > > > > step
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > > > > clients
> > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > > >> protocol, in
> > > > >> > > > which
> > > > >> > > > > > we
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > window
> > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > > old
> > > > >> > > > protocols /
> > > > >> > > > > > rpcs
> > > > >> > > > > > > > > are
> > > > >> > > > > > > > > > > >> used
> > > > >> > > > > > > > > > > >> > by
> > > > >> > > > > > > > > > > >> > > > > >> members
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > >> coordinator could
> > > > >> > > > > > "mimic"
> > > > >> > > > > > > > > the old
> > > > >> > > > > > > > > > > >> > behavior
> > > > >> > > > > > > > > > > >> > > > > >> while
> > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > something
> > > > >> we
> > > > >> > > > need to
> > > > >> > > > > > > > > present
> > > > >> > > > > > > > > > > >> about.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > published
> > > a
> > > > >> new
> > > > >> > > > version
> > > > >> > > > > > of KIP
> > > > >> > > > > > > > > which
> > > > >> > > > > > > > > > > >> > includes
> > > > >> > > > > > > > > > > >> > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about this.
> > See
> > > the
> > > > >> > > > "Supporting
> > > > >> > > > > > > > > Online
> > > > >> > > > > > > > > > > >> > Consumer
> > > > >> > > > > > > > > > > >> > > > > Group
> > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > "Compatibility,
> > > > >> > > > Deprecation,
> > > > >> > > > > > and
> > > > >> > > > > > > > > Migration
> > > > >> > > > > > > > > > > >> > Plan". I
> > > > >> > > > > > > > > > > >> > > > > >> think
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> > through
> > > a
> > > > >> few cases
> > > > >> > > > > > now but
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > overall idea
> > > > >> > > > > > > > > > > >> > > > > >> and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > > > >> understandable.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> > topic
> > > > >> ids. So
> > > > >> > > > far as
> > > > >> > > > > > > > > KIP-516 the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > >> are
> > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > admin
> > > > >> client,
> > > > >> > > > but
> > > > >> > > > > > they
> > > > >> > > > > > > > > are not
> > > > >> > > > > > > > > > > >> > exposed
> > > > >> > > > > > > > > > > >> > > > > via
> > > > >> > > > > > > > > > > >> > > > > >> any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > > think
> > > > >> the
> > > > >> > > > > > question is,
> > > > >> > > > > > > > > first
> > > > >> > > > > > > > > > > >> > should we
> > > > >> > > > > > > > > > > >> > > > > >> let
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > maintaining
> > > > >> the
> > > > >> > > > names
> > > > >> > > > > > -> ids
> > > > >> > > > > > > > > mapping
> > > > >> > > > > > > > > > > >> > itself
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > > > >> existing
> > > > >> > > > RPCs
> > > > >> > > > > > and the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> RPCs
> > > > >> > > > > > > > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > > ever
> > > > >> consider
> > > > >> > > > > > exposing
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > > (both
> > > > >> > > > > > > > > subscribe/assign, as
> > > > >> > > > > > > > > > > >> well
> > > > >> > > > > > > > > > > >> > as in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > > like
> > > > >> topic
> > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we
> > > would
> > > > >> include
> > > > >> > > > > > converting
> > > > >> > > > > > > > > all
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > offsets
> > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic
> > ids
> > > in
> > > > >> this
> > > > >> > > > KIP,
> > > > >> > > > > > the
> > > > >> > > > > > > > > consumer
> > > > >> > > > > > > > > > > >> > would be
> > > > >> > > > > > > > > > > >> > > > > >> able
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with
> > topic
> > > > >> ids. That
> > > > >> > > > > > being
> > > > >> > > > > > > > > said, it
> > > > >> > > > > > > > > > > >> > still has
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> > > various
> > > > >> APIs so
> > > > >> > > > > > having a
> > > > >> > > > > > > > > > > >> mapping
> > > > >> > > > > > > > > > > >> > in the
> > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to
> > me.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a
> > strong
> > > > >> opinion on
> > > > >> > > > > > this.
> > > > >> > > > > > > > > Here I
> > > > >> > > > > > > > > > > >> > wonder if
> > > > >> > > > > > > > > > > >> > > > > >> this
> > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of
> > this
> > > > >> KIP. I
> > > > >> > > > would
> > > > >> > > > > > rather
> > > > >> > > > > > > > > focus
> > > > >> > > > > > > > > > > >> on
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> > consider
> > > > >> this
> > > > >> > > > > > separately if
> > > > >> > > > > > > > > we see
> > > > >> > > > > > > > > > > >> > value in
> > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > Ismael's
> > > > >> point about
> > > > >> > > > > > using
> > > > >> > > > > > > > > topic ids
> > > > >> > > > > > > > > > > >> > in the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> ConsumerGroupHeartbeatRequest, I
> > > > >> > > > think
> > > > >> > > > > > that
> > > > >> > > > > > > > > there
> > > > >> > > > > > > > > > > >> is
> > > > >> > > > > > > > > > > >> > one
> > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> > > consumer
> > > > >> will
> > > > >> > > > have
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> opportunity
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists
> > before
> > > > >> passing
> > > > >> > > > them
> > > > >> > > > > > into
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> group
> > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously,
> > the
> > > > >> > > > coordinator
> > > > >> > > > > > will
> > > > >> > > > > > > > > also
> > > > >> > > > > > > > > > > >> notice
> > > > >> > > > > > > > > > > >> > it but
> > > > >> > > > > > > > > > > >> > > > > >> it
> > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way
> > to
> > > > >> reject an
> > > > >> > > > > > invalid
> > > > >> > > > > > > > > topic in
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > > David
> > > > >> on all
> > > > >> > > > other
> > > > >> > > > > > minor
> > > > >> > > > > > > > > > > >> questions
> > > > >> > > > > > > > > > > >> > > > > except
> > > > >> > > > > > > > > > > >> > > > > >> for
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > > > >> question:
> > > > >> > > > > > personally
> > > > >> > > > > > > > > I think
> > > > >> > > > > > > > > > > >> > it's not
> > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > subscribe
> > > > >> API with
> > > > >> > > > > > Pattern,
> > > > >> > > > > > > > > but
> > > > >> > > > > > > > > > > >> > instead we
> > > > >> > > > > > > > > > > >> > > > > >> still
> > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > documenting
> > > > >> that our
> > > > >> > > > > > > > > subscription may
> > > > >> > > > > > > > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > incompatible
> > > > >> case is a
> > > > >> > > > > > very rare
> > > > >> > > > > > > > > > > >> scenario
> > > > >> > > > > > > > > > > >> > I felt
> > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > > based
> > > > >> > > > > > subscription may
> > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I
> > have
> > > to
> > > > >> look at
> > > > >> > > > the
> > > > >> > > > > > > > > differences
> > > > >> > > > > > > > > > > >> > between
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > > > >> understand the
> > > > >> > > > > > potential
> > > > >> > > > > > > > > issues.
> > > > >> > > > > > > > > > > >> My
> > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work for
> > all
> > > the
> > > > >> basic
> > > > >> > > > > > regular
> > > > >> > > > > > > > > > > >> > expressions. The
> > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> > > mainly
> > > > >> about
> > > > >> > > > the
> > > > >> > > > > > various
> > > > >> > > > > > > > > > > >> character
> > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other
> > people
> > > > >> think
> > > > >> > > > about
> > > > >> > > > > > this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022
> > at
> > > > >> 11:28 PM
> > > > >> > > > > > Guozhang
> > > > >> > > > > > > > > Wang <
> > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I
> > think
> > > on
> > > > >> the high
> > > > >> > > > > > level
> > > > >> > > > > > > > > there are
> > > > >> > > > > > > > > > > >> > two meta
> > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit
> > > more:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > path,
> > > > >> especially
> > > > >> > > > > > the last
> > > > >> > > > > > > > > step
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > > > > clients
> > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > > >> protocol, in
> > > > >> > > > which
> > > > >> > > > > > we
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > window
> > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > > old
> > > > >> > > > protocols /
> > > > >> > > > > > rpcs
> > > > >> > > > > > > > > are
> > > > >> > > > > > > > > > > >> used
> > > > >> > > > > > > > > > > >> > by
> > > > >> > > > > > > > > > > >> > > > > >> members
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > >> coordinator could
> > > > >> > > > > > "mimic"
> > > > >> > > > > > > > > the old
> > > > >> > > > > > > > > > > >> > behavior
> > > > >> > > > > > > > > > > >> > > > > >> while
> > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > something
> > > > >> we
> > > > >> > > > need to
> > > > >> > > > > > > > > present
> > > > >> > > > > > > > > > > >> about.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> > topic
> > > > >> ids. So
> > > > >> > > > far as
> > > > >> > > > > > > > > KIP-516 the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > >> are
> > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > admin
> > > > >> client,
> > > > >> > > > but
> > > > >> > > > > > they
> > > > >> > > > > > > > > are not
> > > > >> > > > > > > > > > > >> > exposed
> > > > >> > > > > > > > > > > >> > > > > via
> > > > >> > > > > > > > > > > >> > > > > >> any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > > think
> > > > >> the
> > > > >> > > > > > question is,
> > > > >> > > > > > > > > first
> > > > >> > > > > > > > > > > >> > should we
> > > > >> > > > > > > > > > > >> > > > > >> let
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > maintaining
> > > > >> the
> > > > >> > > > names
> > > > >> > > > > > -> ids
> > > > >> > > > > > > > > mapping
> > > > >> > > > > > > > > > > >> > itself
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > > > >> existing
> > > > >> > > > RPCs
> > > > >> > > > > > and the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> RPCs
> > > > >> > > > > > > > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > > ever
> > > > >> consider
> > > > >> > > > > > exposing
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > > (both
> > > > >> > > > > > > > > subscribe/assign, as
> > > > >> > > > > > > > > > > >> well
> > > > >> > > > > > > > > > > >> > as in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > > like
> > > > >> topic
> > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > > David
> > > > >> on all
> > > > >> > > > other
> > > > >> > > > > > minor
> > > > >> > > > > > > > > > > >> questions
> > > > >> > > > > > > > > > > >> > > > > except
> > > > >> > > > > > > > > > > >> > > > > >> for
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > > > >> question:
> > > > >> > > > > > personally
> > > > >> > > > > > > > > I think
> > > > >> > > > > > > > > > > >> > it's not
> > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > subscribe
> > > > >> API with
> > > > >> > > > > > Pattern,
> > > > >> > > > > > > > > but
> > > > >> > > > > > > > > > > >> > instead we
> > > > >> > > > > > > > > > > >> > > > > >> still
> > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > documenting
> > > > >> that our
> > > > >> > > > > > > > > subscription may
> > > > >> > > > > > > > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > incompatible
> > > > >> case is a
> > > > >> > > > > > very rare
> > > > >> > > > > > > > > > > >> scenario
> > > > >> > > > > > > > > > > >> > I felt
> > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > > based
> > > > >> > > > > > subscription may
> > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022
> > > at
> > > > >> 5:23 AM
> > > > >> > > > > > David Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > > > >> feedback. Let
> > > > >> > > > me
> > > > >> > > > > > answer
> > > > >> > > > > > > > > your
> > > > >> > > > > > > > > > > >> > questions
> > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> > > > >> premature to
> > > > >> > > > > > talk about
> > > > >> > > > > > > > > > > >> target
> > > > >> > > > > > > > > > > >> > > > > versions
> > > > >> > > > > > > > > > > >> > > > > >> for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > > > >> existing group
> > > > >> > > > > > protocol.
> > > > >> > > > > > > > > > > >> Unlike
> > > > >> > > > > > > > > > > >> > KRaft,
> > > > >> > > > > > > > > > > >> > > > > >> this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol
> > > and
> > > > >> hence
> > > > >> > > > > > > > > deprecation/removal
> > > > >> > > > > > > > > > > >> > will be
> > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > > > >> applications
> > > > >> > > > migrate
> > > > >> > > > > > to
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense.
> > I
> > > will
> > > > >> > > > remove it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says
> > we
> > > > >> intend to
> > > > >> > > > > > release
> > > > >> > > > > > > > > this in
> > > > >> > > > > > > > > > > >> > 4.x, but
> > > > >> > > > > > > > > > > >> > > > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> > > that
> > > > >> as a
> > > > >> > > > way to
> > > > >> > > > > > > > > estimate
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > we'd
> > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > > protocol, I
> > > > >> also
> > > > >> > > > > > suggest
> > > > >> > > > > > > > > removing
> > > > >> > > > > > > > > > > >> > this part.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> > > > >> reasoning. As
> > > > >> > > > > > > > > explained, I
> > > > >> > > > > > > > > > > >> plan
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in
> > Java
> > > > >> while we
> > > > >> > > > > > implement
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> This
> > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> > > will
> > > > >> be
> > > > >> > > > slightly
> > > > >> > > > > > > > > different
> > > > >> > > > > > > > > > > >> > (e.g.
> > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I
> > wanted
> > > to
> > > > >> tighten
> > > > >> > > > the
> > > > >> > > > > > > > > switch from
> > > > >> > > > > > > > > > > >> > the old
> > > > >> > > > > > > > > > > >> > > > > >> group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the
> > > new
> > > > >> group
> > > > >> > > > > > coordinator
> > > > >> > > > > > > > > to a
> > > > >> > > > > > > > > > > >> > major
> > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would
> > > be to
> > > > >> use a
> > > > >> > > > flag
> > > > >> > > > > > to do
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > switch
> > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> > > upgrade.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to
> > > flesh
> > > > >> out the
> > > > >> > > > > > details of
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > migration
> > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we
> > > will
> > > > >> support
> > > > >> > > > > > online
> > > > >> > > > > > > > > > > >> migrations.
> > > > >> > > > > > > > > > > >> > Is
> > > > >> > > > > > > > > > > >> > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> > > > >> detail. It
> > > > >> > > > could
> > > > >> > > > > > also
> > > > >> > > > > > > > > be done
> > > > >> > > > > > > > > > > >> > as a
> > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> > support
> > > > >> online
> > > > >> > > > > > migrations
> > > > >> > > > > > > > > for the
> > > > >> > > > > > > > > > > >> > group.
> > > > >> > > > > > > > > > > >> > > > > That
> > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the
> > > old
> > > > >> protocol
> > > > >> > > > > > will be
> > > > >> > > > > > > > > able to
> > > > >> > > > > > > > > > > >> > switch to
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> > > explain
> > > > >> how that
> > > > >> > > > > > will work
> > > > >> > > > > > > > > > > >> though.
> > > > >> > > > > > > > > > > >> > It is
> > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster
> > must
> > > be
> > > > >> > > > upgraded or
> > > > >> > > > > > > > > rolled to a
> > > > >> > > > > > > > > > > >> > software
> > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > coordinator.
> > > > >> Both the
> > > > >> > > > > > old and
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > > > >> protocol and
> > > > >> > > > rely
> > > > >> > > > > > on the
> > > > >> > > > > > > > > same
> > > > >> > > > > > > > > > > >> > persisted
> > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> > > together.
> > > > >> This
> > > > >> > > > point
> > > > >> > > > > > is an
> > > > >> > > > > > > > > > > >> offline
> > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > > > >> because it
> > > > >> > > > would
> > > > >> > > > > > require
> > > > >> > > > > > > > > > > >> > shutting down
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > > > >> starting up the
> > > > >> > > > > > new one
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> that
> > > > >> > > > > > > > > > > >> > would
> > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > > > >> metadata
> > > > >> > > > > > version/IBP
> > > > >> > > > > > > > > must be
> > > > >> > > > > > > > > > > >> > upgraded
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> X
> > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> > > > >> protocol.
> > > > >> > > > This
> > > > >> > > > > > cannot
> > > > >> > > > > > > > > be done
> > > > >> > > > > > > > > > > >> > before
> > > > >> > > > > > > > > > > >> > > > > 1)
> > > > >> > > > > > > > > > > >> > > > > >> is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because
> > > the
> > > > >> old
> > > > >> > > > > > coordinator
> > > > >> > > > > > > > > doesn't
> > > > >> > > > > > > > > > > >> > support
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers
> > > must be
> > > > >> > > > upgraded
> > > > >> > > > > > to a
> > > > >> > > > > > > > > version
> > > > >> > > > > > > > > > > >> > > > > supporting
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must
> > have
> > > > >> KIP-792).
> > > > >> > > > If
> > > > >> > > > > > the
> > > > >> > > > > > > > > consumer
> > > > >> > > > > > > > > > > >> is
> > > > >> > > > > > > > > > > >> > > > > already
> > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be
> > > done at
> > > > >> this
> > > > >> > > > point.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers
> > > must
> > > > >> be rolled
> > > > >> > > > > > with the
> > > > >> > > > > > > > > > > >> feature
> > > > >> > > > > > > > > > > >> > flag
> > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> > > > >> automatically
> > > > >> > > > > > > > > converted when
> > > > >> > > > > > > > > > > >> > the first
> > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> > > protocol
> > > > >> joins
> > > > >> > > > the
> > > > >> > > > > > group.
> > > > >> > > > > > > > > While
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > members
> > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> > > being
> > > > >> > > > upgraded,
> > > > >> > > > > > the old
> > > > >> > > > > > > > > > > >> protocol
> > > > >> > > > > > > > > > > >> > is
> > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all
> > > of
> > > > >> this in
> > > > >> > > > the
> > > > >> > > > > > KIP.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy
> > > that we
> > > > >> are
> > > > >> > > > > > pushing the
> > > > >> > > > > > > > > pattern
> > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems
> > like
> > > > >> there
> > > > >> > > > could be
> > > > >> > > > > > some
> > > > >> > > > > > > > > tricky
> > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism
> > > for
> > > > >> users to
> > > > >> > > > > > detect
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> they
> > > > >> > > > > > > > > > > >> > need to
> > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching
> > > to
> > > > >> the new
> > > > >> > > > > > protocol?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am
> > a
> > > bit
> > > > >> more
> > > > >> > > > > > optimistic
> > > > >> > > > > > > > > than
> > > > >> > > > > > > > > > > >> you
> > > > >> > > > > > > > > > > >> > on this
> > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > > > >> majority of
> > > > >> > > > the
> > > > >> > > > > > cases
> > > > >> > > > > > > > > are
> > > > >> > > > > > > > > > > >> simple
> > > > >> > > > > > > > > > > >> > > > > regexes
> > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> > > > >> engine. The
> > > > >> > > > > > > > > coordinator will
> > > > >> > > > > > > > > > > >> > verify
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> > > consumer
> > > > >> if the
> > > > >> > > > > > regex is
> > > > >> > > > > > > > > not
> > > > >> > > > > > > > > > > >> valid.
> > > > >> > > > > > > > > > > >> > > > > Coming
> > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in
> > > the
> > > > >> worst
> > > > >> > > > case,
> > > > >> > > > > > the
> > > > >> > > > > > > > > first
> > > > >> > > > > > > > > > > >> > upgraded
> > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> > > > >> rejected.
> > > > >> > > > This
> > > > >> > > > > > should
> > > > >> > > > > > > > > be used
> > > > >> > > > > > > > > > > >> > as the
> > > > >> > > > > > > > > > > >> > > > > >> last
> > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > > customers to
> > > > >> > > > validate
> > > > >> > > > > > their
> > > > >> > > > > > > > > regex
> > > > >> > > > > > > > > > > >> > before
> > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to
> > > test
> > > > >> them with
> > > > >> > > > > > another
> > > > >> > > > > > > > > group.
> > > > >> > > > > > > > > > > >> For
> > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> > > > >> pre-prod
> > > > >> > > > > > environment.
> > > > >> > > > > > > > > > > >> Another
> > > > >> > > > > > > > > > > >> > way
> > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > >> consumer-group tool
> > > > >> > > > to
> > > > >> > > > > > > > > provide a
> > > > >> > > > > > > > > > > >> regex
> > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would
> > > this be
> > > > >> > > > enough in
> > > > >> > > > > > your
> > > > >> > > > > > > > > > > >> opinion?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to
> > the
> > > last
> > > > >> > > > question,
> > > > >> > > > > > will
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> Java
> > > > >> > > > > > > > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > > > >> current regex
> > > > >> > > > > > engine for
> > > > >> > > > > > > > > > > >> > compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy
> > to
> > > > >> keep using
> > > > >> > > > > > client
> > > > >> > > > > > > > > based
> > > > >> > > > > > > > > > > >> > regex at
> > > > >> > > > > > > > > > > >> > > > > >> first
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations
> > simple
> > > and
> > > > >> then
> > > > >> > > > > > migrate to
> > > > >> > > > > > > > > server
> > > > >> > > > > > > > > > > >> > based
> > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> > > point
> > > > >> but I
> > > > >> > > > am
> > > > >> > > > > > > > > concerned that
> > > > >> > > > > > > > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay
> > in
> > > > >> this mode.
> > > > >> > > > That
> > > > >> > > > > > > > > would go
> > > > >> > > > > > > > > > > >> > against our
> > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> > > client
> > > > >> because
> > > > >> > > > we
> > > > >> > > > > > would
> > > > >> > > > > > > > > have to
> > > > >> > > > > > > > > > > >> > continue
> > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on
> > the
> > > > >> client
> > > > >> > > > side. I
> > > > >> > > > > > would
> > > > >> > > > > > > > > rather
> > > > >> > > > > > > > > > > >> > not do
> > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say
> > > that
> > > > >> the group
> > > > >> > > > > > > > > coordinator
> > > > >> > > > > > > > > > > >> will
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > configurations
> > > > >> and that
> > > > >> > > > the
> > > > >> > > > > > > > > > > >> configurations
> > > > >> > > > > > > > > > > >> > will be
> > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > deleted.
> > > > >> Will a
> > > > >> > > > > > transition to
> > > > >> > > > > > > > > DEAD
> > > > >> > > > > > > > > > > >> > trigger
> > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > > >> > > > configurations
> > > > >> > > > > > will be
> > > > >> > > > > > > > > > > >> deleted
> > > > >> > > > > > > > > > > >> > when
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > > > >> together.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the
> > > choice to
> > > > >> store
> > > > >> > > > the
> > > > >> > > > > > > > > configs in
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > group
> > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list
> > all
> > > > >> cluster
> > > > >> > > > > > configs and
> > > > >> > > > > > > > > their
> > > > >> > > > > > > > > > > >> > values?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so.
> > > The
> > > > >> group
> > > > >> > > > > > > > > configurations are
> > > > >> > > > > > > > > > > >> > overrides
> > > > >> > > > > > > > > > > >> > > > > >> of
> > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you
> > > want to
> > > > >> know
> > > > >> > > > all
> > > > >> > > > > > the
> > > > >> > > > > > > > > overrides
> > > > >> > > > > > > > > > > >> > though,
> > > > >> > > > > > > > > > > >> > > > > you
> > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the
> > group
> > > > >> > > > coordinators.
> > > > >> > > > > > You
> > > > >> > > > > > > > > cannot
> > > > >> > > > > > > > > > > >> > rely on
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> > > someone
> > > > >> > > > configure a
> > > > >> > > > > > group
> > > > >> > > > > > > > > before
> > > > >> > > > > > > > > > > >> > starting
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > > > >> allowing the
> > > > >> > > > > > explicit
> > > > >> > > > > > > > > creation
> > > > >> > > > > > > > > > > >> of
> > > > >> > > > > > > > > > > >> > > > > groups?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> > > > >> decoupled
> > > > >> > > > from
> > > > >> > > > > > the
> > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > >> > lifecycle.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group
> > will
> > > be
> > > > >> > > > > > automatically
> > > > >> > > > > > > > > created in
> > > > >> > > > > > > > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > case.
> > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will
> > be
> > > > >> lost after
> > > > >> > > > the
> > > > >> > > > > > > > > retention
> > > > >> > > > > > > > > > > >> > period of
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > > >> > > > Consumer.subscribe
> > > > >> > > > > > method
> > > > >> > > > > > > > > for the
> > > > >> > > > > > > > > > > >> > Java
> > > > >> > > > > > > > > > > >> > > > > >> client
> > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> `java.util.regex.Pattern` of
> > > > >> > > > do
> > > > >> > > > > > we
> > > > >> > > > > > > > > have to
> > > > >> > > > > > > > > > > >> > introduce
> > > > >> > > > > > > > > > > >> > > > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very
> > group
> > > > >> question. I
> > > > >> > > > > > forgot
> > > > >> > > > > > > > > about
> > > > >> > > > > > > > > > > >> that
> > > > >> > > > > > > > > > > >> > one.
> > > > >> > > > > > > > > > > >> > > > > As
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> `java.util.regex.Pattern` is
> > > > >> > > > not
> > > > >> > > > > > fully
> > > > >> > > > > > > > > > > >> compatible
> > > > >> > > > > > > > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> > > > >> might be
> > > > >> > > > better
> > > > >> > > > > > to
> > > > >> > > > > > > > > deprecate
> > > > >> > > > > > > > > > > >> > it and
> > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a
> > > string.
> > > > >> We would
> > > > >> > > > > > rely on
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> server
> > > > >> > > > > > > > > > > >> > side
> > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > migration, I
> > > > >> think
> > > > >> > > > that
> > > > >> > > > > > we
> > > > >> > > > > > > > > could
> > > > >> > > > > > > > > > > >> still
> > > > >> > > > > > > > > > > >> > try to
> > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it.
> > > That
> > > > >> should
> > > > >> > > > > > work, I
> > > > >> > > > > > > > > think, in
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> > > > >> Justine
> > > > >> > > > that we
> > > > >> > > > > > > > > should be
> > > > >> > > > > > > > > > > >> > clearer
> > > > >> > > > > > > > > > > >> > > > > about
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > >> > > > IBP/metadata.version
> > > > >> > > > > > from the
> > > > >> > > > > > > > > > > >> feature
> > > > >> > > > > > > > > > > >> > flag.
> > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > > > >> default for
> > > > >> > > > the
> > > > >> > > > > > feature
> > > > >> > > > > > > > > flag
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > true
> > > > >> > > > > > > > > > > >> > > > > >> based
> > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> > > once
> > > > >> we
> > > > >> > > > want to
> > > > >> > > > > > make
> > > > >> > > > > > > > > it the
> > > > >> > > > > > > > > > > >> > default.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> > > that
> > > > >> feature
> > > > >> > > > > > flag
> > > > >> > > > > > > > > mainly
> > > > >> > > > > > > > > > > >> > during the
> > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should
> > not
> > > have
> > > > >> > > > mentioned
> > > > >> > > > > > it, I
> > > > >> > > > > > > > > think,
> > > > >> > > > > > > > > > > >> > because
> > > > >> > > > > > > > > > > >> > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config
> > > for
> > > > >> it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> > > > >> protocol APIs
> > > > >> > > > > > don't
> > > > >> > > > > > > > > mention
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > required
> > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that
> > > for
> > > > >> > > > consistency.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit
> > > odd
> > > > >> that
> > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > >> > > > > > > > > > > >> > > > > requires
> > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems
> > > to do
> > > > >> more
> > > > >> > > > than
> > > > >> > > > > > > > > reading.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is
> > > how the
> > > > >> > > > current
> > > > >> > > > > > > > > protocol works
> > > > >> > > > > > > > > > > >> > though.
> > > > >> > > > > > > > > > > >> > > > > We
> > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read
> > > Group" to
> > > > >> join a
> > > > >> > > > > > group. We
> > > > >> > > > > > > > > could
> > > > >> > > > > > > > > > > >> > consider
> > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure
> > > that
> > > > >> it is
> > > > >> > > > worth
> > > > >> > > > > > it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> > > > >> recreation
> > > > >> > > > > > handled by
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > consumer
> > > > >> > > > > > > > > > > >> > > > > with
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It
> > > would be
> > > > >> good to
> > > > >> > > > > > have a
> > > > >> > > > > > > > > section
> > > > >> > > > > > > > > > > >> on
> > > > >> > > > > > > > > > > >> > this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> > > protocol
> > > > >> > > > > > perspective, the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> topic
> > > > >> > > > > > > > > > > >> > will
> > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it
> > will
> > > > >> treat it
> > > > >> > > > like a
> > > > >> > > > > > > > > topic with
> > > > >> > > > > > > > > > > >> a
> > > > >> > > > > > > > > > > >> > > > > different
> > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that
> > > the
> > > > >> > > > fetch/commit
> > > > >> > > > > > > > > offsets
> > > > >> > > > > > > > > > > >> APIs
> > > > >> > > > > > > > > > > >> > do not
> > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > consumer
> > > > >> would
> > > > >> > > > reuse the
> > > > >> > > > > > > > > offsets
> > > > >> > > > > > > > > > > >> based
> > > > >> > > > > > > > > > > >> > on the
> > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should
> > > update
> > > > >> those
> > > > >> > > > APIs
> > > > >> > > > > > as well
> > > > >> > > > > > > > > in
> > > > >> > > > > > > > > > > >> order
> > > > >> > > > > > > > > > > >> > to be
> > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> > > > >> strengthen
> > > > >> > > > the
> > > > >> > > > > > > > > semantics of
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> > > mentions
> > > > >> we will
> > > > >> > > > > > write the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> > coordinator
> > > > >> > > > > > > > > > > >> > > > > >> in
> > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > > >> implementation
> > > > >> > > > > > detail, do
> > > > >> > > > > > > > > we plan
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > have a
> > > > >> > > > > > > > > > > >> > > > > >> new
> > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> > > > >> scalability
> > > > >> > > > > > goal when
> > > > >> > > > > > > > > it
> > > > >> > > > > > > > > > > >> comes
> > > > >> > > > > > > > > > > >> > to how
> > > > >> > > > > > > > > > > >> > > > > >> many
> > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol
> > can
> > > > >> support?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have
> > > numbers
> > > > >> at the
> > > > >> > > > > > moment. The
> > > > >> > > > > > > > > > > >> protocol
> > > > >> > > > > > > > > > > >> > should
> > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per
> > > group.
> > > > >> We will
> > > > >> > > > > > measure
> > > > >> > > > > > > > > this when
> > > > >> > > > > > > > > > > >> > we have
> > > > >> > > > > > > > > > > >> > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation.
> > Note
> > > > >> that we
> > > > >> > > > might
> > > > >> > > > > > have
> > > > >> > > > > > > > > other
> > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> > > > >> commits).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > > consider
> > > > >> having
> > > > >> > > > > > > > > SubscribedTopidIds
> > > > >> > > > > > > > > > > >> > instead
> > > > >> > > > > > > > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > SubscribedTopicNames
> > > > >> in
> > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> > > > >> resolve the
> > > > >> > > > > > regex on
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > server, we
> > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name?
> > > The
> > > > >> > > > difference
> > > > >> > > > > > is that
> > > > >> > > > > > > > > > > >> sending
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending
> > > the
> > > > >> topic
> > > > >> > > > names
> > > > >> > > > > > is less
> > > > >> > > > > > > > > > > >> > efficient.
> > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> > > easier
> > > > >> to
> > > > >> > > > handle
> > > > >> > > > > > if we
> > > > >> > > > > > > > > have
> > > > >> > > > > > > > > > > >> topic
> > > > >> > > > > > > > > > > >> > ids.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > > > >> consolidate the
> > > > >> > > > > > metadata
> > > > >> > > > > > > > > lookup
> > > > >> > > > > > > > > > > >> > on the
> > > > >> > > > > > > > > > > >> > > > > >> server
> > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do
> > agree
> > > > >> with your
> > > > >> > > > > > point. As
> > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > >> second
> > > > >> > > > > > > > > > > >> > > > > though,
> > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better
> > > here
> > > > >> for the
> > > > >> > > > > > delete and
> > > > >> > > > > > > > > > > >> > recreation
> > > > >> > > > > > > > > > > >> > > > > case.
> > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we
> > may
> > > > >> allow
> > > > >> > > > users to
> > > > >> > > > > > > > > subscribe
> > > > >> > > > > > > > > > > >> with
> > > > >> > > > > > > > > > > >> > topic
> > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future because
> > that
> > > is
> > > > >> the only
> > > > >> > > > > > way to be
> > > > >> > > > > > > > > > > >> really
> > > > >> > > > > > > > > > > >> > robust
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12,
> > > 2022 at
> > > > >> 1:38 PM
> > > > >> > > > > > David
> > > > >> > > > > > > > > Jacot <
> > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> > > > >> comments.
> > > > >> > > > Please
> > > > >> > > > > > find
> > > > >> > > > > > > > > my
> > > > >> > > > > > > > > > > >> answers
> > > > >> > > > > > > > > > > >> > > > > below.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > > > >> protocol
> > > > >> > > > relies on
> > > > >> > > > > > > > > topic IDs
> > > > >> > > > > > > > > > > >> > with the
> > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names
> > based
> > > in
> > > > >> the
> > > > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > >> > > > > > > > > > > >> > > > > >> I
> > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> > > names
> > > > >> is the
> > > > >> > > > > > right call
> > > > >> > > > > > > > > here.
> > > > >> > > > > > > > > > > >> I
> > > > >> > > > > > > > > > > >> > need to
> > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > > > >> Obviously, the
> > > > >> > > > > > KIP does
> > > > >> > > > > > > > > not
> > > > >> > > > > > > > > > > >> > change the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to
> > > use
> > > > >> topic
> > > > >> > > > IDs.
> > > > >> > > > > > This
> > > > >> > > > > > > > > may be
> > > > >> > > > > > > > > > > >> > something
> > > > >> > > > > > > > > > > >> > > > > >> that
> > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though
> > as
> > > it
> > > > >> would
> > > > >> > > > give
> > > > >> > > > > > better
> > > > >> > > > > > > > > > > >> overall
> > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right.
> > I
> > > > >> think that
> > > > >> > > > I
> > > > >> > > > > > should
> > > > >> > > > > > > > > not
> > > > >> > > > > > > > > > > >> have
> > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will
> > remove
> > > > >> it. We can
> > > > >> > > > > > use an
> > > > >> > > > > > > > > internal
> > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > > > >> feature.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> > > types
> > > > >> will be
> > > > >> > > > > > > > > supported. The
> > > > >> > > > > > > > > > > >> > change is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement
> > > is
> > > > >> that the
> > > > >> > > > > > cluster
> > > > >> > > > > > > > > uses
> > > > >> > > > > > > > > > > >> topic
> > > > >> > > > > > > > > > > >> > IDs.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11,
> > > 2022
> > > > >> at 9:53
> > > > >> > > > PM
> > > > >> > > > > > > > > Guozhang
> > > > >> > > > > > > > > > > >> Wang <
> > > > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> > > > >> feedback.
> > > > >> > > > Here
> > > > >> > > > > > are
> > > > >> > > > > > > > > some
> > > > >> > > > > > > > > > > >> replies
> > > > >> > > > > > > > > > > >> > > > > inlined
> > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> > > 2022
> > > > >> at
> > > > >> > > > 2:53 AM
> > > > >> > > > > > > > > Ismael Juma
> > > > >> > > > > > > > > > > >> <
> > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for
> > the
> > > > >> KIP. This
> > > > >> > > > has
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> potential
> > > > >> > > > > > > > > > > >> > to be a
> > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > > >> > > > questions/comments:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think
> > > it's
> > > > >> > > > premature to
> > > > >> > > > > > talk
> > > > >> > > > > > > > > about
> > > > >> > > > > > > > > > > >> > target
> > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of
> > the
> > > > >> existing
> > > > >> > > > group
> > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > > > >> > Unlike
> > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > > protocol
> > > > >> and hence
> > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > >> > > > > > > > > > > >> > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> > > > >> applications
> > > > >> > > > > > migrate
> > > > >> > > > > > > > > to the
> > > > >> > > > > > > > > > > >> new
> > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree
> > > with
> > > > >> you. I
> > > > >> > > > think
> > > > >> > > > > > we can
> > > > >> > > > > > > > > > > >> remove
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > `Compatibility,
> > > > >> > > > Deprecation,
> > > > >> > > > > > and
> > > > >> > > > > > > > > Migration
> > > > >> > > > > > > > > > > >> > Plan` and
> > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> > > decide
> > > > >> in the
> > > > >> > > > > > future
> > > > >> > > > > > > > > about
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > we
> > > > >> > > > > > > > > > > >> > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > > > >> behaviors.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP
> > > says we
> > > > >> > > > intend to
> > > > >> > > > > > > > > release this
> > > > >> > > > > > > > > > > >> > in 4.x,
> > > > >> > > > > > > > > > > >> > > > > >> but
> > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> > > added
> > > > >> that as
> > > > >> > > > a
> > > > >> > > > > > way to
> > > > >> > > > > > > > > > > >> estimate
> > > > >> > > > > > > > > > > >> > when
> > > > >> > > > > > > > > > > >> > > > > we'd
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > > > >> protocol, I
> > > > >> > > > also
> > > > >> > > > > > suggest
> > > > >> > > > > > > > > > > >> removing
> > > > >> > > > > > > > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's
> > > not
> > > > >> > > > specifically
> > > > >> > > > > > > > > related to
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> > > more
> > > > >> for
> > > > >> > > > client
> > > > >> > > > > > > > > upgrades.
> > > > >> > > > > > > > > > > >> I.e.
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation
> > > may
> > > > >> be done
> > > > >> > > > > > first,
> > > > >> > > > > > > > > and then
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > client
> > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it
> > as
> > > > >> "released"
> > > > >> > > > by
> > > > >> > > > > > the
> > > > >> > > > > > > > > time
> > > > >> > > > > > > > > > > >> clients
> > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> > > > >> enable the
> > > > >> > > > > > feature the
> > > > >> > > > > > > > > > > >> clients
> > > > >> > > > > > > > > > > >> > need to
> > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with
> > a
> > > > >> rolling
> > > > >> > > > bounce
> > > > >> > > > > > and
> > > > >> > > > > > > > > then set
> > > > >> > > > > > > > > > > >> > the flag
> > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and
> > > hence
> > > > >> we feel
> > > > >> > > > it's
> > > > >> > > > > > > > > better to be
> > > > >> > > > > > > > > > > >> > released
> > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need
> > to
> > > > >> flesh out
> > > > >> > > > the
> > > > >> > > > > > > > > details of
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > migration
> > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying
> > > we
> > > > >> will
> > > > >> > > > support
> > > > >> > > > > > online
> > > > >> > > > > > > > > > > >> > migrations. Is
> > > > >> > > > > > > > > > > >> > > > > >> that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this
> > > in
> > > > >> detail.
> > > > >> > > > It
> > > > >> > > > > > could
> > > > >> > > > > > > > > also be
> > > > >> > > > > > > > > > > >> > done as a
> > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > > that's
> > > > >> the
> > > > >> > > > part we
> > > > >> > > > > > can
> > > > >> > > > > > > > > be more
> > > > >> > > > > > > > > > > >> > concrete
> > > > >> > > > > > > > > > > >> > > > > >> about
> > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is
> > related
> > > to
> > > > >> your
> > > > >> > > > > > question 2)
> > > > >> > > > > > > > > above).
> > > > >> > > > > > > > > > > >> > We will
> > > > >> > > > > > > > > > > >> > > > > >> work
> > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > > > >> parallel as we
> > > > >> > > > > > solicit
> > > > >> > > > > > > > > more
> > > > >> > > > > > > > > > > >> > feedback.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am
> > happy
> > > > >> that we are
> > > > >> > > > > > pushing
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > pattern
> > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> > > like
> > > > >> there
> > > > >> > > > > > could be
> > > > >> > > > > > > > > some
> > > > >> > > > > > > > > > > >> tricky
> > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > > mechanism
> > > > >> for
> > > > >> > > > users to
> > > > >> > > > > > > > > detect that
> > > > >> > > > > > > > > > > >> > they
> > > > >> > > > > > > > > > > >> > > > > need
> > > > >> > > > > > > > > > > >> > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > > switching
> > > > >> to the
> > > > >> > > > new
> > > > >> > > > > > > > > protocol?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> > > need
> > > > >> some
> > > > >> > > > > > tooling for
> > > > >> > > > > > > > > > > >> non-java
> > > > >> > > > > > > > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> > > > >> client before
> > > > >> > > > > > > > > switching to
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > new
> > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea
> > on
> > > > >> top of my
> > > > >> > > > head
> > > > >> > > > > > > > > though,
> > > > >> > > > > > > > > > > >> maybe
> > > > >> > > > > > > > > > > >> > others
> > > > >> > > > > > > > > > > >> > > > > >> like
> > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related
> > to
> > > the
> > > > >> last
> > > > >> > > > > > question,
> > > > >> > > > > > > > > will the
> > > > >> > > > > > > > > > > >> > Java
> > > > >> > > > > > > > > > > >> > > > > >> client
> > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with
> > the
> > > > >> current
> > > > >> > > > regex
> > > > >> > > > > > > > > engine for
> > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be
> > > handy
> > > > >> to keep
> > > > >> > > > using
> > > > >> > > > > > > > > client
> > > > >> > > > > > > > > > > >> based
> > > > >> > > > > > > > > > > >> > regex
> > > > >> > > > > > > > > > > >> > > > > at
> > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> > > simple
> > > > >> and
> > > > >> > > > then
> > > > >> > > > > > migrate
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > server based
> > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I
> > have
> > > not
> > > > >> thought
> > > > >> > > > > > about
> > > > >> > > > > > > > > that for
> > > > >> > > > > > > > > > > >> > java
> > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What
> > kind
> > > of
> > > > >> > > > > > compatibility
> > > > >> > > > > > > > > issues do
> > > > >> > > > > > > > > > > >> > you have
> > > > >> > > > > > > > > > > >> > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we
> > say
> > > > >> that the
> > > > >> > > > group
> > > > >> > > > > > > > > coordinator
> > > > >> > > > > > > > > > > >> > will be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > > >> configurations and
> > > > >> > > > that
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> > configurations
> > > > >> > > > > > > > > > > >> > > > > will
> > > > >> > > > > > > > > > > >> > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> > > deleted.
> > > > >> Will a
> > > > >> > > > > > > > > transition to
> > > > >> > > > > > > > > > > >> DEAD
> > > > >> > > > > > > > > > > >> > > > > trigger
> > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > configurations?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> > > DEAD
> > > > >> state
> > > > >> > > > is an
> > > > >> > > > > > > > > ending
> > > > >> > > > > > > > > > > >> state
> > > > >> > > > > > > > > > > >> > (we
> > > > >> > > > > > > > > > > >> > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> > > > >> group is
> > > > >> > > > EMPTY
> > > > >> > > > > > and
> > > > >> > > > > > > > > also all
> > > > >> > > > > > > > > > > >> of
> > > > >> > > > > > > > > > > >> > its
> > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited
> > > to
> > > > >> DEAD this
> > > > >> > > > > > group
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> never
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> > > > >> choice to
> > > > >> > > > store
> > > > >> > > > > > the
> > > > >> > > > > > > > > configs
> > > > >> > > > > > > > > > > >> in
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to
> > > list all
> > > > >> > > > cluster
> > > > >> > > > > > configs
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> their
> > > > >> > > > > > > > > > > >> > > > > values?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> > > > >> question,
> > > > >> > > > and our
> > > > >> > > > > > > > > thoughts
> > > > >> > > > > > > > > > > >> are
> > > > >> > > > > > > > > > > >> > that
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > configurations"
> > > are
> > > > >> > > > overrides
> > > > >> > > > > > of the
> > > > >> > > > > > > > > > > >> > cluster-level
> > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> > > > >> group so
> > > > >> > > > when an
> > > > >> > > > > > > > > admin list
> > > > >> > > > > > > > > > > >> > cluster
> > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> > > > >> cluster-level
> > > > >> > > > > > > > > "defaults", not
> > > > >> > > > > > > > > > > >> > showing
> > > > >> > > > > > > > > > > >> > > > > >> any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > customizations.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> > > > >> someone
> > > > >> > > > > > configure a
> > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > >> > before
> > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > considered
> > > > >> allowing
> > > > >> > > > the
> > > > >> > > > > > explicit
> > > > >> > > > > > > > > > > >> > creation of
> > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs
> > could
> > > be
> > > > >> > > > decoupled
> > > > >> > > > > > from
> > > > >> > > > > > > > > the group
> > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs
> > can
> > > be
> > > > >> created
> > > > >> > > > > > before
> > > > >> > > > > > > > > the group
> > > > >> > > > > > > > > > > >> > itself
> > > > >> > > > > > > > > > > >> > > > > as
> > > > >> > > > > > > > > > > >> > > > > >> an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course,
> > > this
> > > > >> > > > requires
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> corresponding
> > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > > coordinator
> > > > >> based on
> > > > >> > > > the
> > > > >> > > > > > group
> > > > >> > > > > > > > > id ---
> > > > >> > > > > > > > > > > >> > the only
> > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> > > > >> itself is
> > > > >> > > > gone
> > > > >> > > > > > we
> > > > >> > > > > > > > > also check
> > > > >> > > > > > > > > > > >> > if there
> > > > >> > > > > > > > > > > >> > > > > >> are
> > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > > related to
> > > > >> that
> > > > >> > > > group
> > > > >> > > > > > and
> > > > >> > > > > > > > > delete
> > > > >> > > > > > > > > > > >> as
> > > > >> > > > > > > > > > > >> > well.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly
> > this
> > > > >> indeed
> > > > >> > > > > > introduces an
> > > > >> > > > > > > > > > > >> > asymmetry on
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of
> > > the
> > > > >> config
> > > > >> > > > > > entities,
> > > > >> > > > > > > > > and we
> > > > >> > > > > > > > > > > >> > would like
> > > > >> > > > > > > > > > > >> > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > > whether we
> > > > >> should
> > > > >> > > > aim
> > > > >> > > > > > for
> > > > >> > > > > > > > > symmetry
> > > > >> > > > > > > > > > > >> > i.e.
> > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and
> > > hence
> > > > >> not
> > > > >> > > > delete
> > > > >> > > > > > them at
> > > > >> > > > > > > > > all
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> group
> > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> > explicit
> > > > >> deletion
> > > > >> > > > > > operations
> > > > >> > > > > > > > > by
> > > > >> > > > > > > > > > > >> > themselves.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > > >> > > > > > Consumer.subscribe
> > > > >> > > > > > > > > method for
> > > > >> > > > > > > > > > > >> > the Java
> > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > `java.util.regex.Pattern` of
> > > > >> > > > > > do we
> > > > >> > > > > > > > > have
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do
> > > not
> > > > >> need to
> > > > >> > > > > > introduce
> > > > >> > > > > > > > > an
> > > > >> > > > > > > > > > > >> > overload, but
> > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > > > >> compatibility
> > > > >> > > > > > issues
> > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > >> may
> > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree
> > > with
> > > > >> Justine
> > > > >> > > > > > that we
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> be
> > > > >> > > > > > > > > > > >> > clearer
> > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > >> > > > > > IBP/metadata.version
> > > > >> > > > > > > > > from the
> > > > >> > > > > > > > > > > >> > feature
> > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch
> > the
> > > > >> default
> > > > >> > > > for
> > > > >> > > > > > the
> > > > >> > > > > > > > > feature
> > > > >> > > > > > > > > > > >> flag
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > true
> > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > metadata.version
> > > > >> once we
> > > > >> > > > > > want to
> > > > >> > > > > > > > > make it
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of
> > the
> > > > >> protocol
> > > > >> > > > APIs
> > > > >> > > > > > don't
> > > > >> > > > > > > > > mention
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> required
> > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add
> > > that
> > > > >> for
> > > > >> > > > > > consistency.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > > > >> certainly do
> > > > >> > > > that.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a
> > > bit
> > > > >> odd that
> > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it
> > > seems
> > > > >> to do
> > > > >> > > > more
> > > > >> > > > > > than
> > > > >> > > > > > > > > reading.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> > > thought
> > > > >> myself
> > > > >> > > > as
> > > > >> > > > > > well,
> > > > >> > > > > > > > > but in
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > end we
> > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > alternative:
> > > > >> adding
> > > > >> > > > > > Write
> > > > >> > > > > > > > > Group
> > > > >> > > > > > > > > > > >> seems
> > > > >> > > > > > > > > > > >> > an
> > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > > > >> elsewhere (we
> > > > >> > > > only
> > > > >> > > > > > have
> > > > >> > > > > > > > > Read /
> > > > >> > > > > > > > > > > >> > Delete
> > > > >> > > > > > > > > > > >> > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would
> > > like to
> > > > >> hear
> > > > >> > > > others
> > > > >> > > > > > > > > thoughts.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> > > topic
> > > > >> > > > recreation
> > > > >> > > > > > > > > handled by
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> > > > >> would be
> > > > >> > > > good
> > > > >> > > > > > to have
> > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > >> > section on
> > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with
> > > regex
> > > > >> > > > > > subscription
> > > > >> > > > > > > > > right? Yes
> > > > >> > > > > > > > > > > >> > we can
> > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> > > basically
> > > > >> the
> > > > >> > > > idea
> > > > >> > > > > > is that
> > > > >> > > > > > > > > > > >> consumer
> > > > >> > > > > > > > > > > >> > would
> > > > >> > > > > > > > > > > >> > > > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > > protocol as
> > > > >> it's
> > > > >> > > > > > handled all
> > > > >> > > > > > > > > by the
> > > > >> > > > > > > > > > > >> > brokers.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> > > > >> mentions we
> > > > >> > > > will
> > > > >> > > > > > write
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> new
> > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > > > >> implementation
> > > > >> > > > > > detail,
> > > > >> > > > > > > > > do we
> > > > >> > > > > > > > > > > >> > plan to
> > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> > > thought
> > > > >> about
> > > > >> > > > > > that. But
> > > > >> > > > > > > > > I think
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > answer
> > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we
> > > have a
> > > > >> > > > scalability
> > > > >> > > > > > goal
> > > > >> > > > > > > > > when it
> > > > >> > > > > > > > > > > >> > comes to
> > > > >> > > > > > > > > > > >> > > > > >> how
> > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > > protocol can
> > > > >> > > > support?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> > group,
> > > I
> > > > >> think we
> > > > >> > > > > > should
> > > > >> > > > > > > > > shoot for
> > > > >> > > > > > > > > > > >> > 1000s of
> > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability
> > > goals
> > > > >> also
> > > > >> > > > depend
> > > > >> > > > > > on the
> > > > >> > > > > > > > > offset
> > > > >> > > > > > > > > > > >> > > > > management
> > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities
> > of
> > > the
> > > > >> > > > coordinator
> > > > >> > > > > > > > > which we
> > > > >> > > > > > > > > > > >> did
> > > > >> > > > > > > > > > > >> > not
> > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> > > > >> number that
> > > > >> > > > > > applies
> > > > >> > > > > > > > > > > >> > universally.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > > > >> consider
> > > > >> > > > having
> > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> SubscribedTopicNames in
> > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we
> > have
> > > to
> > > > >> resolve
> > > > >> > > > the
> > > > >> > > > > > regex
> > > > >> > > > > > > > > on the
> > > > >> > > > > > > > > > > >> > server,
> > > > >> > > > > > > > > > > >> > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic
> > > name?
> > > > >> The
> > > > >> > > > > > difference is
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > sending the
> > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas
> > > sending
> > > > >> the topic
> > > > >> > > > > > names is
> > > > >> > > > > > > > > less
> > > > >> > > > > > > > > > > >> > efficient.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation
> > is
> > > > >> easier to
> > > > >> > > > > > handle if
> > > > >> > > > > > > > > we have
> > > > >> > > > > > > > > > > >> > topic
> > > > >> > > > > > > > > > > >> > > > > ids.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main
> > reason
> > > to
> > > > >> still
> > > > >> > > > let
> > > > >> > > > > > the
> > > > >> > > > > > > > > clients
> > > > >> > > > > > > > > > > >> send
> > > > >> > > > > > > > > > > >> > names
> > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> > > names
> > > > >> -> ids
> > > > >> > > > on
> > > > >> > > > > > the
> > > > >> > > > > > > > > broker /
> > > > >> > > > > > > > > > > >> > admin
> > > > >> > > > > > > > > > > >> > > > > client
> > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we
> > > added
> > > > >> topic id
> > > > >> > > > in
> > > > >> > > > > > > > > KIP-516, we
> > > > >> > > > > > > > > > > >> > never
> > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > consumer/producers
> > > > >> > > > leveraging
> > > > >> > > > > > the
> > > > >> > > > > > > > > related
> > > > >> > > > > > > > > > > >> > newer
> > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> > > topic
> > > > >> id as
> > > > >> > > > empty
> > > > >> > > > > > UUID.
> > > > >> > > > > > > > > We
> > > > >> > > > > > > > > > > >> want
> > > > >> > > > > > > > > > > >> > to keep
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and
> > > only
> > > > >> > > > delegate
> > > > >> > > > > > the
> > > > >> > > > > > > > > reasoning
> > > > >> > > > > > > > > > > >> on
> > > > >> > > > > > > > > > > >> > broker
> > > > >> > > > > > > > > > > >> > > > > >> and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul
> > 6,
> > > > >> 2022 at
> > > > >> > > > 10:45
> > > > >> > > > > > AM
> > > > >> > > > > > > > > David
> > > > >> > > > > > > > > > > >> Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would
> > > like to
> > > > >> start a
> > > > >> > > > > > > > > discussion
> > > > >> > > > > > > > > > > >> thread
> > > > >> > > > > > > > > > > >> > on
> > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation
> > > of
> > > > >> the
> > > > >> > > > Consumer
> > > > >> > > > > > > > > Rebalance
> > > > >> > > > > > > > > > > >> > Protocol.
> > > > >> > > > > > > > > > > >> > > > > >> With
> > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make
> > the
> > > > >> rebalance
> > > > >> > > > > > protocol
> > > > >> > > > > > > > > (for
> > > > >> > > > > > > > > > > >> > consumers)
> > > > >> > > > > > > > > > > >> > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> > > > >> easier to
> > > > >> > > > > > implement for
> > > > >> > > > > > > > > > > >> > clients, and
> > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> > > here:
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please
> > take
> > > a
> > > > >> look and
> > > > >> > > > let
> > > > >> > > > > > me
> > > > >> > > > > > > > > know what
> > > > >> > > > > > > > > > > >> > you
> > > > >> > > > > > > > > > > >> > > > > think.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will
> > > be
> > > > >> away from
> > > > >> > > > > > July
> > > > >> > > > > > > > > 18th to
> > > > >> > > > > > > > > > > >> > August 8th.
> > > > >> > > > > > > > > > > >> > > > > >> That
> > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> > > read
> > > > >> and
> > > > >> > > > digest
> > > > >> > > > > > this
> > > > >> > > > > > > > > long
> > > > >> > > > > > > > > > > >> KIP.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> --
> > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > --
> > > > >> > > > > -- Guozhang
> > > > >> > > >
> > > > >> > >
> > > > >> > >
> > > > >> > > --
> > > > >> > > -- Guozhang
> > > > >>
> > > > >
> > > > >
> > > > > --
> > > > > -- Guozhang
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Jason,

The RE2 syntax is documented here:
https://github.com/google/re2/wiki/Syntax. Based on the revisions of
this document and also the commits in the RE2J
(https://github.com/google/re2j) port of the core C++ library, it
looks like the syntax has not changed during the last few years. The
syntax is mainly the Perl one but with some restrictions due to how
the engine works.

It is interesting to note that a few products out there use RE2 and
they directly redirect to https://github.com/google/re2/wiki/Syntax in
their documentation. For instance, Spanner [1] and Google Sheets [2]
use it.

[1] https://cloud.google.com/spanner/docs/reference/standard-sql/string_functions#regexp_contains
[2] https://support.google.com/docs/answer/3098292

Given that the syntax is directly used in Google products, I suspect
that compatibility is important for the library itself.

It seems quite reasonable to me to use it directly as-is. What do you think?

Best,
David

On Tue, Sep 27, 2022 at 10:20 PM Jason Gustafson
<ja...@confluent.io.invalid> wrote:
>
> Hi David,
>
> Thanks for the response. Replied below:
>
> > I don't have a strong opinion on this one. However, it seems that this
> could be a pandora box if our subset turns out to not be sufficient
> for some existing users. We will have to add more to it and the
> compatibility story between versions could be a real nightmare. One
> thing that we could do for sure is disabling all unicode groups in the
> library. That seems possible and would already reduce the scope.
>
> Changes to the regex syntax seems not a major problem to me. We would bump
> the protocol version as we usually do. Actually that begs the same question
> about RE2. How stable is the syntax and does it differ depending on the
> version of the library? In general, I do agree that we'd be better off not
> having to manage the syntax ourselves if we can convince ourselves that
> compatibility won't be a problem and the client won't eventually need to
> have access to it.
>
> Thanks,
> Jason
>
> On Tue, Sep 27, 2022 at 6:21 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Luke,
> >
> > Thanks for your comments.
> >
> > Your suggestion makes sense. Jun's made the same one. We are also
> > discussing whether we should directly return the group state in the
> > response and drop the prepare assignment api. It seems that the
> > current consensus is to keep it though. If that stays like this, I
> > will apply your suggestion.
> >
> > Thanks,
> > David
> >
> > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com> wrote:
> > >
> > > Hi David,
> > >
> > > Thanks for your patience for answering all these questions!
> > >
> > > One more question from me:
> > > 1. When the selected consumer is going to be in client-side assigning, we
> > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering if we
> > > could use a flag to indicate the assignment is required. The rationale is
> > > when returning error, we lose a chance to catch up with the latest target
> > > assignment or revoke partitions. Since we expected the record fetching is
> > > not blocked by the rebalance/assignment computing, we might want to allow
> > > the members to update the member assignment in each heartbeat.
> > >
> > > Luke
> > >
> > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > <ja...@confluent.io.invalid>
> > > wrote:
> > >
> > > > Hey David,
> > > >
> > > > Thanks for the response.
> > > >
> > > > > Do you think that we could have a case where two members with the
> > same
> > > > static id are active?
> > > >
> > > > I was just thinking about the handoff when we spin up a new node
> > before the
> > > > old one completes shutdown. It's probably ok to fence the old member
> > > > abruptly. I do recall we had a few issues with this handoff in the
> > past. It
> > > > was due to the fact that the replaced node might have been the leader.
> > > > Since we've removed the leader notion, perhaps this kind of issue won't
> > > > resurface.
> > > >
> > > > In regard to the RE2 library. I understand the intent to pass this
> > through
> > > > as a string without any validation so that we don't need the
> > dependence on
> > > > the client. I do wonder if that will be restrictive in practice. For
> > > > example, if a regex changes on the client side, would we need to
> > validate
> > > > against the current assignment to see which partitions to continue
> > > > fetching? Or would we instead invalidate the complete current
> > assignment?
> > > > Perhaps this is a rare case and we can do the latter. There could be
> > more
> > > > situations though.
> > > >
> > > > As an alternative, I wonder if we could define a minimalistic regex
> > syntax
> > > > (a subset of RE2 say) that could map more easily to regex support that
> > > > already exists in clients? For example, in this scenario, we are
> > matching
> > > > topics, so we only need to be able to represent topic characters and
> > not
> > > > the whole smorgasbord of whitespace and unicode options. Would it make
> > > > sense to have a reduced syntax which includes only standard topic
> > > > characters, wildcards, character classes, and repetitions?
> > > >
> > > > Best,
> > > > Jason
> > > >
> > > >
> > > >
> > > >
> > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > >
> > > > > Hi David, thanks for all the detailed explanations. I think they all
> > make
> > > > > sense. Just want to have a couple follow-ups here:
> > > > >
> > > > > > I don't really see the benefits here because server side assignors
> > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > are not supposed to generate metadata nor to receive metadata from
> > the
> > > > > members.
> > > > >
> > > > > Just to clarify I was asking about the `version` of the assignor
> > (i.e. up
> > > > > to what version that the client would support), and I do agree we
> > would
> > > > not
> > > > > need metadata. What I have in mind is that, for some specific
> > built-in
> > > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> > in a
> > > > > newer version we would have a hierarchical rack ID string format,
> > like
> > > > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > > > would still be in old format. In this case, the broker then needs to
> > > > choose
> > > > > the old versioned assignor. I'm probably making something up here for
> > > > rack
> > > > > aware assignors, but I'm wondering if in general such an
> > "auto-downgrade"
> > > > > behavior would be needed still for broker-side assignor, and if yes
> > would
> > > > > "version" still be useful.
> > > > >
> > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is called
> > once
> > > > > when the member transitions to a new epoch. This one contains the
> > full
> > > > > metadata provided by the client side assignor. Then, `onAssignment`
> > > > > can be called max N times where N is the number of partitions pending
> > > > > revocation by other members. Let me try to clarify this in the KIP.
> > > > >
> > > > > Okay, my understanding is that the calling ordering of these
> > callbacks
> > > > > would be like the following:
> > > > >
> > > > > ----------------------------------------
> > > > > onPartitionsRevoked();   // just once, since we do not really need
> > > > > to revoke incrementally.
> > > > >
> > > > > onAssignment();    // the first call, with epoch incremented
> > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > >
> > > > > onAssignment();              // the first onAssignment would bump up
> > the
> > > > > epoch, and the metadata reflected.
> > > > > onPartitionsAssigned();   // each time we get an additional
> > assignment,
> > > > we
> > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > ...
> > > > > onAssignment();
> > > > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > > > encoded
> > > > > metadata would not change, only the incrementally added partitions be
> > > > > reflected
> > > > >
> > > > > Is that the case?
> > > > >
> > > > > I'm wondering if we would still call onAssignment just once, that
> > encodes
> > > > > all the assignment for this rebalance, including all the partitions
> > that
> > > > > should be assigned to the member but not yet assigned since they
> > have not
> > > > > been revoked by others. In that case the call ordering would be:
> > > > >
> > > > > ----------------------------------------
> > > > > onPartitionsRevoked();   // just once
> > > > > onAssignment();    // just once, with epoch incremented, and metadata
> > > > > encoded changed, the "assignment" field also reflect the final target
> > > > > assignment
> > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > incrementally
> > > > > added partitions
> > > > > ...
> > > > > onPartitionsAssigned();
> > > > >
> > > > > The motivation from this is that, most users would only implement the
> > > > > rebalance callback listeners and hence we'd definitely need to make
> > sure
> > > > > the semantics of that does not change much, and the time
> > > > > `onPartitionsAssigned` indicate the time when the partitions are
> > actually
> > > > > assigned to it; while for assignors, the `onAssignment` is used to
> > > > indicate
> > > > > what decision is made regarding for this member, i.e. when the
> > partitions
> > > > > are decided to be given to it, but not necessarily meant that it has
> > been
> > > > > given, since that time should be determined by the time of
> > > > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > > > implementers would not need to reason which `onAssignment` would be
> > the
> > > > > last one for this epoch.
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > <djacot@confluent.io.invalid
> > > > >
> > > > > wrote:
> > > > >
> > > > > > Hi Guozhang,
> > > > > >
> > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > byte
> > > > > > field,
> > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > And
> > > > the
> > > > > > > rationale seems to be assuming that we should always be able to
> > do
> > > > the
> > > > > > > assignment at the broker-side assignor without errors.
> > Personally I
> > > > > think
> > > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > > broker-side assignors, e.g. for some edge cases where some
> > subscribed
> > > > > > > topics are not recognized with the current broker's metadata.
> > What do
> > > > > you
> > > > > > > think?
> > > > > >
> > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > better
> > > > > > to use an exception on the server side. This is what we usually do
> > for
> > > > > > server side plugins. On the client side, we use a field because the
> > > > > > errors are not defined in advance.
> > > > > >
> > > > > > Your comment also makes me think about what we should do when the
> > > > > > server side assignor fails. I suppose that we have to keep the
> > current
> > > > > > assignment until a new event occurs. For instance, in your example,
> > > > > > the coordinator would have to trigger a rebalance when unrecognized
> > > > > > topics are available. This would be part of the metadata
> > monitoring.
> > > > > >
> > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > reason/version/metadata compared with the broker-side
> > GroupMember. I
> > > > > > agree
> > > > > > > that broker-side assignor would not need reason/metadata since
> > they
> > > > are
> > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > E.g.
> > > > is
> > > > > > it
> > > > > > > possible that we evolve our broker-side built-in assignor but
> > the old
> > > > > > > versioned clients would not be able to work with the new
> > version, in
> > > > > > which
> > > > > > > case we need to let the broker being aware of this and upgrade
> > its
> > > > > > behavior
> > > > > > > to cooperate with the clients?
> > > > > >
> > > > > > I don't really see the benefits here because server side assignors
> > > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > > are not supposed to generate metadata nor to receive metadata from
> > the
> > > > > > members.
> > > > > >
> > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > instead
> > > > > > of
> > > > > > > including these three fields, what about just adding the
> > "Metadata"
> > > > > field
> > > > > > > class which has these three fields? Also, there are two
> > "Metadata"
> > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > reason/version/metadata, and the second is just the encoded
> > metadata
> > > > > > bytes.
> > > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > > which
> > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > second
> > > > > > > bytebuffer field as metadataBytes?
> > > > > >
> > > > > > That's a good point. Let me try to rationalize this interface
> > based on
> > > > > > your suggestions.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > > > > wrote:
> > > > > > >
> > > > > > > Hello David,
> > > > > > >
> > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > >
> > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > byte
> > > > > > field,
> > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > And
> > > > the
> > > > > > > rationale seems to be assuming that we should always be able to
> > do
> > > > the
> > > > > > > assignment at the broker-side assignor without errors.
> > Personally I
> > > > > think
> > > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > > broker-side assignors, e.g. for some edge cases where some
> > subscribed
> > > > > > > topics are not recognized with the current broker's metadata.
> > What do
> > > > > you
> > > > > > > think?
> > > > > > >
> > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > reason/version/metadata compared with the broker-side
> > GroupMember. I
> > > > > > agree
> > > > > > > that broker-side assignor would not need reason/metadata since
> > they
> > > > are
> > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > E.g.
> > > > is
> > > > > > it
> > > > > > > possible that we evolve our broker-side built-in assignor but
> > the old
> > > > > > > versioned clients would not be able to work with the new
> > version, in
> > > > > > which
> > > > > > > case we need to let the broker being aware of this and upgrade
> > its
> > > > > > behavior
> > > > > > > to cooperate with the clients?
> > > > > > >
> > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > instead
> > > > > > of
> > > > > > > including these three fields, what about just adding the
> > "Metadata"
> > > > > field
> > > > > > > class which has these three fields? Also, there are two
> > "Metadata"
> > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > reason/version/metadata, and the second is just the encoded
> > metadata
> > > > > > bytes.
> > > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > > which
> > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > second
> > > > > > > bytebuffer field as metadataBytes?
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > > wrote:
> > > > > > >
> > > > > > > > Hello David,
> > > > > > > >
> > > > > > > > Thanks for bringing this question up. I think the main
> > benefits as
> > > > > you
> > > > > > > > listed is 2) above if it stays; just to clarify, we would only
> > be
> > > > > able
> > > > > > to
> > > > > > > > save one round trip if the rebalance is still triggered by the
> > > > > broker;
> > > > > > if
> > > > > > > > the rebalance is triggered by the client then the
> > num.round.trips
> > > > are
> > > > > > the
> > > > > > > > same:
> > > > > > > >
> > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > >
> > > > > > > > T0: client decides to do a new assignment, suppose it has
> > already
> > > > > sent
> > > > > > a
> > > > > > > > HB and hence has to wait for it to return first since only one
> > > > > request
> > > > > > /
> > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > GroupPrepareAssignment request.
> > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > GroupInstallAssignment
> > > > > > > > request.
> > > > > > > >
> > > > > > > > In total, two round trips.
> > > > > > > >
> > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > >
> > > > > > > > T0: client decides to do a new assignment, suppose it has
> > already
> > > > > sent
> > > > > > a
> > > > > > > > HB and hence has to wait for it to return first since only one
> > > > > request
> > > > > > /
> > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > T1: client receives the HB response, and then sends the new HB
> > > > > request
> > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > T2: the HB response with the optional member metadata map is
> > > > > returned.
> > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > GroupInstallAssignment
> > > > > > > > request.
> > > > > > > >
> > > > > > > > In total, two round trips as well.
> > > > > > > >
> > > > > > > > -----------------------------
> > > > > > > >
> > > > > > > > So to complete the full picture here, we'd need to modify both
> > HB
> > > > > > request
> > > > > > > > and response so that the client can also indicate a new
> > rebalance
> > > > via
> > > > > > the
> > > > > > > > HB request as well, right?
> > > > > > > >
> > > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > additional
> > > > > > fields
> > > > > > > > encoding the full member (subscription) metadata and topic
> > metadata
> > > > > as
> > > > > > > > optional fields.
> > > > > > > >
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > >> Hi all,
> > > > > > > >>
> > > > > > > >> During an offline conversation, someone asked why we need the
> > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > > > > instead
> > > > > > > >> provide the group state in the heartbeat response. This has a
> > few
> > > > > > > >> advantages: 1) it does not require using a special error code
> > to
> > > > > > > >> signal that a new assignment is required as the signal would
> > be
> > > > the
> > > > > > > >> provided group state; 2) it removes one round trip when a
> > client
> > > > > side
> > > > > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > > > > >> response's definition quite large. I recall that I went with
> > the
> > > > > > > >> current approach due to this.
> > > > > > > >>
> > > > > > > >> Providing the group state in the heartbeat response is
> > appealing.
> > > > > What
> > > > > > > >> do you guys think?
> > > > > > > >>
> > > > > > > >> Best,
> > > > > > > >> David
> > > > > > > >>
> > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > djacot@confluent.io>
> > > > > > wrote:
> > > > > > > >> >
> > > > > > > >> > Hi Guozhang,
> > > > > > > >> >
> > > > > > > >> > 1. I have added a reference to the relevant chapter instead
> > of
> > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > >> >
> > > > > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> > > > about
> > > > > > when
> > > > > > > >> > a rebalance should be triggered for the non-upgraded members
> > > > using
> > > > > > the
> > > > > > > >> > old protocol. The section mentions that a rebalance must be
> > > > > > triggered
> > > > > > > >> > when a new assignment is installed. This implies that the
> > group
> > > > > > epoch
> > > > > > > >> > was updated either by a native member or a non-upgraded
> > member.
> > > > > For
> > > > > > > >> > the latter, the JoinGroup request would be the trigger. I
> > have
> > > > > > added a
> > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > Handling"
> > > > > > section
> > > > > > > >> > as well. Does that make sense?
> > > > > > > >> >
> > > > > > > >> > Thanks,
> > > > > > > >> > David
> > > > > > > >> >
> > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > wangguoz@gmail.com
> > > > > >
> > > > > > > >> wrote:
> > > > > > > >> > >
> > > > > > > >> > > Hello David,
> > > > > > > >> > >
> > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > clear in
> > > > > the
> > > > > > doc,
> > > > > > > >> > > could we update:
> > > > > > > >> > >
> > > > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > > > coordinator
> > > > > > > >> will
> > > > > > > >> > > trigger rebalance based on the HB's member metadata /
> > reason?
> > > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > > described
> > > > > > in
> > > > > > > >> "Group
> > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > >> > >
> > > > > > > >> > >
> > > > > > > >> > > Guozhang
> > > > > > > >> > >
> > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > wrote:
> > > > > > > >> > >
> > > > > > > >> > > > Hi Guozhang,
> > > > > > > >> > > >
> > > > > > > >> > > > I thought that the assignor will always be consulted
> > when
> > > > the
> > > > > > next
> > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > > > heartbeat.
> > > > > > > >> This
> > > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > > rebalance
> > > > > by
> > > > > > > >> > > > setting the reason to a non-zero value or by changing
> > the
> > > > > > bytes. Do
> > > > > > > >> > > > you think that this is not sufficient? Are you
> > concerned by
> > > > > the
> > > > > > > >> delay?
> > > > > > > >> > > >
> > > > > > > >> > > > Best,
> > > > > > > >> > > > David
> > > > > > > >> > > >
> > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > wangguoz@gmail.com>
> > > > > > > >> wrote:
> > > > > > > >> > > > >
> > > > > > > >> > > > > Hello David,
> > > > > > > >> > > > >
> > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > >> > > > >
> > > > > > > >> > > > > ```
> > > > > > > >> > > > > In this case, a new assignment is triggered by the
> > client
> > > > > side
> > > > > > > >> > > > > assignor. When constructing the HB, the consumer will
> > > > always
> > > > > > > >> consult
> > > > > > > >> > > > > the client side assignor and propagate the
> > information to
> > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > coordinator. In other words, we don't expect users to
> > call
> > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > >> > > > > ```
> > > > > > > >> > > > >
> > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > interface,
> > > > we
> > > > > > > >> actually do
> > > > > > > >> > > > > not have a way yet to instruct how to construct the
> > next
> > > > HB
> > > > > > > >> request, e.g.
> > > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> > with a
> > > > > new
> > > > > > > >> assignment,
> > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > PartitionAssignor
> > > > > > to
> > > > > > > >> indicate
> > > > > > > >> > > > > the next HB telling broker about so. WDYT about adding
> > > > such
> > > > > an
> > > > > > > >> API on the
> > > > > > > >> > > > > PartitionAssignor?
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > Guozhang
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > > > wrote:
> > > > > > > >> > > > >
> > > > > > > >> > > > > > Hi Jun,
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > I have updated the KIP to include your feedback. I
> > have
> > > > > also
> > > > > > > >> tried to
> > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > Best,
> > > > > > > >> > > > > > David
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > djacot@confluent.io
> > > > > > > >> >
> > > > > > > >> > > > wrote:
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Hi Jun,
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > answering
> > > > your
> > > > > > > >> questions
> > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > of
> > > > the
> > > > > > KIP
> > > > > > > >> seem to
> > > > > > > >> > > > be
> > > > > > > >> > > > > > fewer
> > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > support of
> > > > > > > >> wildcard. A few
> > > > > > > >> > > > > > > > comments below.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > I would also add that the KIP removes the global
> > sync
> > > > > > barrier
> > > > > > > >> in the
> > > > > > > >> > > > > > > protocol which is essential to improve group
> > stability
> > > > > and
> > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > client by
> > > > > > moving
> > > > > > > >> most of
> > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > to
> > > > > > support
> > > > > > > >> rolling
> > > > > > > >> > > > > > changing
> > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > > assignor
> > > > > > used
> > > > > > > >> by a
> > > > > > > >> > > > > > > majority of the members. This allows the group to
> > move
> > > > > > from
> > > > > > > >> one
> > > > > > > >> > > > > > > assignor to another by a roll. This is explained
> > in
> > > > the
> > > > > > > >> Assignor
> > > > > > > >> > > > > > > Selection chapter.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > it's
> > > > > > > >> required in
> > > > > > > >> > > > every
> > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > filled?
> > > > > For
> > > > > > > >> example,
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > not
> > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > filled.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > The client is expected to set those fields in
> > case of
> > > > a
> > > > > > > >> connection
> > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > changed
> > > > > since
> > > > > > > >> the last
> > > > > > > >> > > > > > > HB. The server populates those fields as long as
> > the
> > > > > > member
> > > > > > > >> is not
> > > > > > > >> > > > > > > fully reconciled - the member should acknowledge
> > that
> > > > it
> > > > > > has
> > > > > > > >> the
> > > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> > this in
> > > > > the
> > > > > > KIP.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > > affinity
> > > > > > > >> between the
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > and the broker is only considered during
> > fetching,
> > > > but
> > > > > > not
> > > > > > > >> during
> > > > > > > >> > > > > > assigning
> > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > assignment
> > > > > > is
> > > > > > > >> made,
> > > > > > > >> > > > there
> > > > > > > >> > > > > > is
> > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > replicas
> > > > > of
> > > > > > > >> assigned
> > > > > > > >> > > > > > partitions
> > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > should
> > > > > use
> > > > > > > >> this
> > > > > > > >> > > > > > opportunity
> > > > > > > >> > > > > > > > to address this by including rack in
> > GroupMember.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's an interesting idea. I don't see any issue
> > with
> > > > > > adding
> > > > > > > >> the
> > > > > > > >> > > > rack
> > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > know
> > > > how
> > > > > > busy
> > > > > > > >> a group
> > > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> > it
> > > > > seems
> > > > > > > >> that we
> > > > > > > >> > > > could
> > > > > > > >> > > > > > add
> > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > the
> > > > > event
> > > > > > > >> loop is
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > actual work.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > failover
> > > > > > > >> handling? For
> > > > > > > >> > > > > > example,
> > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > with
> > > > > the
> > > > > > > >> wildcard
> > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sure. When the new group coordinator takes over,
> > it
> > > > has
> > > > > > to:
> > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > assignor
> > > > is
> > > > > > used.
> > > > > > > >> We
> > > > > > > >> > > > don't
> > > > > > > >> > > > > > > store the information about the member selected
> > to run
> > > > > the
> > > > > > > >> assignment
> > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > > > > >> subscriptions, and
> > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > document
> > > > > what
> > > > > > > >> the epoch
> > > > > > > >> > > > > > field
> > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > >> > > > > > reflect
> > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> > latest
> > > > > > group
> > > > > > > >> epoch.
> > > > > > > >> > > > > > > When the group state is updated, the group epoch
> > is
> > > > > > bumped so
> > > > > > > >> we use
> > > > > > > >> > > > > > > that one for all the change records related to the
> > > > > update.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > > following
> > > > > > > >> > > > invariants
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> > a
> > > > > member
> > > > > > > >> not to
> > > > > > > >> > > > get any
> > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's right. Here I meant that the members
> > provided
> > > > by
> > > > > > the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > > assignor
> > > > > > can
> > > > > > > >> not make
> > > > > > > >> > > > > > > up new member ids.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > epoch
> > > > > > equals to
> > > > > > > >> 0":
> > > > > > > >> > > > When
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > used?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > A member is expected to abandon all its
> > partitions and
> > > > > > > >> rejoins when
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this
> > case,
> > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > > > coordinator will have removed the member from the
> > > > group.
> > > > > > The
> > > > > > > >> member
> > > > > > > >> > > > > > > can rejoin the group with the same member id but
> > with
> > > > 0
> > > > > as
> > > > > > > >> epoch. Let
> > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> > to
> > > > > > trigger a
> > > > > > > >> > > > > > reassignment
> > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > updating
> > > > > the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > with
> > > > the
> > > > > > > >> deprecation
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > In this case, a new assignment is triggered by the
> > > > > client
> > > > > > side
> > > > > > > >> > > > > > > assignor. When constructing the HB, the consumer
> > will
> > > > > > always
> > > > > > > >> consult
> > > > > > > >> > > > > > > the client side assignor and propagate the
> > information
> > > > > to
> > > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > users to
> > > > > call
> > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > > section
> > > > > > > >> seems to
> > > > > > > >> > > > have
> > > > > > > >> > > > > > > > multiple typos.
> > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > immediately
> > > > > > > >> gets into
> > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > incorrect.
> > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > seems
> > > > to
> > > > > > get
> > > > > > > >> into
> > > > > > > >> > > > > > epoch=3,
> > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > has A -
> > > > > > > >> epoch=2,
> > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> > the
> > > > > > upgrade
> > > > > > > >> from
> > > > > > > >> > > > any
> > > > > > > >> > > > > > old
> > > > > > > >> > > > > > > > version to new one?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > We will support upgrading from the consumer
> > protocol
> > > > > > version
> > > > > > > >> 3,
> > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented
> > yet
> > > > so
> > > > > > the
> > > > > > > >> earliest
> > > > > > > >> > > > > > > version is unknown at the moment. This is
> > explained in
> > > > > the
> > > > > > > >> migration
> > > > > > > >> > > > > > > plan chapter.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> > update the
> > > > > KIP
> > > > > > > >> based on
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > next week.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Best,
> > > > > > > >> > > > > > > David
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > >> > > > wrote:
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Hi, David,
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > of
> > > > the
> > > > > > KIP
> > > > > > > >> seem to
> > > > > > > >> > > > be
> > > > > > > >> > > > > > fewer
> > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > support of
> > > > > > > >> wildcard. A few
> > > > > > > >> > > > > > > > comments below.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > to
> > > > > > support
> > > > > > > >> rolling
> > > > > > > >> > > > > > changing
> > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > it's
> > > > > > > >> required in
> > > > > > > >> > > > every
> > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > filled?
> > > > > For
> > > > > > > >> example,
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > not
> > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > filled.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > > affinity
> > > > > > > >> between the
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > and the broker is only considered during
> > fetching,
> > > > but
> > > > > > not
> > > > > > > >> during
> > > > > > > >> > > > > > assigning
> > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > assignment
> > > > > > is
> > > > > > > >> made,
> > > > > > > >> > > > there
> > > > > > > >> > > > > > is
> > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > replicas
> > > > > of
> > > > > > > >> assigned
> > > > > > > >> > > > > > partitions
> > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > should
> > > > > use
> > > > > > > >> this
> > > > > > > >> > > > > > opportunity
> > > > > > > >> > > > > > > > to address this by including rack in
> > GroupMember.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > know
> > > > how
> > > > > > busy
> > > > > > > >> a group
> > > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> > it
> > > > > seems
> > > > > > > >> that we
> > > > > > > >> > > > could
> > > > > > > >> > > > > > add
> > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > the
> > > > > event
> > > > > > > >> loop is
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > actual work.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > failover
> > > > > > > >> handling? For
> > > > > > > >> > > > > > example,
> > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > with
> > > > > the
> > > > > > > >> wildcard
> > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > document
> > > > > what
> > > > > > > >> the epoch
> > > > > > > >> > > > > > field
> > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > >> > > > > > reflect
> > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > > following
> > > > > > > >> > > > invariants
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> > a
> > > > > member
> > > > > > > >> not to
> > > > > > > >> > > > get any
> > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > epoch
> > > > > > equals to
> > > > > > > >> 0":
> > > > > > > >> > > > When
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > used?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> > to
> > > > > > trigger a
> > > > > > > >> > > > > > reassignment
> > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > updating
> > > > > the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > with
> > > > the
> > > > > > > >> deprecation
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > > section
> > > > > > > >> seems to
> > > > > > > >> > > > have
> > > > > > > >> > > > > > > > multiple typos.
> > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > immediately
> > > > > > > >> gets into
> > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > incorrect.
> > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > seems
> > > > to
> > > > > > get
> > > > > > > >> into
> > > > > > > >> > > > > > epoch=3,
> > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > has A -
> > > > > > > >> epoch=2,
> > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> > the
> > > > > > upgrade
> > > > > > > >> from
> > > > > > > >> > > > any
> > > > > > > >> > > > > > old
> > > > > > > >> > > > > > > > version to new one?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Thanks,
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Jun
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > wrote:
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > Hi all,
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > > > coordinator
> > > > > > > >> in
> > > > > > > >> > > > Java. I
> > > > > > > >> > > > > > > > > discussed this offline with a few folks and
> > folks
> > > > > are
> > > > > > > >> concerned
> > > > > > > >> > > > that
> > > > > > > >> > > > > > > > > we could introduce many regressions in the old
> > > > > > protocol
> > > > > > > >> if we do
> > > > > > > >> > > > so.
> > > > > > > >> > > > > > > > > Therefore, I am going to remove this statement
> > > > from
> > > > > > the
> > > > > > > >> KIP. It
> > > > > > > >> > > > is an
> > > > > > > >> > > > > > > > > implementation detail after all so it does not
> > > > have
> > > > > > to be
> > > > > > > >> > > > decided at
> > > > > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > > > > refactor the
> > > > > > > >> > > > current
> > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > Cheers,
> > > > > > > >> > > > > > > > > David
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > > > > >> djacot@confluent.io
> > > > > > > >> > > > >
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > "Empty,
> > > > > > assigning,
> > > > > > > >> > > > > > reconciling,
> > > > > > > >> > > > > > > > > stable,
> > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > > section,
> > > > > > > >> right?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > This sentence does not refer to those group
> > > > states
> > > > > > but
> > > > > > > >> rather
> > > > > > > >> > > > to a
> > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > refers to
> > > > > the
> > > > > > > >> entire
> > > > > > > >> > > > state of
> > > > > > > >> > > > > > > > > > group coordinator which is replicated via
> > the
> > > > log
> > > > > > > >> layer. I will
> > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> > is
> > > > > > modelled
> > > > > > > >> as an
> > > > > > > >> > > > event
> > > > > > > >> > > > > > > > > loop"?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> > the new
> > > > > > quorum
> > > > > > > >> > > > > > controller. We
> > > > > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > > > > >> __consumer_offsets
> > > > > > > >> > > > > > > > > > partition is assigned to a unique thread
> > and all
> > > > > the
> > > > > > > >> events
> > > > > > > >> > > > (e.g.
> > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by
> > this
> > > > > > > >> thread. This
> > > > > > > >> > > > > > simplify
> > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > simulation
> > > > > > testing
> > > > > > > >> for the
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > >> *__consumer_offsets*
> > > > > > > >> > > > > > > > > partitions?
> > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > owned
> > > > > by
> > > > > > a
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator?
> > > > > > > >> > > > > > > > > For
> > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > > consumer
> > > > > > > >> groups, and
> > > > > > > >> > > > > > both
> > > > > > > >> > > > > > > > > exist in
> > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > state
> > > > > > machine
> > > > > > > >> for it,
> > > > > > > >> > > > or
> > > > > > > >> > > > > > 2?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > > > think.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.4. I know the
> > "*group.coordinator.threads"
> > > > *is
> > > > > > the
> > > > > > > >> number
> > > > > > > >> > > > of
> > > > > > > >> > > > > > threads
> > > > > > > >> > > > > > > > > used
> > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > wondering
> > > > if
> > > > > > the
> > > > > > > >> purpose
> > > > > > > >> > > > of
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > threads
> > > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > > group
> > > > > > (or
> > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > why
> > > > > > should we
> > > > > > > >> need
> > > > > > > >> > > > > > > > > multi-threads
> > > > > > > >> > > > > > > > > > > here?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> > shard
> > > > > the
> > > > > > > >> > > > processing as
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > does
> > > > the
> > > > > > > >> default
> > > > > > > >> > > > session
> > > > > > > >> > > > > > > > > timeout not
> > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > thought
> > > > > > the
> > > > > > > >> min/max
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> > it,
> > > > > no?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > 30s The
> > > > > > > >> timeout to
> > > > > > > >> > > > detect
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > protocol.
> > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > int 45s
> > > > > The
> > > > > > > >> minimum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > int 60s
> > > > > The
> > > > > > > >> maximum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > session
> > > > > > timeout
> > > > > > > >> should
> > > > > > > >> > > > be 45s
> > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > > [range,
> > > > > > > >> uniform],
> > > > > > > >> > > > > > which means
> > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > like to
> > > > > > know
> > > > > > > >> why not
> > > > > > > >> > > > > > uniform
> > > > > > > >> > > > > > > > > one? I
> > > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > > assignor
> > > > > > > >> (former
> > > > > > > >> > > > sticky
> > > > > > > >> > > > > > > > > assinor)
> > > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > > reason
> > > > > > we
> > > > > > > >> choose
> > > > > > > >> > > > range
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > as default?
> > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > uniform
> > > > The
> > > > > > > >> server side
> > > > > > > >> > > > > > assignors.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > The order on the server side has no
> > influence
> > > > > > because
> > > > > > > >> the
> > > > > > > >> > > > client
> > > > > > > >> > > > > > must
> > > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> > There
> > > > is
> > > > > no
> > > > > > > >> default
> > > > > > > >> > > > in the
> > > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > > specified
> > > > > > by
> > > > > > > >> the
> > > > > > > >> > > > client,
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > request is rejected. The default client
> > value
> > > > for
> > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I
> > hope
> > > > > > that my
> > > > > > > >> > > > answers
> > > > > > > >> > > > > > help
> > > > > > > >> > > > > > > > > > to clarify things. I will update the KIP as
> > well
> > > > > > based
> > > > > > > >> on your
> > > > > > > >> > > > > > > > > > feedback.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > >> > > > > > > > > > David
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > > > > >> showuon@gmail.com>
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > mentioned:
> > > > > > > >> > > > > > > > > > > > The new group coordinator will have a
> > state
> > > > > > machine
> > > > > > > >> per
> > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> > each
> > > > > state
> > > > > > > >> machine is
> > > > > > > >> > > > > > modelled
> > > > > > > >> > > > > > > > > as an
> > > > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > > > executed
> > > > > > in
> > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > "Empty,
> > > > > > assigning,
> > > > > > > >> > > > > > reconciling,
> > > > > > > >> > > > > > > > > stable,
> > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > > section,
> > > > > > > >> right?
> > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> > is
> > > > > > modelled
> > > > > > > >> as an
> > > > > > > >> > > > event
> > > > > > > >> > > > > > > > > loop"?
> > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > >> *__consumer_offsets*
> > > > > > > >> > > > > > > > > partitions?
> > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > owned
> > > > > by
> > > > > > a
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator?
> > > > > > > >> > > > > > > > > For
> > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > > consumer
> > > > > > > >> groups, and
> > > > > > > >> > > > > > both
> > > > > > > >> > > > > > > > > exist in
> > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > state
> > > > > > machine
> > > > > > > >> for it,
> > > > > > > >> > > > or
> > > > > > > >> > > > > > 2?
> > > > > > > >> > > > > > > > > > > 1.4. I know the
> > "*group.coordinator.threads"
> > > > *is
> > > > > > the
> > > > > > > >> number
> > > > > > > >> > > > of
> > > > > > > >> > > > > > threads
> > > > > > > >> > > > > > > > > used
> > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > wondering
> > > > if
> > > > > > the
> > > > > > > >> purpose
> > > > > > > >> > > > of
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > threads
> > > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > > group
> > > > > > (or
> > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > why
> > > > > > should we
> > > > > > > >> need
> > > > > > > >> > > > > > > > > multi-threads
> > > > > > > >> > > > > > > > > > > here?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > configs:
> > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > does
> > > > the
> > > > > > > >> default
> > > > > > > >> > > > session
> > > > > > > >> > > > > > > > > timeout not
> > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > thought
> > > > > > the
> > > > > > > >> min/max
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> > it,
> > > > > no?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > 30s The
> > > > > > > >> timeout to
> > > > > > > >> > > > detect
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > protocol.
> > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > int 45s
> > > > > The
> > > > > > > >> minimum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > int 60s
> > > > > The
> > > > > > > >> maximum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > > [range,
> > > > > > > >> uniform],
> > > > > > > >> > > > > > which means
> > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > like to
> > > > > > know
> > > > > > > >> why not
> > > > > > > >> > > > > > uniform
> > > > > > > >> > > > > > > > > one? I
> > > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > > assignor
> > > > > > > >> (former
> > > > > > > >> > > > sticky
> > > > > > > >> > > > > > > > > assinor)
> > > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > > reason
> > > > > > we
> > > > > > > >> choose
> > > > > > > >> > > > range
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > as default?
> > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > uniform
> > > > The
> > > > > > > >> server side
> > > > > > > >> > > > > > assignors.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > >> > > > > > > > > > > Luke
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen
> > <
> > > > > > > >> showuon@gmail.com
> > > > > > > >> > > > >
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > > > > >> integrating with
> > > > > > > >> > > > > > KIP-848,
> > > > > > > >> > > > > > > > > but I
> > > > > > > >> > > > > > > > > > > > think we should have a separate
> > discussion
> > > > > > thread
> > > > > > > >> for the
> > > > > > > >> > > > Kafka
> > > > > > > >> > > > > > > > > Connect
> > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > > > > Consumer
> > > > > > > >> Rebalance
> > > > > > > >> > > > > > Protocol
> > > > > > > >> > > > > > > > > [1],
> > > > > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > > > > consumer
> > > > > > > >> rebalance
> > > > > > > >> > > > > > protocol,
> > > > > > > >> > > > > > > > > WDYT?
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > feedback.
> > > > > > Looks
> > > > > > > >> like
> > > > > > > >> > > > there's
> > > > > > > >> > > > > > > > > agreement on
> > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> > would
> > > > > > revisit
> > > > > > > >> the doc
> > > > > > > >> > > > and
> > > > > > > >> > > > > > see
> > > > > > > >> > > > > > > > > what
> > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David
> > Jacot
> > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > document.
> > > > > > That's
> > > > > > > >> really
> > > > > > > >> > > > > > helpful. I
> > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> > Connect
> > > > > > and the
> > > > > > > >> > > > Consumer
> > > > > > > >> > > > > > could
> > > > > > > >> > > > > > > > > share
> > > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> > main
> > > > > > > >> difference is
> > > > > > > >> > > > that
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > Consumer
> > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > > > whereas
> > > > > > > >> Connect
> > > > > > > >> > > > assigns
> > > > > > > >> > > > > > tasks
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > forward:
> > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > > > > support
> > > > > > > >> different
> > > > > > > >> > > > > > resource
> > > > > > > >> > > > > > > > > types
> > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > Connect.
> > > > > The
> > > > > > > >> dedicated
> > > > > > > >> > > > APIs
> > > > > > > >> > > > > > > > > would be
> > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > different on
> > > > > the
> > > > > > > >> > > > > > content/resources and
> > > > > > > >> > > > > > > > > > > >> > they would rely on the same engine
> > on the
> > > > > > > >> coordinator
> > > > > > > >> > > > side.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2) because
> > I am
> > > > > > not a
> > > > > > > >> fan of
> > > > > > > >> > > > > > > > > overcharging
> > > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> > That
> > > > > being
> > > > > > > >> said, I am
> > > > > > > >> > > > not
> > > > > > > >> > > > > > > > > opposed to
> > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to
> > do
> > > > it.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > discuss
> > > > it
> > > > > > here
> > > > > > > >> for now
> > > > > > > >> > > > in
> > > > > > > >> > > > > > order
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible
> > with
> > > > > what
> > > > > > we
> > > > > > > >> will do
> > > > > > > >> > > > for
> > > > > > > >> > > > > > > > > Connect in
> > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> > > > Jacot
> > > > > <
> > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > > > > through
> > > > > > and
> > > > > > > >> address
> > > > > > > >> > > > > > your
> > > > > > > >> > > > > > > > > comments
> > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > > > > feedback.
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> > Gregory
> > > > > > Harris
> > > > > > > >> <
> > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > wonderful to
> > > > > see
> > > > > > > >> > > > cooperative
> > > > > > > >> > > > > > > > > rebalancing
> > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > > > section
> > > > > > > >> describes how
> > > > > > > >> > > > > > member
> > > > > > > >> > > > > > > > > epoch
> > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > >> > > > > > > > > > > >> > > > advance in step with the group
> > epoch
> > > > > and
> > > > > > > >> assignment
> > > > > > > >> > > > > > epoch
> > > > > > > >> > > > > > > > > values. I
> > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > > > > reasons
> > > > > > > >> described
> > > > > > > >> > > > in
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > KIP. When
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > assigning
> > > > > > > >> partitions to a
> > > > > > > >> > > > > > worker,
> > > > > > > >> > > > > > > > > what
> > > > > > > >> > > > > > > > > > > >> member
> > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > assignment
> > > > > > use?
> > > > > > > >> Are
> > > > > > > >> > > > member
> > > > > > > >> > > > > > epochs
> > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > correspond
> > > > to
> > > > > > > >> multiple
> > > > > > > >> > > > > > different
> > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> > field
> > > > > > opaque
> > > > > > > >> to the
> > > > > > > >> > > > group
> > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > > > assignor
> > > > > > > >> > > > implementations
> > > > > > > >> > > > > > > > > interact
> > > > > > > >> > > > > > > > > > > >> with
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> > common
> > > > > > meaning
> > > > > > > >> agreed
> > > > > > > >> > > > > > upon? If
> > > > > > > >> > > > > > > > > so, what
> > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> > field
> > > > over
> > > > > > > >> including
> > > > > > > >> > > > such
> > > > > > > >> > > > > > > > > functionality
> > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 3. The following is included in
> > the
> > > > > KIP:
> > > > > > > >> "Thanks to
> > > > > > > >> > > > > > this, the
> > > > > > > >> > > > > > > > > input
> > > > > > > >> > > > > > > > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> > > > driven
> > > > > > by
> > > > > > > >> the group
> > > > > > > >> > > > > > > > > coordinator.
> > > > > > > >> > > > > > > > > > > >> The
> > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > responsible for
> > > > > > > >> maintaining
> > > > > > > >> > > > any
> > > > > > > >> > > > > > state
> > > > > > > >> > > > > > > > > besides
> > > > > > > >> > > > > > > > > > > >> its
> > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this
> > mean
> > > > > > that the
> > > > > > > >> > > > > > client-side
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > non-Metadata
> > > > > state
> > > > > > > >> (such as
> > > > > > > >> > > > > > partition
> > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > topics,
> > > > > etc),
> > > > > > or
> > > > > > > >> that
> > > > > > > >> > > > > > additional
> > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > > > > classes
> > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > >> > > > > >
> > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > >> > > > > > that
> > > > > > > >> > > > > > > > > seem to
> > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > possible
> > > > > for
> > > > > > > >> these two
> > > > > > > >> > > > > > > > > implementations
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> > promote
> > > > > > feature
> > > > > > > >> parity
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > > server-side
> > > > > > > >> > > > > > > > > > > >> and
> > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and would
> > also
> > > > > > > >> facilitate
> > > > > > > >> > > > > > operational
> > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > >> > > > > > > > > > > >> > > > certain situations. For example,
> > if a
> > > > > > > >> server-side
> > > > > > > >> > > > > > assignor
> > > > > > > >> > > > > > > > > has some
> > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > deploying
> > > > > the
> > > > > > > >> patched
> > > > > > > >> > > > > > assignor to
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer group
> > to a
> > > > > > > >> client-side
> > > > > > > >> > > > > > assignor
> > > > > > > >> > > > > > > > > may be
> > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > >> > > > > > > > > > > >> > > > and less risky than patching all
> > of
> > > > the
> > > > > > > >> brokers.
> > > > > > > >> > > > With
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > currently
> > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > non-trivial
> > > > > > > >> > > > reimplementation
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > have
> > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs
> > have
> > > > > > diverged
> > > > > > > >> > > > > > significantly,
> > > > > > > >> > > > > > > > > then it
> > > > > > > >> > > > > > > > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> > > > would
> > > > > > not be
> > > > > > > >> > > > possible.
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM
> > Sagar
> > > > <
> > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > > > > discuss
> > > > > > how
> > > > > > > >> Connect
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > need to
> > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > > protocol.
> > > > > > > >> Here's the
> > > > > > > >> > > > page:
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I
> > have
> > > > > > tried
> > > > > > > >> to keep
> > > > > > > >> > > > a
> > > > > > > >> > > > > > format
> > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > > > think.
> > > > > > Also,
> > > > > > > >> do you
> > > > > > > >> > > > > > think this
> > > > > > > >> > > > > > > > > > > >> should
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> > > > thread
> > > > > > or
> > > > > > > >> is this
> > > > > > > >> > > > one
> > > > > > > >> > > > > > fine?
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > > > > Sagar <
> > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> > doc on
> > > > > > Kafka
> > > > > > > >> Streams.
> > > > > > > >> > > > > > Sure, I
> > > > > > > >> > > > > > > > > would do
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> > such a
> > > > > > > >> document.
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47
> > AM
> > > > > > Guozhang
> > > > > > > >> Wang <
> > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> > could
> > > > > come
> > > > > > > >> back with
> > > > > > > >> > > > some
> > > > > > > >> > > > > > > > > analysis on
> > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > > > integration
> > > > > > > >> with
> > > > > > > >> > > > the new
> > > > > > > >> > > > > > > > > protocol;
> > > > > > > >> > > > > > > > > > > >> so
> > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the
> > new
> > > > > > "protocol
> > > > > > > >> type"
> > > > > > > >> > > > we
> > > > > > > >> > > > > > did not
> > > > > > > >> > > > > > > > > yet
> > > > > > > >> > > > > > > > > > > >> think
> > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > implementations.
> > > > > > For
> > > > > > > >> Streams
> > > > > > > >> > > > > > here's a
> > > > > > > >> > > > > > > > > draft of
> > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > > >
> > > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis
> > on
> > > > > > Connect.
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at
> > 10:48 PM
> > > > > > Sagar <
> > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > response.
> > > > The
> > > > > > > >> reason I
> > > > > > > >> > > > thought
> > > > > > > >> > > > > > > > > connect can
> > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that
> > even
> > > > > > today the
> > > > > > > >> > > > connect
> > > > > > > >> > > > > > uses a
> > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > AbstractCoordinator to
> > > > > > > >> empower
> > > > > > > >> > > > > > > > > rebalances of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > > > > >> WorkerCoordinator sets
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata()
> > method
> > > > > by
> > > > > > > >> plumbing
> > > > > > > >> > > > into
> > > > > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to
> > support
> > > > > > > >> connect would
> > > > > > > >> > > > be
> > > > > > > >> > > > > > > > > similar at a
> > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams
> > mainly
> > > > > > because
> > > > > > > >> of the
> > > > > > > >> > > > > > Client
> > > > > > > >> > > > > > > > > side
> > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > > implementation
> > > > > > > >> level, we
> > > > > > > >> > > > > > might
> > > > > > > >> > > > > > > > > need to
> > > > > > > >> > > > > > > > > > > >> make
> > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this
> > new
> > > > > > > >> assignment
> > > > > > > >> > > > protocol
> > > > > > > >> > > > > > like
> > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > request/response and
> > > > > > > >> SyncGroup and
> > > > > > > >> > > > > > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar
> > lines
> > > > > to
> > > > > > > >> streams (or
> > > > > > > >> > > > > > there
> > > > > > > >> > > > > > > > > might be
> > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> > > > detailed
> > > > > > > >> analysis of
> > > > > > > >> > > > the
> > > > > > > >> > > > > > same
> > > > > > > >> > > > > > > > > and we
> > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> > thread for
> > > > > > that
> > > > > > > >> as that
> > > > > > > >> > > > would
> > > > > > > >> > > > > > > > > derail this
> > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if
> > that
> > > > > sounds
> > > > > > > >> good to
> > > > > > > >> > > > you.
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at
> > 5:47
> > > > PM
> > > > > > David
> > > > > > > >> Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > comments.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > > > > >> `Assignment#error`.
> > > > > > > >> > > > > > Sure, I
> > > > > > > >> > > > > > > > > can
> > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > > transition C
> > > > > > from
> > > > > > > >> his
> > > > > > > >> > > > current
> > > > > > > >> > > > > > > > > assignment
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when
> > he
> > > > can
> > > > > > move
> > > > > > > >> to
> > > > > > > >> > > > epoch 3.
> > > > > > > >> > > > > > When
> > > > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is
> > updated
> > > > > and
> > > > > > > >> persisted
> > > > > > > >> > > > > > with all
> > > > > > > >> > > > > > > > > its
> > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if they
> > are
> > > > > not
> > > > > > all
> > > > > > > >> revoked
> > > > > > > >> > > > > > yet. In
> > > > > > > >> > > > > > > > > other
> > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > becomes
> > > > the
> > > > > > target
> > > > > > > >> > > > > > assignment.
> > > > > > > >> > > > > > > > > This is
> > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid
> > having
> > > > > to
> > > > > > > >> write all
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > changes to
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on
> > the
> > > > > > persisted
> > > > > > > >> state
> > > > > > > >> > > > so I
> > > > > > > >> > > > > > > > > understand
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see
> > if I
> > > > can
> > > > > > > >> improve
> > > > > > > >> > > > this in
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it
> > > > could
> > > > > > > >> reuse the
> > > > > > > >> > > > > > protocol
> > > > > > > >> > > > > > > > > with a
> > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in
> > the
> > > > > > > >> protocol. The
> > > > > > > >> > > > > > assignment
> > > > > > > >> > > > > > > > > is
> > > > > > > >> > > > > > > > > > > >> about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> > > > metadata,
> > > > > > could
> > > > > > > >> > > > Connect
> > > > > > > >> > > > > > fit
> > > > > > > >> > > > > > > > > into this?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at
> > 1:55
> > > > > PM
> > > > > > > >> Sagar <
> > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I
> > just
> > > > > had
> > > > > > > >> minor
> > > > > > > >> > > > > > observations:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment
> > Error
> > > > > > > >> section in
> > > > > > > >> > > > Client
> > > > > > > >> > > > > > Side
> > > > > > > >> > > > > > > > > mode
> > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In
> > this
> > > > > > case,
> > > > > > > >> the
> > > > > > > >> > > > client
> > > > > > > >> > > > > > side
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > >> can
> > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > > > > coordinator`.
> > > > > > > >> In this
> > > > > > > >> > > > > > case are
> > > > > > > >> > > > > > > > > you
> > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > > > > > >> AssignmentError
> > > > > > > >> > > > that's
> > > > > > > >> > > > > > > > > listed down
> > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think
> > it
> > > > > would
> > > > > > > >> make sense
> > > > > > > >> > > > to
> > > > > > > >> > > > > > > > > mention this
> > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > > > > > section, I
> > > > > > > >> have a
> > > > > > > >> > > > > > slight
> > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the same.
> > > > > Consider
> > > > > > > >> this step:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the
> > > > group
> > > > > > > >> coordinator
> > > > > > > >> > > > > > > > > transitions him
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > > > > > revoke. It
> > > > > > > >> > > > persists
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > change and
> > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > > > > > (epoch=3)
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > partitions=[foo-0]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > partitions=[foo-2]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > partitions=[foo-1]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > > > > >> partitions=[foo-0,
> > > > > > > >> > > > > > foo-1]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > > > > >> partitions=[foo-2]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > > > > >> partitions=[foo-1]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > > > > > transitions
> > > > > > > >> to
> > > > > > > >> > > > epoch 3
> > > > > > > >> > > > > > but
> > > > > > > >> > > > > > > > > cannot
> > > > > > > >> > > > > > > > > > > >> get
> > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned
> > that
> > > > > > member C
> > > > > > > >> can't
> > > > > > > >> > > > get
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > foo-1
> > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > description
> > > > > > above,
> > > > > > > >> it seems
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > > > already has
> > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to
> > remove
> > > > it
> > > > > > and
> > > > > > > >> > > > populate it
> > > > > > > >> > > > > > only
> > > > > > > >> > > > > > > > > when it
> > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of
> > > > other
> > > > > > > >> places, so
> > > > > > > >> > > > have
> > > > > > > >> > > > > > I
> > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it
> > > > might
> > > > > > be
> > > > > > > >> out of
> > > > > > > >> > > > scope
> > > > > > > >> > > > > > of
> > > > > > > >> > > > > > > > > this
> > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it
> > would
> > > > > > > >> probably be
> > > > > > > >> > > > > > running in
> > > > > > > >> > > > > > > > > client
> > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> > rebalance
> > > > > > > >> protocol as it
> > > > > > > >> > > > has
> > > > > > > >> > > > > > its
> > > > > > > >> > > > > > > > > own
> > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > IncrementalCooperative).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022
> > at
> > > > 5:00
> > > > > > PM
> > > > > > > >> David
> > > > > > > >> > > > Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our
> > goal
> > > > is
> > > > > > to
> > > > > > > >> move
> > > > > > > >> > > > forward
> > > > > > > >> > > > > > with
> > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying
> > on
> > > > one
> > > > > > > >> generic API.
> > > > > > > >> > > > For
> > > > > > > >> > > > > > > > > Connect, we
> > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern
> > and
> > > > > > > >> reuse/share the
> > > > > > > >> > > > core
> > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the
> > > > > schema
> > > > > > > >> registry, I
> > > > > > > >> > > > > > think
> > > > > > > >> > > > > > > > > that we
> > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored
> > API to
> > > > > do
> > > > > > > >> simple
> > > > > > > >> > > > > > > > > membership/leader
> > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15,
> > 2022 at
> > > > > > 10:22
> > > > > > > >> AM Ismael
> > > > > > > >> > > > > > Juma <
> > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> > comments:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > > >> java.util.regex.Pattern
> > > > > > > >> > > > vs
> > > > > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > > differences
> > > > > > in
> > > > > > > >> more
> > > > > > > >> > > > detail
> > > > > > > >> > > > > > before
> > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if
> > people
> > > > > pass
> > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > >> > > > > > > > > > > >> they
> > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be
> > > > honored.
> > > > > > If
> > > > > > > >> we are
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > > > > something
> > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider
> > adding
> > > > > an
> > > > > > > >> overload
> > > > > > > >> > > > with
> > > > > > > >> > > > > > our own
> > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to
> > > > > expose
> > > > > > > >> re2j's at
> > > > > > > >> > > > this
> > > > > > > >> > > > > > > > > point).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic
> > > > ids,
> > > > > > any
> > > > > > > >> major new
> > > > > > > >> > > > > > protocol
> > > > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should
> > handle
> > > > the
> > > > > > topic
> > > > > > > >> > > > > > recreation case
> > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to
> > > > handle.
> > > > > I
> > > > > > > >> agree with
> > > > > > > >> > > > > > David
> > > > > > > >> > > > > > > > > that we'd
> > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > > > > protocols
> > > > > > > >> that don't
> > > > > > > >> > > > > > have it
> > > > > > > >> > > > > > > > > yet and
> > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> > internals
> > > > > > versus
> > > > > > > >> adding
> > > > > > > >> > > > new
> > > > > > > >> > > > > > APIs
> > > > > > > >> > > > > > > > > to the
> > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> > adding
> > > > new
> > > > > > APIs
> > > > > > > >> is
> > > > > > > >> > > > required
> > > > > > > >> > > > > > for
> > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not
> > sure
> > > > > > about
> > > > > > > >> the
> > > > > > > >> > > > > > coordinator
> > > > > > > >> > > > > > > > > storing
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> > configs
> > > > to
> > > > > be
> > > > > > > >> > > > centralized
> > > > > > > >> > > > > > in the
> > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > (auditability,
> > > > > > > >> visibility,
> > > > > > > >> > > > > > consistency,
> > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > automatically
> > > > > > > >> deleting
> > > > > > > >> > > > > > configs in a
> > > > > > > >> > > > > > > > > way
> > > > > > > >> > > > > > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> > > > > property
> > > > > > > >> for modern
> > > > > > > >> > > > > > systems
> > > > > > > >> > > > > > > > > is to
> > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data
> > > > loss
> > > > > > > >> scenarios.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13,
> > 2022
> > > > at
> > > > > > 3:47
> > > > > > > >> PM David
> > > > > > > >> > > > > > Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > Guozhang. My
> > > > > > answers
> > > > > > > >> are
> > > > > > > >> > > > below:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > migration
> > > > > path,
> > > > > > > >> especially
> > > > > > > >> > > > > > the last
> > > > > > > >> > > > > > > > > step
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the
> > new
> > > > > > > >> protocol, in
> > > > > > > >> > > > which
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> > rpcs
> > > > and
> > > > > > old
> > > > > > > >> > > > protocols /
> > > > > > > >> > > > > > rpcs
> > > > > > > >> > > > > > > > > are
> > > > > > > >> > > > > > > > > > > >> used
> > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > > > >> coordinator could
> > > > > > > >> > > > > > "mimic"
> > > > > > > >> > > > > > > > > the old
> > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> > is
> > > > > > something
> > > > > > > >> we
> > > > > > > >> > > > need to
> > > > > > > >> > > > > > > > > present
> > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > > > > published
> > > > > > a
> > > > > > > >> new
> > > > > > > >> > > > version
> > > > > > > >> > > > > > of KIP
> > > > > > > >> > > > > > > > > which
> > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about
> > this.
> > > > > See
> > > > > > the
> > > > > > > >> > > > "Supporting
> > > > > > > >> > > > > > > > > Online
> > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > "Compatibility,
> > > > > > > >> > > > Deprecation,
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > Migration
> > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> > > > > through
> > > > > > a
> > > > > > > >> few cases
> > > > > > > >> > > > > > now but
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> > should be
> > > > > > > >> understandable.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> > of
> > > > > topic
> > > > > > > >> ids. So
> > > > > > > >> > > > far as
> > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> > and
> > > > > admin
> > > > > > > >> client,
> > > > > > > >> > > > but
> > > > > > > >> > > > > > they
> > > > > > > >> > > > > > > > > are not
> > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> > yet.
> > > > I
> > > > > > think
> > > > > > > >> the
> > > > > > > >> > > > > > question is,
> > > > > > > >> > > > > > > > > first
> > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > > maintaining
> > > > > > > >> the
> > > > > > > >> > > > names
> > > > > > > >> > > > > > -> ids
> > > > > > > >> > > > > > > > > mapping
> > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > augmented
> > > > > > > >> existing
> > > > > > > >> > > > RPCs
> > > > > > > >> > > > > > and the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > should
> > > > we
> > > > > > ever
> > > > > > > >> consider
> > > > > > > >> > > > > > exposing
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > > > well
> > > > > > (both
> > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > >> > > > > > > > > > > >> well
> > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> > cases
> > > > > > like
> > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming
> > that we
> > > > > > would
> > > > > > > >> include
> > > > > > > >> > > > > > converting
> > > > > > > >> > > > > > > > > all
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using
> > topic
> > > > > ids
> > > > > > in
> > > > > > > >> this
> > > > > > > >> > > > KIP,
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate
> > with
> > > > > topic
> > > > > > > >> ids. That
> > > > > > > >> > > > > > being
> > > > > > > >> > > > > > > > > said, it
> > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > names in
> > > > > > various
> > > > > > > >> APIs so
> > > > > > > >> > > > > > having a
> > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > inevitable to
> > > > > me.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have
> > a
> > > > > strong
> > > > > > > >> opinion on
> > > > > > > >> > > > > > this.
> > > > > > > >> > > > > > > > > Here I
> > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> > scope of
> > > > > this
> > > > > > > >> KIP. I
> > > > > > > >> > > > would
> > > > > > > >> > > > > > rather
> > > > > > > >> > > > > > > > > focus
> > > > > > > >> > > > > > > > > > > >> on
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> > > > > consider
> > > > > > > >> this
> > > > > > > >> > > > > > separately if
> > > > > > > >> > > > > > > > > we see
> > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > > > > Ismael's
> > > > > > > >> point about
> > > > > > > >> > > > > > using
> > > > > > > >> > > > > > > > > topic ids
> > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > >> > > > think
> > > > > > > >> > > > > > that
> > > > > > > >> > > > > > > > > there
> > > > > > > >> > > > > > > > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it.
> > The
> > > > > > consumer
> > > > > > > >> will
> > > > > > > >> > > > have
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > exists
> > > > > before
> > > > > > > >> passing
> > > > > > > >> > > > them
> > > > > > > >> > > > > > into
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> group
> > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> > Obviously,
> > > > > the
> > > > > > > >> > > > coordinator
> > > > > > > >> > > > > > will
> > > > > > > >> > > > > > > > > also
> > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have
> > a
> > > > way
> > > > > to
> > > > > > > >> reject an
> > > > > > > >> > > > > > invalid
> > > > > > > >> > > > > > > > > topic in
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> > with
> > > > > > David
> > > > > > > >> on all
> > > > > > > >> > > > other
> > > > > > > >> > > > > > minor
> > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > `subscribe(Pattern)`
> > > > > > > >> question:
> > > > > > > >> > > > > > personally
> > > > > > > >> > > > > > > > > I think
> > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > > > subscribe
> > > > > > > >> API with
> > > > > > > >> > > > > > Pattern,
> > > > > > > >> > > > > > > > > but
> > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > documenting
> > > > > > > >> that our
> > > > > > > >> > > > > > > > > subscription may
> > > > > > > >> > > > > > > > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > incompatible
> > > > > > > >> case is a
> > > > > > > >> > > > > > very rare
> > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > `String`
> > > > > > based
> > > > > > > >> > > > > > subscription may
> > > > > > > >> > > > > > > > > be
> > > > > > > >> > > > > > > > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > regexes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> > work. I
> > > > > have
> > > > > > to
> > > > > > > >> look at
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > differences
> > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> > better
> > > > > > > >> understand the
> > > > > > > >> > > > > > potential
> > > > > > > >> > > > > > > > > issues.
> > > > > > > >> > > > > > > > > > > >> My
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work
> > for
> > > > > all
> > > > > > the
> > > > > > > >> basic
> > > > > > > >> > > > > > regular
> > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two
> > are
> > > > > > mainly
> > > > > > > >> about
> > > > > > > >> > > > the
> > > > > > > >> > > > > > various
> > > > > > > >> > > > > > > > > > > >> character
> > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what
> > other
> > > > > people
> > > > > > > >> think
> > > > > > > >> > > > about
> > > > > > > >> > > > > > this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12,
> > 2022
> > > > > at
> > > > > > > >> 11:28 PM
> > > > > > > >> > > > > > Guozhang
> > > > > > > >> > > > > > > > > Wang <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David!
> > I
> > > > > think
> > > > > > on
> > > > > > > >> the high
> > > > > > > >> > > > > > level
> > > > > > > >> > > > > > > > > there are
> > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize
> > a
> > > > bit
> > > > > > more:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > migration
> > > > > path,
> > > > > > > >> especially
> > > > > > > >> > > > > > the last
> > > > > > > >> > > > > > > > > step
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the
> > new
> > > > > > > >> protocol, in
> > > > > > > >> > > > which
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> > rpcs
> > > > and
> > > > > > old
> > > > > > > >> > > > protocols /
> > > > > > > >> > > > > > rpcs
> > > > > > > >> > > > > > > > > are
> > > > > > > >> > > > > > > > > > > >> used
> > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > > > >> coordinator could
> > > > > > > >> > > > > > "mimic"
> > > > > > > >> > > > > > > > > the old
> > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> > is
> > > > > > something
> > > > > > > >> we
> > > > > > > >> > > > need to
> > > > > > > >> > > > > > > > > present
> > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> > of
> > > > > topic
> > > > > > > >> ids. So
> > > > > > > >> > > > far as
> > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> > and
> > > > > admin
> > > > > > > >> client,
> > > > > > > >> > > > but
> > > > > > > >> > > > > > they
> > > > > > > >> > > > > > > > > are not
> > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> > yet.
> > > > I
> > > > > > think
> > > > > > > >> the
> > > > > > > >> > > > > > question is,
> > > > > > > >> > > > > > > > > first
> > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > > maintaining
> > > > > > > >> the
> > > > > > > >> > > > names
> > > > > > > >> > > > > > -> ids
> > > > > > > >> > > > > > > > > mapping
> > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > augmented
> > > > > > > >> existing
> > > > > > > >> > > > RPCs
> > > > > > > >> > > > > > and the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > should
> > > > we
> > > > > > ever
> > > > > > > >> consider
> > > > > > > >> > > > > > exposing
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > > > well
> > > > > > (both
> > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > >> > > > > > > > > > > >> well
> > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> > cases
> > > > > > like
> > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> > with
> > > > > > David
> > > > > > > >> on all
> > > > > > > >> > > > other
> > > > > > > >> > > > > > minor
> > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > `subscribe(Pattern)`
> > > > > > > >> question:
> > > > > > > >> > > > > > personally
> > > > > > > >> > > > > > > > > I think
> > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > > > subscribe
> > > > > > > >> API with
> > > > > > > >> > > > > > Pattern,
> > > > > > > >> > > > > > > > > but
> > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > documenting
> > > > > > > >> that our
> > > > > > > >> > > > > > > > > subscription may
> > > > > > > >> > > > > > > > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > incompatible
> > > > > > > >> case is a
> > > > > > > >> > > > > > very rare
> > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > `String`
> > > > > > based
> > > > > > > >> > > > > > subscription may
> > > > > > > >> > > > > > > > > be
> > > > > > > >> > > > > > > > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > regexes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul
> > 12,
> > > > 2022
> > > > > > at
> > > > > > > >> 5:23 AM
> > > > > > > >> > > > > > David Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for
> > your
> > > > > > > >> feedback. Let
> > > > > > > >> > > > me
> > > > > > > >> > > > > > answer
> > > > > > > >> > > > > > > > > your
> > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think
> > > > it's
> > > > > > > >> premature to
> > > > > > > >> > > > > > talk about
> > > > > > > >> > > > > > > > > > > >> target
> > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal
> > of the
> > > > > > > >> existing group
> > > > > > > >> > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> > > > protocol
> > > > > > and
> > > > > > > >> hence
> > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> > quickly
> > > > > > > >> applications
> > > > > > > >> > > > migrate
> > > > > > > >> > > > > > to
> > > > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes
> > > > sense.
> > > > > I
> > > > > > will
> > > > > > > >> > > > remove it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP
> > > > says
> > > > > we
> > > > > > > >> intend to
> > > > > > > >> > > > > > release
> > > > > > > >> > > > > > > > > this in
> > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we
> > > > added
> > > > > > that
> > > > > > > >> as a
> > > > > > > >> > > > way to
> > > > > > > >> > > > > > > > > estimate
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > > > > > protocol, I
> > > > > > > >> also
> > > > > > > >> > > > > > suggest
> > > > > > > >> > > > > > > > > removing
> > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > explain
> > > > my
> > > > > > > >> reasoning. As
> > > > > > > >> > > > > > > > > explained, I
> > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> > in
> > > > > Java
> > > > > > > >> while we
> > > > > > > >> > > > > > implement
> > > > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> > > > internals
> > > > > > will
> > > > > > > >> be
> > > > > > > >> > > > slightly
> > > > > > > >> > > > > > > > > different
> > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I
> > > > > wanted
> > > > > > to
> > > > > > > >> tighten
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > switch from
> > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> > to
> > > > the
> > > > > > new
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator
> > > > > > > >> > > > > > > > > to a
> > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative
> > > > would
> > > > > > be to
> > > > > > > >> use a
> > > > > > > >> > > > flag
> > > > > > > >> > > > > > to do
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> > software
> > > > > > upgrade.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We
> > need to
> > > > > > flesh
> > > > > > > >> out the
> > > > > > > >> > > > > > details of
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're
> > saying
> > > > we
> > > > > > will
> > > > > > > >> support
> > > > > > > >> > > > > > online
> > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain
> > this
> > > > in
> > > > > > > >> detail. It
> > > > > > > >> > > > could
> > > > > > > >> > > > > > also
> > > > > > > >> > > > > > > > > be done
> > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> > > > > support
> > > > > > > >> online
> > > > > > > >> > > > > > migrations
> > > > > > > >> > > > > > > > > for the
> > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group
> > using
> > > > the
> > > > > > old
> > > > > > > >> protocol
> > > > > > > >> > > > > > will be
> > > > > > > >> > > > > > > > > able to
> > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > briefly
> > > > > > explain
> > > > > > > >> how that
> > > > > > > >> > > > > > will work
> > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step
> > > > process:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> > cluster
> > > > > must
> > > > > > be
> > > > > > > >> > > > upgraded or
> > > > > > > >> > > > > > > > > rolled to a
> > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > > > > coordinator.
> > > > > > > >> Both the
> > > > > > > >> > > > > > old and
> > > > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the
> > old
> > > > > > > >> protocol and
> > > > > > > >> > > > rely
> > > > > > > >> > > > > > on the
> > > > > > > >> > > > > > > > > same
> > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can
> > work
> > > > > > together.
> > > > > > > >> This
> > > > > > > >> > > > point
> > > > > > > >> > > > > > is an
> > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one
> > live
> > > > > > > >> because it
> > > > > > > >> > > > would
> > > > > > > >> > > > > > require
> > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> > and
> > > > > > > >> starting up the
> > > > > > > >> > > > > > new one
> > > > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > unavailabilities.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> > cluster's
> > > > > > > >> metadata
> > > > > > > >> > > > > > version/IBP
> > > > > > > >> > > > > > > > > must be
> > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable
> > the
> > > > new
> > > > > > > >> protocol.
> > > > > > > >> > > > This
> > > > > > > >> > > > > > cannot
> > > > > > > >> > > > > > > > > be done
> > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated
> > > > because
> > > > > > the
> > > > > > > >> old
> > > > > > > >> > > > > > coordinator
> > > > > > > >> > > > > > > > > doesn't
> > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> > consumers
> > > > > > must be
> > > > > > > >> > > > upgraded
> > > > > > > >> > > > > > to a
> > > > > > > >> > > > > > > > > version
> > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> > (must
> > > > > have
> > > > > > > >> KIP-792).
> > > > > > > >> > > > If
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing
> > must be
> > > > > > done at
> > > > > > > >> this
> > > > > > > >> > > > point.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> > consumers
> > > > > > must
> > > > > > > >> be rolled
> > > > > > > >> > > > > > with the
> > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer
> > group
> > > > is
> > > > > > > >> automatically
> > > > > > > >> > > > > > > > > converted when
> > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the
> > new
> > > > > > protocol
> > > > > > > >> joins
> > > > > > > >> > > > the
> > > > > > > >> > > > > > group.
> > > > > > > >> > > > > > > > > While
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > protocol are
> > > > > > being
> > > > > > > >> > > > upgraded,
> > > > > > > >> > > > > > the old
> > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > clarify
> > > > all
> > > > > > of
> > > > > > > >> this in
> > > > > > > >> > > > the
> > > > > > > >> > > > > > KIP.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am
> > happy
> > > > > > that we
> > > > > > > >> are
> > > > > > > >> > > > > > pushing the
> > > > > > > >> > > > > > > > > pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it
> > seems
> > > > > like
> > > > > > > >> there
> > > > > > > >> > > > could be
> > > > > > > >> > > > > > some
> > > > > > > >> > > > > > > > > tricky
> > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> > > > mechanism
> > > > > > for
> > > > > > > >> users to
> > > > > > > >> > > > > > detect
> > > > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> they
> > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> > > > switching
> > > > > > to
> > > > > > > >> the new
> > > > > > > >> > > > > > protocol?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think
> > that I
> > > > am
> > > > > a
> > > > > > bit
> > > > > > > >> more
> > > > > > > >> > > > > > optimistic
> > > > > > > >> > > > > > > > > than
> > > > > > > >> > > > > > > > > > > >> you
> > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe
> > that the
> > > > > > > >> majority of
> > > > > > > >> > > > the
> > > > > > > >> > > > > > cases
> > > > > > > >> > > > > > > > > are
> > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with
> > the
> > > > new
> > > > > > > >> engine. The
> > > > > > > >> > > > > > > > > coordinator will
> > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject
> > the
> > > > > > consumer
> > > > > > > >> if the
> > > > > > > >> > > > > > regex is
> > > > > > > >> > > > > > > > > not
> > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> > path,
> > > > in
> > > > > > the
> > > > > > > >> worst
> > > > > > > >> > > > case,
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > first
> > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group
> > will
> > > > be
> > > > > > > >> rejected.
> > > > > > > >> > > > This
> > > > > > > >> > > > > > should
> > > > > > > >> > > > > > > > > be used
> > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > > > > > customers to
> > > > > > > >> > > > validate
> > > > > > > >> > > > > > their
> > > > > > > >> > > > > > > > > regex
> > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would
> > be to
> > > > > > test
> > > > > > > >> them with
> > > > > > > >> > > > > > another
> > > > > > > >> > > > > > > > > group.
> > > > > > > >> > > > > > > > > > > >> For
> > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be
> > done
> > > > in a
> > > > > > > >> pre-prod
> > > > > > > >> > > > > > environment.
> > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > > > > >> consumer-group tool
> > > > > > > >> > > > to
> > > > > > > >> > > > > > > > > provide a
> > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism.
> > Would
> > > > > > this be
> > > > > > > >> > > > enough in
> > > > > > > >> > > > > > your
> > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > Related to
> > > > > the
> > > > > > last
> > > > > > > >> > > > question,
> > > > > > > >> > > > > > will
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick
> > with the
> > > > > > > >> current regex
> > > > > > > >> > > > > > engine for
> > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be
> > > > handy
> > > > > to
> > > > > > > >> keep using
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > based
> > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations
> > > > > simple
> > > > > > and
> > > > > > > >> then
> > > > > > > >> > > > > > migrate to
> > > > > > > >> > > > > > > > > server
> > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand
> > > > your
> > > > > > point
> > > > > > > >> but I
> > > > > > > >> > > > am
> > > > > > > >> > > > > > > > > concerned that
> > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually
> > stay
> > > > > in
> > > > > > > >> this mode.
> > > > > > > >> > > > That
> > > > > > > >> > > > > > > > > would go
> > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying
> > the
> > > > > > client
> > > > > > > >> because
> > > > > > > >> > > > we
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > have to
> > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > metadata on
> > > > > the
> > > > > > > >> client
> > > > > > > >> > > > side. I
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > rather
> > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When
> > we say
> > > > > > that
> > > > > > > >> the group
> > > > > > > >> > > > > > > > > coordinator
> > > > > > > >> > > > > > > > > > > >> will
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > configurations
> > > > > > > >> and that
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > > > > deleted.
> > > > > > > >> Will a
> > > > > > > >> > > > > > transition to
> > > > > > > >> > > > > > > > > DEAD
> > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > configurations?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> > right.
> > > > The
> > > > > > > >> > > > configurations
> > > > > > > >> > > > > > will be
> > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted.
> > They go
> > > > > > > >> together.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will
> > the
> > > > > > choice to
> > > > > > > >> store
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > configs in
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to
> > list
> > > > > all
> > > > > > > >> cluster
> > > > > > > >> > > > > > configs and
> > > > > > > >> > > > > > > > > their
> > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't
> > think
> > > > so.
> > > > > > The
> > > > > > > >> group
> > > > > > > >> > > > > > > > > configurations are
> > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If
> > you
> > > > > > want to
> > > > > > > >> know
> > > > > > > >> > > > all
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > overrides
> > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all
> > the
> > > > > group
> > > > > > > >> > > > coordinators.
> > > > > > > >> > > > > > You
> > > > > > > >> > > > > > > > > cannot
> > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > instance.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How
> > would
> > > > > > someone
> > > > > > > >> > > > configure a
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > before
> > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > considered
> > > > > > > >> allowing the
> > > > > > > >> > > > > > explicit
> > > > > > > >> > > > > > > > > creation
> > > > > > > >> > > > > > > > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs
> > could
> > > > be
> > > > > > > >> decoupled
> > > > > > > >> > > > from
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The
> > group
> > > > > will
> > > > > > be
> > > > > > > >> > > > > > automatically
> > > > > > > >> > > > > > > > > created in
> > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs
> > will
> > > > > be
> > > > > > > >> lost after
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > retention
> > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will
> > the
> > > > > > > >> > > > Consumer.subscribe
> > > > > > > >> > > > > > method
> > > > > > > >> > > > > > > > > for the
> > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > >> > > > do
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > have to
> > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a
> > very
> > > > > group
> > > > > > > >> question. I
> > > > > > > >> > > > > > forgot
> > > > > > > >> > > > > > > > > about
> > > > > > > >> > > > > > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > >> > > > not
> > > > > > > >> > > > > > fully
> > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to
> > use,
> > > > it
> > > > > > > >> might be
> > > > > > > >> > > > better
> > > > > > > >> > > > > > to
> > > > > > > >> > > > > > > > > deprecate
> > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes
> > a
> > > > > > string.
> > > > > > > >> We would
> > > > > > > >> > > > > > rely on
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> server
> > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > > > > migration, I
> > > > > > > >> think
> > > > > > > >> > > > that
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > could
> > > > > > > >> > > > > > > > > > > >> still
> > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and
> > use
> > > > it.
> > > > > > That
> > > > > > > >> should
> > > > > > > >> > > > > > work, I
> > > > > > > >> > > > > > > > > think, in
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I
> > agree
> > > > with
> > > > > > > >> Justine
> > > > > > > >> > > > that we
> > > > > > > >> > > > > > > > > should be
> > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > > > > >> > > > IBP/metadata.version
> > > > > > > >> > > > > > from the
> > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> > switch the
> > > > > > > >> default for
> > > > > > > >> > > > the
> > > > > > > >> > > > > > feature
> > > > > > > >> > > > > > > > > flag
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > metadata.version
> > > > > > once
> > > > > > > >> we
> > > > > > > >> > > > want to
> > > > > > > >> > > > > > make
> > > > > > > >> > > > > > > > > it the
> > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was
> > to
> > > > use
> > > > > > that
> > > > > > > >> feature
> > > > > > > >> > > > > > flag
> > > > > > > >> > > > > > > > > mainly
> > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I
> > should
> > > > > not
> > > > > > have
> > > > > > > >> > > > mentioned
> > > > > > > >> > > > > > it, I
> > > > > > > >> > > > > > > > > think,
> > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal
> > > > config
> > > > > > for
> > > > > > > >> it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some
> > of
> > > > the
> > > > > > > >> protocol APIs
> > > > > > > >> > > > > > don't
> > > > > > > >> > > > > > > > > mention
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to
> > add
> > > > that
> > > > > > for
> > > > > > > >> > > > consistency.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is
> > a
> > > > bit
> > > > > > odd
> > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it
> > > > seems
> > > > > > to do
> > > > > > > >> more
> > > > > > > >> > > > than
> > > > > > > >> > > > > > > > > reading.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree.
> > This is
> > > > > > how the
> > > > > > > >> > > > current
> > > > > > > >> > > > > > > > > protocol works
> > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require
> > "Read
> > > > > > Group" to
> > > > > > > >> join a
> > > > > > > >> > > > > > group. We
> > > > > > > >> > > > > > > > > could
> > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not
> > > > sure
> > > > > > that
> > > > > > > >> it is
> > > > > > > >> > > > worth
> > > > > > > >> > > > > > it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is
> > > > topic
> > > > > > > >> recreation
> > > > > > > >> > > > > > handled by
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol?
> > It
> > > > > > would be
> > > > > > > >> good to
> > > > > > > >> > > > > > have a
> > > > > > > >> > > > > > > > > section
> > > > > > > >> > > > > > > > > > > >> on
> > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From
> > a
> > > > > > protocol
> > > > > > > >> > > > > > perspective, the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so
> > it
> > > > > will
> > > > > > > >> treat it
> > > > > > > >> > > > like a
> > > > > > > >> > > > > > > > > topic with
> > > > > > > >> > > > > > > > > > > >> a
> > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue
> > is
> > > > that
> > > > > > the
> > > > > > > >> > > > fetch/commit
> > > > > > > >> > > > > > > > > offsets
> > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > > > > consumer
> > > > > > > >> would
> > > > > > > >> > > > reuse the
> > > > > > > >> > > > > > > > > offsets
> > > > > > > >> > > > > > > > > > > >> based
> > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > should
> > > > > > update
> > > > > > > >> those
> > > > > > > >> > > > APIs
> > > > > > > >> > > > > > as well
> > > > > > > >> > > > > > > > > in
> > > > > > > >> > > > > > > > > > > >> order
> > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That
> > > > would
> > > > > > > >> strengthen
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > semantics of
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The
> > KIP
> > > > > > mentions
> > > > > > > >> we will
> > > > > > > >> > > > > > write the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > > > > > >> implementation
> > > > > > > >> > > > > > detail, do
> > > > > > > >> > > > > > > > > we plan
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we
> > > > have a
> > > > > > > >> scalability
> > > > > > > >> > > > > > goal when
> > > > > > > >> > > > > > > > > it
> > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group
> > protocol
> > > > > can
> > > > > > > >> support?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't
> > have
> > > > > > numbers
> > > > > > > >> at the
> > > > > > > >> > > > > > moment. The
> > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members
> > per
> > > > > > group.
> > > > > > > >> We will
> > > > > > > >> > > > > > measure
> > > > > > > >> > > > > > > > > this when
> > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > implementation.
> > > > > Note
> > > > > > > >> that we
> > > > > > > >> > > > might
> > > > > > > >> > > > > > have
> > > > > > > >> > > > > > > > > other
> > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g.
> > > > offset
> > > > > > > >> commits).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > > > > > consider
> > > > > > > >> having
> > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > SubscribedTopicNames
> > > > > > > >> in
> > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we
> > have
> > > > to
> > > > > > > >> resolve the
> > > > > > > >> > > > > > regex on
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic
> > > > name?
> > > > > > The
> > > > > > > >> > > > difference
> > > > > > > >> > > > > > is that
> > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas
> > > > sending
> > > > > > the
> > > > > > > >> topic
> > > > > > > >> > > > names
> > > > > > > >> > > > > > is less
> > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > recreation is
> > > > > > easier
> > > > > > > >> to
> > > > > > > >> > > > handle
> > > > > > > >> > > > > > if we
> > > > > > > >> > > > > > > > > have
> > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea
> > was to
> > > > > > > >> consolidate the
> > > > > > > >> > > > > > metadata
> > > > > > > >> > > > > > > > > lookup
> > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I
> > do
> > > > > agree
> > > > > > > >> with your
> > > > > > > >> > > > > > point. As
> > > > > > > >> > > > > > > > > a
> > > > > > > >> > > > > > > > > > > >> second
> > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be
> > > > better
> > > > > > here
> > > > > > > >> for the
> > > > > > > >> > > > > > delete and
> > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose
> > that we
> > > > > may
> > > > > > > >> allow
> > > > > > > >> > > > users to
> > > > > > > >> > > > > > > > > subscribe
> > > > > > > >> > > > > > > > > > > >> with
> > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> > because
> > > > > that
> > > > > > is
> > > > > > > >> the only
> > > > > > > >> > > > > > way to be
> > > > > > > >> > > > > > > > > > > >> really
> > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul
> > 12,
> > > > > > 2022 at
> > > > > > > >> 1:38 PM
> > > > > > > >> > > > > > David
> > > > > > > >> > > > > > > > > Jacot <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > Justine,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for
> > > > your
> > > > > > > >> comments.
> > > > > > > >> > > > Please
> > > > > > > >> > > > > > find
> > > > > > > >> > > > > > > > > my
> > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes,
> > the new
> > > > > > > >> protocol
> > > > > > > >> > > > relies on
> > > > > > > >> > > > > > > > > topic IDs
> > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic
> > names
> > > > > based
> > > > > > in
> > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using
> > topic
> > > > > > names
> > > > > > > >> is the
> > > > > > > >> > > > > > right call
> > > > > > > >> > > > > > > > > here.
> > > > > > > >> > > > > > > > > > > >> I
> > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little
> > more.
> > > > > > > >> Obviously, the
> > > > > > > >> > > > > > KIP does
> > > > > > > >> > > > > > > > > not
> > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets
> > RPCs
> > > > to
> > > > > > use
> > > > > > > >> topic
> > > > > > > >> > > > IDs.
> > > > > > > >> > > > > > This
> > > > > > > >> > > > > > > > > may be
> > > > > > > >> > > > > > > > > > > >> > something
> > > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include
> > though
> > > > > as
> > > > > > it
> > > > > > > >> would
> > > > > > > >> > > > give
> > > > > > > >> > > > > > better
> > > > > > > >> > > > > > > > > > > >> overall
> > > > > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're
> > > > right.
> > > > > I
> > > > > > > >> think that
> > > > > > > >> > > > I
> > > > > > > >> > > > > > should
> > > > > > > >> > > > > > > > > not
> > > > > > > >> > > > > > > > > > > >> have
> > > > > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I
> > will
> > > > > remove
> > > > > > > >> it. We can
> > > > > > > >> > > > > > use an
> > > > > > > >> > > > > > > > > internal
> > > > > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > developing the
> > > > > > > >> feature.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both
> > cluster
> > > > > > types
> > > > > > > >> will be
> > > > > > > >> > > > > > > > > supported. The
> > > > > > > >> > > > > > > > > > > >> > change is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only
> > > > requirement
> > > > > > is
> > > > > > > >> that the
> > > > > > > >> > > > > > cluster
> > > > > > > >> > > > > > > > > uses
> > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > IDs.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon,
> > Jul
> > > > 11,
> > > > > > 2022
> > > > > > > >> at 9:53
> > > > > > > >> > > > PM
> > > > > > > >> > > > > > > > > Guozhang
> > > > > > > >> > > > > > > > > > > >> Wang <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi
> > Ismael,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks
> > for
> > > > the
> > > > > > > >> feedback.
> > > > > > > >> > > > Here
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > > some
> > > > > > > >> > > > > > > > > > > >> replies
> > > > > > > >> > > > > > > > > > > >> > > > > inlined
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat,
> > Jul
> > > > 9,
> > > > > > 2022
> > > > > > > >> at
> > > > > > > >> > > > 2:53 AM
> > > > > > > >> > > > > > > > > Ismael Juma
> > > > > > > >> > > > > > > > > > > >> <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > Thanks for
> > > > > the
> > > > > > > >> KIP. This
> > > > > > > >> > > > has
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> potential
> > > > > > > >> > > > > > > > > > > >> > to be a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > improvement. A
> > > > few
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > initial
> > > > > > > >> > > > questions/comments:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I
> > think
> > > > > > it's
> > > > > > > >> > > > premature to
> > > > > > > >> > > > > > talk
> > > > > > > >> > > > > > > > > about
> > > > > > > >> > > > > > > > > > > >> > target
> > > > > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation
> > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > removal of
> > > > > the
> > > > > > > >> existing
> > > > > > > >> > > > group
> > > > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> > Unlike
> > > > > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > > > > > protocol
> > > > > > > >> and hence
> > > > > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how
> > > > quickly
> > > > > > > >> applications
> > > > > > > >> > > > > > migrate
> > > > > > > >> > > > > > > > > to the
> > > > > > > >> > > > > > > > > > > >> new
> > > > > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I
> > agree
> > > > > > with
> > > > > > > >> you. I
> > > > > > > >> > > > think
> > > > > > > >> > > > > > we can
> > > > > > > >> > > > > > > > > > > >> remove
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > `Compatibility,
> > > > > > > >> > > > Deprecation,
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > Migration
> > > > > > > >> > > > > > > > > > > >> > Plan` and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we
> > will
> > > > > > decide
> > > > > > > >> in the
> > > > > > > >> > > > > > future
> > > > > > > >> > > > > > > > > about
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > we
> > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > protocol and
> > > > > > > >> behaviors.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2.
> > The KIP
> > > > > > says we
> > > > > > > >> > > > intend to
> > > > > > > >> > > > > > > > > release this
> > > > > > > >> > > > > > > > > > > >> > in 4.x,
> > > > > > > >> > > > > > > > > > > >> > > > > >> but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why.
> > If we
> > > > > > added
> > > > > > > >> that as
> > > > > > > >> > > > a
> > > > > > > >> > > > > > way to
> > > > > > > >> > > > > > > > > > > >> estimate
> > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > >> > > > > > > > > > > >> > > > > we'd
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > group
> > > > > > > >> protocol, I
> > > > > > > >> > > > also
> > > > > > > >> > > > > > suggest
> > > > > > > >> > > > > > > > > > > >> removing
> > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> > > > that's
> > > > > > not
> > > > > > > >> > > > specifically
> > > > > > > >> > > > > > > > > related to
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan,
> > but
> > > > it's
> > > > > > more
> > > > > > > >> for
> > > > > > > >> > > > client
> > > > > > > >> > > > > > > > > upgrades.
> > > > > > > >> > > > > > > > > > > >> I.e.
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > implementation
> > > > > > may
> > > > > > > >> be done
> > > > > > > >> > > > > > first,
> > > > > > > >> > > > > > > > > and then
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only
> > mark it
> > > > > as
> > > > > > > >> "released"
> > > > > > > >> > > > by
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > time
> > > > > > > >> > > > > > > > > > > >> clients
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that
> > time,
> > > > to
> > > > > > > >> enable the
> > > > > > > >> > > > > > feature the
> > > > > > > >> > > > > > > > > > > >> clients
> > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode
> > > > with
> > > > > a
> > > > > > > >> rolling
> > > > > > > >> > > > bounce
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > then set
> > > > > > > >> > > > > > > > > > > >> > the flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce,
> > and
> > > > > > hence
> > > > > > > >> we feel
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > > > > better to be
> > > > > > > >> > > > > > > > > > > >> > released
> > > > > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We
> > need
> > > > > to
> > > > > > > >> flesh out
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > details of
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > migration
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're
> > > > saying
> > > > > > we
> > > > > > > >> will
> > > > > > > >> > > > support
> > > > > > > >> > > > > > online
> > > > > > > >> > > > > > > > > > > >> > migrations. Is
> > > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > explain
> > > > this
> > > > > > in
> > > > > > > >> detail.
> > > > > > > >> > > > It
> > > > > > > >> > > > > > could
> > > > > > > >> > > > > > > > > also be
> > > > > > > >> > > > > > > > > > > >> > done as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > easier.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> > think
> > > > > > that's
> > > > > > > >> the
> > > > > > > >> > > > part we
> > > > > > > >> > > > > > can
> > > > > > > >> > > > > > > > > be more
> > > > > > > >> > > > > > > > > > > >> > concrete
> > > > > > > >> > > > > > > > > > > >> > > > > >> about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is
> > > > > related
> > > > > > to
> > > > > > > >> your
> > > > > > > >> > > > > > question 2)
> > > > > > > >> > > > > > > > > above).
> > > > > > > >> > > > > > > > > > > >> > We will
> > > > > > > >> > > > > > > > > > > >> > > > > >> work
> > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > explicit in
> > > > > > > >> parallel as we
> > > > > > > >> > > > > > solicit
> > > > > > > >> > > > > > > > > more
> > > > > > > >> > > > > > > > > > > >> > feedback.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I
> > am
> > > > > happy
> > > > > > > >> that we are
> > > > > > > >> > > > > > pushing
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it
> > > > seems
> > > > > > like
> > > > > > > >> there
> > > > > > > >> > > > > > could be
> > > > > > > >> > > > > > > > > some
> > > > > > > >> > > > > > > > > > > >> tricky
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > > > > > mechanism
> > > > > > > >> for
> > > > > > > >> > > > users to
> > > > > > > >> > > > > > > > > detect that
> > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > >> > > > > > > > > > > >> > > > > need
> > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > > > > > switching
> > > > > > > >> to the
> > > > > > > >> > > > new
> > > > > > > >> > > > > > > > > protocol?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> > think
> > > > we
> > > > > > need
> > > > > > > >> some
> > > > > > > >> > > > > > tooling for
> > > > > > > >> > > > > > > > > > > >> non-java
> > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > "dry-run"
> > > > the
> > > > > > > >> client before
> > > > > > > >> > > > > > > > > switching to
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific
> > > > idea
> > > > > on
> > > > > > > >> top of my
> > > > > > > >> > > > head
> > > > > > > >> > > > > > > > > though,
> > > > > > > >> > > > > > > > > > > >> maybe
> > > > > > > >> > > > > > > > > > > >> > others
> > > > > > > >> > > > > > > > > > > >> > > > > >> like
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in
> > > > here?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5.
> > Related
> > > > > to
> > > > > > the
> > > > > > > >> last
> > > > > > > >> > > > > > question,
> > > > > > > >> > > > > > > > > will the
> > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick
> > with
> > > > > the
> > > > > > > >> current
> > > > > > > >> > > > regex
> > > > > > > >> > > > > > > > > engine for
> > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it
> > may be
> > > > > > handy
> > > > > > > >> to keep
> > > > > > > >> > > > using
> > > > > > > >> > > > > > > > > client
> > > > > > > >> > > > > > > > > > > >> based
> > > > > > > >> > > > > > > > > > > >> > regex
> > > > > > > >> > > > > > > > > > > >> > > > > at
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > migrations
> > > > > > simple
> > > > > > > >> and
> > > > > > > >> > > > then
> > > > > > > >> > > > > > migrate
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > server based
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > Honestly I
> > > > > have
> > > > > > not
> > > > > > > >> thought
> > > > > > > >> > > > > > about
> > > > > > > >> > > > > > > > > that for
> > > > > > > >> > > > > > > > > > > >> > java
> > > > > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that.
> > What
> > > > > kind
> > > > > > of
> > > > > > > >> > > > > > compatibility
> > > > > > > >> > > > > > > > > issues do
> > > > > > > >> > > > > > > > > > > >> > you have
> > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6.
> > When we
> > > > > say
> > > > > > > >> that the
> > > > > > > >> > > > group
> > > > > > > >> > > > > > > > > coordinator
> > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > > > > > >> configurations and
> > > > > > > >> > > > that
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> > configurations
> > > > > > > >> > > > > > > > > > > >> > > > > will
> > > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > is
> > > > > > deleted.
> > > > > > > >> Will a
> > > > > > > >> > > > > > > > > transition to
> > > > > > > >> > > > > > > > > > > >> DEAD
> > > > > > > >> > > > > > > > > > > >> > > > > trigger
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > configurations?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes,
> > since
> > > > the
> > > > > > DEAD
> > > > > > > >> state
> > > > > > > >> > > > is an
> > > > > > > >> > > > > > > > > ending
> > > > > > > >> > > > > > > > > > > >> state
> > > > > > > >> > > > > > > > > > > >> > (we
> > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state
> > when
> > > > the
> > > > > > > >> group is
> > > > > > > >> > > > EMPTY
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > also all
> > > > > > > >> > > > > > > > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > its
> > > > > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's
> > > > transited
> > > > > > to
> > > > > > > >> DEAD this
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> never
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7.
> > Will
> > > > the
> > > > > > > >> choice to
> > > > > > > >> > > > store
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > configs
> > > > > > > >> > > > > > > > > > > >> in
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > harder to
> > > > > > list all
> > > > > > > >> > > > cluster
> > > > > > > >> > > > > > configs
> > > > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> their
> > > > > > > >> > > > > > > > > > > >> > > > > values?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a
> > > > good
> > > > > > > >> question,
> > > > > > > >> > > > and our
> > > > > > > >> > > > > > > > > thoughts
> > > > > > > >> > > > > > > > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > that
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > configurations"
> > > > > > are
> > > > > > > >> > > > overrides
> > > > > > > >> > > > > > of the
> > > > > > > >> > > > > > > > > > > >> > cluster-level
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > customized
> > > > per
> > > > > > > >> group so
> > > > > > > >> > > > when an
> > > > > > > >> > > > > > > > > admin list
> > > > > > > >> > > > > > > > > > > >> > cluster
> > > > > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list
> > just
> > > > the
> > > > > > > >> cluster-level
> > > > > > > >> > > > > > > > > "defaults", not
> > > > > > > >> > > > > > > > > > > >> > showing
> > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > customizations.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How
> > > > would
> > > > > > > >> someone
> > > > > > > >> > > > > > configure a
> > > > > > > >> > > > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers?
> > Have
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > > > > considered
> > > > > > > >> allowing
> > > > > > > >> > > > the
> > > > > > > >> > > > > > explicit
> > > > > > > >> > > > > > > > > > > >> > creation of
> > > > > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Alternatively,
> > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > configs
> > > > > could
> > > > > > be
> > > > > > > >> > > > decoupled
> > > > > > > >> > > > > > from
> > > > > > > >> > > > > > > > > the group
> > > > > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The
> > configs
> > > > > can
> > > > > > be
> > > > > > > >> created
> > > > > > > >> > > > > > before
> > > > > > > >> > > > > > > > > the group
> > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > >> > > > > > > > > > > >> > > > > as
> > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of
> > > > course,
> > > > > > this
> > > > > > > >> > > > requires
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> corresponding
> > > > > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > > > > > coordinator
> > > > > > > >> based on
> > > > > > > >> > > > the
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > id ---
> > > > > > > >> > > > > > > > > > > >> > the only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the
> > > > group
> > > > > > > >> itself is
> > > > > > > >> > > > gone
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > also check
> > > > > > > >> > > > > > > > > > > >> > if there
> > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > configuration
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > > > > > related to
> > > > > > > >> that
> > > > > > > >> > > > group
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > delete
> > > > > > > >> > > > > > > > > > > >> as
> > > > > > > >> > > > > > > > > > > >> > well.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > Admittedly
> > > > > this
> > > > > > > >> indeed
> > > > > > > >> > > > > > introduces an
> > > > > > > >> > > > > > > > > > > >> > asymmetry on
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > lifecycles
> > > > of
> > > > > > the
> > > > > > > >> config
> > > > > > > >> > > > > > entities,
> > > > > > > >> > > > > > > > > and we
> > > > > > > >> > > > > > > > > > > >> > would like
> > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > > > > > whether we
> > > > > > > >> should
> > > > > > > >> > > > aim
> > > > > > > >> > > > > > for
> > > > > > > >> > > > > > > > > symmetry
> > > > > > > >> > > > > > > > > > > >> > i.e.
> > > > > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs
> > and
> > > > > > hence
> > > > > > > >> not
> > > > > > > >> > > > delete
> > > > > > > >> > > > > > them at
> > > > > > > >> > > > > > > > > all
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> > > > > explicit
> > > > > > > >> deletion
> > > > > > > >> > > > > > operations
> > > > > > > >> > > > > > > > > by
> > > > > > > >> > > > > > > > > > > >> > themselves.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9.
> > Will
> > > > the
> > > > > > > >> > > > > > Consumer.subscribe
> > > > > > > >> > > > > > > > > method for
> > > > > > > >> > > > > > > > > > > >> > the Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > `java.util.regex.Pattern` of
> > > > > > > >> > > > > > do we
> > > > > > > >> > > > > > > > > have
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> > we
> > > > do
> > > > > > not
> > > > > > > >> need to
> > > > > > > >> > > > > > introduce
> > > > > > > >> > > > > > > > > an
> > > > > > > >> > > > > > > > > > > >> > overload, but
> > > > > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be
> > some
> > > > > > > >> compatibility
> > > > > > > >> > > > > > issues
> > > > > > > >> > > > > > > > > that we
> > > > > > > >> > > > > > > > > > > >> may
> > > > > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I
> > > > agree
> > > > > > with
> > > > > > > >> Justine
> > > > > > > >> > > > > > that we
> > > > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > switch to
> > > > > > > >> > > > > > IBP/metadata.version
> > > > > > > >> > > > > > > > > from the
> > > > > > > >> > > > > > > > > > > >> > feature
> > > > > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can
> > switch
> > > > > the
> > > > > > > >> default
> > > > > > > >> > > > for
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > feature
> > > > > > > >> > > > > > > > > > > >> flag
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > true
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > metadata.version
> > > > > > > >> once we
> > > > > > > >> > > > > > want to
> > > > > > > >> > > > > > > > > make it
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11.
> > Some of
> > > > > the
> > > > > > > >> protocol
> > > > > > > >> > > > APIs
> > > > > > > >> > > > > > don't
> > > > > > > >> > > > > > > > > mention
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> required
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good
> > to
> > > > add
> > > > > > that
> > > > > > > >> for
> > > > > > > >> > > > > > consistency.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we
> > can
> > > > > > > >> certainly do
> > > > > > > >> > > > that.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12.
> > It is
> > > > a
> > > > > > bit
> > > > > > > >> odd that
> > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > though it
> > > > > > seems
> > > > > > > >> to do
> > > > > > > >> > > > more
> > > > > > > >> > > > > > than
> > > > > > > >> > > > > > > > > reading.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had
> > that
> > > > > > thought
> > > > > > > >> myself
> > > > > > > >> > > > as
> > > > > > > >> > > > > > well,
> > > > > > > >> > > > > > > > > but in
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > end we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > > > > alternative:
> > > > > > > >> adding
> > > > > > > >> > > > > > Write
> > > > > > > >> > > > > > > > > Group
> > > > > > > >> > > > > > > > > > > >> seems
> > > > > > > >> > > > > > > > > > > >> > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not
> > have it
> > > > > > > >> elsewhere (we
> > > > > > > >> > > > only
> > > > > > > >> > > > > > have
> > > > > > > >> > > > > > > > > Read /
> > > > > > > >> > > > > > > > > > > >> > Delete
> > > > > > > >> > > > > > > > > > > >> > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far).
> > Would
> > > > > > like to
> > > > > > > >> hear
> > > > > > > >> > > > others
> > > > > > > >> > > > > > > > > thoughts.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13.
> > How is
> > > > > > topic
> > > > > > > >> > > > recreation
> > > > > > > >> > > > > > > > > handled by
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > protocol?
> > > > It
> > > > > > > >> would be
> > > > > > > >> > > > good
> > > > > > > >> > > > > > to have
> > > > > > > >> > > > > > > > > a
> > > > > > > >> > > > > > > > > > > >> > section on
> > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean
> > > > with
> > > > > > regex
> > > > > > > >> > > > > > subscription
> > > > > > > >> > > > > > > > > right? Yes
> > > > > > > >> > > > > > > > > > > >> > we can
> > > > > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that,
> > but
> > > > > > basically
> > > > > > > >> the
> > > > > > > >> > > > idea
> > > > > > > >> > > > > > is that
> > > > > > > >> > > > > > > > > > > >> consumer
> > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > >> > > > > > > > > > > >> > > > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > > > > > protocol as
> > > > > > > >> it's
> > > > > > > >> > > > > > handled all
> > > > > > > >> > > > > > > > > by the
> > > > > > > >> > > > > > > > > > > >> > brokers.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14.
> > The
> > > > KIP
> > > > > > > >> mentions we
> > > > > > > >> > > > will
> > > > > > > >> > > > > > write
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> new
> > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this
> > is an
> > > > > > > >> implementation
> > > > > > > >> > > > > > detail,
> > > > > > > >> > > > > > > > > do we
> > > > > > > >> > > > > > > > > > > >> > plan to
> > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for
> > it?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have
> > not
> > > > > > thought
> > > > > > > >> about
> > > > > > > >> > > > > > that. But
> > > > > > > >> > > > > > > > > I think
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > answer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15.
> > Do we
> > > > > > have a
> > > > > > > >> > > > scalability
> > > > > > > >> > > > > > goal
> > > > > > > >> > > > > > > > > when it
> > > > > > > >> > > > > > > > > > > >> > comes to
> > > > > > > >> > > > > > > > > > > >> > > > > >> how
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > > > > > protocol can
> > > > > > > >> > > > support?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> > > > > group,
> > > > > > I
> > > > > > > >> think we
> > > > > > > >> > > > > > should
> > > > > > > >> > > > > > > > > shoot for
> > > > > > > >> > > > > > > > > > > >> > 1000s of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > scalability
> > > > > > goals
> > > > > > > >> also
> > > > > > > >> > > > depend
> > > > > > > >> > > > > > on the
> > > > > > > >> > > > > > > > > offset
> > > > > > > >> > > > > > > > > > > >> > > > > management
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > capabilities
> > > > > of
> > > > > > the
> > > > > > > >> > > > coordinator
> > > > > > > >> > > > > > > > > which we
> > > > > > > >> > > > > > > > > > > >> did
> > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to
> > > > give a
> > > > > > > >> number that
> > > > > > > >> > > > > > applies
> > > > > > > >> > > > > > > > > > > >> > universally.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16.
> > Did we
> > > > > > > >> consider
> > > > > > > >> > > > having
> > > > > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> SubscribedTopicNames in
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > ConsumerGroupHeartbeatRequest?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since
> > we
> > > > > have
> > > > > > to
> > > > > > > >> resolve
> > > > > > > >> > > > the
> > > > > > > >> > > > > > regex
> > > > > > > >> > > > > > > > > on the
> > > > > > > >> > > > > > > > > > > >> > server,
> > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > topic
> > > > > > name?
> > > > > > > >> The
> > > > > > > >> > > > > > difference is
> > > > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > sending the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > whereas
> > > > > > sending
> > > > > > > >> the topic
> > > > > > > >> > > > > > names is
> > > > > > > >> > > > > > > > > less
> > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > recreation
> > > > > is
> > > > > > > >> easier to
> > > > > > > >> > > > > > handle if
> > > > > > > >> > > > > > > > > we have
> > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > >> > > > > > > > > > > >> > > > > ids.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main
> > > > > reason
> > > > > > to
> > > > > > > >> still
> > > > > > > >> > > > let
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > clients
> > > > > > > >> > > > > > > > > > > >> send
> > > > > > > >> > > > > > > > > > > >> > names
> > > > > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > reasoning of
> > > > > > names
> > > > > > > >> -> ids
> > > > > > > >> > > > on
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > broker /
> > > > > > > >> > > > > > > > > > > >> > admin
> > > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > although we
> > > > > > added
> > > > > > > >> topic id
> > > > > > > >> > > > in
> > > > > > > >> > > > > > > > > KIP-516, we
> > > > > > > >> > > > > > > > > > > >> > never
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > consumer/producers
> > > > > > > >> > > > leveraging
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > related
> > > > > > > >> > > > > > > > > > > >> > newer
> > > > > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just
> > set the
> > > > > > topic
> > > > > > > >> id as
> > > > > > > >> > > > empty
> > > > > > > >> > > > > > UUID.
> > > > > > > >> > > > > > > > > We
> > > > > > > >> > > > > > > > > > > >> want
> > > > > > > >> > > > > > > > > > > >> > to keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > consumer/producer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be
> > thin
> > > > and
> > > > > > only
> > > > > > > >> > > > delegate
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > reasoning
> > > > > > > >> > > > > > > > > > > >> on
> > > > > > > >> > > > > > > > > > > >> > broker
> > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > Thanks,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On
> > Wed,
> > > > Jul
> > > > > 6,
> > > > > > > >> 2022 at
> > > > > > > >> > > > 10:45
> > > > > > > >> > > > > > AM
> > > > > > > >> > > > > > > > > David
> > > > > > > >> > > > > > > > > > > >> Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi
> > all,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I
> > would
> > > > > > like to
> > > > > > > >> start a
> > > > > > > >> > > > > > > > > discussion
> > > > > > > >> > > > > > > > > > > >> thread
> > > > > > > >> > > > > > > > > > > >> > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > Generation
> > > > > > of
> > > > > > > >> the
> > > > > > > >> > > > Consumer
> > > > > > > >> > > > > > > > > Rebalance
> > > > > > > >> > > > > > > > > > > >> > Protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> With
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to
> > make
> > > > > the
> > > > > > > >> rebalance
> > > > > > > >> > > > > > protocol
> > > > > > > >> > > > > > > > > (for
> > > > > > > >> > > > > > > > > > > >> > consumers)
> > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > scalable,
> > > > > > > >> easier to
> > > > > > > >> > > > > > implement for
> > > > > > > >> > > > > > > > > > > >> > clients, and
> > > > > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > operators.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The
> > KIP
> > > > is
> > > > > > here:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > Please
> > > > > take
> > > > > > a
> > > > > > > >> look and
> > > > > > > >> > > > let
> > > > > > > >> > > > > > me
> > > > > > > >> > > > > > > > > know what
> > > > > > > >> > > > > > > > > > > >> > you
> > > > > > > >> > > > > > > > > > > >> > > > > think.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS:
> > I
> > > > will
> > > > > > be
> > > > > > > >> away from
> > > > > > > >> > > > > > July
> > > > > > > >> > > > > > > > > 18th to
> > > > > > > >> > > > > > > > > > > >> > August 8th.
> > > > > > > >> > > > > > > > > > > >> > > > > >> That
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of
> > time
> > > > to
> > > > > > read
> > > > > > > >> and
> > > > > > > >> > > > digest
> > > > > > > >> > > > > > this
> > > > > > > >> > > > > > > > > long
> > > > > > > >> > > > > > > > > > > >> KIP.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> --
> > > > > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > --
> > > > > > > >> > > > > -- Guozhang
> > > > > > > >> > > >
> > > > > > > >> > >
> > > > > > > >> > >
> > > > > > > >> > > --
> > > > > > > >> > > -- Guozhang
> > > > > > > >>
> > > > > > > >
> > > > > > > >
> > > > > > > > --
> > > > > > > > -- Guozhang
> > > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > --
> > > > > > > -- Guozhang
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > > -- Guozhang
> > > > >
> > > >
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Jason Gustafson <ja...@confluent.io.INVALID>]

Hey David,

I think I missed the fact that the RE2/J library is pure Java. I had
initially thought that we were using a native library. That does sound
quite a bit better since we can control the version and there is no
additional installation requirement for users. I guess if there are any big
changes to the library in the future, we can consider bumping the protocol
at the time we upgrade.

Best,
Jason

On Wed, Sep 28, 2022 at 12:36 PM Guozhang Wang <wa...@gmail.com> wrote:

> Hello Jun/David,
>
> Just to chime in here for Jun's comment #50, today Streams does not have a
> way to map topic names to topic IDs, and in its topology description topics
> are encoded with names.
>
> What I originally thought was that, on the consumer side, we'd need to
> maintain this mapping from topic names to topic IDs anyways, and Streams
> should leverage on the consumer's functionality to get the corresponding
> topic names for the given topic IDs. But when I read through the RPC again
> I realized it may not be the case, since:
>
> 1) in HB request, when clients send the subscription, they are still
> sending them in terms of topic names. That is, the consumer would not try
> to keep all the mappings from topic names to topic ids for the subscribed
> topics.
> 2) in HB responses, the assignment's topics are encoded as IDs already. And
> fetch requests can directly rely on that since they just need topic IDs and
> do not care what actual topic names are.
> 3) in prepare-assignment responses, we return subscriptions as topic IDs as
> well.
>
> In a word, the consumer does not try to maintain the mapping of topic names
> to IDs at all, and even if they do, they will not expose it in e.g.
> assignor APIs. But for client-side assignors like Streams, if they do not
> maintain the mapping from topic IDs to topic names, certain logic such as
> co-partitioning indeed cannot be executed.
>
> So I think we'd need to consider 1) use topic names in subscriptions in the
> prepare-assignment responses, 2) use topic Names in client-side
> PartitionAssignor#GroupMember#subscribedTopics and
> PartitionAssignor#TopicMetadata#topicName.
>
> On Wed, Sep 28, 2022 at 1:26 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Jun,
> >
> > Thanks for your comments.
> >
> > 48. Okay. There is a typo in the text:
> > > Note that the persisted target assignment of B already includes foo-2
> > here
> > but B does not get it yet because foo-2 is not free.
> >
> > It should be:
> > > Note that the persisted member assignment of B already includes foo-2
> > here
> > but B does not get it yet because foo-2 is not free.
> >
> > The examples are showing the persisted state in the __consumer_offsets
> > topics. When a member transitions to the next epoch, we directly
> > persist the whole target assignment in order to reduce the number of
> > writes and we manage dependencies in memory. So "B - epoch=2,
> > partitions=[foo-2]" is actually correct from a persisted record
> > perspective. However, the in-memory state would be different because
> > foo-2 is not yet available as you pointed out. The in-memory state
> > would like as follow:
> >
> > Group (epoch=2)
> > A
> > B
> > Target Assignment (epoch=2)
> > A - partitions=[foo-0, foo-1]
> > B - partitions=[foo-2]
> > Member Assignment
> > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending partitions=[]
> > B - epoch=2, partitions=[], pending partitions=[foo-2]
> >
> > I suppose that the confusion comes from this. I will go and convert
> > all the examples to their in-memory representation in order to reduce
> > the confusion.
> >
> > 50. Okay. I have to admit that I am not so familiar with the internals
> > of KStreams. If I understand correctly, you're saying that KStreams
> > may not have topic ids in its topology so topic names are missing here
> > to match the topics. I was hoping that KStreams would have topic ids
> > in the topology as well. If that is not the case, we can also add
> > topic names on our side. I will check this.
> >
> > 53. It seems that I was not clear. Sorry for that. I did not mean to
> > say that MemberAssignment.metadata is not used by the group
> > coordinator. Let me re-explain the flow.
> >
> > PartitionAssignor.assign() returns an Assignment. Assignment contains
> > MemberAssignments where each member has MemberAssignment.metadata. The
> > Assignment is installed by the client with the
> > ConsumerGroupInstallAssignment API and this overrides the current
> > target assignment in the coordinator. Then the group coordinator uses
> > it to reconcile members. Therefore the target assignment will
> > eventually become the current member assignment and it will be
> > delivered to the member, including the original
> > MemberAssignment.metadata defines by the assignor.
> >
> > I hope that this helps to clarify how MemberAssignment.metadata is
> > used afterwards.
> >
> > Best,
> > David
> >
> > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <ju...@confluent.io.invalid>
> wrote:
> > > Hi, David,
> > >
> > > Thanks for the reply. A few more follow up questions.
> > >
> > > 48. Hmm, so member assignment is what the client sees. Then, I am not
> > sure
> > > about the following. The text says B cannot get foo-2 yet. But in
> Member
> > > Assignment, B has foo-2.
> > >
> > > At this point B can transition to epoch 2 but cannot get foo-2 until A
> > > revokes it.
> > >
> > > Note that the persisted target assignment of B already includes foo-2
> > here
> > > but B does not get it yet because foo-2 is not free.
> > >
> > > Group (epoch=2)
> > > A
> > > B
> > > Target Assignment (epoch=2)
> > > A - partitions=[foo-0, foo-1]
> > > B - partitions=[foo-2]
> > > Member Assignment
> > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > B - epoch=2, partitions=[foo-2]
> > >
> > >
> > > 50. Hmm, I thought KStreams doesn't use range assignor all the time. It
> > > does the assignment based on the topology. For example, if you are
> > joining
> > > two topics A and B and deriving the join result in a 3rd topic C, which
> > > will be aggregated. You want to use range to co-locate the partitions
> > from
> > > topic A and B. However, for topic C, the partition could potentially be
> > > assigned in round robin. Without the topic name, KStreams won't know
> > which
> > > topics need to be co-located.
> > >
> > > 53. If MemberAssignment.metadata is not used by the group coordinator,
> > > where is it being used?
> > >
> > > Thanks,
> > >
> > > Jun
> > >
> > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > <ja...@confluent.io.invalid>
> > > wrote:
> > >
> > > > Hi David,
> > > >
> > > > Thanks for the response. Replied below:
> > > >
> > > > > I don't have a strong opinion on this one. However, it seems that
> > this
> > > > could be a pandora box if our subset turns out to not be sufficient
> > > > for some existing users. We will have to add more to it and the
> > > > compatibility story between versions could be a real nightmare. One
> > > > thing that we could do for sure is disabling all unicode groups in
> the
> > > > library. That seems possible and would already reduce the scope.
> > > >
> > > > Changes to the regex syntax seems not a major problem to me. We would
> > bump
> > > > the protocol version as we usually do. Actually that begs the same
> > question
> > > > about RE2. How stable is the syntax and does it differ depending on
> the
> > > > version of the library? In general, I do agree that we'd be better
> off
> > not
> > > > having to manage the syntax ourselves if we can convince ourselves
> that
> > > > compatibility won't be a problem and the client won't eventually need
> > to
> > > > have access to it.
> > > >
> > > > Thanks,
> > > > Jason
> > > >
> > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi Luke,
> > > > >
> > > > > Thanks for your comments.
> > > > >
> > > > > Your suggestion makes sense. Jun's made the same one. We are also
> > > > > discussing whether we should directly return the group state in the
> > > > > response and drop the prepare assignment api. It seems that the
> > > > > current consensus is to keep it though. If that stays like this, I
> > > > > will apply your suggestion.
> > > > >
> > > > > Thanks,
> > > > > David
> > > > >
> > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com>
> > wrote:
> > > > > >
> > > > > > Hi David,
> > > > > >
> > > > > > Thanks for your patience for answering all these questions!
> > > > > >
> > > > > > One more question from me:
> > > > > > 1. When the selected consumer is going to be in client-side
> > assigning,
> > > > we
> > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> > wondering if
> > > > we
> > > > > > could use a flag to indicate the assignment is required. The
> > rationale
> > > > is
> > > > > > when returning error, we lose a chance to catch up with the
> latest
> > > > target
> > > > > > assignment or revoke partitions. Since we expected the record
> > fetching
> > > > is
> > > > > > not blocked by the rebalance/assignment computing, we might want
> to
> > > > allow
> > > > > > the members to update the member assignment in each heartbeat.
> > > > > >
> > > > > > Luke
> > > > > >
> > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > <ja...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > > > Hey David,
> > > > > > >
> > > > > > > Thanks for the response.
> > > > > > >
> > > > > > > > Do you think that we could have a case where two members with
> > the
> > > > > same
> > > > > > > static id are active?
> > > > > > >
> > > > > > > I was just thinking about the handoff when we spin up a new
> node
> > > > > before the
> > > > > > > old one completes shutdown. It's probably ok to fence the old
> > member
> > > > > > > abruptly. I do recall we had a few issues with this handoff in
> > the
> > > > > past. It
> > > > > > > was due to the fact that the replaced node might have been the
> > > > leader.
> > > > > > > Since we've removed the leader notion, perhaps this kind of
> issue
> > > > won't
> > > > > > > resurface.
> > > > > > >
> > > > > > > In regard to the RE2 library. I understand the intent to pass
> > this
> > > > > through
> > > > > > > as a string without any validation so that we don't need the
> > > > > dependence on
> > > > > > > the client. I do wonder if that will be restrictive in
> practice.
> > For
> > > > > > > example, if a regex changes on the client side, would we need
> to
> > > > > validate
> > > > > > > against the current assignment to see which partitions to
> > continue
> > > > > > > fetching? Or would we instead invalidate the complete current
> > > > > assignment?
> > > > > > > Perhaps this is a rare case and we can do the latter. There
> > could be
> > > > > more
> > > > > > > situations though.
> > > > > > >
> > > > > > > As an alternative, I wonder if we could define a minimalistic
> > regex
> > > > > syntax
> > > > > > > (a subset of RE2 say) that could map more easily to regex
> support
> > > > that
> > > > > > > already exists in clients? For example, in this scenario, we
> are
> > > > > matching
> > > > > > > topics, so we only need to be able to represent topic
> characters
> > and
> > > > > not
> > > > > > > the whole smorgasbord of whitespace and unicode options. Would
> it
> > > > make
> > > > > > > sense to have a reduced syntax which includes only standard
> topic
> > > > > > > characters, wildcards, character classes, and repetitions?
> > > > > > >
> > > > > > > Best,
> > > > > > > Jason
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > wrote:
> > > > > > >
> > > > > > > > Hi David, thanks for all the detailed explanations. I think
> > they
> > > > all
> > > > > make
> > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > >
> > > > > > > > > I don't really see the benefits here because server side
> > > > assignors
> > > > > > > > don't have metadata at all. They only assign
> topic-partitions.
> > They
> > > > > > > > are not supposed to generate metadata nor to receive metadata
> > from
> > > > > the
> > > > > > > > members.
> > > > > > > >
> > > > > > > > Just to clarify I was asking about the `version` of the
> > assignor
> > > > > (i.e. up
> > > > > > > > to what version that the client would support), and I do
> agree
> > we
> > > > > would
> > > > > > > not
> > > > > > > > need metadata. What I have in mind is that, for some specific
> > > > > built-in
> > > > > > > > broker-assignors, e.g. rack-aware assignors, if it's possible
> > that
> > > > > in a
> > > > > > > > newer version we would have a hierarchical rack ID string
> > format,
> > > > > like
> > > > > > > > "tier1-tier2" etc, but if some client has not upgraded their
> > rack
> > > > ID
> > > > > > > > would still be in old format. In this case, the broker then
> > needs
> > > > to
> > > > > > > choose
> > > > > > > > the old versioned assignor. I'm probably making something up
> > here
> > > > for
> > > > > > > rack
> > > > > > > > aware assignors, but I'm wondering if in general such an
> > > > > "auto-downgrade"
> > > > > > > > behavior would be needed still for broker-side assignor, and
> > if yes
> > > > > would
> > > > > > > > "version" still be useful.
> > > > > > > >
> > > > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> > called
> > > > > once
> > > > > > > > when the member transitions to a new epoch. This one contains
> > the
> > > > > full
> > > > > > > > metadata provided by the client side assignor. Then,
> > `onAssignment`
> > > > > > > > can be called max N times where N is the number of partitions
> > > > pending
> > > > > > > > revocation by other members. Let me try to clarify this in
> the
> > KIP.
> > > > > > > >
> > > > > > > > Okay, my understanding is that the calling ordering of these
> > > > > callbacks
> > > > > > > > would be like the following:
> > > > > > > >
> > > > > > > > ----------------------------------------
> > > > > > > > onPartitionsRevoked();   // just once, since we do not really
> > need
> > > > > > > > to revoke incrementally.
> > > > > > > >
> > > > > > > > onAssignment();    // the first call, with epoch incremented
> > > > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > > > >
> > > > > > > > onAssignment();              // the first onAssignment would
> > bump
> > > > up
> > > > > the
> > > > > > > > epoch, and the metadata reflected.
> > > > > > > > onPartitionsAssigned();   // each time we get an additional
> > > > > assignment,
> > > > > > > we
> > > > > > > > call onAssignment and then paired with an
> onPartitionsAssigned
> > > > > > > > ...
> > > > > > > > onAssignment();
> > > > > > > > onPartitionsAssigned();   // on each of the onAssignment
> > calls, the
> > > > > > > encoded
> > > > > > > > metadata would not change, only the incrementally added
> > partitions
> > > > be
> > > > > > > > reflected
> > > > > > > >
> > > > > > > > Is that the case?
> > > > > > > >
> > > > > > > > I'm wondering if we would still call onAssignment just once,
> > that
> > > > > encodes
> > > > > > > > all the assignment for this rebalance, including all the
> > partitions
> > > > > that
> > > > > > > > should be assigned to the member but not yet assigned since
> > they
> > > > > have not
> > > > > > > > been revoked by others. In that case the call ordering would
> > be:
> > > > > > > >
> > > > > > > > ----------------------------------------
> > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > onAssignment();    // just once, with epoch incremented, and
> > > > metadata
> > > > > > > > encoded changed, the "assignment" field also reflect the
> final
> > > > target
> > > > > > > > assignment
> > > > > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > > > > incrementally
> > > > > > > > added partitions
> > > > > > > > ...
> > > > > > > > onPartitionsAssigned();
> > > > > > > >
> > > > > > > > The motivation from this is that, most users would only
> > implement
> > > > the
> > > > > > > > rebalance callback listeners and hence we'd definitely need
> to
> > make
> > > > > sure
> > > > > > > > the semantics of that does not change much, and the time
> > > > > > > > `onPartitionsAssigned` indicate the time when the partitions
> > are
> > > > > actually
> > > > > > > > assigned to it; while for assignors, the `onAssignment` is
> > used to
> > > > > > > indicate
> > > > > > > > what decision is made regarding for this member, i.e. when
> the
> > > > > partitions
> > > > > > > > are decided to be given to it, but not necessarily meant that
> > it
> > > > has
> > > > > been
> > > > > > > > given, since that time should be determined by the time of
> > > > > > > > `onPartitionsAssigned`. The benefits though, would be that
> > assignor
> > > > > > > > implementers would not need to reason which `onAssignment`
> > would be
> > > > > the
> > > > > > > > last one for this epoch.
> > > > > > > >
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > <djacot@confluent.io.invalid
> > > > > > > >
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi Guozhang,
> > > > > > > > >
> > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> > Error
> > > > > byte
> > > > > > > > > field,
> > > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> > not.
> > > > > And
> > > > > > > the
> > > > > > > > > > rationale seems to be assuming that we should always be
> > able to
> > > > > do
> > > > > > > the
> > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > Personally I
> > > > > > > > think
> > > > > > > > > > it's still potentially beneficial to add the Error field
> > even
> > > > for
> > > > > > > > > > broker-side assignors, e.g. for some edge cases where
> some
> > > > > subscribed
> > > > > > > > > > topics are not recognized with the current broker's
> > metadata.
> > > > > What do
> > > > > > > > you
> > > > > > > > > > think?
> > > > > > > > >
> > > > > > > > > Yeah, that seems reasonable. However, I wonder if it would
> be
> > > > > better
> > > > > > > > > to use an exception on the server side. This is what we
> > usually
> > > > do
> > > > > for
> > > > > > > > > server side plugins. On the client side, we use a field
> > because
> > > > the
> > > > > > > > > errors are not defined in advance.
> > > > > > > > >
> > > > > > > > > Your comment also makes me think about what we should do
> > when the
> > > > > > > > > server side assignor fails. I suppose that we have to keep
> > the
> > > > > current
> > > > > > > > > assignment until a new event occurs. For instance, in your
> > > > example,
> > > > > > > > > the coordinator would have to trigger a rebalance when
> > > > unrecognized
> > > > > > > > > topics are available. This would be part of the metadata
> > > > > monitoring.
> > > > > > > > >
> > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > fields
> > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > GroupMember. I
> > > > > > > > > agree
> > > > > > > > > > that broker-side assignor would not need reason/metadata
> > since
> > > > > they
> > > > > > > are
> > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > version?
> > > > > E.g.
> > > > > > > is
> > > > > > > > > it
> > > > > > > > > > possible that we evolve our broker-side built-in assignor
> > but
> > > > > the old
> > > > > > > > > > versioned clients would not be able to work with the new
> > > > > version, in
> > > > > > > > > which
> > > > > > > > > > case we need to let the broker being aware of this and
> > upgrade
> > > > > its
> > > > > > > > > behavior
> > > > > > > > > > to cooperate with the clients?
> > > > > > > > >
> > > > > > > > > I don't really see the benefits here because server side
> > > > assignors
> > > > > > > > > don't have metadata at all. They only assign
> > topic-partitions.
> > > > They
> > > > > > > > > are not supposed to generate metadata nor to receive
> metadata
> > > > from
> > > > > the
> > > > > > > > > members.
> > > > > > > > >
> > > > > > > > > > 3) Also related to 2) above, for the client-side
> > "GroupMember",
> > > > > > > instead
> > > > > > > > > of
> > > > > > > > > > including these three fields, what about just adding the
> > > > > "Metadata"
> > > > > > > > field
> > > > > > > > > > class which has these three fields? Also, there are two
> > > > > "Metadata"
> > > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > > reason/version/metadata, and the second is just the
> encoded
> > > > > metadata
> > > > > > > > > bytes.
> > > > > > > > > > I'm wondering what about just naming the first as
> > > > memberMetadata,
> > > > > > > which
> > > > > > > > > > then has a bytebuffer field Metadata, or instead naming
> the
> > > > > second
> > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > >
> > > > > > > > > That's a good point. Let me try to rationalize this
> interface
> > > > > based on
> > > > > > > > > your suggestions.
> > > > > > > > >
> > > > > > > > > Best,
> > > > > > > > > David
> > > > > > > > >
> > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > wangguoz@gmail.com>
> > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Hello David,
> > > > > > > > > >
> > > > > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > > > > >
> > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> > Error
> > > > > byte
> > > > > > > > > field,
> > > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> > not.
> > > > > And
> > > > > > > the
> > > > > > > > > > rationale seems to be assuming that we should always be
> > able to
> > > > > do
> > > > > > > the
> > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > Personally I
> > > > > > > > think
> > > > > > > > > > it's still potentially beneficial to add the Error field
> > even
> > > > for
> > > > > > > > > > broker-side assignors, e.g. for some edge cases where
> some
> > > > > subscribed
> > > > > > > > > > topics are not recognized with the current broker's
> > metadata.
> > > > > What do
> > > > > > > > you
> > > > > > > > > > think?
> > > > > > > > > >
> > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > fields
> > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > GroupMember. I
> > > > > > > > > agree
> > > > > > > > > > that broker-side assignor would not need reason/metadata
> > since
> > > > > they
> > > > > > > are
> > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > version?
> > > > > E.g.
> > > > > > > is
> > > > > > > > > it
> > > > > > > > > > possible that we evolve our broker-side built-in assignor
> > but
> > > > > the old
> > > > > > > > > > versioned clients would not be able to work with the new
> > > > > version, in
> > > > > > > > > which
> > > > > > > > > > case we need to let the broker being aware of this and
> > upgrade
> > > > > its
> > > > > > > > > behavior
> > > > > > > > > > to cooperate with the clients?
> > > > > > > > > >
> > > > > > > > > > 3) Also related to 2) above, for the client-side
> > "GroupMember",
> > > > > > > instead
> > > > > > > > > of
> > > > > > > > > > including these three fields, what about just adding the
> > > > > "Metadata"
> > > > > > > > field
> > > > > > > > > > class which has these three fields? Also, there are two
> > > > > "Metadata"
> > > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > > reason/version/metadata, and the second is just the
> encoded
> > > > > metadata
> > > > > > > > > bytes.
> > > > > > > > > > I'm wondering what about just naming the first as
> > > > memberMetadata,
> > > > > > > which
> > > > > > > > > > then has a bytebuffer field Metadata, or instead naming
> the
> > > > > second
> > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > >
> > > > > > > > > > Guozhang
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hello David,
> > > > > > > > > > >
> > > > > > > > > > > Thanks for bringing this question up. I think the main
> > > > > benefits as
> > > > > > > > you
> > > > > > > > > > > listed is 2) above if it stays; just to clarify, we
> would
> > > > only
> > > > > be
> > > > > > > > able
> > > > > > > > > to
> > > > > > > > > > > save one round trip if the rebalance is still triggered
> > by
> > > > the
> > > > > > > > broker;
> > > > > > > > > if
> > > > > > > > > > > the rebalance is triggered by the client then the
> > > > > num.round.trips
> > > > > > > are
> > > > > > > > > the
> > > > > > > > > > > same:
> > > > > > > > > > >
> > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > >
> > > > > > > > > > > T0: client decides to do a new assignment, suppose it
> has
> > > > > already
> > > > > > > > sent
> > > > > > > > > a
> > > > > > > > > > > HB and hence has to wait for it to return first since
> > only
> > > > one
> > > > > > > > request
> > > > > > > > > /
> > > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > request.
> > > > > > > > > > >
> > > > > > > > > > > In total, two round trips.
> > > > > > > > > > >
> > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > >
> > > > > > > > > > > T0: client decides to do a new assignment, suppose it
> has
> > > > > already
> > > > > > > > sent
> > > > > > > > > a
> > > > > > > > > > > HB and hence has to wait for it to return first since
> > only
> > > > one
> > > > > > > > request
> > > > > > > > > /
> > > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > > T1: client receives the HB response, and then sends the
> > new
> > > > HB
> > > > > > > > request
> > > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > > T2: the HB response with the optional member metadata
> > map is
> > > > > > > > returned.
> > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > request.
> > > > > > > > > > >
> > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > >
> > > > > > > > > > > -----------------------------
> > > > > > > > > > >
> > > > > > > > > > > So to complete the full picture here, we'd need to
> modify
> > > > both
> > > > > HB
> > > > > > > > > request
> > > > > > > > > > > and response so that the client can also indicate a new
> > > > > rebalance
> > > > > > > via
> > > > > > > > > the
> > > > > > > > > > > HB request as well, right?
> > > > > > > > > > >
> > > > > > > > > > > Assuming all above is true, I think it's okay to merge
> > the
> > > > > > > > > > > GroupPrepareAssignment into HB given that we can make
> the
> > > > > > > additional
> > > > > > > > > fields
> > > > > > > > > > > encoding the full member (subscription) metadata and
> > topic
> > > > > metadata
> > > > > > > > as
> > > > > > > > > > > optional fields.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Guozhang
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > >> Hi all,
> > > > > > > > > > >>
> > > > > > > > > > >> During an offline conversation, someone asked why we
> > need
> > > > the
> > > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that
> we
> > > > could
> > > > > > > > instead
> > > > > > > > > > >> provide the group state in the heartbeat response.
> This
> > has
> > > > a
> > > > > few
> > > > > > > > > > >> advantages: 1) it does not require using a special
> error
> > > > code
> > > > > to
> > > > > > > > > > >> signal that a new assignment is required as the signal
> > would
> > > > > be
> > > > > > > the
> > > > > > > > > > >> provided group state; 2) it removes one round trip
> when
> > a
> > > > > client
> > > > > > > > side
> > > > > > > > > > >> assignor is used. The downside is that it makes the
> > > > heartbeat
> > > > > > > > > > >> response's definition quite large. I recall that I
> went
> > with
> > > > > the
> > > > > > > > > > >> current approach due to this.
> > > > > > > > > > >>
> > > > > > > > > > >> Providing the group state in the heartbeat response is
> > > > > appealing.
> > > > > > > > What
> > > > > > > > > > >> do you guys think?
> > > > > > > > > > >>
> > > > > > > > > > >> Best,
> > > > > > > > > > >> David
> > > > > > > > > > >>
> > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > > djacot@confluent.io>
> > > > > > > > > wrote:
> > > > > > > > > > >> >
> > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > >> >
> > > > > > > > > > >> > 1. I have added a reference to the relevant chapter
> > > > instead
> > > > > of
> > > > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > > > >> >
> > > > > > > > > > >> > 2. The "Rebalance Triggers" section you are
> referring
> > to
> > > > is
> > > > > > > about
> > > > > > > > > when
> > > > > > > > > > >> > a rebalance should be triggered for the non-upgraded
> > > > members
> > > > > > > using
> > > > > > > > > the
> > > > > > > > > > >> > old protocol. The section mentions that a rebalance
> > must
> > > > be
> > > > > > > > > triggered
> > > > > > > > > > >> > when a new assignment is installed. This implies
> that
> > the
> > > > > group
> > > > > > > > > epoch
> > > > > > > > > > >> > was updated either by a native member or a
> > non-upgraded
> > > > > member.
> > > > > > > > For
> > > > > > > > > > >> > the latter, the JoinGroup request would be the
> > trigger. I
> > > > > have
> > > > > > > > > added a
> > > > > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > > > > Handling"
> > > > > > > > > section
> > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > >> >
> > > > > > > > > > >> > Thanks,
> > > > > > > > > > >> > David
> > > > > > > > > > >> >
> > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > > > wangguoz@gmail.com
> > > > > > > > >
> > > > > > > > > > >> wrote:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Hello David,
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Alright I think that's sufficient. Just to make
> that
> > > > > clear in
> > > > > > > > the
> > > > > > > > > doc,
> > > > > > > > > > >> > > could we update:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > 1) the heartbeat request handling section, stating
> > when
> > > > > > > > > coordinator
> > > > > > > > > > >> will
> > > > > > > > > > >> > > trigger rebalance based on the HB's member
> metadata
> > /
> > > > > reason?
> > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to include
> what
> > we
> > > > > > > described
> > > > > > > > > in
> > > > > > > > > > >> "Group
> > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > > > >> > >
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Guozhang
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > wrote:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > I thought that the assignor will always be
> > consulted
> > > > > when
> > > > > > > the
> > > > > > > > > next
> > > > > > > > > > >> > > > heartbeat request is constructed. In other
> words,
> > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be called for
> > every
> > > > > > > > heartbeat.
> > > > > > > > > > >> This
> > > > > > > > > > >> > > > gives the opportunity for the assignor to
> enforce
> > a
> > > > > > > rebalance
> > > > > > > > by
> > > > > > > > > > >> > > > setting the reason to a non-zero value or by
> > changing
> > > > > the
> > > > > > > > > bytes. Do
> > > > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > > > concerned by
> > > > > > > > the
> > > > > > > > > > >> delay?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Best,
> > > > > > > > > > >> > > > David
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > >> wrote:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > ```
> > > > > > > > > > >> > > > > In this case, a new assignment is triggered by
> > the
> > > > > client
> > > > > > > > side
> > > > > > > > > > >> > > > > assignor. When constructing the HB, the
> consumer
> > > > will
> > > > > > > always
> > > > > > > > > > >> consult
> > > > > > > > > > >> > > > > the client side assignor and propagate the
> > > > > information to
> > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > coordinator. In other words, we don't expect
> > users
> > > > to
> > > > > call
> > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > >> > > > > ```
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > > > > interface,
> > > > > > > we
> > > > > > > > > > >> actually do
> > > > > > > > > > >> > > > > not have a way yet to instruct how to
> construct
> > the
> > > > > next
> > > > > > > HB
> > > > > > > > > > >> request, e.g.
> > > > > > > > > > >> > > > > when the assignor wants to enforce a new
> > rebalance
> > > > > with a
> > > > > > > > new
> > > > > > > > > > >> assignment,
> > > > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > > > PartitionAssignor
> > > > > > > > > to
> > > > > > > > > > >> indicate
> > > > > > > > > > >> > > > > the next HB telling broker about so. WDYT
> about
> > > > adding
> > > > > > > such
> > > > > > > > an
> > > > > > > > > > >> API on the
> > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > I have updated the KIP to include your
> > feedback. I
> > > > > have
> > > > > > > > also
> > > > > > > > > > >> tried to
> > > > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > >> > > > > > David
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > > > > djacot@confluent.io
> > > > > > > > > > >> >
> > > > > > > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > > > > answering
> > > > > > > your
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > > > inline and I will update the KIP next
> week.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > benefits
> > > > > of
> > > > > > > the
> > > > > > > > > KIP
> > > > > > > > > > >> seem to
> > > > > > > > > > >> > > > be
> > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > > support of
> > > > > > > > > > >> wildcard. A few
> > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > I would also add that the KIP removes the
> > global
> > > > > sync
> > > > > > > > > barrier
> > > > > > > > > > >> in the
> > > > > > > > > > >> > > > > > > protocol which is essential to improve
> group
> > > > > stability
> > > > > > > > and
> > > > > > > > > > >> > > > > > > scalability, and the KIP also simplifies
> the
> > > > > client by
> > > > > > > > > moving
> > > > > > > > > > >> most of
> > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do
> we
> > plan
> > > > > to
> > > > > > > > > support
> > > > > > > > > > >> rolling
> > > > > > > > > > >> > > > > > changing
> > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > consumers?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Definitely. The group coordinator will use
> > the
> > > > > > > assignor
> > > > > > > > > used
> > > > > > > > > > >> by a
> > > > > > > > > > >> > > > > > > majority of the members. This allows the
> > group
> > > > to
> > > > > move
> > > > > > > > > from
> > > > > > > > > > >> one
> > > > > > > > > > >> > > > > > > assignor to another by a roll. This is
> > explained
> > > > > in
> > > > > > > the
> > > > > > > > > > >> Assignor
> > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > whether
> > > > > it's
> > > > > > > > > > >> required in
> > > > > > > > > > >> > > > every
> > > > > > > > > > >> > > > > > > > request or the scenarios when it needs
> to
> > be
> > > > > filled?
> > > > > > > > For
> > > > > > > > > > >> example,
> > > > > > > > > > >> > > > it's
> > > > > > > > > > >> > > > > > not
> > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs
> to
> > be
> > > > > filled.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > The client is expected to set those fields
> > in
> > > > > case of
> > > > > > > a
> > > > > > > > > > >> connection
> > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields
> have
> > > > > changed
> > > > > > > > since
> > > > > > > > > > >> the last
> > > > > > > > > > >> > > > > > > HB. The server populates those fields as
> > long as
> > > > > the
> > > > > > > > > member
> > > > > > > > > > >> is not
> > > > > > > > > > >> > > > > > > fully reconciled - the member should
> > acknowledge
> > > > > that
> > > > > > > it
> > > > > > > > > has
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > expected epoch and assignment. I will
> > clarify
> > > > > this in
> > > > > > > > the
> > > > > > > > > KIP.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol,
> the
> > rack
> > > > > > > > affinity
> > > > > > > > > > >> between the
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > > fetching,
> > > > > > > but
> > > > > > > > > not
> > > > > > > > > > >> during
> > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> > the
> > > > > > > > assignment
> > > > > > > > > is
> > > > > > > > > > >> made,
> > > > > > > > > > >> > > > there
> > > > > > > > > > >> > > > > > is
> > > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> > no
> > > > > replicas
> > > > > > > > of
> > > > > > > > > > >> assigned
> > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> > if we
> > > > > should
> > > > > > > > use
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > GroupMember.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > That's an interesting idea. I don't see
> any
> > > > issue
> > > > > with
> > > > > > > > > adding
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > rack
> > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > useful to
> > > > > know
> > > > > > > how
> > > > > > > > > busy
> > > > > > > > > > >> a group
> > > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > > model,
> > > > > it
> > > > > > > > seems
> > > > > > > > > > >> that we
> > > > > > > > > > >> > > > could
> > > > > > > > > > >> > > > > > add
> > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> > time
> > > > > the
> > > > > > > > event
> > > > > > > > > > >> loop is
> > > > > > > > > > >> > > > doing
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > That's a great idea. I will add it.
> Thanks.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> coordinator
> > > > > failover
> > > > > > > > > > >> handling? For
> > > > > > > > > > >> > > > > > example,
> > > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> > group
> > > > > with
> > > > > > > > the
> > > > > > > > > > >> wildcard
> > > > > > > > > > >> > > > > > > > subscription now has a new matching
> topic?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Sure. When the new group coordinator takes
> > over,
> > > > > it
> > > > > > > has
> > > > > > > > > to:
> > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > >> > > > > > > * Trigger a new assignment if a client
> side
> > > > > assignor
> > > > > > > is
> > > > > > > > > used.
> > > > > > > > > > >> We
> > > > > > > > > > >> > > > don't
> > > > > > > > > > >> > > > > > > store the information about the member
> > selected
> > > > > to run
> > > > > > > > the
> > > > > > > > > > >> assignment
> > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > > > wildcard
> > > > > > > > > > >> subscriptions, and
> > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could
> we
> > > > > document
> > > > > > > > what
> > > > > > > > > > >> the epoch
> > > > > > > > > > >> > > > > > field
> > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > one in
> > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Sure. I will clarify that but it is always
> > the
> > > > > latest
> > > > > > > > > group
> > > > > > > > > > >> epoch.
> > > > > > > > > > >> > > > > > > When the group state is updated, the group
> > epoch
> > > > > is
> > > > > > > > > bumped so
> > > > > > > > > > >> we use
> > > > > > > > > > >> > > > > > > that one for all the change records
> related
> > to
> > > > the
> > > > > > > > update.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > that
> > > > the
> > > > > > > > > following
> > > > > > > > > > >> > > > invariants
> > > > > > > > > > >> > > > > > are
> > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > possible
> > > > for
> > > > > a
> > > > > > > > member
> > > > > > > > > > >> not to
> > > > > > > > > > >> > > > get any
> > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > That's right. Here I meant that the
> members
> > > > > provided
> > > > > > > by
> > > > > > > > > the
> > > > > > > > > > >> assignor
> > > > > > > > > > >> > > > > > > in the assignment must exist in the group.
> > The
> > > > > > > assignor
> > > > > > > > > can
> > > > > > > > > > >> not make
> > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > member
> > > > > epoch
> > > > > > > > > equals to
> > > > > > > > > > >> 0":
> > > > > > > > > > >> > > > When
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > would be
> > > > > used?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > > > partitions and
> > > > > > > > > > >> rejoins when
> > > > > > > > > > >> > > > it
> > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In
> > this
> > > > > case,
> > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > coordinator will have removed the member
> > from
> > > > the
> > > > > > > group.
> > > > > > > > > The
> > > > > > > > > > >> member
> > > > > > > > > > >> > > > > > > can rejoin the group with the same member
> > id but
> > > > > with
> > > > > > > 0
> > > > > > > > as
> > > > > > > > > > >> epoch. Let
> > > > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > > ability
> > > > > to
> > > > > > > > > trigger a
> > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> > by
> > > > > updating
> > > > > > > > the
> > > > > > > > > > >> assignor
> > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > conflicting
> > > > > with
> > > > > > > the
> > > > > > > > > > >> deprecation
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > In this case, a new assignment is
> triggered
> > by
> > > > the
> > > > > > > > client
> > > > > > > > > side
> > > > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> > consumer
> > > > > will
> > > > > > > > > always
> > > > > > > > > > >> consult
> > > > > > > > > > >> > > > > > > the client side assignor and propagate the
> > > > > information
> > > > > > > > to
> > > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > coordinator. In other words, we don't
> expect
> > > > > users to
> > > > > > > > call
> > > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> > But
> > > > the
> > > > > > > > section
> > > > > > > > > > >> seems to
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> > 2, B
> > > > > > > > > immediately
> > > > > > > > > > >> gets into
> > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which
> seems
> > > > > > > incorrect.
> > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> > 3, C
> > > > > seems
> > > > > > > to
> > > > > > > > > get
> > > > > > > > > > >> into
> > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > still
> > > > > has A -
> > > > > > > > > > >> epoch=2,
> > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > > support
> > > > > the
> > > > > > > > > upgrade
> > > > > > > > > > >> from
> > > > > > > > > > >> > > > any
> > > > > > > > > > >> > > > > > old
> > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > We will support upgrading from the
> consumer
> > > > > protocol
> > > > > > > > > version
> > > > > > > > > > >> 3,
> > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > > > implemented
> > > > > yet
> > > > > > > so
> > > > > > > > > the
> > > > > > > > > > >> earliest
> > > > > > > > > > >> > > > > > > version is unknown at the moment. This is
> > > > > explained in
> > > > > > > > the
> > > > > > > > > > >> migration
> > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I
> will
> > > > > update the
> > > > > > > > KIP
> > > > > > > > > > >> based on
> > > > > > > > > > >> > > > it
> > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > >> > > > > > > David
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > benefits
> > > > > of
> > > > > > > the
> > > > > > > > > KIP
> > > > > > > > > > >> seem to
> > > > > > > > > > >> > > > be
> > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > > support of
> > > > > > > > > > >> wildcard. A few
> > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do
> we
> > plan
> > > > > to
> > > > > > > > > support
> > > > > > > > > > >> rolling
> > > > > > > > > > >> > > > > > changing
> > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > consumers?
> > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > whether
> > > > > it's
> > > > > > > > > > >> required in
> > > > > > > > > > >> > > > every
> > > > > > > > > > >> > > > > > > > request or the scenarios when it needs
> to
> > be
> > > > > filled?
> > > > > > > > For
> > > > > > > > > > >> example,
> > > > > > > > > > >> > > > it's
> > > > > > > > > > >> > > > > > not
> > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs
> to
> > be
> > > > > filled.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol,
> the
> > rack
> > > > > > > > affinity
> > > > > > > > > > >> between the
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > > fetching,
> > > > > > > but
> > > > > > > > > not
> > > > > > > > > > >> during
> > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> > the
> > > > > > > > assignment
> > > > > > > > > is
> > > > > > > > > > >> made,
> > > > > > > > > > >> > > > there
> > > > > > > > > > >> > > > > > is
> > > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> > no
> > > > > replicas
> > > > > > > > of
> > > > > > > > > > >> assigned
> > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> > if we
> > > > > should
> > > > > > > > use
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > GroupMember.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > useful to
> > > > > know
> > > > > > > how
> > > > > > > > > busy
> > > > > > > > > > >> a group
> > > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > > model,
> > > > > it
> > > > > > > > seems
> > > > > > > > > > >> that we
> > > > > > > > > > >> > > > could
> > > > > > > > > > >> > > > > > add
> > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> > time
> > > > > the
> > > > > > > > event
> > > > > > > > > > >> loop is
> > > > > > > > > > >> > > > doing
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> coordinator
> > > > > failover
> > > > > > > > > > >> handling? For
> > > > > > > > > > >> > > > > > example,
> > > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> > group
> > > > > with
> > > > > > > > the
> > > > > > > > > > >> wildcard
> > > > > > > > > > >> > > > > > > > subscription now has a new matching
> topic?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could
> we
> > > > > document
> > > > > > > > what
> > > > > > > > > > >> the epoch
> > > > > > > > > > >> > > > > > field
> > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > one in
> > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > that
> > > > the
> > > > > > > > > following
> > > > > > > > > > >> > > > invariants
> > > > > > > > > > >> > > > > > are
> > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > possible
> > > > for
> > > > > a
> > > > > > > > member
> > > > > > > > > > >> not to
> > > > > > > > > > >> > > > get any
> > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > member
> > > > > epoch
> > > > > > > > > equals to
> > > > > > > > > > >> 0":
> > > > > > > > > > >> > > > When
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > would be
> > > > > used?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > > ability
> > > > > to
> > > > > > > > > trigger a
> > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> > by
> > > > > updating
> > > > > > > > the
> > > > > > > > > > >> assignor
> > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > conflicting
> > > > > with
> > > > > > > the
> > > > > > > > > > >> deprecation
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> > But
> > > > the
> > > > > > > > section
> > > > > > > > > > >> seems to
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> > 2, B
> > > > > > > > > immediately
> > > > > > > > > > >> gets into
> > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which
> seems
> > > > > > > incorrect.
> > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> > 3, C
> > > > > seems
> > > > > > > to
> > > > > > > > > get
> > > > > > > > > > >> into
> > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > still
> > > > > has A -
> > > > > > > > > > >> epoch=2,
> > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > > support
> > > > > the
> > > > > > > > > upgrade
> > > > > > > > > > >> from
> > > > > > > > > > >> > > > any
> > > > > > > > > > >> > > > > > old
> > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David
> > Jacot
> > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > re-implement the
> > > > > > > > > coordinator
> > > > > > > > > > >> in
> > > > > > > > > > >> > > > Java. I
> > > > > > > > > > >> > > > > > > > > discussed this offline with a few
> folks
> > and
> > > > > folks
> > > > > > > > are
> > > > > > > > > > >> concerned
> > > > > > > > > > >> > > > that
> > > > > > > > > > >> > > > > > > > > we could introduce many regressions in
> > the
> > > > old
> > > > > > > > > protocol
> > > > > > > > > > >> if we do
> > > > > > > > > > >> > > > so.
> > > > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > > > statement
> > > > > > > from
> > > > > > > > > the
> > > > > > > > > > >> KIP. It
> > > > > > > > > > >> > > > is an
> > > > > > > > > > >> > > > > > > > > implementation detail after all so it
> > does
> > > > not
> > > > > > > have
> > > > > > > > > to be
> > > > > > > > > > >> > > > decided at
> > > > > > > > > > >> > > > > > > > > this stage. We will likely start by
> > trying
> > > > to
> > > > > > > > > refactor the
> > > > > > > > > > >> > > > current
> > > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David
> > Jacot
> > > > <
> > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine
> are:
> > > > > "Empty,
> > > > > > > > > assigning,
> > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> > States
> > > > > > > > section,
> > > > > > > > > > >> right?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > This sentence does not refer to
> those
> > > > group
> > > > > > > states
> > > > > > > > > but
> > > > > > > > > > >> rather
> > > > > > > > > > >> > > > to a
> > > > > > > > > > >> > > > > > > > > > state machine replication (SMR).
> This
> > > > > refers to
> > > > > > > > the
> > > > > > > > > > >> entire
> > > > > > > > > > >> > > > state of
> > > > > > > > > > >> > > > > > > > > > group coordinator which is
> replicated
> > via
> > > > > the
> > > > > > > log
> > > > > > > > > > >> layer. I will
> > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > > machine
> > > > > is
> > > > > > > > > modelled
> > > > > > > > > > >> as an
> > > > > > > > > > >> > > > event
> > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > The idea is to follow a model
> similar
> > to
> > > > > the new
> > > > > > > > > quorum
> > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > >> > > > > > > > > > will have N threads to process
> events.
> > > > Each
> > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > >> > > > > > > > > > partition is assigned to a unique
> > thread
> > > > > and all
> > > > > > > > the
> > > > > > > > > > >> events
> > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> > processed
> > > > by
> > > > > this
> > > > > > > > > > >> thread. This
> > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > > > > simulation
> > > > > > > > > testing
> > > > > > > > > > >> for the
> > > > > > > > > > >> > > > > > group
> > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> machine
> > per
> > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> > group"
> > > > > owned
> > > > > > > > by
> > > > > > > > > a
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> > owns 2
> > > > > > > > consumer
> > > > > > > > > > >> groups, and
> > > > > > > > > > >> > > > > > both
> > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > have 1
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > >> for it,
> > > > > > > > > > >> > > > or
> > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> > there, I
> > > > > > > think.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > "*group.coordinator.threads"
> > > > > > > *is
> > > > > > > > > the
> > > > > > > > > > >> number
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > threads
> > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > > wondering
> > > > > > > if
> > > > > > > > > the
> > > > > > > > > > >> purpose
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > > consumer
> > > > > > > group
> > > > > > > > > (or
> > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > computation,
> > > > > why
> > > > > > > > > should we
> > > > > > > > > > >> need
> > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an
> > ability to
> > > > > shard
> > > > > > > > the
> > > > > > > > > > >> > > > processing as
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> > why
> > > > > does
> > > > > > > the
> > > > > > > > > > >> default
> > > > > > > > > > >> > > > session
> > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > max(60s)? I
> > > > > > > thought
> > > > > > > > > the
> > > > > > > > > > >> min/max
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > bound
> > > > of
> > > > > it,
> > > > > > > > no?
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> > int
> > > > > 30s The
> > > > > > > > > > >> timeout to
> > > > > > > > > > >> > > > detect
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > group
> > > > > > > protocol.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.min.session.timeout.ms
> > > > > int 45s
> > > > > > > > The
> > > > > > > > > > >> minimum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.max.session.timeout.ms
> > > > > int 60s
> > > > > > > > The
> > > > > > > > > > >> maximum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The
> default
> > > > > session
> > > > > > > > > timeout
> > > > > > > > > > >> should
> > > > > > > > > > >> > > > be 45s
> > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > assignor
> > > > are
> > > > > > > > [range,
> > > > > > > > > > >> uniform],
> > > > > > > > > > >> > > > > > which means
> > > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> > I'd
> > > > > like to
> > > > > > > > > know
> > > > > > > > > > >> why not
> > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > > uniform
> > > > > > > > assignor
> > > > > > > > > > >> (former
> > > > > > > > > > >> > > > sticky
> > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > >> > > > > > > > > > > for better evenly distribution.
> Any
> > > > other
> > > > > > > reason
> > > > > > > > > we
> > > > > > > > > > >> choose
> > > > > > > > > > >> > > > range
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List
> range,
> > > > > uniform
> > > > > > > The
> > > > > > > > > > >> server side
> > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > The order on the server side has no
> > > > > influence
> > > > > > > > > because
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > client
> > > > > > > > > > >> > > > > > must
> > > > > > > > > > >> > > > > > > > > > chose the selector that he wants to
> > use.
> > > > > There
> > > > > > > is
> > > > > > > > no
> > > > > > > > > > >> default
> > > > > > > > > > >> > > > in the
> > > > > > > > > > >> > > > > > > > > > current proposal. If the assignor is
> > not
> > > > > > > specified
> > > > > > > > > by
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > client,
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > request is rejected. The default
> > client
> > > > > value
> > > > > > > for
> > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> > > > though.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > Thanks for your very good comments,
> > Luke.
> > > > I
> > > > > hope
> > > > > > > > > that my
> > > > > > > > > > >> > > > answers
> > > > > > > > > > >> > > > > > help
> > > > > > > > > > >> > > > > > > > > > to clarify things. I will update the
> > KIP
> > > > as
> > > > > well
> > > > > > > > > based
> > > > > > > > > > >> on your
> > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke
> > Chen
> > > > <
> > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section,
> you
> > > > > > > mentioned:
> > > > > > > > > > >> > > > > > > > > > > > The new group coordinator will
> > have a
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > >> per
> > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions,
> > where
> > > > > each
> > > > > > > > state
> > > > > > > > > > >> machine is
> > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > >> > > > > > > > > > > event loop. Those state machines
> > will be
> > > > > > > > executed
> > > > > > > > > in
> > > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads*
> threads.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine
> are:
> > > > > "Empty,
> > > > > > > > > assigning,
> > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> > States
> > > > > > > > section,
> > > > > > > > > > >> right?
> > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > > machine
> > > > > is
> > > > > > > > > modelled
> > > > > > > > > > >> as an
> > > > > > > > > > >> > > > event
> > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> machine
> > per
> > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> > group"
> > > > > owned
> > > > > > > > by
> > > > > > > > > a
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> > owns 2
> > > > > > > > consumer
> > > > > > > > > > >> groups, and
> > > > > > > > > > >> > > > > > both
> > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > have 1
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > >> for it,
> > > > > > > > > > >> > > > or
> > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > "*group.coordinator.threads"
> > > > > > > *is
> > > > > > > > > the
> > > > > > > > > > >> number
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > threads
> > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > > wondering
> > > > > > > if
> > > > > > > > > the
> > > > > > > > > > >> purpose
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > > consumer
> > > > > > > group
> > > > > > > > > (or
> > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > computation,
> > > > > why
> > > > > > > > > should we
> > > > > > > > > > >> need
> > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2. For the default value in the
> new
> > > > > configs:
> > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> > why
> > > > > does
> > > > > > > the
> > > > > > > > > > >> default
> > > > > > > > > > >> > > > session
> > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > max(60s)? I
> > > > > > > thought
> > > > > > > > > the
> > > > > > > > > > >> min/max
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > bound
> > > > of
> > > > > it,
> > > > > > > > no?
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> > int
> > > > > 30s The
> > > > > > > > > > >> timeout to
> > > > > > > > > > >> > > > detect
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > group
> > > > > > > protocol.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.min.session.timeout.ms
> > > > > int 45s
> > > > > > > > The
> > > > > > > > > > >> minimum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.max.session.timeout.ms
> > > > > int 60s
> > > > > > > > The
> > > > > > > > > > >> maximum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > assignor
> > > > are
> > > > > > > > [range,
> > > > > > > > > > >> uniform],
> > > > > > > > > > >> > > > > > which means
> > > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> > I'd
> > > > > like to
> > > > > > > > > know
> > > > > > > > > > >> why not
> > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > > uniform
> > > > > > > > assignor
> > > > > > > > > > >> (former
> > > > > > > > > > >> > > > sticky
> > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > >> > > > > > > > > > > for better evenly distribution.
> Any
> > > > other
> > > > > > > reason
> > > > > > > > > we
> > > > > > > > > > >> choose
> > > > > > > > > > >> > > > range
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List
> range,
> > > > > uniform
> > > > > > > The
> > > > > > > > > > >> server side
> > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM
> Luke
> > > > Chen
> > > > > <
> > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> > > > Connect
> > > > > > > > > > >> integrating with
> > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > >> > > > > > > > > > > > think we should have a separate
> > > > > discussion
> > > > > > > > > thread
> > > > > > > > > > >> for the
> > > > > > > > > > >> > > > Kafka
> > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect
> > With
> > > > New
> > > > > > > > Consumer
> > > > > > > > > > >> Rebalance
> > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > >> > > > > > > > > > > > and let this discussion thread
> > focus
> > > > on
> > > > > > > > consumer
> > > > > > > > > > >> rebalance
> > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > >
> > > > > > > > > > >>
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM
> > Sagar
> > > > <
> > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for
> the
> > > > > feedback.
> > > > > > > > > Looks
> > > > > > > > > > >> like
> > > > > > > > > > >> > > > there's
> > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > >> > > > > > > > > > > >> using separate APIs for
> Connect.
> > I
> > > > > would
> > > > > > > > > revisit
> > > > > > > > > > >> the doc
> > > > > > > > > > >> > > > and
> > > > > > > > > > >> > > > > > see
> > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM
> > David
> > > > > Jacot
> > > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and
> the
> > > > > document.
> > > > > > > > > That's
> > > > > > > > > > >> really
> > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that
> > both
> > > > > Connect
> > > > > > > > > and the
> > > > > > > > > > >> > > > Consumer
> > > > > > > > > > >> > > > > > could
> > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > >> > > > > > > > > > > >> > the same underlying "engine".
> > The
> > > > > main
> > > > > > > > > > >> difference is
> > > > > > > > > > >> > > > that
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> > members
> > > > > > > whereas
> > > > > > > > > > >> Connect
> > > > > > > > > > >> > > > assigns
> > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to
> move
> > > > > forward:
> > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed
> > APIs
> > > > to
> > > > > > > > support
> > > > > > > > > > >> different
> > > > > > > > > > >> > > > > > resource
> > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks,
> > etc.); or
> > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs
> > for
> > > > > Connect.
> > > > > > > > The
> > > > > > > > > > >> dedicated
> > > > > > > > > > >> > > > APIs
> > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > > > different on
> > > > > > > > the
> > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> > engine
> > > > > on the
> > > > > > > > > > >> coordinator
> > > > > > > > > > >> > > > side.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> > > > because
> > > > > I am
> > > > > > > > > not a
> > > > > > > > > > >> fan of
> > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> > purposes.
> > > > > That
> > > > > > > > being
> > > > > > > > > > >> said, I am
> > > > > > > > > > >> > > > not
> > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant
> > way to
> > > > > do
> > > > > > > it.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > I think that we can continue
> to
> > > > > discuss
> > > > > > > it
> > > > > > > > > here
> > > > > > > > > > >> for now
> > > > > > > > > > >> > > > in
> > > > > > > > > > >> > > > > > order
> > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> > compatible
> > > > > with
> > > > > > > > what
> > > > > > > > > we
> > > > > > > > > > >> will do
> > > > > > > > > > >> > > > for
> > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41
> PM
> > > > David
> > > > > > > Jacot
> > > > > > > > <
> > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I
> > will
> > > > go
> > > > > > > > through
> > > > > > > > > and
> > > > > > > > > > >> address
> > > > > > > > > > >> > > > > > your
> > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks
> > for
> > > > your
> > > > > > > > > feedback.
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at
> 10:05
> > PM
> > > > > Gregory
> > > > > > > > > Harris
> > > > > > > > > > >> <
> > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > > > > wonderful to
> > > > > > > > see
> > > > > > > > > > >> > > > cooperative
> > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> > Alternatives'
> > > > > > > section
> > > > > > > > > > >> describes how
> > > > > > > > > > >> > > > > > member
> > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step with the
> > group
> > > > > epoch
> > > > > > > > and
> > > > > > > > > > >> assignment
> > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea
> > for
> > > > the
> > > > > > > > reasons
> > > > > > > > > > >> described
> > > > > > > > > > >> > > > in
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > > > > assigning
> > > > > > > > > > >> partitions to a
> > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each
> incremental
> > > > > > > assignment
> > > > > > > > > use?
> > > > > > > > > > >> Are
> > > > > > > > > > >> > > > member
> > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > > > > correspond
> > > > > > > to
> > > > > > > > > > >> multiple
> > > > > > > > > > >> > > > > > different
> > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's
> > 'Reason'
> > > > > field
> > > > > > > > > opaque
> > > > > > > > > > >> to the
> > > > > > > > > > >> > > > group
> > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> > client-side
> > > > > > > assignor
> > > > > > > > > > >> > > > implementations
> > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is
> > its
> > > > > common
> > > > > > > > > meaning
> > > > > > > > > > >> agreed
> > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct
> > Reason
> > > > > field
> > > > > > > over
> > > > > > > > > > >> including
> > > > > > > > > > >> > > > such
> > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> > included in
> > > > > the
> > > > > > > > KIP:
> > > > > > > > > > >> "Thanks to
> > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > > > entirely
> > > > > > > driven
> > > > > > > > > by
> > > > > > > > > > >> the group
> > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > > > responsible for
> > > > > > > > > > >> maintaining
> > > > > > > > > > >> > > > any
> > > > > > > > > > >> > > > > > state
> > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions."
> Does
> > this
> > > > > mean
> > > > > > > > > that the
> > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > > > non-Metadata
> > > > > > > > state
> > > > > > > > > > >> (such as
> > > > > > > > > > >> > > > > > partition
> > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics,
> config
> > > > > topics,
> > > > > > > > etc),
> > > > > > > > > or
> > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > additional
> > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > > > separate
> > > > > > > > classes
> > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > >> > > > > >
> > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > >> > > >
> > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > >> > > > > > that
> > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is
> > it
> > > > > possible
> > > > > > > > for
> > > > > > > > > > >> these two
> > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve
> > to
> > > > > promote
> > > > > > > > > feature
> > > > > > > > > > >> parity
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors,
> and
> > > > would
> > > > > also
> > > > > > > > > > >> facilitate
> > > > > > > > > > >> > > > > > operational
> > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > > > example,
> > > > > if a
> > > > > > > > > > >> server-side
> > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a
> patch,
> > > > > deploying
> > > > > > > > the
> > > > > > > > > > >> patched
> > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > > > > > > > >> > > > and switching one
> consumer
> > > > group
> > > > > to a
> > > > > > > > > > >> client-side
> > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> > patching
> > > > all
> > > > > of
> > > > > > > the
> > > > > > > > > > >> brokers.
> > > > > > > > > > >> > > > With
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > > > > non-trivial
> > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two
> > APIs
> > > > > have
> > > > > > > > > diverged
> > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > reimplementation
> > > > > > > would
> > > > > > > > > not be
> > > > > > > > > > >> > > > possible.
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at
> > 8:39 AM
> > > > > Sagar
> > > > > > > <
> > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence
> > page
> > > > to
> > > > > > > > discuss
> > > > > > > > > how
> > > > > > > > > > >> Connect
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> > rebalance
> > > > > > > protocol.
> > > > > > > > > > >> Here's the
> > > > > > > > > > >> > > > page:
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > >
> > > > > > > > > > >>
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty
> longish
> > and
> > > > I
> > > > > have
> > > > > > > > > tried
> > > > > > > > > > >> to keep
> > > > > > > > > > >> > > > a
> > > > > > > > > > >> > > > > > format
> > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know
> > what you
> > > > > > > think.
> > > > > > > > > Also,
> > > > > > > > > > >> do you
> > > > > > > > > > >> > > > > > think this
> > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > > > discussion
> > > > > > > thread
> > > > > > > > > or
> > > > > > > > > > >> is this
> > > > > > > > > > >> > > > one
> > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at
> > 7:37
> > > > AM
> > > > > > > > Sagar <
> > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for
> > the
> > > > > doc on
> > > > > > > > > Kafka
> > > > > > > > > > >> Streams.
> > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up
> > with
> > > > > such a
> > > > > > > > > > >> document.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022
> at
> > > > 4:47
> > > > > AM
> > > > > > > > > Guozhang
> > > > > > > > > > >> Wang <
> > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if
> > you
> > > > > could
> > > > > > > > come
> > > > > > > > > > >> back with
> > > > > > > > > > >> > > > some
> > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the
> Connect
> > side
> > > > > > > > > integration
> > > > > > > > > > >> with
> > > > > > > > > > >> > > > the new
> > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging
> on
> > the
> > > > > new
> > > > > > > > > "protocol
> > > > > > > > > > >> type"
> > > > > > > > > > >> > > > we
> > > > > > > > > > >> > > > > > did not
> > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > > > implementations.
> > > > > > > > > For
> > > > > > > > > > >> Streams
> > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > >
> > > > > > > > > > >>
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > > >
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your
> > analysis
> > > > > on
> > > > > > > > > Connect.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022
> > at
> > > > > 10:48 PM
> > > > > > > > > Sagar <
> > > > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > > > > response.
> > > > > > > The
> > > > > > > > > > >> reason I
> > > > > > > > > > >> > > > thought
> > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is
> > that
> > > > > even
> > > > > > > > > today the
> > > > > > > > > > >> > > > connect
> > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > > > AbstractCoordinator to
> > > > > > > > > > >> empower
> > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors.
> > The
> > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> > metadata()
> > > > > method
> > > > > > > > by
> > > > > > > > > > >> plumbing
> > > > > > > > > > >> > > > into
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > JoinGroupRequestProtocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the
> changes
> > to
> > > > > support
> > > > > > > > > > >> connect would
> > > > > > > > > > >> > > > be
> > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in
> > streams
> > > > > mainly
> > > > > > > > > because
> > > > > > > > > > >> of the
> > > > > > > > > > >> > > > > > Client
> > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At
> an
> > > > > > > > implementation
> > > > > > > > > > >> level, we
> > > > > > > > > > >> > > > > > might
> > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get
> onto
> > this
> > > > > new
> > > > > > > > > > >> assignment
> > > > > > > > > > >> > > > protocol
> > > > > > > > > > >> > > > > > like
> > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > > request/response and
> > > > > > > > > > >> SyncGroup and
> > > > > > > > > > >> > > > > > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on
> > similar
> > > > > lines
> > > > > > > > to
> > > > > > > > > > >> streams (or
> > > > > > > > > > >> > > > > > there
> > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> > perform a
> > > > > > > detailed
> > > > > > > > > > >> analysis of
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > same
> > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate
> discussion
> > > > > thread for
> > > > > > > > > that
> > > > > > > > > > >> as that
> > > > > > > > > > >> > > > would
> > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me
> know
> > if
> > > > > that
> > > > > > > > sounds
> > > > > > > > > > >> good to
> > > > > > > > > > >> > > > you.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15,
> > 2022 at
> > > > > 5:47
> > > > > > > PM
> > > > > > > > > David
> > > > > > > > > > >> Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > > > > comments.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That
> > refers to
> > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is
> to
> > > > > > > transition C
> > > > > > > > > from
> > > > > > > > > > >> his
> > > > > > > > > > >> > > > current
> > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target
> assignment
> > when
> > > > > he
> > > > > > > can
> > > > > > > > > move
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > >> > > > > > When
> > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> assignment
> > is
> > > > > updated
> > > > > > > > and
> > > > > > > > > > >> persisted
> > > > > > > > > > >> > > > > > with all
> > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even
> if
> > > > they
> > > > > are
> > > > > > > > not
> > > > > > > > > all
> > > > > > > > > > >> revoked
> > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> assignment
> > > > > becomes
> > > > > > > the
> > > > > > > > > target
> > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to
> > avoid
> > > > > having
> > > > > > > > to
> > > > > > > > > > >> write all
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are
> > based on
> > > > > the
> > > > > > > > > persisted
> > > > > > > > > > >> state
> > > > > > > > > > >> > > > so I
> > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let
> me
> > see
> > > > > if I
> > > > > > > can
> > > > > > > > > > >> improve
> > > > > > > > > > >> > > > this in
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding
> > Connect,
> > > > it
> > > > > > > could
> > > > > > > > > > >> reuse the
> > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it
> > fits in
> > > > > the
> > > > > > > > > > >> protocol. The
> > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > topicid-partitions +
> > > > > > > metadata,
> > > > > > > > > could
> > > > > > > > > > >> > > > Connect
> > > > > > > > > > >> > > > > > fit
> > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15,
> > 2022
> > > > at
> > > > > 1:55
> > > > > > > > PM
> > > > > > > > > > >> Sagar <
> > > > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the
> > KIP.
> > > > I
> > > > > just
> > > > > > > > had
> > > > > > > > > > >> minor
> > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the
> > Assignment
> > > > > Error
> > > > > > > > > > >> section in
> > > > > > > > > > >> > > > Client
> > > > > > > > > > >> > > > > > Side
> > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned
> > => `In
> > > > > this
> > > > > > > > > case,
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > client
> > > > > > > > > > >> > > > > > side
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the
> > group
> > > > > > > > > coordinator`.
> > > > > > > > > > >> In this
> > > > > > > > > > >> > > > > > case are
> > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> > returning
> > > > an
> > > > > > > > > > >> AssignmentError
> > > > > > > > > > >> > > > that's
> > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you
> > think
> > > > > it
> > > > > > > > would
> > > > > > > > > > >> make sense
> > > > > > > > > > >> > > > to
> > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case
> > > > Studies
> > > > > > > > > section, I
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > > slight
> > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have
> the
> > > > same.
> > > > > > > > Consider
> > > > > > > > > > >> this step:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> > heartbeats,
> > > > the
> > > > > > > group
> > > > > > > > > > >> coordinator
> > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> > partitions
> > > > to
> > > > > > > > > revoke. It
> > > > > > > > > > >> > > > persists
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group
> > (epoch=3)
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > > > Assignment
> > > > > > > > > (epoch=3)
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > > > partitions=[foo-0]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > > > partitions=[foo-2]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > > > partitions=[foo-1]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > > > Assignment
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > epoch=2,
> > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > epoch=3,
> > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > epoch=3,
> > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> > heartbeats,
> > > > it
> > > > > > > > > transitions
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > >> > > > > > but
> > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's
> > mentioned
> > > > > that
> > > > > > > > > member C
> > > > > > > > > > >> can't
> > > > > > > > > > >> > > > get
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > > > description
> > > > > > > > > above,
> > > > > > > > > > >> it seems
> > > > > > > > > > >> > > > it
> > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be
> better
> > to
> > > > > remove
> > > > > > > it
> > > > > > > > > and
> > > > > > > > > > >> > > > populate it
> > > > > > > > > > >> > > > > > only
> > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in
> a
> > lot
> > > > of
> > > > > > > other
> > > > > > > > > > >> places, so
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > I
> > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding
> > connect ,
> > > > it
> > > > > > > might
> > > > > > > > > be
> > > > > > > > > > >> out of
> > > > > > > > > > >> > > > scope
> > > > > > > > > > >> > > > > > of
> > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> > understood it
> > > > > would
> > > > > > > > > > >> probably be
> > > > > > > > > > >> > > > > > running in
> > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the
> new
> > > > > rebalance
> > > > > > > > > > >> protocol as it
> > > > > > > > > > >> > > > has
> > > > > > > > > > >> > > > > > its
> > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > IncrementalCooperative).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul
> 15,
> > 2022
> > > > > at
> > > > > > > 5:00
> > > > > > > > > PM
> > > > > > > > > > >> David
> > > > > > > > > > >> > > > Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> > Hector! Our
> > > > > goal
> > > > > > > is
> > > > > > > > > to
> > > > > > > > > > >> move
> > > > > > > > > > >> > > > forward
> > > > > > > > > > >> > > > > > with
> > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of
> > relying
> > > > > on
> > > > > > > one
> > > > > > > > > > >> generic API.
> > > > > > > > > > >> > > > For
> > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same
> > pattern
> > > > > and
> > > > > > > > > > >> reuse/share the
> > > > > > > > > > >> > > > core
> > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side.
> > For
> > > > the
> > > > > > > > schema
> > > > > > > > > > >> registry, I
> > > > > > > > > > >> > > > > > think
> > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a
> > tailored
> > > > > API to
> > > > > > > > do
> > > > > > > > > > >> simple
> > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul
> > 15,
> > > > > 2022 at
> > > > > > > > > 10:22
> > > > > > > > > > >> AM Ismael
> > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three
> quick
> > > > > comments:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1.
> Regarding
> > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > >> > > > vs
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > com.google.re2j.Pattern,
> > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document
> the
> > > > > > > differences
> > > > > > > > > in
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > detail
> > > > > > > > > > >> > > > > > before
> > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said,
> > if
> > > > > people
> > > > > > > > pass
> > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics
> > to be
> > > > > > > honored.
> > > > > > > > > If
> > > > > > > > > > >> we are
> > > > > > > > > > >> > > > doing
> > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should
> > consider
> > > > > adding
> > > > > > > > an
> > > > > > > > > > >> overload
> > > > > > > > > > >> > > > with
> > > > > > > > > > >> > > > > > our own
> > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd
> > want
> > > > to
> > > > > > > > expose
> > > > > > > > > > >> re2j's at
> > > > > > > > > > >> > > > this
> > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2.
> Regarding
> > > > topic
> > > > > > > ids,
> > > > > > > > > any
> > > > > > > > > > >> major new
> > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and
> > should
> > > > > handle
> > > > > > > the
> > > > > > > > > topic
> > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we
> > need to
> > > > > > > handle.
> > > > > > > > I
> > > > > > > > > > >> agree with
> > > > > > > > > > >> > > > > > David
> > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the
> > relevant
> > > > > > > > protocols
> > > > > > > > > > >> that don't
> > > > > > > > > > >> > > > > > have it
> > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on
> the
> > > > > internals
> > > > > > > > > versus
> > > > > > > > > > >> adding
> > > > > > > > > > >> > > > new
> > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find
> that
> > > > > adding
> > > > > > > new
> > > > > > > > > APIs
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > required
> > > > > > > > > > >> > > > > > for
> > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am
> > still
> > > > not
> > > > > sure
> > > > > > > > > about
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful
> for
> > > > > configs
> > > > > > > to
> > > > > > > > be
> > > > > > > > > > >> > > > centralized
> > > > > > > > > > >> > > > > > in the
> > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > > > (auditability,
> > > > > > > > > > >> visibility,
> > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > > > > automatically
> > > > > > > > > > >> deleting
> > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> recovered. A
> > > > good
> > > > > > > > property
> > > > > > > > > > >> for modern
> > > > > > > > > > >> > > > > > systems
> > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > unrecoverable
> > > > data
> > > > > > > loss
> > > > > > > > > > >> scenarios.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed,
> Jul
> > 13,
> > > > > 2022
> > > > > > > at
> > > > > > > > > 3:47
> > > > > > > > > > >> PM David
> > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > > > Guozhang. My
> > > > > > > > > answers
> > > > > > > > > > >> are
> > > > > > > > > > >> > > > below:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > > migration
> > > > > > > > path,
> > > > > > > > > > >> especially
> > > > > > > > > > >> > > > > > the last
> > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > enable
> > > > the
> > > > > new
> > > > > > > > > > >> protocol, in
> > > > > > > > > > >> > > > which
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > protocols /
> > > > > rpcs
> > > > > > > and
> > > > > > > > > old
> > > > > > > > > > >> > > > protocols /
> > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> > How
> > > > the
> > > > > > > > > > >> coordinator could
> > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > protocol
> > > > > is
> > > > > > > > > something
> > > > > > > > > > >> we
> > > > > > > > > > >> > > > need to
> > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I
> > just
> > > > > > > > published
> > > > > > > > > a
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > version
> > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details
> > about
> > > > > this.
> > > > > > > > See
> > > > > > > > > the
> > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > > > > "Compatibility,
> > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > >> > > > > > and
> > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have
> to
> > > > think
> > > > > > > > through
> > > > > > > > > a
> > > > > > > > > > >> few cases
> > > > > > > > > > >> > > > > > now but
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> mechanism
> > > > > should be
> > > > > > > > > > >> understandable.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> > usage
> > > > > of
> > > > > > > > topic
> > > > > > > > > > >> ids. So
> > > > > > > > > > >> > > > far as
> > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part
> of
> > RPCs
> > > > > and
> > > > > > > > admin
> > > > > > > > > > >> client,
> > > > > > > > > > >> > > > but
> > > > > > > > > > >> > > > > > they
> > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > consumers
> > > > > yet.
> > > > > > > I
> > > > > > > > > think
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> > to be
> > > > > > > > > maintaining
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > names
> > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> > the
> > > > > > > augmented
> > > > > > > > > > >> existing
> > > > > > > > > > >> > > > RPCs
> > > > > > > > > > >> > > > > > and the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > secondly,
> > > > > should
> > > > > > > we
> > > > > > > > > ever
> > > > > > > > > > >> consider
> > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> > APIs
> > > > as
> > > > > > > well
> > > > > > > > > (both
> > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > listener for
> > > > > cases
> > > > > > > > > like
> > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> > Assuming
> > > > > that we
> > > > > > > > > would
> > > > > > > > > > >> include
> > > > > > > > > > >> > > > > > converting
> > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to
> > using
> > > > > topic
> > > > > > > > ids
> > > > > > > > > in
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > KIP,
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully
> > operate
> > > > > with
> > > > > > > > topic
> > > > > > > > > > >> ids. That
> > > > > > > > > > >> > > > > > being
> > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> topics
> > > > > names in
> > > > > > > > > various
> > > > > > > > > > >> APIs so
> > > > > > > > > > >> > > > > > having a
> > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > > > inevitable to
> > > > > > > > me.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I
> don't
> > > > have
> > > > > a
> > > > > > > > strong
> > > > > > > > > > >> opinion on
> > > > > > > > > > >> > > > > > this.
> > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond
> the
> > > > > scope of
> > > > > > > > this
> > > > > > > > > > >> KIP. I
> > > > > > > > > > >> > > > would
> > > > > > > > > > >> > > > > > rather
> > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and
> > we
> > > > can
> > > > > > > > consider
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming
> > back to
> > > > > > > > Ismael's
> > > > > > > > > > >> point about
> > > > > > > > > > >> > > > > > using
> > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > >> > > > think
> > > > > > > > > > >> > > > > > that
> > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour
> of
> > it.
> > > > > The
> > > > > > > > > consumer
> > > > > > > > > > >> will
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> topics
> > > > > exists
> > > > > > > > before
> > > > > > > > > > >> passing
> > > > > > > > > > >> > > > them
> > > > > > > > > > >> > > > > > into
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> protocol.
> > > > > Obviously,
> > > > > > > > the
> > > > > > > > > > >> > > > coordinator
> > > > > > > > > > >> > > > > > will
> > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not
> really
> > > > have
> > > > > a
> > > > > > > way
> > > > > > > > to
> > > > > > > > > > >> reject an
> > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > agreeing
> > > > > with
> > > > > > > > > David
> > > > > > > > > > >> on all
> > > > > > > > > > >> > > > other
> > > > > > > > > > >> > > > > > minor
> > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > `subscribe(Pattern)`
> > > > > > > > > > >> question:
> > > > > > > > > > >> > > > > > personally
> > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > deprecate
> > > > the
> > > > > > > > > subscribe
> > > > > > > > > > >> API with
> > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> > just
> > > > > > > > documenting
> > > > > > > > > > >> that our
> > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> > the
> > > > > > > > incompatible
> > > > > > > > > > >> case is a
> > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > overloaded
> > > > > > > `String`
> > > > > > > > > based
> > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> invalid
> > > > > regexes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That
> could
> > > > > work. I
> > > > > > > > have
> > > > > > > > > to
> > > > > > > > > > >> look at
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines
> to
> > > > > better
> > > > > > > > > > >> understand the
> > > > > > > > > > >> > > > > > potential
> > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> would
> > > > work
> > > > > for
> > > > > > > > all
> > > > > > > > > the
> > > > > > > > > > >> basic
> > > > > > > > > > >> > > > > > regular
> > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between
> > the
> > > > two
> > > > > are
> > > > > > > > > mainly
> > > > > > > > > > >> about
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > various
> > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder
> > what
> > > > > other
> > > > > > > > people
> > > > > > > > > > >> think
> > > > > > > > > > >> > > > about
> > > > > > > > > > >> > > > > > this.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue,
> > Jul
> > > > 12,
> > > > > 2022
> > > > > > > > at
> > > > > > > > > > >> 11:28 PM
> > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> wangguoz@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks
> > > > David!
> > > > > I
> > > > > > > > think
> > > > > > > > > on
> > > > > > > > > > >> the high
> > > > > > > > > > >> > > > > > level
> > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > concretize
> > > > > a
> > > > > > > bit
> > > > > > > > > more:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > > migration
> > > > > > > > path,
> > > > > > > > > > >> especially
> > > > > > > > > > >> > > > > > the last
> > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > enable
> > > > the
> > > > > new
> > > > > > > > > > >> protocol, in
> > > > > > > > > > >> > > > which
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > protocols /
> > > > > rpcs
> > > > > > > and
> > > > > > > > > old
> > > > > > > > > > >> > > > protocols /
> > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> > How
> > > > the
> > > > > > > > > > >> coordinator could
> > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > protocol
> > > > > is
> > > > > > > > > something
> > > > > > > > > > >> we
> > > > > > > > > > >> > > > need to
> > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> > usage
> > > > > of
> > > > > > > > topic
> > > > > > > > > > >> ids. So
> > > > > > > > > > >> > > > far as
> > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part
> of
> > RPCs
> > > > > and
> > > > > > > > admin
> > > > > > > > > > >> client,
> > > > > > > > > > >> > > > but
> > > > > > > > > > >> > > > > > they
> > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > consumers
> > > > > yet.
> > > > > > > I
> > > > > > > > > think
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> > to be
> > > > > > > > > maintaining
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > names
> > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> > the
> > > > > > > augmented
> > > > > > > > > > >> existing
> > > > > > > > > > >> > > > RPCs
> > > > > > > > > > >> > > > > > and the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > secondly,
> > > > > should
> > > > > > > we
> > > > > > > > > ever
> > > > > > > > > > >> consider
> > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> > APIs
> > > > as
> > > > > > > well
> > > > > > > > > (both
> > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > listener for
> > > > > cases
> > > > > > > > > like
> > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > agreeing
> > > > > with
> > > > > > > > > David
> > > > > > > > > > >> on all
> > > > > > > > > > >> > > > other
> > > > > > > > > > >> > > > > > minor
> > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > `subscribe(Pattern)`
> > > > > > > > > > >> question:
> > > > > > > > > > >> > > > > > personally
> > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > deprecate
> > > > the
> > > > > > > > > subscribe
> > > > > > > > > > >> API with
> > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> > just
> > > > > > > > documenting
> > > > > > > > > > >> that our
> > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> > the
> > > > > > > > incompatible
> > > > > > > > > > >> case is a
> > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > overloaded
> > > > > > > `String`
> > > > > > > > > based
> > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> invalid
> > > > > regexes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> Guozhang
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On
> Tue,
> > Jul
> > > > > 12,
> > > > > > > 2022
> > > > > > > > > at
> > > > > > > > > > >> 5:23 AM
> > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi
> > Ismael,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Thanks for
> > > > > your
> > > > > > > > > > >> feedback. Let
> > > > > > > > > > >> > > > me
> > > > > > > > > > >> > > > > > answer
> > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 1. I
> > > > think
> > > > > > > it's
> > > > > > > > > > >> premature to
> > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > deprecation
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > removal
> > > > > of the
> > > > > > > > > > >> existing group
> > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > client
> > > > > > > protocol
> > > > > > > > > and
> > > > > > > > > > >> hence
> > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> dependent
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> how
> > > > > quickly
> > > > > > > > > > >> applications
> > > > > > > > > > >> > > > migrate
> > > > > > > > > > >> > > > > > to
> > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That
> > makes
> > > > > > > sense.
> > > > > > > > I
> > > > > > > > > will
> > > > > > > > > > >> > > > remove it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2.
> > The
> > > > KIP
> > > > > > > says
> > > > > > > > we
> > > > > > > > > > >> intend to
> > > > > > > > > > >> > > > > > release
> > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> clear
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > why. If
> > > > we
> > > > > > > added
> > > > > > > > > that
> > > > > > > > > > >> as a
> > > > > > > > > > >> > > > way to
> > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> the
> > > > group
> > > > > > > > > protocol, I
> > > > > > > > > > >> also
> > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let
> me
> > > > > explain
> > > > > > > my
> > > > > > > > > > >> reasoning. As
> > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > coordinator
> > > > > in
> > > > > > > > Java
> > > > > > > > > > >> while we
> > > > > > > > > > >> > > > > > implement
> > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > the
> > > > > > > internals
> > > > > > > > > will
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > slightly
> > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > Therefore, I
> > > > > > > > wanted
> > > > > > > > > to
> > > > > > > > > > >> tighten
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > coordinator
> > > > > to
> > > > > > > the
> > > > > > > > > new
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > alternative
> > > > > > > would
> > > > > > > > > be to
> > > > > > > > > > >> use a
> > > > > > > > > > >> > > > flag
> > > > > > > > > > >> > > > > > to do
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> the
> > > > > software
> > > > > > > > > upgrade.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3.
> > We
> > > > > need to
> > > > > > > > > flesh
> > > > > > > > > > >> out the
> > > > > > > > > > >> > > > > > details of
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > we're
> > > > > saying
> > > > > > > we
> > > > > > > > > will
> > > > > > > > > > >> support
> > > > > > > > > > >> > > > > > online
> > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > explain
> > > > > this
> > > > > > > in
> > > > > > > > > > >> detail. It
> > > > > > > > > > >> > > > could
> > > > > > > > > > >> > > > > > also
> > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > easier.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes,
> > we
> > > > will
> > > > > > > > support
> > > > > > > > > > >> online
> > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > group
> > > > > using
> > > > > > > the
> > > > > > > > > old
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > > > will be
> > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let
> me
> > > > > briefly
> > > > > > > > > explain
> > > > > > > > > > >> how that
> > > > > > > > > > >> > > > > > will work
> > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four
> > step
> > > > > > > process:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1.
> The
> > > > > cluster
> > > > > > > > must
> > > > > > > > > be
> > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > group
> > > > > > > > > coordinator.
> > > > > > > > > > >> Both the
> > > > > > > > > > >> > > > > > old and
> > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > support
> > > > the
> > > > > old
> > > > > > > > > > >> protocol and
> > > > > > > > > > >> > > > rely
> > > > > > > > > > >> > > > > > on the
> > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they
> > can
> > > > > work
> > > > > > > > > together.
> > > > > > > > > > >> This
> > > > > > > > > > >> > > > point
> > > > > > > > > > >> > > > > > is an
> > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > this
> > > > one
> > > > > live
> > > > > > > > > > >> because it
> > > > > > > > > > >> > > > would
> > > > > > > > > > >> > > > > > require
> > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > coordinator
> > > > > and
> > > > > > > > > > >> starting up the
> > > > > > > > > > >> > > > > > new one
> > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > unavailabilities.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2.
> The
> > > > > cluster's
> > > > > > > > > > >> metadata
> > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > enable
> > > > > the
> > > > > > > new
> > > > > > > > > > >> protocol.
> > > > > > > > > > >> > > > This
> > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > terminated
> > > > > > > because
> > > > > > > > > the
> > > > > > > > > > >> old
> > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3.
> The
> > > > > consumers
> > > > > > > > > must be
> > > > > > > > > > >> > > > upgraded
> > > > > > > > > > >> > > > > > to a
> > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > migration
> > > > > (must
> > > > > > > > have
> > > > > > > > > > >> KIP-792).
> > > > > > > > > > >> > > > If
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Nothing
> > > > > must be
> > > > > > > > > done at
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > point.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4.
> The
> > > > > consumers
> > > > > > > > > must
> > > > > > > > > > >> be rolled
> > > > > > > > > > >> > > > > > with the
> > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > consumer
> > > > > group
> > > > > > > is
> > > > > > > > > > >> automatically
> > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> using
> > the
> > > > > new
> > > > > > > > > protocol
> > > > > > > > > > >> joins
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > group.
> > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > > > > protocol are
> > > > > > > > > being
> > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > >> > > > > > the old
> > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > one.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let
> me
> > > > > clarify
> > > > > > > all
> > > > > > > > > of
> > > > > > > > > > >> this in
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4.
> > I am
> > > > > happy
> > > > > > > > > that we
> > > > > > > > > > >> are
> > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > server,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> but
> > it
> > > > > seems
> > > > > > > > like
> > > > > > > > > > >> there
> > > > > > > > > > >> > > > could be
> > > > > > > > > > >> > > > > > some
> > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > have a
> > > > > > > mechanism
> > > > > > > > > for
> > > > > > > > > > >> users to
> > > > > > > > > > >> > > > > > detect
> > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > before
> > > > > > > switching
> > > > > > > > > to
> > > > > > > > > > >> the new
> > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > think
> > > > > that I
> > > > > > > am
> > > > > > > > a
> > > > > > > > > bit
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > believe
> > > > > that the
> > > > > > > > > > >> majority of
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > cases
> > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work
> > with
> > > > > the
> > > > > > > new
> > > > > > > > > > >> engine. The
> > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > reject
> > > > > the
> > > > > > > > > consumer
> > > > > > > > > > >> if the
> > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > migration
> > > > > path,
> > > > > > > in
> > > > > > > > > the
> > > > > > > > > > >> worst
> > > > > > > > > > >> > > > case,
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > group
> > > > > will
> > > > > > > be
> > > > > > > > > > >> rejected.
> > > > > > > > > > >> > > > This
> > > > > > > > > > >> > > > > > should
> > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> would
> > say.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One
> > way
> > > > for
> > > > > > > > > customers to
> > > > > > > > > > >> > > > validate
> > > > > > > > > > >> > > > > > their
> > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod
> > would
> > > > > be to
> > > > > > > > > test
> > > > > > > > > > >> them with
> > > > > > > > > > >> > > > > > another
> > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> could
> > be
> > > > > done
> > > > > > > in a
> > > > > > > > > > >> pre-prod
> > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > extend the
> > > > > > > > > > >> consumer-group tool
> > > > > > > > > > >> > > > to
> > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > mechanism.
> > > > > Would
> > > > > > > > > this be
> > > > > > > > > > >> > > > enough in
> > > > > > > > > > >> > > > > > your
> > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > > > > Related to
> > > > > > > > the
> > > > > > > > > last
> > > > > > > > > > >> > > > question,
> > > > > > > > > > >> > > > > > will
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > stick
> > > > > with the
> > > > > > > > > > >> current regex
> > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > example,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it
> > may
> > > > be
> > > > > > > handy
> > > > > > > > to
> > > > > > > > > > >> keep using
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > migrations
> > > > > > > > simple
> > > > > > > > > and
> > > > > > > > > > >> then
> > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> step.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > understand
> > > > > > > your
> > > > > > > > > point
> > > > > > > > > > >> but I
> > > > > > > > > > >> > > > am
> > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > actually
> > > > > stay
> > > > > > > > in
> > > > > > > > > > >> this mode.
> > > > > > > > > > >> > > > That
> > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > simplifying
> > > > > the
> > > > > > > > > client
> > > > > > > > > > >> because
> > > > > > > > > > >> > > > we
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > metadata on
> > > > > > > > the
> > > > > > > > > > >> client
> > > > > > > > > > >> > > > side. I
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6.
> > When
> > > > > we say
> > > > > > > > > that
> > > > > > > > > > >> the group
> > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> the
> > > > > > > > configurations
> > > > > > > > > > >> and that
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > group is
> > > > > > > > deleted.
> > > > > > > > > > >> Will a
> > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > configurations?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> That's
> > > > > right.
> > > > > > > The
> > > > > > > > > > >> > > > configurations
> > > > > > > > > > >> > > > > > will be
> > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > deleted.
> > > > > They go
> > > > > > > > > > >> together.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7.
> > Will
> > > > > the
> > > > > > > > > choice to
> > > > > > > > > > >> store
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > harder
> > > > to
> > > > > list
> > > > > > > > all
> > > > > > > > > > >> cluster
> > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > don't
> > > > > think
> > > > > > > so.
> > > > > > > > > The
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > configs.
> > > > If
> > > > > you
> > > > > > > > > want to
> > > > > > > > > > >> know
> > > > > > > > > > >> > > > all
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > ask all
> > > > > the
> > > > > > > > group
> > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > >> > > > > > You
> > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > instance.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8.
> > How
> > > > > would
> > > > > > > > > someone
> > > > > > > > > > >> > > > configure a
> > > > > > > > > > >> > > > > > group
> > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > > > > considered
> > > > > > > > > > >> allowing the
> > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Alternatively,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > configs
> > > > > could
> > > > > > > be
> > > > > > > > > > >> decoupled
> > > > > > > > > > >> > > > from
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > The
> > > > > group
> > > > > > > > will
> > > > > > > > > be
> > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > configs
> > > > > will
> > > > > > > > be
> > > > > > > > > > >> lost after
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > passes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9.
> > Will
> > > > > the
> > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > >> > > > > > method
> > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > >> > > > do
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > That's a
> > > > > very
> > > > > > > > group
> > > > > > > > > > >> question. I
> > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > >> > > > not
> > > > > > > > > > >> > > > > > fully
> > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > plan to
> > > > > use,
> > > > > > > it
> > > > > > > > > > >> might be
> > > > > > > > > > >> > > > better
> > > > > > > > > > >> > > > > > to
> > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> which
> > > > takes
> > > > > a
> > > > > > > > > string.
> > > > > > > > > > >> We would
> > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > During the
> > > > > > > > > migration, I
> > > > > > > > > > >> think
> > > > > > > > > > >> > > > that
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> regex
> > and
> > > > > use
> > > > > > > it.
> > > > > > > > > That
> > > > > > > > > > >> should
> > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> cases.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 10.
> > I
> > > > > agree
> > > > > > > with
> > > > > > > > > > >> Justine
> > > > > > > > > > >> > > > that we
> > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > switch
> > > > to
> > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > >> > > > > > from the
> > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> can
> > > > > switch the
> > > > > > > > > > >> default for
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > feature
> > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > metadata.version
> > > > > > > > > once
> > > > > > > > > > >> we
> > > > > > > > > > >> > > > want to
> > > > > > > > > > >> > > > > > make
> > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My
> > plan
> > > > was
> > > > > to
> > > > > > > use
> > > > > > > > > that
> > > > > > > > > > >> feature
> > > > > > > > > > >> > > > > > flag
> > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > phase. I
> > > > > should
> > > > > > > > not
> > > > > > > > > have
> > > > > > > > > > >> > > > mentioned
> > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > > > internal
> > > > > > > config
> > > > > > > > > for
> > > > > > > > > > >> it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 11.
> > Some
> > > > > of
> > > > > > > the
> > > > > > > > > > >> protocol APIs
> > > > > > > > > > >> > > > > > don't
> > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > good to
> > > > > add
> > > > > > > that
> > > > > > > > > for
> > > > > > > > > > >> > > > consistency.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Noted.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 12.
> > It
> > > > is
> > > > > a
> > > > > > > bit
> > > > > > > > > odd
> > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > though
> > > > it
> > > > > > > seems
> > > > > > > > > to do
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > than
> > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > agree.
> > > > > This is
> > > > > > > > > how the
> > > > > > > > > > >> > > > current
> > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > require
> > > > > "Read
> > > > > > > > > Group" to
> > > > > > > > > > >> join a
> > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but
> I
> > am
> > > > not
> > > > > > > sure
> > > > > > > > > that
> > > > > > > > > > >> it is
> > > > > > > > > > >> > > > worth
> > > > > > > > > > >> > > > > > it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 13.
> > How
> > > > is
> > > > > > > topic
> > > > > > > > > > >> recreation
> > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > protocol?
> > > > > It
> > > > > > > > > would be
> > > > > > > > > > >> good to
> > > > > > > > > > >> > > > > > have a
> > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Noted.
> > > > From
> > > > > a
> > > > > > > > > protocol
> > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> topic
> > id
> > > > so
> > > > > it
> > > > > > > > will
> > > > > > > > > > >> treat it
> > > > > > > > > > >> > > > like a
> > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only
> > issue
> > > > > is
> > > > > > > that
> > > > > > > > > the
> > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs
> > so the
> > > > > > > > consumer
> > > > > > > > > > >> would
> > > > > > > > > > >> > > > reuse the
> > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > we
> > > > > should
> > > > > > > > > update
> > > > > > > > > > >> those
> > > > > > > > > > >> > > > APIs
> > > > > > > > > > >> > > > > > as well
> > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > end.
> > > > That
> > > > > > > would
> > > > > > > > > > >> strengthen
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 14.
> > The
> > > > > KIP
> > > > > > > > > mentions
> > > > > > > > > > >> we will
> > > > > > > > > > >> > > > > > write the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> though
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > this is
> > > > an
> > > > > > > > > > >> implementation
> > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> for
> > it?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 15.
> > Do
> > > > we
> > > > > > > have a
> > > > > > > > > > >> scalability
> > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > group
> > > > > protocol
> > > > > > > > can
> > > > > > > > > > >> support?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We
> > don't
> > > > > have
> > > > > > > > > numbers
> > > > > > > > > > >> at the
> > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of
> > members
> > > > > per
> > > > > > > > > group.
> > > > > > > > > > >> We will
> > > > > > > > > > >> > > > > > measure
> > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > implementation.
> > > > > > > > Note
> > > > > > > > > > >> that we
> > > > > > > > > > >> > > > might
> > > > > > > > > > >> > > > > > have
> > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road
> > (e.g.
> > > > > > > offset
> > > > > > > > > > >> commits).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 16.
> > Did
> > > > we
> > > > > > > > > consider
> > > > > > > > > > >> having
> > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > SubscribedTopicNames
> > > > > > > > > > >> in
> > > > > > > > > > >> > > > > > > > > > > >> >
> ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > since we
> > > > > have
> > > > > > > to
> > > > > > > > > > >> resolve the
> > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> the
> > > > topic
> > > > > > > name?
> > > > > > > > > The
> > > > > > > > > > >> > > > difference
> > > > > > > > > > >> > > > > > is that
> > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > efficient
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > whereas
> > > > > > > sending
> > > > > > > > > the
> > > > > > > > > > >> topic
> > > > > > > > > > >> > > > names
> > > > > > > > > > >> > > > > > is less
> > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > recreation is
> > > > > > > > > easier
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > handle
> > > > > > > > > > >> > > > > > if we
> > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The
> > idea
> > > > > was to
> > > > > > > > > > >> consolidate the
> > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> paths
> > but
> > > > I
> > > > > do
> > > > > > > > agree
> > > > > > > > > > >> with your
> > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids
> > may be
> > > > > > > better
> > > > > > > > > here
> > > > > > > > > > >> for the
> > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > suppose
> > > > > that we
> > > > > > > > may
> > > > > > > > > > >> allow
> > > > > > > > > > >> > > > users to
> > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> future
> > > > > because
> > > > > > > > that
> > > > > > > > > is
> > > > > > > > > > >> the only
> > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > re-creation.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On
> > Tue,
> > > > Jul
> > > > > 12,
> > > > > > > > > 2022 at
> > > > > > > > > > >> 1:38 PM
> > > > > > > > > > >> > > > > > David
> > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > djacot@confluent.io>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > > > > Justine,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > Thanks
> > > > for
> > > > > > > your
> > > > > > > > > > >> comments.
> > > > > > > > > > >> > > > Please
> > > > > > > > > > >> > > > > > find
> > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > -
> > Yes,
> > > > > the new
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > relies on
> > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > topic
> > > > > names
> > > > > > > > based
> > > > > > > > > in
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if
> > using
> > > > > topic
> > > > > > > > > names
> > > > > > > > > > >> is the
> > > > > > > > > > >> > > > > > right call
> > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a
> > little
> > > > > more.
> > > > > > > > > > >> Obviously, the
> > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > >> > > > > > > > > > > >>
> >
>
>
> --
> -- Guozhang
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Guozhang,

Thanks. I have updated those assignor interfaces to not include topic
ids. I agree with you that it makes more sense to not expose them at
the moment. This also keeps the Consumer API consistent as we don't
expose them anywhere else.

Best,
David

On Tue, Oct 4, 2022 at 6:30 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hi David,
>
> I made another pass on the doc, and assuming we will change the
> "TopicIdPartition" fields to "TopicPartition" for the consumer-side APIs, I
> think this proposal looks great for me.
>
> Guozhang
>
> On Mon, Oct 3, 2022 at 9:36 AM Guozhang Wang <wa...@gmail.com> wrote:
>
> > 3. Ah thanks, that's my mis-read then. I think keeping the current class
> > and the params of `onAssignment` is fine then.
> >
> > On Mon, Oct 3, 2022 at 4:32 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> >> 3. ConsumerGroupMetadata is already defined in the consumer [1]. I
> >> realized that we pass it in the current assignor so I thought that we
> >> could reuse it. The alternative would be to add the same information
> >> in a class which includes the error and the member assignment as well.
> >>
> >> [1]
> >> https://github.com/apache/kafka/blob/99b9b3e84f4e98c3f07714e1de6a139a004cbc5b/clients/src/main/java/org/apache/kafka/clients/consumer/ConsumerGroupMetadata.java
> >>
> >> Best,
> >> David
> >>
> >> On Mon, Oct 3, 2022 at 6:15 AM Guozhang Wang <wa...@gmail.com> wrote:
> >> >
> >> > 1. Yeah that makes sense. I originally concerned if there's any
> >> consistency
> >> > issues if the HB request with encoded topic IDs and the GPA/GIA RPCs
> >> with
> >> > encoded topic IDs use mappings from two different brokers, and one of
> >> them
> >> > is stale, but after thinking twice about it I think as long as we handle
> >> > e.g. `unknown topic ids` from HB responses then we should be fine.
> >> >
> >> > 3. Ack, in that case I think we also need to define
> >> "ConsumerGroupMetadata"
> >> > which is passed in `onAssignment`. We have a schema defined in the
> >> > persistent storage for different purposes, but not in the consumer APIs
> >> > yet. I think that would just include a single "Map<String, Metadata>"?
> >> >
> >> > On Fri, Sep 30, 2022 at 5:51 AM David Jacot <djacot@confluent.io.invalid
> >> >
> >> > wrote:
> >> >
> >> > > Hey Guozhang,
> >> > >
> >> > > Thanks for your comments.
> >> > >
> >> > > 1. That's a very good question. I think that it is better to separate
> >> > > the two concerns here. In my opinion, the GPA should only be used to
> >> > > feed the assignor. I see the GPA as a way to call a remote function so
> >> > > the GPA provides all its inputs to ensure that the assignor does not
> >> > > depend on any local state. Knowing this, I would not use the GPA to
> >> > > update the local state either. That is not its purpose. The consumer
> >> > > is responsible to update its local metadata cache either when it gets
> >> > > an unknown topic in the user-provided subscription or in the
> >> > > coordinator-provided assignment. The consumer has to do this anyway to
> >> > > get all topic metadata (e.g. leaders).
> >> > >
> >> > > 2. Interesting thought. Handling the translation is definitely
> >> > > possible. I wonder if there are cases where having the topic ids is
> >> > > beneficial. I am thinking about when a topic is deleted and recreated
> >> > > right away. In this case, the current member assignment would still
> >> > > have the old topic id and we would not be able to resolve it anymore.
> >> > > I suppose that we should directly remove it from the current
> >> > > assignment on the server side in this case and only add the new one.
> >> > > As I first thought, that seems OK. The major downside is that the user
> >> > > of the assignor would not know that the topic is different from what
> >> > > it may have cached locally.
> >> > >
> >> > > We could say that we don't expose topic ids externally for now in
> >> > > order to stay consistent with the other interfaces but it would be
> >> > > annoying to add it back if we want to expose topic ids in the future.
> >> > > I personally believe that we will have to expose them at some point.
> >> > > For instance, in the rebalance listener, it would be useful to know
> >> > > that a topic has been deleted and recreated. The only way to do this
> >> > > is to know the topic id. Let me think a little more about this and
> >> > > come back to you.
> >> > >
> >> > > 3. Sorry for this one. I was a typo. PartitionAssingor#assign should
> >> > > return GroupAssignment.
> >> > >
> >> > > Best,
> >> > > David
> >> > >
> >> > > On Thu, Sep 29, 2022 at 6:06 PM Guozhang Wang <wa...@gmail.com>
> >> wrote:
> >> > > >
> >> > > > Thanks David, I read through the updated wiki, and I have a couple
> >> > > > follow-up questions:
> >> > > >
> >> > > > 1. It seems we would have two potential sources to update the
> >> consumer's
> >> > > > internally maintained topic name -> topic ID mapping, which is used
> >> for
> >> > > > encoding subscribed topic Ids in the HB request, and offset commit
> >> / list
> >> > > > offset requests, etc. One is from metadata responses, and another
> >> one is
> >> > > > from the GPA (group prepare assignment). On the other hand, the
> >> assignor
> >> > > > APIs directly take parsed AssignmentMemberSpec and
> >> > > AssignmentTopicMetadata
> >> > > > from GPA responses directly instead of trying to consult the
> >> consumer's
> >> > > own
> >> > > > cached mapping. So just to clarify, consumers would also update its
> >> > > > internal mapping from the GPA responses as well, right?
> >> > > >
> >> > > > 2. I'm wondering if it's necessary to expose the topic IDs to the
> >> > > assignor
> >> > > > developers. What they need to do at the moment is to take the
> >> mapping
> >> > > from
> >> > > > AssignmentTopicMetadata, and translate their decided assignment into
> >> > > topic
> >> > > > IDs when returning the MemberAssignment. We can instead just let
> >> consumer
> >> > > > to do such translation for them internally and still let the
> >> assignor
> >> > > > developers to only deal with the topic names. Of course that means
> >> the
> >> > > > client side assignor APIs would be a bit further diverged from the
> >> > > > broker-side assignor such that:
> >> > > >
> >> > > > a) Topic metadata would be encoded just as String -> num.partitions;
> >> > > > b) AssignmentMemberSpec contains subscribedTopicNames and
> >> topicPartitions
> >> > > > as target assignments;
> >> > > > c) Returned MemberAssignment contains list of topicPartitions
> >> instead of
> >> > > > topicIdPartitions.
> >> > > >
> >> > > > Internally, consumers would convert the IDs to names when calling
> >> the
> >> > > > assign() function, and convert them back when sending the request;
> >> and
> >> > > > similarly it would convert the IDs to names when calling
> >> onAssignment()
> >> > > as
> >> > > > well.
> >> > > >
> >> > > > In short, I'm wondering if we should bite the bullet for the users,
> >> > > eating
> >> > > > more complexities internally than pushing that to users.
> >> > > >
> >> > > > 3. This is a nit one: the return type of `Assignment
> >> > > assign(AssignmentSpec
> >> > > > assignmentSpec);` seems not defined in the doc. I think it would
> >> just
> >> > > > include two fields, as:
> >> > > >
> >> > > > class Assignment {
> >> > > >         GroupAssignment groupAssignment;
> >> > > >
> >> > > >         ConsumerGroupMetadata consumerGroupMetadata;
> >> > > >     }
> >> > > >
> >> > > > Is that right?
> >> > > >
> >> > > > Guozhang
> >> > > >
> >> > > > On Thu, Sep 29, 2022 at 1:26 AM David Jacot
> >> <djacot@confluent.io.invalid
> >> > > >
> >> > > > wrote:
> >> > > >
> >> > > > > Hi Guozhang,
> >> > > > >
> >> > > > > Thanks for your input.
> >> > > > >
> >> > > > > > 1) in HB request, when clients send the subscription, they are
> >> still
> >> > > > > > sending them in terms of topic names. That is, the consumer
> >> would
> >> > > not try
> >> > > > > > to keep all the mappings from topic names to topic ids for the
> >> > > subscribed
> >> > > > > > topics.
> >> > > > >
> >> > > > > This is a mistake of mine. The heartbeat request should use Topic
> >> Ids.
> >> > > > > I forgot to change it. The idea is that the client will get the
> >> > > > > metadata first to translate the topic names to topic ids and thus
> >> also
> >> > > > > have the mapping in its local metadata cache.
> >> > > > >
> >> > > > > > 2) in HB responses, the assignment's topics are encoded as IDs
> >> > > already.
> >> > > > > And
> >> > > > > > fetch requests can directly rely on that since they just need
> >> topic
> >> > > IDs
> >> > > > > and
> >> > > > > > do not care what actual topic names are.
> >> > > > >
> >> > > > > That's right.
> >> > > > >
> >> > > > > > 3) in prepare-assignment responses, we return subscriptions as
> >> topic
> >> > > IDs
> >> > > > > as
> >> > > > > > well.
> >> > > > >
> >> > > > > That's right.
> >> > > > >
> >> > > > > > In a word, the consumer does not try to maintain the mapping of
> >> topic
> >> > > > > names
> >> > > > > > to IDs at all, and even if they do, they will not expose it in
> >> e.g.
> >> > > > > > assignor APIs. But for client-side assignors like Streams, if
> >> they
> >> > > do not
> >> > > > > > maintain the mapping from topic IDs to topic names, certain
> >> logic
> >> > > such as
> >> > > > > > co-partitioning indeed cannot be executed.
> >> > > > >
> >> > > > > As I said, the consumer will have the mapping in its local
> >> metadata
> >> > > > > cache. It needs to get the partition metadata anyway and the
> >> topic id
> >> > > > > is part of it. However, I do agree that this mapping is not
> >> exposed.
> >> > > > >
> >> > > > > > So I think we'd need to consider 1) use topic names in
> >> subscriptions
> >> > > in
> >> > > > > the
> >> > > > > > prepare-assignment responses, 2) use topic Names in client-side
> >> > > > > > PartitionAssignor#GroupMember#subscribedTopics and
> >> > > > > > PartitionAssignor#TopicMetadata#topicName.
> >> > > > >
> >> > > > > 1) makes sense to me. I think that we could actually provide both
> >> > > > > names and ids. Regarding 2), I think that I am slightly in favor
> >> of
> >> > > > > using both names and ids as well. Let me try to update the
> >> protocol
> >> > > > > and the interface towards this.
> >> > > > >
> >> > > > > Thanks,
> >> > > > > David
> >> > > > >
> >> > > > > On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wangguoz@gmail.com
> >> >
> >> > > wrote:
> >> > > > > >
> >> > > > > > Hello Jun/David,
> >> > > > > >
> >> > > > > > Just to chime in here for Jun's comment #50, today Streams does
> >> not
> >> > > have
> >> > > > > a
> >> > > > > > way to map topic names to topic IDs, and in its topology
> >> description
> >> > > > > topics
> >> > > > > > are encoded with names.
> >> > > > > >
> >> > > > > > What I originally thought was that, on the consumer side, we'd
> >> need
> >> > > to
> >> > > > > > maintain this mapping from topic names to topic IDs anyways, and
> >> > > Streams
> >> > > > > > should leverage on the consumer's functionality to get the
> >> > > corresponding
> >> > > > > > topic names for the given topic IDs. But when I read through
> >> the RPC
> >> > > > > again
> >> > > > > > I realized it may not be the case, since:
> >> > > > > >
> >> > > > > > 1) in HB request, when clients send the subscription, they are
> >> still
> >> > > > > > sending them in terms of topic names. That is, the consumer
> >> would
> >> > > not try
> >> > > > > > to keep all the mappings from topic names to topic ids for the
> >> > > subscribed
> >> > > > > > topics.
> >> > > > > > 2) in HB responses, the assignment's topics are encoded as IDs
> >> > > already.
> >> > > > > And
> >> > > > > > fetch requests can directly rely on that since they just need
> >> topic
> >> > > IDs
> >> > > > > and
> >> > > > > > do not care what actual topic names are.
> >> > > > > > 3) in prepare-assignment responses, we return subscriptions as
> >> topic
> >> > > IDs
> >> > > > > as
> >> > > > > > well.
> >> > > > > >
> >> > > > > > In a word, the consumer does not try to maintain the mapping of
> >> topic
> >> > > > > names
> >> > > > > > to IDs at all, and even if they do, they will not expose it in
> >> e.g.
> >> > > > > > assignor APIs. But for client-side assignors like Streams, if
> >> they
> >> > > do not
> >> > > > > > maintain the mapping from topic IDs to topic names, certain
> >> logic
> >> > > such as
> >> > > > > > co-partitioning indeed cannot be executed.
> >> > > > > >
> >> > > > > > So I think we'd need to consider 1) use topic names in
> >> subscriptions
> >> > > in
> >> > > > > the
> >> > > > > > prepare-assignment responses, 2) use topic Names in client-side
> >> > > > > > PartitionAssignor#GroupMember#subscribedTopics and
> >> > > > > > PartitionAssignor#TopicMetadata#topicName.
> >> > > > > >
> >> > > > > > On Wed, Sep 28, 2022 at 1:26 AM David Jacot
> >> > > <djacot@confluent.io.invalid
> >> > > > > >
> >> > > > > > wrote:
> >> > > > > >
> >> > > > > > > Hi Jun,
> >> > > > > > >
> >> > > > > > > Thanks for your comments.
> >> > > > > > >
> >> > > > > > > 48. Okay. There is a typo in the text:
> >> > > > > > > > Note that the persisted target assignment of B already
> >> includes
> >> > > foo-2
> >> > > > > > > here
> >> > > > > > > but B does not get it yet because foo-2 is not free.
> >> > > > > > >
> >> > > > > > > It should be:
> >> > > > > > > > Note that the persisted member assignment of B already
> >> includes
> >> > > foo-2
> >> > > > > > > here
> >> > > > > > > but B does not get it yet because foo-2 is not free.
> >> > > > > > >
> >> > > > > > > The examples are showing the persisted state in the
> >> > > __consumer_offsets
> >> > > > > > > topics. When a member transitions to the next epoch, we
> >> directly
> >> > > > > > > persist the whole target assignment in order to reduce the
> >> number
> >> > > of
> >> > > > > > > writes and we manage dependencies in memory. So "B - epoch=2,
> >> > > > > > > partitions=[foo-2]" is actually correct from a persisted
> >> record
> >> > > > > > > perspective. However, the in-memory state would be different
> >> > > because
> >> > > > > > > foo-2 is not yet available as you pointed out. The in-memory
> >> state
> >> > > > > > > would like as follow:
> >> > > > > > >
> >> > > > > > > Group (epoch=2)
> >> > > > > > > A
> >> > > > > > > B
> >> > > > > > > Target Assignment (epoch=2)
> >> > > > > > > A - partitions=[foo-0, foo-1]
> >> > > > > > > B - partitions=[foo-2]
> >> > > > > > > Member Assignment
> >> > > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending
> >> > > partitions=[]
> >> > > > > > > B - epoch=2, partitions=[], pending partitions=[foo-2]
> >> > > > > > >
> >> > > > > > > I suppose that the confusion comes from this. I will go and
> >> convert
> >> > > > > > > all the examples to their in-memory representation in order to
> >> > > reduce
> >> > > > > > > the confusion.
> >> > > > > > >
> >> > > > > > > 50. Okay. I have to admit that I am not so familiar with the
> >> > > internals
> >> > > > > > > of KStreams. If I understand correctly, you're saying that
> >> KStreams
> >> > > > > > > may not have topic ids in its topology so topic names are
> >> missing
> >> > > here
> >> > > > > > > to match the topics. I was hoping that KStreams would have
> >> topic
> >> > > ids
> >> > > > > > > in the topology as well. If that is not the case, we can also
> >> add
> >> > > > > > > topic names on our side. I will check this.
> >> > > > > > >
> >> > > > > > > 53. It seems that I was not clear. Sorry for that. I did not
> >> mean
> >> > > to
> >> > > > > > > say that MemberAssignment.metadata is not used by the group
> >> > > > > > > coordinator. Let me re-explain the flow.
> >> > > > > > >
> >> > > > > > > PartitionAssignor.assign() returns an Assignment. Assignment
> >> > > contains
> >> > > > > > > MemberAssignments where each member has
> >> MemberAssignment.metadata.
> >> > > The
> >> > > > > > > Assignment is installed by the client with the
> >> > > > > > > ConsumerGroupInstallAssignment API and this overrides the
> >> current
> >> > > > > > > target assignment in the coordinator. Then the group
> >> coordinator
> >> > > uses
> >> > > > > > > it to reconcile members. Therefore the target assignment will
> >> > > > > > > eventually become the current member assignment and it will be
> >> > > > > > > delivered to the member, including the original
> >> > > > > > > MemberAssignment.metadata defines by the assignor.
> >> > > > > > >
> >> > > > > > > I hope that this helps to clarify how
> >> MemberAssignment.metadata is
> >> > > > > > > used afterwards.
> >> > > > > > >
> >> > > > > > > Best,
> >> > > > > > > David
> >> > > > > > >
> >> > > > > > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao
> >> <jun@confluent.io.invalid
> >> > > >
> >> > > > > wrote:
> >> > > > > > > > Hi, David,
> >> > > > > > > >
> >> > > > > > > > Thanks for the reply. A few more follow up questions.
> >> > > > > > > >
> >> > > > > > > > 48. Hmm, so member assignment is what the client sees.
> >> Then, I
> >> > > am not
> >> > > > > > > sure
> >> > > > > > > > about the following. The text says B cannot get foo-2 yet.
> >> But in
> >> > > > > Member
> >> > > > > > > > Assignment, B has foo-2.
> >> > > > > > > >
> >> > > > > > > > At this point B can transition to epoch 2 but cannot get
> >> foo-2
> >> > > until
> >> > > > > A
> >> > > > > > > > revokes it.
> >> > > > > > > >
> >> > > > > > > > Note that the persisted target assignment of B already
> >> includes
> >> > > foo-2
> >> > > > > > > here
> >> > > > > > > > but B does not get it yet because foo-2 is not free.
> >> > > > > > > >
> >> > > > > > > > Group (epoch=2)
> >> > > > > > > > A
> >> > > > > > > > B
> >> > > > > > > > Target Assignment (epoch=2)
> >> > > > > > > > A - partitions=[foo-0, foo-1]
> >> > > > > > > > B - partitions=[foo-2]
> >> > > > > > > > Member Assignment
> >> > > > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> >> > > > > > > > B - epoch=2, partitions=[foo-2]
> >> > > > > > > >
> >> > > > > > > >
> >> > > > > > > > 50. Hmm, I thought KStreams doesn't use range assignor all
> >> the
> >> > > time.
> >> > > > > It
> >> > > > > > > > does the assignment based on the topology. For example, if
> >> you
> >> > > are
> >> > > > > > > joining
> >> > > > > > > > two topics A and B and deriving the join result in a 3rd
> >> topic C,
> >> > > > > which
> >> > > > > > > > will be aggregated. You want to use range to co-locate the
> >> > > partitions
> >> > > > > > > from
> >> > > > > > > > topic A and B. However, for topic C, the partition could
> >> > > potentially
> >> > > > > be
> >> > > > > > > > assigned in round robin. Without the topic name, KStreams
> >> won't
> >> > > know
> >> > > > > > > which
> >> > > > > > > > topics need to be co-located.
> >> > > > > > > >
> >> > > > > > > > 53. If MemberAssignment.metadata is not used by the group
> >> > > > > coordinator,
> >> > > > > > > > where is it being used?
> >> > > > > > > >
> >> > > > > > > > Thanks,
> >> > > > > > > >
> >> > > > > > > > Jun
> >> > > > > > > >
> >> > > > > > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> >> > > > > > > <ja...@confluent.io.invalid>
> >> > > > > > > > wrote:
> >> > > > > > > >
> >> > > > > > > > > Hi David,
> >> > > > > > > > >
> >> > > > > > > > > Thanks for the response. Replied below:
> >> > > > > > > > >
> >> > > > > > > > > > I don't have a strong opinion on this one. However, it
> >> seems
> >> > > that
> >> > > > > > > this
> >> > > > > > > > > could be a pandora box if our subset turns out to not be
> >> > > sufficient
> >> > > > > > > > > for some existing users. We will have to add more to it
> >> and the
> >> > > > > > > > > compatibility story between versions could be a real
> >> > > nightmare. One
> >> > > > > > > > > thing that we could do for sure is disabling all unicode
> >> > > groups in
> >> > > > > the
> >> > > > > > > > > library. That seems possible and would already reduce the
> >> > > scope.
> >> > > > > > > > >
> >> > > > > > > > > Changes to the regex syntax seems not a major problem to
> >> me. We
> >> > > > > would
> >> > > > > > > bump
> >> > > > > > > > > the protocol version as we usually do. Actually that begs
> >> the
> >> > > same
> >> > > > > > > question
> >> > > > > > > > > about RE2. How stable is the syntax and does it differ
> >> > > depending
> >> > > > > on the
> >> > > > > > > > > version of the library? In general, I do agree that we'd
> >> be
> >> > > better
> >> > > > > off
> >> > > > > > > not
> >> > > > > > > > > having to manage the syntax ourselves if we can convince
> >> > > ourselves
> >> > > > > that
> >> > > > > > > > > compatibility won't be a problem and the client won't
> >> > > eventually
> >> > > > > need
> >> > > > > > > to
> >> > > > > > > > > have access to it.
> >> > > > > > > > >
> >> > > > > > > > > Thanks,
> >> > > > > > > > > Jason
> >> > > > > > > > >
> >> > > > > > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> >> > > > > > > <dj...@confluent.io.invalid>
> >> > > > > > > > > wrote:
> >> > > > > > > > >
> >> > > > > > > > > > Hi Luke,
> >> > > > > > > > > >
> >> > > > > > > > > > Thanks for your comments.
> >> > > > > > > > > >
> >> > > > > > > > > > Your suggestion makes sense. Jun's made the same one.
> >> We are
> >> > > also
> >> > > > > > > > > > discussing whether we should directly return the group
> >> state
> >> > > in
> >> > > > > the
> >> > > > > > > > > > response and drop the prepare assignment api. It seems
> >> that
> >> > > the
> >> > > > > > > > > > current consensus is to keep it though. If that stays
> >> like
> >> > > this,
> >> > > > > I
> >> > > > > > > > > > will apply your suggestion.
> >> > > > > > > > > >
> >> > > > > > > > > > Thanks,
> >> > > > > > > > > > David
> >> > > > > > > > > >
> >> > > > > > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <
> >> > > showuon@gmail.com>
> >> > > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > Hi David,
> >> > > > > > > > > > >
> >> > > > > > > > > > > Thanks for your patience for answering all these
> >> questions!
> >> > > > > > > > > > >
> >> > > > > > > > > > > One more question from me:
> >> > > > > > > > > > > 1. When the selected consumer is going to be in
> >> client-side
> >> > > > > > > assigning,
> >> > > > > > > > > we
> >> > > > > > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response.
> >> I'm
> >> > > > > > > wondering if
> >> > > > > > > > > we
> >> > > > > > > > > > > could use a flag to indicate the assignment is
> >> required.
> >> > > The
> >> > > > > > > rationale
> >> > > > > > > > > is
> >> > > > > > > > > > > when returning error, we lose a chance to catch up
> >> with the
> >> > > > > latest
> >> > > > > > > > > target
> >> > > > > > > > > > > assignment or revoke partitions. Since we expected the
> >> > > record
> >> > > > > > > fetching
> >> > > > > > > > > is
> >> > > > > > > > > > > not blocked by the rebalance/assignment computing, we
> >> might
> >> > > > > want to
> >> > > > > > > > > allow
> >> > > > > > > > > > > the members to update the member assignment in each
> >> > > heartbeat.
> >> > > > > > > > > > >
> >> > > > > > > > > > > Luke
> >> > > > > > > > > > >
> >> > > > > > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> >> > > > > > > > > > <ja...@confluent.io.invalid>
> >> > > > > > > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > > Hey David,
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Thanks for the response.
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > > Do you think that we could have a case where two
> >> > > members
> >> > > > > with
> >> > > > > > > the
> >> > > > > > > > > > same
> >> > > > > > > > > > > > static id are active?
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > I was just thinking about the handoff when we spin
> >> up a
> >> > > new
> >> > > > > node
> >> > > > > > > > > > before the
> >> > > > > > > > > > > > old one completes shutdown. It's probably ok to
> >> fence
> >> > > the old
> >> > > > > > > member
> >> > > > > > > > > > > > abruptly. I do recall we had a few issues with this
> >> > > handoff
> >> > > > > in
> >> > > > > > > the
> >> > > > > > > > > > past. It
> >> > > > > > > > > > > > was due to the fact that the replaced node might
> >> have
> >> > > been
> >> > > > > the
> >> > > > > > > > > leader.
> >> > > > > > > > > > > > Since we've removed the leader notion, perhaps this
> >> kind
> >> > > of
> >> > > > > issue
> >> > > > > > > > > won't
> >> > > > > > > > > > > > resurface.
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > In regard to the RE2 library. I understand the
> >> intent to
> >> > > pass
> >> > > > > > > this
> >> > > > > > > > > > through
> >> > > > > > > > > > > > as a string without any validation so that we don't
> >> need
> >> > > the
> >> > > > > > > > > > dependence on
> >> > > > > > > > > > > > the client. I do wonder if that will be restrictive
> >> in
> >> > > > > practice.
> >> > > > > > > For
> >> > > > > > > > > > > > example, if a regex changes on the client side,
> >> would we
> >> > > > > need to
> >> > > > > > > > > > validate
> >> > > > > > > > > > > > against the current assignment to see which
> >> partitions to
> >> > > > > > > continue
> >> > > > > > > > > > > > fetching? Or would we instead invalidate the
> >> complete
> >> > > current
> >> > > > > > > > > > assignment?
> >> > > > > > > > > > > > Perhaps this is a rare case and we can do the
> >> latter.
> >> > > There
> >> > > > > > > could be
> >> > > > > > > > > > more
> >> > > > > > > > > > > > situations though.
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > As an alternative, I wonder if we could define a
> >> > > minimalistic
> >> > > > > > > regex
> >> > > > > > > > > > syntax
> >> > > > > > > > > > > > (a subset of RE2 say) that could map more easily to
> >> regex
> >> > > > > support
> >> > > > > > > > > that
> >> > > > > > > > > > > > already exists in clients? For example, in this
> >> > > scenario, we
> >> > > > > are
> >> > > > > > > > > > matching
> >> > > > > > > > > > > > topics, so we only need to be able to represent
> >> topic
> >> > > > > characters
> >> > > > > > > and
> >> > > > > > > > > > not
> >> > > > > > > > > > > > the whole smorgasbord of whitespace and unicode
> >> options.
> >> > > > > Would it
> >> > > > > > > > > make
> >> > > > > > > > > > > > sense to have a reduced syntax which includes only
> >> > > standard
> >> > > > > topic
> >> > > > > > > > > > > > characters, wildcards, character classes, and
> >> > > repetitions?
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Best,
> >> > > > > > > > > > > > Jason
> >> > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> >> > > > > > > wangguoz@gmail.com>
> >> > > > > > > > > > wrote:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > > Hi David, thanks for all the detailed
> >> explanations. I
> >> > > think
> >> > > > > > > they
> >> > > > > > > > > all
> >> > > > > > > > > > make
> >> > > > > > > > > > > > > sense. Just want to have a couple follow-ups here:
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > > I don't really see the benefits here because
> >> server
> >> > > side
> >> > > > > > > > > assignors
> >> > > > > > > > > > > > > don't have metadata at all. They only assign
> >> > > > > topic-partitions.
> >> > > > > > > They
> >> > > > > > > > > > > > > are not supposed to generate metadata nor to
> >> receive
> >> > > > > metadata
> >> > > > > > > from
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > members.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Just to clarify I was asking about the `version`
> >> of the
> >> > > > > > > assignor
> >> > > > > > > > > > (i.e. up
> >> > > > > > > > > > > > > to what version that the client would support),
> >> and I
> >> > > do
> >> > > > > agree
> >> > > > > > > we
> >> > > > > > > > > > would
> >> > > > > > > > > > > > not
> >> > > > > > > > > > > > > need metadata. What I have in mind is that, for
> >> some
> >> > > > > specific
> >> > > > > > > > > > built-in
> >> > > > > > > > > > > > > broker-assignors, e.g. rack-aware assignors, if
> >> it's
> >> > > > > possible
> >> > > > > > > that
> >> > > > > > > > > > in a
> >> > > > > > > > > > > > > newer version we would have a hierarchical rack ID
> >> > > string
> >> > > > > > > format,
> >> > > > > > > > > > like
> >> > > > > > > > > > > > > "tier1-tier2" etc, but if some client has not
> >> upgraded
> >> > > > > their
> >> > > > > > > rack
> >> > > > > > > > > ID
> >> > > > > > > > > > > > > would still be in old format. In this case, the
> >> broker
> >> > > then
> >> > > > > > > needs
> >> > > > > > > > > to
> >> > > > > > > > > > > > choose
> >> > > > > > > > > > > > > the old versioned assignor. I'm probably making
> >> > > something
> >> > > > > up
> >> > > > > > > here
> >> > > > > > > > > for
> >> > > > > > > > > > > > rack
> >> > > > > > > > > > > > > aware assignors, but I'm wondering if in general
> >> such
> >> > > an
> >> > > > > > > > > > "auto-downgrade"
> >> > > > > > > > > > > > > behavior would be needed still for broker-side
> >> > > assignor,
> >> > > > > and
> >> > > > > > > if yes
> >> > > > > > > > > > would
> >> > > > > > > > > > > > > "version" still be useful.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > > Yeah, that's right. Within a rebalance,
> >> > > `onAssignment` is
> >> > > > > > > called
> >> > > > > > > > > > once
> >> > > > > > > > > > > > > when the member transitions to a new epoch. This
> >> one
> >> > > > > contains
> >> > > > > > > the
> >> > > > > > > > > > full
> >> > > > > > > > > > > > > metadata provided by the client side assignor.
> >> Then,
> >> > > > > > > `onAssignment`
> >> > > > > > > > > > > > > can be called max N times where N is the number of
> >> > > > > partitions
> >> > > > > > > > > pending
> >> > > > > > > > > > > > > revocation by other members. Let me try to clarify
> >> > > this in
> >> > > > > the
> >> > > > > > > KIP.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Okay, my understanding is that the calling
> >> ordering of
> >> > > > > these
> >> > > > > > > > > > callbacks
> >> > > > > > > > > > > > > would be like the following:
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > ----------------------------------------
> >> > > > > > > > > > > > > onPartitionsRevoked();   // just once, since we
> >> do not
> >> > > > > really
> >> > > > > > > need
> >> > > > > > > > > > > > > to revoke incrementally.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > onAssignment();    // the first call, with epoch
> >> > > > > incremented
> >> > > > > > > > > > > > > onPartitionsAssigned();   // paired with the
> >> > > onAssignment
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > onAssignment();              // the first
> >> onAssignment
> >> > > > > would
> >> > > > > > > bump
> >> > > > > > > > > up
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > epoch, and the metadata reflected.
> >> > > > > > > > > > > > > onPartitionsAssigned();   // each time we get an
> >> > > additional
> >> > > > > > > > > > assignment,
> >> > > > > > > > > > > > we
> >> > > > > > > > > > > > > call onAssignment and then paired with an
> >> > > > > onPartitionsAssigned
> >> > > > > > > > > > > > > ...
> >> > > > > > > > > > > > > onAssignment();
> >> > > > > > > > > > > > > onPartitionsAssigned();   // on each of the
> >> > > onAssignment
> >> > > > > > > calls, the
> >> > > > > > > > > > > > encoded
> >> > > > > > > > > > > > > metadata would not change, only the incrementally
> >> added
> >> > > > > > > partitions
> >> > > > > > > > > be
> >> > > > > > > > > > > > > reflected
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Is that the case?
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > I'm wondering if we would still call onAssignment
> >> just
> >> > > > > once,
> >> > > > > > > that
> >> > > > > > > > > > encodes
> >> > > > > > > > > > > > > all the assignment for this rebalance, including
> >> all
> >> > > the
> >> > > > > > > partitions
> >> > > > > > > > > > that
> >> > > > > > > > > > > > > should be assigned to the member but not yet
> >> assigned
> >> > > since
> >> > > > > > > they
> >> > > > > > > > > > have not
> >> > > > > > > > > > > > > been revoked by others. In that case the call
> >> ordering
> >> > > > > would
> >> > > > > > > be:
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > ----------------------------------------
> >> > > > > > > > > > > > > onPartitionsRevoked();   // just once
> >> > > > > > > > > > > > > onAssignment();    // just once, with epoch
> >> > > incremented,
> >> > > > > and
> >> > > > > > > > > metadata
> >> > > > > > > > > > > > > encoded changed, the "assignment" field also
> >> reflect
> >> > > the
> >> > > > > final
> >> > > > > > > > > target
> >> > > > > > > > > > > > > assignment
> >> > > > > > > > > > > > > onPartitionsAssigned();   // multiple times, which
> >> > > > > represent
> >> > > > > > > > > > > > incrementally
> >> > > > > > > > > > > > > added partitions
> >> > > > > > > > > > > > > ...
> >> > > > > > > > > > > > > onPartitionsAssigned();
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > The motivation from this is that, most users
> >> would only
> >> > > > > > > implement
> >> > > > > > > > > the
> >> > > > > > > > > > > > > rebalance callback listeners and hence we'd
> >> definitely
> >> > > > > need to
> >> > > > > > > make
> >> > > > > > > > > > sure
> >> > > > > > > > > > > > > the semantics of that does not change much, and
> >> the
> >> > > time
> >> > > > > > > > > > > > > `onPartitionsAssigned` indicate the time when the
> >> > > > > partitions
> >> > > > > > > are
> >> > > > > > > > > > actually
> >> > > > > > > > > > > > > assigned to it; while for assignors, the
> >> > > `onAssignment` is
> >> > > > > > > used to
> >> > > > > > > > > > > > indicate
> >> > > > > > > > > > > > > what decision is made regarding for this member,
> >> i.e.
> >> > > when
> >> > > > > the
> >> > > > > > > > > > partitions
> >> > > > > > > > > > > > > are decided to be given to it, but not necessarily
> >> > > meant
> >> > > > > that
> >> > > > > > > it
> >> > > > > > > > > has
> >> > > > > > > > > > been
> >> > > > > > > > > > > > > given, since that time should be determined by the
> >> > > time of
> >> > > > > > > > > > > > > `onPartitionsAssigned`. The benefits though,
> >> would be
> >> > > that
> >> > > > > > > assignor
> >> > > > > > > > > > > > > implementers would not need to reason which
> >> > > `onAssignment`
> >> > > > > > > would be
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > last one for this epoch.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Guozhang
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Guozhang
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> >> > > > > > > > > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > wrote:
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > > Hi Guozhang,
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > 1) The client-side
> >> "PartitionAssignor#Assignment"
> >> > > has
> >> > > > > an
> >> > > > > > > Error
> >> > > > > > > > > > byte
> >> > > > > > > > > > > > > > field,
> >> > > > > > > > > > > > > > > while the broker-side
> >> > > "PartitionAssignor#Assignment"
> >> > > > > does
> >> > > > > > > not.
> >> > > > > > > > > > And
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > rationale seems to be assuming that we should
> >> > > always be
> >> > > > > > > able to
> >> > > > > > > > > > do
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > assignment at the broker-side assignor without
> >> > > errors.
> >> > > > > > > > > > Personally I
> >> > > > > > > > > > > > > think
> >> > > > > > > > > > > > > > > it's still potentially beneficial to add the
> >> Error
> >> > > > > field
> >> > > > > > > even
> >> > > > > > > > > for
> >> > > > > > > > > > > > > > > broker-side assignors, e.g. for some edge
> >> cases
> >> > > where
> >> > > > > some
> >> > > > > > > > > > subscribed
> >> > > > > > > > > > > > > > > topics are not recognized with the current
> >> broker's
> >> > > > > > > metadata.
> >> > > > > > > > > > What do
> >> > > > > > > > > > > > > you
> >> > > > > > > > > > > > > > > think?
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > Yeah, that seems reasonable. However, I wonder
> >> if it
> >> > > > > would be
> >> > > > > > > > > > better
> >> > > > > > > > > > > > > > to use an exception on the server side. This is
> >> what
> >> > > we
> >> > > > > > > usually
> >> > > > > > > > > do
> >> > > > > > > > > > for
> >> > > > > > > > > > > > > > server side plugins. On the client side, we use
> >> a
> >> > > field
> >> > > > > > > because
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > errors are not defined in advance.
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > Your comment also makes me think about what we
> >> > > should do
> >> > > > > > > when the
> >> > > > > > > > > > > > > > server side assignor fails. I suppose that we
> >> have to
> >> > > > > keep
> >> > > > > > > the
> >> > > > > > > > > > current
> >> > > > > > > > > > > > > > assignment until a new event occurs. For
> >> instance, in
> >> > > > > your
> >> > > > > > > > > example,
> >> > > > > > > > > > > > > > the coordinator would have to trigger a
> >> rebalance
> >> > > when
> >> > > > > > > > > unrecognized
> >> > > > > > > > > > > > > > topics are available. This would be part of the
> >> > > metadata
> >> > > > > > > > > > monitoring.
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> >> > > additional
> >> > > > > > > fields
> >> > > > > > > > > > > > > > > reason/version/metadata compared with the
> >> > > broker-side
> >> > > > > > > > > > GroupMember. I
> >> > > > > > > > > > > > > > agree
> >> > > > > > > > > > > > > > > that broker-side assignor would not need
> >> > > > > reason/metadata
> >> > > > > > > since
> >> > > > > > > > > > they
> >> > > > > > > > > > > > are
> >> > > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but
> >> what
> >> > > about
> >> > > > > > > version?
> >> > > > > > > > > > E.g.
> >> > > > > > > > > > > > is
> >> > > > > > > > > > > > > > it
> >> > > > > > > > > > > > > > > possible that we evolve our broker-side
> >> built-in
> >> > > > > assignor
> >> > > > > > > but
> >> > > > > > > > > > the old
> >> > > > > > > > > > > > > > > versioned clients would not be able to work
> >> with
> >> > > the
> >> > > > > new
> >> > > > > > > > > > version, in
> >> > > > > > > > > > > > > > which
> >> > > > > > > > > > > > > > > case we need to let the broker being aware of
> >> this
> >> > > and
> >> > > > > > > upgrade
> >> > > > > > > > > > its
> >> > > > > > > > > > > > > > behavior
> >> > > > > > > > > > > > > > > to cooperate with the clients?
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > I don't really see the benefits here because
> >> server
> >> > > side
> >> > > > > > > > > assignors
> >> > > > > > > > > > > > > > don't have metadata at all. They only assign
> >> > > > > > > topic-partitions.
> >> > > > > > > > > They
> >> > > > > > > > > > > > > > are not supposed to generate metadata nor to
> >> receive
> >> > > > > metadata
> >> > > > > > > > > from
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > members.
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > 3) Also related to 2) above, for the
> >> client-side
> >> > > > > > > "GroupMember",
> >> > > > > > > > > > > > instead
> >> > > > > > > > > > > > > > of
> >> > > > > > > > > > > > > > > including these three fields, what about just
> >> > > adding
> >> > > > > the
> >> > > > > > > > > > "Metadata"
> >> > > > > > > > > > > > > field
> >> > > > > > > > > > > > > > > class which has these three fields? Also,
> >> there
> >> > > are two
> >> > > > > > > > > > "Metadata"
> >> > > > > > > > > > > > > > > currently in the APIs, the first is a class
> >> that
> >> > > > > encodes
> >> > > > > > > > > > > > > > > reason/version/metadata, and the second is
> >> just the
> >> > > > > encoded
> >> > > > > > > > > > metadata
> >> > > > > > > > > > > > > > bytes.
> >> > > > > > > > > > > > > > > I'm wondering what about just naming the
> >> first as
> >> > > > > > > > > memberMetadata,
> >> > > > > > > > > > > > which
> >> > > > > > > > > > > > > > > then has a bytebuffer field Metadata, or
> >> instead
> >> > > > > naming the
> >> > > > > > > > > > second
> >> > > > > > > > > > > > > > > bytebuffer field as metadataBytes?
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > That's a good point. Let me try to rationalize
> >> this
> >> > > > > interface
> >> > > > > > > > > > based on
> >> > > > > > > > > > > > > > your suggestions.
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > Best,
> >> > > > > > > > > > > > > > David
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> >> > > > > > > > > wangguoz@gmail.com>
> >> > > > > > > > > > > > > wrote:
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > Hello David,
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > Just had a few more nit questions about the
> >> public
> >> > > > > APIs:
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > 1) The client-side
> >> "PartitionAssignor#Assignment"
> >> > > has
> >> > > > > an
> >> > > > > > > Error
> >> > > > > > > > > > byte
> >> > > > > > > > > > > > > > field,
> >> > > > > > > > > > > > > > > while the broker-side
> >> > > "PartitionAssignor#Assignment"
> >> > > > > does
> >> > > > > > > not.
> >> > > > > > > > > > And
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > rationale seems to be assuming that we should
> >> > > always be
> >> > > > > > > able to
> >> > > > > > > > > > do
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > assignment at the broker-side assignor without
> >> > > errors.
> >> > > > > > > > > > Personally I
> >> > > > > > > > > > > > > think
> >> > > > > > > > > > > > > > > it's still potentially beneficial to add the
> >> Error
> >> > > > > field
> >> > > > > > > even
> >> > > > > > > > > for
> >> > > > > > > > > > > > > > > broker-side assignors, e.g. for some edge
> >> cases
> >> > > where
> >> > > > > some
> >> > > > > > > > > > subscribed
> >> > > > > > > > > > > > > > > topics are not recognized with the current
> >> broker's
> >> > > > > > > metadata.
> >> > > > > > > > > > What do
> >> > > > > > > > > > > > > you
> >> > > > > > > > > > > > > > > think?
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> >> > > additional
> >> > > > > > > fields
> >> > > > > > > > > > > > > > > reason/version/metadata compared with the
> >> > > broker-side
> >> > > > > > > > > > GroupMember. I
> >> > > > > > > > > > > > > > agree
> >> > > > > > > > > > > > > > > that broker-side assignor would not need
> >> > > > > reason/metadata
> >> > > > > > > since
> >> > > > > > > > > > they
> >> > > > > > > > > > > > are
> >> > > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but
> >> what
> >> > > about
> >> > > > > > > version?
> >> > > > > > > > > > E.g.
> >> > > > > > > > > > > > is
> >> > > > > > > > > > > > > > it
> >> > > > > > > > > > > > > > > possible that we evolve our broker-side
> >> built-in
> >> > > > > assignor
> >> > > > > > > but
> >> > > > > > > > > > the old
> >> > > > > > > > > > > > > > > versioned clients would not be able to work
> >> with
> >> > > the
> >> > > > > new
> >> > > > > > > > > > version, in
> >> > > > > > > > > > > > > > which
> >> > > > > > > > > > > > > > > case we need to let the broker being aware of
> >> this
> >> > > and
> >> > > > > > > upgrade
> >> > > > > > > > > > its
> >> > > > > > > > > > > > > > behavior
> >> > > > > > > > > > > > > > > to cooperate with the clients?
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > 3) Also related to 2) above, for the
> >> client-side
> >> > > > > > > "GroupMember",
> >> > > > > > > > > > > > instead
> >> > > > > > > > > > > > > > of
> >> > > > > > > > > > > > > > > including these three fields, what about just
> >> > > adding
> >> > > > > the
> >> > > > > > > > > > "Metadata"
> >> > > > > > > > > > > > > field
> >> > > > > > > > > > > > > > > class which has these three fields? Also,
> >> there
> >> > > are two
> >> > > > > > > > > > "Metadata"
> >> > > > > > > > > > > > > > > currently in the APIs, the first is a class
> >> that
> >> > > > > encodes
> >> > > > > > > > > > > > > > > reason/version/metadata, and the second is
> >> just the
> >> > > > > encoded
> >> > > > > > > > > > metadata
> >> > > > > > > > > > > > > > bytes.
> >> > > > > > > > > > > > > > > I'm wondering what about just naming the
> >> first as
> >> > > > > > > > > memberMetadata,
> >> > > > > > > > > > > > which
> >> > > > > > > > > > > > > > > then has a bytebuffer field Metadata, or
> >> instead
> >> > > > > naming the
> >> > > > > > > > > > second
> >> > > > > > > > > > > > > > > bytebuffer field as metadataBytes?
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > Guozhang
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang
> >> Wang <
> >> > > > > > > > > > wangguoz@gmail.com>
> >> > > > > > > > > > > > > > wrote:
> >> > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > Hello David,
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > Thanks for bringing this question up. I
> >> think the
> >> > > > > main
> >> > > > > > > > > > benefits as
> >> > > > > > > > > > > > > you
> >> > > > > > > > > > > > > > > > listed is 2) above if it stays; just to
> >> clarify,
> >> > > we
> >> > > > > would
> >> > > > > > > > > only
> >> > > > > > > > > > be
> >> > > > > > > > > > > > > able
> >> > > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > > save one round trip if the rebalance is
> >> still
> >> > > > > triggered
> >> > > > > > > by
> >> > > > > > > > > the
> >> > > > > > > > > > > > > broker;
> >> > > > > > > > > > > > > > if
> >> > > > > > > > > > > > > > > > the rebalance is triggered by the client
> >> then the
> >> > > > > > > > > > num.round.trips
> >> > > > > > > > > > > > are
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > > same:
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > 1) With GroupPrepareAssignment:
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > T0: client decides to do a new assignment,
> >> > > suppose
> >> > > > > it has
> >> > > > > > > > > > already
> >> > > > > > > > > > > > > sent
> >> > > > > > > > > > > > > > a
> >> > > > > > > > > > > > > > > > HB and hence has to wait for it to return
> >> first
> >> > > since
> >> > > > > > > only
> >> > > > > > > > > one
> >> > > > > > > > > > > > > request
> >> > > > > > > > > > > > > > /
> >> > > > > > > > > > > > > > > > response can be inflight with the
> >> coordinator's
> >> > > > > socket.
> >> > > > > > > > > > > > > > > > T1: client receives the HB response, and
> >> then
> >> > > sends
> >> > > > > the
> >> > > > > > > > > > > > > > > > GroupPrepareAssignment request.
> >> > > > > > > > > > > > > > > > T2: the GroupPrepareAssignment response is
> >> > > returned.
> >> > > > > > > > > > > > > > > > T3: it calculates the new assignment, and
> >> sends a
> >> > > > > > > > > > > > > > GroupInstallAssignment
> >> > > > > > > > > > > > > > > > request.
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > In total, two round trips.
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > 2) Without GroupPrepareAssignment:
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > T0: client decides to do a new assignment,
> >> > > suppose
> >> > > > > it has
> >> > > > > > > > > > already
> >> > > > > > > > > > > > > sent
> >> > > > > > > > > > > > > > a
> >> > > > > > > > > > > > > > > > HB and hence has to wait for it to return
> >> first
> >> > > since
> >> > > > > > > only
> >> > > > > > > > > one
> >> > > > > > > > > > > > > request
> >> > > > > > > > > > > > > > /
> >> > > > > > > > > > > > > > > > response can be inflight with the
> >> coordinator's
> >> > > > > socket.
> >> > > > > > > > > > > > > > > > T1: client receives the HB response, and
> >> then
> >> > > sends
> >> > > > > the
> >> > > > > > > new
> >> > > > > > > > > HB
> >> > > > > > > > > > > > > request
> >> > > > > > > > > > > > > > > > with the flag indicating a new rebalance
> >> needed..
> >> > > > > > > > > > > > > > > > T2: the HB response with the optional member
> >> > > metadata
> >> > > > > > > map is
> >> > > > > > > > > > > > > returned.
> >> > > > > > > > > > > > > > > > T3: it calculates the new assignment, and
> >> sends a
> >> > > > > > > > > > > > > > GroupInstallAssignment
> >> > > > > > > > > > > > > > > > request.
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > In total, two round trips as well.
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > -----------------------------
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > So to complete the full picture here, we'd
> >> need
> >> > > to
> >> > > > > modify
> >> > > > > > > > > both
> >> > > > > > > > > > HB
> >> > > > > > > > > > > > > > request
> >> > > > > > > > > > > > > > > > and response so that the client can also
> >> > > indicate a
> >> > > > > new
> >> > > > > > > > > > rebalance
> >> > > > > > > > > > > > via
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > > HB request as well, right?
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > Assuming all above is true, I think it's
> >> okay to
> >> > > > > merge
> >> > > > > > > the
> >> > > > > > > > > > > > > > > > GroupPrepareAssignment into HB given that
> >> we can
> >> > > > > make the
> >> > > > > > > > > > > > additional
> >> > > > > > > > > > > > > > fields
> >> > > > > > > > > > > > > > > > encoding the full member (subscription)
> >> metadata
> >> > > and
> >> > > > > > > topic
> >> > > > > > > > > > metadata
> >> > > > > > > > > > > > > as
> >> > > > > > > > > > > > > > > > optional fields.
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > Guozhang
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> >> > > > > > > > > > > > > > <dj...@confluent.io.invalid>
> >> > > > > > > > > > > > > > > > wrote:
> >> > > > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> Hi all,
> >> > > > > > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> During an offline conversation, someone
> >> asked
> >> > > why we
> >> > > > > > > need
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and
> >> suggested
> >> > > > > that we
> >> > > > > > > > > could
> >> > > > > > > > > > > > > instead
> >> > > > > > > > > > > > > > > >> provide the group state in the heartbeat
> >> > > response.
> >> > > > > This
> >> > > > > > > has
> >> > > > > > > > > a
> >> > > > > > > > > > few
> >> > > > > > > > > > > > > > > >> advantages: 1) it does not require using a
> >> > > special
> >> > > > > error
> >> > > > > > > > > code
> >> > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> signal that a new assignment is required
> >> as the
> >> > > > > signal
> >> > > > > > > would
> >> > > > > > > > > > be
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> provided group state; 2) it removes one
> >> round
> >> > > trip
> >> > > > > when
> >> > > > > > > a
> >> > > > > > > > > > client
> >> > > > > > > > > > > > > side
> >> > > > > > > > > > > > > > > >> assignor is used. The downside is that it
> >> makes
> >> > > the
> >> > > > > > > > > heartbeat
> >> > > > > > > > > > > > > > > >> response's definition quite large. I recall
> >> > > that I
> >> > > > > went
> >> > > > > > > with
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> current approach due to this.
> >> > > > > > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> Providing the group state in the heartbeat
> >> > > response
> >> > > > > is
> >> > > > > > > > > > appealing.
> >> > > > > > > > > > > > > What
> >> > > > > > > > > > > > > > > >> do you guys think?
> >> > > > > > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> Best,
> >> > > > > > > > > > > > > > > >> David
> >> > > > > > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David
> >> Jacot <
> >> > > > > > > > > > djacot@confluent.io>
> >> > > > > > > > > > > > > > wrote:
> >> > > > > > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > Hi Guozhang,
> >> > > > > > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > 1. I have added a reference to the
> >> relevant
> >> > > > > chapter
> >> > > > > > > > > instead
> >> > > > > > > > > > of
> >> > > > > > > > > > > > > > > >> > repeating the whole thing. Does that
> >> work for
> >> > > you?
> >> > > > > > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you
> >> are
> >> > > > > referring
> >> > > > > > > to
> >> > > > > > > > > is
> >> > > > > > > > > > > > about
> >> > > > > > > > > > > > > > when
> >> > > > > > > > > > > > > > > >> > a rebalance should be triggered for the
> >> > > > > non-upgraded
> >> > > > > > > > > members
> >> > > > > > > > > > > > using
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > old protocol. The section mentions that a
> >> > > > > rebalance
> >> > > > > > > must
> >> > > > > > > > > be
> >> > > > > > > > > > > > > > triggered
> >> > > > > > > > > > > > > > > >> > when a new assignment is installed. This
> >> > > implies
> >> > > > > that
> >> > > > > > > the
> >> > > > > > > > > > group
> >> > > > > > > > > > > > > > epoch
> >> > > > > > > > > > > > > > > >> > was updated either by a native member or
> >> a
> >> > > > > > > non-upgraded
> >> > > > > > > > > > member.
> >> > > > > > > > > > > > > For
> >> > > > > > > > > > > > > > > >> > the latter, the JoinGroup request would
> >> be the
> >> > > > > > > trigger. I
> >> > > > > > > > > > have
> >> > > > > > > > > > > > > > added a
> >> > > > > > > > > > > > > > > >> > reference to the relevant chapter in the
> >> > > > > "JoinGroup
> >> > > > > > > > > > Handling"
> >> > > > > > > > > > > > > > section
> >> > > > > > > > > > > > > > > >> > as well. Does that make sense?
> >> > > > > > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > Thanks,
> >> > > > > > > > > > > > > > > >> > David
> >> > > > > > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang
> >> Wang
> >> > > <
> >> > > > > > > > > > > > wangguoz@gmail.com
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> wrote:
> >> > > > > > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > Hello David,
> >> > > > > > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > Alright I think that's sufficient.
> >> Just to
> >> > > make
> >> > > > > that
> >> > > > > > > > > > clear in
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > doc,
> >> > > > > > > > > > > > > > > >> > > could we update:
> >> > > > > > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > 1) the heartbeat request handling
> >> section,
> >> > > > > stating
> >> > > > > > > when
> >> > > > > > > > > > > > > > coordinator
> >> > > > > > > > > > > > > > > >> will
> >> > > > > > > > > > > > > > > >> > > trigger rebalance based on the HB's
> >> member
> >> > > > > metadata
> >> > > > > > > /
> >> > > > > > > > > > reason?
> >> > > > > > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to
> >> > > include
> >> > > > > what
> >> > > > > > > we
> >> > > > > > > > > > > > described
> >> > > > > > > > > > > > > > in
> >> > > > > > > > > > > > > > > >> "Group
> >> > > > > > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as
> >> > > well?
> >> > > > > > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > Guozhang
> >> > > > > > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David
> >> Jacot
> >> > > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> >> > > > > > > > > > > > > > > >> > > wrote:
> >> > > > > > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > Hi Guozhang,
> >> > > > > > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > I thought that the assignor will
> >> always be
> >> > > > > > > consulted
> >> > > > > > > > > > when
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > next
> >> > > > > > > > > > > > > > > >> > > > heartbeat request is constructed. In
> >> other
> >> > > > > words,
> >> > > > > > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be
> >> > > called
> >> > > > > for
> >> > > > > > > every
> >> > > > > > > > > > > > > heartbeat.
> >> > > > > > > > > > > > > > > >> This
> >> > > > > > > > > > > > > > > >> > > > gives the opportunity for the
> >> assignor to
> >> > > > > enforce
> >> > > > > > > a
> >> > > > > > > > > > > > rebalance
> >> > > > > > > > > > > > > by
> >> > > > > > > > > > > > > > > >> > > > setting the reason to a non-zero
> >> value or
> >> > > by
> >> > > > > > > changing
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > bytes. Do
> >> > > > > > > > > > > > > > > >> > > > you think that this is not
> >> sufficient?
> >> > > Are you
> >> > > > > > > > > > concerned by
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> delay?
> >> > > > > > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > Best,
> >> > > > > > > > > > > > > > > >> > > > David
> >> > > > > > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM
> >> Guozhang
> >> > > Wang <
> >> > > > > > > > > > > > > > wangguoz@gmail.com>
> >> > > > > > > > > > > > > > > >> wrote:
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > Hello David,
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > One of Jun's comments make me
> >> thinking:
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > ```
> >> > > > > > > > > > > > > > > >> > > > > In this case, a new assignment is
> >> > > triggered
> >> > > > > by
> >> > > > > > > the
> >> > > > > > > > > > client
> >> > > > > > > > > > > > > side
> >> > > > > > > > > > > > > > > >> > > > > assignor. When constructing the
> >> HB, the
> >> > > > > consumer
> >> > > > > > > > > will
> >> > > > > > > > > > > > always
> >> > > > > > > > > > > > > > > >> consult
> >> > > > > > > > > > > > > > > >> > > > > the client side assignor and
> >> propagate
> >> > > the
> >> > > > > > > > > > information to
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > coordinator. In other words, we
> >> don't
> >> > > expect
> >> > > > > > > users
> >> > > > > > > > > to
> >> > > > > > > > > > call
> >> > > > > > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> >> > > > > > > > > > > > > > > >> > > > > ```
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > As I looked at the current
> >> > > > > PartitionAssignor's
> >> > > > > > > > > > interface,
> >> > > > > > > > > > > > we
> >> > > > > > > > > > > > > > > >> actually do
> >> > > > > > > > > > > > > > > >> > > > > not have a way yet to instruct how
> >> to
> >> > > > > construct
> >> > > > > > > the
> >> > > > > > > > > > next
> >> > > > > > > > > > > > HB
> >> > > > > > > > > > > > > > > >> request, e.g.
> >> > > > > > > > > > > > > > > >> > > > > when the assignor wants to enforce
> >> a new
> >> > > > > > > rebalance
> >> > > > > > > > > > with a
> >> > > > > > > > > > > > > new
> >> > > > > > > > > > > > > > > >> assignment,
> >> > > > > > > > > > > > > > > >> > > > > we'd need some customizable APIs
> >> inside
> >> > > the
> >> > > > > > > > > > > > > PartitionAssignor
> >> > > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> indicate
> >> > > > > > > > > > > > > > > >> > > > > the next HB telling broker about
> >> so.
> >> > > WDYT
> >> > > > > about
> >> > > > > > > > > adding
> >> > > > > > > > > > > > such
> >> > > > > > > > > > > > > an
> >> > > > > > > > > > > > > > > >> API on the
> >> > > > > > > > > > > > > > > >> > > > > PartitionAssignor?
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > Guozhang
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM
> >> David
> >> > > Jacot
> >> > > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> >> > > > > > > > > > > > > > > >> > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > Hi Jun,
> >> > > > > > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > I have updated the KIP to
> >> include your
> >> > > > > > > feedback. I
> >> > > > > > > > > > have
> >> > > > > > > > > > > > > also
> >> > > > > > > > > > > > > > > >> tried to
> >> > > > > > > > > > > > > > > >> > > > > > clarify the parts which were not
> >> > > cleared.
> >> > > > > > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > Best,
> >> > > > > > > > > > > > > > > >> > > > > > David
> >> > > > > > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM
> >> David
> >> > > > > Jacot <
> >> > > > > > > > > > > > > > djacot@confluent.io
> >> > > > > > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Hi Jun,
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let
> >> me
> >> > > start
> >> > > > > by
> >> > > > > > > > > > answering
> >> > > > > > > > > > > > your
> >> > > > > > > > > > > > > > > >> questions
> >> > > > > > > > > > > > > > > >> > > > > > > inline and I will update the
> >> KIP
> >> > > next
> >> > > > > week.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall,
> >> the
> >> > > main
> >> > > > > > > benefits
> >> > > > > > > > > > of
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > KIP
> >> > > > > > > > > > > > > > > >> seem to
> >> > > > > > > > > > > > > > > >> > > > be
> >> > > > > > > > > > > > > > > >> > > > > > fewer
> >> > > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and
> >> more
> >> > > > > efficient
> >> > > > > > > > > > support of
> >> > > > > > > > > > > > > > > >> wildcard. A few
> >> > > > > > > > > > > > > > > >> > > > > > > > comments below.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > I would also add that the KIP
> >> > > removes
> >> > > > > the
> >> > > > > > > global
> >> > > > > > > > > > sync
> >> > > > > > > > > > > > > > barrier
> >> > > > > > > > > > > > > > > >> in the
> >> > > > > > > > > > > > > > > >> > > > > > > protocol which is essential to
> >> > > improve
> >> > > > > group
> >> > > > > > > > > > stability
> >> > > > > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> > > > > > > scalability, and the KIP also
> >> > > > > simplifies the
> >> > > > > > > > > > client by
> >> > > > > > > > > > > > > > moving
> >> > > > > > > > > > > > > > > >> most of
> >> > > > > > > > > > > > > > > >> > > > > > > the logic to the server side.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 30.
> >> ConsumerGroupHeartbeatRequest
> >> > > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> >> > > singleton.
> >> > > > > Do we
> >> > > > > > > plan
> >> > > > > > > > > > to
> >> > > > > > > > > > > > > > support
> >> > > > > > > > > > > > > > > >> rolling
> >> > > > > > > > > > > > > > > >> > > > > > changing
> >> > > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in
> >> the
> >> > > > > > > consumers?
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Definitely. The group
> >> coordinator
> >> > > will
> >> > > > > use
> >> > > > > > > the
> >> > > > > > > > > > > > assignor
> >> > > > > > > > > > > > > > used
> >> > > > > > > > > > > > > > > >> by a
> >> > > > > > > > > > > > > > > >> > > > > > > majority of the members. This
> >> > > allows the
> >> > > > > > > group
> >> > > > > > > > > to
> >> > > > > > > > > > move
> >> > > > > > > > > > > > > > from
> >> > > > > > > > > > > > > > > >> one
> >> > > > > > > > > > > > > > > >> > > > > > > assignor to another by a roll.
> >> This
> >> > > is
> >> > > > > > > explained
> >> > > > > > > > > > in
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> Assignor
> >> > > > > > > > > > > > > > > >> > > > > > > Selection chapter.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could
> >> you
> >> > > explain
> >> > > > > > > whether
> >> > > > > > > > > > it's
> >> > > > > > > > > > > > > > > >> required in
> >> > > > > > > > > > > > > > > >> > > > every
> >> > > > > > > > > > > > > > > >> > > > > > > > request or the scenarios
> >> when it
> >> > > > > needs to
> >> > > > > > > be
> >> > > > > > > > > > filled?
> >> > > > > > > > > > > > > For
> >> > > > > > > > > > > > > > > >> example,
> >> > > > > > > > > > > > > > > >> > > > it's
> >> > > > > > > > > > > > > > > >> > > > > > not
> >> > > > > > > > > > > > > > > >> > > > > > > > clear to me when
> >> TopicPartitions
> >> > > > > needs to
> >> > > > > > > be
> >> > > > > > > > > > filled.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > The client is expected to set
> >> those
> >> > > > > fields
> >> > > > > > > in
> >> > > > > > > > > > case of
> >> > > > > > > > > > > > a
> >> > > > > > > > > > > > > > > >> connection
> >> > > > > > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when
> >> the
> >> > > fields
> >> > > > > have
> >> > > > > > > > > > changed
> >> > > > > > > > > > > > > since
> >> > > > > > > > > > > > > > > >> the last
> >> > > > > > > > > > > > > > > >> > > > > > > HB. The server populates those
> >> > > fields as
> >> > > > > > > long as
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > member
> >> > > > > > > > > > > > > > > >> is not
> >> > > > > > > > > > > > > > > >> > > > > > > fully reconciled - the member
> >> should
> >> > > > > > > acknowledge
> >> > > > > > > > > > that
> >> > > > > > > > > > > > it
> >> > > > > > > > > > > > > > has
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > expected epoch and assignment.
> >> I
> >> > > will
> >> > > > > > > clarify
> >> > > > > > > > > > this in
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > KIP.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> >> > > protocol,
> >> > > > > the
> >> > > > > > > rack
> >> > > > > > > > > > > > > affinity
> >> > > > > > > > > > > > > > > >> between the
> >> > > > > > > > > > > > > > > >> > > > > > client
> >> > > > > > > > > > > > > > > >> > > > > > > > and the broker is only
> >> considered
> >> > > > > during
> >> > > > > > > > > > fetching,
> >> > > > > > > > > > > > but
> >> > > > > > > > > > > > > > not
> >> > > > > > > > > > > > > > > >> during
> >> > > > > > > > > > > > > > > >> > > > > > assigning
> >> > > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> >> > > Sometimes,
> >> > > > > once
> >> > > > > > > the
> >> > > > > > > > > > > > > assignment
> >> > > > > > > > > > > > > > is
> >> > > > > > > > > > > > > > > >> made,
> >> > > > > > > > > > > > > > > >> > > > there
> >> > > > > > > > > > > > > > > >> > > > > > is
> >> > > > > > > > > > > > > > > >> > > > > > > > no opportunity for read
> >> affinity
> >> > > > > because
> >> > > > > > > no
> >> > > > > > > > > > replicas
> >> > > > > > > > > > > > > of
> >> > > > > > > > > > > > > > > >> assigned
> >> > > > > > > > > > > > > > > >> > > > > > partitions
> >> > > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> >> > > > > wondering
> >> > > > > > > if we
> >> > > > > > > > > > should
> >> > > > > > > > > > > > > use
> >> > > > > > > > > > > > > > > >> this
> >> > > > > > > > > > > > > > > >> > > > > > opportunity
> >> > > > > > > > > > > > > > > >> > > > > > > > to address this by including
> >> rack
> >> > > in
> >> > > > > > > > > > GroupMember.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > That's an interesting idea. I
> >> don't
> >> > > see
> >> > > > > any
> >> > > > > > > > > issue
> >> > > > > > > > > > with
> >> > > > > > > > > > > > > > adding
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > rack
> >> > > > > > > > > > > > > > > >> > > > > > > to the members. I will do so.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side,
> >> often,
> >> > > it's
> >> > > > > > > useful to
> >> > > > > > > > > > know
> >> > > > > > > > > > > > how
> >> > > > > > > > > > > > > > busy
> >> > > > > > > > > > > > > > > >> a group
> >> > > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> >> > > event
> >> > > > > loop
> >> > > > > > > > > model,
> >> > > > > > > > > > it
> >> > > > > > > > > > > > > seems
> >> > > > > > > > > > > > > > > >> that we
> >> > > > > > > > > > > > > > > >> > > > could
> >> > > > > > > > > > > > > > > >> > > > > > add
> >> > > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the
> >> fraction
> >> > > of
> >> > > > > the
> >> > > > > > > time
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > event
> >> > > > > > > > > > > > > > > >> loop is
> >> > > > > > > > > > > > > > > >> > > > doing
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > actual work.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > That's a great idea. I will
> >> add it.
> >> > > > > Thanks.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> >> > > > > coordinator
> >> > > > > > > > > > failover
> >> > > > > > > > > > > > > > > >> handling? For
> >> > > > > > > > > > > > > > > >> > > > > > example,
> >> > > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the
> >> check
> >> > > if
> >> > > > > any
> >> > > > > > > group
> >> > > > > > > > > > with
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> wildcard
> >> > > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> >> > > matching
> >> > > > > topic?
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Sure. When the new group
> >> coordinator
> >> > > > > takes
> >> > > > > > > over,
> >> > > > > > > > > > it
> >> > > > > > > > > > > > has
> >> > > > > > > > > > > > > > to:
> >> > > > > > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> >> > > > > > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a
> >> > > client
> >> > > > > side
> >> > > > > > > > > > assignor
> >> > > > > > > > > > > > is
> >> > > > > > > > > > > > > > used.
> >> > > > > > > > > > > > > > > >> We
> >> > > > > > > > > > > > > > > >> > > > don't
> >> > > > > > > > > > > > > > > >> > > > > > > store the information about the
> >> > > member
> >> > > > > > > selected
> >> > > > > > > > > > to run
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> assignment
> >> > > > > > > > > > > > > > > >> > > > > > > so we have to start a new one.
> >> > > > > > > > > > > > > > > >> > > > > > > * Update the topics metadata,
> >> > > verify the
> >> > > > > > > > > wildcard
> >> > > > > > > > > > > > > > > >> subscriptions, and
> >> > > > > > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 34.
> >> ConsumerGroupMetadataValue,
> >> > > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> ConsumerGroupMemberMetadataValue:
> >> > > > > Could we
> >> > > > > > > > > > document
> >> > > > > > > > > > > > > what
> >> > > > > > > > > > > > > > > >> the epoch
> >> > > > > > > > > > > > > > > >> > > > > > field
> >> > > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does
> >> the
> >> > > epoch
> >> > > > > in
> >> > > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> >> > > > > > > > > > > > > > > >> > > > > > reflect
> >> > > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> >> > > about the
> >> > > > > > > one in
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > ConsumerGroupPartitionMetadataValue
> >> > > > > and
> >> > > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Sure. I will clarify that but
> >> it is
> >> > > > > always
> >> > > > > > > the
> >> > > > > > > > > > latest
> >> > > > > > > > > > > > > > group
> >> > > > > > > > > > > > > > > >> epoch.
> >> > > > > > > > > > > > > > > >> > > > > > > When the group state is
> >> updated, the
> >> > > > > group
> >> > > > > > > epoch
> >> > > > > > > > > > is
> >> > > > > > > > > > > > > > bumped so
> >> > > > > > > > > > > > > > > >> we use
> >> > > > > > > > > > > > > > > >> > > > > > > that one for all the change
> >> records
> >> > > > > related
> >> > > > > > > to
> >> > > > > > > > > the
> >> > > > > > > > > > > > > update.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator
> >> will
> >> > > ensure
> >> > > > > > > that
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > following
> >> > > > > > > > > > > > > > > >> > > > invariants
> >> > > > > > > > > > > > > > > >> > > > > > are
> >> > > > > > > > > > > > > > > >> > > > > > > > met: ... All members
> >> exists." It's
> >> > > > > > > possible
> >> > > > > > > > > for
> >> > > > > > > > > > a
> >> > > > > > > > > > > > > member
> >> > > > > > > > > > > > > > > >> not to
> >> > > > > > > > > > > > > > > >> > > > get any
> >> > > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > That's right. Here I meant
> >> that the
> >> > > > > members
> >> > > > > > > > > > provided
> >> > > > > > > > > > > > by
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> assignor
> >> > > > > > > > > > > > > > > >> > > > > > > in the assignment must exist
> >> in the
> >> > > > > group.
> >> > > > > > > The
> >> > > > > > > > > > > > assignor
> >> > > > > > > > > > > > > > can
> >> > > > > > > > > > > > > > > >> not make
> >> > > > > > > > > > > > > > > >> > > > > > > up new member ids.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> >> > > with a
> >> > > > > > > member
> >> > > > > > > > > > epoch
> >> > > > > > > > > > > > > > equals to
> >> > > > > > > > > > > > > > > >> 0":
> >> > > > > > > > > > > > > > > >> > > > When
> >> > > > > > > > > > > > > > > >> > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what
> >> member
> >> > > id
> >> > > > > > > would be
> >> > > > > > > > > > used?
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > A member is expected to
> >> abandon all
> >> > > its
> >> > > > > > > > > > partitions and
> >> > > > > > > > > > > > > > > >> rejoins when
> >> > > > > > > > > > > > > > > >> > > > it
> >> > > > > > > > > > > > > > > >> > > > > > > receives the
> >> FENCED_MEMBER_EPOCH
> >> > > error.
> >> > > > > In
> >> > > > > > > this
> >> > > > > > > > > > case,
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > > coordinator will have removed
> >> the
> >> > > member
> >> > > > > > > from
> >> > > > > > > > > the
> >> > > > > > > > > > > > group.
> >> > > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> member
> >> > > > > > > > > > > > > > > >> > > > > > > can rejoin the group with the
> >> same
> >> > > > > member
> >> > > > > > > id but
> >> > > > > > > > > > with
> >> > > > > > > > > > > > 0
> >> > > > > > > > > > > > > as
> >> > > > > > > > > > > > > > > >> epoch. Let
> >> > > > > > > > > > > > > > > >> > > > > > > me see if I can clarify this
> >> in the
> >> > > KIP.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users
> >> will
> >> > > have
> >> > > > > the
> >> > > > > > > > > ability
> >> > > > > > > > > > to
> >> > > > > > > > > > > > > > trigger a
> >> > > > > > > > > > > > > > > >> > > > > > reassignment
> >> > > > > > > > > > > > > > > >> > > > > > > > by either providing a
> >> non-zero
> >> > > reason
> >> > > > > or
> >> > > > > > > by
> >> > > > > > > > > > updating
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems
> >> to be
> >> > > > > > > conflicting
> >> > > > > > > > > > with
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> deprecation
> >> > > > > > > > > > > > > > > >> > > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > In this case, a new assignment
> >> is
> >> > > > > triggered
> >> > > > > > > by
> >> > > > > > > > > the
> >> > > > > > > > > > > > > client
> >> > > > > > > > > > > > > > side
> >> > > > > > > > > > > > > > > >> > > > > > > assignor. When constructing
> >> the HB,
> >> > > the
> >> > > > > > > consumer
> >> > > > > > > > > > will
> >> > > > > > > > > > > > > > always
> >> > > > > > > > > > > > > > > >> consult
> >> > > > > > > > > > > > > > > >> > > > > > > the client side assignor and
> >> > > propagate
> >> > > > > the
> >> > > > > > > > > > information
> >> > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > > coordinator. In other words, we
> >> > > don't
> >> > > > > expect
> >> > > > > > > > > > users to
> >> > > > > > > > > > > > > call
> >> > > > > > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance
> >> anymore.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment
> >> examples are
> >> > > > > nice.
> >> > > > > > > But
> >> > > > > > > > > the
> >> > > > > > > > > > > > > section
> >> > > > > > > > > > > > > > > >> seems to
> >> > > > > > > > > > > > > > > >> > > > have
> >> > > > > > > > > > > > > > > >> > > > > > > > multiple typos.
> >> > > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group
> >> transitions to
> >> > > > > epoch
> >> > > > > > > 2, B
> >> > > > > > > > > > > > > > immediately
> >> > > > > > > > > > > > > > > >> gets into
> >> > > > > > > > > > > > > > > >> > > > > > > > "epoch=1,
> >> partitions=[foo-2]",
> >> > > which
> >> > > > > seems
> >> > > > > > > > > > > > incorrect.
> >> > > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group
> >> transitions to
> >> > > > > epoch
> >> > > > > > > 3, C
> >> > > > > > > > > > seems
> >> > > > > > > > > > > > to
> >> > > > > > > > > > > > > > get
> >> > > > > > > > > > > > > > > >> into
> >> > > > > > > > > > > > > > > >> > > > > > epoch=3,
> >> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> >> > > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to
> >> epoch
> >> > > 3, C
> >> > > > > > > still
> >> > > > > > > > > > has A -
> >> > > > > > > > > > > > > > > >> epoch=2,
> >> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Sorry for that! I will revise
> >> them.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of
> >> consumers:
> >> > > Do
> >> > > > > we
> >> > > > > > > > > support
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > upgrade
> >> > > > > > > > > > > > > > > >> from
> >> > > > > > > > > > > > > > > >> > > > any
> >> > > > > > > > > > > > > > > >> > > > > > old
> >> > > > > > > > > > > > > > > >> > > > > > > > version to new one?
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > We will support upgrading from
> >> the
> >> > > > > consumer
> >> > > > > > > > > > protocol
> >> > > > > > > > > > > > > > version
> >> > > > > > > > > > > > > > > >> 3,
> >> > > > > > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792
> >> is
> >> > > not
> >> > > > > > > > > implemented
> >> > > > > > > > > > yet
> >> > > > > > > > > > > > so
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> earliest
> >> > > > > > > > > > > > > > > >> > > > > > > version is unknown at the
> >> moment.
> >> > > This
> >> > > > > is
> >> > > > > > > > > > explained in
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> migration
> >> > > > > > > > > > > > > > > >> > > > > > > plan chapter.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Thanks again for your feedback,
> >> > > Jun. I
> >> > > > > will
> >> > > > > > > > > > update the
> >> > > > > > > > > > > > > KIP
> >> > > > > > > > > > > > > > > >> based on
> >> > > > > > > > > > > > > > > >> > > > it
> >> > > > > > > > > > > > > > > >> > > > > > > next week.
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > Best,
> >> > > > > > > > > > > > > > > >> > > > > > > David
> >> > > > > > > > > > > > > > > >> > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM
> >> Jun
> >> > > Rao
> >> > > > > > > > > > > > > > > >> <ju...@confluent.io.invalid>
> >> > > > > > > > > > > > > > > >> > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > Hi, David,
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall,
> >> the
> >> > > main
> >> > > > > > > benefits
> >> > > > > > > > > > of
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > KIP
> >> > > > > > > > > > > > > > > >> seem to
> >> > > > > > > > > > > > > > > >> > > > be
> >> > > > > > > > > > > > > > > >> > > > > > fewer
> >> > > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and
> >> more
> >> > > > > efficient
> >> > > > > > > > > > support of
> >> > > > > > > > > > > > > > > >> wildcard. A few
> >> > > > > > > > > > > > > > > >> > > > > > > > comments below.
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 30.
> >> ConsumerGroupHeartbeatRequest
> >> > > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> >> > > singleton.
> >> > > > > Do we
> >> > > > > > > plan
> >> > > > > > > > > > to
> >> > > > > > > > > > > > > > support
> >> > > > > > > > > > > > > > > >> rolling
> >> > > > > > > > > > > > > > > >> > > > > > changing
> >> > > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in
> >> the
> >> > > > > > > consumers?
> >> > > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could
> >> you
> >> > > explain
> >> > > > > > > whether
> >> > > > > > > > > > it's
> >> > > > > > > > > > > > > > > >> required in
> >> > > > > > > > > > > > > > > >> > > > every
> >> > > > > > > > > > > > > > > >> > > > > > > > request or the scenarios
> >> when it
> >> > > > > needs to
> >> > > > > > > be
> >> > > > > > > > > > filled?
> >> > > > > > > > > > > > > For
> >> > > > > > > > > > > > > > > >> example,
> >> > > > > > > > > > > > > > > >> > > > it's
> >> > > > > > > > > > > > > > > >> > > > > > not
> >> > > > > > > > > > > > > > > >> > > > > > > > clear to me when
> >> TopicPartitions
> >> > > > > needs to
> >> > > > > > > be
> >> > > > > > > > > > filled.
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> >> > > protocol,
> >> > > > > the
> >> > > > > > > rack
> >> > > > > > > > > > > > > affinity
> >> > > > > > > > > > > > > > > >> between the
> >> > > > > > > > > > > > > > > >> > > > > > client
> >> > > > > > > > > > > > > > > >> > > > > > > > and the broker is only
> >> considered
> >> > > > > during
> >> > > > > > > > > > fetching,
> >> > > > > > > > > > > > but
> >> > > > > > > > > > > > > > not
> >> > > > > > > > > > > > > > > >> during
> >> > > > > > > > > > > > > > > >> > > > > > assigning
> >> > > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> >> > > Sometimes,
> >> > > > > once
> >> > > > > > > the
> >> > > > > > > > > > > > > assignment
> >> > > > > > > > > > > > > > is
> >> > > > > > > > > > > > > > > >> made,
> >> > > > > > > > > > > > > > > >> > > > there
> >> > > > > > > > > > > > > > > >> > > > > > is
> >> > > > > > > > > > > > > > > >> > > > > > > > no opportunity for read
> >> affinity
> >> > > > > because
> >> > > > > > > no
> >> > > > > > > > > > replicas
> >> > > > > > > > > > > > > of
> >> > > > > > > > > > > > > > > >> assigned
> >> > > > > > > > > > > > > > > >> > > > > > partitions
> >> > > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> >> > > > > wondering
> >> > > > > > > if we
> >> > > > > > > > > > should
> >> > > > > > > > > > > > > use
> >> > > > > > > > > > > > > > > >> this
> >> > > > > > > > > > > > > > > >> > > > > > opportunity
> >> > > > > > > > > > > > > > > >> > > > > > > > to address this by including
> >> rack
> >> > > in
> >> > > > > > > > > > GroupMember.
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side,
> >> often,
> >> > > it's
> >> > > > > > > useful to
> >> > > > > > > > > > know
> >> > > > > > > > > > > > how
> >> > > > > > > > > > > > > > busy
> >> > > > > > > > > > > > > > > >> a group
> >> > > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> >> > > event
> >> > > > > loop
> >> > > > > > > > > model,
> >> > > > > > > > > > it
> >> > > > > > > > > > > > > seems
> >> > > > > > > > > > > > > > > >> that we
> >> > > > > > > > > > > > > > > >> > > > could
> >> > > > > > > > > > > > > > > >> > > > > > add
> >> > > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the
> >> fraction
> >> > > of
> >> > > > > the
> >> > > > > > > time
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > event
> >> > > > > > > > > > > > > > > >> loop is
> >> > > > > > > > > > > > > > > >> > > > doing
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > actual work.
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> >> > > > > coordinator
> >> > > > > > > > > > failover
> >> > > > > > > > > > > > > > > >> handling? For
> >> > > > > > > > > > > > > > > >> > > > > > example,
> >> > > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the
> >> check
> >> > > if
> >> > > > > any
> >> > > > > > > group
> >> > > > > > > > > > with
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> wildcard
> >> > > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> >> > > matching
> >> > > > > topic?
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 34.
> >> ConsumerGroupMetadataValue,
> >> > > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> ConsumerGroupMemberMetadataValue:
> >> > > > > Could we
> >> > > > > > > > > > document
> >> > > > > > > > > > > > > what
> >> > > > > > > > > > > > > > > >> the epoch
> >> > > > > > > > > > > > > > > >> > > > > > field
> >> > > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does
> >> the
> >> > > epoch
> >> > > > > in
> >> > > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> >> > > > > > > > > > > > > > > >> > > > > > reflect
> >> > > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> >> > > about the
> >> > > > > > > one in
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > ConsumerGroupPartitionMetadataValue
> >> > > > > and
> >> > > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator
> >> will
> >> > > ensure
> >> > > > > > > that
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > following
> >> > > > > > > > > > > > > > > >> > > > invariants
> >> > > > > > > > > > > > > > > >> > > > > > are
> >> > > > > > > > > > > > > > > >> > > > > > > > met: ... All members
> >> exists." It's
> >> > > > > > > possible
> >> > > > > > > > > for
> >> > > > > > > > > > a
> >> > > > > > > > > > > > > member
> >> > > > > > > > > > > > > > > >> not to
> >> > > > > > > > > > > > > > > >> > > > get any
> >> > > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> >> > > with a
> >> > > > > > > member
> >> > > > > > > > > > epoch
> >> > > > > > > > > > > > > > equals to
> >> > > > > > > > > > > > > > > >> 0":
> >> > > > > > > > > > > > > > > >> > > > When
> >> > > > > > > > > > > > > > > >> > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what
> >> member
> >> > > id
> >> > > > > > > would be
> >> > > > > > > > > > used?
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users
> >> will
> >> > > have
> >> > > > > the
> >> > > > > > > > > ability
> >> > > > > > > > > > to
> >> > > > > > > > > > > > > > trigger a
> >> > > > > > > > > > > > > > > >> > > > > > reassignment
> >> > > > > > > > > > > > > > > >> > > > > > > > by either providing a
> >> non-zero
> >> > > reason
> >> > > > > or
> >> > > > > > > by
> >> > > > > > > > > > updating
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems
> >> to be
> >> > > > > > > conflicting
> >> > > > > > > > > > with
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> deprecation
> >> > > > > > > > > > > > > > > >> > > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment
> >> examples are
> >> > > > > nice.
> >> > > > > > > But
> >> > > > > > > > > the
> >> > > > > > > > > > > > > section
> >> > > > > > > > > > > > > > > >> seems to
> >> > > > > > > > > > > > > > > >> > > > have
> >> > > > > > > > > > > > > > > >> > > > > > > > multiple typos.
> >> > > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group
> >> transitions to
> >> > > > > epoch
> >> > > > > > > 2, B
> >> > > > > > > > > > > > > > immediately
> >> > > > > > > > > > > > > > > >> gets into
> >> > > > > > > > > > > > > > > >> > > > > > > > "epoch=1,
> >> partitions=[foo-2]",
> >> > > which
> >> > > > > seems
> >> > > > > > > > > > > > incorrect.
> >> > > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group
> >> transitions to
> >> > > > > epoch
> >> > > > > > > 3, C
> >> > > > > > > > > > seems
> >> > > > > > > > > > > > to
> >> > > > > > > > > > > > > > get
> >> > > > > > > > > > > > > > > >> into
> >> > > > > > > > > > > > > > > >> > > > > > epoch=3,
> >> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> >> > > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to
> >> epoch
> >> > > 3, C
> >> > > > > > > still
> >> > > > > > > > > > has A -
> >> > > > > > > > > > > > > > > >> epoch=2,
> >> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of
> >> consumers:
> >> > > Do
> >> > > > > we
> >> > > > > > > > > support
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > upgrade
> >> > > > > > > > > > > > > > > >> from
> >> > > > > > > > > > > > > > > >> > > > any
> >> > > > > > > > > > > > > > > >> > > > > > old
> >> > > > > > > > > > > > > > > >> > > > > > > > version to new one?
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > Thanks,
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > Jun
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20
> >> AM
> >> > > David
> >> > > > > > > Jacot
> >> > > > > > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> >> > > > > > > > > > > > > > > >> > > > > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > Hi all,
> >> > > > > > > > > > > > > > > >> > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > The KIP states that we will
> >> > > > > > > re-implement the
> >> > > > > > > > > > > > > > coordinator
> >> > > > > > > > > > > > > > > >> in
> >> > > > > > > > > > > > > > > >> > > > Java. I
> >> > > > > > > > > > > > > > > >> > > > > > > > > discussed this offline
> >> with a
> >> > > few
> >> > > > > folks
> >> > > > > > > and
> >> > > > > > > > > > folks
> >> > > > > > > > > > > > > are
> >> > > > > > > > > > > > > > > >> concerned
> >> > > > > > > > > > > > > > > >> > > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > we could introduce many
> >> > > regressions
> >> > > > > in
> >> > > > > > > the
> >> > > > > > > > > old
> >> > > > > > > > > > > > > > protocol
> >> > > > > > > > > > > > > > > >> if we do
> >> > > > > > > > > > > > > > > >> > > > so.
> >> > > > > > > > > > > > > > > >> > > > > > > > > Therefore, I am going to
> >> remove
> >> > > this
> >> > > > > > > > > statement
> >> > > > > > > > > > > > from
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> KIP. It
> >> > > > > > > > > > > > > > > >> > > > is an
> >> > > > > > > > > > > > > > > >> > > > > > > > > implementation detail
> >> after all
> >> > > so
> >> > > > > it
> >> > > > > > > does
> >> > > > > > > > > not
> >> > > > > > > > > > > > have
> >> > > > > > > > > > > > > > to be
> >> > > > > > > > > > > > > > > >> > > > decided at
> >> > > > > > > > > > > > > > > >> > > > > > > > > this stage. We will likely
> >> > > start by
> >> > > > > > > trying
> >> > > > > > > > > to
> >> > > > > > > > > > > > > > refactor the
> >> > > > > > > > > > > > > > > >> > > > current
> >> > > > > > > > > > > > > > > >> > > > > > > > > implementation as a first
> >> step.
> >> > > > > > > > > > > > > > > >> > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > Cheers,
> >> > > > > > > > > > > > > > > >> > > > > > > > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at
> >> 3:52 PM
> >> > > > > David
> >> > > > > > > Jacot
> >> > > > > > > > > <
> >> > > > > > > > > > > > > > > >> djacot@confluent.io
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > Hi Luke,
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> >> > > machine
> >> > > > > are:
> >> > > > > > > > > > "Empty,
> >> > > > > > > > > > > > > > assigning,
> >> > > > > > > > > > > > > > > >> > > > > > reconciling,
> >> > > > > > > > > > > > > > > >> > > > > > > > > stable,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in
> >> Consumer
> >> > > > > Group
> >> > > > > > > States
> >> > > > > > > > > > > > > section,
> >> > > > > > > > > > > > > > > >> right?
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > This sentence does not
> >> refer
> >> > > to
> >> > > > > those
> >> > > > > > > > > group
> >> > > > > > > > > > > > states
> >> > > > > > > > > > > > > > but
> >> > > > > > > > > > > > > > > >> rather
> >> > > > > > > > > > > > > > > >> > > > to a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > state machine replication
> >> > > (SMR).
> >> > > > > This
> >> > > > > > > > > > refers to
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> entire
> >> > > > > > > > > > > > > > > >> > > > state of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > group coordinator which
> >> is
> >> > > > > replicated
> >> > > > > > > via
> >> > > > > > > > > > the
> >> > > > > > > > > > > > log
> >> > > > > > > > > > > > > > > >> layer. I will
> >> > > > > > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean
> >> "each
> >> > > > > state
> >> > > > > > > > > machine
> >> > > > > > > > > > is
> >> > > > > > > > > > > > > > modelled
> >> > > > > > > > > > > > > > > >> as an
> >> > > > > > > > > > > > > > > >> > > > event
> >> > > > > > > > > > > > > > > >> > > > > > > > > loop"?
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > The idea is to follow a
> >> model
> >> > > > > similar
> >> > > > > > > to
> >> > > > > > > > > > the new
> >> > > > > > > > > > > > > > quorum
> >> > > > > > > > > > > > > > > >> > > > > > controller. We
> >> > > > > > > > > > > > > > > >> > > > > > > > > > will have N threads to
> >> process
> >> > > > > events.
> >> > > > > > > > > Each
> >> > > > > > > > > > > > > > > >> __consumer_offsets
> >> > > > > > > > > > > > > > > >> > > > > > > > > > partition is assigned to
> >> a
> >> > > unique
> >> > > > > > > thread
> >> > > > > > > > > > and all
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> events
> >> > > > > > > > > > > > > > > >> > > > (e.g.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > requests, callbacks,
> >> etc.) are
> >> > > > > > > processed
> >> > > > > > > > > by
> >> > > > > > > > > > this
> >> > > > > > > > > > > > > > > >> thread. This
> >> > > > > > > > > > > > > > > >> > > > > > simplify
> >> > > > > > > > > > > > > > > >> > > > > > > > > > concurrency and will
> >> enable
> >> > > us to
> >> > > > > do
> >> > > > > > > > > > simulation
> >> > > > > > > > > > > > > > testing
> >> > > > > > > > > > > > > > > >> for the
> >> > > > > > > > > > > > > > > >> > > > > > group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > coordinator.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a
> >> state
> >> > > > > machine
> >> > > > > > > per
> >> > > > > > > > > > > > > > > >> *__consumer_offsets*
> >> > > > > > > > > > > > > > > >> > > > > > > > > partitions?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine
> >> "per
> >> > > > > consumer
> >> > > > > > > group"
> >> > > > > > > > > > owned
> >> > > > > > > > > > > > > by
> >> > > > > > > > > > > > > > a
> >> > > > > > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > coordinator?
> >> > > > > > > > > > > > > > > >> > > > > > > > > For
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> >> > > > > coordinator
> >> > > > > > > owns 2
> >> > > > > > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> groups, and
> >> > > > > > > > > > > > > > > >> > > > > > both
> >> > > > > > > > > > > > > > > >> > > > > > > > > exist in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> >> > > will we
> >> > > > > > > have 1
> >> > > > > > > > > > state
> >> > > > > > > > > > > > > > machine
> >> > > > > > > > > > > > > > > >> for it,
> >> > > > > > > > > > > > > > > >> > > > or
> >> > > > > > > > > > > > > > > >> > > > > > 2?
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion
> >> comes
> >> > > from
> >> > > > > > > there, I
> >> > > > > > > > > > > > think.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> >> > > > > > > > > > "*group.coordinator.threads"
> >> > > > > > > > > > > > *is
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> number
> >> > > > > > > > > > > > > > > >> > > > of
> >> > > > > > > > > > > > > > > >> > > > > > threads
> >> > > > > > > > > > > > > > > >> > > > > > > > > used
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > to run the state
> >> machines.
> >> > > But
> >> > > > > I'm
> >> > > > > > > > > > wondering
> >> > > > > > > > > > > > if
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> purpose
> >> > > > > > > > > > > > > > > >> > > > of
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > threads
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the
> >> state of
> >> > > > > each
> >> > > > > > > > > consumer
> >> > > > > > > > > > > > group
> >> > > > > > > > > > > > > > (or
> >> > > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no
> >> heavy
> >> > > > > > > computation,
> >> > > > > > > > > > why
> >> > > > > > > > > > > > > > should we
> >> > > > > > > > > > > > > > > >> need
> >> > > > > > > > > > > > > > > >> > > > > > > > > multi-threads
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > here?
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to
> >> have
> >> > > an
> >> > > > > > > ability to
> >> > > > > > > > > > shard
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > processing as
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > computation could be
> >> heavy.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer
> >> session
> >> > > > > timeout,
> >> > > > > > > why
> >> > > > > > > > > > does
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> default
> >> > > > > > > > > > > > > > > >> > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > timeout not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > locate between min
> >> (45s) and
> >> > > > > > > max(60s)? I
> >> > > > > > > > > > > > thought
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> min/max
> >> > > > > > > > > > > > > > > >> > > > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> >> > > lower/upper
> >> > > > > > > bound
> >> > > > > > > > > of
> >> > > > > > > > > > it,
> >> > > > > > > > > > > > > no?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > group.consumer.session.timeout.ms
> >> > > > > > > int
> >> > > > > > > > > > 30s The
> >> > > > > > > > > > > > > > > >> timeout to
> >> > > > > > > > > > > > > > > >> > > > detect
> >> > > > > > > > > > > > > > > >> > > > > > client
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> >> > > consumer
> >> > > > > > > group
> >> > > > > > > > > > > > protocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > group.consumer.min.session.timeout.ms
> >> > > > > > > > > > int 45s
> >> > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> minimum
> >> > > > > > > > > > > > > > > >> > > > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > group.consumer.max.session.timeout.ms
> >> > > > > > > > > > int 60s
> >> > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> maximum
> >> > > > > > > > > > > > > > > >> > > > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > This is indeed a
> >> mistake. The
> >> > > > > default
> >> > > > > > > > > > session
> >> > > > > > > > > > > > > > timeout
> >> > > > > > > > > > > > > > > >> should
> >> > > > > > > > > > > > > > > >> > > > be 45s
> >> > > > > > > > > > > > > > > >> > > > > > > > > > (the current default).
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default
> >> server side
> >> > > > > > > assignor
> >> > > > > > > > > are
> >> > > > > > > > > > > > > [range,
> >> > > > > > > > > > > > > > > >> uniform],
> >> > > > > > > > > > > > > > > >> > > > > > which means
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to
> >> "range"
> >> > > > > assignor.
> >> > > > > > > I'd
> >> > > > > > > > > > like to
> >> > > > > > > > > > > > > > know
> >> > > > > > > > > > > > > > > >> why not
> >> > > > > > > > > > > > > > > >> > > > > > uniform
> >> > > > > > > > > > > > > > > >> > > > > > > > > one? I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users
> >> will
> >> > > > > choose
> >> > > > > > > > > uniform
> >> > > > > > > > > > > > > assignor
> >> > > > > > > > > > > > > > > >> (former
> >> > > > > > > > > > > > > > > >> > > > sticky
> >> > > > > > > > > > > > > > > >> > > > > > > > > assinor)
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> >> > > distribution.
> >> > > > > Any
> >> > > > > > > > > other
> >> > > > > > > > > > > > reason
> >> > > > > > > > > > > > > > we
> >> > > > > > > > > > > > > > > >> choose
> >> > > > > > > > > > > > > > > >> > > > range
> >> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > as default?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> group.consumer.assignors
> >> > > List
> >> > > > > range,
> >> > > > > > > > > > uniform
> >> > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> server side
> >> > > > > > > > > > > > > > > >> > > > > > assignors.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > The order on the server
> >> side
> >> > > has
> >> > > > > no
> >> > > > > > > > > > influence
> >> > > > > > > > > > > > > > because
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > client
> >> > > > > > > > > > > > > > > >> > > > > > must
> >> > > > > > > > > > > > > > > >> > > > > > > > > > chose the selector that
> >> he
> >> > > wants
> >> > > > > to
> >> > > > > > > use.
> >> > > > > > > > > > There
> >> > > > > > > > > > > > is
> >> > > > > > > > > > > > > no
> >> > > > > > > > > > > > > > > >> default
> >> > > > > > > > > > > > > > > >> > > > in the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > current proposal. If the
> >> > > assignor
> >> > > > > is
> >> > > > > > > not
> >> > > > > > > > > > > > specified
> >> > > > > > > > > > > > > > by
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > client,
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > request is rejected. The
> >> > > default
> >> > > > > > > client
> >> > > > > > > > > > value
> >> > > > > > > > > > > > for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor`
> >> is
> >> > > > > `uniform`
> >> > > > > > > > > though.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > Thanks for your very good
> >> > > > > comments,
> >> > > > > > > Luke.
> >> > > > > > > > > I
> >> > > > > > > > > > hope
> >> > > > > > > > > > > > > > that my
> >> > > > > > > > > > > > > > > >> > > > answers
> >> > > > > > > > > > > > > > > >> > > > > > help
> >> > > > > > > > > > > > > > > >> > > > > > > > > > to clarify things. I will
> >> > > update
> >> > > > > the
> >> > > > > > > KIP
> >> > > > > > > > > as
> >> > > > > > > > > > well
> >> > > > > > > > > > > > > > based
> >> > > > > > > > > > > > > > > >> on your
> >> > > > > > > > > > > > > > > >> > > > > > > > > > feedback.
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > Cheers,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at
> >> 9:29
> >> > > AM
> >> > > > > Luke
> >> > > > > > > Chen
> >> > > > > > > > > <
> >> > > > > > > > > > > > > > > >> showuon@gmail.com>
> >> > > > > > > > > > > > > > > >> > > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > Hi David,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > Some more questions:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator
> >> > > > > section, you
> >> > > > > > > > > > > > mentioned:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > The new group
> >> coordinator
> >> > > will
> >> > > > > > > have a
> >> > > > > > > > > > state
> >> > > > > > > > > > > > > > machine
> >> > > > > > > > > > > > > > > >> per
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets*
> >> > > partitions,
> >> > > > > > > where
> >> > > > > > > > > > each
> >> > > > > > > > > > > > > state
> >> > > > > > > > > > > > > > > >> machine is
> >> > > > > > > > > > > > > > > >> > > > > > modelled
> >> > > > > > > > > > > > > > > >> > > > > > > > > as an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > event loop. Those state
> >> > > machines
> >> > > > > > > will be
> >> > > > > > > > > > > > > executed
> >> > > > > > > > > > > > > > in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> *group.coordinator.threads*
> >> > > > > threads.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> >> > > machine
> >> > > > > are:
> >> > > > > > > > > > "Empty,
> >> > > > > > > > > > > > > > assigning,
> >> > > > > > > > > > > > > > > >> > > > > > reconciling,
> >> > > > > > > > > > > > > > > >> > > > > > > > > stable,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in
> >> Consumer
> >> > > > > Group
> >> > > > > > > States
> >> > > > > > > > > > > > > section,
> >> > > > > > > > > > > > > > > >> right?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean
> >> "each
> >> > > > > state
> >> > > > > > > > > machine
> >> > > > > > > > > > is
> >> > > > > > > > > > > > > > modelled
> >> > > > > > > > > > > > > > > >> as an
> >> > > > > > > > > > > > > > > >> > > > event
> >> > > > > > > > > > > > > > > >> > > > > > > > > loop"?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a
> >> state
> >> > > > > machine
> >> > > > > > > per
> >> > > > > > > > > > > > > > > >> *__consumer_offsets*
> >> > > > > > > > > > > > > > > >> > > > > > > > > partitions?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine
> >> "per
> >> > > > > consumer
> >> > > > > > > group"
> >> > > > > > > > > > owned
> >> > > > > > > > > > > > > by
> >> > > > > > > > > > > > > > a
> >> > > > > > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > coordinator?
> >> > > > > > > > > > > > > > > >> > > > > > > > > For
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> >> > > > > coordinator
> >> > > > > > > owns 2
> >> > > > > > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> groups, and
> >> > > > > > > > > > > > > > > >> > > > > > both
> >> > > > > > > > > > > > > > > >> > > > > > > > > exist in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> >> > > will we
> >> > > > > > > have 1
> >> > > > > > > > > > state
> >> > > > > > > > > > > > > > machine
> >> > > > > > > > > > > > > > > >> for it,
> >> > > > > > > > > > > > > > > >> > > > or
> >> > > > > > > > > > > > > > > >> > > > > > 2?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> >> > > > > > > > > > "*group.coordinator.threads"
> >> > > > > > > > > > > > *is
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> number
> >> > > > > > > > > > > > > > > >> > > > of
> >> > > > > > > > > > > > > > > >> > > > > > threads
> >> > > > > > > > > > > > > > > >> > > > > > > > > used
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > to run the state
> >> machines.
> >> > > But
> >> > > > > I'm
> >> > > > > > > > > > wondering
> >> > > > > > > > > > > > if
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> purpose
> >> > > > > > > > > > > > > > > >> > > > of
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > threads
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the
> >> state of
> >> > > > > each
> >> > > > > > > > > consumer
> >> > > > > > > > > > > > group
> >> > > > > > > > > > > > > > (or
> >> > > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no
> >> heavy
> >> > > > > > > computation,
> >> > > > > > > > > > why
> >> > > > > > > > > > > > > > should we
> >> > > > > > > > > > > > > > > >> need
> >> > > > > > > > > > > > > > > >> > > > > > > > > multi-threads
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > here?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 2. For the default
> >> value in
> >> > > the
> >> > > > > new
> >> > > > > > > > > > configs:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer
> >> session
> >> > > > > timeout,
> >> > > > > > > why
> >> > > > > > > > > > does
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> default
> >> > > > > > > > > > > > > > > >> > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > timeout not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > locate between min
> >> (45s) and
> >> > > > > > > max(60s)? I
> >> > > > > > > > > > > > thought
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> min/max
> >> > > > > > > > > > > > > > > >> > > > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> >> > > lower/upper
> >> > > > > > > bound
> >> > > > > > > > > of
> >> > > > > > > > > > it,
> >> > > > > > > > > > > > > no?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > group.consumer.session.timeout.ms
> >> > > > > > > int
> >> > > > > > > > > > 30s The
> >> > > > > > > > > > > > > > > >> timeout to
> >> > > > > > > > > > > > > > > >> > > > detect
> >> > > > > > > > > > > > > > > >> > > > > > client
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> >> > > consumer
> >> > > > > > > group
> >> > > > > > > > > > > > protocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > group.consumer.min.session.timeout.ms
> >> > > > > > > > > > int 45s
> >> > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> minimum
> >> > > > > > > > > > > > > > > >> > > > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > group.consumer.max.session.timeout.ms
> >> > > > > > > > > > int 60s
> >> > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> maximum
> >> > > > > > > > > > > > > > > >> > > > > > session
> >> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default
> >> server side
> >> > > > > > > assignor
> >> > > > > > > > > are
> >> > > > > > > > > > > > > [range,
> >> > > > > > > > > > > > > > > >> uniform],
> >> > > > > > > > > > > > > > > >> > > > > > which means
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to
> >> "range"
> >> > > > > assignor.
> >> > > > > > > I'd
> >> > > > > > > > > > like to
> >> > > > > > > > > > > > > > know
> >> > > > > > > > > > > > > > > >> why not
> >> > > > > > > > > > > > > > > >> > > > > > uniform
> >> > > > > > > > > > > > > > > >> > > > > > > > > one? I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users
> >> will
> >> > > > > choose
> >> > > > > > > > > uniform
> >> > > > > > > > > > > > > assignor
> >> > > > > > > > > > > > > > > >> (former
> >> > > > > > > > > > > > > > > >> > > > sticky
> >> > > > > > > > > > > > > > > >> > > > > > > > > assinor)
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> >> > > distribution.
> >> > > > > Any
> >> > > > > > > > > other
> >> > > > > > > > > > > > reason
> >> > > > > > > > > > > > > > we
> >> > > > > > > > > > > > > > > >> choose
> >> > > > > > > > > > > > > > > >> > > > range
> >> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > as default?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> group.consumer.assignors
> >> > > List
> >> > > > > range,
> >> > > > > > > > > > uniform
> >> > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> server side
> >> > > > > > > > > > > > > > > >> > > > > > assignors.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > Thank you.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > Luke
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at
> >> > > 2:10 PM
> >> > > > > Luke
> >> > > > > > > > > Chen
> >> > > > > > > > > > <
> >> > > > > > > > > > > > > > > >> showuon@gmail.com
> >> > > > > > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts
> >> about
> >> > > > > Kafka
> >> > > > > > > > > Connect
> >> > > > > > > > > > > > > > > >> integrating with
> >> > > > > > > > > > > > > > > >> > > > > > KIP-848,
> >> > > > > > > > > > > > > > > >> > > > > > > > > but I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > think we should have
> >> a
> >> > > > > separate
> >> > > > > > > > > > discussion
> >> > > > > > > > > > > > > > thread
> >> > > > > > > > > > > > > > > >> for the
> >> > > > > > > > > > > > > > > >> > > > Kafka
> >> > > > > > > > > > > > > > > >> > > > > > > > > Connect
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating
> >> Kafka
> >> > > Connect
> >> > > > > > > With
> >> > > > > > > > > New
> >> > > > > > > > > > > > > Consumer
> >> > > > > > > > > > > > > > > >> Rebalance
> >> > > > > > > > > > > > > > > >> > > > > > Protocol
> >> > > > > > > > > > > > > > > >> > > > > > > > > [1],
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > and let this
> >> discussion
> >> > > thread
> >> > > > > > > focus
> >> > > > > > > > > on
> >> > > > > > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> rebalance
> >> > > > > > > > > > > > > > > >> > > > > > protocol,
> >> > > > > > > > > > > > > > > >> > > > > > > > > WDYT?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > [1]
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >>
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > >
> >> > > > > > > > >
> >> > > > > > >
> >> > > > >
> >> > >
> >> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > Thank you.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > Luke
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022
> >> at
> >> > > 9:31
> >> > > > > PM
> >> > > > > > > Sagar
> >> > > > > > > > > <
> >> > > > > > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> >> > > > > > > > > > > > > > > >> > > > > > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Thank you
> >> Guozhang/David
> >> > > for
> >> > > > > the
> >> > > > > > > > > > feedback.
> >> > > > > > > > > > > > > > Looks
> >> > > > > > > > > > > > > > > >> like
> >> > > > > > > > > > > > > > > >> > > > there's
> >> > > > > > > > > > > > > > > >> > > > > > > > > agreement on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs
> >> for
> >> > > > > Connect.
> >> > > > > > > I
> >> > > > > > > > > > would
> >> > > > > > > > > > > > > > revisit
> >> > > > > > > > > > > > > > > >> the doc
> >> > > > > > > > > > > > > > > >> > > > and
> >> > > > > > > > > > > > > > > >> > > > > > see
> >> > > > > > > > > > > > > > > >> > > > > > > > > what
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> changes are to be
> >> made.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022
> >> at
> >> > > 7:11
> >> > > > > PM
> >> > > > > > > David
> >> > > > > > > > > > Jacot
> >> > > > > > > > > > > > > > > >> > > > > > > > >
> >> <dj...@confluent.io.invalid>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the
> >> feedback
> >> > > > > and the
> >> > > > > > > > > > document.
> >> > > > > > > > > > > > > > That's
> >> > > > > > > > > > > > > > > >> really
> >> > > > > > > > > > > > > > > >> > > > > > helpful. I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will take a look
> >> at it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems
> >> to me
> >> > > > > that
> >> > > > > > > both
> >> > > > > > > > > > Connect
> >> > > > > > > > > > > > > > and the
> >> > > > > > > > > > > > > > > >> > > > Consumer
> >> > > > > > > > > > > > > > > >> > > > > > could
> >> > > > > > > > > > > > > > > >> > > > > > > > > share
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the same
> >> underlying
> >> > > > > "engine".
> >> > > > > > > The
> >> > > > > > > > > > main
> >> > > > > > > > > > > > > > > >> difference is
> >> > > > > > > > > > > > > > > >> > > > that
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > Consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigns
> >> > > topic-partitions to
> >> > > > > > > members
> >> > > > > > > > > > > > whereas
> >> > > > > > > > > > > > > > > >> Connect
> >> > > > > > > > > > > > > > > >> > > > assigns
> >> > > > > > > > > > > > > > > >> > > > > > tasks
> >> > > > > > > > > > > > > > > >> > > > > > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two
> >> > > ways to
> >> > > > > move
> >> > > > > > > > > > forward:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the
> >> new
> >> > > > > proposed
> >> > > > > > > APIs
> >> > > > > > > > > to
> >> > > > > > > > > > > > > support
> >> > > > > > > > > > > > > > > >> different
> >> > > > > > > > > > > > > > > >> > > > > > resource
> >> > > > > > > > > > > > > > > >> > > > > > > > > types
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions,
> >> > > tasks,
> >> > > > > > > etc.); or
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new
> >> dedicated
> >> > > > > APIs
> >> > > > > > > for
> >> > > > > > > > > > Connect.
> >> > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> dedicated
> >> > > > > > > > > > > > > > > >> > > > APIs
> >> > > > > > > > > > > > > > > >> > > > > > > > > would be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new
> >> > > ones but
> >> > > > > > > > > > different on
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > content/resources and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > they would rely
> >> on the
> >> > > same
> >> > > > > > > engine
> >> > > > > > > > > > on the
> >> > > > > > > > > > > > > > > >> coordinator
> >> > > > > > > > > > > > > > > >> > > > side.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean
> >> > > towards
> >> > > > > 2)
> >> > > > > > > > > because
> >> > > > > > > > > > I am
> >> > > > > > > > > > > > > > not a
> >> > > > > > > > > > > > > > > >> fan of
> >> > > > > > > > > > > > > > > >> > > > > > > > > overcharging
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve
> >> different
> >> > > > > > > purposes.
> >> > > > > > > > > > That
> >> > > > > > > > > > > > > being
> >> > > > > > > > > > > > > > > >> said, I am
> >> > > > > > > > > > > > > > > >> > > > not
> >> > > > > > > > > > > > > > > >> > > > > > > > > opposed to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find
> >> an
> >> > > > > elegant
> >> > > > > > > way to
> >> > > > > > > > > > do
> >> > > > > > > > > > > > it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I think that we
> >> can
> >> > > > > continue to
> >> > > > > > > > > > discuss
> >> > > > > > > > > > > > it
> >> > > > > > > > > > > > > > here
> >> > > > > > > > > > > > > > > >> for now
> >> > > > > > > > > > > > > > > >> > > > in
> >> > > > > > > > > > > > > > > >> > > > > > order
> >> > > > > > > > > > > > > > > >> > > > > > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this
> >> KIP is
> >> > > > > > > compatible
> >> > > > > > > > > > with
> >> > > > > > > > > > > > > what
> >> > > > > > > > > > > > > > we
> >> > > > > > > > > > > > > > > >> will do
> >> > > > > > > > > > > > > > > >> > > > for
> >> > > > > > > > > > > > > > > >> > > > > > > > > Connect in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the future.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Best,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8,
> >> 2022 at
> >> > > > > 2:41 PM
> >> > > > > > > > > David
> >> > > > > > > > > > > > Jacot
> >> > > > > > > > > > > > > <
> >> > > > > > > > > > > > > > > >> > > > > > djacot@confluent.io>
> >> > > > > > > > > > > > > > > >> > > > > > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from
> >> > > vacation.
> >> > > > > I
> >> > > > > > > will
> >> > > > > > > > > go
> >> > > > > > > > > > > > > through
> >> > > > > > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> address
> >> > > > > > > > > > > > > > > >> > > > > > your
> >> > > > > > > > > > > > > > > >> > > > > > > > > comments
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming
> >> days.
> >> > > > > Thanks
> >> > > > > > > for
> >> > > > > > > > > your
> >> > > > > > > > > > > > > > feedback.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3,
> >> 2022
> >> > > at
> >> > > > > 10:05
> >> > > > > > > PM
> >> > > > > > > > > > Gregory
> >> > > > > > > > > > > > > > Harris
> >> > > > > > > > > > > > > > > >> <
> >> > > > > > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for
> >> the KIP,
> >> > > > > it's
> >> > > > > > > > > > wonderful to
> >> > > > > > > > > > > > > see
> >> > > > > > > > > > > > > > > >> > > > cooperative
> >> > > > > > > > > > > > > > > >> > > > > > > > > rebalancing
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > making it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > down the
> >> stack!
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few
> >> > > questions:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The
> >> 'Rejected
> >> > > > > > > Alternatives'
> >> > > > > > > > > > > > section
> >> > > > > > > > > > > > > > > >> describes how
> >> > > > > > > > > > > > > > > >> > > > > > member
> >> > > > > > > > > > > > > > > >> > > > > > > > > epoch
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in
> >> step
> >> > > with
> >> > > > > the
> >> > > > > > > group
> >> > > > > > > > > > epoch
> >> > > > > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> assignment
> >> > > > > > > > > > > > > > > >> > > > > > epoch
> >> > > > > > > > > > > > > > > >> > > > > > > > > values. I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is
> >> a good
> >> > > > > idea
> >> > > > > > > for
> >> > > > > > > > > the
> >> > > > > > > > > > > > > reasons
> >> > > > > > > > > > > > > > > >> described
> >> > > > > > > > > > > > > > > >> > > > in
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > KIP. When
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is
> >> > > > > incrementally
> >> > > > > > > > > > assigning
> >> > > > > > > > > > > > > > > >> partitions to a
> >> > > > > > > > > > > > > > > >> > > > > > worker,
> >> > > > > > > > > > > > > > > >> > > > > > > > > what
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> member
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does
> >> each
> >> > > > > incremental
> >> > > > > > > > > > > > assignment
> >> > > > > > > > > > > > > > use?
> >> > > > > > > > > > > > > > > >> Are
> >> > > > > > > > > > > > > > > >> > > > member
> >> > > > > > > > > > > > > > > >> > > > > > epochs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> re-used,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > a single
> >> member
> >> > > epoch
> >> > > > > can
> >> > > > > > > > > > correspond
> >> > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> multiple
> >> > > > > > > > > > > > > > > >> > > > > > different
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > larger)
> >> > > assignments?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the
> >> > > Assignor's
> >> > > > > > > 'Reason'
> >> > > > > > > > > > field
> >> > > > > > > > > > > > > > opaque
> >> > > > > > > > > > > > > > > >> to the
> >> > > > > > > > > > > > > > > >> > > > group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should
> >> custom
> >> > > > > > > client-side
> >> > > > > > > > > > > > assignor
> >> > > > > > > > > > > > > > > >> > > > implementations
> >> > > > > > > > > > > > > > > >> > > > > > > > > interact
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field,
> >> and
> >> > > how
> >> > > > > is
> >> > > > > > > its
> >> > > > > > > > > > common
> >> > > > > > > > > > > > > > meaning
> >> > > > > > > > > > > > > > > >> agreed
> >> > > > > > > > > > > > > > > >> > > > > > upon? If
> >> > > > > > > > > > > > > > > >> > > > > > > > > so, what
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > is the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a
> >> > > distinct
> >> > > > > > > Reason
> >> > > > > > > > > > field
> >> > > > > > > > > > > > over
> >> > > > > > > > > > > > > > > >> including
> >> > > > > > > > > > > > > > > >> > > > such
> >> > > > > > > > > > > > > > > >> > > > > > > > > functionality
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque
> >> metadata?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The
> >> following is
> >> > > > > > > included in
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > KIP:
> >> > > > > > > > > > > > > > > >> "Thanks to
> >> > > > > > > > > > > > > > > >> > > > > > this, the
> >> > > > > > > > > > > > > > > >> > > > > > > > > input
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client side
> >> > > assignor is
> >> > > > > > > > > entirely
> >> > > > > > > > > > > > driven
> >> > > > > > > > > > > > > > by
> >> > > > > > > > > > > > > > > >> the group
> >> > > > > > > > > > > > > > > >> > > > > > > > > coordinator.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> The
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no
> >> > > longer
> >> > > > > > > > > > responsible for
> >> > > > > > > > > > > > > > > >> maintaining
> >> > > > > > > > > > > > > > > >> > > > any
> >> > > > > > > > > > > > > > > >> > > > > > state
> >> > > > > > > > > > > > > > > >> > > > > > > > > besides
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> its
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned
> >> > > partitions."
> >> > > > > Does
> >> > > > > > > this
> >> > > > > > > > > > mean
> >> > > > > > > > > > > > > > that the
> >> > > > > > > > > > > > > > > >> > > > > > client-side
> >> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> MAY
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate
> >> > > additional
> >> > > > > > > > > > non-Metadata
> >> > > > > > > > > > > > > state
> >> > > > > > > > > > > > > > > >> (such as
> >> > > > > > > > > > > > > > > >> > > > > > partition
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory
> >> metrics,
> >> > > > > config
> >> > > > > > > > > > topics,
> >> > > > > > > > > > > > > etc),
> >> > > > > > > > > > > > > > or
> >> > > > > > > > > > > > > > > >> that
> >> > > > > > > > > > > > > > > >> > > > > > additional
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD
> >> NOT be
> >> > > > > used?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that
> >> > > there are
> >> > > > > > > > > separate
> >> > > > > > > > > > > > > classes
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > for
> >> > > > > > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and
> >> > > > > > > > > > > > > > > >> > > >
> >> > > > > > > org.apache.kafka.clients.consumer.PartitionAssignor
> >> > > > > > > > > > > > > > > >> > > > > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > seem to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap
> >> > > significantly.
> >> > > > > Is
> >> > > > > > > it
> >> > > > > > > > > > possible
> >> > > > > > > > > > > > > for
> >> > > > > > > > > > > > > > > >> these two
> >> > > > > > > > > > > > > > > >> > > > > > > > > implementations
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This
> >> would
> >> > > > > serve
> >> > > > > > > to
> >> > > > > > > > > > promote
> >> > > > > > > > > > > > > > feature
> >> > > > > > > > > > > > > > > >> parity
> >> > > > > > > > > > > > > > > >> > > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > server-side
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side
> >> > > assignors,
> >> > > > > and
> >> > > > > > > > > would
> >> > > > > > > > > > also
> >> > > > > > > > > > > > > > > >> facilitate
> >> > > > > > > > > > > > > > > >> > > > > > operational
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > certain
> >> > > situations. For
> >> > > > > > > > > example,
> >> > > > > > > > > > if a
> >> > > > > > > > > > > > > > > >> server-side
> >> > > > > > > > > > > > > > > >> > > > > > assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > has some
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > poor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and
> >> needs
> >> > > a
> >> > > > > patch,
> >> > > > > > > > > > deploying
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> patched
> >> > > > > > > > > > > > > > > >> > > > > > assignor to
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching
> >> one
> >> > > > > consumer
> >> > > > > > > > > group
> >> > > > > > > > > > to a
> >> > > > > > > > > > > > > > > >> client-side
> >> > > > > > > > > > > > > > > >> > > > > > assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > may be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > faster
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and less
> >> risky than
> >> > > > > > > patching
> >> > > > > > > > > all
> >> > > > > > > > > > of
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> brokers.
> >> > > > > > > > > > > > > > > >> > > > With
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > currently
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed
> >> distinct
> >> > > > > APIs, a
> >> > > > > > > > > > non-trivial
> >> > > > > > > > > > > > > > > >> > > > reimplementation
> >> > > > > > > > > > > > > > > >> > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > have
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled,
> >> and if
> >> > > the
> >> > > > > two
> >> > > > > > > APIs
> >> > > > > > > > > > have
> >> > > > > > > > > > > > > > diverged
> >> > > > > > > > > > > > > > > >> > > > > > significantly,
> >> > > > > > > > > > > > > > > >> > > > > > > > > then it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that
> >> a
> >> > > > > > > > > reimplementation
> >> > > > > > > > > > > > would
> >> > > > > > > > > > > > > > not be
> >> > > > > > > > > > > > > > > >> > > > possible.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > --
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > gharris1727@gmail.com
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > github.com/gharris1727
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3,
> >> > > 2022 at
> >> > > > > > > 8:39 AM
> >> > > > > > > > > > Sagar
> >> > > > > > > > > > > > <
> >> > > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi
> >> > > Guozhang/David,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a
> >> > > > > confluence
> >> > > > > > > page
> >> > > > > > > > > to
> >> > > > > > > > > > > > > discuss
> >> > > > > > > > > > > > > > how
> >> > > > > > > > > > > > > > > >> Connect
> >> > > > > > > > > > > > > > > >> > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > need to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > change
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on
> >> the new
> >> > > > > > > rebalance
> >> > > > > > > > > > > > protocol.
> >> > > > > > > > > > > > > > > >> Here's the
> >> > > > > > > > > > > > > > > >> > > > page:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >>
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > >
> >> > > > > > > > >
> >> > > > > > >
> >> > > > >
> >> > >
> >> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also
> >> pretty
> >> > > > > longish
> >> > > > > > > and
> >> > > > > > > > > I
> >> > > > > > > > > > have
> >> > > > > > > > > > > > > > tried
> >> > > > > > > > > > > > > > > >> to keep
> >> > > > > > > > > > > > > > > >> > > > a
> >> > > > > > > > > > > > > > > >> > > > > > format
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> similar to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848.
> >> Let me
> >> > > know
> >> > > > > > > what you
> >> > > > > > > > > > > > think.
> >> > > > > > > > > > > > > > Also,
> >> > > > > > > > > > > > > > > >> do you
> >> > > > > > > > > > > > > > > >> > > > > > think this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a
> >> > > separate
> >> > > > > > > > > discussion
> >> > > > > > > > > > > > thread
> >> > > > > > > > > > > > > > or
> >> > > > > > > > > > > > > > > >> is this
> >> > > > > > > > > > > > > > > >> > > > one
> >> > > > > > > > > > > > > > > >> > > > > > fine?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul
> >> 26,
> >> > > 2022
> >> > > > > at
> >> > > > > > > 7:37
> >> > > > > > > > > AM
> >> > > > > > > > > > > > > Sagar <
> >> > > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello
> >> Guozhang,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you
> >> so
> >> > > much
> >> > > > > for
> >> > > > > > > the
> >> > > > > > > > > > doc on
> >> > > > > > > > > > > > > > Kafka
> >> > > > > > > > > > > > > > > >> Streams.
> >> > > > > > > > > > > > > > > >> > > > > > Sure, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > would do
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis
> >> and
> >> > > come
> >> > > > > up
> >> > > > > > > with
> >> > > > > > > > > > such a
> >> > > > > > > > > > > > > > > >> document.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue,
> >> Jul 26,
> >> > > > > 2022 at
> >> > > > > > > > > 4:47
> >> > > > > > > > > > AM
> >> > > > > > > > > > > > > > Guozhang
> >> > > > > > > > > > > > > > > >> Wang <
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello
> >> Sagar,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would
> >> be
> >> > > great
> >> > > > > if
> >> > > > > > > you
> >> > > > > > > > > > could
> >> > > > > > > > > > > > > come
> >> > > > > > > > > > > > > > > >> back with
> >> > > > > > > > > > > > > > > >> > > > some
> >> > > > > > > > > > > > > > > >> > > > > > > > > analysis on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > how to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> implement the
> >> > > > > Connect
> >> > > > > > > side
> >> > > > > > > > > > > > > > integration
> >> > > > > > > > > > > > > > > >> with
> >> > > > > > > > > > > > > > > >> > > > the new
> >> > > > > > > > > > > > > > > >> > > > > > > > > protocol;
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> so
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > far
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides
> >> > > > > leveraging on
> >> > > > > > > the
> >> > > > > > > > > > new
> >> > > > > > > > > > > > > > "protocol
> >> > > > > > > > > > > > > > > >> type"
> >> > > > > > > > > > > > > > > >> > > > we
> >> > > > > > > > > > > > > > > >> > > > > > did not
> >> > > > > > > > > > > > > > > >> > > > > > > > > yet
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> Connect
> >> > > side
> >> > > > > > > > > > > > implementations.
> >> > > > > > > > > > > > > > For
> >> > > > > > > > > > > > > > > >> Streams
> >> > > > > > > > > > > > > > > >> > > > > > here's a
> >> > > > > > > > > > > > > > > >> > > > > > > > > draft of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> integration
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > >
> >> > > > > > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > >
> >> > > > > > > > > > > > > > > >>
> >> > > > > > > > > > > > > >
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > >
> >> > > > > > > > >
> >> > > > > > >
> >> > > > >
> >> > >
> >> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI
> >> for
> >> > > your
> >> > > > > > > analysis
> >> > > > > > > > > > on
> >> > > > > > > > > > > > > > Connect.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue,
> >> Jul
> >> > > 19,
> >> > > > > 2022
> >> > > > > > > at
> >> > > > > > > > > > 10:48 PM
> >> > > > > > > > > > > > > > Sagar <
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> > > sagarmeansocean@gmail.com
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi
> >> David,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank
> >> you
> >> > > for
> >> > > > > your
> >> > > > > > > > > > response.
> >> > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> reason I
> >> > > > > > > > > > > > > > > >> > > > thought
> >> > > > > > > > > > > > > > > >> > > > > > > > > connect can
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > also fit
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this
> >> new
> >> > > scheme
> >> > > > > is
> >> > > > > > > that
> >> > > > > > > > > > even
> >> > > > > > > > > > > > > > today the
> >> > > > > > > > > > > > > > > >> > > > connect
> >> > > > > > > > > > > > > > > >> > > > > > uses a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> WorkerCoordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> extending
> >> > > from
> >> > > > > > > > > > > > > > AbstractCoordinator to
> >> > > > > > > > > > > > > > > >> empower
> >> > > > > > > > > > > > > > > >> > > > > > > > > rebalances of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > tasks/connectors.
> >> > > > > > > The
> >> > > > > > > > > > > > > > > >> WorkerCoordinator sets
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and
> >> uses the
> >> > > > > > > metadata()
> >> > > > > > > > > > method
> >> > > > > > > > > > > > > by
> >> > > > > > > > > > > > > > > >> plumbing
> >> > > > > > > > > > > > > > > >> > > > into
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > JoinGroupRequestProtocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I
> >> think the
> >> > > > > changes
> >> > > > > > > to
> >> > > > > > > > > > support
> >> > > > > > > > > > > > > > > >> connect would
> >> > > > > > > > > > > > > > > >> > > > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > similar at a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > high
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> >> changes
> >> > > in
> >> > > > > > > streams
> >> > > > > > > > > > mainly
> >> > > > > > > > > > > > > > because
> >> > > > > > > > > > > > > > > >> of the
> >> > > > > > > > > > > > > > > >> > > > > > Client
> >> > > > > > > > > > > > > > > >> > > > > > > > > side
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assignors
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in
> >> > > both.
> >> > > > > At an
> >> > > > > > > > > > > > > implementation
> >> > > > > > > > > > > > > > > >> level, we
> >> > > > > > > > > > > > > > > >> > > > > > might
> >> > > > > > > > > > > > > > > >> > > > > > > > > need to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> make
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > a lot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> changes to
> >> > > get
> >> > > > > onto
> >> > > > > > > this
> >> > > > > > > > > > new
> >> > > > > > > > > > > > > > > >> assignment
> >> > > > > > > > > > > > > > > >> > > > protocol
> >> > > > > > > > > > > > > > > >> > > > > > like
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> enhancing
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> JoinGroup
> >> > > > > > > > > > request/response and
> >> > > > > > > > > > > > > > > >> SyncGroup and
> >> > > > > > > > > > > > > > > >> > > > > > using
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > ConsumerGroupHeartbeat
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc
> >> > > again on
> >> > > > > > > similar
> >> > > > > > > > > > lines
> >> > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> streams (or
> >> > > > > > > > > > > > > > > >> > > > > > there
> >> > > > > > > > > > > > > > > >> > > > > > > > > might be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> deviations). I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would
> >> try to
> >> > > > > > > perform a
> >> > > > > > > > > > > > detailed
> >> > > > > > > > > > > > > > > >> analysis of
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > same
> >> > > > > > > > > > > > > > > >> > > > > > > > > and we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > can have
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> separate
> >> > > > > discussion
> >> > > > > > > > > > thread for
> >> > > > > > > > > > > > > > that
> >> > > > > > > > > > > > > > > >> as that
> >> > > > > > > > > > > > > > > >> > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > derail this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> thread. Let
> >> > > me
> >> > > > > know
> >> > > > > > > if
> >> > > > > > > > > > that
> >> > > > > > > > > > > > > sounds
> >> > > > > > > > > > > > > > > >> good to
> >> > > > > > > > > > > > > > > >> > > > you.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On
> >> Fri, Jul
> >> > > 15,
> >> > > > > > > 2022 at
> >> > > > > > > > > > 5:47
> >> > > > > > > > > > > > PM
> >> > > > > > > > > > > > > > David
> >> > > > > > > > > > > > > > > >> Jacot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi
> >> Sagar,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> Thanks for
> >> > > > > your
> >> > > > > > > > > > comments.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1)
> >> Yes.
> >> > > That
> >> > > > > > > refers to
> >> > > > > > > > > > > > > > > >> `Assignment#error`.
> >> > > > > > > > > > > > > > > >> > > > > > Sure, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > can
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2)
> >> The
> >> > > idea
> >> > > > > is to
> >> > > > > > > > > > > > transition C
> >> > > > > > > > > > > > > > from
> >> > > > > > > > > > > > > > > >> his
> >> > > > > > > > > > > > > > > >> > > > current
> >> > > > > > > > > > > > > > > >> > > > > > > > > assignment
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > his
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> target
> >> > > > > assignment
> >> > > > > > > when
> >> > > > > > > > > > he
> >> > > > > > > > > > > > can
> >> > > > > > > > > > > > > > move
> >> > > > > > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > epoch 3.
> >> > > > > > > > > > > > > > > >> > > > > > When
> >> > > > > > > > > > > > > > > >> > > > > > > > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > happens,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> member
> >> > > > > assignment
> >> > > > > > > is
> >> > > > > > > > > > updated
> >> > > > > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> persisted
> >> > > > > > > > > > > > > > > >> > > > > > with all
> >> > > > > > > > > > > > > > > >> > > > > > > > > its
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigned
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> partitions
> >> > > > > even if
> >> > > > > > > > > they
> >> > > > > > > > > > are
> >> > > > > > > > > > > > > not
> >> > > > > > > > > > > > > > all
> >> > > > > > > > > > > > > > > >> revoked
> >> > > > > > > > > > > > > > > >> > > > > > yet. In
> >> > > > > > > > > > > > > > > >> > > > > > > > > other
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > words, the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> member
> >> > > > > assignment
> >> > > > > > > > > > becomes
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > target
> >> > > > > > > > > > > > > > > >> > > > > > assignment.
> >> > > > > > > > > > > > > > > >> > > > > > > > > This is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > basically
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > optimization
> >> > > > > to
> >> > > > > > > avoid
> >> > > > > > > > > > having
> >> > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> write all
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > changes to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > log.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> examples
> >> > > are
> >> > > > > > > based on
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > persisted
> >> > > > > > > > > > > > > > > >> state
> >> > > > > > > > > > > > > > > >> > > > so I
> >> > > > > > > > > > > > > > > >> > > > > > > > > understand
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> confusion.
> >> > > > > Let me
> >> > > > > > > see
> >> > > > > > > > > > if I
> >> > > > > > > > > > > > can
> >> > > > > > > > > > > > > > > >> improve
> >> > > > > > > > > > > > > > > >> > > > this in
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > description.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3)
> >> > > Regarding
> >> > > > > > > Connect,
> >> > > > > > > > > it
> >> > > > > > > > > > > > could
> >> > > > > > > > > > > > > > > >> reuse the
> >> > > > > > > > > > > > > > > >> > > > > > protocol
> >> > > > > > > > > > > > > > > >> > > > > > > > > with a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client side
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> assignor
> >> > > if it
> >> > > > > > > fits in
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> protocol. The
> >> > > > > > > > > > > > > > > >> > > > > > assignment
> >> > > > > > > > > > > > > > > >> > > > > > > > > is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > topicid-partitions +
> >> > > > > > > > > > > > metadata,
> >> > > > > > > > > > > > > > could
> >> > > > > > > > > > > > > > > >> > > > Connect
> >> > > > > > > > > > > > > > > >> > > > > > fit
> >> > > > > > > > > > > > > > > >> > > > > > > > > into this?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On
> >> Fri,
> >> > > Jul
> >> > > > > 15,
> >> > > > > > > 2022
> >> > > > > > > > > at
> >> > > > > > > > > > 1:55
> >> > > > > > > > > > > > > PM
> >> > > > > > > > > > > > > > > >> Sagar <
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > sagarmeansocean@gmail.com>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi
> >> > > David,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> Thanks
> >> > > for
> >> > > > > the
> >> > > > > > > KIP.
> >> > > > > > > > > I
> >> > > > > > > > > > just
> >> > > > > > > > > > > > > had
> >> > > > > > > > > > > > > > > >> minor
> >> > > > > > > > > > > > > > > >> > > > > > observations:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1)
> >> In
> >> > > the
> >> > > > > > > Assignment
> >> > > > > > > > > > Error
> >> > > > > > > > > > > > > > > >> section in
> >> > > > > > > > > > > > > > > >> > > > Client
> >> > > > > > > > > > > > > > > >> > > > > > Side
> >> > > > > > > > > > > > > > > >> > > > > > > > > mode
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> process,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you
> >> > > > > mentioned
> >> > > > > > > => `In
> >> > > > > > > > > > this
> >> > > > > > > > > > > > > > case,
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > client
> >> > > > > > > > > > > > > > > >> > > > > > side
> >> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> can
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> error
> >> > > to the
> >> > > > > > > group
> >> > > > > > > > > > > > > > coordinator`.
> >> > > > > > > > > > > > > > > >> In this
> >> > > > > > > > > > > > > > > >> > > > > > case are
> >> > > > > > > > > > > > > > > >> > > > > > > > > you
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > referring to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> Assignor
> >> > > > > > > returning
> >> > > > > > > > > an
> >> > > > > > > > > > > > > > > >> AssignmentError
> >> > > > > > > > > > > > > > > >> > > > that's
> >> > > > > > > > > > > > > > > >> > > > > > > > > listed down
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > towards
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If
> >> yes,
> >> > > do
> >> > > > > you
> >> > > > > > > think
> >> > > > > > > > > > it
> >> > > > > > > > > > > > > would
> >> > > > > > > > > > > > > > > >> make sense
> >> > > > > > > > > > > > > > > >> > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > mention this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2)
> >> In
> >> > > the
> >> > > > > Case
> >> > > > > > > > > Studies
> >> > > > > > > > > > > > > > section, I
> >> > > > > > > > > > > > > > > >> have a
> >> > > > > > > > > > > > > > > >> > > > > > slight
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> confusion,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> others
> >> > > have
> >> > > > > the
> >> > > > > > > > > same.
> >> > > > > > > > > > > > > Consider
> >> > > > > > > > > > > > > > > >> this step:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> When B
> >> > > > > > > heartbeats,
> >> > > > > > > > > the
> >> > > > > > > > > > > > group
> >> > > > > > > > > > > > > > > >> coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > transitions him
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > epoch
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> because
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B
> >> has no
> >> > > > > > > partitions
> >> > > > > > > > > to
> >> > > > > > > > > > > > > > revoke. It
> >> > > > > > > > > > > > > > > >> > > > persists
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > change and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > reply.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> -
> >> > > Group
> >> > > > > > > (epoch=3)
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > A
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > B
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > C
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> -
> >> > > Target
> >> > > > > > > > > Assignment
> >> > > > > > > > > > > > > > (epoch=3)
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > A -
> >> > > > > > > > > > > > > partitions=[foo-0]
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > B -
> >> > > > > > > > > > > > > partitions=[foo-2]
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > C -
> >> > > > > > > > > > > > > partitions=[foo-1]
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> -
> >> > > Member
> >> > > > > > > > > Assignment
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > A -
> >> > > > > > > epoch=2,
> >> > > > > > > > > > > > > > > >> partitions=[foo-0,
> >> > > > > > > > > > > > > > > >> > > > > > foo-1]
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > B -
> >> > > > > > > epoch=3,
> >> > > > > > > > > > > > > > > >> partitions=[foo-2]
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >>    -
> >> > > C -
> >> > > > > > > epoch=3,
> >> > > > > > > > > > > > > > > >> partitions=[foo-1]
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> When C
> >> > > > > > > heartbeats,
> >> > > > > > > > > it
> >> > > > > > > > > > > > > > transitions
> >> > > > > > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > epoch 3
> >> > > > > > > > > > > > > > > >> > > > > > but
> >> > > > > > > > > > > > > > > >> > > > > > > > > cannot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> get
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > Here,it's
> >> > > > > > > mentioned
> >> > > > > > > > > > that
> >> > > > > > > > > > > > > > member C
> >> > > > > > > > > > > > > > > >> can't
> >> > > > > > > > > > > > > > > >> > > > get
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > foo-1
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > partition
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> based
> >> > > on the
> >> > > > > > > > > > description
> >> > > > > > > > > > > > > > above,
> >> > > > > > > > > > > > > > > >> it seems
> >> > > > > > > > > > > > > > > >> > > > it
> >> > > > > > > > > > > > > > > >> > > > > > > > > already has
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Do you
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> would be
> >> > > > > better
> >> > > > > > > to
> >> > > > > > > > > > remove
> >> > > > > > > > > > > > it
> >> > > > > > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> > > > populate it
> >> > > > > > > > > > > > > > > >> > > > > > only
> >> > > > > > > > > > > > > > > >> > > > > > > > > when it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > actually
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I
> >> see
> >> > > this
> >> > > > > in a
> >> > > > > > > lot
> >> > > > > > > > > of
> >> > > > > > > > > > > > other
> >> > > > > > > > > > > > > > > >> places, so
> >> > > > > > > > > > > > > > > >> > > > have
> >> > > > > > > > > > > > > > > >> > > > > > I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> understood it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> incorrectly
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > Regarding
> >> > > > > > > connect ,
> >> > > > > > > > > it
> >> > > > > > > > > > > > might
> >> > > > > > > > > > > > > > be
> >> > > > > > > > > > > > > > > >> out of
> >> > > > > > > > > > > > > > > >> > > > scope
> >> > > > > > > > > > > > > > > >> > > > > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> what I
> >> > > > > > > understood it
> >> > > > > > > > > > would
> >> > > > > > > > > > > > > > > >> probably be
> >> > > > > > > > > > > > > > > >> > > > > > running in
> >> > > > > > > > > > > > > > > >> > > > > > > > > client
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> even on
> >> > > the
> >> > > > > new
> >> > > > > > > > > > rebalance
> >> > > > > > > > > > > > > > > >> protocol as it
> >> > > > > > > > > > > > > > > >> > > > has
> >> > > > > > > > > > > > > > > >> > > > > > its
> >> > > > > > > > > > > > > > > >> > > > > > > > > own
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Custom
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > Assignors(Eager
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> >> > > > > > > > > > > > IncrementalCooperative).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> Thanks!
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> Sagar.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On
> >> Fri,
> >> > > Jul
> >> > > > > 15,
> >> > > > > > > 2022
> >> > > > > > > > > > at
> >> > > > > > > > > > > > 5:00
> >> > > > > > > > > > > > > > PM
> >> > > > > > > > > > > > > > > >> David
> >> > > > > > > > > > > > > > > >> > > > Jacot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> Thanks
> >> > > > > > > Hector! Our
> >> > > > > > > > > > goal
> >> > > > > > > > > > > > is
> >> > > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> move
> >> > > > > > > > > > > > > > > >> > > > forward
> >> > > > > > > > > > > > > > > >> > > > > > with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > instead of
> >> > > > > > > relying
> >> > > > > > > > > > on
> >> > > > > > > > > > > > one
> >> > > > > > > > > > > > > > > >> generic API.
> >> > > > > > > > > > > > > > > >> > > > For
> >> > > > > > > > > > > > > > > >> > > > > > > > > Connect, we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > can apply
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> exact
> >> > > same
> >> > > > > > > pattern
> >> > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> reuse/share the
> >> > > > > > > > > > > > > > > >> > > > core
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> server
> >> > > > > side.
> >> > > > > > > For
> >> > > > > > > > > the
> >> > > > > > > > > > > > > schema
> >> > > > > > > > > > > > > > > >> registry, I
> >> > > > > > > > > > > > > > > >> > > > > > think
> >> > > > > > > > > > > > > > > >> > > > > > > > > that we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > having a
> >> > > > > > > tailored
> >> > > > > > > > > > API to
> >> > > > > > > > > > > > > do
> >> > > > > > > > > > > > > > > >> simple
> >> > > > > > > > > > > > > > > >> > > > > > > > > membership/leader
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > election.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> Best,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> David
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> On
> >> > > Fri,
> >> > > > > Jul
> >> > > > > > > 15,
> >> > > > > > > > > > 2022 at
> >> > > > > > > > > > > > > > 10:22
> >> > > > > > > > > > > > > > > >> AM Ismael
> >> > > > > > > > > > > > > > > >> > > > > > Juma <
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > Three
> >> > > > > quick
> >> > > > > > > > > > comments:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > 1.
> >> > > > > Regarding
> >> > > > > > > > > > > > > > > >> java.util.regex.Pattern
> >> > > > > > > > > > > > > > > >> > > > vs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > com.google.re2j.Pattern,
> >> > > > > > > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > document the
> >> > > > > > > > > > > > differences
> >> > > > > > > > > > > > > > in
> >> > > > > > > > > > > > > > > >> more
> >> > > > > > > > > > > > > > > >> > > > detail
> >> > > > > > > > > > > > > > > >> > > > > > before
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> deciding
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > another.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > That
> >> > > > > said,
> >> > > > > > > if
> >> > > > > > > > > > people
> >> > > > > > > > > > > > > pass
> >> > > > > > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> they
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > expect
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > semantics
> >> > > > > > > to be
> >> > > > > > > > > > > > honored.
> >> > > > > > > > > > > > > > If
> >> > > > > > > > > > > > > > > >> we are
> >> > > > > > > > > > > > > > > >> > > > doing
> >> > > > > > > > > > > > > > > >> > > > > > > > > something
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > different,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > should
> >> > > > > > > consider
> >> > > > > > > > > > adding
> >> > > > > > > > > > > > > an
> >> > > > > > > > > > > > > > > >> overload
> >> > > > > > > > > > > > > > > >> > > > with
> >> > > > > > > > > > > > > > > >> > > > > > our own
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Pattern
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > class
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > think
> >> > > > > we'd
> >> > > > > > > want
> >> > > > > > > > > to
> >> > > > > > > > > > > > > expose
> >> > > > > > > > > > > > > > > >> re2j's at
> >> > > > > > > > > > > > > > > >> > > > this
> >> > > > > > > > > > > > > > > >> > > > > > > > > point).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > 2.
> >> > > > > Regarding
> >> > > > > > > > > topic
> >> > > > > > > > > > > > ids,
> >> > > > > > > > > > > > > > any
> >> > > > > > > > > > > > > > > >> major new
> >> > > > > > > > > > > > > > > >> > > > > > protocol
> >> > > > > > > > > > > > > > > >> > > > > > > > > should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > it
> >> > > and
> >> > > > > > > should
> >> > > > > > > > > > handle
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > topic
> >> > > > > > > > > > > > > > > >> > > > > > recreation case
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > part we
> >> > > > > > > need to
> >> > > > > > > > > > > > handle.
> >> > > > > > > > > > > > > I
> >> > > > > > > > > > > > > > > >> agree with
> >> > > > > > > > > > > > > > > >> > > > > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > that we'd
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > want to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > to
> >> > > the
> >> > > > > > > relevant
> >> > > > > > > > > > > > > protocols
> >> > > > > > > > > > > > > > > >> that don't
> >> > > > > > > > > > > > > > > >> > > > > > have it
> >> > > > > > > > > > > > > > > >> > > > > > > > > yet and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > that we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> probably
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > focus
> >> > > > > on the
> >> > > > > > > > > > internals
> >> > > > > > > > > > > > > > versus
> >> > > > > > > > > > > > > > > >> adding
> >> > > > > > > > > > > > > > > >> > > > new
> >> > > > > > > > > > > > > > > >> > > > > > APIs
> >> > > > > > > > > > > > > > > >> > > > > > > > > to the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > (unless
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > we
> >> > > find
> >> > > > > that
> >> > > > > > > > > > adding
> >> > > > > > > > > > > > new
> >> > > > > > > > > > > > > > APIs
> >> > > > > > > > > > > > > > > >> is
> >> > > > > > > > > > > > > > > >> > > > required
> >> > > > > > > > > > > > > > > >> > > > > > for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> reasonable
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> semantics).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > 3.
> >> > > I am
> >> > > > > > > still
> >> > > > > > > > > not
> >> > > > > > > > > > sure
> >> > > > > > > > > > > > > > about
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > storing
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > powerful for
> >> > > > > > > > > > configs
> >> > > > > > > > > > > > to
> >> > > > > > > > > > > > > be
> >> > > > > > > > > > > > > > > >> > > > centralized
> >> > > > > > > > > > > > > > > >> > > > > > in the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> metadata
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > log for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> various
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > reasons
> >> > > > > > > > > > (auditability,
> >> > > > > > > > > > > > > > > >> visibility,
> >> > > > > > > > > > > > > > > >> > > > > > consistency,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> etc.).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> Similarly, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > sure
> >> > > > > about
> >> > > > > > > > > > > > automatically
> >> > > > > > > > > > > > > > > >> deleting
> >> > > > > > > > > > > > > > > >> > > > > > configs in a
> >> > > > > > > > > > > > > > > >> > > > > > > > > way
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > they
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > recovered. A
> >> > > > > > > > > good
> >> > > > > > > > > > > > > property
> >> > > > > > > > > > > > > > > >> for modern
> >> > > > > > > > > > > > > > > >> > > > > > systems
> >> > > > > > > > > > > > > > > >> > > > > > > > > is to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > minimize
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> number
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > unrecoverable
> >> > > > > > > > > data
> >> > > > > > > > > > > > loss
> >> > > > > > > > > > > > > > > >> scenarios.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > Ismael
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > On
> >> > > Wed,
> >> > > > > Jul
> >> > > > > > > 13,
> >> > > > > > > > > > 2022
> >> > > > > > > > > > > > at
> >> > > > > > > > > > > > > > 3:47
> >> > > > > > > > > > > > > > > >> PM David
> >> > > > > > > > > > > > > > > >> > > > > > Jacot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > Thanks
> >> > > > > > > > > > Guozhang. My
> >> > > > > > > > > > > > > > answers
> >> > > > > > > > > > > > > > > >> are
> >> > > > > > > > > > > > > > > >> > > > below:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > 1)
> >> > > > > the
> >> > > > > > > > > > migration
> >> > > > > > > > > > > > > path,
> >> > > > > > > > > > > > > > > >> especially
> >> > > > > > > > > > > > > > > >> > > > > > the last
> >> > > > > > > > > > > > > > > >> > > > > > > > > step
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > flag
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > to
> >> > > > > > > enable
> >> > > > > > > > > the
> >> > > > > > > > > > new
> >> > > > > > > > > > > > > > > >> protocol, in
> >> > > > > > > > > > > > > > > >> > > > which
> >> > > > > > > > > > > > > > > >> > > > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> both
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > protocols /
> >> > > > > > > > > > rpcs
> >> > > > > > > > > > > > and
> >> > > > > > > > > > > > > > old
> >> > > > > > > > > > > > > > > >> > > > protocols /
> >> > > > > > > > > > > > > > > >> > > > > > rpcs
> >> > > > > > > > > > > > > > > >> > > > > > > > > are
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> used
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> same
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > group.
> >> > > > > > > How
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> coordinator could
> >> > > > > > > > > > > > > > > >> > > > > > "mimic"
> >> > > > > > > > > > > > > > > >> > > > > > > > > the old
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> using
> >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > new
> >> > > > > > > protocol
> >> > > > > > > > > > is
> >> > > > > > > > > > > > > > something
> >> > > > > > > > > > > > > > > >> we
> >> > > > > > > > > > > > > > > >> > > > need to
> >> > > > > > > > > > > > > > > >> > > > > > > > > present
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > Noted. I
> >> > > > > > > just
> >> > > > > > > > > > > > > published
> >> > > > > > > > > > > > > > a
> >> > > > > > > > > > > > > > > >> new
> >> > > > > > > > > > > > > > > >> > > > version
> >> > > > > > > > > > > > > > > >> > > > > > of KIP
> >> > > > > > > > > > > > > > > >> > > > > > > > > which
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > includes
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > details
> >> > > > > > > about
> >> > > > > > > > > > this.
> >> > > > > > > > > > > > > See
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > "Supporting
> >> > > > > > > > > > > > > > > >> > > > > > > > > Online
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> Upgrade"
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > and
> >> > > > > the
> >> > > > > > > > > > > > > "Compatibility,
> >> > > > > > > > > > > > > > > >> > > > Deprecation,
> >> > > > > > > > > > > > > > > >> > > > > > and
> >> > > > > > > > > > > > > > > >> > > > > > > > > Migration
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > I
> >> > > > > have to
> >> > > > > > > > > think
> >> > > > > > > > > > > > > through
> >> > > > > > > > > > > > > > a
> >> > > > > > > > > > > > > > > >> few cases
> >> > > > > > > > > > > > > > > >> > > > > > now but
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > mechanism
> >> > > > > > > > > > should be
> >> > > > > > > > > > > > > > > >> understandable.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > 2)
> >> > > > > the
> >> > > > > > > usage
> >> > > > > > > > > > of
> >> > > > > > > > > > > > > topic
> >> > > > > > > > > > > > > > > >> ids. So
> >> > > > > > > > > > > > > > > >> > > > far as
> >> > > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> used
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > as
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > part of
> >> > > > > > > RPCs
> >> > > > > > > > > > and
> >> > > > > > > > > > > > > admin
> >> > > > > > > > > > > > > > > >> client,
> >> > > > > > > > > > > > > > > >> > > > but
> >> > > > > > > > > > > > > > > >> > > > > > they
> >> > > > > > > > > > > > > > > >> > > > > > > > > are not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> public
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > APIs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > to
> >> > > > > > > consumers
> >> > > > > > > > > > yet.
> >> > > > > > > > > > > > I
> >> > > > > > > > > > > > > > think
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > question is,
> >> > > > > > > > > > > > > > > >> > > > > > > > > first
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > client
> >> > > > > > > to be
> >> > > > > > > > > > > > > > maintaining
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > names
> >> > > > > > > > > > > > > > > >> > > > > > -> ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > mapping
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > leverage
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > on
> >> > > > > all
> >> > > > > > > the
> >> > > > > > > > > > > > augmented
> >> > > > > > > > > > > > > > > >> existing
> >> > > > > > > > > > > > > > > >> > > > RPCs
> >> > > > > > > > > > > > > > > >> > > > > > and the
> >> > > > > > > > > > > > > > > >> > > > > > > > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> ids;
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > secondly,
> >> > > > > > > > > > should
> >> > > > > > > > > > > > we
> >> > > > > > > > > > > > > > ever
> >> > > > > > > > > > > > > > > >> consider
> >> > > > > > > > > > > > > > > >> > > > > > exposing
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > public
> >> > > > > > > APIs
> >> > > > > > > > > as
> >> > > > > > > > > > > > well
> >> > > > > > > > > > > > > > (both
> >> > > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> well
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > rebalance
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > listener for
> >> > > > > > > > > > cases
> >> > > > > > > > > > > > > > like
> >> > > > > > > > > > > > > > > >> topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > deletion-and-recreation).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > a)
> >> > > > > > > Assuming
> >> > > > > > > > > > that we
> >> > > > > > > > > > > > > > would
> >> > > > > > > > > > > > > > > >> include
> >> > > > > > > > > > > > > > > >> > > > > > converting
> >> > > > > > > > > > > > > > > >> > > > > > > > > all
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > offsets
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> related
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > RPCs
> >> > > > > to
> >> > > > > > > using
> >> > > > > > > > > > topic
> >> > > > > > > > > > > > > ids
> >> > > > > > > > > > > > > > in
> >> > > > > > > > > > > > > > > >> this
> >> > > > > > > > > > > > > > > >> > > > KIP,
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > would be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > fully
> >> > > > > > > operate
> >> > > > > > > > > > with
> >> > > > > > > > > > > > > topic
> >> > > > > > > > > > > > > > > >> ids. That
> >> > > > > > > > > > > > > > > >> > > > > > being
> >> > > > > > > > > > > > > > > >> > > > > > > > > said, it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > still has
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> provide
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > the
> >> > > > > topics
> >> > > > > > > > > > names in
> >> > > > > > > > > > > > > > various
> >> > > > > > > > > > > > > > > >> APIs so
> >> > > > > > > > > > > > > > > >> > > > > > having a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> mapping
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > seems
> >> > > > > > > > > > inevitable to
> >> > > > > > > > > > > > > me.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > b) I
> >> > > > > don't
> >> > > > > > > > > have
> >> > > > > > > > > > a
> >> > > > > > > > > > > > > strong
> >> > > > > > > > > > > > > > > >> opinion on
> >> > > > > > > > > > > > > > > >> > > > > > this.
> >> > > > > > > > > > > > > > > >> > > > > > > > > Here I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > beyond the
> >> > > > > > > > > > scope of
> >> > > > > > > > > > > > > this
> >> > > > > > > > > > > > > > > >> KIP. I
> >> > > > > > > > > > > > > > > >> > > > would
> >> > > > > > > > > > > > > > > >> > > > > > rather
> >> > > > > > > > > > > > > > > >> > > > > > > > > focus
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> internals
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > here
> >> > > > > and
> >> > > > > > > we
> >> > > > > > > > > can
> >> > > > > > > > > > > > > consider
> >> > > > > > > > > > > > > > > >> this
> >> > > > > > > > > > > > > > > >> > > > > > separately if
> >> > > > > > > > > > > > > > > >> > > > > > > > > we see
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > value in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > Coming
> >> > > > > > > back to
> >> > > > > > > > > > > > > Ismael's
> >> > > > > > > > > > > > > > > >> point about
> >> > > > > > > > > > > > > > > >> > > > > > using
> >> > > > > > > > > > > > > > > >> > > > > > > > > topic ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> >> > > > > > > > > > > > > > > >> > > > think
> >> > > > > > > > > > > > > > > >> > > > > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > there
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> advantage
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > favour of
> >> > > > > > > it.
> >> > > > > > > > > > The
> >> > > > > > > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> will
> >> > > > > > > > > > > > > > > >> > > > have
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> opportunity
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > the
> >> > > > > topics
> >> > > > > > > > > > exists
> >> > > > > > > > > > > > > before
> >> > > > > > > > > > > > > > > >> passing
> >> > > > > > > > > > > > > > > >> > > > them
> >> > > > > > > > > > > > > > > >> > > > > > into
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > protocol.
> >> > > > > > > > > > Obviously,
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > coordinator
> >> > > > > > > > > > > > > > > >> > > > > > will
> >> > > > > > > > > > > > > > > >> > > > > > > > > also
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> notice
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it but
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > not
> >> > > > > really
> >> > > > > > > > > have
> >> > > > > > > > > > a
> >> > > > > > > > > > > > way
> >> > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> reject an
> >> > > > > > > > > > > > > > > >> > > > > > invalid
> >> > > > > > > > > > > > > > > >> > > > > > > > > topic in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > I'm
> >> > > > > > > agreeing
> >> > > > > > > > > > with
> >> > > > > > > > > > > > > > David
> >> > > > > > > > > > > > > > > >> on all
> >> > > > > > > > > > > > > > > >> > > > other
> >> > > > > > > > > > > > > > > >> > > > > > minor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > `subscribe(Pattern)`
> >> > > > > > > > > > > > > > > >> question:
> >> > > > > > > > > > > > > > > >> > > > > > personally
> >> > > > > > > > > > > > > > > >> > > > > > > > > I think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> necessary
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > deprecate
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > subscribe
> >> > > > > > > > > > > > > > > >> API with
> >> > > > > > > > > > > > > > > >> > > > > > Pattern,
> >> > > > > > > > > > > > > > > >> > > > > > > > > but
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > Pattern
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > while
> >> > > > > > > just
> >> > > > > > > > > > > > > documenting
> >> > > > > > > > > > > > > > > >> that our
> >> > > > > > > > > > > > > > > >> > > > > > > > > subscription may
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected
> >> by
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > server.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > Since
> >> > > > > > > the
> >> > > > > > > > > > > > > incompatible
> >> > > > > > > > > > > > > > > >> case is a
> >> > > > > > > > > > > > > > > >> > > > > > very rare
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > overloaded
> >> > > > > > > > > > > > `String`
> >> > > > > > > > > > > > > > based
> >> > > > > > > > > > > > > > > >> > > > > > subscription may
> >> > > > > > > > > > > > > > > >> > > > > > > > > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> more
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> vulnerable
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > various
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > invalid
> >> > > > > > > > > > regexes.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > That
> >> > > > > could
> >> > > > > > > > > > work. I
> >> > > > > > > > > > > > > have
> >> > > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> look at
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > differences
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > between
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > engines to
> >> > > > > > > > > > better
> >> > > > > > > > > > > > > > > >> understand the
> >> > > > > > > > > > > > > > > >> > > > > > potential
> >> > > > > > > > > > > > > > > >> > > > > > > > > issues.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> My
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > understanding
> >> > > > > is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > that
> >> > > > > would
> >> > > > > > > > > work
> >> > > > > > > > > > for
> >> > > > > > > > > > > > > all
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> basic
> >> > > > > > > > > > > > > > > >> > > > > > regular
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > differences
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > between
> >> > > > > > > the
> >> > > > > > > > > two
> >> > > > > > > > > > are
> >> > > > > > > > > > > > > > mainly
> >> > > > > > > > > > > > > > > >> about
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > various
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> character
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes.
> >> I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > wonder
> >> > > > > > > what
> >> > > > > > > > > > other
> >> > > > > > > > > > > > > people
> >> > > > > > > > > > > > > > > >> think
> >> > > > > > > > > > > > > > > >> > > > about
> >> > > > > > > > > > > > > > > >> > > > > > this.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > Best,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > On
> >> > > > > Tue,
> >> > > > > > > Jul
> >> > > > > > > > > 12,
> >> > > > > > > > > > 2022
> >> > > > > > > > > > > > > at
> >> > > > > > > > > > > > > > > >> 11:28 PM
> >> > > > > > > > > > > > > > > >> > > > > > Guozhang
> >> > > > > > > > > > > > > > > >> > > > > > > > > Wang <
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> > > > > wangguoz@gmail.com
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > Thanks
> >> > > > > > > > > David!
> >> > > > > > > > > > I
> >> > > > > > > > > > > > > think
> >> > > > > > > > > > > > > > on
> >> > > > > > > > > > > > > > > >> the high
> >> > > > > > > > > > > > > > > >> > > > > > level
> >> > > > > > > > > > > > > > > >> > > > > > > > > there are
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > two meta
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> points we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> need
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > to
> >> > > > > > > > > concretize
> >> > > > > > > > > > a
> >> > > > > > > > > > > > bit
> >> > > > > > > > > > > > > > more:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > 1)
> >> > > > > the
> >> > > > > > > > > > migration
> >> > > > > > > > > > > > > path,
> >> > > > > > > > > > > > > > > >> especially
> >> > > > > > > > > > > > > > > >> > > > > > the last
> >> > > > > > > > > > > > > > > >> > > > > > > > > step
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > flag
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > to
> >> > > > > > > enable
> >> > > > > > > > > the
> >> > > > > > > > > > new
> >> > > > > > > > > > > > > > > >> protocol, in
> >> > > > > > > > > > > > > > > >> > > > which
> >> > > > > > > > > > > > > > > >> > > > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> both
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > protocols /
> >> > > > > > > > > > rpcs
> >> > > > > > > > > > > > and
> >> > > > > > > > > > > > > > old
> >> > > > > > > > > > > > > > > >> > > > protocols /
> >> > > > > > > > > > > > > > > >> > > > > > rpcs
> >> > > > > > > > > > > > > > > >> > > > > > > > > are
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> used
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> same
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > group.
> >> > > > > > > How
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> coordinator could
> >> > > > > > > > > > > > > > > >> > > > > > "mimic"
> >> > > > > > > > > > > > > > > >> > > > > > > > > the old
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> using
> >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > new
> >> > > > > > > protocol
> >> > > > > > > > > > is
> >> > > > > > > > > > > > > > something
> >> > > > > > > > > > > > > > > >> we
> >> > > > > > > > > > > > > > > >> > > > need to
> >> > > > > > > > > > > > > > > >> > > > > > > > > present
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > 2)
> >> > > > > the
> >> > > > > > > usage
> >> > > > > > > > > > of
> >> > > > > > > > > > > > > topic
> >> > > > > > > > > > > > > > > >> ids. So
> >> > > > > > > > > > > > > > > >> > > > far as
> >> > > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> used
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > as
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > part of
> >> > > > > > > RPCs
> >> > > > > > > > > > and
> >> > > > > > > > > > > > > admin
> >> > > > > > > > > > > > > > > >> client,
> >> > > > > > > > > > > > > > > >> > > > but
> >> > > > > > > > > > > > > > > >> > > > > > they
> >> > > > > > > > > > > > > > > >> > > > > > > > > are not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> public
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > APIs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > to
> >> > > > > > > consumers
> >> > > > > > > > > > yet.
> >> > > > > > > > > > > > I
> >> > > > > > > > > > > > > > think
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > question is,
> >> > > > > > > > > > > > > > > >> > > > > > > > > first
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > client
> >> > > > > > > to be
> >> > > > > > > > > > > > > > maintaining
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > names
> >> > > > > > > > > > > > > > > >> > > > > > -> ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > mapping
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > leverage
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > on
> >> > > > > all
> >> > > > > > > the
> >> > > > > > > > > > > > augmented
> >> > > > > > > > > > > > > > > >> existing
> >> > > > > > > > > > > > > > > >> > > > RPCs
> >> > > > > > > > > > > > > > > >> > > > > > and the
> >> > > > > > > > > > > > > > > >> > > > > > > > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> ids;
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > secondly,
> >> > > > > > > > > > should
> >> > > > > > > > > > > > we
> >> > > > > > > > > > > > > > ever
> >> > > > > > > > > > > > > > > >> consider
> >> > > > > > > > > > > > > > > >> > > > > > exposing
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > public
> >> > > > > > > APIs
> >> > > > > > > > > as
> >> > > > > > > > > > > > well
> >> > > > > > > > > > > > > > (both
> >> > > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> well
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > rebalance
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > listener for
> >> > > > > > > > > > cases
> >> > > > > > > > > > > > > > like
> >> > > > > > > > > > > > > > > >> topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > deletion-and-recreation).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > I'm
> >> > > > > > > agreeing
> >> > > > > > > > > > with
> >> > > > > > > > > > > > > > David
> >> > > > > > > > > > > > > > > >> on all
> >> > > > > > > > > > > > > > > >> > > > other
> >> > > > > > > > > > > > > > > >> > > > > > minor
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > `subscribe(Pattern)`
> >> > > > > > > > > > > > > > > >> question:
> >> > > > > > > > > > > > > > > >> > > > > > personally
> >> > > > > > > > > > > > > > > >> > > > > > > > > I think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> necessary
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > deprecate
> >> > > > > > > > > the
> >> > > > > > > > > > > > > > subscribe
> >> > > > > > > > > > > > > > > >> API with
> >> > > > > > > > > > > > > > > >> > > > > > Pattern,
> >> > > > > > > > > > > > > > > >> > > > > > > > > but
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > Pattern
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > while
> >> > > > > > > just
> >> > > > > > > > > > > > > documenting
> >> > > > > > > > > > > > > > > >> that our
> >> > > > > > > > > > > > > > > >> > > > > > > > > subscription may
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected
> >> by
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > server.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > Since
> >> > > > > > > the
> >> > > > > > > > > > > > > incompatible
> >> > > > > > > > > > > > > > > >> case is a
> >> > > > > > > > > > > > > > > >> > > > > > very rare
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > overloaded
> >> > > > > > > > > > > > `String`
> >> > > > > > > > > > > > > > based
> >> > > > > > > > > > > > > > > >> > > > > > subscription may
> >> > > > > > > > > > > > > > > >> > > > > > > > > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> more
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> vulnerable
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > various
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > invalid
> >> > > > > > > > > > regexes.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > Guozhang
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > On
> >> > > > > Tue,
> >> > > > > > > Jul
> >> > > > > > > > > > 12,
> >> > > > > > > > > > > > 2022
> >> > > > > > > > > > > > > > at
> >> > > > > > > > > > > > > > > >> 5:23 AM
> >> > > > > > > > > > > > > > > >> > > > > > David Jacot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > > > > > > > > > > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > Hi
> >> > > > > > > Ismael,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > Thanks for
> >> > > > > > > > > > your
> >> > > > > > > > > > > > > > > >> feedback. Let
> >> > > > > > > > > > > > > > > >> > > > me
> >> > > > > > > > > > > > > > > >> > > > > > answer
> >> > > > > > > > > > > > > > > >> > > > > > > > > your
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > questions
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 1. I
> >> > > > > > > > > think
> >> > > > > > > > > > > > it's
> >> > > > > > > > > > > > > > > >> premature to
> >> > > > > > > > > > > > > > > >> > > > > > talk about
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> target
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > deprecation
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > removal
> >> > > > > > > > > > of the
> >> > > > > > > > > > > > > > > >> existing group
> >> > > > > > > > > > > > > > > >> > > > > > protocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Unlike
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > affects a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > core
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > client
> >> > > > > > > > > > > > protocol
> >> > > > > > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> hence
> >> > > > > > > > > > > > > > > >> > > > > > > > > deprecation/removal
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > dependent
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > how
> >> > > > > > > > > > quickly
> >> > > > > > > > > > > > > > > >> applications
> >> > > > > > > > > > > > > > > >> > > > migrate
> >> > > > > > > > > > > > > > > >> > > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > the new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > That
> >> > > > > > > makes
> >> > > > > > > > > > > > sense.
> >> > > > > > > > > > > > > I
> >> > > > > > > > > > > > > > will
> >> > > > > > > > > > > > > > > >> > > > remove it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 2.
> >> > > > > > > The
> >> > > > > > > > > KIP
> >> > > > > > > > > > > > says
> >> > > > > > > > > > > > > we
> >> > > > > > > > > > > > > > > >> intend to
> >> > > > > > > > > > > > > > > >> > > > > > release
> >> > > > > > > > > > > > > > > >> > > > > > > > > this in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> wasn't
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> made
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > clear
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > why. If
> >> > > > > > > > > we
> >> > > > > > > > > > > > added
> >> > > > > > > > > > > > > > that
> >> > > > > > > > > > > > > > > >> as a
> >> > > > > > > > > > > > > > > >> > > > way to
> >> > > > > > > > > > > > > > > >> > > > > > > > > estimate
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > we'd
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> deprecate
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > remove
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > the
> >> > > > > > > > > group
> >> > > > > > > > > > > > > > protocol, I
> >> > > > > > > > > > > > > > > >> also
> >> > > > > > > > > > > > > > > >> > > > > > suggest
> >> > > > > > > > > > > > > > > >> > > > > > > > > removing
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this part.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Let me
> >> > > > > > > > > > explain
> >> > > > > > > > > > > > my
> >> > > > > > > > > > > > > > > >> reasoning. As
> >> > > > > > > > > > > > > > > >> > > > > > > > > explained, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> plan
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > coordinator
> >> > > > > > > > > > in
> >> > > > > > > > > > > > > Java
> >> > > > > > > > > > > > > > > >> while we
> >> > > > > > > > > > > > > > > >> > > > > > implement
> >> > > > > > > > > > > > > > > >> > > > > > > > > the new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > that
> >> > > > > > > the
> >> > > > > > > > > > > > internals
> >> > > > > > > > > > > > > > will
> >> > > > > > > > > > > > > > > >> be
> >> > > > > > > > > > > > > > > >> > > > slightly
> >> > > > > > > > > > > > > > > >> > > > > > > > > different
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > model).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > Therefore, I
> >> > > > > > > > > > > > > wanted
> >> > > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> tighten
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > switch from
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the old
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > coordinator
> >> > > > > > > > > > to
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > new
> >> > > > > > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > to a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > major
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > alternative
> >> > > > > > > > > > > > would
> >> > > > > > > > > > > > > > be to
> >> > > > > > > > > > > > > > > >> use a
> >> > > > > > > > > > > > > > > >> > > > flag
> >> > > > > > > > > > > > > > > >> > > > > > to do
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > relying
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > on the
> >> > > > > > > > > > software
> >> > > > > > > > > > > > > > upgrade.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 3.
> >> > > > > > > We
> >> > > > > > > > > > need to
> >> > > > > > > > > > > > > > flesh
> >> > > > > > > > > > > > > > > >> out the
> >> > > > > > > > > > > > > > > >> > > > > > details of
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > migration
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> sounds
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > like
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > we're
> >> > > > > > > > > > saying
> >> > > > > > > > > > > > we
> >> > > > > > > > > > > > > > will
> >> > > > > > > > > > > > > > > >> support
> >> > > > > > > > > > > > > > > >> > > > > > online
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> migrations.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> correct?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> We
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > explain
> >> > > > > > > > > > this
> >> > > > > > > > > > > > in
> >> > > > > > > > > > > > > > > >> detail. It
> >> > > > > > > > > > > > > > > >> > > > could
> >> > > > > > > > > > > > > > > >> > > > > > also
> >> > > > > > > > > > > > > > > >> > > > > > > > > be done
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> if
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > it's
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > easier.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Yes,
> >> > > > > > > we
> >> > > > > > > > > will
> >> > > > > > > > > > > > > support
> >> > > > > > > > > > > > > > > >> online
> >> > > > > > > > > > > > > > > >> > > > > > migrations
> >> > > > > > > > > > > > > > > >> > > > > > > > > for the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > group.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > a
> >> > > > > > > group
> >> > > > > > > > > > using
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > old
> >> > > > > > > > > > > > > > > >> protocol
> >> > > > > > > > > > > > > > > >> > > > > > will be
> >> > > > > > > > > > > > > > > >> > > > > > > > > able to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > protocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Let me
> >> > > > > > > > > > briefly
> >> > > > > > > > > > > > > > explain
> >> > > > > > > > > > > > > > > >> how that
> >> > > > > > > > > > > > > > > >> > > > > > will work
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> though.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > It is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> basically
> >> > > a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > four
> >> > > > > > > step
> >> > > > > > > > > > > > process:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > 1. The
> >> > > > > > > > > > cluster
> >> > > > > > > > > > > > > must
> >> > > > > > > > > > > > > > be
> >> > > > > > > > > > > > > > > >> > > > upgraded or
> >> > > > > > > > > > > > > > > >> > > > > > > > > rolled to a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > software
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> supporting
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > new
> >> > > > > > > group
> >> > > > > > > > > > > > > > coordinator.
> >> > > > > > > > > > > > > > > >> Both the
> >> > > > > > > > > > > > > > > >> > > > > > old and
> >> > > > > > > > > > > > > > > >> > > > > > > > > the new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > support
> >> > > > > > > > > the
> >> > > > > > > > > > old
> >> > > > > > > > > > > > > > > >> protocol and
> >> > > > > > > > > > > > > > > >> > > > rely
> >> > > > > > > > > > > > > > > >> > > > > > on the
> >> > > > > > > > > > > > > > > >> > > > > > > > > same
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > persisted
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> metadata
> >> > > so
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > they
> >> > > > > > > can
> >> > > > > > > > > > work
> >> > > > > > > > > > > > > > together.
> >> > > > > > > > > > > > > > > >> This
> >> > > > > > > > > > > > > > > >> > > > point
> >> > > > > > > > > > > > > > > >> > > > > > is an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> offline
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> migration.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> cannot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > do
> >> > > > > > > this
> >> > > > > > > > > one
> >> > > > > > > > > > live
> >> > > > > > > > > > > > > > > >> because it
> >> > > > > > > > > > > > > > > >> > > > would
> >> > > > > > > > > > > > > > > >> > > > > > require
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> current
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > coordinator
> >> > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> starting up the
> >> > > > > > > > > > > > > > > >> > > > > > new one
> >> > > > > > > > > > > > > > > >> > > > > > > > > and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > unavailabilities.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > 2. The
> >> > > > > > > > > > cluster's
> >> > > > > > > > > > > > > > > >> metadata
> >> > > > > > > > > > > > > > > >> > > > > > version/IBP
> >> > > > > > > > > > > > > > > >> > > > > > > > > must be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> order
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > to
> >> > > > > > > enable
> >> > > > > > > > > > the
> >> > > > > > > > > > > > new
> >> > > > > > > > > > > > > > > >> protocol.
> >> > > > > > > > > > > > > > > >> > > > This
> >> > > > > > > > > > > > > > > >> > > > > > cannot
> >> > > > > > > > > > > > > > > >> > > > > > > > > be done
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > terminated
> >> > > > > > > > > > > > because
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> old
> >> > > > > > > > > > > > > > > >> > > > > > coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > doesn't
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > support
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > protocol.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > 3. The
> >> > > > > > > > > > consumers
> >> > > > > > > > > > > > > > must be
> >> > > > > > > > > > > > > > > >> > > > upgraded
> >> > > > > > > > > > > > > > > >> > > > > > to a
> >> > > > > > > > > > > > > > > >> > > > > > > > > version
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> online
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > migration
> >> > > > > > > > > > (must
> >> > > > > > > > > > > > > have
> >> > > > > > > > > > > > > > > >> KIP-792).
> >> > > > > > > > > > > > > > > >> > > > If
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> there.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > Nothing
> >> > > > > > > > > > must be
> >> > > > > > > > > > > > > > done at
> >> > > > > > > > > > > > > > > >> this
> >> > > > > > > > > > > > > > > >> > > > point.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > 4. The
> >> > > > > > > > > > consumers
> >> > > > > > > > > > > > > > must
> >> > > > > > > > > > > > > > > >> be rolled
> >> > > > > > > > > > > > > > > >> > > > > > with the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> The
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > consumer
> >> > > > > > > > > > group
> >> > > > > > > > > > > > is
> >> > > > > > > > > > > > > > > >> automatically
> >> > > > > > > > > > > > > > > >> > > > > > > > > converted when
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the first
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > using
> >> > > > > > > the
> >> > > > > > > > > > new
> >> > > > > > > > > > > > > > protocol
> >> > > > > > > > > > > > > > > >> joins
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > group.
> >> > > > > > > > > > > > > > > >> > > > > > > > > While
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> using the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > old
> >> > > > > > > > > > protocol are
> >> > > > > > > > > > > > > > being
> >> > > > > > > > > > > > > > > >> > > > upgraded,
> >> > > > > > > > > > > > > > > >> > > > > > the old
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > new
> >> > > > > > > one.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Let me
> >> > > > > > > > > > clarify
> >> > > > > > > > > > > > all
> >> > > > > > > > > > > > > > of
> >> > > > > > > > > > > > > > > >> this in
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > KIP.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 4.
> >> > > > > > > I am
> >> > > > > > > > > > happy
> >> > > > > > > > > > > > > > that we
> >> > > > > > > > > > > > > > > >> are
> >> > > > > > > > > > > > > > > >> > > > > > pushing the
> >> > > > > > > > > > > > > > > >> > > > > > > > > pattern
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> subscriptions
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > server,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > but
> >> > > > > > > it
> >> > > > > > > > > > seems
> >> > > > > > > > > > > > > like
> >> > > > > > > > > > > > > > > >> there
> >> > > > > > > > > > > > > > > >> > > > could be
> >> > > > > > > > > > > > > > > >> > > > > > some
> >> > > > > > > > > > > > > > > >> > > > > > > > > tricky
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> compatibility
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > issues.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > Will
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > have a
> >> > > > > > > > > > > > mechanism
> >> > > > > > > > > > > > > > for
> >> > > > > > > > > > > > > > > >> users to
> >> > > > > > > > > > > > > > > >> > > > > > detect
> >> > > > > > > > > > > > > > > >> > > > > > > > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> they
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> their
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > regex
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > before
> >> > > > > > > > > > > > switching
> >> > > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> the new
> >> > > > > > > > > > > > > > > >> > > > > > protocol?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > I
> >> > > > > > > think
> >> > > > > > > > > > that I
> >> > > > > > > > > > > > am
> >> > > > > > > > > > > > > a
> >> > > > > > > > > > > > > > bit
> >> > > > > > > > > > > > > > > >> more
> >> > > > > > > > > > > > > > > >> > > > > > optimistic
> >> > > > > > > > > > > > > > > >> > > > > > > > > than
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> you
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> point. I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > believe
> >> > > > > > > > > > that the
> >> > > > > > > > > > > > > > > >> majority of
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > cases
> >> > > > > > > > > > > > > > > >> > > > > > > > > are
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> simple
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > work
> >> > > > > > > with
> >> > > > > > > > > > the
> >> > > > > > > > > > > > new
> >> > > > > > > > > > > > > > > >> engine. The
> >> > > > > > > > > > > > > > > >> > > > > > > > > coordinator will
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > verify
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> anyway
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > and
> >> > > > > > > reject
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> if the
> >> > > > > > > > > > > > > > > >> > > > > > regex is
> >> > > > > > > > > > > > > > > >> > > > > > > > > not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> valid.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > migration
> >> > > > > > > > > > path,
> >> > > > > > > > > > > > in
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> worst
> >> > > > > > > > > > > > > > > >> > > > case,
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > first
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > joining
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > the
> >> > > > > > > group
> >> > > > > > > > > > will
> >> > > > > > > > > > > > be
> >> > > > > > > > > > > > > > > >> rejected.
> >> > > > > > > > > > > > > > > >> > > > This
> >> > > > > > > > > > > > > > > >> > > > > > should
> >> > > > > > > > > > > > > > > >> > > > > > > > > be used
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > defence, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > would
> >> > > > > > > say.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > One
> >> > > > > > > way
> >> > > > > > > > > for
> >> > > > > > > > > > > > > > customers to
> >> > > > > > > > > > > > > > > >> > > > validate
> >> > > > > > > > > > > > > > > >> > > > > > their
> >> > > > > > > > > > > > > > > >> > > > > > > > > regex
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> upgrading
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> their
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > prod
> >> > > > > > > would
> >> > > > > > > > > > be to
> >> > > > > > > > > > > > > > test
> >> > > > > > > > > > > > > > > >> them with
> >> > > > > > > > > > > > > > > >> > > > > > another
> >> > > > > > > > > > > > > > > >> > > > > > > > > group.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> For
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > could
> >> > > > > > > be
> >> > > > > > > > > > done
> >> > > > > > > > > > > > in a
> >> > > > > > > > > > > > > > > >> pre-prod
> >> > > > > > > > > > > > > > > >> > > > > > environment.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Another
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > way
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > extend the
> >> > > > > > > > > > > > > > > >> consumer-group tool
> >> > > > > > > > > > > > > > > >> > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > provide a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> regex
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> >> validation
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > mechanism.
> >> > > > > > > > > > Would
> >> > > > > > > > > > > > > > this be
> >> > > > > > > > > > > > > > > >> > > > enough in
> >> > > > > > > > > > > > > > > >> > > > > > your
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> opinion?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 5.
> >> > > > > > > > > > Related to
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > last
> >> > > > > > > > > > > > > > > >> > > > question,
> >> > > > > > > > > > > > > > > >> > > > > > will
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Java
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > users
> >> > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > stick
> >> > > > > > > > > > with the
> >> > > > > > > > > > > > > > > >> current regex
> >> > > > > > > > > > > > > > > >> > > > > > engine for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> reasons?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> For
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > example,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > it
> >> > > > > > > may
> >> > > > > > > > > be
> >> > > > > > > > > > > > handy
> >> > > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> keep using
> >> > > > > > > > > > > > > > > >> > > > > > client
> >> > > > > > > > > > > > > > > >> > > > > > > > > based
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > regex at
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > migrations
> >> > > > > > > > > > > > > simple
> >> > > > > > > > > > > > > > and
> >> > > > > > > > > > > > > > > >> then
> >> > > > > > > > > > > > > > > >> > > > > > migrate to
> >> > > > > > > > > > > > > > > >> > > > > > > > > server
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > based
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > second
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > step.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > I
> >> > > > > > > > > understand
> >> > > > > > > > > > > > your
> >> > > > > > > > > > > > > > point
> >> > > > > > > > > > > > > > > >> but I
> >> > > > > > > > > > > > > > > >> > > > am
> >> > > > > > > > > > > > > > > >> > > > > > > > > concerned that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> users
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > to
> >> > > > > > > > > actually
> >> > > > > > > > > > stay
> >> > > > > > > > > > > > > in
> >> > > > > > > > > > > > > > > >> this mode.
> >> > > > > > > > > > > > > > > >> > > > That
> >> > > > > > > > > > > > > > > >> > > > > > > > > would go
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > against our
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > simplifying
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > client
> >> > > > > > > > > > > > > > > >> because
> >> > > > > > > > > > > > > > > >> > > > we
> >> > > > > > > > > > > > > > > >> > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > have to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > continue
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> monitoring
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > the
> >> > > > > > > > > > metadata on
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> client
> >> > > > > > > > > > > > > > > >> > > > side. I
> >> > > > > > > > > > > > > > > >> > > > > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > rather
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > not do
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 6.
> >> > > > > > > When
> >> > > > > > > > > > we say
> >> > > > > > > > > > > > > > that
> >> > > > > > > > > > > > > > > >> the group
> >> > > > > > > > > > > > > > > >> > > > > > > > > coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> will
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > responsible
> >> > > > > for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > storing
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > the
> >> > > > > > > > > > > > > configurations
> >> > > > > > > > > > > > > > > >> and that
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> configurations
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> deleted
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> when
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > group is
> >> > > > > > > > > > > > > deleted.
> >> > > > > > > > > > > > > > > >> Will a
> >> > > > > > > > > > > > > > > >> > > > > > transition to
> >> > > > > > > > > > > > > > > >> > > > > > > > > DEAD
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > trigger
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> >> deletion
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > > > configurations?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > That's
> >> > > > > > > > > > right.
> >> > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> > > > configurations
> >> > > > > > > > > > > > > > > >> > > > > > will be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> deleted
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > when
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> group is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > deleted.
> >> > > > > > > > > > They go
> >> > > > > > > > > > > > > > > >> together.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 7.
> >> > > > > > > Will
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > choice to
> >> > > > > > > > > > > > > > > >> store
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > configs in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > make
> >> > > > > it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > harder
> >> > > > > > > > > to
> >> > > > > > > > > > list
> >> > > > > > > > > > > > > all
> >> > > > > > > > > > > > > > > >> cluster
> >> > > > > > > > > > > > > > > >> > > > > > configs and
> >> > > > > > > > > > > > > > > >> > > > > > > > > their
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > values?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > I
> >> > > > > > > don't
> >> > > > > > > > > > think
> >> > > > > > > > > > > > so.
> >> > > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> group
> >> > > > > > > > > > > > > > > >> > > > > > > > > configurations are
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > overrides
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> cluster
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > configs.
> >> > > > > > > > > If
> >> > > > > > > > > > you
> >> > > > > > > > > > > > > > want to
> >> > > > > > > > > > > > > > > >> know
> >> > > > > > > > > > > > > > > >> > > > all
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > overrides
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > though,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> have
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > to
> >> > > > > > > ask all
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > group
> >> > > > > > > > > > > > > > > >> > > > coordinators.
> >> > > > > > > > > > > > > > > >> > > > > > You
> >> > > > > > > > > > > > > > > >> > > > > > > > > cannot
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > rely on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> metadata
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> log
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > for
> >> > > > > > > > > > instance.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 8.
> >> > > > > > > How
> >> > > > > > > > > > would
> >> > > > > > > > > > > > > > someone
> >> > > > > > > > > > > > > > > >> > > > configure a
> >> > > > > > > > > > > > > > > >> > > > > > group
> >> > > > > > > > > > > > > > > >> > > > > > > > > before
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > starting
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > consumers?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Have
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > we
> >> > > > > > > > > > considered
> >> > > > > > > > > > > > > > > >> allowing the
> >> > > > > > > > > > > > > > > >> > > > > > explicit
> >> > > > > > > > > > > > > > > >> > > > > > > > > creation
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > > > > > Alternatively,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > configs
> >> > > > > > > > > > could
> >> > > > > > > > > > > > be
> >> > > > > > > > > > > > > > > >> decoupled
> >> > > > > > > > > > > > > > > >> > > > from
> >> > > > > > > > > > > > > > > >> > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Yes.
> >> > > > > > > The
> >> > > > > > > > > > group
> >> > > > > > > > > > > > > will
> >> > > > > > > > > > > > > > be
> >> > > > > > > > > > > > > > > >> > > > > > automatically
> >> > > > > > > > > > > > > > > >> > > > > > > > > created in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> However,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > the
> >> > > > > > > > > configs
> >> > > > > > > > > > will
> >> > > > > > > > > > > > > be
> >> > > > > > > > > > > > > > > >> lost after
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > retention
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > period of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > passes.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 9.
> >> > > > > > > Will
> >> > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > Consumer.subscribe
> >> > > > > > > > > > > > > > > >> > > > > > method
> >> > > > > > > > > > > > > > > >> > > > > > > > > for the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> take
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > > > > > > >> `java.util.regex.Pattern` of
> >> > > > > > > > > > > > > > > >> > > > do
> >> > > > > > > > > > > > > > > >> > > > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > have to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > introduce
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> overload?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > That's a
> >> > > > > > > > > > very
> >> > > > > > > > > > > > > group
> >> > > > > > > > > > > > > > > >> question. I
> >> > > > > > > > > > > > > > > >> > > > > > forgot
> >> > > > > > > > > > > > > > > >> > > > > > > > > about
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> `java.util.regex.Pattern` is
> >> > > > > > > > > > > > > > > >> > > > not
> >> > > > > > > > > > > > > > > >> > > > > > fully
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> compatible
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> engine
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > we
> >> > > > > > > plan to
> >> > > > > > > > > > use,
> >> > > > > > > > > > > > it
> >> > > > > > > > > > > > > > > >> might be
> >> > > > > > > > > > > > > > > >> > > > better
> >> > > > > > > > > > > > > > > >> > > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > deprecate
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > overload
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > which
> >> > > > > > > > > takes
> >> > > > > > > > > > a
> >> > > > > > > > > > > > > > string.
> >> > > > > > > > > > > > > > > >> We would
> >> > > > > > > > > > > > > > > >> > > > > > rely on
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> server
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > validation.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > During the
> >> > > > > > > > > > > > > > migration, I
> >> > > > > > > > > > > > > > > >> think
> >> > > > > > > > > > > > > > > >> > > > that
> >> > > > > > > > > > > > > > > >> > > > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > could
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> still
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > try to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> toString
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > regex
> >> > > > > > > and
> >> > > > > > > > > > use
> >> > > > > > > > > > > > it.
> >> > > > > > > > > > > > > > That
> >> > > > > > > > > > > > > > > >> should
> >> > > > > > > > > > > > > > > >> > > > > > work, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > think, in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > cases.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 10.
> >> > > > > > > I
> >> > > > > > > > > > agree
> >> > > > > > > > > > > > with
> >> > > > > > > > > > > > > > > >> Justine
> >> > > > > > > > > > > > > > > >> > > > that we
> >> > > > > > > > > > > > > > > >> > > > > > > > > should be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > clearer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> reason
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > switch
> >> > > > > > > > > to
> >> > > > > > > > > > > > > > > >> > > > IBP/metadata.version
> >> > > > > > > > > > > > > > > >> > > > > > from the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> mean
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > can
> >> > > > > > > > > > switch the
> >> > > > > > > > > > > > > > > >> default for
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > feature
> >> > > > > > > > > > > > > > > >> > > > > > > > > flag
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > true
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > > > metadata.version
> >> > > > > > > > > > > > > > once
> >> > > > > > > > > > > > > > > >> we
> >> > > > > > > > > > > > > > > >> > > > want to
> >> > > > > > > > > > > > > > > >> > > > > > make
> >> > > > > > > > > > > > > > > >> > > > > > > > > it the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > default.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > My
> >> > > > > > > plan
> >> > > > > > > > > was
> >> > > > > > > > > > to
> >> > > > > > > > > > > > use
> >> > > > > > > > > > > > > > that
> >> > > > > > > > > > > > > > > >> feature
> >> > > > > > > > > > > > > > > >> > > > > > flag
> >> > > > > > > > > > > > > > > >> > > > > > > > > mainly
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > during the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > development
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > phase. I
> >> > > > > > > > > > should
> >> > > > > > > > > > > > > not
> >> > > > > > > > > > > > > > have
> >> > > > > > > > > > > > > > > >> > > > mentioned
> >> > > > > > > > > > > > > > > >> > > > > > it, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > think,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > because
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> use
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > an
> >> > > > > > > > > internal
> >> > > > > > > > > > > > config
> >> > > > > > > > > > > > > > for
> >> > > > > > > > > > > > > > > >> it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 11.
> >> > > > > > > Some
> >> > > > > > > > > > of
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> protocol APIs
> >> > > > > > > > > > > > > > > >> > > > > > don't
> >> > > > > > > > > > > > > > > >> > > > > > > > > mention
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> ACLs, it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > would
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > good to
> >> > > > > > > > > > add
> >> > > > > > > > > > > > that
> >> > > > > > > > > > > > > > for
> >> > > > > > > > > > > > > > > >> > > > consistency.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Noted.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 12.
> >> > > > > > > It
> >> > > > > > > > > is
> >> > > > > > > > > > a
> >> > > > > > > > > > > > bit
> >> > > > > > > > > > > > > > odd
> >> > > > > > > > > > > > > > > >> that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> ConsumerGroupHeartbeat
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> Group"
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > even
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > though
> >> > > > > > > > > it
> >> > > > > > > > > > > > seems
> >> > > > > > > > > > > > > > to do
> >> > > > > > > > > > > > > > > >> more
> >> > > > > > > > > > > > > > > >> > > > than
> >> > > > > > > > > > > > > > > >> > > > > > > > > reading.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > I
> >> > > > > > > agree.
> >> > > > > > > > > > This is
> >> > > > > > > > > > > > > > how the
> >> > > > > > > > > > > > > > > >> > > > current
> >> > > > > > > > > > > > > > > >> > > > > > > > > protocol works
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > though.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > require
> >> > > > > > > > > > "Read
> >> > > > > > > > > > > > > > Group" to
> >> > > > > > > > > > > > > > > >> join a
> >> > > > > > > > > > > > > > > >> > > > > > group. We
> >> > > > > > > > > > > > > > > >> > > > > > > > > could
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > consider
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> changing
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> this
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > but I
> >> > > > > > > am
> >> > > > > > > > > not
> >> > > > > > > > > > > > sure
> >> > > > > > > > > > > > > > that
> >> > > > > > > > > > > > > > > >> it is
> >> > > > > > > > > > > > > > > >> > > > worth
> >> > > > > > > > > > > > > > > >> > > > > > it.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 13.
> >> > > > > > > How
> >> > > > > > > > > is
> >> > > > > > > > > > > > topic
> >> > > > > > > > > > > > > > > >> recreation
> >> > > > > > > > > > > > > > > >> > > > > > handled by
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > consumer
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > group
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > protocol?
> >> > > > > > > > > > It
> >> > > > > > > > > > > > > > would be
> >> > > > > > > > > > > > > > > >> good to
> >> > > > > > > > > > > > > > > >> > > > > > have a
> >> > > > > > > > > > > > > > > >> > > > > > > > > section
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> on
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Noted.
> >> > > > > > > > > From
> >> > > > > > > > > > a
> >> > > > > > > > > > > > > > protocol
> >> > > > > > > > > > > > > > > >> > > > > > perspective, the
> >> > > > > > > > > > > > > > > >> > > > > > > > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > topic
> >> > > > > > > id
> >> > > > > > > > > so
> >> > > > > > > > > > it
> >> > > > > > > > > > > > > will
> >> > > > > > > > > > > > > > > >> treat it
> >> > > > > > > > > > > > > > > >> > > > like a
> >> > > > > > > > > > > > > > > >> > > > > > > > > topic with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> The
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > only
> >> > > > > > > issue
> >> > > > > > > > > > is
> >> > > > > > > > > > > > that
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > fetch/commit
> >> > > > > > > > > > > > > > > >> > > > > > > > > offsets
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> APIs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > do not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > IDs
> >> > > > > > > so the
> >> > > > > > > > > > > > > consumer
> >> > > > > > > > > > > > > > > >> would
> >> > > > > > > > > > > > > > > >> > > > reuse the
> >> > > > > > > > > > > > > > > >> > > > > > > > > offsets
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> based
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > that
> >> > > > > > > we
> >> > > > > > > > > > should
> >> > > > > > > > > > > > > > update
> >> > > > > > > > > > > > > > > >> those
> >> > > > > > > > > > > > > > > >> > > > APIs
> >> > > > > > > > > > > > > > > >> > > > > > as well
> >> > > > > > > > > > > > > > > >> > > > > > > > > in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> order
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> consistent
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> end
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > to
> >> > > > > > > end.
> >> > > > > > > > > That
> >> > > > > > > > > > > > would
> >> > > > > > > > > > > > > > > >> strengthen
> >> > > > > > > > > > > > > > > >> > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > semantics of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 14.
> >> > > > > > > The
> >> > > > > > > > > > KIP
> >> > > > > > > > > > > > > > mentions
> >> > > > > > > > > > > > > > > >> we will
> >> > > > > > > > > > > > > > > >> > > > > > write the
> >> > > > > > > > > > > > > > > >> > > > > > > > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> Even
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > though
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > this is
> >> > > > > > > > > an
> >> > > > > > > > > > > > > > > >> implementation
> >> > > > > > > > > > > > > > > >> > > > > > detail, do
> >> > > > > > > > > > > > > > > >> > > > > > > > > we plan
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > have a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> gradle
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > module
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > for
> >> > > > > > > it?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Yes.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 15.
> >> > > > > > > Do
> >> > > > > > > > > we
> >> > > > > > > > > > > > have a
> >> > > > > > > > > > > > > > > >> scalability
> >> > > > > > > > > > > > > > > >> > > > > > goal when
> >> > > > > > > > > > > > > > > >> > > > > > > > > it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> comes
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to how
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > members
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > new
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > group
> >> > > > > > > > > > protocol
> >> > > > > > > > > > > > > can
> >> > > > > > > > > > > > > > > >> support?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > We
> >> > > > > > > don't
> >> > > > > > > > > > have
> >> > > > > > > > > > > > > > numbers
> >> > > > > > > > > > > > > > > >> at the
> >> > > > > > > > > > > > > > > >> > > > > > moment. The
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> support
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> 1000s
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > of
> >> > > > > > > members
> >> > > > > > > > > > per
> >> > > > > > > > > > > > > > group.
> >> > > > > > > > > > > > > > > >> We will
> >> > > > > > > > > > > > > > > >> > > > > > measure
> >> > > > > > > > > > > > > > > >> > > > > > > > > this when
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > we have
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > implementation.
> >> > > > > > > > > > > > > Note
> >> > > > > > > > > > > > > > > >> that we
> >> > > > > > > > > > > > > > > >> > > > might
> >> > > > > > > > > > > > > > > >> > > > > > have
> >> > > > > > > > > > > > > > > >> > > > > > > > > other
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > road
> >> > > > > > > (e.g.
> >> > > > > > > > > > > > offset
> >> > > > > > > > > > > > > > > >> commits).
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > 16.
> >> > > > > > > Did
> >> > > > > > > > > we
> >> > > > > > > > > > > > > > consider
> >> > > > > > > > > > > > > > > >> having
> >> > > > > > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > > > > > SubscribedTopicNames
> >> > > > > > > > > > > > > > > >> in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> >> > > > > ConsumerGroupHeartbeatRequest?
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> idea
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > since we
> >> > > > > > > > > > have
> >> > > > > > > > > > > > to
> >> > > > > > > > > > > > > > > >> resolve the
> >> > > > > > > > > > > > > > > >> > > > > > regex on
> >> > > > > > > > > > > > > > > >> > > > > > > > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > server, we
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> same
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > the
> >> > > > > > > > > topic
> >> > > > > > > > > > > > name?
> >> > > > > > > > > > > > > > The
> >> > > > > > > > > > > > > > > >> > > > difference
> >> > > > > > > > > > > > > > > >> > > > > > is that
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> sending
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > efficient
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > whereas
> >> > > > > > > > > > > > sending
> >> > > > > > > > > > > > > > the
> >> > > > > > > > > > > > > > > >> topic
> >> > > > > > > > > > > > > > > >> > > > names
> >> > > > > > > > > > > > > > > >> > > > > > is less
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > Furthermore,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > delete
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > recreation is
> >> > > > > > > > > > > > > > easier
> >> > > > > > > > > > > > > > > >> to
> >> > > > > > > > > > > > > > > >> > > > handle
> >> > > > > > > > > > > > > > > >> > > > > > if we
> >> > > > > > > > > > > > > > > >> > > > > > > > > have
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ids.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > The
> >> > > > > > > idea
> >> > > > > > > > > > was to
> >> > > > > > > > > > > > > > > >> consolidate the
> >> > > > > > > > > > > > > > > >> > > > > > metadata
> >> > > > > > > > > > > > > > > >> > > > > > > > > lookup
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> both
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > paths
> >> > > > > > > but
> >> > > > > > > > > I
> >> > > > > > > > > > do
> >> > > > > > > > > > > > > agree
> >> > > > > > > > > > > > > > > >> with your
> >> > > > > > > > > > > > > > > >> > > > > > point. As
> >> > > > > > > > > > > > > > > >> > > > > > > > > a
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> second
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > ids
> >> > > > > > > may be
> >> > > > > > > > > > > > better
> >> > > > > > > > > > > > > > here
> >> > > > > > > > > > > > > > > >> for the
> >> > > > > > > > > > > > > > > >> > > > > > delete and
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > recreation
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> Also, I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > suppose
> >> > > > > > > > > > that we
> >> > > > > > > > > > > > > may
> >> > > > > > > > > > > > > > > >> allow
> >> > > > > > > > > > > > > > > >> > > > users to
> >> > > > > > > > > > > > > > > >> > > > > > > > > subscribe
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> with
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > future
> >> > > > > > > > > > because
> >> > > > > > > > > > > > > that
> >> > > > > > > > > > > > > > is
> >> > > > > > > > > > > > > > > >> the only
> >> > > > > > > > > > > > > > > >> > > > > > way to be
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> really
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > robust
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > re-creation.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > Best,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > On
> >> > > > > > > Tue,
> >> > > > > > > > > Jul
> >> > > > > > > > > > 12,
> >> > > > > > > > > > > > > > 2022 at
> >> > > > > > > > > > > > > > > >> 1:38 PM
> >> > > > > > > > > > > > > > > >> > > > > > David
> >> > > > > > > > > > > > > > > >> > > > > > > > > Jacot <
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > djacot@confluent.io>
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > >
> >> > > wrote:
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > Hi
> >> > > > > > > > > > Justine,
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > Thanks
> >> > > > > > > > > for
> >> > > > > > > > > > > > your
> >> > > > > > > > > > > > > > > >> comments.
> >> > > > > > > > > > > > > > > >> > > > Please
> >> > > > > > > > > > > > > > > >> > > > > > find
> >> > > > > > > > > > > > > > > >> > > > > > > > > my
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> answers
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > -
> >> > > > > > > Yes,
> >> > > > > > > > > > the new
> >> > > > > > > > > > > > > > > >> protocol
> >> > > > > > > > > > > > > > > >> > > > relies on
> >> > > > > > > > > > > > > > > >> > > > > > > > > topic IDs
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> >> exception
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> of the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > > > topic
> >> > > > > > > > > > names
> >> > > > > > > > > > > > > based
> >> > > > > > > > > > > > > > in
> >> > > > > > > > > > > > > > > >> the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > ConsumerGroupHeartbeatRequest.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> sure
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > if
> >> > > > > > > using
> >> > > > > > > > > > topic
> >> > > > > > > > > > > > > > names
> >> > > > > > > > > > > > > > > >> is the
> >> > > > > > > > > > > > > > > >> > > > > > right call
> >> > > > > > > > > > > > > > > >> > > > > > > > > here.
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> I
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> about
> >> > > it
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > >
> >> > > > >
> >> > > > > a
> >> > > > > > > little
> >> > > > > > > > > > more.
> >> > > > > > > > > > > > > > > >> Obviously, the
> >> > > > > > > > > > > > > > > >> > > > > > KIP does
> >> > > > > > > > > > > > > > > >> > > > > > > > > not
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > change the
> >> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> >> > > > > > >
> >> > > > > >
> >> > > > > >
> >> > > > > > --
> >> > > > > > -- Guozhang
> >> > > > >
> >> > > >
> >> > > >
> >> > > > --
> >> > > > -- Guozhang
> >> > >
> >> >
> >> >
> >> > --
> >> > -- Guozhang
> >>
> >
> >
> > --
> > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hi David,

I made another pass on the doc, and assuming we will change the
"TopicIdPartition" fields to "TopicPartition" for the consumer-side APIs, I
think this proposal looks great for me.

Guozhang

On Mon, Oct 3, 2022 at 9:36 AM Guozhang Wang <wa...@gmail.com> wrote:

> 3. Ah thanks, that's my mis-read then. I think keeping the current class
> and the params of `onAssignment` is fine then.
>
> On Mon, Oct 3, 2022 at 4:32 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
>> 3. ConsumerGroupMetadata is already defined in the consumer [1]. I
>> realized that we pass it in the current assignor so I thought that we
>> could reuse it. The alternative would be to add the same information
>> in a class which includes the error and the member assignment as well.
>>
>> [1]
>> https://github.com/apache/kafka/blob/99b9b3e84f4e98c3f07714e1de6a139a004cbc5b/clients/src/main/java/org/apache/kafka/clients/consumer/ConsumerGroupMetadata.java
>>
>> Best,
>> David
>>
>> On Mon, Oct 3, 2022 at 6:15 AM Guozhang Wang <wa...@gmail.com> wrote:
>> >
>> > 1. Yeah that makes sense. I originally concerned if there's any
>> consistency
>> > issues if the HB request with encoded topic IDs and the GPA/GIA RPCs
>> with
>> > encoded topic IDs use mappings from two different brokers, and one of
>> them
>> > is stale, but after thinking twice about it I think as long as we handle
>> > e.g. `unknown topic ids` from HB responses then we should be fine.
>> >
>> > 3. Ack, in that case I think we also need to define
>> "ConsumerGroupMetadata"
>> > which is passed in `onAssignment`. We have a schema defined in the
>> > persistent storage for different purposes, but not in the consumer APIs
>> > yet. I think that would just include a single "Map<String, Metadata>"?
>> >
>> > On Fri, Sep 30, 2022 at 5:51 AM David Jacot <djacot@confluent.io.invalid
>> >
>> > wrote:
>> >
>> > > Hey Guozhang,
>> > >
>> > > Thanks for your comments.
>> > >
>> > > 1. That's a very good question. I think that it is better to separate
>> > > the two concerns here. In my opinion, the GPA should only be used to
>> > > feed the assignor. I see the GPA as a way to call a remote function so
>> > > the GPA provides all its inputs to ensure that the assignor does not
>> > > depend on any local state. Knowing this, I would not use the GPA to
>> > > update the local state either. That is not its purpose. The consumer
>> > > is responsible to update its local metadata cache either when it gets
>> > > an unknown topic in the user-provided subscription or in the
>> > > coordinator-provided assignment. The consumer has to do this anyway to
>> > > get all topic metadata (e.g. leaders).
>> > >
>> > > 2. Interesting thought. Handling the translation is definitely
>> > > possible. I wonder if there are cases where having the topic ids is
>> > > beneficial. I am thinking about when a topic is deleted and recreated
>> > > right away. In this case, the current member assignment would still
>> > > have the old topic id and we would not be able to resolve it anymore.
>> > > I suppose that we should directly remove it from the current
>> > > assignment on the server side in this case and only add the new one.
>> > > As I first thought, that seems OK. The major downside is that the user
>> > > of the assignor would not know that the topic is different from what
>> > > it may have cached locally.
>> > >
>> > > We could say that we don't expose topic ids externally for now in
>> > > order to stay consistent with the other interfaces but it would be
>> > > annoying to add it back if we want to expose topic ids in the future.
>> > > I personally believe that we will have to expose them at some point.
>> > > For instance, in the rebalance listener, it would be useful to know
>> > > that a topic has been deleted and recreated. The only way to do this
>> > > is to know the topic id. Let me think a little more about this and
>> > > come back to you.
>> > >
>> > > 3. Sorry for this one. I was a typo. PartitionAssingor#assign should
>> > > return GroupAssignment.
>> > >
>> > > Best,
>> > > David
>> > >
>> > > On Thu, Sep 29, 2022 at 6:06 PM Guozhang Wang <wa...@gmail.com>
>> wrote:
>> > > >
>> > > > Thanks David, I read through the updated wiki, and I have a couple
>> > > > follow-up questions:
>> > > >
>> > > > 1. It seems we would have two potential sources to update the
>> consumer's
>> > > > internally maintained topic name -> topic ID mapping, which is used
>> for
>> > > > encoding subscribed topic Ids in the HB request, and offset commit
>> / list
>> > > > offset requests, etc. One is from metadata responses, and another
>> one is
>> > > > from the GPA (group prepare assignment). On the other hand, the
>> assignor
>> > > > APIs directly take parsed AssignmentMemberSpec and
>> > > AssignmentTopicMetadata
>> > > > from GPA responses directly instead of trying to consult the
>> consumer's
>> > > own
>> > > > cached mapping. So just to clarify, consumers would also update its
>> > > > internal mapping from the GPA responses as well, right?
>> > > >
>> > > > 2. I'm wondering if it's necessary to expose the topic IDs to the
>> > > assignor
>> > > > developers. What they need to do at the moment is to take the
>> mapping
>> > > from
>> > > > AssignmentTopicMetadata, and translate their decided assignment into
>> > > topic
>> > > > IDs when returning the MemberAssignment. We can instead just let
>> consumer
>> > > > to do such translation for them internally and still let the
>> assignor
>> > > > developers to only deal with the topic names. Of course that means
>> the
>> > > > client side assignor APIs would be a bit further diverged from the
>> > > > broker-side assignor such that:
>> > > >
>> > > > a) Topic metadata would be encoded just as String -> num.partitions;
>> > > > b) AssignmentMemberSpec contains subscribedTopicNames and
>> topicPartitions
>> > > > as target assignments;
>> > > > c) Returned MemberAssignment contains list of topicPartitions
>> instead of
>> > > > topicIdPartitions.
>> > > >
>> > > > Internally, consumers would convert the IDs to names when calling
>> the
>> > > > assign() function, and convert them back when sending the request;
>> and
>> > > > similarly it would convert the IDs to names when calling
>> onAssignment()
>> > > as
>> > > > well.
>> > > >
>> > > > In short, I'm wondering if we should bite the bullet for the users,
>> > > eating
>> > > > more complexities internally than pushing that to users.
>> > > >
>> > > > 3. This is a nit one: the return type of `Assignment
>> > > assign(AssignmentSpec
>> > > > assignmentSpec);` seems not defined in the doc. I think it would
>> just
>> > > > include two fields, as:
>> > > >
>> > > > class Assignment {
>> > > >         GroupAssignment groupAssignment;
>> > > >
>> > > >         ConsumerGroupMetadata consumerGroupMetadata;
>> > > >     }
>> > > >
>> > > > Is that right?
>> > > >
>> > > > Guozhang
>> > > >
>> > > > On Thu, Sep 29, 2022 at 1:26 AM David Jacot
>> <djacot@confluent.io.invalid
>> > > >
>> > > > wrote:
>> > > >
>> > > > > Hi Guozhang,
>> > > > >
>> > > > > Thanks for your input.
>> > > > >
>> > > > > > 1) in HB request, when clients send the subscription, they are
>> still
>> > > > > > sending them in terms of topic names. That is, the consumer
>> would
>> > > not try
>> > > > > > to keep all the mappings from topic names to topic ids for the
>> > > subscribed
>> > > > > > topics.
>> > > > >
>> > > > > This is a mistake of mine. The heartbeat request should use Topic
>> Ids.
>> > > > > I forgot to change it. The idea is that the client will get the
>> > > > > metadata first to translate the topic names to topic ids and thus
>> also
>> > > > > have the mapping in its local metadata cache.
>> > > > >
>> > > > > > 2) in HB responses, the assignment's topics are encoded as IDs
>> > > already.
>> > > > > And
>> > > > > > fetch requests can directly rely on that since they just need
>> topic
>> > > IDs
>> > > > > and
>> > > > > > do not care what actual topic names are.
>> > > > >
>> > > > > That's right.
>> > > > >
>> > > > > > 3) in prepare-assignment responses, we return subscriptions as
>> topic
>> > > IDs
>> > > > > as
>> > > > > > well.
>> > > > >
>> > > > > That's right.
>> > > > >
>> > > > > > In a word, the consumer does not try to maintain the mapping of
>> topic
>> > > > > names
>> > > > > > to IDs at all, and even if they do, they will not expose it in
>> e.g.
>> > > > > > assignor APIs. But for client-side assignors like Streams, if
>> they
>> > > do not
>> > > > > > maintain the mapping from topic IDs to topic names, certain
>> logic
>> > > such as
>> > > > > > co-partitioning indeed cannot be executed.
>> > > > >
>> > > > > As I said, the consumer will have the mapping in its local
>> metadata
>> > > > > cache. It needs to get the partition metadata anyway and the
>> topic id
>> > > > > is part of it. However, I do agree that this mapping is not
>> exposed.
>> > > > >
>> > > > > > So I think we'd need to consider 1) use topic names in
>> subscriptions
>> > > in
>> > > > > the
>> > > > > > prepare-assignment responses, 2) use topic Names in client-side
>> > > > > > PartitionAssignor#GroupMember#subscribedTopics and
>> > > > > > PartitionAssignor#TopicMetadata#topicName.
>> > > > >
>> > > > > 1) makes sense to me. I think that we could actually provide both
>> > > > > names and ids. Regarding 2), I think that I am slightly in favor
>> of
>> > > > > using both names and ids as well. Let me try to update the
>> protocol
>> > > > > and the interface towards this.
>> > > > >
>> > > > > Thanks,
>> > > > > David
>> > > > >
>> > > > > On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wangguoz@gmail.com
>> >
>> > > wrote:
>> > > > > >
>> > > > > > Hello Jun/David,
>> > > > > >
>> > > > > > Just to chime in here for Jun's comment #50, today Streams does
>> not
>> > > have
>> > > > > a
>> > > > > > way to map topic names to topic IDs, and in its topology
>> description
>> > > > > topics
>> > > > > > are encoded with names.
>> > > > > >
>> > > > > > What I originally thought was that, on the consumer side, we'd
>> need
>> > > to
>> > > > > > maintain this mapping from topic names to topic IDs anyways, and
>> > > Streams
>> > > > > > should leverage on the consumer's functionality to get the
>> > > corresponding
>> > > > > > topic names for the given topic IDs. But when I read through
>> the RPC
>> > > > > again
>> > > > > > I realized it may not be the case, since:
>> > > > > >
>> > > > > > 1) in HB request, when clients send the subscription, they are
>> still
>> > > > > > sending them in terms of topic names. That is, the consumer
>> would
>> > > not try
>> > > > > > to keep all the mappings from topic names to topic ids for the
>> > > subscribed
>> > > > > > topics.
>> > > > > > 2) in HB responses, the assignment's topics are encoded as IDs
>> > > already.
>> > > > > And
>> > > > > > fetch requests can directly rely on that since they just need
>> topic
>> > > IDs
>> > > > > and
>> > > > > > do not care what actual topic names are.
>> > > > > > 3) in prepare-assignment responses, we return subscriptions as
>> topic
>> > > IDs
>> > > > > as
>> > > > > > well.
>> > > > > >
>> > > > > > In a word, the consumer does not try to maintain the mapping of
>> topic
>> > > > > names
>> > > > > > to IDs at all, and even if they do, they will not expose it in
>> e.g.
>> > > > > > assignor APIs. But for client-side assignors like Streams, if
>> they
>> > > do not
>> > > > > > maintain the mapping from topic IDs to topic names, certain
>> logic
>> > > such as
>> > > > > > co-partitioning indeed cannot be executed.
>> > > > > >
>> > > > > > So I think we'd need to consider 1) use topic names in
>> subscriptions
>> > > in
>> > > > > the
>> > > > > > prepare-assignment responses, 2) use topic Names in client-side
>> > > > > > PartitionAssignor#GroupMember#subscribedTopics and
>> > > > > > PartitionAssignor#TopicMetadata#topicName.
>> > > > > >
>> > > > > > On Wed, Sep 28, 2022 at 1:26 AM David Jacot
>> > > <djacot@confluent.io.invalid
>> > > > > >
>> > > > > > wrote:
>> > > > > >
>> > > > > > > Hi Jun,
>> > > > > > >
>> > > > > > > Thanks for your comments.
>> > > > > > >
>> > > > > > > 48. Okay. There is a typo in the text:
>> > > > > > > > Note that the persisted target assignment of B already
>> includes
>> > > foo-2
>> > > > > > > here
>> > > > > > > but B does not get it yet because foo-2 is not free.
>> > > > > > >
>> > > > > > > It should be:
>> > > > > > > > Note that the persisted member assignment of B already
>> includes
>> > > foo-2
>> > > > > > > here
>> > > > > > > but B does not get it yet because foo-2 is not free.
>> > > > > > >
>> > > > > > > The examples are showing the persisted state in the
>> > > __consumer_offsets
>> > > > > > > topics. When a member transitions to the next epoch, we
>> directly
>> > > > > > > persist the whole target assignment in order to reduce the
>> number
>> > > of
>> > > > > > > writes and we manage dependencies in memory. So "B - epoch=2,
>> > > > > > > partitions=[foo-2]" is actually correct from a persisted
>> record
>> > > > > > > perspective. However, the in-memory state would be different
>> > > because
>> > > > > > > foo-2 is not yet available as you pointed out. The in-memory
>> state
>> > > > > > > would like as follow:
>> > > > > > >
>> > > > > > > Group (epoch=2)
>> > > > > > > A
>> > > > > > > B
>> > > > > > > Target Assignment (epoch=2)
>> > > > > > > A - partitions=[foo-0, foo-1]
>> > > > > > > B - partitions=[foo-2]
>> > > > > > > Member Assignment
>> > > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending
>> > > partitions=[]
>> > > > > > > B - epoch=2, partitions=[], pending partitions=[foo-2]
>> > > > > > >
>> > > > > > > I suppose that the confusion comes from this. I will go and
>> convert
>> > > > > > > all the examples to their in-memory representation in order to
>> > > reduce
>> > > > > > > the confusion.
>> > > > > > >
>> > > > > > > 50. Okay. I have to admit that I am not so familiar with the
>> > > internals
>> > > > > > > of KStreams. If I understand correctly, you're saying that
>> KStreams
>> > > > > > > may not have topic ids in its topology so topic names are
>> missing
>> > > here
>> > > > > > > to match the topics. I was hoping that KStreams would have
>> topic
>> > > ids
>> > > > > > > in the topology as well. If that is not the case, we can also
>> add
>> > > > > > > topic names on our side. I will check this.
>> > > > > > >
>> > > > > > > 53. It seems that I was not clear. Sorry for that. I did not
>> mean
>> > > to
>> > > > > > > say that MemberAssignment.metadata is not used by the group
>> > > > > > > coordinator. Let me re-explain the flow.
>> > > > > > >
>> > > > > > > PartitionAssignor.assign() returns an Assignment. Assignment
>> > > contains
>> > > > > > > MemberAssignments where each member has
>> MemberAssignment.metadata.
>> > > The
>> > > > > > > Assignment is installed by the client with the
>> > > > > > > ConsumerGroupInstallAssignment API and this overrides the
>> current
>> > > > > > > target assignment in the coordinator. Then the group
>> coordinator
>> > > uses
>> > > > > > > it to reconcile members. Therefore the target assignment will
>> > > > > > > eventually become the current member assignment and it will be
>> > > > > > > delivered to the member, including the original
>> > > > > > > MemberAssignment.metadata defines by the assignor.
>> > > > > > >
>> > > > > > > I hope that this helps to clarify how
>> MemberAssignment.metadata is
>> > > > > > > used afterwards.
>> > > > > > >
>> > > > > > > Best,
>> > > > > > > David
>> > > > > > >
>> > > > > > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao
>> <jun@confluent.io.invalid
>> > > >
>> > > > > wrote:
>> > > > > > > > Hi, David,
>> > > > > > > >
>> > > > > > > > Thanks for the reply. A few more follow up questions.
>> > > > > > > >
>> > > > > > > > 48. Hmm, so member assignment is what the client sees.
>> Then, I
>> > > am not
>> > > > > > > sure
>> > > > > > > > about the following. The text says B cannot get foo-2 yet.
>> But in
>> > > > > Member
>> > > > > > > > Assignment, B has foo-2.
>> > > > > > > >
>> > > > > > > > At this point B can transition to epoch 2 but cannot get
>> foo-2
>> > > until
>> > > > > A
>> > > > > > > > revokes it.
>> > > > > > > >
>> > > > > > > > Note that the persisted target assignment of B already
>> includes
>> > > foo-2
>> > > > > > > here
>> > > > > > > > but B does not get it yet because foo-2 is not free.
>> > > > > > > >
>> > > > > > > > Group (epoch=2)
>> > > > > > > > A
>> > > > > > > > B
>> > > > > > > > Target Assignment (epoch=2)
>> > > > > > > > A - partitions=[foo-0, foo-1]
>> > > > > > > > B - partitions=[foo-2]
>> > > > > > > > Member Assignment
>> > > > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
>> > > > > > > > B - epoch=2, partitions=[foo-2]
>> > > > > > > >
>> > > > > > > >
>> > > > > > > > 50. Hmm, I thought KStreams doesn't use range assignor all
>> the
>> > > time.
>> > > > > It
>> > > > > > > > does the assignment based on the topology. For example, if
>> you
>> > > are
>> > > > > > > joining
>> > > > > > > > two topics A and B and deriving the join result in a 3rd
>> topic C,
>> > > > > which
>> > > > > > > > will be aggregated. You want to use range to co-locate the
>> > > partitions
>> > > > > > > from
>> > > > > > > > topic A and B. However, for topic C, the partition could
>> > > potentially
>> > > > > be
>> > > > > > > > assigned in round robin. Without the topic name, KStreams
>> won't
>> > > know
>> > > > > > > which
>> > > > > > > > topics need to be co-located.
>> > > > > > > >
>> > > > > > > > 53. If MemberAssignment.metadata is not used by the group
>> > > > > coordinator,
>> > > > > > > > where is it being used?
>> > > > > > > >
>> > > > > > > > Thanks,
>> > > > > > > >
>> > > > > > > > Jun
>> > > > > > > >
>> > > > > > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
>> > > > > > > <ja...@confluent.io.invalid>
>> > > > > > > > wrote:
>> > > > > > > >
>> > > > > > > > > Hi David,
>> > > > > > > > >
>> > > > > > > > > Thanks for the response. Replied below:
>> > > > > > > > >
>> > > > > > > > > > I don't have a strong opinion on this one. However, it
>> seems
>> > > that
>> > > > > > > this
>> > > > > > > > > could be a pandora box if our subset turns out to not be
>> > > sufficient
>> > > > > > > > > for some existing users. We will have to add more to it
>> and the
>> > > > > > > > > compatibility story between versions could be a real
>> > > nightmare. One
>> > > > > > > > > thing that we could do for sure is disabling all unicode
>> > > groups in
>> > > > > the
>> > > > > > > > > library. That seems possible and would already reduce the
>> > > scope.
>> > > > > > > > >
>> > > > > > > > > Changes to the regex syntax seems not a major problem to
>> me. We
>> > > > > would
>> > > > > > > bump
>> > > > > > > > > the protocol version as we usually do. Actually that begs
>> the
>> > > same
>> > > > > > > question
>> > > > > > > > > about RE2. How stable is the syntax and does it differ
>> > > depending
>> > > > > on the
>> > > > > > > > > version of the library? In general, I do agree that we'd
>> be
>> > > better
>> > > > > off
>> > > > > > > not
>> > > > > > > > > having to manage the syntax ourselves if we can convince
>> > > ourselves
>> > > > > that
>> > > > > > > > > compatibility won't be a problem and the client won't
>> > > eventually
>> > > > > need
>> > > > > > > to
>> > > > > > > > > have access to it.
>> > > > > > > > >
>> > > > > > > > > Thanks,
>> > > > > > > > > Jason
>> > > > > > > > >
>> > > > > > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
>> > > > > > > <dj...@confluent.io.invalid>
>> > > > > > > > > wrote:
>> > > > > > > > >
>> > > > > > > > > > Hi Luke,
>> > > > > > > > > >
>> > > > > > > > > > Thanks for your comments.
>> > > > > > > > > >
>> > > > > > > > > > Your suggestion makes sense. Jun's made the same one.
>> We are
>> > > also
>> > > > > > > > > > discussing whether we should directly return the group
>> state
>> > > in
>> > > > > the
>> > > > > > > > > > response and drop the prepare assignment api. It seems
>> that
>> > > the
>> > > > > > > > > > current consensus is to keep it though. If that stays
>> like
>> > > this,
>> > > > > I
>> > > > > > > > > > will apply your suggestion.
>> > > > > > > > > >
>> > > > > > > > > > Thanks,
>> > > > > > > > > > David
>> > > > > > > > > >
>> > > > > > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <
>> > > showuon@gmail.com>
>> > > > > > > wrote:
>> > > > > > > > > > >
>> > > > > > > > > > > Hi David,
>> > > > > > > > > > >
>> > > > > > > > > > > Thanks for your patience for answering all these
>> questions!
>> > > > > > > > > > >
>> > > > > > > > > > > One more question from me:
>> > > > > > > > > > > 1. When the selected consumer is going to be in
>> client-side
>> > > > > > > assigning,
>> > > > > > > > > we
>> > > > > > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response.
>> I'm
>> > > > > > > wondering if
>> > > > > > > > > we
>> > > > > > > > > > > could use a flag to indicate the assignment is
>> required.
>> > > The
>> > > > > > > rationale
>> > > > > > > > > is
>> > > > > > > > > > > when returning error, we lose a chance to catch up
>> with the
>> > > > > latest
>> > > > > > > > > target
>> > > > > > > > > > > assignment or revoke partitions. Since we expected the
>> > > record
>> > > > > > > fetching
>> > > > > > > > > is
>> > > > > > > > > > > not blocked by the rebalance/assignment computing, we
>> might
>> > > > > want to
>> > > > > > > > > allow
>> > > > > > > > > > > the members to update the member assignment in each
>> > > heartbeat.
>> > > > > > > > > > >
>> > > > > > > > > > > Luke
>> > > > > > > > > > >
>> > > > > > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
>> > > > > > > > > > <ja...@confluent.io.invalid>
>> > > > > > > > > > > wrote:
>> > > > > > > > > > >
>> > > > > > > > > > > > Hey David,
>> > > > > > > > > > > >
>> > > > > > > > > > > > Thanks for the response.
>> > > > > > > > > > > >
>> > > > > > > > > > > > > Do you think that we could have a case where two
>> > > members
>> > > > > with
>> > > > > > > the
>> > > > > > > > > > same
>> > > > > > > > > > > > static id are active?
>> > > > > > > > > > > >
>> > > > > > > > > > > > I was just thinking about the handoff when we spin
>> up a
>> > > new
>> > > > > node
>> > > > > > > > > > before the
>> > > > > > > > > > > > old one completes shutdown. It's probably ok to
>> fence
>> > > the old
>> > > > > > > member
>> > > > > > > > > > > > abruptly. I do recall we had a few issues with this
>> > > handoff
>> > > > > in
>> > > > > > > the
>> > > > > > > > > > past. It
>> > > > > > > > > > > > was due to the fact that the replaced node might
>> have
>> > > been
>> > > > > the
>> > > > > > > > > leader.
>> > > > > > > > > > > > Since we've removed the leader notion, perhaps this
>> kind
>> > > of
>> > > > > issue
>> > > > > > > > > won't
>> > > > > > > > > > > > resurface.
>> > > > > > > > > > > >
>> > > > > > > > > > > > In regard to the RE2 library. I understand the
>> intent to
>> > > pass
>> > > > > > > this
>> > > > > > > > > > through
>> > > > > > > > > > > > as a string without any validation so that we don't
>> need
>> > > the
>> > > > > > > > > > dependence on
>> > > > > > > > > > > > the client. I do wonder if that will be restrictive
>> in
>> > > > > practice.
>> > > > > > > For
>> > > > > > > > > > > > example, if a regex changes on the client side,
>> would we
>> > > > > need to
>> > > > > > > > > > validate
>> > > > > > > > > > > > against the current assignment to see which
>> partitions to
>> > > > > > > continue
>> > > > > > > > > > > > fetching? Or would we instead invalidate the
>> complete
>> > > current
>> > > > > > > > > > assignment?
>> > > > > > > > > > > > Perhaps this is a rare case and we can do the
>> latter.
>> > > There
>> > > > > > > could be
>> > > > > > > > > > more
>> > > > > > > > > > > > situations though.
>> > > > > > > > > > > >
>> > > > > > > > > > > > As an alternative, I wonder if we could define a
>> > > minimalistic
>> > > > > > > regex
>> > > > > > > > > > syntax
>> > > > > > > > > > > > (a subset of RE2 say) that could map more easily to
>> regex
>> > > > > support
>> > > > > > > > > that
>> > > > > > > > > > > > already exists in clients? For example, in this
>> > > scenario, we
>> > > > > are
>> > > > > > > > > > matching
>> > > > > > > > > > > > topics, so we only need to be able to represent
>> topic
>> > > > > characters
>> > > > > > > and
>> > > > > > > > > > not
>> > > > > > > > > > > > the whole smorgasbord of whitespace and unicode
>> options.
>> > > > > Would it
>> > > > > > > > > make
>> > > > > > > > > > > > sense to have a reduced syntax which includes only
>> > > standard
>> > > > > topic
>> > > > > > > > > > > > characters, wildcards, character classes, and
>> > > repetitions?
>> > > > > > > > > > > >
>> > > > > > > > > > > > Best,
>> > > > > > > > > > > > Jason
>> > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
>> > > > > > > wangguoz@gmail.com>
>> > > > > > > > > > wrote:
>> > > > > > > > > > > >
>> > > > > > > > > > > > > Hi David, thanks for all the detailed
>> explanations. I
>> > > think
>> > > > > > > they
>> > > > > > > > > all
>> > > > > > > > > > make
>> > > > > > > > > > > > > sense. Just want to have a couple follow-ups here:
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > > I don't really see the benefits here because
>> server
>> > > side
>> > > > > > > > > assignors
>> > > > > > > > > > > > > don't have metadata at all. They only assign
>> > > > > topic-partitions.
>> > > > > > > They
>> > > > > > > > > > > > > are not supposed to generate metadata nor to
>> receive
>> > > > > metadata
>> > > > > > > from
>> > > > > > > > > > the
>> > > > > > > > > > > > > members.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > Just to clarify I was asking about the `version`
>> of the
>> > > > > > > assignor
>> > > > > > > > > > (i.e. up
>> > > > > > > > > > > > > to what version that the client would support),
>> and I
>> > > do
>> > > > > agree
>> > > > > > > we
>> > > > > > > > > > would
>> > > > > > > > > > > > not
>> > > > > > > > > > > > > need metadata. What I have in mind is that, for
>> some
>> > > > > specific
>> > > > > > > > > > built-in
>> > > > > > > > > > > > > broker-assignors, e.g. rack-aware assignors, if
>> it's
>> > > > > possible
>> > > > > > > that
>> > > > > > > > > > in a
>> > > > > > > > > > > > > newer version we would have a hierarchical rack ID
>> > > string
>> > > > > > > format,
>> > > > > > > > > > like
>> > > > > > > > > > > > > "tier1-tier2" etc, but if some client has not
>> upgraded
>> > > > > their
>> > > > > > > rack
>> > > > > > > > > ID
>> > > > > > > > > > > > > would still be in old format. In this case, the
>> broker
>> > > then
>> > > > > > > needs
>> > > > > > > > > to
>> > > > > > > > > > > > choose
>> > > > > > > > > > > > > the old versioned assignor. I'm probably making
>> > > something
>> > > > > up
>> > > > > > > here
>> > > > > > > > > for
>> > > > > > > > > > > > rack
>> > > > > > > > > > > > > aware assignors, but I'm wondering if in general
>> such
>> > > an
>> > > > > > > > > > "auto-downgrade"
>> > > > > > > > > > > > > behavior would be needed still for broker-side
>> > > assignor,
>> > > > > and
>> > > > > > > if yes
>> > > > > > > > > > would
>> > > > > > > > > > > > > "version" still be useful.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > > Yeah, that's right. Within a rebalance,
>> > > `onAssignment` is
>> > > > > > > called
>> > > > > > > > > > once
>> > > > > > > > > > > > > when the member transitions to a new epoch. This
>> one
>> > > > > contains
>> > > > > > > the
>> > > > > > > > > > full
>> > > > > > > > > > > > > metadata provided by the client side assignor.
>> Then,
>> > > > > > > `onAssignment`
>> > > > > > > > > > > > > can be called max N times where N is the number of
>> > > > > partitions
>> > > > > > > > > pending
>> > > > > > > > > > > > > revocation by other members. Let me try to clarify
>> > > this in
>> > > > > the
>> > > > > > > KIP.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > Okay, my understanding is that the calling
>> ordering of
>> > > > > these
>> > > > > > > > > > callbacks
>> > > > > > > > > > > > > would be like the following:
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > ----------------------------------------
>> > > > > > > > > > > > > onPartitionsRevoked();   // just once, since we
>> do not
>> > > > > really
>> > > > > > > need
>> > > > > > > > > > > > > to revoke incrementally.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > onAssignment();    // the first call, with epoch
>> > > > > incremented
>> > > > > > > > > > > > > onPartitionsAssigned();   // paired with the
>> > > onAssignment
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > onAssignment();              // the first
>> onAssignment
>> > > > > would
>> > > > > > > bump
>> > > > > > > > > up
>> > > > > > > > > > the
>> > > > > > > > > > > > > epoch, and the metadata reflected.
>> > > > > > > > > > > > > onPartitionsAssigned();   // each time we get an
>> > > additional
>> > > > > > > > > > assignment,
>> > > > > > > > > > > > we
>> > > > > > > > > > > > > call onAssignment and then paired with an
>> > > > > onPartitionsAssigned
>> > > > > > > > > > > > > ...
>> > > > > > > > > > > > > onAssignment();
>> > > > > > > > > > > > > onPartitionsAssigned();   // on each of the
>> > > onAssignment
>> > > > > > > calls, the
>> > > > > > > > > > > > encoded
>> > > > > > > > > > > > > metadata would not change, only the incrementally
>> added
>> > > > > > > partitions
>> > > > > > > > > be
>> > > > > > > > > > > > > reflected
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > Is that the case?
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > I'm wondering if we would still call onAssignment
>> just
>> > > > > once,
>> > > > > > > that
>> > > > > > > > > > encodes
>> > > > > > > > > > > > > all the assignment for this rebalance, including
>> all
>> > > the
>> > > > > > > partitions
>> > > > > > > > > > that
>> > > > > > > > > > > > > should be assigned to the member but not yet
>> assigned
>> > > since
>> > > > > > > they
>> > > > > > > > > > have not
>> > > > > > > > > > > > > been revoked by others. In that case the call
>> ordering
>> > > > > would
>> > > > > > > be:
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > ----------------------------------------
>> > > > > > > > > > > > > onPartitionsRevoked();   // just once
>> > > > > > > > > > > > > onAssignment();    // just once, with epoch
>> > > incremented,
>> > > > > and
>> > > > > > > > > metadata
>> > > > > > > > > > > > > encoded changed, the "assignment" field also
>> reflect
>> > > the
>> > > > > final
>> > > > > > > > > target
>> > > > > > > > > > > > > assignment
>> > > > > > > > > > > > > onPartitionsAssigned();   // multiple times, which
>> > > > > represent
>> > > > > > > > > > > > incrementally
>> > > > > > > > > > > > > added partitions
>> > > > > > > > > > > > > ...
>> > > > > > > > > > > > > onPartitionsAssigned();
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > The motivation from this is that, most users
>> would only
>> > > > > > > implement
>> > > > > > > > > the
>> > > > > > > > > > > > > rebalance callback listeners and hence we'd
>> definitely
>> > > > > need to
>> > > > > > > make
>> > > > > > > > > > sure
>> > > > > > > > > > > > > the semantics of that does not change much, and
>> the
>> > > time
>> > > > > > > > > > > > > `onPartitionsAssigned` indicate the time when the
>> > > > > partitions
>> > > > > > > are
>> > > > > > > > > > actually
>> > > > > > > > > > > > > assigned to it; while for assignors, the
>> > > `onAssignment` is
>> > > > > > > used to
>> > > > > > > > > > > > indicate
>> > > > > > > > > > > > > what decision is made regarding for this member,
>> i.e.
>> > > when
>> > > > > the
>> > > > > > > > > > partitions
>> > > > > > > > > > > > > are decided to be given to it, but not necessarily
>> > > meant
>> > > > > that
>> > > > > > > it
>> > > > > > > > > has
>> > > > > > > > > > been
>> > > > > > > > > > > > > given, since that time should be determined by the
>> > > time of
>> > > > > > > > > > > > > `onPartitionsAssigned`. The benefits though,
>> would be
>> > > that
>> > > > > > > assignor
>> > > > > > > > > > > > > implementers would not need to reason which
>> > > `onAssignment`
>> > > > > > > would be
>> > > > > > > > > > the
>> > > > > > > > > > > > > last one for this epoch.
>> > > > > > > > > > > > >
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > Guozhang
>> > > > > > > > > > > > >
>> > > > > > > > > > > > >
>> > > > > > > > > > > > >
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > Guozhang
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
>> > > > > > > > > > <djacot@confluent.io.invalid
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > wrote:
>> > > > > > > > > > > > >
>> > > > > > > > > > > > > > Hi Guozhang,
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > 1) The client-side
>> "PartitionAssignor#Assignment"
>> > > has
>> > > > > an
>> > > > > > > Error
>> > > > > > > > > > byte
>> > > > > > > > > > > > > > field,
>> > > > > > > > > > > > > > > while the broker-side
>> > > "PartitionAssignor#Assignment"
>> > > > > does
>> > > > > > > not.
>> > > > > > > > > > And
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > rationale seems to be assuming that we should
>> > > always be
>> > > > > > > able to
>> > > > > > > > > > do
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > assignment at the broker-side assignor without
>> > > errors.
>> > > > > > > > > > Personally I
>> > > > > > > > > > > > > think
>> > > > > > > > > > > > > > > it's still potentially beneficial to add the
>> Error
>> > > > > field
>> > > > > > > even
>> > > > > > > > > for
>> > > > > > > > > > > > > > > broker-side assignors, e.g. for some edge
>> cases
>> > > where
>> > > > > some
>> > > > > > > > > > subscribed
>> > > > > > > > > > > > > > > topics are not recognized with the current
>> broker's
>> > > > > > > metadata.
>> > > > > > > > > > What do
>> > > > > > > > > > > > > you
>> > > > > > > > > > > > > > > think?
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > Yeah, that seems reasonable. However, I wonder
>> if it
>> > > > > would be
>> > > > > > > > > > better
>> > > > > > > > > > > > > > to use an exception on the server side. This is
>> what
>> > > we
>> > > > > > > usually
>> > > > > > > > > do
>> > > > > > > > > > for
>> > > > > > > > > > > > > > server side plugins. On the client side, we use
>> a
>> > > field
>> > > > > > > because
>> > > > > > > > > the
>> > > > > > > > > > > > > > errors are not defined in advance.
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > Your comment also makes me think about what we
>> > > should do
>> > > > > > > when the
>> > > > > > > > > > > > > > server side assignor fails. I suppose that we
>> have to
>> > > > > keep
>> > > > > > > the
>> > > > > > > > > > current
>> > > > > > > > > > > > > > assignment until a new event occurs. For
>> instance, in
>> > > > > your
>> > > > > > > > > example,
>> > > > > > > > > > > > > > the coordinator would have to trigger a
>> rebalance
>> > > when
>> > > > > > > > > unrecognized
>> > > > > > > > > > > > > > topics are available. This would be part of the
>> > > metadata
>> > > > > > > > > > monitoring.
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
>> > > additional
>> > > > > > > fields
>> > > > > > > > > > > > > > > reason/version/metadata compared with the
>> > > broker-side
>> > > > > > > > > > GroupMember. I
>> > > > > > > > > > > > > > agree
>> > > > > > > > > > > > > > > that broker-side assignor would not need
>> > > > > reason/metadata
>> > > > > > > since
>> > > > > > > > > > they
>> > > > > > > > > > > > are
>> > > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but
>> what
>> > > about
>> > > > > > > version?
>> > > > > > > > > > E.g.
>> > > > > > > > > > > > is
>> > > > > > > > > > > > > > it
>> > > > > > > > > > > > > > > possible that we evolve our broker-side
>> built-in
>> > > > > assignor
>> > > > > > > but
>> > > > > > > > > > the old
>> > > > > > > > > > > > > > > versioned clients would not be able to work
>> with
>> > > the
>> > > > > new
>> > > > > > > > > > version, in
>> > > > > > > > > > > > > > which
>> > > > > > > > > > > > > > > case we need to let the broker being aware of
>> this
>> > > and
>> > > > > > > upgrade
>> > > > > > > > > > its
>> > > > > > > > > > > > > > behavior
>> > > > > > > > > > > > > > > to cooperate with the clients?
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > I don't really see the benefits here because
>> server
>> > > side
>> > > > > > > > > assignors
>> > > > > > > > > > > > > > don't have metadata at all. They only assign
>> > > > > > > topic-partitions.
>> > > > > > > > > They
>> > > > > > > > > > > > > > are not supposed to generate metadata nor to
>> receive
>> > > > > metadata
>> > > > > > > > > from
>> > > > > > > > > > the
>> > > > > > > > > > > > > > members.
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > 3) Also related to 2) above, for the
>> client-side
>> > > > > > > "GroupMember",
>> > > > > > > > > > > > instead
>> > > > > > > > > > > > > > of
>> > > > > > > > > > > > > > > including these three fields, what about just
>> > > adding
>> > > > > the
>> > > > > > > > > > "Metadata"
>> > > > > > > > > > > > > field
>> > > > > > > > > > > > > > > class which has these three fields? Also,
>> there
>> > > are two
>> > > > > > > > > > "Metadata"
>> > > > > > > > > > > > > > > currently in the APIs, the first is a class
>> that
>> > > > > encodes
>> > > > > > > > > > > > > > > reason/version/metadata, and the second is
>> just the
>> > > > > encoded
>> > > > > > > > > > metadata
>> > > > > > > > > > > > > > bytes.
>> > > > > > > > > > > > > > > I'm wondering what about just naming the
>> first as
>> > > > > > > > > memberMetadata,
>> > > > > > > > > > > > which
>> > > > > > > > > > > > > > > then has a bytebuffer field Metadata, or
>> instead
>> > > > > naming the
>> > > > > > > > > > second
>> > > > > > > > > > > > > > > bytebuffer field as metadataBytes?
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > That's a good point. Let me try to rationalize
>> this
>> > > > > interface
>> > > > > > > > > > based on
>> > > > > > > > > > > > > > your suggestions.
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > Best,
>> > > > > > > > > > > > > > David
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
>> > > > > > > > > wangguoz@gmail.com>
>> > > > > > > > > > > > > wrote:
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > Hello David,
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > Just had a few more nit questions about the
>> public
>> > > > > APIs:
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > 1) The client-side
>> "PartitionAssignor#Assignment"
>> > > has
>> > > > > an
>> > > > > > > Error
>> > > > > > > > > > byte
>> > > > > > > > > > > > > > field,
>> > > > > > > > > > > > > > > while the broker-side
>> > > "PartitionAssignor#Assignment"
>> > > > > does
>> > > > > > > not.
>> > > > > > > > > > And
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > rationale seems to be assuming that we should
>> > > always be
>> > > > > > > able to
>> > > > > > > > > > do
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > assignment at the broker-side assignor without
>> > > errors.
>> > > > > > > > > > Personally I
>> > > > > > > > > > > > > think
>> > > > > > > > > > > > > > > it's still potentially beneficial to add the
>> Error
>> > > > > field
>> > > > > > > even
>> > > > > > > > > for
>> > > > > > > > > > > > > > > broker-side assignors, e.g. for some edge
>> cases
>> > > where
>> > > > > some
>> > > > > > > > > > subscribed
>> > > > > > > > > > > > > > > topics are not recognized with the current
>> broker's
>> > > > > > > metadata.
>> > > > > > > > > > What do
>> > > > > > > > > > > > > you
>> > > > > > > > > > > > > > > think?
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
>> > > additional
>> > > > > > > fields
>> > > > > > > > > > > > > > > reason/version/metadata compared with the
>> > > broker-side
>> > > > > > > > > > GroupMember. I
>> > > > > > > > > > > > > > agree
>> > > > > > > > > > > > > > > that broker-side assignor would not need
>> > > > > reason/metadata
>> > > > > > > since
>> > > > > > > > > > they
>> > > > > > > > > > > > are
>> > > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but
>> what
>> > > about
>> > > > > > > version?
>> > > > > > > > > > E.g.
>> > > > > > > > > > > > is
>> > > > > > > > > > > > > > it
>> > > > > > > > > > > > > > > possible that we evolve our broker-side
>> built-in
>> > > > > assignor
>> > > > > > > but
>> > > > > > > > > > the old
>> > > > > > > > > > > > > > > versioned clients would not be able to work
>> with
>> > > the
>> > > > > new
>> > > > > > > > > > version, in
>> > > > > > > > > > > > > > which
>> > > > > > > > > > > > > > > case we need to let the broker being aware of
>> this
>> > > and
>> > > > > > > upgrade
>> > > > > > > > > > its
>> > > > > > > > > > > > > > behavior
>> > > > > > > > > > > > > > > to cooperate with the clients?
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > 3) Also related to 2) above, for the
>> client-side
>> > > > > > > "GroupMember",
>> > > > > > > > > > > > instead
>> > > > > > > > > > > > > > of
>> > > > > > > > > > > > > > > including these three fields, what about just
>> > > adding
>> > > > > the
>> > > > > > > > > > "Metadata"
>> > > > > > > > > > > > > field
>> > > > > > > > > > > > > > > class which has these three fields? Also,
>> there
>> > > are two
>> > > > > > > > > > "Metadata"
>> > > > > > > > > > > > > > > currently in the APIs, the first is a class
>> that
>> > > > > encodes
>> > > > > > > > > > > > > > > reason/version/metadata, and the second is
>> just the
>> > > > > encoded
>> > > > > > > > > > metadata
>> > > > > > > > > > > > > > bytes.
>> > > > > > > > > > > > > > > I'm wondering what about just naming the
>> first as
>> > > > > > > > > memberMetadata,
>> > > > > > > > > > > > which
>> > > > > > > > > > > > > > > then has a bytebuffer field Metadata, or
>> instead
>> > > > > naming the
>> > > > > > > > > > second
>> > > > > > > > > > > > > > > bytebuffer field as metadataBytes?
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > Guozhang
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang
>> Wang <
>> > > > > > > > > > wangguoz@gmail.com>
>> > > > > > > > > > > > > > wrote:
>> > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > Hello David,
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > Thanks for bringing this question up. I
>> think the
>> > > > > main
>> > > > > > > > > > benefits as
>> > > > > > > > > > > > > you
>> > > > > > > > > > > > > > > > listed is 2) above if it stays; just to
>> clarify,
>> > > we
>> > > > > would
>> > > > > > > > > only
>> > > > > > > > > > be
>> > > > > > > > > > > > > able
>> > > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > > save one round trip if the rebalance is
>> still
>> > > > > triggered
>> > > > > > > by
>> > > > > > > > > the
>> > > > > > > > > > > > > broker;
>> > > > > > > > > > > > > > if
>> > > > > > > > > > > > > > > > the rebalance is triggered by the client
>> then the
>> > > > > > > > > > num.round.trips
>> > > > > > > > > > > > are
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > > same:
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > 1) With GroupPrepareAssignment:
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > T0: client decides to do a new assignment,
>> > > suppose
>> > > > > it has
>> > > > > > > > > > already
>> > > > > > > > > > > > > sent
>> > > > > > > > > > > > > > a
>> > > > > > > > > > > > > > > > HB and hence has to wait for it to return
>> first
>> > > since
>> > > > > > > only
>> > > > > > > > > one
>> > > > > > > > > > > > > request
>> > > > > > > > > > > > > > /
>> > > > > > > > > > > > > > > > response can be inflight with the
>> coordinator's
>> > > > > socket.
>> > > > > > > > > > > > > > > > T1: client receives the HB response, and
>> then
>> > > sends
>> > > > > the
>> > > > > > > > > > > > > > > > GroupPrepareAssignment request.
>> > > > > > > > > > > > > > > > T2: the GroupPrepareAssignment response is
>> > > returned.
>> > > > > > > > > > > > > > > > T3: it calculates the new assignment, and
>> sends a
>> > > > > > > > > > > > > > GroupInstallAssignment
>> > > > > > > > > > > > > > > > request.
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > In total, two round trips.
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > 2) Without GroupPrepareAssignment:
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > T0: client decides to do a new assignment,
>> > > suppose
>> > > > > it has
>> > > > > > > > > > already
>> > > > > > > > > > > > > sent
>> > > > > > > > > > > > > > a
>> > > > > > > > > > > > > > > > HB and hence has to wait for it to return
>> first
>> > > since
>> > > > > > > only
>> > > > > > > > > one
>> > > > > > > > > > > > > request
>> > > > > > > > > > > > > > /
>> > > > > > > > > > > > > > > > response can be inflight with the
>> coordinator's
>> > > > > socket.
>> > > > > > > > > > > > > > > > T1: client receives the HB response, and
>> then
>> > > sends
>> > > > > the
>> > > > > > > new
>> > > > > > > > > HB
>> > > > > > > > > > > > > request
>> > > > > > > > > > > > > > > > with the flag indicating a new rebalance
>> needed..
>> > > > > > > > > > > > > > > > T2: the HB response with the optional member
>> > > metadata
>> > > > > > > map is
>> > > > > > > > > > > > > returned.
>> > > > > > > > > > > > > > > > T3: it calculates the new assignment, and
>> sends a
>> > > > > > > > > > > > > > GroupInstallAssignment
>> > > > > > > > > > > > > > > > request.
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > In total, two round trips as well.
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > -----------------------------
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > So to complete the full picture here, we'd
>> need
>> > > to
>> > > > > modify
>> > > > > > > > > both
>> > > > > > > > > > HB
>> > > > > > > > > > > > > > request
>> > > > > > > > > > > > > > > > and response so that the client can also
>> > > indicate a
>> > > > > new
>> > > > > > > > > > rebalance
>> > > > > > > > > > > > via
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > > HB request as well, right?
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > Assuming all above is true, I think it's
>> okay to
>> > > > > merge
>> > > > > > > the
>> > > > > > > > > > > > > > > > GroupPrepareAssignment into HB given that
>> we can
>> > > > > make the
>> > > > > > > > > > > > additional
>> > > > > > > > > > > > > > fields
>> > > > > > > > > > > > > > > > encoding the full member (subscription)
>> metadata
>> > > and
>> > > > > > > topic
>> > > > > > > > > > metadata
>> > > > > > > > > > > > > as
>> > > > > > > > > > > > > > > > optional fields.
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > Guozhang
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
>> > > > > > > > > > > > > > <dj...@confluent.io.invalid>
>> > > > > > > > > > > > > > > > wrote:
>> > > > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> Hi all,
>> > > > > > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> During an offline conversation, someone
>> asked
>> > > why we
>> > > > > > > need
>> > > > > > > > > the
>> > > > > > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and
>> suggested
>> > > > > that we
>> > > > > > > > > could
>> > > > > > > > > > > > > instead
>> > > > > > > > > > > > > > > >> provide the group state in the heartbeat
>> > > response.
>> > > > > This
>> > > > > > > has
>> > > > > > > > > a
>> > > > > > > > > > few
>> > > > > > > > > > > > > > > >> advantages: 1) it does not require using a
>> > > special
>> > > > > error
>> > > > > > > > > code
>> > > > > > > > > > to
>> > > > > > > > > > > > > > > >> signal that a new assignment is required
>> as the
>> > > > > signal
>> > > > > > > would
>> > > > > > > > > > be
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> provided group state; 2) it removes one
>> round
>> > > trip
>> > > > > when
>> > > > > > > a
>> > > > > > > > > > client
>> > > > > > > > > > > > > side
>> > > > > > > > > > > > > > > >> assignor is used. The downside is that it
>> makes
>> > > the
>> > > > > > > > > heartbeat
>> > > > > > > > > > > > > > > >> response's definition quite large. I recall
>> > > that I
>> > > > > went
>> > > > > > > with
>> > > > > > > > > > the
>> > > > > > > > > > > > > > > >> current approach due to this.
>> > > > > > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> Providing the group state in the heartbeat
>> > > response
>> > > > > is
>> > > > > > > > > > appealing.
>> > > > > > > > > > > > > What
>> > > > > > > > > > > > > > > >> do you guys think?
>> > > > > > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> Best,
>> > > > > > > > > > > > > > > >> David
>> > > > > > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David
>> Jacot <
>> > > > > > > > > > djacot@confluent.io>
>> > > > > > > > > > > > > > wrote:
>> > > > > > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > Hi Guozhang,
>> > > > > > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > 1. I have added a reference to the
>> relevant
>> > > > > chapter
>> > > > > > > > > instead
>> > > > > > > > > > of
>> > > > > > > > > > > > > > > >> > repeating the whole thing. Does that
>> work for
>> > > you?
>> > > > > > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you
>> are
>> > > > > referring
>> > > > > > > to
>> > > > > > > > > is
>> > > > > > > > > > > > about
>> > > > > > > > > > > > > > when
>> > > > > > > > > > > > > > > >> > a rebalance should be triggered for the
>> > > > > non-upgraded
>> > > > > > > > > members
>> > > > > > > > > > > > using
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> > old protocol. The section mentions that a
>> > > > > rebalance
>> > > > > > > must
>> > > > > > > > > be
>> > > > > > > > > > > > > > triggered
>> > > > > > > > > > > > > > > >> > when a new assignment is installed. This
>> > > implies
>> > > > > that
>> > > > > > > the
>> > > > > > > > > > group
>> > > > > > > > > > > > > > epoch
>> > > > > > > > > > > > > > > >> > was updated either by a native member or
>> a
>> > > > > > > non-upgraded
>> > > > > > > > > > member.
>> > > > > > > > > > > > > For
>> > > > > > > > > > > > > > > >> > the latter, the JoinGroup request would
>> be the
>> > > > > > > trigger. I
>> > > > > > > > > > have
>> > > > > > > > > > > > > > added a
>> > > > > > > > > > > > > > > >> > reference to the relevant chapter in the
>> > > > > "JoinGroup
>> > > > > > > > > > Handling"
>> > > > > > > > > > > > > > section
>> > > > > > > > > > > > > > > >> > as well. Does that make sense?
>> > > > > > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > Thanks,
>> > > > > > > > > > > > > > > >> > David
>> > > > > > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang
>> Wang
>> > > <
>> > > > > > > > > > > > wangguoz@gmail.com
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> wrote:
>> > > > > > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > Hello David,
>> > > > > > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > Alright I think that's sufficient.
>> Just to
>> > > make
>> > > > > that
>> > > > > > > > > > clear in
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > doc,
>> > > > > > > > > > > > > > > >> > > could we update:
>> > > > > > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > 1) the heartbeat request handling
>> section,
>> > > > > stating
>> > > > > > > when
>> > > > > > > > > > > > > > coordinator
>> > > > > > > > > > > > > > > >> will
>> > > > > > > > > > > > > > > >> > > trigger rebalance based on the HB's
>> member
>> > > > > metadata
>> > > > > > > /
>> > > > > > > > > > reason?
>> > > > > > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to
>> > > include
>> > > > > what
>> > > > > > > we
>> > > > > > > > > > > > described
>> > > > > > > > > > > > > > in
>> > > > > > > > > > > > > > > >> "Group
>> > > > > > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as
>> > > well?
>> > > > > > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > Guozhang
>> > > > > > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David
>> Jacot
>> > > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
>> > > > > > > > > > > > > > > >> > > wrote:
>> > > > > > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > Hi Guozhang,
>> > > > > > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > I thought that the assignor will
>> always be
>> > > > > > > consulted
>> > > > > > > > > > when
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > next
>> > > > > > > > > > > > > > > >> > > > heartbeat request is constructed. In
>> other
>> > > > > words,
>> > > > > > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be
>> > > called
>> > > > > for
>> > > > > > > every
>> > > > > > > > > > > > > heartbeat.
>> > > > > > > > > > > > > > > >> This
>> > > > > > > > > > > > > > > >> > > > gives the opportunity for the
>> assignor to
>> > > > > enforce
>> > > > > > > a
>> > > > > > > > > > > > rebalance
>> > > > > > > > > > > > > by
>> > > > > > > > > > > > > > > >> > > > setting the reason to a non-zero
>> value or
>> > > by
>> > > > > > > changing
>> > > > > > > > > > the
>> > > > > > > > > > > > > > bytes. Do
>> > > > > > > > > > > > > > > >> > > > you think that this is not
>> sufficient?
>> > > Are you
>> > > > > > > > > > concerned by
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> delay?
>> > > > > > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > Best,
>> > > > > > > > > > > > > > > >> > > > David
>> > > > > > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM
>> Guozhang
>> > > Wang <
>> > > > > > > > > > > > > > wangguoz@gmail.com>
>> > > > > > > > > > > > > > > >> wrote:
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > Hello David,
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > One of Jun's comments make me
>> thinking:
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > ```
>> > > > > > > > > > > > > > > >> > > > > In this case, a new assignment is
>> > > triggered
>> > > > > by
>> > > > > > > the
>> > > > > > > > > > client
>> > > > > > > > > > > > > side
>> > > > > > > > > > > > > > > >> > > > > assignor. When constructing the
>> HB, the
>> > > > > consumer
>> > > > > > > > > will
>> > > > > > > > > > > > always
>> > > > > > > > > > > > > > > >> consult
>> > > > > > > > > > > > > > > >> > > > > the client side assignor and
>> propagate
>> > > the
>> > > > > > > > > > information to
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > coordinator. In other words, we
>> don't
>> > > expect
>> > > > > > > users
>> > > > > > > > > to
>> > > > > > > > > > call
>> > > > > > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
>> > > > > > > > > > > > > > > >> > > > > ```
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > As I looked at the current
>> > > > > PartitionAssignor's
>> > > > > > > > > > interface,
>> > > > > > > > > > > > we
>> > > > > > > > > > > > > > > >> actually do
>> > > > > > > > > > > > > > > >> > > > > not have a way yet to instruct how
>> to
>> > > > > construct
>> > > > > > > the
>> > > > > > > > > > next
>> > > > > > > > > > > > HB
>> > > > > > > > > > > > > > > >> request, e.g.
>> > > > > > > > > > > > > > > >> > > > > when the assignor wants to enforce
>> a new
>> > > > > > > rebalance
>> > > > > > > > > > with a
>> > > > > > > > > > > > > new
>> > > > > > > > > > > > > > > >> assignment,
>> > > > > > > > > > > > > > > >> > > > > we'd need some customizable APIs
>> inside
>> > > the
>> > > > > > > > > > > > > PartitionAssignor
>> > > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> indicate
>> > > > > > > > > > > > > > > >> > > > > the next HB telling broker about
>> so.
>> > > WDYT
>> > > > > about
>> > > > > > > > > adding
>> > > > > > > > > > > > such
>> > > > > > > > > > > > > an
>> > > > > > > > > > > > > > > >> API on the
>> > > > > > > > > > > > > > > >> > > > > PartitionAssignor?
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > Guozhang
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM
>> David
>> > > Jacot
>> > > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
>> > > > > > > > > > > > > > > >> > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > Hi Jun,
>> > > > > > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > I have updated the KIP to
>> include your
>> > > > > > > feedback. I
>> > > > > > > > > > have
>> > > > > > > > > > > > > also
>> > > > > > > > > > > > > > > >> tried to
>> > > > > > > > > > > > > > > >> > > > > > clarify the parts which were not
>> > > cleared.
>> > > > > > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > Best,
>> > > > > > > > > > > > > > > >> > > > > > David
>> > > > > > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM
>> David
>> > > > > Jacot <
>> > > > > > > > > > > > > > djacot@confluent.io
>> > > > > > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Hi Jun,
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let
>> me
>> > > start
>> > > > > by
>> > > > > > > > > > answering
>> > > > > > > > > > > > your
>> > > > > > > > > > > > > > > >> questions
>> > > > > > > > > > > > > > > >> > > > > > > inline and I will update the
>> KIP
>> > > next
>> > > > > week.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall,
>> the
>> > > main
>> > > > > > > benefits
>> > > > > > > > > > of
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > KIP
>> > > > > > > > > > > > > > > >> seem to
>> > > > > > > > > > > > > > > >> > > > be
>> > > > > > > > > > > > > > > >> > > > > > fewer
>> > > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and
>> more
>> > > > > efficient
>> > > > > > > > > > support of
>> > > > > > > > > > > > > > > >> wildcard. A few
>> > > > > > > > > > > > > > > >> > > > > > > > comments below.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > I would also add that the KIP
>> > > removes
>> > > > > the
>> > > > > > > global
>> > > > > > > > > > sync
>> > > > > > > > > > > > > > barrier
>> > > > > > > > > > > > > > > >> in the
>> > > > > > > > > > > > > > > >> > > > > > > protocol which is essential to
>> > > improve
>> > > > > group
>> > > > > > > > > > stability
>> > > > > > > > > > > > > and
>> > > > > > > > > > > > > > > >> > > > > > > scalability, and the KIP also
>> > > > > simplifies the
>> > > > > > > > > > client by
>> > > > > > > > > > > > > > moving
>> > > > > > > > > > > > > > > >> most of
>> > > > > > > > > > > > > > > >> > > > > > > the logic to the server side.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 30.
>> ConsumerGroupHeartbeatRequest
>> > > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
>> > > singleton.
>> > > > > Do we
>> > > > > > > plan
>> > > > > > > > > > to
>> > > > > > > > > > > > > > support
>> > > > > > > > > > > > > > > >> rolling
>> > > > > > > > > > > > > > > >> > > > > > changing
>> > > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in
>> the
>> > > > > > > consumers?
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Definitely. The group
>> coordinator
>> > > will
>> > > > > use
>> > > > > > > the
>> > > > > > > > > > > > assignor
>> > > > > > > > > > > > > > used
>> > > > > > > > > > > > > > > >> by a
>> > > > > > > > > > > > > > > >> > > > > > > majority of the members. This
>> > > allows the
>> > > > > > > group
>> > > > > > > > > to
>> > > > > > > > > > move
>> > > > > > > > > > > > > > from
>> > > > > > > > > > > > > > > >> one
>> > > > > > > > > > > > > > > >> > > > > > > assignor to another by a roll.
>> This
>> > > is
>> > > > > > > explained
>> > > > > > > > > > in
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> Assignor
>> > > > > > > > > > > > > > > >> > > > > > > Selection chapter.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could
>> you
>> > > explain
>> > > > > > > whether
>> > > > > > > > > > it's
>> > > > > > > > > > > > > > > >> required in
>> > > > > > > > > > > > > > > >> > > > every
>> > > > > > > > > > > > > > > >> > > > > > > > request or the scenarios
>> when it
>> > > > > needs to
>> > > > > > > be
>> > > > > > > > > > filled?
>> > > > > > > > > > > > > For
>> > > > > > > > > > > > > > > >> example,
>> > > > > > > > > > > > > > > >> > > > it's
>> > > > > > > > > > > > > > > >> > > > > > not
>> > > > > > > > > > > > > > > >> > > > > > > > clear to me when
>> TopicPartitions
>> > > > > needs to
>> > > > > > > be
>> > > > > > > > > > filled.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > The client is expected to set
>> those
>> > > > > fields
>> > > > > > > in
>> > > > > > > > > > case of
>> > > > > > > > > > > > a
>> > > > > > > > > > > > > > > >> connection
>> > > > > > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when
>> the
>> > > fields
>> > > > > have
>> > > > > > > > > > changed
>> > > > > > > > > > > > > since
>> > > > > > > > > > > > > > > >> the last
>> > > > > > > > > > > > > > > >> > > > > > > HB. The server populates those
>> > > fields as
>> > > > > > > long as
>> > > > > > > > > > the
>> > > > > > > > > > > > > > member
>> > > > > > > > > > > > > > > >> is not
>> > > > > > > > > > > > > > > >> > > > > > > fully reconciled - the member
>> should
>> > > > > > > acknowledge
>> > > > > > > > > > that
>> > > > > > > > > > > > it
>> > > > > > > > > > > > > > has
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > expected epoch and assignment.
>> I
>> > > will
>> > > > > > > clarify
>> > > > > > > > > > this in
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > KIP.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
>> > > protocol,
>> > > > > the
>> > > > > > > rack
>> > > > > > > > > > > > > affinity
>> > > > > > > > > > > > > > > >> between the
>> > > > > > > > > > > > > > > >> > > > > > client
>> > > > > > > > > > > > > > > >> > > > > > > > and the broker is only
>> considered
>> > > > > during
>> > > > > > > > > > fetching,
>> > > > > > > > > > > > but
>> > > > > > > > > > > > > > not
>> > > > > > > > > > > > > > > >> during
>> > > > > > > > > > > > > > > >> > > > > > assigning
>> > > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
>> > > Sometimes,
>> > > > > once
>> > > > > > > the
>> > > > > > > > > > > > > assignment
>> > > > > > > > > > > > > > is
>> > > > > > > > > > > > > > > >> made,
>> > > > > > > > > > > > > > > >> > > > there
>> > > > > > > > > > > > > > > >> > > > > > is
>> > > > > > > > > > > > > > > >> > > > > > > > no opportunity for read
>> affinity
>> > > > > because
>> > > > > > > no
>> > > > > > > > > > replicas
>> > > > > > > > > > > > > of
>> > > > > > > > > > > > > > > >> assigned
>> > > > > > > > > > > > > > > >> > > > > > partitions
>> > > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
>> > > > > wondering
>> > > > > > > if we
>> > > > > > > > > > should
>> > > > > > > > > > > > > use
>> > > > > > > > > > > > > > > >> this
>> > > > > > > > > > > > > > > >> > > > > > opportunity
>> > > > > > > > > > > > > > > >> > > > > > > > to address this by including
>> rack
>> > > in
>> > > > > > > > > > GroupMember.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > That's an interesting idea. I
>> don't
>> > > see
>> > > > > any
>> > > > > > > > > issue
>> > > > > > > > > > with
>> > > > > > > > > > > > > > adding
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > rack
>> > > > > > > > > > > > > > > >> > > > > > > to the members. I will do so.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side,
>> often,
>> > > it's
>> > > > > > > useful to
>> > > > > > > > > > know
>> > > > > > > > > > > > how
>> > > > > > > > > > > > > > busy
>> > > > > > > > > > > > > > > >> a group
>> > > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
>> > > event
>> > > > > loop
>> > > > > > > > > model,
>> > > > > > > > > > it
>> > > > > > > > > > > > > seems
>> > > > > > > > > > > > > > > >> that we
>> > > > > > > > > > > > > > > >> > > > could
>> > > > > > > > > > > > > > > >> > > > > > add
>> > > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the
>> fraction
>> > > of
>> > > > > the
>> > > > > > > time
>> > > > > > > > > > the
>> > > > > > > > > > > > > event
>> > > > > > > > > > > > > > > >> loop is
>> > > > > > > > > > > > > > > >> > > > doing
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > actual work.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > That's a great idea. I will
>> add it.
>> > > > > Thanks.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
>> > > > > coordinator
>> > > > > > > > > > failover
>> > > > > > > > > > > > > > > >> handling? For
>> > > > > > > > > > > > > > > >> > > > > > example,
>> > > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the
>> check
>> > > if
>> > > > > any
>> > > > > > > group
>> > > > > > > > > > with
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> wildcard
>> > > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
>> > > matching
>> > > > > topic?
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Sure. When the new group
>> coordinator
>> > > > > takes
>> > > > > > > over,
>> > > > > > > > > > it
>> > > > > > > > > > > > has
>> > > > > > > > > > > > > > to:
>> > > > > > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
>> > > > > > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a
>> > > client
>> > > > > side
>> > > > > > > > > > assignor
>> > > > > > > > > > > > is
>> > > > > > > > > > > > > > used.
>> > > > > > > > > > > > > > > >> We
>> > > > > > > > > > > > > > > >> > > > don't
>> > > > > > > > > > > > > > > >> > > > > > > store the information about the
>> > > member
>> > > > > > > selected
>> > > > > > > > > > to run
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> assignment
>> > > > > > > > > > > > > > > >> > > > > > > so we have to start a new one.
>> > > > > > > > > > > > > > > >> > > > > > > * Update the topics metadata,
>> > > verify the
>> > > > > > > > > wildcard
>> > > > > > > > > > > > > > > >> subscriptions, and
>> > > > > > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 34.
>> ConsumerGroupMetadataValue,
>> > > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
>> > > > > > > > > > > > > > > >> > > > > > > >
>> ConsumerGroupMemberMetadataValue:
>> > > > > Could we
>> > > > > > > > > > document
>> > > > > > > > > > > > > what
>> > > > > > > > > > > > > > > >> the epoch
>> > > > > > > > > > > > > > > >> > > > > > field
>> > > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does
>> the
>> > > epoch
>> > > > > in
>> > > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
>> > > > > > > > > > > > > > > >> > > > > > reflect
>> > > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
>> > > about the
>> > > > > > > one in
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > ConsumerGroupPartitionMetadataValue
>> > > > > and
>> > > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Sure. I will clarify that but
>> it is
>> > > > > always
>> > > > > > > the
>> > > > > > > > > > latest
>> > > > > > > > > > > > > > group
>> > > > > > > > > > > > > > > >> epoch.
>> > > > > > > > > > > > > > > >> > > > > > > When the group state is
>> updated, the
>> > > > > group
>> > > > > > > epoch
>> > > > > > > > > > is
>> > > > > > > > > > > > > > bumped so
>> > > > > > > > > > > > > > > >> we use
>> > > > > > > > > > > > > > > >> > > > > > > that one for all the change
>> records
>> > > > > related
>> > > > > > > to
>> > > > > > > > > the
>> > > > > > > > > > > > > update.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator
>> will
>> > > ensure
>> > > > > > > that
>> > > > > > > > > the
>> > > > > > > > > > > > > > following
>> > > > > > > > > > > > > > > >> > > > invariants
>> > > > > > > > > > > > > > > >> > > > > > are
>> > > > > > > > > > > > > > > >> > > > > > > > met: ... All members
>> exists." It's
>> > > > > > > possible
>> > > > > > > > > for
>> > > > > > > > > > a
>> > > > > > > > > > > > > member
>> > > > > > > > > > > > > > > >> not to
>> > > > > > > > > > > > > > > >> > > > get any
>> > > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > That's right. Here I meant
>> that the
>> > > > > members
>> > > > > > > > > > provided
>> > > > > > > > > > > > by
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> assignor
>> > > > > > > > > > > > > > > >> > > > > > > in the assignment must exist
>> in the
>> > > > > group.
>> > > > > > > The
>> > > > > > > > > > > > assignor
>> > > > > > > > > > > > > > can
>> > > > > > > > > > > > > > > >> not make
>> > > > > > > > > > > > > > > >> > > > > > > up new member ids.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
>> > > with a
>> > > > > > > member
>> > > > > > > > > > epoch
>> > > > > > > > > > > > > > equals to
>> > > > > > > > > > > > > > > >> 0":
>> > > > > > > > > > > > > > > >> > > > When
>> > > > > > > > > > > > > > > >> > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what
>> member
>> > > id
>> > > > > > > would be
>> > > > > > > > > > used?
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > A member is expected to
>> abandon all
>> > > its
>> > > > > > > > > > partitions and
>> > > > > > > > > > > > > > > >> rejoins when
>> > > > > > > > > > > > > > > >> > > > it
>> > > > > > > > > > > > > > > >> > > > > > > receives the
>> FENCED_MEMBER_EPOCH
>> > > error.
>> > > > > In
>> > > > > > > this
>> > > > > > > > > > case,
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > > coordinator will have removed
>> the
>> > > member
>> > > > > > > from
>> > > > > > > > > the
>> > > > > > > > > > > > group.
>> > > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> member
>> > > > > > > > > > > > > > > >> > > > > > > can rejoin the group with the
>> same
>> > > > > member
>> > > > > > > id but
>> > > > > > > > > > with
>> > > > > > > > > > > > 0
>> > > > > > > > > > > > > as
>> > > > > > > > > > > > > > > >> epoch. Let
>> > > > > > > > > > > > > > > >> > > > > > > me see if I can clarify this
>> in the
>> > > KIP.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users
>> will
>> > > have
>> > > > > the
>> > > > > > > > > ability
>> > > > > > > > > > to
>> > > > > > > > > > > > > > trigger a
>> > > > > > > > > > > > > > > >> > > > > > reassignment
>> > > > > > > > > > > > > > > >> > > > > > > > by either providing a
>> non-zero
>> > > reason
>> > > > > or
>> > > > > > > by
>> > > > > > > > > > updating
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> assignor
>> > > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems
>> to be
>> > > > > > > conflicting
>> > > > > > > > > > with
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> deprecation
>> > > > > > > > > > > > > > > >> > > > of
>> > > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > In this case, a new assignment
>> is
>> > > > > triggered
>> > > > > > > by
>> > > > > > > > > the
>> > > > > > > > > > > > > client
>> > > > > > > > > > > > > > side
>> > > > > > > > > > > > > > > >> > > > > > > assignor. When constructing
>> the HB,
>> > > the
>> > > > > > > consumer
>> > > > > > > > > > will
>> > > > > > > > > > > > > > always
>> > > > > > > > > > > > > > > >> consult
>> > > > > > > > > > > > > > > >> > > > > > > the client side assignor and
>> > > propagate
>> > > > > the
>> > > > > > > > > > information
>> > > > > > > > > > > > > to
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > > coordinator. In other words, we
>> > > don't
>> > > > > expect
>> > > > > > > > > > users to
>> > > > > > > > > > > > > call
>> > > > > > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance
>> anymore.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment
>> examples are
>> > > > > nice.
>> > > > > > > But
>> > > > > > > > > the
>> > > > > > > > > > > > > section
>> > > > > > > > > > > > > > > >> seems to
>> > > > > > > > > > > > > > > >> > > > have
>> > > > > > > > > > > > > > > >> > > > > > > > multiple typos.
>> > > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group
>> transitions to
>> > > > > epoch
>> > > > > > > 2, B
>> > > > > > > > > > > > > > immediately
>> > > > > > > > > > > > > > > >> gets into
>> > > > > > > > > > > > > > > >> > > > > > > > "epoch=1,
>> partitions=[foo-2]",
>> > > which
>> > > > > seems
>> > > > > > > > > > > > incorrect.
>> > > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group
>> transitions to
>> > > > > epoch
>> > > > > > > 3, C
>> > > > > > > > > > seems
>> > > > > > > > > > > > to
>> > > > > > > > > > > > > > get
>> > > > > > > > > > > > > > > >> into
>> > > > > > > > > > > > > > > >> > > > > > epoch=3,
>> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
>> > > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to
>> epoch
>> > > 3, C
>> > > > > > > still
>> > > > > > > > > > has A -
>> > > > > > > > > > > > > > > >> epoch=2,
>> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Sorry for that! I will revise
>> them.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of
>> consumers:
>> > > Do
>> > > > > we
>> > > > > > > > > support
>> > > > > > > > > > the
>> > > > > > > > > > > > > > upgrade
>> > > > > > > > > > > > > > > >> from
>> > > > > > > > > > > > > > > >> > > > any
>> > > > > > > > > > > > > > > >> > > > > > old
>> > > > > > > > > > > > > > > >> > > > > > > > version to new one?
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > We will support upgrading from
>> the
>> > > > > consumer
>> > > > > > > > > > protocol
>> > > > > > > > > > > > > > version
>> > > > > > > > > > > > > > > >> 3,
>> > > > > > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792
>> is
>> > > not
>> > > > > > > > > implemented
>> > > > > > > > > > yet
>> > > > > > > > > > > > so
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> earliest
>> > > > > > > > > > > > > > > >> > > > > > > version is unknown at the
>> moment.
>> > > This
>> > > > > is
>> > > > > > > > > > explained in
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> migration
>> > > > > > > > > > > > > > > >> > > > > > > plan chapter.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Thanks again for your feedback,
>> > > Jun. I
>> > > > > will
>> > > > > > > > > > update the
>> > > > > > > > > > > > > KIP
>> > > > > > > > > > > > > > > >> based on
>> > > > > > > > > > > > > > > >> > > > it
>> > > > > > > > > > > > > > > >> > > > > > > next week.
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > Best,
>> > > > > > > > > > > > > > > >> > > > > > > David
>> > > > > > > > > > > > > > > >> > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM
>> Jun
>> > > Rao
>> > > > > > > > > > > > > > > >> <ju...@confluent.io.invalid>
>> > > > > > > > > > > > > > > >> > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > Hi, David,
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall,
>> the
>> > > main
>> > > > > > > benefits
>> > > > > > > > > > of
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > KIP
>> > > > > > > > > > > > > > > >> seem to
>> > > > > > > > > > > > > > > >> > > > be
>> > > > > > > > > > > > > > > >> > > > > > fewer
>> > > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and
>> more
>> > > > > efficient
>> > > > > > > > > > support of
>> > > > > > > > > > > > > > > >> wildcard. A few
>> > > > > > > > > > > > > > > >> > > > > > > > comments below.
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 30.
>> ConsumerGroupHeartbeatRequest
>> > > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
>> > > singleton.
>> > > > > Do we
>> > > > > > > plan
>> > > > > > > > > > to
>> > > > > > > > > > > > > > support
>> > > > > > > > > > > > > > > >> rolling
>> > > > > > > > > > > > > > > >> > > > > > changing
>> > > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in
>> the
>> > > > > > > consumers?
>> > > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could
>> you
>> > > explain
>> > > > > > > whether
>> > > > > > > > > > it's
>> > > > > > > > > > > > > > > >> required in
>> > > > > > > > > > > > > > > >> > > > every
>> > > > > > > > > > > > > > > >> > > > > > > > request or the scenarios
>> when it
>> > > > > needs to
>> > > > > > > be
>> > > > > > > > > > filled?
>> > > > > > > > > > > > > For
>> > > > > > > > > > > > > > > >> example,
>> > > > > > > > > > > > > > > >> > > > it's
>> > > > > > > > > > > > > > > >> > > > > > not
>> > > > > > > > > > > > > > > >> > > > > > > > clear to me when
>> TopicPartitions
>> > > > > needs to
>> > > > > > > be
>> > > > > > > > > > filled.
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
>> > > protocol,
>> > > > > the
>> > > > > > > rack
>> > > > > > > > > > > > > affinity
>> > > > > > > > > > > > > > > >> between the
>> > > > > > > > > > > > > > > >> > > > > > client
>> > > > > > > > > > > > > > > >> > > > > > > > and the broker is only
>> considered
>> > > > > during
>> > > > > > > > > > fetching,
>> > > > > > > > > > > > but
>> > > > > > > > > > > > > > not
>> > > > > > > > > > > > > > > >> during
>> > > > > > > > > > > > > > > >> > > > > > assigning
>> > > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
>> > > Sometimes,
>> > > > > once
>> > > > > > > the
>> > > > > > > > > > > > > assignment
>> > > > > > > > > > > > > > is
>> > > > > > > > > > > > > > > >> made,
>> > > > > > > > > > > > > > > >> > > > there
>> > > > > > > > > > > > > > > >> > > > > > is
>> > > > > > > > > > > > > > > >> > > > > > > > no opportunity for read
>> affinity
>> > > > > because
>> > > > > > > no
>> > > > > > > > > > replicas
>> > > > > > > > > > > > > of
>> > > > > > > > > > > > > > > >> assigned
>> > > > > > > > > > > > > > > >> > > > > > partitions
>> > > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
>> > > > > wondering
>> > > > > > > if we
>> > > > > > > > > > should
>> > > > > > > > > > > > > use
>> > > > > > > > > > > > > > > >> this
>> > > > > > > > > > > > > > > >> > > > > > opportunity
>> > > > > > > > > > > > > > > >> > > > > > > > to address this by including
>> rack
>> > > in
>> > > > > > > > > > GroupMember.
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side,
>> often,
>> > > it's
>> > > > > > > useful to
>> > > > > > > > > > know
>> > > > > > > > > > > > how
>> > > > > > > > > > > > > > busy
>> > > > > > > > > > > > > > > >> a group
>> > > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
>> > > event
>> > > > > loop
>> > > > > > > > > model,
>> > > > > > > > > > it
>> > > > > > > > > > > > > seems
>> > > > > > > > > > > > > > > >> that we
>> > > > > > > > > > > > > > > >> > > > could
>> > > > > > > > > > > > > > > >> > > > > > add
>> > > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the
>> fraction
>> > > of
>> > > > > the
>> > > > > > > time
>> > > > > > > > > > the
>> > > > > > > > > > > > > event
>> > > > > > > > > > > > > > > >> loop is
>> > > > > > > > > > > > > > > >> > > > doing
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > actual work.
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
>> > > > > coordinator
>> > > > > > > > > > failover
>> > > > > > > > > > > > > > > >> handling? For
>> > > > > > > > > > > > > > > >> > > > > > example,
>> > > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the
>> check
>> > > if
>> > > > > any
>> > > > > > > group
>> > > > > > > > > > with
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> wildcard
>> > > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
>> > > matching
>> > > > > topic?
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 34.
>> ConsumerGroupMetadataValue,
>> > > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
>> > > > > > > > > > > > > > > >> > > > > > > >
>> ConsumerGroupMemberMetadataValue:
>> > > > > Could we
>> > > > > > > > > > document
>> > > > > > > > > > > > > what
>> > > > > > > > > > > > > > > >> the epoch
>> > > > > > > > > > > > > > > >> > > > > > field
>> > > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does
>> the
>> > > epoch
>> > > > > in
>> > > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
>> > > > > > > > > > > > > > > >> > > > > > reflect
>> > > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
>> > > about the
>> > > > > > > one in
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > ConsumerGroupPartitionMetadataValue
>> > > > > and
>> > > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator
>> will
>> > > ensure
>> > > > > > > that
>> > > > > > > > > the
>> > > > > > > > > > > > > > following
>> > > > > > > > > > > > > > > >> > > > invariants
>> > > > > > > > > > > > > > > >> > > > > > are
>> > > > > > > > > > > > > > > >> > > > > > > > met: ... All members
>> exists." It's
>> > > > > > > possible
>> > > > > > > > > for
>> > > > > > > > > > a
>> > > > > > > > > > > > > member
>> > > > > > > > > > > > > > > >> not to
>> > > > > > > > > > > > > > > >> > > > get any
>> > > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
>> > > with a
>> > > > > > > member
>> > > > > > > > > > epoch
>> > > > > > > > > > > > > > equals to
>> > > > > > > > > > > > > > > >> 0":
>> > > > > > > > > > > > > > > >> > > > When
>> > > > > > > > > > > > > > > >> > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what
>> member
>> > > id
>> > > > > > > would be
>> > > > > > > > > > used?
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users
>> will
>> > > have
>> > > > > the
>> > > > > > > > > ability
>> > > > > > > > > > to
>> > > > > > > > > > > > > > trigger a
>> > > > > > > > > > > > > > > >> > > > > > reassignment
>> > > > > > > > > > > > > > > >> > > > > > > > by either providing a
>> non-zero
>> > > reason
>> > > > > or
>> > > > > > > by
>> > > > > > > > > > updating
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> assignor
>> > > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems
>> to be
>> > > > > > > conflicting
>> > > > > > > > > > with
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> deprecation
>> > > > > > > > > > > > > > > >> > > > of
>> > > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment
>> examples are
>> > > > > nice.
>> > > > > > > But
>> > > > > > > > > the
>> > > > > > > > > > > > > section
>> > > > > > > > > > > > > > > >> seems to
>> > > > > > > > > > > > > > > >> > > > have
>> > > > > > > > > > > > > > > >> > > > > > > > multiple typos.
>> > > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group
>> transitions to
>> > > > > epoch
>> > > > > > > 2, B
>> > > > > > > > > > > > > > immediately
>> > > > > > > > > > > > > > > >> gets into
>> > > > > > > > > > > > > > > >> > > > > > > > "epoch=1,
>> partitions=[foo-2]",
>> > > which
>> > > > > seems
>> > > > > > > > > > > > incorrect.
>> > > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group
>> transitions to
>> > > > > epoch
>> > > > > > > 3, C
>> > > > > > > > > > seems
>> > > > > > > > > > > > to
>> > > > > > > > > > > > > > get
>> > > > > > > > > > > > > > > >> into
>> > > > > > > > > > > > > > > >> > > > > > epoch=3,
>> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
>> > > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to
>> epoch
>> > > 3, C
>> > > > > > > still
>> > > > > > > > > > has A -
>> > > > > > > > > > > > > > > >> epoch=2,
>> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of
>> consumers:
>> > > Do
>> > > > > we
>> > > > > > > > > support
>> > > > > > > > > > the
>> > > > > > > > > > > > > > upgrade
>> > > > > > > > > > > > > > > >> from
>> > > > > > > > > > > > > > > >> > > > any
>> > > > > > > > > > > > > > > >> > > > > > old
>> > > > > > > > > > > > > > > >> > > > > > > > version to new one?
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > Thanks,
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > Jun
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20
>> AM
>> > > David
>> > > > > > > Jacot
>> > > > > > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
>> > > > > > > > > > > > > > > >> > > > > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > Hi all,
>> > > > > > > > > > > > > > > >> > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > The KIP states that we will
>> > > > > > > re-implement the
>> > > > > > > > > > > > > > coordinator
>> > > > > > > > > > > > > > > >> in
>> > > > > > > > > > > > > > > >> > > > Java. I
>> > > > > > > > > > > > > > > >> > > > > > > > > discussed this offline
>> with a
>> > > few
>> > > > > folks
>> > > > > > > and
>> > > > > > > > > > folks
>> > > > > > > > > > > > > are
>> > > > > > > > > > > > > > > >> concerned
>> > > > > > > > > > > > > > > >> > > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > we could introduce many
>> > > regressions
>> > > > > in
>> > > > > > > the
>> > > > > > > > > old
>> > > > > > > > > > > > > > protocol
>> > > > > > > > > > > > > > > >> if we do
>> > > > > > > > > > > > > > > >> > > > so.
>> > > > > > > > > > > > > > > >> > > > > > > > > Therefore, I am going to
>> remove
>> > > this
>> > > > > > > > > statement
>> > > > > > > > > > > > from
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> KIP. It
>> > > > > > > > > > > > > > > >> > > > is an
>> > > > > > > > > > > > > > > >> > > > > > > > > implementation detail
>> after all
>> > > so
>> > > > > it
>> > > > > > > does
>> > > > > > > > > not
>> > > > > > > > > > > > have
>> > > > > > > > > > > > > > to be
>> > > > > > > > > > > > > > > >> > > > decided at
>> > > > > > > > > > > > > > > >> > > > > > > > > this stage. We will likely
>> > > start by
>> > > > > > > trying
>> > > > > > > > > to
>> > > > > > > > > > > > > > refactor the
>> > > > > > > > > > > > > > > >> > > > current
>> > > > > > > > > > > > > > > >> > > > > > > > > implementation as a first
>> step.
>> > > > > > > > > > > > > > > >> > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > Cheers,
>> > > > > > > > > > > > > > > >> > > > > > > > > David
>> > > > > > > > > > > > > > > >> > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at
>> 3:52 PM
>> > > > > David
>> > > > > > > Jacot
>> > > > > > > > > <
>> > > > > > > > > > > > > > > >> djacot@confluent.io
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > Hi Luke,
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
>> > > machine
>> > > > > are:
>> > > > > > > > > > "Empty,
>> > > > > > > > > > > > > > assigning,
>> > > > > > > > > > > > > > > >> > > > > > reconciling,
>> > > > > > > > > > > > > > > >> > > > > > > > > stable,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in
>> Consumer
>> > > > > Group
>> > > > > > > States
>> > > > > > > > > > > > > section,
>> > > > > > > > > > > > > > > >> right?
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > This sentence does not
>> refer
>> > > to
>> > > > > those
>> > > > > > > > > group
>> > > > > > > > > > > > states
>> > > > > > > > > > > > > > but
>> > > > > > > > > > > > > > > >> rather
>> > > > > > > > > > > > > > > >> > > > to a
>> > > > > > > > > > > > > > > >> > > > > > > > > > state machine replication
>> > > (SMR).
>> > > > > This
>> > > > > > > > > > refers to
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> entire
>> > > > > > > > > > > > > > > >> > > > state of
>> > > > > > > > > > > > > > > >> > > > > > > > > > group coordinator which
>> is
>> > > > > replicated
>> > > > > > > via
>> > > > > > > > > > the
>> > > > > > > > > > > > log
>> > > > > > > > > > > > > > > >> layer. I will
>> > > > > > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean
>> "each
>> > > > > state
>> > > > > > > > > machine
>> > > > > > > > > > is
>> > > > > > > > > > > > > > modelled
>> > > > > > > > > > > > > > > >> as an
>> > > > > > > > > > > > > > > >> > > > event
>> > > > > > > > > > > > > > > >> > > > > > > > > loop"?
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > The idea is to follow a
>> model
>> > > > > similar
>> > > > > > > to
>> > > > > > > > > > the new
>> > > > > > > > > > > > > > quorum
>> > > > > > > > > > > > > > > >> > > > > > controller. We
>> > > > > > > > > > > > > > > >> > > > > > > > > > will have N threads to
>> process
>> > > > > events.
>> > > > > > > > > Each
>> > > > > > > > > > > > > > > >> __consumer_offsets
>> > > > > > > > > > > > > > > >> > > > > > > > > > partition is assigned to
>> a
>> > > unique
>> > > > > > > thread
>> > > > > > > > > > and all
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> events
>> > > > > > > > > > > > > > > >> > > > (e.g.
>> > > > > > > > > > > > > > > >> > > > > > > > > > requests, callbacks,
>> etc.) are
>> > > > > > > processed
>> > > > > > > > > by
>> > > > > > > > > > this
>> > > > > > > > > > > > > > > >> thread. This
>> > > > > > > > > > > > > > > >> > > > > > simplify
>> > > > > > > > > > > > > > > >> > > > > > > > > > concurrency and will
>> enable
>> > > us to
>> > > > > do
>> > > > > > > > > > simulation
>> > > > > > > > > > > > > > testing
>> > > > > > > > > > > > > > > >> for the
>> > > > > > > > > > > > > > > >> > > > > > group
>> > > > > > > > > > > > > > > >> > > > > > > > > > coordinator.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a
>> state
>> > > > > machine
>> > > > > > > per
>> > > > > > > > > > > > > > > >> *__consumer_offsets*
>> > > > > > > > > > > > > > > >> > > > > > > > > partitions?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine
>> "per
>> > > > > consumer
>> > > > > > > group"
>> > > > > > > > > > owned
>> > > > > > > > > > > > > by
>> > > > > > > > > > > > > > a
>> > > > > > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > coordinator?
>> > > > > > > > > > > > > > > >> > > > > > > > > For
>> > > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
>> > > > > coordinator
>> > > > > > > owns 2
>> > > > > > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> groups, and
>> > > > > > > > > > > > > > > >> > > > > > both
>> > > > > > > > > > > > > > > >> > > > > > > > > exist in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
>> > > will we
>> > > > > > > have 1
>> > > > > > > > > > state
>> > > > > > > > > > > > > > machine
>> > > > > > > > > > > > > > > >> for it,
>> > > > > > > > > > > > > > > >> > > > or
>> > > > > > > > > > > > > > > >> > > > > > 2?
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion
>> comes
>> > > from
>> > > > > > > there, I
>> > > > > > > > > > > > think.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
>> > > > > > > > > > "*group.coordinator.threads"
>> > > > > > > > > > > > *is
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> number
>> > > > > > > > > > > > > > > >> > > > of
>> > > > > > > > > > > > > > > >> > > > > > threads
>> > > > > > > > > > > > > > > >> > > > > > > > > used
>> > > > > > > > > > > > > > > >> > > > > > > > > > > to run the state
>> machines.
>> > > But
>> > > > > I'm
>> > > > > > > > > > wondering
>> > > > > > > > > > > > if
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> purpose
>> > > > > > > > > > > > > > > >> > > > of
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > threads
>> > > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the
>> state of
>> > > > > each
>> > > > > > > > > consumer
>> > > > > > > > > > > > group
>> > > > > > > > > > > > > > (or
>> > > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
>> > > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no
>> heavy
>> > > > > > > computation,
>> > > > > > > > > > why
>> > > > > > > > > > > > > > should we
>> > > > > > > > > > > > > > > >> need
>> > > > > > > > > > > > > > > >> > > > > > > > > multi-threads
>> > > > > > > > > > > > > > > >> > > > > > > > > > > here?
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to
>> have
>> > > an
>> > > > > > > ability to
>> > > > > > > > > > shard
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > processing as
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > computation could be
>> heavy.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer
>> session
>> > > > > timeout,
>> > > > > > > why
>> > > > > > > > > > does
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> default
>> > > > > > > > > > > > > > > >> > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > timeout not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > locate between min
>> (45s) and
>> > > > > > > max(60s)? I
>> > > > > > > > > > > > thought
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> min/max
>> > > > > > > > > > > > > > > >> > > > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
>> > > lower/upper
>> > > > > > > bound
>> > > > > > > > > of
>> > > > > > > > > > it,
>> > > > > > > > > > > > > no?
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > group.consumer.session.timeout.ms
>> > > > > > > int
>> > > > > > > > > > 30s The
>> > > > > > > > > > > > > > > >> timeout to
>> > > > > > > > > > > > > > > >> > > > detect
>> > > > > > > > > > > > > > > >> > > > > > client
>> > > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
>> > > consumer
>> > > > > > > group
>> > > > > > > > > > > > protocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > group.consumer.min.session.timeout.ms
>> > > > > > > > > > int 45s
>> > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> minimum
>> > > > > > > > > > > > > > > >> > > > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > group.consumer.max.session.timeout.ms
>> > > > > > > > > > int 60s
>> > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> maximum
>> > > > > > > > > > > > > > > >> > > > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > This is indeed a
>> mistake. The
>> > > > > default
>> > > > > > > > > > session
>> > > > > > > > > > > > > > timeout
>> > > > > > > > > > > > > > > >> should
>> > > > > > > > > > > > > > > >> > > > be 45s
>> > > > > > > > > > > > > > > >> > > > > > > > > > (the current default).
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default
>> server side
>> > > > > > > assignor
>> > > > > > > > > are
>> > > > > > > > > > > > > [range,
>> > > > > > > > > > > > > > > >> uniform],
>> > > > > > > > > > > > > > > >> > > > > > which means
>> > > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to
>> "range"
>> > > > > assignor.
>> > > > > > > I'd
>> > > > > > > > > > like to
>> > > > > > > > > > > > > > know
>> > > > > > > > > > > > > > > >> why not
>> > > > > > > > > > > > > > > >> > > > > > uniform
>> > > > > > > > > > > > > > > >> > > > > > > > > one? I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users
>> will
>> > > > > choose
>> > > > > > > > > uniform
>> > > > > > > > > > > > > assignor
>> > > > > > > > > > > > > > > >> (former
>> > > > > > > > > > > > > > > >> > > > sticky
>> > > > > > > > > > > > > > > >> > > > > > > > > assinor)
>> > > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
>> > > distribution.
>> > > > > Any
>> > > > > > > > > other
>> > > > > > > > > > > > reason
>> > > > > > > > > > > > > > we
>> > > > > > > > > > > > > > > >> choose
>> > > > > > > > > > > > > > > >> > > > range
>> > > > > > > > > > > > > > > >> > > > > > > > > assignor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > as default?
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> group.consumer.assignors
>> > > List
>> > > > > range,
>> > > > > > > > > > uniform
>> > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> server side
>> > > > > > > > > > > > > > > >> > > > > > assignors.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > The order on the server
>> side
>> > > has
>> > > > > no
>> > > > > > > > > > influence
>> > > > > > > > > > > > > > because
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > client
>> > > > > > > > > > > > > > > >> > > > > > must
>> > > > > > > > > > > > > > > >> > > > > > > > > > chose the selector that
>> he
>> > > wants
>> > > > > to
>> > > > > > > use.
>> > > > > > > > > > There
>> > > > > > > > > > > > is
>> > > > > > > > > > > > > no
>> > > > > > > > > > > > > > > >> default
>> > > > > > > > > > > > > > > >> > > > in the
>> > > > > > > > > > > > > > > >> > > > > > > > > > current proposal. If the
>> > > assignor
>> > > > > is
>> > > > > > > not
>> > > > > > > > > > > > specified
>> > > > > > > > > > > > > > by
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > client,
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > request is rejected. The
>> > > default
>> > > > > > > client
>> > > > > > > > > > value
>> > > > > > > > > > > > for
>> > > > > > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor`
>> is
>> > > > > `uniform`
>> > > > > > > > > though.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > Thanks for your very good
>> > > > > comments,
>> > > > > > > Luke.
>> > > > > > > > > I
>> > > > > > > > > > hope
>> > > > > > > > > > > > > > that my
>> > > > > > > > > > > > > > > >> > > > answers
>> > > > > > > > > > > > > > > >> > > > > > help
>> > > > > > > > > > > > > > > >> > > > > > > > > > to clarify things. I will
>> > > update
>> > > > > the
>> > > > > > > KIP
>> > > > > > > > > as
>> > > > > > > > > > well
>> > > > > > > > > > > > > > based
>> > > > > > > > > > > > > > > >> on your
>> > > > > > > > > > > > > > > >> > > > > > > > > > feedback.
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > Cheers,
>> > > > > > > > > > > > > > > >> > > > > > > > > > David
>> > > > > > > > > > > > > > > >> > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at
>> 9:29
>> > > AM
>> > > > > Luke
>> > > > > > > Chen
>> > > > > > > > > <
>> > > > > > > > > > > > > > > >> showuon@gmail.com>
>> > > > > > > > > > > > > > > >> > > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > Hi David,
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > Some more questions:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator
>> > > > > section, you
>> > > > > > > > > > > > mentioned:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > The new group
>> coordinator
>> > > will
>> > > > > > > have a
>> > > > > > > > > > state
>> > > > > > > > > > > > > > machine
>> > > > > > > > > > > > > > > >> per
>> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets*
>> > > partitions,
>> > > > > > > where
>> > > > > > > > > > each
>> > > > > > > > > > > > > state
>> > > > > > > > > > > > > > > >> machine is
>> > > > > > > > > > > > > > > >> > > > > > modelled
>> > > > > > > > > > > > > > > >> > > > > > > > > as an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > event loop. Those state
>> > > machines
>> > > > > > > will be
>> > > > > > > > > > > > > executed
>> > > > > > > > > > > > > > in
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> *group.coordinator.threads*
>> > > > > threads.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
>> > > machine
>> > > > > are:
>> > > > > > > > > > "Empty,
>> > > > > > > > > > > > > > assigning,
>> > > > > > > > > > > > > > > >> > > > > > reconciling,
>> > > > > > > > > > > > > > > >> > > > > > > > > stable,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in
>> Consumer
>> > > > > Group
>> > > > > > > States
>> > > > > > > > > > > > > section,
>> > > > > > > > > > > > > > > >> right?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean
>> "each
>> > > > > state
>> > > > > > > > > machine
>> > > > > > > > > > is
>> > > > > > > > > > > > > > modelled
>> > > > > > > > > > > > > > > >> as an
>> > > > > > > > > > > > > > > >> > > > event
>> > > > > > > > > > > > > > > >> > > > > > > > > loop"?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a
>> state
>> > > > > machine
>> > > > > > > per
>> > > > > > > > > > > > > > > >> *__consumer_offsets*
>> > > > > > > > > > > > > > > >> > > > > > > > > partitions?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine
>> "per
>> > > > > consumer
>> > > > > > > group"
>> > > > > > > > > > owned
>> > > > > > > > > > > > > by
>> > > > > > > > > > > > > > a
>> > > > > > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > coordinator?
>> > > > > > > > > > > > > > > >> > > > > > > > > For
>> > > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
>> > > > > coordinator
>> > > > > > > owns 2
>> > > > > > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> groups, and
>> > > > > > > > > > > > > > > >> > > > > > both
>> > > > > > > > > > > > > > > >> > > > > > > > > exist in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
>> > > will we
>> > > > > > > have 1
>> > > > > > > > > > state
>> > > > > > > > > > > > > > machine
>> > > > > > > > > > > > > > > >> for it,
>> > > > > > > > > > > > > > > >> > > > or
>> > > > > > > > > > > > > > > >> > > > > > 2?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
>> > > > > > > > > > "*group.coordinator.threads"
>> > > > > > > > > > > > *is
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> number
>> > > > > > > > > > > > > > > >> > > > of
>> > > > > > > > > > > > > > > >> > > > > > threads
>> > > > > > > > > > > > > > > >> > > > > > > > > used
>> > > > > > > > > > > > > > > >> > > > > > > > > > > to run the state
>> machines.
>> > > But
>> > > > > I'm
>> > > > > > > > > > wondering
>> > > > > > > > > > > > if
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> purpose
>> > > > > > > > > > > > > > > >> > > > of
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > threads
>> > > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the
>> state of
>> > > > > each
>> > > > > > > > > consumer
>> > > > > > > > > > > > group
>> > > > > > > > > > > > > > (or
>> > > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
>> > > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no
>> heavy
>> > > > > > > computation,
>> > > > > > > > > > why
>> > > > > > > > > > > > > > should we
>> > > > > > > > > > > > > > > >> need
>> > > > > > > > > > > > > > > >> > > > > > > > > multi-threads
>> > > > > > > > > > > > > > > >> > > > > > > > > > > here?
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 2. For the default
>> value in
>> > > the
>> > > > > new
>> > > > > > > > > > configs:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer
>> session
>> > > > > timeout,
>> > > > > > > why
>> > > > > > > > > > does
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> default
>> > > > > > > > > > > > > > > >> > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > timeout not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > locate between min
>> (45s) and
>> > > > > > > max(60s)? I
>> > > > > > > > > > > > thought
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> min/max
>> > > > > > > > > > > > > > > >> > > > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
>> > > lower/upper
>> > > > > > > bound
>> > > > > > > > > of
>> > > > > > > > > > it,
>> > > > > > > > > > > > > no?
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > group.consumer.session.timeout.ms
>> > > > > > > int
>> > > > > > > > > > 30s The
>> > > > > > > > > > > > > > > >> timeout to
>> > > > > > > > > > > > > > > >> > > > detect
>> > > > > > > > > > > > > > > >> > > > > > client
>> > > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
>> > > consumer
>> > > > > > > group
>> > > > > > > > > > > > protocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > group.consumer.min.session.timeout.ms
>> > > > > > > > > > int 45s
>> > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> minimum
>> > > > > > > > > > > > > > > >> > > > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > group.consumer.max.session.timeout.ms
>> > > > > > > > > > int 60s
>> > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> maximum
>> > > > > > > > > > > > > > > >> > > > > > session
>> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default
>> server side
>> > > > > > > assignor
>> > > > > > > > > are
>> > > > > > > > > > > > > [range,
>> > > > > > > > > > > > > > > >> uniform],
>> > > > > > > > > > > > > > > >> > > > > > which means
>> > > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to
>> "range"
>> > > > > assignor.
>> > > > > > > I'd
>> > > > > > > > > > like to
>> > > > > > > > > > > > > > know
>> > > > > > > > > > > > > > > >> why not
>> > > > > > > > > > > > > > > >> > > > > > uniform
>> > > > > > > > > > > > > > > >> > > > > > > > > one? I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users
>> will
>> > > > > choose
>> > > > > > > > > uniform
>> > > > > > > > > > > > > assignor
>> > > > > > > > > > > > > > > >> (former
>> > > > > > > > > > > > > > > >> > > > sticky
>> > > > > > > > > > > > > > > >> > > > > > > > > assinor)
>> > > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
>> > > distribution.
>> > > > > Any
>> > > > > > > > > other
>> > > > > > > > > > > > reason
>> > > > > > > > > > > > > > we
>> > > > > > > > > > > > > > > >> choose
>> > > > > > > > > > > > > > > >> > > > range
>> > > > > > > > > > > > > > > >> > > > > > > > > assignor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > as default?
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> group.consumer.assignors
>> > > List
>> > > > > range,
>> > > > > > > > > > uniform
>> > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> server side
>> > > > > > > > > > > > > > > >> > > > > > assignors.
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > Thank you.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > Luke
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at
>> > > 2:10 PM
>> > > > > Luke
>> > > > > > > > > Chen
>> > > > > > > > > > <
>> > > > > > > > > > > > > > > >> showuon@gmail.com
>> > > > > > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts
>> about
>> > > > > Kafka
>> > > > > > > > > Connect
>> > > > > > > > > > > > > > > >> integrating with
>> > > > > > > > > > > > > > > >> > > > > > KIP-848,
>> > > > > > > > > > > > > > > >> > > > > > > > > but I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > think we should have
>> a
>> > > > > separate
>> > > > > > > > > > discussion
>> > > > > > > > > > > > > > thread
>> > > > > > > > > > > > > > > >> for the
>> > > > > > > > > > > > > > > >> > > > Kafka
>> > > > > > > > > > > > > > > >> > > > > > > > > Connect
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating
>> Kafka
>> > > Connect
>> > > > > > > With
>> > > > > > > > > New
>> > > > > > > > > > > > > Consumer
>> > > > > > > > > > > > > > > >> Rebalance
>> > > > > > > > > > > > > > > >> > > > > > Protocol
>> > > > > > > > > > > > > > > >> > > > > > > > > [1],
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > and let this
>> discussion
>> > > thread
>> > > > > > > focus
>> > > > > > > > > on
>> > > > > > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> rebalance
>> > > > > > > > > > > > > > > >> > > > > > protocol,
>> > > > > > > > > > > > > > > >> > > > > > > > > WDYT?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > [1]
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >>
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > >
>> > > > > > > > >
>> > > > > > >
>> > > > >
>> > >
>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > Thank you.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > Luke
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022
>> at
>> > > 9:31
>> > > > > PM
>> > > > > > > Sagar
>> > > > > > > > > <
>> > > > > > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
>> > > > > > > > > > > > > > > >> > > > > > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Thank you
>> Guozhang/David
>> > > for
>> > > > > the
>> > > > > > > > > > feedback.
>> > > > > > > > > > > > > > Looks
>> > > > > > > > > > > > > > > >> like
>> > > > > > > > > > > > > > > >> > > > there's
>> > > > > > > > > > > > > > > >> > > > > > > > > agreement on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs
>> for
>> > > > > Connect.
>> > > > > > > I
>> > > > > > > > > > would
>> > > > > > > > > > > > > > revisit
>> > > > > > > > > > > > > > > >> the doc
>> > > > > > > > > > > > > > > >> > > > and
>> > > > > > > > > > > > > > > >> > > > > > see
>> > > > > > > > > > > > > > > >> > > > > > > > > what
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> changes are to be
>> made.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022
>> at
>> > > 7:11
>> > > > > PM
>> > > > > > > David
>> > > > > > > > > > Jacot
>> > > > > > > > > > > > > > > >> > > > > > > > >
>> <dj...@confluent.io.invalid>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the
>> feedback
>> > > > > and the
>> > > > > > > > > > document.
>> > > > > > > > > > > > > > That's
>> > > > > > > > > > > > > > > >> really
>> > > > > > > > > > > > > > > >> > > > > > helpful. I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will take a look
>> at it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems
>> to me
>> > > > > that
>> > > > > > > both
>> > > > > > > > > > Connect
>> > > > > > > > > > > > > > and the
>> > > > > > > > > > > > > > > >> > > > Consumer
>> > > > > > > > > > > > > > > >> > > > > > could
>> > > > > > > > > > > > > > > >> > > > > > > > > share
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the same
>> underlying
>> > > > > "engine".
>> > > > > > > The
>> > > > > > > > > > main
>> > > > > > > > > > > > > > > >> difference is
>> > > > > > > > > > > > > > > >> > > > that
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > Consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigns
>> > > topic-partitions to
>> > > > > > > members
>> > > > > > > > > > > > whereas
>> > > > > > > > > > > > > > > >> Connect
>> > > > > > > > > > > > > > > >> > > > assigns
>> > > > > > > > > > > > > > > >> > > > > > tasks
>> > > > > > > > > > > > > > > >> > > > > > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two
>> > > ways to
>> > > > > move
>> > > > > > > > > > forward:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the
>> new
>> > > > > proposed
>> > > > > > > APIs
>> > > > > > > > > to
>> > > > > > > > > > > > > support
>> > > > > > > > > > > > > > > >> different
>> > > > > > > > > > > > > > > >> > > > > > resource
>> > > > > > > > > > > > > > > >> > > > > > > > > types
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions,
>> > > tasks,
>> > > > > > > etc.); or
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new
>> dedicated
>> > > > > APIs
>> > > > > > > for
>> > > > > > > > > > Connect.
>> > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> dedicated
>> > > > > > > > > > > > > > > >> > > > APIs
>> > > > > > > > > > > > > > > >> > > > > > > > > would be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new
>> > > ones but
>> > > > > > > > > > different on
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > content/resources and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > they would rely
>> on the
>> > > same
>> > > > > > > engine
>> > > > > > > > > > on the
>> > > > > > > > > > > > > > > >> coordinator
>> > > > > > > > > > > > > > > >> > > > side.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean
>> > > towards
>> > > > > 2)
>> > > > > > > > > because
>> > > > > > > > > > I am
>> > > > > > > > > > > > > > not a
>> > > > > > > > > > > > > > > >> fan of
>> > > > > > > > > > > > > > > >> > > > > > > > > overcharging
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve
>> different
>> > > > > > > purposes.
>> > > > > > > > > > That
>> > > > > > > > > > > > > being
>> > > > > > > > > > > > > > > >> said, I am
>> > > > > > > > > > > > > > > >> > > > not
>> > > > > > > > > > > > > > > >> > > > > > > > > opposed to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find
>> an
>> > > > > elegant
>> > > > > > > way to
>> > > > > > > > > > do
>> > > > > > > > > > > > it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I think that we
>> can
>> > > > > continue to
>> > > > > > > > > > discuss
>> > > > > > > > > > > > it
>> > > > > > > > > > > > > > here
>> > > > > > > > > > > > > > > >> for now
>> > > > > > > > > > > > > > > >> > > > in
>> > > > > > > > > > > > > > > >> > > > > > order
>> > > > > > > > > > > > > > > >> > > > > > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this
>> KIP is
>> > > > > > > compatible
>> > > > > > > > > > with
>> > > > > > > > > > > > > what
>> > > > > > > > > > > > > > we
>> > > > > > > > > > > > > > > >> will do
>> > > > > > > > > > > > > > > >> > > > for
>> > > > > > > > > > > > > > > >> > > > > > > > > Connect in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the future.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Best,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > David
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8,
>> 2022 at
>> > > > > 2:41 PM
>> > > > > > > > > David
>> > > > > > > > > > > > Jacot
>> > > > > > > > > > > > > <
>> > > > > > > > > > > > > > > >> > > > > > djacot@confluent.io>
>> > > > > > > > > > > > > > > >> > > > > > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from
>> > > vacation.
>> > > > > I
>> > > > > > > will
>> > > > > > > > > go
>> > > > > > > > > > > > > through
>> > > > > > > > > > > > > > and
>> > > > > > > > > > > > > > > >> address
>> > > > > > > > > > > > > > > >> > > > > > your
>> > > > > > > > > > > > > > > >> > > > > > > > > comments
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming
>> days.
>> > > > > Thanks
>> > > > > > > for
>> > > > > > > > > your
>> > > > > > > > > > > > > > feedback.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > David
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3,
>> 2022
>> > > at
>> > > > > 10:05
>> > > > > > > PM
>> > > > > > > > > > Gregory
>> > > > > > > > > > > > > > Harris
>> > > > > > > > > > > > > > > >> <
>> > > > > > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for
>> the KIP,
>> > > > > it's
>> > > > > > > > > > wonderful to
>> > > > > > > > > > > > > see
>> > > > > > > > > > > > > > > >> > > > cooperative
>> > > > > > > > > > > > > > > >> > > > > > > > > rebalancing
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > making it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > down the
>> stack!
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few
>> > > questions:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The
>> 'Rejected
>> > > > > > > Alternatives'
>> > > > > > > > > > > > section
>> > > > > > > > > > > > > > > >> describes how
>> > > > > > > > > > > > > > > >> > > > > > member
>> > > > > > > > > > > > > > > >> > > > > > > > > epoch
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in
>> step
>> > > with
>> > > > > the
>> > > > > > > group
>> > > > > > > > > > epoch
>> > > > > > > > > > > > > and
>> > > > > > > > > > > > > > > >> assignment
>> > > > > > > > > > > > > > > >> > > > > > epoch
>> > > > > > > > > > > > > > > >> > > > > > > > > values. I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is
>> a good
>> > > > > idea
>> > > > > > > for
>> > > > > > > > > the
>> > > > > > > > > > > > > reasons
>> > > > > > > > > > > > > > > >> described
>> > > > > > > > > > > > > > > >> > > > in
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > KIP. When
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is
>> > > > > incrementally
>> > > > > > > > > > assigning
>> > > > > > > > > > > > > > > >> partitions to a
>> > > > > > > > > > > > > > > >> > > > > > worker,
>> > > > > > > > > > > > > > > >> > > > > > > > > what
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> member
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does
>> each
>> > > > > incremental
>> > > > > > > > > > > > assignment
>> > > > > > > > > > > > > > use?
>> > > > > > > > > > > > > > > >> Are
>> > > > > > > > > > > > > > > >> > > > member
>> > > > > > > > > > > > > > > >> > > > > > epochs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> re-used,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > a single
>> member
>> > > epoch
>> > > > > can
>> > > > > > > > > > correspond
>> > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> multiple
>> > > > > > > > > > > > > > > >> > > > > > different
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > larger)
>> > > assignments?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the
>> > > Assignor's
>> > > > > > > 'Reason'
>> > > > > > > > > > field
>> > > > > > > > > > > > > > opaque
>> > > > > > > > > > > > > > > >> to the
>> > > > > > > > > > > > > > > >> > > > group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should
>> custom
>> > > > > > > client-side
>> > > > > > > > > > > > assignor
>> > > > > > > > > > > > > > > >> > > > implementations
>> > > > > > > > > > > > > > > >> > > > > > > > > interact
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field,
>> and
>> > > how
>> > > > > is
>> > > > > > > its
>> > > > > > > > > > common
>> > > > > > > > > > > > > > meaning
>> > > > > > > > > > > > > > > >> agreed
>> > > > > > > > > > > > > > > >> > > > > > upon? If
>> > > > > > > > > > > > > > > >> > > > > > > > > so, what
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > is the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a
>> > > distinct
>> > > > > > > Reason
>> > > > > > > > > > field
>> > > > > > > > > > > > over
>> > > > > > > > > > > > > > > >> including
>> > > > > > > > > > > > > > > >> > > > such
>> > > > > > > > > > > > > > > >> > > > > > > > > functionality
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque
>> metadata?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The
>> following is
>> > > > > > > included in
>> > > > > > > > > > the
>> > > > > > > > > > > > > KIP:
>> > > > > > > > > > > > > > > >> "Thanks to
>> > > > > > > > > > > > > > > >> > > > > > this, the
>> > > > > > > > > > > > > > > >> > > > > > > > > input
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client side
>> > > assignor is
>> > > > > > > > > entirely
>> > > > > > > > > > > > driven
>> > > > > > > > > > > > > > by
>> > > > > > > > > > > > > > > >> the group
>> > > > > > > > > > > > > > > >> > > > > > > > > coordinator.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> The
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no
>> > > longer
>> > > > > > > > > > responsible for
>> > > > > > > > > > > > > > > >> maintaining
>> > > > > > > > > > > > > > > >> > > > any
>> > > > > > > > > > > > > > > >> > > > > > state
>> > > > > > > > > > > > > > > >> > > > > > > > > besides
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> its
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned
>> > > partitions."
>> > > > > Does
>> > > > > > > this
>> > > > > > > > > > mean
>> > > > > > > > > > > > > > that the
>> > > > > > > > > > > > > > > >> > > > > > client-side
>> > > > > > > > > > > > > > > >> > > > > > > > > assignor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> MAY
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate
>> > > additional
>> > > > > > > > > > non-Metadata
>> > > > > > > > > > > > > state
>> > > > > > > > > > > > > > > >> (such as
>> > > > > > > > > > > > > > > >> > > > > > partition
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory
>> metrics,
>> > > > > config
>> > > > > > > > > > topics,
>> > > > > > > > > > > > > etc),
>> > > > > > > > > > > > > > or
>> > > > > > > > > > > > > > > >> that
>> > > > > > > > > > > > > > > >> > > > > > additional
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD
>> NOT be
>> > > > > used?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that
>> > > there are
>> > > > > > > > > separate
>> > > > > > > > > > > > > classes
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > for
>> > > > > > > > > > > > > > > >> > > > > >
>> > > > > > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and
>> > > > > > > > > > > > > > > >> > > >
>> > > > > > > org.apache.kafka.clients.consumer.PartitionAssignor
>> > > > > > > > > > > > > > > >> > > > > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > seem to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap
>> > > significantly.
>> > > > > Is
>> > > > > > > it
>> > > > > > > > > > possible
>> > > > > > > > > > > > > for
>> > > > > > > > > > > > > > > >> these two
>> > > > > > > > > > > > > > > >> > > > > > > > > implementations
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This
>> would
>> > > > > serve
>> > > > > > > to
>> > > > > > > > > > promote
>> > > > > > > > > > > > > > feature
>> > > > > > > > > > > > > > > >> parity
>> > > > > > > > > > > > > > > >> > > > of
>> > > > > > > > > > > > > > > >> > > > > > > > > server-side
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side
>> > > assignors,
>> > > > > and
>> > > > > > > > > would
>> > > > > > > > > > also
>> > > > > > > > > > > > > > > >> facilitate
>> > > > > > > > > > > > > > > >> > > > > > operational
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > certain
>> > > situations. For
>> > > > > > > > > example,
>> > > > > > > > > > if a
>> > > > > > > > > > > > > > > >> server-side
>> > > > > > > > > > > > > > > >> > > > > > assignor
>> > > > > > > > > > > > > > > >> > > > > > > > > has some
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > poor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and
>> needs
>> > > a
>> > > > > patch,
>> > > > > > > > > > deploying
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> patched
>> > > > > > > > > > > > > > > >> > > > > > assignor to
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching
>> one
>> > > > > consumer
>> > > > > > > > > group
>> > > > > > > > > > to a
>> > > > > > > > > > > > > > > >> client-side
>> > > > > > > > > > > > > > > >> > > > > > assignor
>> > > > > > > > > > > > > > > >> > > > > > > > > may be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > faster
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and less
>> risky than
>> > > > > > > patching
>> > > > > > > > > all
>> > > > > > > > > > of
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> brokers.
>> > > > > > > > > > > > > > > >> > > > With
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > currently
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed
>> distinct
>> > > > > APIs, a
>> > > > > > > > > > non-trivial
>> > > > > > > > > > > > > > > >> > > > reimplementation
>> > > > > > > > > > > > > > > >> > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > have
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled,
>> and if
>> > > the
>> > > > > two
>> > > > > > > APIs
>> > > > > > > > > > have
>> > > > > > > > > > > > > > diverged
>> > > > > > > > > > > > > > > >> > > > > > significantly,
>> > > > > > > > > > > > > > > >> > > > > > > > > then it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that
>> a
>> > > > > > > > > reimplementation
>> > > > > > > > > > > > would
>> > > > > > > > > > > > > > not be
>> > > > > > > > > > > > > > > >> > > > possible.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > --
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > gharris1727@gmail.com
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > github.com/gharris1727
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3,
>> > > 2022 at
>> > > > > > > 8:39 AM
>> > > > > > > > > > Sagar
>> > > > > > > > > > > > <
>> > > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi
>> > > Guozhang/David,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a
>> > > > > confluence
>> > > > > > > page
>> > > > > > > > > to
>> > > > > > > > > > > > > discuss
>> > > > > > > > > > > > > > how
>> > > > > > > > > > > > > > > >> Connect
>> > > > > > > > > > > > > > > >> > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > need to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > change
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on
>> the new
>> > > > > > > rebalance
>> > > > > > > > > > > > protocol.
>> > > > > > > > > > > > > > > >> Here's the
>> > > > > > > > > > > > > > > >> > > > page:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >>
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > >
>> > > > > > > > >
>> > > > > > >
>> > > > >
>> > >
>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also
>> pretty
>> > > > > longish
>> > > > > > > and
>> > > > > > > > > I
>> > > > > > > > > > have
>> > > > > > > > > > > > > > tried
>> > > > > > > > > > > > > > > >> to keep
>> > > > > > > > > > > > > > > >> > > > a
>> > > > > > > > > > > > > > > >> > > > > > format
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> similar to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848.
>> Let me
>> > > know
>> > > > > > > what you
>> > > > > > > > > > > > think.
>> > > > > > > > > > > > > > Also,
>> > > > > > > > > > > > > > > >> do you
>> > > > > > > > > > > > > > > >> > > > > > think this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a
>> > > separate
>> > > > > > > > > discussion
>> > > > > > > > > > > > thread
>> > > > > > > > > > > > > > or
>> > > > > > > > > > > > > > > >> is this
>> > > > > > > > > > > > > > > >> > > > one
>> > > > > > > > > > > > > > > >> > > > > > fine?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul
>> 26,
>> > > 2022
>> > > > > at
>> > > > > > > 7:37
>> > > > > > > > > AM
>> > > > > > > > > > > > > Sagar <
>> > > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello
>> Guozhang,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you
>> so
>> > > much
>> > > > > for
>> > > > > > > the
>> > > > > > > > > > doc on
>> > > > > > > > > > > > > > Kafka
>> > > > > > > > > > > > > > > >> Streams.
>> > > > > > > > > > > > > > > >> > > > > > Sure, I
>> > > > > > > > > > > > > > > >> > > > > > > > > would do
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis
>> and
>> > > come
>> > > > > up
>> > > > > > > with
>> > > > > > > > > > such a
>> > > > > > > > > > > > > > > >> document.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue,
>> Jul 26,
>> > > > > 2022 at
>> > > > > > > > > 4:47
>> > > > > > > > > > AM
>> > > > > > > > > > > > > > Guozhang
>> > > > > > > > > > > > > > > >> Wang <
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello
>> Sagar,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would
>> be
>> > > great
>> > > > > if
>> > > > > > > you
>> > > > > > > > > > could
>> > > > > > > > > > > > > come
>> > > > > > > > > > > > > > > >> back with
>> > > > > > > > > > > > > > > >> > > > some
>> > > > > > > > > > > > > > > >> > > > > > > > > analysis on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > how to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> implement the
>> > > > > Connect
>> > > > > > > side
>> > > > > > > > > > > > > > integration
>> > > > > > > > > > > > > > > >> with
>> > > > > > > > > > > > > > > >> > > > the new
>> > > > > > > > > > > > > > > >> > > > > > > > > protocol;
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> so
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > far
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides
>> > > > > leveraging on
>> > > > > > > the
>> > > > > > > > > > new
>> > > > > > > > > > > > > > "protocol
>> > > > > > > > > > > > > > > >> type"
>> > > > > > > > > > > > > > > >> > > > we
>> > > > > > > > > > > > > > > >> > > > > > did not
>> > > > > > > > > > > > > > > >> > > > > > > > > yet
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> Connect
>> > > side
>> > > > > > > > > > > > implementations.
>> > > > > > > > > > > > > > For
>> > > > > > > > > > > > > > > >> Streams
>> > > > > > > > > > > > > > > >> > > > > > here's a
>> > > > > > > > > > > > > > > >> > > > > > > > > draft of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> integration
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > >
>> > > > > > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > >
>> > > > > > > > > > > > > > > >>
>> > > > > > > > > > > > > >
>> > > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > >
>> > > > > > > > >
>> > > > > > >
>> > > > >
>> > >
>> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI
>> for
>> > > your
>> > > > > > > analysis
>> > > > > > > > > > on
>> > > > > > > > > > > > > > Connect.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue,
>> Jul
>> > > 19,
>> > > > > 2022
>> > > > > > > at
>> > > > > > > > > > 10:48 PM
>> > > > > > > > > > > > > > Sagar <
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> > > sagarmeansocean@gmail.com
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi
>> David,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank
>> you
>> > > for
>> > > > > your
>> > > > > > > > > > response.
>> > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> reason I
>> > > > > > > > > > > > > > > >> > > > thought
>> > > > > > > > > > > > > > > >> > > > > > > > > connect can
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > also fit
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this
>> new
>> > > scheme
>> > > > > is
>> > > > > > > that
>> > > > > > > > > > even
>> > > > > > > > > > > > > > today the
>> > > > > > > > > > > > > > > >> > > > connect
>> > > > > > > > > > > > > > > >> > > > > > uses a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> WorkerCoordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> extending
>> > > from
>> > > > > > > > > > > > > > AbstractCoordinator to
>> > > > > > > > > > > > > > > >> empower
>> > > > > > > > > > > > > > > >> > > > > > > > > rebalances of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > tasks/connectors.
>> > > > > > > The
>> > > > > > > > > > > > > > > >> WorkerCoordinator sets
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and
>> uses the
>> > > > > > > metadata()
>> > > > > > > > > > method
>> > > > > > > > > > > > > by
>> > > > > > > > > > > > > > > >> plumbing
>> > > > > > > > > > > > > > > >> > > > into
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > > > JoinGroupRequestProtocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I
>> think the
>> > > > > changes
>> > > > > > > to
>> > > > > > > > > > support
>> > > > > > > > > > > > > > > >> connect would
>> > > > > > > > > > > > > > > >> > > > be
>> > > > > > > > > > > > > > > >> > > > > > > > > similar at a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > high
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
>> changes
>> > > in
>> > > > > > > streams
>> > > > > > > > > > mainly
>> > > > > > > > > > > > > > because
>> > > > > > > > > > > > > > > >> of the
>> > > > > > > > > > > > > > > >> > > > > > Client
>> > > > > > > > > > > > > > > >> > > > > > > > > side
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assignors
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in
>> > > both.
>> > > > > At an
>> > > > > > > > > > > > > implementation
>> > > > > > > > > > > > > > > >> level, we
>> > > > > > > > > > > > > > > >> > > > > > might
>> > > > > > > > > > > > > > > >> > > > > > > > > need to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> make
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > a lot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> changes to
>> > > get
>> > > > > onto
>> > > > > > > this
>> > > > > > > > > > new
>> > > > > > > > > > > > > > > >> assignment
>> > > > > > > > > > > > > > > >> > > > protocol
>> > > > > > > > > > > > > > > >> > > > > > like
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> enhancing
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> JoinGroup
>> > > > > > > > > > request/response and
>> > > > > > > > > > > > > > > >> SyncGroup and
>> > > > > > > > > > > > > > > >> > > > > > using
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > ConsumerGroupHeartbeat
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc
>> > > again on
>> > > > > > > similar
>> > > > > > > > > > lines
>> > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> streams (or
>> > > > > > > > > > > > > > > >> > > > > > there
>> > > > > > > > > > > > > > > >> > > > > > > > > might be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> deviations). I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would
>> try to
>> > > > > > > perform a
>> > > > > > > > > > > > detailed
>> > > > > > > > > > > > > > > >> analysis of
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > same
>> > > > > > > > > > > > > > > >> > > > > > > > > and we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > can have
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> separate
>> > > > > discussion
>> > > > > > > > > > thread for
>> > > > > > > > > > > > > > that
>> > > > > > > > > > > > > > > >> as that
>> > > > > > > > > > > > > > > >> > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > derail this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> thread. Let
>> > > me
>> > > > > know
>> > > > > > > if
>> > > > > > > > > > that
>> > > > > > > > > > > > > sounds
>> > > > > > > > > > > > > > > >> good to
>> > > > > > > > > > > > > > > >> > > > you.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On
>> Fri, Jul
>> > > 15,
>> > > > > > > 2022 at
>> > > > > > > > > > 5:47
>> > > > > > > > > > > > PM
>> > > > > > > > > > > > > > David
>> > > > > > > > > > > > > > > >> Jacot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > <djacot@confluent.io.invalid
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi
>> Sagar,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> Thanks for
>> > > > > your
>> > > > > > > > > > comments.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1)
>> Yes.
>> > > That
>> > > > > > > refers to
>> > > > > > > > > > > > > > > >> `Assignment#error`.
>> > > > > > > > > > > > > > > >> > > > > > Sure, I
>> > > > > > > > > > > > > > > >> > > > > > > > > can
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2)
>> The
>> > > idea
>> > > > > is to
>> > > > > > > > > > > > transition C
>> > > > > > > > > > > > > > from
>> > > > > > > > > > > > > > > >> his
>> > > > > > > > > > > > > > > >> > > > current
>> > > > > > > > > > > > > > > >> > > > > > > > > assignment
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > his
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> target
>> > > > > assignment
>> > > > > > > when
>> > > > > > > > > > he
>> > > > > > > > > > > > can
>> > > > > > > > > > > > > > move
>> > > > > > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > epoch 3.
>> > > > > > > > > > > > > > > >> > > > > > When
>> > > > > > > > > > > > > > > >> > > > > > > > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > happens,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> member
>> > > > > assignment
>> > > > > > > is
>> > > > > > > > > > updated
>> > > > > > > > > > > > > and
>> > > > > > > > > > > > > > > >> persisted
>> > > > > > > > > > > > > > > >> > > > > > with all
>> > > > > > > > > > > > > > > >> > > > > > > > > its
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigned
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> partitions
>> > > > > even if
>> > > > > > > > > they
>> > > > > > > > > > are
>> > > > > > > > > > > > > not
>> > > > > > > > > > > > > > all
>> > > > > > > > > > > > > > > >> revoked
>> > > > > > > > > > > > > > > >> > > > > > yet. In
>> > > > > > > > > > > > > > > >> > > > > > > > > other
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > words, the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> member
>> > > > > assignment
>> > > > > > > > > > becomes
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > target
>> > > > > > > > > > > > > > > >> > > > > > assignment.
>> > > > > > > > > > > > > > > >> > > > > > > > > This is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > basically
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > optimization
>> > > > > to
>> > > > > > > avoid
>> > > > > > > > > > having
>> > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> write all
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > changes to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > log.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> examples
>> > > are
>> > > > > > > based on
>> > > > > > > > > > the
>> > > > > > > > > > > > > > persisted
>> > > > > > > > > > > > > > > >> state
>> > > > > > > > > > > > > > > >> > > > so I
>> > > > > > > > > > > > > > > >> > > > > > > > > understand
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> confusion.
>> > > > > Let me
>> > > > > > > see
>> > > > > > > > > > if I
>> > > > > > > > > > > > can
>> > > > > > > > > > > > > > > >> improve
>> > > > > > > > > > > > > > > >> > > > this in
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > description.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3)
>> > > Regarding
>> > > > > > > Connect,
>> > > > > > > > > it
>> > > > > > > > > > > > could
>> > > > > > > > > > > > > > > >> reuse the
>> > > > > > > > > > > > > > > >> > > > > > protocol
>> > > > > > > > > > > > > > > >> > > > > > > > > with a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client side
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> assignor
>> > > if it
>> > > > > > > fits in
>> > > > > > > > > > the
>> > > > > > > > > > > > > > > >> protocol. The
>> > > > > > > > > > > > > > > >> > > > > > assignment
>> > > > > > > > > > > > > > > >> > > > > > > > > is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > topicid-partitions +
>> > > > > > > > > > > > metadata,
>> > > > > > > > > > > > > > could
>> > > > > > > > > > > > > > > >> > > > Connect
>> > > > > > > > > > > > > > > >> > > > > > fit
>> > > > > > > > > > > > > > > >> > > > > > > > > into this?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On
>> Fri,
>> > > Jul
>> > > > > 15,
>> > > > > > > 2022
>> > > > > > > > > at
>> > > > > > > > > > 1:55
>> > > > > > > > > > > > > PM
>> > > > > > > > > > > > > > > >> Sagar <
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > sagarmeansocean@gmail.com>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi
>> > > David,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> Thanks
>> > > for
>> > > > > the
>> > > > > > > KIP.
>> > > > > > > > > I
>> > > > > > > > > > just
>> > > > > > > > > > > > > had
>> > > > > > > > > > > > > > > >> minor
>> > > > > > > > > > > > > > > >> > > > > > observations:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1)
>> In
>> > > the
>> > > > > > > Assignment
>> > > > > > > > > > Error
>> > > > > > > > > > > > > > > >> section in
>> > > > > > > > > > > > > > > >> > > > Client
>> > > > > > > > > > > > > > > >> > > > > > Side
>> > > > > > > > > > > > > > > >> > > > > > > > > mode
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> process,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you
>> > > > > mentioned
>> > > > > > > => `In
>> > > > > > > > > > this
>> > > > > > > > > > > > > > case,
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > client
>> > > > > > > > > > > > > > > >> > > > > > side
>> > > > > > > > > > > > > > > >> > > > > > > > > assignor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> can
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> error
>> > > to the
>> > > > > > > group
>> > > > > > > > > > > > > > coordinator`.
>> > > > > > > > > > > > > > > >> In this
>> > > > > > > > > > > > > > > >> > > > > > case are
>> > > > > > > > > > > > > > > >> > > > > > > > > you
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > referring to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> Assignor
>> > > > > > > returning
>> > > > > > > > > an
>> > > > > > > > > > > > > > > >> AssignmentError
>> > > > > > > > > > > > > > > >> > > > that's
>> > > > > > > > > > > > > > > >> > > > > > > > > listed down
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > towards
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If
>> yes,
>> > > do
>> > > > > you
>> > > > > > > think
>> > > > > > > > > > it
>> > > > > > > > > > > > > would
>> > > > > > > > > > > > > > > >> make sense
>> > > > > > > > > > > > > > > >> > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > mention this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2)
>> In
>> > > the
>> > > > > Case
>> > > > > > > > > Studies
>> > > > > > > > > > > > > > section, I
>> > > > > > > > > > > > > > > >> have a
>> > > > > > > > > > > > > > > >> > > > > > slight
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> confusion,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> others
>> > > have
>> > > > > the
>> > > > > > > > > same.
>> > > > > > > > > > > > > Consider
>> > > > > > > > > > > > > > > >> this step:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> When B
>> > > > > > > heartbeats,
>> > > > > > > > > the
>> > > > > > > > > > > > group
>> > > > > > > > > > > > > > > >> coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > transitions him
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > epoch
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> because
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B
>> has no
>> > > > > > > partitions
>> > > > > > > > > to
>> > > > > > > > > > > > > > revoke. It
>> > > > > > > > > > > > > > > >> > > > persists
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > change and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > reply.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> -
>> > > Group
>> > > > > > > (epoch=3)
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > A
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > B
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > C
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> -
>> > > Target
>> > > > > > > > > Assignment
>> > > > > > > > > > > > > > (epoch=3)
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > A -
>> > > > > > > > > > > > > partitions=[foo-0]
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > B -
>> > > > > > > > > > > > > partitions=[foo-2]
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > C -
>> > > > > > > > > > > > > partitions=[foo-1]
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> -
>> > > Member
>> > > > > > > > > Assignment
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > A -
>> > > > > > > epoch=2,
>> > > > > > > > > > > > > > > >> partitions=[foo-0,
>> > > > > > > > > > > > > > > >> > > > > > foo-1]
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > B -
>> > > > > > > epoch=3,
>> > > > > > > > > > > > > > > >> partitions=[foo-2]
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>>    -
>> > > C -
>> > > > > > > epoch=3,
>> > > > > > > > > > > > > > > >> partitions=[foo-1]
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> When C
>> > > > > > > heartbeats,
>> > > > > > > > > it
>> > > > > > > > > > > > > > transitions
>> > > > > > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > epoch 3
>> > > > > > > > > > > > > > > >> > > > > > but
>> > > > > > > > > > > > > > > >> > > > > > > > > cannot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> get
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > Here,it's
>> > > > > > > mentioned
>> > > > > > > > > > that
>> > > > > > > > > > > > > > member C
>> > > > > > > > > > > > > > > >> can't
>> > > > > > > > > > > > > > > >> > > > get
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > foo-1
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > partition
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> based
>> > > on the
>> > > > > > > > > > description
>> > > > > > > > > > > > > > above,
>> > > > > > > > > > > > > > > >> it seems
>> > > > > > > > > > > > > > > >> > > > it
>> > > > > > > > > > > > > > > >> > > > > > > > > already has
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Do you
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> would be
>> > > > > better
>> > > > > > > to
>> > > > > > > > > > remove
>> > > > > > > > > > > > it
>> > > > > > > > > > > > > > and
>> > > > > > > > > > > > > > > >> > > > populate it
>> > > > > > > > > > > > > > > >> > > > > > only
>> > > > > > > > > > > > > > > >> > > > > > > > > when it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > actually
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I
>> see
>> > > this
>> > > > > in a
>> > > > > > > lot
>> > > > > > > > > of
>> > > > > > > > > > > > other
>> > > > > > > > > > > > > > > >> places, so
>> > > > > > > > > > > > > > > >> > > > have
>> > > > > > > > > > > > > > > >> > > > > > I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> understood it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> incorrectly
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > Regarding
>> > > > > > > connect ,
>> > > > > > > > > it
>> > > > > > > > > > > > might
>> > > > > > > > > > > > > > be
>> > > > > > > > > > > > > > > >> out of
>> > > > > > > > > > > > > > > >> > > > scope
>> > > > > > > > > > > > > > > >> > > > > > of
>> > > > > > > > > > > > > > > >> > > > > > > > > this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> what I
>> > > > > > > understood it
>> > > > > > > > > > would
>> > > > > > > > > > > > > > > >> probably be
>> > > > > > > > > > > > > > > >> > > > > > running in
>> > > > > > > > > > > > > > > >> > > > > > > > > client
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> even on
>> > > the
>> > > > > new
>> > > > > > > > > > rebalance
>> > > > > > > > > > > > > > > >> protocol as it
>> > > > > > > > > > > > > > > >> > > > has
>> > > > > > > > > > > > > > > >> > > > > > its
>> > > > > > > > > > > > > > > >> > > > > > > > > own
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Custom
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > Assignors(Eager
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
>> > > > > > > > > > > > IncrementalCooperative).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> Thanks!
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> Sagar.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On
>> Fri,
>> > > Jul
>> > > > > 15,
>> > > > > > > 2022
>> > > > > > > > > > at
>> > > > > > > > > > > > 5:00
>> > > > > > > > > > > > > > PM
>> > > > > > > > > > > > > > > >> David
>> > > > > > > > > > > > > > > >> > > > Jacot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > <djacot@confluent.io.invalid
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> Thanks
>> > > > > > > Hector! Our
>> > > > > > > > > > goal
>> > > > > > > > > > > > is
>> > > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> move
>> > > > > > > > > > > > > > > >> > > > forward
>> > > > > > > > > > > > > > > >> > > > > > with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > instead of
>> > > > > > > relying
>> > > > > > > > > > on
>> > > > > > > > > > > > one
>> > > > > > > > > > > > > > > >> generic API.
>> > > > > > > > > > > > > > > >> > > > For
>> > > > > > > > > > > > > > > >> > > > > > > > > Connect, we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > can apply
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> exact
>> > > same
>> > > > > > > pattern
>> > > > > > > > > > and
>> > > > > > > > > > > > > > > >> reuse/share the
>> > > > > > > > > > > > > > > >> > > > core
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> server
>> > > > > side.
>> > > > > > > For
>> > > > > > > > > the
>> > > > > > > > > > > > > schema
>> > > > > > > > > > > > > > > >> registry, I
>> > > > > > > > > > > > > > > >> > > > > > think
>> > > > > > > > > > > > > > > >> > > > > > > > > that we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > having a
>> > > > > > > tailored
>> > > > > > > > > > API to
>> > > > > > > > > > > > > do
>> > > > > > > > > > > > > > > >> simple
>> > > > > > > > > > > > > > > >> > > > > > > > > membership/leader
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > election.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> Best,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> David
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> On
>> > > Fri,
>> > > > > Jul
>> > > > > > > 15,
>> > > > > > > > > > 2022 at
>> > > > > > > > > > > > > > 10:22
>> > > > > > > > > > > > > > > >> AM Ismael
>> > > > > > > > > > > > > > > >> > > > > > Juma <
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > Three
>> > > > > quick
>> > > > > > > > > > comments:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > 1.
>> > > > > Regarding
>> > > > > > > > > > > > > > > >> java.util.regex.Pattern
>> > > > > > > > > > > > > > > >> > > > vs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> > > > > > > com.google.re2j.Pattern,
>> > > > > > > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > document the
>> > > > > > > > > > > > differences
>> > > > > > > > > > > > > > in
>> > > > > > > > > > > > > > > >> more
>> > > > > > > > > > > > > > > >> > > > detail
>> > > > > > > > > > > > > > > >> > > > > > before
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> deciding
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > another.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > That
>> > > > > said,
>> > > > > > > if
>> > > > > > > > > > people
>> > > > > > > > > > > > > pass
>> > > > > > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> they
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > expect
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > semantics
>> > > > > > > to be
>> > > > > > > > > > > > honored.
>> > > > > > > > > > > > > > If
>> > > > > > > > > > > > > > > >> we are
>> > > > > > > > > > > > > > > >> > > > doing
>> > > > > > > > > > > > > > > >> > > > > > > > > something
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > different,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > should
>> > > > > > > consider
>> > > > > > > > > > adding
>> > > > > > > > > > > > > an
>> > > > > > > > > > > > > > > >> overload
>> > > > > > > > > > > > > > > >> > > > with
>> > > > > > > > > > > > > > > >> > > > > > our own
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Pattern
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > class
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > think
>> > > > > we'd
>> > > > > > > want
>> > > > > > > > > to
>> > > > > > > > > > > > > expose
>> > > > > > > > > > > > > > > >> re2j's at
>> > > > > > > > > > > > > > > >> > > > this
>> > > > > > > > > > > > > > > >> > > > > > > > > point).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > 2.
>> > > > > Regarding
>> > > > > > > > > topic
>> > > > > > > > > > > > ids,
>> > > > > > > > > > > > > > any
>> > > > > > > > > > > > > > > >> major new
>> > > > > > > > > > > > > > > >> > > > > > protocol
>> > > > > > > > > > > > > > > >> > > > > > > > > should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > it
>> > > and
>> > > > > > > should
>> > > > > > > > > > handle
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > topic
>> > > > > > > > > > > > > > > >> > > > > > recreation case
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > part we
>> > > > > > > need to
>> > > > > > > > > > > > handle.
>> > > > > > > > > > > > > I
>> > > > > > > > > > > > > > > >> agree with
>> > > > > > > > > > > > > > > >> > > > > > David
>> > > > > > > > > > > > > > > >> > > > > > > > > that we'd
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > want to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> ids
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > to
>> > > the
>> > > > > > > relevant
>> > > > > > > > > > > > > protocols
>> > > > > > > > > > > > > > > >> that don't
>> > > > > > > > > > > > > > > >> > > > > > have it
>> > > > > > > > > > > > > > > >> > > > > > > > > yet and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > that we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> probably
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > focus
>> > > > > on the
>> > > > > > > > > > internals
>> > > > > > > > > > > > > > versus
>> > > > > > > > > > > > > > > >> adding
>> > > > > > > > > > > > > > > >> > > > new
>> > > > > > > > > > > > > > > >> > > > > > APIs
>> > > > > > > > > > > > > > > >> > > > > > > > > to the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > (unless
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > we
>> > > find
>> > > > > that
>> > > > > > > > > > adding
>> > > > > > > > > > > > new
>> > > > > > > > > > > > > > APIs
>> > > > > > > > > > > > > > > >> is
>> > > > > > > > > > > > > > > >> > > > required
>> > > > > > > > > > > > > > > >> > > > > > for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> reasonable
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> semantics).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > 3.
>> > > I am
>> > > > > > > still
>> > > > > > > > > not
>> > > > > > > > > > sure
>> > > > > > > > > > > > > > about
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > storing
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > powerful for
>> > > > > > > > > > configs
>> > > > > > > > > > > > to
>> > > > > > > > > > > > > be
>> > > > > > > > > > > > > > > >> > > > centralized
>> > > > > > > > > > > > > > > >> > > > > > in the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> metadata
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > log for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> various
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > reasons
>> > > > > > > > > > (auditability,
>> > > > > > > > > > > > > > > >> visibility,
>> > > > > > > > > > > > > > > >> > > > > > consistency,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> etc.).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> Similarly, I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > sure
>> > > > > about
>> > > > > > > > > > > > automatically
>> > > > > > > > > > > > > > > >> deleting
>> > > > > > > > > > > > > > > >> > > > > > configs in a
>> > > > > > > > > > > > > > > >> > > > > > > > > way
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > they
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > recovered. A
>> > > > > > > > > good
>> > > > > > > > > > > > > property
>> > > > > > > > > > > > > > > >> for modern
>> > > > > > > > > > > > > > > >> > > > > > systems
>> > > > > > > > > > > > > > > >> > > > > > > > > is to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > minimize
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> number
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > unrecoverable
>> > > > > > > > > data
>> > > > > > > > > > > > loss
>> > > > > > > > > > > > > > > >> scenarios.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > Ismael
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > On
>> > > Wed,
>> > > > > Jul
>> > > > > > > 13,
>> > > > > > > > > > 2022
>> > > > > > > > > > > > at
>> > > > > > > > > > > > > > 3:47
>> > > > > > > > > > > > > > > >> PM David
>> > > > > > > > > > > > > > > >> > > > > > Jacot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > <djacot@confluent.io.invalid
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > Thanks
>> > > > > > > > > > Guozhang. My
>> > > > > > > > > > > > > > answers
>> > > > > > > > > > > > > > > >> are
>> > > > > > > > > > > > > > > >> > > > below:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > 1)
>> > > > > the
>> > > > > > > > > > migration
>> > > > > > > > > > > > > path,
>> > > > > > > > > > > > > > > >> especially
>> > > > > > > > > > > > > > > >> > > > > > the last
>> > > > > > > > > > > > > > > >> > > > > > > > > step
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > flag
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > to
>> > > > > > > enable
>> > > > > > > > > the
>> > > > > > > > > > new
>> > > > > > > > > > > > > > > >> protocol, in
>> > > > > > > > > > > > > > > >> > > > which
>> > > > > > > > > > > > > > > >> > > > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> both
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > protocols /
>> > > > > > > > > > rpcs
>> > > > > > > > > > > > and
>> > > > > > > > > > > > > > old
>> > > > > > > > > > > > > > > >> > > > protocols /
>> > > > > > > > > > > > > > > >> > > > > > rpcs
>> > > > > > > > > > > > > > > >> > > > > > > > > are
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> used
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> same
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > group.
>> > > > > > > How
>> > > > > > > > > the
>> > > > > > > > > > > > > > > >> coordinator could
>> > > > > > > > > > > > > > > >> > > > > > "mimic"
>> > > > > > > > > > > > > > > >> > > > > > > > > the old
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> using
>> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > new
>> > > > > > > protocol
>> > > > > > > > > > is
>> > > > > > > > > > > > > > something
>> > > > > > > > > > > > > > > >> we
>> > > > > > > > > > > > > > > >> > > > need to
>> > > > > > > > > > > > > > > >> > > > > > > > > present
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > Noted. I
>> > > > > > > just
>> > > > > > > > > > > > > published
>> > > > > > > > > > > > > > a
>> > > > > > > > > > > > > > > >> new
>> > > > > > > > > > > > > > > >> > > > version
>> > > > > > > > > > > > > > > >> > > > > > of KIP
>> > > > > > > > > > > > > > > >> > > > > > > > > which
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > includes
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > details
>> > > > > > > about
>> > > > > > > > > > this.
>> > > > > > > > > > > > > See
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > "Supporting
>> > > > > > > > > > > > > > > >> > > > > > > > > Online
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> Upgrade"
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > and
>> > > > > the
>> > > > > > > > > > > > > "Compatibility,
>> > > > > > > > > > > > > > > >> > > > Deprecation,
>> > > > > > > > > > > > > > > >> > > > > > and
>> > > > > > > > > > > > > > > >> > > > > > > > > Migration
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > I
>> > > > > have to
>> > > > > > > > > think
>> > > > > > > > > > > > > through
>> > > > > > > > > > > > > > a
>> > > > > > > > > > > > > > > >> few cases
>> > > > > > > > > > > > > > > >> > > > > > now but
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > mechanism
>> > > > > > > > > > should be
>> > > > > > > > > > > > > > > >> understandable.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > 2)
>> > > > > the
>> > > > > > > usage
>> > > > > > > > > > of
>> > > > > > > > > > > > > topic
>> > > > > > > > > > > > > > > >> ids. So
>> > > > > > > > > > > > > > > >> > > > far as
>> > > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> used
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > as
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > part of
>> > > > > > > RPCs
>> > > > > > > > > > and
>> > > > > > > > > > > > > admin
>> > > > > > > > > > > > > > > >> client,
>> > > > > > > > > > > > > > > >> > > > but
>> > > > > > > > > > > > > > > >> > > > > > they
>> > > > > > > > > > > > > > > >> > > > > > > > > are not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> public
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > APIs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > to
>> > > > > > > consumers
>> > > > > > > > > > yet.
>> > > > > > > > > > > > I
>> > > > > > > > > > > > > > think
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > question is,
>> > > > > > > > > > > > > > > >> > > > > > > > > first
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > client
>> > > > > > > to be
>> > > > > > > > > > > > > > maintaining
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > names
>> > > > > > > > > > > > > > > >> > > > > > -> ids
>> > > > > > > > > > > > > > > >> > > > > > > > > mapping
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > leverage
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > on
>> > > > > all
>> > > > > > > the
>> > > > > > > > > > > > augmented
>> > > > > > > > > > > > > > > >> existing
>> > > > > > > > > > > > > > > >> > > > RPCs
>> > > > > > > > > > > > > > > >> > > > > > and the
>> > > > > > > > > > > > > > > >> > > > > > > > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> ids;
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > secondly,
>> > > > > > > > > > should
>> > > > > > > > > > > > we
>> > > > > > > > > > > > > > ever
>> > > > > > > > > > > > > > > >> consider
>> > > > > > > > > > > > > > > >> > > > > > exposing
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > public
>> > > > > > > APIs
>> > > > > > > > > as
>> > > > > > > > > > > > well
>> > > > > > > > > > > > > > (both
>> > > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> well
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > rebalance
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > listener for
>> > > > > > > > > > cases
>> > > > > > > > > > > > > > like
>> > > > > > > > > > > > > > > >> topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > deletion-and-recreation).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > a)
>> > > > > > > Assuming
>> > > > > > > > > > that we
>> > > > > > > > > > > > > > would
>> > > > > > > > > > > > > > > >> include
>> > > > > > > > > > > > > > > >> > > > > > converting
>> > > > > > > > > > > > > > > >> > > > > > > > > all
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > offsets
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> related
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > RPCs
>> > > > > to
>> > > > > > > using
>> > > > > > > > > > topic
>> > > > > > > > > > > > > ids
>> > > > > > > > > > > > > > in
>> > > > > > > > > > > > > > > >> this
>> > > > > > > > > > > > > > > >> > > > KIP,
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > would be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > fully
>> > > > > > > operate
>> > > > > > > > > > with
>> > > > > > > > > > > > > topic
>> > > > > > > > > > > > > > > >> ids. That
>> > > > > > > > > > > > > > > >> > > > > > being
>> > > > > > > > > > > > > > > >> > > > > > > > > said, it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > still has
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> provide
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > the
>> > > > > topics
>> > > > > > > > > > names in
>> > > > > > > > > > > > > > various
>> > > > > > > > > > > > > > > >> APIs so
>> > > > > > > > > > > > > > > >> > > > > > having a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> mapping
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > seems
>> > > > > > > > > > inevitable to
>> > > > > > > > > > > > > me.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > b) I
>> > > > > don't
>> > > > > > > > > have
>> > > > > > > > > > a
>> > > > > > > > > > > > > strong
>> > > > > > > > > > > > > > > >> opinion on
>> > > > > > > > > > > > > > > >> > > > > > this.
>> > > > > > > > > > > > > > > >> > > > > > > > > Here I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > beyond the
>> > > > > > > > > > scope of
>> > > > > > > > > > > > > this
>> > > > > > > > > > > > > > > >> KIP. I
>> > > > > > > > > > > > > > > >> > > > would
>> > > > > > > > > > > > > > > >> > > > > > rather
>> > > > > > > > > > > > > > > >> > > > > > > > > focus
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> internals
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > here
>> > > > > and
>> > > > > > > we
>> > > > > > > > > can
>> > > > > > > > > > > > > consider
>> > > > > > > > > > > > > > > >> this
>> > > > > > > > > > > > > > > >> > > > > > separately if
>> > > > > > > > > > > > > > > >> > > > > > > > > we see
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > value in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > Coming
>> > > > > > > back to
>> > > > > > > > > > > > > Ismael's
>> > > > > > > > > > > > > > > >> point about
>> > > > > > > > > > > > > > > >> > > > > > using
>> > > > > > > > > > > > > > > >> > > > > > > > > topic ids
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
>> > > > > > > > > > > > > > > >> > > > think
>> > > > > > > > > > > > > > > >> > > > > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > there
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> advantage
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > favour of
>> > > > > > > it.
>> > > > > > > > > > The
>> > > > > > > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> will
>> > > > > > > > > > > > > > > >> > > > have
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> opportunity
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > the
>> > > > > topics
>> > > > > > > > > > exists
>> > > > > > > > > > > > > before
>> > > > > > > > > > > > > > > >> passing
>> > > > > > > > > > > > > > > >> > > > them
>> > > > > > > > > > > > > > > >> > > > > > into
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > protocol.
>> > > > > > > > > > Obviously,
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > coordinator
>> > > > > > > > > > > > > > > >> > > > > > will
>> > > > > > > > > > > > > > > >> > > > > > > > > also
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> notice
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it but
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > not
>> > > > > really
>> > > > > > > > > have
>> > > > > > > > > > a
>> > > > > > > > > > > > way
>> > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> reject an
>> > > > > > > > > > > > > > > >> > > > > > invalid
>> > > > > > > > > > > > > > > >> > > > > > > > > topic in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > I'm
>> > > > > > > agreeing
>> > > > > > > > > > with
>> > > > > > > > > > > > > > David
>> > > > > > > > > > > > > > > >> on all
>> > > > > > > > > > > > > > > >> > > > other
>> > > > > > > > > > > > > > > >> > > > > > minor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > `subscribe(Pattern)`
>> > > > > > > > > > > > > > > >> question:
>> > > > > > > > > > > > > > > >> > > > > > personally
>> > > > > > > > > > > > > > > >> > > > > > > > > I think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> necessary
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > deprecate
>> > > > > > > > > the
>> > > > > > > > > > > > > > subscribe
>> > > > > > > > > > > > > > > >> API with
>> > > > > > > > > > > > > > > >> > > > > > Pattern,
>> > > > > > > > > > > > > > > >> > > > > > > > > but
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > Pattern
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > while
>> > > > > > > just
>> > > > > > > > > > > > > documenting
>> > > > > > > > > > > > > > > >> that our
>> > > > > > > > > > > > > > > >> > > > > > > > > subscription may
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected
>> by
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > server.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > Since
>> > > > > > > the
>> > > > > > > > > > > > > incompatible
>> > > > > > > > > > > > > > > >> case is a
>> > > > > > > > > > > > > > > >> > > > > > very rare
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > overloaded
>> > > > > > > > > > > > `String`
>> > > > > > > > > > > > > > based
>> > > > > > > > > > > > > > > >> > > > > > subscription may
>> > > > > > > > > > > > > > > >> > > > > > > > > be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> more
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> vulnerable
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > various
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > invalid
>> > > > > > > > > > regexes.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > That
>> > > > > could
>> > > > > > > > > > work. I
>> > > > > > > > > > > > > have
>> > > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> look at
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > differences
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > between
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > engines to
>> > > > > > > > > > better
>> > > > > > > > > > > > > > > >> understand the
>> > > > > > > > > > > > > > > >> > > > > > potential
>> > > > > > > > > > > > > > > >> > > > > > > > > issues.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> My
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > understanding
>> > > > > is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > that
>> > > > > would
>> > > > > > > > > work
>> > > > > > > > > > for
>> > > > > > > > > > > > > all
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> basic
>> > > > > > > > > > > > > > > >> > > > > > regular
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > differences
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > between
>> > > > > > > the
>> > > > > > > > > two
>> > > > > > > > > > are
>> > > > > > > > > > > > > > mainly
>> > > > > > > > > > > > > > > >> about
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > various
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> character
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes.
>> I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > wonder
>> > > > > > > what
>> > > > > > > > > > other
>> > > > > > > > > > > > > people
>> > > > > > > > > > > > > > > >> think
>> > > > > > > > > > > > > > > >> > > > about
>> > > > > > > > > > > > > > > >> > > > > > this.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > Best,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > David
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > On
>> > > > > Tue,
>> > > > > > > Jul
>> > > > > > > > > 12,
>> > > > > > > > > > 2022
>> > > > > > > > > > > > > at
>> > > > > > > > > > > > > > > >> 11:28 PM
>> > > > > > > > > > > > > > > >> > > > > > Guozhang
>> > > > > > > > > > > > > > > >> > > > > > > > > Wang <
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> > > > > wangguoz@gmail.com
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > Thanks
>> > > > > > > > > David!
>> > > > > > > > > > I
>> > > > > > > > > > > > > think
>> > > > > > > > > > > > > > on
>> > > > > > > > > > > > > > > >> the high
>> > > > > > > > > > > > > > > >> > > > > > level
>> > > > > > > > > > > > > > > >> > > > > > > > > there are
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > two meta
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> points we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> need
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > to
>> > > > > > > > > concretize
>> > > > > > > > > > a
>> > > > > > > > > > > > bit
>> > > > > > > > > > > > > > more:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > 1)
>> > > > > the
>> > > > > > > > > > migration
>> > > > > > > > > > > > > path,
>> > > > > > > > > > > > > > > >> especially
>> > > > > > > > > > > > > > > >> > > > > > the last
>> > > > > > > > > > > > > > > >> > > > > > > > > step
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > flag
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > to
>> > > > > > > enable
>> > > > > > > > > the
>> > > > > > > > > > new
>> > > > > > > > > > > > > > > >> protocol, in
>> > > > > > > > > > > > > > > >> > > > which
>> > > > > > > > > > > > > > > >> > > > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> both
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > protocols /
>> > > > > > > > > > rpcs
>> > > > > > > > > > > > and
>> > > > > > > > > > > > > > old
>> > > > > > > > > > > > > > > >> > > > protocols /
>> > > > > > > > > > > > > > > >> > > > > > rpcs
>> > > > > > > > > > > > > > > >> > > > > > > > > are
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> used
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> same
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > group.
>> > > > > > > How
>> > > > > > > > > the
>> > > > > > > > > > > > > > > >> coordinator could
>> > > > > > > > > > > > > > > >> > > > > > "mimic"
>> > > > > > > > > > > > > > > >> > > > > > > > > the old
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> using
>> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > new
>> > > > > > > protocol
>> > > > > > > > > > is
>> > > > > > > > > > > > > > something
>> > > > > > > > > > > > > > > >> we
>> > > > > > > > > > > > > > > >> > > > need to
>> > > > > > > > > > > > > > > >> > > > > > > > > present
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > 2)
>> > > > > the
>> > > > > > > usage
>> > > > > > > > > > of
>> > > > > > > > > > > > > topic
>> > > > > > > > > > > > > > > >> ids. So
>> > > > > > > > > > > > > > > >> > > > far as
>> > > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> used
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > as
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > part of
>> > > > > > > RPCs
>> > > > > > > > > > and
>> > > > > > > > > > > > > admin
>> > > > > > > > > > > > > > > >> client,
>> > > > > > > > > > > > > > > >> > > > but
>> > > > > > > > > > > > > > > >> > > > > > they
>> > > > > > > > > > > > > > > >> > > > > > > > > are not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> public
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > APIs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > to
>> > > > > > > consumers
>> > > > > > > > > > yet.
>> > > > > > > > > > > > I
>> > > > > > > > > > > > > > think
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > question is,
>> > > > > > > > > > > > > > > >> > > > > > > > > first
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > client
>> > > > > > > to be
>> > > > > > > > > > > > > > maintaining
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > names
>> > > > > > > > > > > > > > > >> > > > > > -> ids
>> > > > > > > > > > > > > > > >> > > > > > > > > mapping
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > leverage
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > on
>> > > > > all
>> > > > > > > the
>> > > > > > > > > > > > augmented
>> > > > > > > > > > > > > > > >> existing
>> > > > > > > > > > > > > > > >> > > > RPCs
>> > > > > > > > > > > > > > > >> > > > > > and the
>> > > > > > > > > > > > > > > >> > > > > > > > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> ids;
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > secondly,
>> > > > > > > > > > should
>> > > > > > > > > > > > we
>> > > > > > > > > > > > > > ever
>> > > > > > > > > > > > > > > >> consider
>> > > > > > > > > > > > > > > >> > > > > > exposing
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > public
>> > > > > > > APIs
>> > > > > > > > > as
>> > > > > > > > > > > > well
>> > > > > > > > > > > > > > (both
>> > > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> well
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > rebalance
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > listener for
>> > > > > > > > > > cases
>> > > > > > > > > > > > > > like
>> > > > > > > > > > > > > > > >> topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > deletion-and-recreation).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > I'm
>> > > > > > > agreeing
>> > > > > > > > > > with
>> > > > > > > > > > > > > > David
>> > > > > > > > > > > > > > > >> on all
>> > > > > > > > > > > > > > > >> > > > other
>> > > > > > > > > > > > > > > >> > > > > > minor
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > `subscribe(Pattern)`
>> > > > > > > > > > > > > > > >> question:
>> > > > > > > > > > > > > > > >> > > > > > personally
>> > > > > > > > > > > > > > > >> > > > > > > > > I think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> necessary
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > deprecate
>> > > > > > > > > the
>> > > > > > > > > > > > > > subscribe
>> > > > > > > > > > > > > > > >> API with
>> > > > > > > > > > > > > > > >> > > > > > Pattern,
>> > > > > > > > > > > > > > > >> > > > > > > > > but
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > Pattern
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > while
>> > > > > > > just
>> > > > > > > > > > > > > documenting
>> > > > > > > > > > > > > > > >> that our
>> > > > > > > > > > > > > > > >> > > > > > > > > subscription may
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected
>> by
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > server.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > Since
>> > > > > > > the
>> > > > > > > > > > > > > incompatible
>> > > > > > > > > > > > > > > >> case is a
>> > > > > > > > > > > > > > > >> > > > > > very rare
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > overloaded
>> > > > > > > > > > > > `String`
>> > > > > > > > > > > > > > based
>> > > > > > > > > > > > > > > >> > > > > > subscription may
>> > > > > > > > > > > > > > > >> > > > > > > > > be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> more
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> vulnerable
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > various
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > invalid
>> > > > > > > > > > regexes.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > Guozhang
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > On
>> > > > > Tue,
>> > > > > > > Jul
>> > > > > > > > > > 12,
>> > > > > > > > > > > > 2022
>> > > > > > > > > > > > > > at
>> > > > > > > > > > > > > > > >> 5:23 AM
>> > > > > > > > > > > > > > > >> > > > > > David Jacot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > > > > > > > > > > > <djacot@confluent.io.invalid
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > Hi
>> > > > > > > Ismael,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > Thanks for
>> > > > > > > > > > your
>> > > > > > > > > > > > > > > >> feedback. Let
>> > > > > > > > > > > > > > > >> > > > me
>> > > > > > > > > > > > > > > >> > > > > > answer
>> > > > > > > > > > > > > > > >> > > > > > > > > your
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > questions
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 1. I
>> > > > > > > > > think
>> > > > > > > > > > > > it's
>> > > > > > > > > > > > > > > >> premature to
>> > > > > > > > > > > > > > > >> > > > > > talk about
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> target
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > deprecation
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > removal
>> > > > > > > > > > of the
>> > > > > > > > > > > > > > > >> existing group
>> > > > > > > > > > > > > > > >> > > > > > protocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Unlike
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > affects a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > core
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > client
>> > > > > > > > > > > > protocol
>> > > > > > > > > > > > > > and
>> > > > > > > > > > > > > > > >> hence
>> > > > > > > > > > > > > > > >> > > > > > > > > deprecation/removal
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > dependent
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > how
>> > > > > > > > > > quickly
>> > > > > > > > > > > > > > > >> applications
>> > > > > > > > > > > > > > > >> > > > migrate
>> > > > > > > > > > > > > > > >> > > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > the new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > That
>> > > > > > > makes
>> > > > > > > > > > > > sense.
>> > > > > > > > > > > > > I
>> > > > > > > > > > > > > > will
>> > > > > > > > > > > > > > > >> > > > remove it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 2.
>> > > > > > > The
>> > > > > > > > > KIP
>> > > > > > > > > > > > says
>> > > > > > > > > > > > > we
>> > > > > > > > > > > > > > > >> intend to
>> > > > > > > > > > > > > > > >> > > > > > release
>> > > > > > > > > > > > > > > >> > > > > > > > > this in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> wasn't
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> made
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > clear
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > why. If
>> > > > > > > > > we
>> > > > > > > > > > > > added
>> > > > > > > > > > > > > > that
>> > > > > > > > > > > > > > > >> as a
>> > > > > > > > > > > > > > > >> > > > way to
>> > > > > > > > > > > > > > > >> > > > > > > > > estimate
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > we'd
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> deprecate
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > remove
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > the
>> > > > > > > > > group
>> > > > > > > > > > > > > > protocol, I
>> > > > > > > > > > > > > > > >> also
>> > > > > > > > > > > > > > > >> > > > > > suggest
>> > > > > > > > > > > > > > > >> > > > > > > > > removing
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this part.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Let me
>> > > > > > > > > > explain
>> > > > > > > > > > > > my
>> > > > > > > > > > > > > > > >> reasoning. As
>> > > > > > > > > > > > > > > >> > > > > > > > > explained, I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> plan
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > coordinator
>> > > > > > > > > > in
>> > > > > > > > > > > > > Java
>> > > > > > > > > > > > > > > >> while we
>> > > > > > > > > > > > > > > >> > > > > > implement
>> > > > > > > > > > > > > > > >> > > > > > > > > the new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > that
>> > > > > > > the
>> > > > > > > > > > > > internals
>> > > > > > > > > > > > > > will
>> > > > > > > > > > > > > > > >> be
>> > > > > > > > > > > > > > > >> > > > slightly
>> > > > > > > > > > > > > > > >> > > > > > > > > different
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > model).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > Therefore, I
>> > > > > > > > > > > > > wanted
>> > > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> tighten
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > switch from
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the old
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > coordinator
>> > > > > > > > > > to
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > new
>> > > > > > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > to a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > major
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > alternative
>> > > > > > > > > > > > would
>> > > > > > > > > > > > > > be to
>> > > > > > > > > > > > > > > >> use a
>> > > > > > > > > > > > > > > >> > > > flag
>> > > > > > > > > > > > > > > >> > > > > > to do
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > relying
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > on the
>> > > > > > > > > > software
>> > > > > > > > > > > > > > upgrade.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 3.
>> > > > > > > We
>> > > > > > > > > > need to
>> > > > > > > > > > > > > > flesh
>> > > > > > > > > > > > > > > >> out the
>> > > > > > > > > > > > > > > >> > > > > > details of
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > migration
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> sounds
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > like
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > we're
>> > > > > > > > > > saying
>> > > > > > > > > > > > we
>> > > > > > > > > > > > > > will
>> > > > > > > > > > > > > > > >> support
>> > > > > > > > > > > > > > > >> > > > > > online
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> migrations.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> correct?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> We
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > explain
>> > > > > > > > > > this
>> > > > > > > > > > > > in
>> > > > > > > > > > > > > > > >> detail. It
>> > > > > > > > > > > > > > > >> > > > could
>> > > > > > > > > > > > > > > >> > > > > > also
>> > > > > > > > > > > > > > > >> > > > > > > > > be done
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> if
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > it's
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > easier.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Yes,
>> > > > > > > we
>> > > > > > > > > will
>> > > > > > > > > > > > > support
>> > > > > > > > > > > > > > > >> online
>> > > > > > > > > > > > > > > >> > > > > > migrations
>> > > > > > > > > > > > > > > >> > > > > > > > > for the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > group.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > a
>> > > > > > > group
>> > > > > > > > > > using
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > old
>> > > > > > > > > > > > > > > >> protocol
>> > > > > > > > > > > > > > > >> > > > > > will be
>> > > > > > > > > > > > > > > >> > > > > > > > > able to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > protocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Let me
>> > > > > > > > > > briefly
>> > > > > > > > > > > > > > explain
>> > > > > > > > > > > > > > > >> how that
>> > > > > > > > > > > > > > > >> > > > > > will work
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> though.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > It is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> basically
>> > > a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > four
>> > > > > > > step
>> > > > > > > > > > > > process:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > 1. The
>> > > > > > > > > > cluster
>> > > > > > > > > > > > > must
>> > > > > > > > > > > > > > be
>> > > > > > > > > > > > > > > >> > > > upgraded or
>> > > > > > > > > > > > > > > >> > > > > > > > > rolled to a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > software
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> supporting
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > new
>> > > > > > > group
>> > > > > > > > > > > > > > coordinator.
>> > > > > > > > > > > > > > > >> Both the
>> > > > > > > > > > > > > > > >> > > > > > old and
>> > > > > > > > > > > > > > > >> > > > > > > > > the new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > support
>> > > > > > > > > the
>> > > > > > > > > > old
>> > > > > > > > > > > > > > > >> protocol and
>> > > > > > > > > > > > > > > >> > > > rely
>> > > > > > > > > > > > > > > >> > > > > > on the
>> > > > > > > > > > > > > > > >> > > > > > > > > same
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > persisted
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> metadata
>> > > so
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > they
>> > > > > > > can
>> > > > > > > > > > work
>> > > > > > > > > > > > > > together.
>> > > > > > > > > > > > > > > >> This
>> > > > > > > > > > > > > > > >> > > > point
>> > > > > > > > > > > > > > > >> > > > > > is an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> offline
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> migration.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> cannot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > do
>> > > > > > > this
>> > > > > > > > > one
>> > > > > > > > > > live
>> > > > > > > > > > > > > > > >> because it
>> > > > > > > > > > > > > > > >> > > > would
>> > > > > > > > > > > > > > > >> > > > > > require
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> current
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > coordinator
>> > > > > > > > > > and
>> > > > > > > > > > > > > > > >> starting up the
>> > > > > > > > > > > > > > > >> > > > > > new one
>> > > > > > > > > > > > > > > >> > > > > > > > > and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > would
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > unavailabilities.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > 2. The
>> > > > > > > > > > cluster's
>> > > > > > > > > > > > > > > >> metadata
>> > > > > > > > > > > > > > > >> > > > > > version/IBP
>> > > > > > > > > > > > > > > >> > > > > > > > > must be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> order
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > to
>> > > > > > > enable
>> > > > > > > > > > the
>> > > > > > > > > > > > new
>> > > > > > > > > > > > > > > >> protocol.
>> > > > > > > > > > > > > > > >> > > > This
>> > > > > > > > > > > > > > > >> > > > > > cannot
>> > > > > > > > > > > > > > > >> > > > > > > > > be done
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > terminated
>> > > > > > > > > > > > because
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> old
>> > > > > > > > > > > > > > > >> > > > > > coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > doesn't
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > support
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > protocol.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > 3. The
>> > > > > > > > > > consumers
>> > > > > > > > > > > > > > must be
>> > > > > > > > > > > > > > > >> > > > upgraded
>> > > > > > > > > > > > > > > >> > > > > > to a
>> > > > > > > > > > > > > > > >> > > > > > > > > version
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> online
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > migration
>> > > > > > > > > > (must
>> > > > > > > > > > > > > have
>> > > > > > > > > > > > > > > >> KIP-792).
>> > > > > > > > > > > > > > > >> > > > If
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> there.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > Nothing
>> > > > > > > > > > must be
>> > > > > > > > > > > > > > done at
>> > > > > > > > > > > > > > > >> this
>> > > > > > > > > > > > > > > >> > > > point.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > 4. The
>> > > > > > > > > > consumers
>> > > > > > > > > > > > > > must
>> > > > > > > > > > > > > > > >> be rolled
>> > > > > > > > > > > > > > > >> > > > > > with the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> The
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > consumer
>> > > > > > > > > > group
>> > > > > > > > > > > > is
>> > > > > > > > > > > > > > > >> automatically
>> > > > > > > > > > > > > > > >> > > > > > > > > converted when
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the first
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > using
>> > > > > > > the
>> > > > > > > > > > new
>> > > > > > > > > > > > > > protocol
>> > > > > > > > > > > > > > > >> joins
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > group.
>> > > > > > > > > > > > > > > >> > > > > > > > > While
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> using the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > old
>> > > > > > > > > > protocol are
>> > > > > > > > > > > > > > being
>> > > > > > > > > > > > > > > >> > > > upgraded,
>> > > > > > > > > > > > > > > >> > > > > > the old
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > new
>> > > > > > > one.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Let me
>> > > > > > > > > > clarify
>> > > > > > > > > > > > all
>> > > > > > > > > > > > > > of
>> > > > > > > > > > > > > > > >> this in
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > KIP.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 4.
>> > > > > > > I am
>> > > > > > > > > > happy
>> > > > > > > > > > > > > > that we
>> > > > > > > > > > > > > > > >> are
>> > > > > > > > > > > > > > > >> > > > > > pushing the
>> > > > > > > > > > > > > > > >> > > > > > > > > pattern
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> subscriptions
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > server,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > but
>> > > > > > > it
>> > > > > > > > > > seems
>> > > > > > > > > > > > > like
>> > > > > > > > > > > > > > > >> there
>> > > > > > > > > > > > > > > >> > > > could be
>> > > > > > > > > > > > > > > >> > > > > > some
>> > > > > > > > > > > > > > > >> > > > > > > > > tricky
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> compatibility
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > issues.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > Will
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > have a
>> > > > > > > > > > > > mechanism
>> > > > > > > > > > > > > > for
>> > > > > > > > > > > > > > > >> users to
>> > > > > > > > > > > > > > > >> > > > > > detect
>> > > > > > > > > > > > > > > >> > > > > > > > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> they
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> their
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > regex
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > before
>> > > > > > > > > > > > switching
>> > > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> the new
>> > > > > > > > > > > > > > > >> > > > > > protocol?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > I
>> > > > > > > think
>> > > > > > > > > > that I
>> > > > > > > > > > > > am
>> > > > > > > > > > > > > a
>> > > > > > > > > > > > > > bit
>> > > > > > > > > > > > > > > >> more
>> > > > > > > > > > > > > > > >> > > > > > optimistic
>> > > > > > > > > > > > > > > >> > > > > > > > > than
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> you
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> point. I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > believe
>> > > > > > > > > > that the
>> > > > > > > > > > > > > > > >> majority of
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > cases
>> > > > > > > > > > > > > > > >> > > > > > > > > are
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> simple
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > work
>> > > > > > > with
>> > > > > > > > > > the
>> > > > > > > > > > > > new
>> > > > > > > > > > > > > > > >> engine. The
>> > > > > > > > > > > > > > > >> > > > > > > > > coordinator will
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > verify
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> anyway
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > and
>> > > > > > > reject
>> > > > > > > > > > the
>> > > > > > > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> if the
>> > > > > > > > > > > > > > > >> > > > > > regex is
>> > > > > > > > > > > > > > > >> > > > > > > > > not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> valid.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > migration
>> > > > > > > > > > path,
>> > > > > > > > > > > > in
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> worst
>> > > > > > > > > > > > > > > >> > > > case,
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > first
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > joining
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > the
>> > > > > > > group
>> > > > > > > > > > will
>> > > > > > > > > > > > be
>> > > > > > > > > > > > > > > >> rejected.
>> > > > > > > > > > > > > > > >> > > > This
>> > > > > > > > > > > > > > > >> > > > > > should
>> > > > > > > > > > > > > > > >> > > > > > > > > be used
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > defence, I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > would
>> > > > > > > say.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > One
>> > > > > > > way
>> > > > > > > > > for
>> > > > > > > > > > > > > > customers to
>> > > > > > > > > > > > > > > >> > > > validate
>> > > > > > > > > > > > > > > >> > > > > > their
>> > > > > > > > > > > > > > > >> > > > > > > > > regex
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> upgrading
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> their
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > prod
>> > > > > > > would
>> > > > > > > > > > be to
>> > > > > > > > > > > > > > test
>> > > > > > > > > > > > > > > >> them with
>> > > > > > > > > > > > > > > >> > > > > > another
>> > > > > > > > > > > > > > > >> > > > > > > > > group.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> For
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > could
>> > > > > > > be
>> > > > > > > > > > done
>> > > > > > > > > > > > in a
>> > > > > > > > > > > > > > > >> pre-prod
>> > > > > > > > > > > > > > > >> > > > > > environment.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Another
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > way
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > extend the
>> > > > > > > > > > > > > > > >> consumer-group tool
>> > > > > > > > > > > > > > > >> > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > provide a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> regex
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
>> validation
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > mechanism.
>> > > > > > > > > > Would
>> > > > > > > > > > > > > > this be
>> > > > > > > > > > > > > > > >> > > > enough in
>> > > > > > > > > > > > > > > >> > > > > > your
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> opinion?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 5.
>> > > > > > > > > > Related to
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > last
>> > > > > > > > > > > > > > > >> > > > question,
>> > > > > > > > > > > > > > > >> > > > > > will
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Java
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > users
>> > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > stick
>> > > > > > > > > > with the
>> > > > > > > > > > > > > > > >> current regex
>> > > > > > > > > > > > > > > >> > > > > > engine for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> reasons?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> For
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > example,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > it
>> > > > > > > may
>> > > > > > > > > be
>> > > > > > > > > > > > handy
>> > > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> keep using
>> > > > > > > > > > > > > > > >> > > > > > client
>> > > > > > > > > > > > > > > >> > > > > > > > > based
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > regex at
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > migrations
>> > > > > > > > > > > > > simple
>> > > > > > > > > > > > > > and
>> > > > > > > > > > > > > > > >> then
>> > > > > > > > > > > > > > > >> > > > > > migrate to
>> > > > > > > > > > > > > > > >> > > > > > > > > server
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > based
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > second
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > step.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > I
>> > > > > > > > > understand
>> > > > > > > > > > > > your
>> > > > > > > > > > > > > > point
>> > > > > > > > > > > > > > > >> but I
>> > > > > > > > > > > > > > > >> > > > am
>> > > > > > > > > > > > > > > >> > > > > > > > > concerned that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> users
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > to
>> > > > > > > > > actually
>> > > > > > > > > > stay
>> > > > > > > > > > > > > in
>> > > > > > > > > > > > > > > >> this mode.
>> > > > > > > > > > > > > > > >> > > > That
>> > > > > > > > > > > > > > > >> > > > > > > > > would go
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > against our
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > simplifying
>> > > > > > > > > > the
>> > > > > > > > > > > > > > client
>> > > > > > > > > > > > > > > >> because
>> > > > > > > > > > > > > > > >> > > > we
>> > > > > > > > > > > > > > > >> > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > have to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > continue
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> monitoring
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > the
>> > > > > > > > > > metadata on
>> > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> client
>> > > > > > > > > > > > > > > >> > > > side. I
>> > > > > > > > > > > > > > > >> > > > > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > rather
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > not do
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 6.
>> > > > > > > When
>> > > > > > > > > > we say
>> > > > > > > > > > > > > > that
>> > > > > > > > > > > > > > > >> the group
>> > > > > > > > > > > > > > > >> > > > > > > > > coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> will
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > responsible
>> > > > > for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > storing
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > the
>> > > > > > > > > > > > > configurations
>> > > > > > > > > > > > > > > >> and that
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> configurations
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> deleted
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> when
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > group is
>> > > > > > > > > > > > > deleted.
>> > > > > > > > > > > > > > > >> Will a
>> > > > > > > > > > > > > > > >> > > > > > transition to
>> > > > > > > > > > > > > > > >> > > > > > > > > DEAD
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > trigger
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
>> deletion
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > > > configurations?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > That's
>> > > > > > > > > > right.
>> > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> > > > configurations
>> > > > > > > > > > > > > > > >> > > > > > will be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> deleted
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > when
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> group is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > deleted.
>> > > > > > > > > > They go
>> > > > > > > > > > > > > > > >> together.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 7.
>> > > > > > > Will
>> > > > > > > > > > the
>> > > > > > > > > > > > > > choice to
>> > > > > > > > > > > > > > > >> store
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > configs in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > make
>> > > > > it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > harder
>> > > > > > > > > to
>> > > > > > > > > > list
>> > > > > > > > > > > > > all
>> > > > > > > > > > > > > > > >> cluster
>> > > > > > > > > > > > > > > >> > > > > > configs and
>> > > > > > > > > > > > > > > >> > > > > > > > > their
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > values?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > I
>> > > > > > > don't
>> > > > > > > > > > think
>> > > > > > > > > > > > so.
>> > > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> group
>> > > > > > > > > > > > > > > >> > > > > > > > > configurations are
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > overrides
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> cluster
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > configs.
>> > > > > > > > > If
>> > > > > > > > > > you
>> > > > > > > > > > > > > > want to
>> > > > > > > > > > > > > > > >> know
>> > > > > > > > > > > > > > > >> > > > all
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > overrides
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > though,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> have
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > to
>> > > > > > > ask all
>> > > > > > > > > > the
>> > > > > > > > > > > > > group
>> > > > > > > > > > > > > > > >> > > > coordinators.
>> > > > > > > > > > > > > > > >> > > > > > You
>> > > > > > > > > > > > > > > >> > > > > > > > > cannot
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > rely on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> metadata
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> log
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > for
>> > > > > > > > > > instance.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 8.
>> > > > > > > How
>> > > > > > > > > > would
>> > > > > > > > > > > > > > someone
>> > > > > > > > > > > > > > > >> > > > configure a
>> > > > > > > > > > > > > > > >> > > > > > group
>> > > > > > > > > > > > > > > >> > > > > > > > > before
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > starting
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > consumers?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Have
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > we
>> > > > > > > > > > considered
>> > > > > > > > > > > > > > > >> allowing the
>> > > > > > > > > > > > > > > >> > > > > > explicit
>> > > > > > > > > > > > > > > >> > > > > > > > > creation
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > > > > > Alternatively,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > configs
>> > > > > > > > > > could
>> > > > > > > > > > > > be
>> > > > > > > > > > > > > > > >> decoupled
>> > > > > > > > > > > > > > > >> > > > from
>> > > > > > > > > > > > > > > >> > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Yes.
>> > > > > > > The
>> > > > > > > > > > group
>> > > > > > > > > > > > > will
>> > > > > > > > > > > > > > be
>> > > > > > > > > > > > > > > >> > > > > > automatically
>> > > > > > > > > > > > > > > >> > > > > > > > > created in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> However,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > the
>> > > > > > > > > configs
>> > > > > > > > > > will
>> > > > > > > > > > > > > be
>> > > > > > > > > > > > > > > >> lost after
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > retention
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > period of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > passes.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 9.
>> > > > > > > Will
>> > > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > Consumer.subscribe
>> > > > > > > > > > > > > > > >> > > > > > method
>> > > > > > > > > > > > > > > >> > > > > > > > > for the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> take
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > > > > > > >> `java.util.regex.Pattern` of
>> > > > > > > > > > > > > > > >> > > > do
>> > > > > > > > > > > > > > > >> > > > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > have to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > introduce
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> overload?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > That's a
>> > > > > > > > > > very
>> > > > > > > > > > > > > group
>> > > > > > > > > > > > > > > >> question. I
>> > > > > > > > > > > > > > > >> > > > > > forgot
>> > > > > > > > > > > > > > > >> > > > > > > > > about
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> `java.util.regex.Pattern` is
>> > > > > > > > > > > > > > > >> > > > not
>> > > > > > > > > > > > > > > >> > > > > > fully
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> compatible
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> engine
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > we
>> > > > > > > plan to
>> > > > > > > > > > use,
>> > > > > > > > > > > > it
>> > > > > > > > > > > > > > > >> might be
>> > > > > > > > > > > > > > > >> > > > better
>> > > > > > > > > > > > > > > >> > > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > deprecate
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > overload
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > which
>> > > > > > > > > takes
>> > > > > > > > > > a
>> > > > > > > > > > > > > > string.
>> > > > > > > > > > > > > > > >> We would
>> > > > > > > > > > > > > > > >> > > > > > rely on
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> server
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > validation.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > During the
>> > > > > > > > > > > > > > migration, I
>> > > > > > > > > > > > > > > >> think
>> > > > > > > > > > > > > > > >> > > > that
>> > > > > > > > > > > > > > > >> > > > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > could
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> still
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > try to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> toString
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > regex
>> > > > > > > and
>> > > > > > > > > > use
>> > > > > > > > > > > > it.
>> > > > > > > > > > > > > > That
>> > > > > > > > > > > > > > > >> should
>> > > > > > > > > > > > > > > >> > > > > > work, I
>> > > > > > > > > > > > > > > >> > > > > > > > > think, in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
>> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > cases.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 10.
>> > > > > > > I
>> > > > > > > > > > agree
>> > > > > > > > > > > > with
>> > > > > > > > > > > > > > > >> Justine
>> > > > > > > > > > > > > > > >> > > > that we
>> > > > > > > > > > > > > > > >> > > > > > > > > should be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > clearer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> reason
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > switch
>> > > > > > > > > to
>> > > > > > > > > > > > > > > >> > > > IBP/metadata.version
>> > > > > > > > > > > > > > > >> > > > > > from the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> mean
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > can
>> > > > > > > > > > switch the
>> > > > > > > > > > > > > > > >> default for
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > feature
>> > > > > > > > > > > > > > > >> > > > > > > > > flag
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > true
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > > > metadata.version
>> > > > > > > > > > > > > > once
>> > > > > > > > > > > > > > > >> we
>> > > > > > > > > > > > > > > >> > > > want to
>> > > > > > > > > > > > > > > >> > > > > > make
>> > > > > > > > > > > > > > > >> > > > > > > > > it the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > default.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > My
>> > > > > > > plan
>> > > > > > > > > was
>> > > > > > > > > > to
>> > > > > > > > > > > > use
>> > > > > > > > > > > > > > that
>> > > > > > > > > > > > > > > >> feature
>> > > > > > > > > > > > > > > >> > > > > > flag
>> > > > > > > > > > > > > > > >> > > > > > > > > mainly
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > during the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > development
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > phase. I
>> > > > > > > > > > should
>> > > > > > > > > > > > > not
>> > > > > > > > > > > > > > have
>> > > > > > > > > > > > > > > >> > > > mentioned
>> > > > > > > > > > > > > > > >> > > > > > it, I
>> > > > > > > > > > > > > > > >> > > > > > > > > think,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > because
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> use
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > an
>> > > > > > > > > internal
>> > > > > > > > > > > > config
>> > > > > > > > > > > > > > for
>> > > > > > > > > > > > > > > >> it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 11.
>> > > > > > > Some
>> > > > > > > > > > of
>> > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> protocol APIs
>> > > > > > > > > > > > > > > >> > > > > > don't
>> > > > > > > > > > > > > > > >> > > > > > > > > mention
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> ACLs, it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > would
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > good to
>> > > > > > > > > > add
>> > > > > > > > > > > > that
>> > > > > > > > > > > > > > for
>> > > > > > > > > > > > > > > >> > > > consistency.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Noted.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 12.
>> > > > > > > It
>> > > > > > > > > is
>> > > > > > > > > > a
>> > > > > > > > > > > > bit
>> > > > > > > > > > > > > > odd
>> > > > > > > > > > > > > > > >> that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> ConsumerGroupHeartbeat
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> Group"
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > even
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > though
>> > > > > > > > > it
>> > > > > > > > > > > > seems
>> > > > > > > > > > > > > > to do
>> > > > > > > > > > > > > > > >> more
>> > > > > > > > > > > > > > > >> > > > than
>> > > > > > > > > > > > > > > >> > > > > > > > > reading.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > I
>> > > > > > > agree.
>> > > > > > > > > > This is
>> > > > > > > > > > > > > > how the
>> > > > > > > > > > > > > > > >> > > > current
>> > > > > > > > > > > > > > > >> > > > > > > > > protocol works
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > though.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > require
>> > > > > > > > > > "Read
>> > > > > > > > > > > > > > Group" to
>> > > > > > > > > > > > > > > >> join a
>> > > > > > > > > > > > > > > >> > > > > > group. We
>> > > > > > > > > > > > > > > >> > > > > > > > > could
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > consider
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> changing
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> this
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > but I
>> > > > > > > am
>> > > > > > > > > not
>> > > > > > > > > > > > sure
>> > > > > > > > > > > > > > that
>> > > > > > > > > > > > > > > >> it is
>> > > > > > > > > > > > > > > >> > > > worth
>> > > > > > > > > > > > > > > >> > > > > > it.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 13.
>> > > > > > > How
>> > > > > > > > > is
>> > > > > > > > > > > > topic
>> > > > > > > > > > > > > > > >> recreation
>> > > > > > > > > > > > > > > >> > > > > > handled by
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > consumer
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > group
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > protocol?
>> > > > > > > > > > It
>> > > > > > > > > > > > > > would be
>> > > > > > > > > > > > > > > >> good to
>> > > > > > > > > > > > > > > >> > > > > > have a
>> > > > > > > > > > > > > > > >> > > > > > > > > section
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> on
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Noted.
>> > > > > > > > > From
>> > > > > > > > > > a
>> > > > > > > > > > > > > > protocol
>> > > > > > > > > > > > > > > >> > > > > > perspective, the
>> > > > > > > > > > > > > > > >> > > > > > > > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > topic
>> > > > > > > id
>> > > > > > > > > so
>> > > > > > > > > > it
>> > > > > > > > > > > > > will
>> > > > > > > > > > > > > > > >> treat it
>> > > > > > > > > > > > > > > >> > > > like a
>> > > > > > > > > > > > > > > >> > > > > > > > > topic with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> The
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > only
>> > > > > > > issue
>> > > > > > > > > > is
>> > > > > > > > > > > > that
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > fetch/commit
>> > > > > > > > > > > > > > > >> > > > > > > > > offsets
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> APIs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > do not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > IDs
>> > > > > > > so the
>> > > > > > > > > > > > > consumer
>> > > > > > > > > > > > > > > >> would
>> > > > > > > > > > > > > > > >> > > > reuse the
>> > > > > > > > > > > > > > > >> > > > > > > > > offsets
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> based
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > that
>> > > > > > > we
>> > > > > > > > > > should
>> > > > > > > > > > > > > > update
>> > > > > > > > > > > > > > > >> those
>> > > > > > > > > > > > > > > >> > > > APIs
>> > > > > > > > > > > > > > > >> > > > > > as well
>> > > > > > > > > > > > > > > >> > > > > > > > > in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> order
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> consistent
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> end
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > to
>> > > > > > > end.
>> > > > > > > > > That
>> > > > > > > > > > > > would
>> > > > > > > > > > > > > > > >> strengthen
>> > > > > > > > > > > > > > > >> > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > semantics of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 14.
>> > > > > > > The
>> > > > > > > > > > KIP
>> > > > > > > > > > > > > > mentions
>> > > > > > > > > > > > > > > >> we will
>> > > > > > > > > > > > > > > >> > > > > > write the
>> > > > > > > > > > > > > > > >> > > > > > > > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> Even
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > though
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > this is
>> > > > > > > > > an
>> > > > > > > > > > > > > > > >> implementation
>> > > > > > > > > > > > > > > >> > > > > > detail, do
>> > > > > > > > > > > > > > > >> > > > > > > > > we plan
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > have a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> gradle
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > module
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > for
>> > > > > > > it?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Yes.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 15.
>> > > > > > > Do
>> > > > > > > > > we
>> > > > > > > > > > > > have a
>> > > > > > > > > > > > > > > >> scalability
>> > > > > > > > > > > > > > > >> > > > > > goal when
>> > > > > > > > > > > > > > > >> > > > > > > > > it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> comes
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to how
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > members
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > new
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > group
>> > > > > > > > > > protocol
>> > > > > > > > > > > > > can
>> > > > > > > > > > > > > > > >> support?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > We
>> > > > > > > don't
>> > > > > > > > > > have
>> > > > > > > > > > > > > > numbers
>> > > > > > > > > > > > > > > >> at the
>> > > > > > > > > > > > > > > >> > > > > > moment. The
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> support
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> 1000s
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > of
>> > > > > > > members
>> > > > > > > > > > per
>> > > > > > > > > > > > > > group.
>> > > > > > > > > > > > > > > >> We will
>> > > > > > > > > > > > > > > >> > > > > > measure
>> > > > > > > > > > > > > > > >> > > > > > > > > this when
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > we have
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > implementation.
>> > > > > > > > > > > > > Note
>> > > > > > > > > > > > > > > >> that we
>> > > > > > > > > > > > > > > >> > > > might
>> > > > > > > > > > > > > > > >> > > > > > have
>> > > > > > > > > > > > > > > >> > > > > > > > > other
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > road
>> > > > > > > (e.g.
>> > > > > > > > > > > > offset
>> > > > > > > > > > > > > > > >> commits).
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > 16.
>> > > > > > > Did
>> > > > > > > > > we
>> > > > > > > > > > > > > > consider
>> > > > > > > > > > > > > > > >> having
>> > > > > > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > > > > > SubscribedTopicNames
>> > > > > > > > > > > > > > > >> in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
>> > > > > ConsumerGroupHeartbeatRequest?
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> idea
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > since we
>> > > > > > > > > > have
>> > > > > > > > > > > > to
>> > > > > > > > > > > > > > > >> resolve the
>> > > > > > > > > > > > > > > >> > > > > > regex on
>> > > > > > > > > > > > > > > >> > > > > > > > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > server, we
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> same
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > the
>> > > > > > > > > topic
>> > > > > > > > > > > > name?
>> > > > > > > > > > > > > > The
>> > > > > > > > > > > > > > > >> > > > difference
>> > > > > > > > > > > > > > > >> > > > > > is that
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> sending
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > efficient
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > whereas
>> > > > > > > > > > > > sending
>> > > > > > > > > > > > > > the
>> > > > > > > > > > > > > > > >> topic
>> > > > > > > > > > > > > > > >> > > > names
>> > > > > > > > > > > > > > > >> > > > > > is less
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > Furthermore,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > delete
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > recreation is
>> > > > > > > > > > > > > > easier
>> > > > > > > > > > > > > > > >> to
>> > > > > > > > > > > > > > > >> > > > handle
>> > > > > > > > > > > > > > > >> > > > > > if we
>> > > > > > > > > > > > > > > >> > > > > > > > > have
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ids.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > The
>> > > > > > > idea
>> > > > > > > > > > was to
>> > > > > > > > > > > > > > > >> consolidate the
>> > > > > > > > > > > > > > > >> > > > > > metadata
>> > > > > > > > > > > > > > > >> > > > > > > > > lookup
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> both
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > paths
>> > > > > > > but
>> > > > > > > > > I
>> > > > > > > > > > do
>> > > > > > > > > > > > > agree
>> > > > > > > > > > > > > > > >> with your
>> > > > > > > > > > > > > > > >> > > > > > point. As
>> > > > > > > > > > > > > > > >> > > > > > > > > a
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> second
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > ids
>> > > > > > > may be
>> > > > > > > > > > > > better
>> > > > > > > > > > > > > > here
>> > > > > > > > > > > > > > > >> for the
>> > > > > > > > > > > > > > > >> > > > > > delete and
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > recreation
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> Also, I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > suppose
>> > > > > > > > > > that we
>> > > > > > > > > > > > > may
>> > > > > > > > > > > > > > > >> allow
>> > > > > > > > > > > > > > > >> > > > users to
>> > > > > > > > > > > > > > > >> > > > > > > > > subscribe
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> with
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > future
>> > > > > > > > > > because
>> > > > > > > > > > > > > that
>> > > > > > > > > > > > > > is
>> > > > > > > > > > > > > > > >> the only
>> > > > > > > > > > > > > > > >> > > > > > way to be
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> really
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > robust
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > re-creation.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > Best,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > David
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > On
>> > > > > > > Tue,
>> > > > > > > > > Jul
>> > > > > > > > > > 12,
>> > > > > > > > > > > > > > 2022 at
>> > > > > > > > > > > > > > > >> 1:38 PM
>> > > > > > > > > > > > > > > >> > > > > > David
>> > > > > > > > > > > > > > > >> > > > > > > > > Jacot <
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > djacot@confluent.io>
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > >
>> > > wrote:
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > Hi
>> > > > > > > > > > Justine,
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > Thanks
>> > > > > > > > > for
>> > > > > > > > > > > > your
>> > > > > > > > > > > > > > > >> comments.
>> > > > > > > > > > > > > > > >> > > > Please
>> > > > > > > > > > > > > > > >> > > > > > find
>> > > > > > > > > > > > > > > >> > > > > > > > > my
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> answers
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > -
>> > > > > > > Yes,
>> > > > > > > > > > the new
>> > > > > > > > > > > > > > > >> protocol
>> > > > > > > > > > > > > > > >> > > > relies on
>> > > > > > > > > > > > > > > >> > > > > > > > > topic IDs
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
>> exception
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> of the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > > > topic
>> > > > > > > > > > names
>> > > > > > > > > > > > > based
>> > > > > > > > > > > > > > in
>> > > > > > > > > > > > > > > >> the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
>> > > > > > > > > ConsumerGroupHeartbeatRequest.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> sure
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > if
>> > > > > > > using
>> > > > > > > > > > topic
>> > > > > > > > > > > > > > names
>> > > > > > > > > > > > > > > >> is the
>> > > > > > > > > > > > > > > >> > > > > > right call
>> > > > > > > > > > > > > > > >> > > > > > > > > here.
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> I
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> about
>> > > it
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
>> > > >
>> > > > >
>> > > > > a
>> > > > > > > little
>> > > > > > > > > > more.
>> > > > > > > > > > > > > > > >> Obviously, the
>> > > > > > > > > > > > > > > >> > > > > > KIP does
>> > > > > > > > > > > > > > > >> > > > > > > > > not
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > change the
>> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
>> > > > > > >
>> > > > > >
>> > > > > >
>> > > > > > --
>> > > > > > -- Guozhang
>> > > > >
>> > > >
>> > > >
>> > > > --
>> > > > -- Guozhang
>> > >
>> >
>> >
>> > --
>> > -- Guozhang
>>
>
>
> --
> -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

3. Ah thanks, that's my mis-read then. I think keeping the current class
and the params of `onAssignment` is fine then.

On Mon, Oct 3, 2022 at 4:32 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> 3. ConsumerGroupMetadata is already defined in the consumer [1]. I
> realized that we pass it in the current assignor so I thought that we
> could reuse it. The alternative would be to add the same information
> in a class which includes the error and the member assignment as well.
>
> [1]
> https://github.com/apache/kafka/blob/99b9b3e84f4e98c3f07714e1de6a139a004cbc5b/clients/src/main/java/org/apache/kafka/clients/consumer/ConsumerGroupMetadata.java
>
> Best,
> David
>
> On Mon, Oct 3, 2022 at 6:15 AM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > 1. Yeah that makes sense. I originally concerned if there's any
> consistency
> > issues if the HB request with encoded topic IDs and the GPA/GIA RPCs with
> > encoded topic IDs use mappings from two different brokers, and one of
> them
> > is stale, but after thinking twice about it I think as long as we handle
> > e.g. `unknown topic ids` from HB responses then we should be fine.
> >
> > 3. Ack, in that case I think we also need to define
> "ConsumerGroupMetadata"
> > which is passed in `onAssignment`. We have a schema defined in the
> > persistent storage for different purposes, but not in the consumer APIs
> > yet. I think that would just include a single "Map<String, Metadata>"?
> >
> > On Fri, Sep 30, 2022 at 5:51 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hey Guozhang,
> > >
> > > Thanks for your comments.
> > >
> > > 1. That's a very good question. I think that it is better to separate
> > > the two concerns here. In my opinion, the GPA should only be used to
> > > feed the assignor. I see the GPA as a way to call a remote function so
> > > the GPA provides all its inputs to ensure that the assignor does not
> > > depend on any local state. Knowing this, I would not use the GPA to
> > > update the local state either. That is not its purpose. The consumer
> > > is responsible to update its local metadata cache either when it gets
> > > an unknown topic in the user-provided subscription or in the
> > > coordinator-provided assignment. The consumer has to do this anyway to
> > > get all topic metadata (e.g. leaders).
> > >
> > > 2. Interesting thought. Handling the translation is definitely
> > > possible. I wonder if there are cases where having the topic ids is
> > > beneficial. I am thinking about when a topic is deleted and recreated
> > > right away. In this case, the current member assignment would still
> > > have the old topic id and we would not be able to resolve it anymore.
> > > I suppose that we should directly remove it from the current
> > > assignment on the server side in this case and only add the new one.
> > > As I first thought, that seems OK. The major downside is that the user
> > > of the assignor would not know that the topic is different from what
> > > it may have cached locally.
> > >
> > > We could say that we don't expose topic ids externally for now in
> > > order to stay consistent with the other interfaces but it would be
> > > annoying to add it back if we want to expose topic ids in the future.
> > > I personally believe that we will have to expose them at some point.
> > > For instance, in the rebalance listener, it would be useful to know
> > > that a topic has been deleted and recreated. The only way to do this
> > > is to know the topic id. Let me think a little more about this and
> > > come back to you.
> > >
> > > 3. Sorry for this one. I was a typo. PartitionAssingor#assign should
> > > return GroupAssignment.
> > >
> > > Best,
> > > David
> > >
> > > On Thu, Sep 29, 2022 at 6:06 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > >
> > > > Thanks David, I read through the updated wiki, and I have a couple
> > > > follow-up questions:
> > > >
> > > > 1. It seems we would have two potential sources to update the
> consumer's
> > > > internally maintained topic name -> topic ID mapping, which is used
> for
> > > > encoding subscribed topic Ids in the HB request, and offset commit /
> list
> > > > offset requests, etc. One is from metadata responses, and another
> one is
> > > > from the GPA (group prepare assignment). On the other hand, the
> assignor
> > > > APIs directly take parsed AssignmentMemberSpec and
> > > AssignmentTopicMetadata
> > > > from GPA responses directly instead of trying to consult the
> consumer's
> > > own
> > > > cached mapping. So just to clarify, consumers would also update its
> > > > internal mapping from the GPA responses as well, right?
> > > >
> > > > 2. I'm wondering if it's necessary to expose the topic IDs to the
> > > assignor
> > > > developers. What they need to do at the moment is to take the mapping
> > > from
> > > > AssignmentTopicMetadata, and translate their decided assignment into
> > > topic
> > > > IDs when returning the MemberAssignment. We can instead just let
> consumer
> > > > to do such translation for them internally and still let the assignor
> > > > developers to only deal with the topic names. Of course that means
> the
> > > > client side assignor APIs would be a bit further diverged from the
> > > > broker-side assignor such that:
> > > >
> > > > a) Topic metadata would be encoded just as String -> num.partitions;
> > > > b) AssignmentMemberSpec contains subscribedTopicNames and
> topicPartitions
> > > > as target assignments;
> > > > c) Returned MemberAssignment contains list of topicPartitions
> instead of
> > > > topicIdPartitions.
> > > >
> > > > Internally, consumers would convert the IDs to names when calling the
> > > > assign() function, and convert them back when sending the request;
> and
> > > > similarly it would convert the IDs to names when calling
> onAssignment()
> > > as
> > > > well.
> > > >
> > > > In short, I'm wondering if we should bite the bullet for the users,
> > > eating
> > > > more complexities internally than pushing that to users.
> > > >
> > > > 3. This is a nit one: the return type of `Assignment
> > > assign(AssignmentSpec
> > > > assignmentSpec);` seems not defined in the doc. I think it would just
> > > > include two fields, as:
> > > >
> > > > class Assignment {
> > > >         GroupAssignment groupAssignment;
> > > >
> > > >         ConsumerGroupMetadata consumerGroupMetadata;
> > > >     }
> > > >
> > > > Is that right?
> > > >
> > > > Guozhang
> > > >
> > > > On Thu, Sep 29, 2022 at 1:26 AM David Jacot
> <djacot@confluent.io.invalid
> > > >
> > > > wrote:
> > > >
> > > > > Hi Guozhang,
> > > > >
> > > > > Thanks for your input.
> > > > >
> > > > > > 1) in HB request, when clients send the subscription, they are
> still
> > > > > > sending them in terms of topic names. That is, the consumer would
> > > not try
> > > > > > to keep all the mappings from topic names to topic ids for the
> > > subscribed
> > > > > > topics.
> > > > >
> > > > > This is a mistake of mine. The heartbeat request should use Topic
> Ids.
> > > > > I forgot to change it. The idea is that the client will get the
> > > > > metadata first to translate the topic names to topic ids and thus
> also
> > > > > have the mapping in its local metadata cache.
> > > > >
> > > > > > 2) in HB responses, the assignment's topics are encoded as IDs
> > > already.
> > > > > And
> > > > > > fetch requests can directly rely on that since they just need
> topic
> > > IDs
> > > > > and
> > > > > > do not care what actual topic names are.
> > > > >
> > > > > That's right.
> > > > >
> > > > > > 3) in prepare-assignment responses, we return subscriptions as
> topic
> > > IDs
> > > > > as
> > > > > > well.
> > > > >
> > > > > That's right.
> > > > >
> > > > > > In a word, the consumer does not try to maintain the mapping of
> topic
> > > > > names
> > > > > > to IDs at all, and even if they do, they will not expose it in
> e.g.
> > > > > > assignor APIs. But for client-side assignors like Streams, if
> they
> > > do not
> > > > > > maintain the mapping from topic IDs to topic names, certain logic
> > > such as
> > > > > > co-partitioning indeed cannot be executed.
> > > > >
> > > > > As I said, the consumer will have the mapping in its local metadata
> > > > > cache. It needs to get the partition metadata anyway and the topic
> id
> > > > > is part of it. However, I do agree that this mapping is not
> exposed.
> > > > >
> > > > > > So I think we'd need to consider 1) use topic names in
> subscriptions
> > > in
> > > > > the
> > > > > > prepare-assignment responses, 2) use topic Names in client-side
> > > > > > PartitionAssignor#GroupMember#subscribedTopics and
> > > > > > PartitionAssignor#TopicMetadata#topicName.
> > > > >
> > > > > 1) makes sense to me. I think that we could actually provide both
> > > > > names and ids. Regarding 2), I think that I am slightly in favor of
> > > > > using both names and ids as well. Let me try to update the protocol
> > > > > and the interface towards this.
> > > > >
> > > > > Thanks,
> > > > > David
> > > > >
> > > > > On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > > > >
> > > > > > Hello Jun/David,
> > > > > >
> > > > > > Just to chime in here for Jun's comment #50, today Streams does
> not
> > > have
> > > > > a
> > > > > > way to map topic names to topic IDs, and in its topology
> description
> > > > > topics
> > > > > > are encoded with names.
> > > > > >
> > > > > > What I originally thought was that, on the consumer side, we'd
> need
> > > to
> > > > > > maintain this mapping from topic names to topic IDs anyways, and
> > > Streams
> > > > > > should leverage on the consumer's functionality to get the
> > > corresponding
> > > > > > topic names for the given topic IDs. But when I read through the
> RPC
> > > > > again
> > > > > > I realized it may not be the case, since:
> > > > > >
> > > > > > 1) in HB request, when clients send the subscription, they are
> still
> > > > > > sending them in terms of topic names. That is, the consumer would
> > > not try
> > > > > > to keep all the mappings from topic names to topic ids for the
> > > subscribed
> > > > > > topics.
> > > > > > 2) in HB responses, the assignment's topics are encoded as IDs
> > > already.
> > > > > And
> > > > > > fetch requests can directly rely on that since they just need
> topic
> > > IDs
> > > > > and
> > > > > > do not care what actual topic names are.
> > > > > > 3) in prepare-assignment responses, we return subscriptions as
> topic
> > > IDs
> > > > > as
> > > > > > well.
> > > > > >
> > > > > > In a word, the consumer does not try to maintain the mapping of
> topic
> > > > > names
> > > > > > to IDs at all, and even if they do, they will not expose it in
> e.g.
> > > > > > assignor APIs. But for client-side assignors like Streams, if
> they
> > > do not
> > > > > > maintain the mapping from topic IDs to topic names, certain logic
> > > such as
> > > > > > co-partitioning indeed cannot be executed.
> > > > > >
> > > > > > So I think we'd need to consider 1) use topic names in
> subscriptions
> > > in
> > > > > the
> > > > > > prepare-assignment responses, 2) use topic Names in client-side
> > > > > > PartitionAssignor#GroupMember#subscribedTopics and
> > > > > > PartitionAssignor#TopicMetadata#topicName.
> > > > > >
> > > > > > On Wed, Sep 28, 2022 at 1:26 AM David Jacot
> > > <djacot@confluent.io.invalid
> > > > > >
> > > > > > wrote:
> > > > > >
> > > > > > > Hi Jun,
> > > > > > >
> > > > > > > Thanks for your comments.
> > > > > > >
> > > > > > > 48. Okay. There is a typo in the text:
> > > > > > > > Note that the persisted target assignment of B already
> includes
> > > foo-2
> > > > > > > here
> > > > > > > but B does not get it yet because foo-2 is not free.
> > > > > > >
> > > > > > > It should be:
> > > > > > > > Note that the persisted member assignment of B already
> includes
> > > foo-2
> > > > > > > here
> > > > > > > but B does not get it yet because foo-2 is not free.
> > > > > > >
> > > > > > > The examples are showing the persisted state in the
> > > __consumer_offsets
> > > > > > > topics. When a member transitions to the next epoch, we
> directly
> > > > > > > persist the whole target assignment in order to reduce the
> number
> > > of
> > > > > > > writes and we manage dependencies in memory. So "B - epoch=2,
> > > > > > > partitions=[foo-2]" is actually correct from a persisted record
> > > > > > > perspective. However, the in-memory state would be different
> > > because
> > > > > > > foo-2 is not yet available as you pointed out. The in-memory
> state
> > > > > > > would like as follow:
> > > > > > >
> > > > > > > Group (epoch=2)
> > > > > > > A
> > > > > > > B
> > > > > > > Target Assignment (epoch=2)
> > > > > > > A - partitions=[foo-0, foo-1]
> > > > > > > B - partitions=[foo-2]
> > > > > > > Member Assignment
> > > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending
> > > partitions=[]
> > > > > > > B - epoch=2, partitions=[], pending partitions=[foo-2]
> > > > > > >
> > > > > > > I suppose that the confusion comes from this. I will go and
> convert
> > > > > > > all the examples to their in-memory representation in order to
> > > reduce
> > > > > > > the confusion.
> > > > > > >
> > > > > > > 50. Okay. I have to admit that I am not so familiar with the
> > > internals
> > > > > > > of KStreams. If I understand correctly, you're saying that
> KStreams
> > > > > > > may not have topic ids in its topology so topic names are
> missing
> > > here
> > > > > > > to match the topics. I was hoping that KStreams would have
> topic
> > > ids
> > > > > > > in the topology as well. If that is not the case, we can also
> add
> > > > > > > topic names on our side. I will check this.
> > > > > > >
> > > > > > > 53. It seems that I was not clear. Sorry for that. I did not
> mean
> > > to
> > > > > > > say that MemberAssignment.metadata is not used by the group
> > > > > > > coordinator. Let me re-explain the flow.
> > > > > > >
> > > > > > > PartitionAssignor.assign() returns an Assignment. Assignment
> > > contains
> > > > > > > MemberAssignments where each member has
> MemberAssignment.metadata.
> > > The
> > > > > > > Assignment is installed by the client with the
> > > > > > > ConsumerGroupInstallAssignment API and this overrides the
> current
> > > > > > > target assignment in the coordinator. Then the group
> coordinator
> > > uses
> > > > > > > it to reconcile members. Therefore the target assignment will
> > > > > > > eventually become the current member assignment and it will be
> > > > > > > delivered to the member, including the original
> > > > > > > MemberAssignment.metadata defines by the assignor.
> > > > > > >
> > > > > > > I hope that this helps to clarify how
> MemberAssignment.metadata is
> > > > > > > used afterwards.
> > > > > > >
> > > > > > > Best,
> > > > > > > David
> > > > > > >
> > > > > > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao
> <jun@confluent.io.invalid
> > > >
> > > > > wrote:
> > > > > > > > Hi, David,
> > > > > > > >
> > > > > > > > Thanks for the reply. A few more follow up questions.
> > > > > > > >
> > > > > > > > 48. Hmm, so member assignment is what the client sees. Then,
> I
> > > am not
> > > > > > > sure
> > > > > > > > about the following. The text says B cannot get foo-2 yet.
> But in
> > > > > Member
> > > > > > > > Assignment, B has foo-2.
> > > > > > > >
> > > > > > > > At this point B can transition to epoch 2 but cannot get
> foo-2
> > > until
> > > > > A
> > > > > > > > revokes it.
> > > > > > > >
> > > > > > > > Note that the persisted target assignment of B already
> includes
> > > foo-2
> > > > > > > here
> > > > > > > > but B does not get it yet because foo-2 is not free.
> > > > > > > >
> > > > > > > > Group (epoch=2)
> > > > > > > > A
> > > > > > > > B
> > > > > > > > Target Assignment (epoch=2)
> > > > > > > > A - partitions=[foo-0, foo-1]
> > > > > > > > B - partitions=[foo-2]
> > > > > > > > Member Assignment
> > > > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > > > > > > B - epoch=2, partitions=[foo-2]
> > > > > > > >
> > > > > > > >
> > > > > > > > 50. Hmm, I thought KStreams doesn't use range assignor all
> the
> > > time.
> > > > > It
> > > > > > > > does the assignment based on the topology. For example, if
> you
> > > are
> > > > > > > joining
> > > > > > > > two topics A and B and deriving the join result in a 3rd
> topic C,
> > > > > which
> > > > > > > > will be aggregated. You want to use range to co-locate the
> > > partitions
> > > > > > > from
> > > > > > > > topic A and B. However, for topic C, the partition could
> > > potentially
> > > > > be
> > > > > > > > assigned in round robin. Without the topic name, KStreams
> won't
> > > know
> > > > > > > which
> > > > > > > > topics need to be co-located.
> > > > > > > >
> > > > > > > > 53. If MemberAssignment.metadata is not used by the group
> > > > > coordinator,
> > > > > > > > where is it being used?
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > >
> > > > > > > > Jun
> > > > > > > >
> > > > > > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > > > > > > <ja...@confluent.io.invalid>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi David,
> > > > > > > > >
> > > > > > > > > Thanks for the response. Replied below:
> > > > > > > > >
> > > > > > > > > > I don't have a strong opinion on this one. However, it
> seems
> > > that
> > > > > > > this
> > > > > > > > > could be a pandora box if our subset turns out to not be
> > > sufficient
> > > > > > > > > for some existing users. We will have to add more to it
> and the
> > > > > > > > > compatibility story between versions could be a real
> > > nightmare. One
> > > > > > > > > thing that we could do for sure is disabling all unicode
> > > groups in
> > > > > the
> > > > > > > > > library. That seems possible and would already reduce the
> > > scope.
> > > > > > > > >
> > > > > > > > > Changes to the regex syntax seems not a major problem to
> me. We
> > > > > would
> > > > > > > bump
> > > > > > > > > the protocol version as we usually do. Actually that begs
> the
> > > same
> > > > > > > question
> > > > > > > > > about RE2. How stable is the syntax and does it differ
> > > depending
> > > > > on the
> > > > > > > > > version of the library? In general, I do agree that we'd be
> > > better
> > > > > off
> > > > > > > not
> > > > > > > > > having to manage the syntax ourselves if we can convince
> > > ourselves
> > > > > that
> > > > > > > > > compatibility won't be a problem and the client won't
> > > eventually
> > > > > need
> > > > > > > to
> > > > > > > > > have access to it.
> > > > > > > > >
> > > > > > > > > Thanks,
> > > > > > > > > Jason
> > > > > > > > >
> > > > > > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hi Luke,
> > > > > > > > > >
> > > > > > > > > > Thanks for your comments.
> > > > > > > > > >
> > > > > > > > > > Your suggestion makes sense. Jun's made the same one. We
> are
> > > also
> > > > > > > > > > discussing whether we should directly return the group
> state
> > > in
> > > > > the
> > > > > > > > > > response and drop the prepare assignment api. It seems
> that
> > > the
> > > > > > > > > > current consensus is to keep it though. If that stays
> like
> > > this,
> > > > > I
> > > > > > > > > > will apply your suggestion.
> > > > > > > > > >
> > > > > > > > > > Thanks,
> > > > > > > > > > David
> > > > > > > > > >
> > > > > > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <
> > > showuon@gmail.com>
> > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > Hi David,
> > > > > > > > > > >
> > > > > > > > > > > Thanks for your patience for answering all these
> questions!
> > > > > > > > > > >
> > > > > > > > > > > One more question from me:
> > > > > > > > > > > 1. When the selected consumer is going to be in
> client-side
> > > > > > > assigning,
> > > > > > > > > we
> > > > > > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response.
> I'm
> > > > > > > wondering if
> > > > > > > > > we
> > > > > > > > > > > could use a flag to indicate the assignment is
> required.
> > > The
> > > > > > > rationale
> > > > > > > > > is
> > > > > > > > > > > when returning error, we lose a chance to catch up
> with the
> > > > > latest
> > > > > > > > > target
> > > > > > > > > > > assignment or revoke partitions. Since we expected the
> > > record
> > > > > > > fetching
> > > > > > > > > is
> > > > > > > > > > > not blocked by the rebalance/assignment computing, we
> might
> > > > > want to
> > > > > > > > > allow
> > > > > > > > > > > the members to update the member assignment in each
> > > heartbeat.
> > > > > > > > > > >
> > > > > > > > > > > Luke
> > > > > > > > > > >
> > > > > > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > > > > > > <ja...@confluent.io.invalid>
> > > > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > > Hey David,
> > > > > > > > > > > >
> > > > > > > > > > > > Thanks for the response.
> > > > > > > > > > > >
> > > > > > > > > > > > > Do you think that we could have a case where two
> > > members
> > > > > with
> > > > > > > the
> > > > > > > > > > same
> > > > > > > > > > > > static id are active?
> > > > > > > > > > > >
> > > > > > > > > > > > I was just thinking about the handoff when we spin
> up a
> > > new
> > > > > node
> > > > > > > > > > before the
> > > > > > > > > > > > old one completes shutdown. It's probably ok to fence
> > > the old
> > > > > > > member
> > > > > > > > > > > > abruptly. I do recall we had a few issues with this
> > > handoff
> > > > > in
> > > > > > > the
> > > > > > > > > > past. It
> > > > > > > > > > > > was due to the fact that the replaced node might have
> > > been
> > > > > the
> > > > > > > > > leader.
> > > > > > > > > > > > Since we've removed the leader notion, perhaps this
> kind
> > > of
> > > > > issue
> > > > > > > > > won't
> > > > > > > > > > > > resurface.
> > > > > > > > > > > >
> > > > > > > > > > > > In regard to the RE2 library. I understand the
> intent to
> > > pass
> > > > > > > this
> > > > > > > > > > through
> > > > > > > > > > > > as a string without any validation so that we don't
> need
> > > the
> > > > > > > > > > dependence on
> > > > > > > > > > > > the client. I do wonder if that will be restrictive
> in
> > > > > practice.
> > > > > > > For
> > > > > > > > > > > > example, if a regex changes on the client side,
> would we
> > > > > need to
> > > > > > > > > > validate
> > > > > > > > > > > > against the current assignment to see which
> partitions to
> > > > > > > continue
> > > > > > > > > > > > fetching? Or would we instead invalidate the complete
> > > current
> > > > > > > > > > assignment?
> > > > > > > > > > > > Perhaps this is a rare case and we can do the latter.
> > > There
> > > > > > > could be
> > > > > > > > > > more
> > > > > > > > > > > > situations though.
> > > > > > > > > > > >
> > > > > > > > > > > > As an alternative, I wonder if we could define a
> > > minimalistic
> > > > > > > regex
> > > > > > > > > > syntax
> > > > > > > > > > > > (a subset of RE2 say) that could map more easily to
> regex
> > > > > support
> > > > > > > > > that
> > > > > > > > > > > > already exists in clients? For example, in this
> > > scenario, we
> > > > > are
> > > > > > > > > > matching
> > > > > > > > > > > > topics, so we only need to be able to represent topic
> > > > > characters
> > > > > > > and
> > > > > > > > > > not
> > > > > > > > > > > > the whole smorgasbord of whitespace and unicode
> options.
> > > > > Would it
> > > > > > > > > make
> > > > > > > > > > > > sense to have a reduced syntax which includes only
> > > standard
> > > > > topic
> > > > > > > > > > > > characters, wildcards, character classes, and
> > > repetitions?
> > > > > > > > > > > >
> > > > > > > > > > > > Best,
> > > > > > > > > > > > Jason
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > > > > > > wangguoz@gmail.com>
> > > > > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > > Hi David, thanks for all the detailed
> explanations. I
> > > think
> > > > > > > they
> > > > > > > > > all
> > > > > > > > > > make
> > > > > > > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > > > > > > >
> > > > > > > > > > > > > > I don't really see the benefits here because
> server
> > > side
> > > > > > > > > assignors
> > > > > > > > > > > > > don't have metadata at all. They only assign
> > > > > topic-partitions.
> > > > > > > They
> > > > > > > > > > > > > are not supposed to generate metadata nor to
> receive
> > > > > metadata
> > > > > > > from
> > > > > > > > > > the
> > > > > > > > > > > > > members.
> > > > > > > > > > > > >
> > > > > > > > > > > > > Just to clarify I was asking about the `version`
> of the
> > > > > > > assignor
> > > > > > > > > > (i.e. up
> > > > > > > > > > > > > to what version that the client would support),
> and I
> > > do
> > > > > agree
> > > > > > > we
> > > > > > > > > > would
> > > > > > > > > > > > not
> > > > > > > > > > > > > need metadata. What I have in mind is that, for
> some
> > > > > specific
> > > > > > > > > > built-in
> > > > > > > > > > > > > broker-assignors, e.g. rack-aware assignors, if
> it's
> > > > > possible
> > > > > > > that
> > > > > > > > > > in a
> > > > > > > > > > > > > newer version we would have a hierarchical rack ID
> > > string
> > > > > > > format,
> > > > > > > > > > like
> > > > > > > > > > > > > "tier1-tier2" etc, but if some client has not
> upgraded
> > > > > their
> > > > > > > rack
> > > > > > > > > ID
> > > > > > > > > > > > > would still be in old format. In this case, the
> broker
> > > then
> > > > > > > needs
> > > > > > > > > to
> > > > > > > > > > > > choose
> > > > > > > > > > > > > the old versioned assignor. I'm probably making
> > > something
> > > > > up
> > > > > > > here
> > > > > > > > > for
> > > > > > > > > > > > rack
> > > > > > > > > > > > > aware assignors, but I'm wondering if in general
> such
> > > an
> > > > > > > > > > "auto-downgrade"
> > > > > > > > > > > > > behavior would be needed still for broker-side
> > > assignor,
> > > > > and
> > > > > > > if yes
> > > > > > > > > > would
> > > > > > > > > > > > > "version" still be useful.
> > > > > > > > > > > > >
> > > > > > > > > > > > > > Yeah, that's right. Within a rebalance,
> > > `onAssignment` is
> > > > > > > called
> > > > > > > > > > once
> > > > > > > > > > > > > when the member transitions to a new epoch. This
> one
> > > > > contains
> > > > > > > the
> > > > > > > > > > full
> > > > > > > > > > > > > metadata provided by the client side assignor.
> Then,
> > > > > > > `onAssignment`
> > > > > > > > > > > > > can be called max N times where N is the number of
> > > > > partitions
> > > > > > > > > pending
> > > > > > > > > > > > > revocation by other members. Let me try to clarify
> > > this in
> > > > > the
> > > > > > > KIP.
> > > > > > > > > > > > >
> > > > > > > > > > > > > Okay, my understanding is that the calling
> ordering of
> > > > > these
> > > > > > > > > > callbacks
> > > > > > > > > > > > > would be like the following:
> > > > > > > > > > > > >
> > > > > > > > > > > > > ----------------------------------------
> > > > > > > > > > > > > onPartitionsRevoked();   // just once, since we do
> not
> > > > > really
> > > > > > > need
> > > > > > > > > > > > > to revoke incrementally.
> > > > > > > > > > > > >
> > > > > > > > > > > > > onAssignment();    // the first call, with epoch
> > > > > incremented
> > > > > > > > > > > > > onPartitionsAssigned();   // paired with the
> > > onAssignment
> > > > > > > > > > > > >
> > > > > > > > > > > > > onAssignment();              // the first
> onAssignment
> > > > > would
> > > > > > > bump
> > > > > > > > > up
> > > > > > > > > > the
> > > > > > > > > > > > > epoch, and the metadata reflected.
> > > > > > > > > > > > > onPartitionsAssigned();   // each time we get an
> > > additional
> > > > > > > > > > assignment,
> > > > > > > > > > > > we
> > > > > > > > > > > > > call onAssignment and then paired with an
> > > > > onPartitionsAssigned
> > > > > > > > > > > > > ...
> > > > > > > > > > > > > onAssignment();
> > > > > > > > > > > > > onPartitionsAssigned();   // on each of the
> > > onAssignment
> > > > > > > calls, the
> > > > > > > > > > > > encoded
> > > > > > > > > > > > > metadata would not change, only the incrementally
> added
> > > > > > > partitions
> > > > > > > > > be
> > > > > > > > > > > > > reflected
> > > > > > > > > > > > >
> > > > > > > > > > > > > Is that the case?
> > > > > > > > > > > > >
> > > > > > > > > > > > > I'm wondering if we would still call onAssignment
> just
> > > > > once,
> > > > > > > that
> > > > > > > > > > encodes
> > > > > > > > > > > > > all the assignment for this rebalance, including
> all
> > > the
> > > > > > > partitions
> > > > > > > > > > that
> > > > > > > > > > > > > should be assigned to the member but not yet
> assigned
> > > since
> > > > > > > they
> > > > > > > > > > have not
> > > > > > > > > > > > > been revoked by others. In that case the call
> ordering
> > > > > would
> > > > > > > be:
> > > > > > > > > > > > >
> > > > > > > > > > > > > ----------------------------------------
> > > > > > > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > > > > > > onAssignment();    // just once, with epoch
> > > incremented,
> > > > > and
> > > > > > > > > metadata
> > > > > > > > > > > > > encoded changed, the "assignment" field also
> reflect
> > > the
> > > > > final
> > > > > > > > > target
> > > > > > > > > > > > > assignment
> > > > > > > > > > > > > onPartitionsAssigned();   // multiple times, which
> > > > > represent
> > > > > > > > > > > > incrementally
> > > > > > > > > > > > > added partitions
> > > > > > > > > > > > > ...
> > > > > > > > > > > > > onPartitionsAssigned();
> > > > > > > > > > > > >
> > > > > > > > > > > > > The motivation from this is that, most users would
> only
> > > > > > > implement
> > > > > > > > > the
> > > > > > > > > > > > > rebalance callback listeners and hence we'd
> definitely
> > > > > need to
> > > > > > > make
> > > > > > > > > > sure
> > > > > > > > > > > > > the semantics of that does not change much, and the
> > > time
> > > > > > > > > > > > > `onPartitionsAssigned` indicate the time when the
> > > > > partitions
> > > > > > > are
> > > > > > > > > > actually
> > > > > > > > > > > > > assigned to it; while for assignors, the
> > > `onAssignment` is
> > > > > > > used to
> > > > > > > > > > > > indicate
> > > > > > > > > > > > > what decision is made regarding for this member,
> i.e.
> > > when
> > > > > the
> > > > > > > > > > partitions
> > > > > > > > > > > > > are decided to be given to it, but not necessarily
> > > meant
> > > > > that
> > > > > > > it
> > > > > > > > > has
> > > > > > > > > > been
> > > > > > > > > > > > > given, since that time should be determined by the
> > > time of
> > > > > > > > > > > > > `onPartitionsAssigned`. The benefits though, would
> be
> > > that
> > > > > > > assignor
> > > > > > > > > > > > > implementers would not need to reason which
> > > `onAssignment`
> > > > > > > would be
> > > > > > > > > > the
> > > > > > > > > > > > > last one for this epoch.
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > >
> > > > > > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >
> > > > > > > > > > > > > wrote:
> > > > > > > > > > > > >
> > > > > > > > > > > > > > Hi Guozhang,
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 1) The client-side
> "PartitionAssignor#Assignment"
> > > has
> > > > > an
> > > > > > > Error
> > > > > > > > > > byte
> > > > > > > > > > > > > > field,
> > > > > > > > > > > > > > > while the broker-side
> > > "PartitionAssignor#Assignment"
> > > > > does
> > > > > > > not.
> > > > > > > > > > And
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > rationale seems to be assuming that we should
> > > always be
> > > > > > > able to
> > > > > > > > > > do
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > assignment at the broker-side assignor without
> > > errors.
> > > > > > > > > > Personally I
> > > > > > > > > > > > > think
> > > > > > > > > > > > > > > it's still potentially beneficial to add the
> Error
> > > > > field
> > > > > > > even
> > > > > > > > > for
> > > > > > > > > > > > > > > broker-side assignors, e.g. for some edge cases
> > > where
> > > > > some
> > > > > > > > > > subscribed
> > > > > > > > > > > > > > > topics are not recognized with the current
> broker's
> > > > > > > metadata.
> > > > > > > > > > What do
> > > > > > > > > > > > > you
> > > > > > > > > > > > > > > think?
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Yeah, that seems reasonable. However, I wonder
> if it
> > > > > would be
> > > > > > > > > > better
> > > > > > > > > > > > > > to use an exception on the server side. This is
> what
> > > we
> > > > > > > usually
> > > > > > > > > do
> > > > > > > > > > for
> > > > > > > > > > > > > > server side plugins. On the client side, we use a
> > > field
> > > > > > > because
> > > > > > > > > the
> > > > > > > > > > > > > > errors are not defined in advance.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Your comment also makes me think about what we
> > > should do
> > > > > > > when the
> > > > > > > > > > > > > > server side assignor fails. I suppose that we
> have to
> > > > > keep
> > > > > > > the
> > > > > > > > > > current
> > > > > > > > > > > > > > assignment until a new event occurs. For
> instance, in
> > > > > your
> > > > > > > > > example,
> > > > > > > > > > > > > > the coordinator would have to trigger a rebalance
> > > when
> > > > > > > > > unrecognized
> > > > > > > > > > > > > > topics are available. This would be part of the
> > > metadata
> > > > > > > > > > monitoring.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> > > additional
> > > > > > > fields
> > > > > > > > > > > > > > > reason/version/metadata compared with the
> > > broker-side
> > > > > > > > > > GroupMember. I
> > > > > > > > > > > > > > agree
> > > > > > > > > > > > > > > that broker-side assignor would not need
> > > > > reason/metadata
> > > > > > > since
> > > > > > > > > > they
> > > > > > > > > > > > are
> > > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but
> what
> > > about
> > > > > > > version?
> > > > > > > > > > E.g.
> > > > > > > > > > > > is
> > > > > > > > > > > > > > it
> > > > > > > > > > > > > > > possible that we evolve our broker-side
> built-in
> > > > > assignor
> > > > > > > but
> > > > > > > > > > the old
> > > > > > > > > > > > > > > versioned clients would not be able to work
> with
> > > the
> > > > > new
> > > > > > > > > > version, in
> > > > > > > > > > > > > > which
> > > > > > > > > > > > > > > case we need to let the broker being aware of
> this
> > > and
> > > > > > > upgrade
> > > > > > > > > > its
> > > > > > > > > > > > > > behavior
> > > > > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > I don't really see the benefits here because
> server
> > > side
> > > > > > > > > assignors
> > > > > > > > > > > > > > don't have metadata at all. They only assign
> > > > > > > topic-partitions.
> > > > > > > > > They
> > > > > > > > > > > > > > are not supposed to generate metadata nor to
> receive
> > > > > metadata
> > > > > > > > > from
> > > > > > > > > > the
> > > > > > > > > > > > > > members.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 3) Also related to 2) above, for the
> client-side
> > > > > > > "GroupMember",
> > > > > > > > > > > > instead
> > > > > > > > > > > > > > of
> > > > > > > > > > > > > > > including these three fields, what about just
> > > adding
> > > > > the
> > > > > > > > > > "Metadata"
> > > > > > > > > > > > > field
> > > > > > > > > > > > > > > class which has these three fields? Also, there
> > > are two
> > > > > > > > > > "Metadata"
> > > > > > > > > > > > > > > currently in the APIs, the first is a class
> that
> > > > > encodes
> > > > > > > > > > > > > > > reason/version/metadata, and the second is
> just the
> > > > > encoded
> > > > > > > > > > metadata
> > > > > > > > > > > > > > bytes.
> > > > > > > > > > > > > > > I'm wondering what about just naming the first
> as
> > > > > > > > > memberMetadata,
> > > > > > > > > > > > which
> > > > > > > > > > > > > > > then has a bytebuffer field Metadata, or
> instead
> > > > > naming the
> > > > > > > > > > second
> > > > > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > That's a good point. Let me try to rationalize
> this
> > > > > interface
> > > > > > > > > > based on
> > > > > > > > > > > > > > your suggestions.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Best,
> > > > > > > > > > > > > > David
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Hello David,
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Just had a few more nit questions about the
> public
> > > > > APIs:
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 1) The client-side
> "PartitionAssignor#Assignment"
> > > has
> > > > > an
> > > > > > > Error
> > > > > > > > > > byte
> > > > > > > > > > > > > > field,
> > > > > > > > > > > > > > > while the broker-side
> > > "PartitionAssignor#Assignment"
> > > > > does
> > > > > > > not.
> > > > > > > > > > And
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > rationale seems to be assuming that we should
> > > always be
> > > > > > > able to
> > > > > > > > > > do
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > assignment at the broker-side assignor without
> > > errors.
> > > > > > > > > > Personally I
> > > > > > > > > > > > > think
> > > > > > > > > > > > > > > it's still potentially beneficial to add the
> Error
> > > > > field
> > > > > > > even
> > > > > > > > > for
> > > > > > > > > > > > > > > broker-side assignors, e.g. for some edge cases
> > > where
> > > > > some
> > > > > > > > > > subscribed
> > > > > > > > > > > > > > > topics are not recognized with the current
> broker's
> > > > > > > metadata.
> > > > > > > > > > What do
> > > > > > > > > > > > > you
> > > > > > > > > > > > > > > think?
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> > > additional
> > > > > > > fields
> > > > > > > > > > > > > > > reason/version/metadata compared with the
> > > broker-side
> > > > > > > > > > GroupMember. I
> > > > > > > > > > > > > > agree
> > > > > > > > > > > > > > > that broker-side assignor would not need
> > > > > reason/metadata
> > > > > > > since
> > > > > > > > > > they
> > > > > > > > > > > > are
> > > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but
> what
> > > about
> > > > > > > version?
> > > > > > > > > > E.g.
> > > > > > > > > > > > is
> > > > > > > > > > > > > > it
> > > > > > > > > > > > > > > possible that we evolve our broker-side
> built-in
> > > > > assignor
> > > > > > > but
> > > > > > > > > > the old
> > > > > > > > > > > > > > > versioned clients would not be able to work
> with
> > > the
> > > > > new
> > > > > > > > > > version, in
> > > > > > > > > > > > > > which
> > > > > > > > > > > > > > > case we need to let the broker being aware of
> this
> > > and
> > > > > > > upgrade
> > > > > > > > > > its
> > > > > > > > > > > > > > behavior
> > > > > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 3) Also related to 2) above, for the
> client-side
> > > > > > > "GroupMember",
> > > > > > > > > > > > instead
> > > > > > > > > > > > > > of
> > > > > > > > > > > > > > > including these three fields, what about just
> > > adding
> > > > > the
> > > > > > > > > > "Metadata"
> > > > > > > > > > > > > field
> > > > > > > > > > > > > > > class which has these three fields? Also, there
> > > are two
> > > > > > > > > > "Metadata"
> > > > > > > > > > > > > > > currently in the APIs, the first is a class
> that
> > > > > encodes
> > > > > > > > > > > > > > > reason/version/metadata, and the second is
> just the
> > > > > encoded
> > > > > > > > > > metadata
> > > > > > > > > > > > > > bytes.
> > > > > > > > > > > > > > > I'm wondering what about just naming the first
> as
> > > > > > > > > memberMetadata,
> > > > > > > > > > > > which
> > > > > > > > > > > > > > > then has a bytebuffer field Metadata, or
> instead
> > > > > naming the
> > > > > > > > > > second
> > > > > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang
> <
> > > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > Hello David,
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > Thanks for bringing this question up. I
> think the
> > > > > main
> > > > > > > > > > benefits as
> > > > > > > > > > > > > you
> > > > > > > > > > > > > > > > listed is 2) above if it stays; just to
> clarify,
> > > we
> > > > > would
> > > > > > > > > only
> > > > > > > > > > be
> > > > > > > > > > > > > able
> > > > > > > > > > > > > > to
> > > > > > > > > > > > > > > > save one round trip if the rebalance is still
> > > > > triggered
> > > > > > > by
> > > > > > > > > the
> > > > > > > > > > > > > broker;
> > > > > > > > > > > > > > if
> > > > > > > > > > > > > > > > the rebalance is triggered by the client
> then the
> > > > > > > > > > num.round.trips
> > > > > > > > > > > > are
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > > same:
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > T0: client decides to do a new assignment,
> > > suppose
> > > > > it has
> > > > > > > > > > already
> > > > > > > > > > > > > sent
> > > > > > > > > > > > > > a
> > > > > > > > > > > > > > > > HB and hence has to wait for it to return
> first
> > > since
> > > > > > > only
> > > > > > > > > one
> > > > > > > > > > > > > request
> > > > > > > > > > > > > > /
> > > > > > > > > > > > > > > > response can be inflight with the
> coordinator's
> > > > > socket.
> > > > > > > > > > > > > > > > T1: client receives the HB response, and then
> > > sends
> > > > > the
> > > > > > > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > > > > > > T2: the GroupPrepareAssignment response is
> > > returned.
> > > > > > > > > > > > > > > > T3: it calculates the new assignment, and
> sends a
> > > > > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > > > > request.
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > In total, two round trips.
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > T0: client decides to do a new assignment,
> > > suppose
> > > > > it has
> > > > > > > > > > already
> > > > > > > > > > > > > sent
> > > > > > > > > > > > > > a
> > > > > > > > > > > > > > > > HB and hence has to wait for it to return
> first
> > > since
> > > > > > > only
> > > > > > > > > one
> > > > > > > > > > > > > request
> > > > > > > > > > > > > > /
> > > > > > > > > > > > > > > > response can be inflight with the
> coordinator's
> > > > > socket.
> > > > > > > > > > > > > > > > T1: client receives the HB response, and then
> > > sends
> > > > > the
> > > > > > > new
> > > > > > > > > HB
> > > > > > > > > > > > > request
> > > > > > > > > > > > > > > > with the flag indicating a new rebalance
> needed..
> > > > > > > > > > > > > > > > T2: the HB response with the optional member
> > > metadata
> > > > > > > map is
> > > > > > > > > > > > > returned.
> > > > > > > > > > > > > > > > T3: it calculates the new assignment, and
> sends a
> > > > > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > > > > request.
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > -----------------------------
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > So to complete the full picture here, we'd
> need
> > > to
> > > > > modify
> > > > > > > > > both
> > > > > > > > > > HB
> > > > > > > > > > > > > > request
> > > > > > > > > > > > > > > > and response so that the client can also
> > > indicate a
> > > > > new
> > > > > > > > > > rebalance
> > > > > > > > > > > > via
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > > HB request as well, right?
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > Assuming all above is true, I think it's
> okay to
> > > > > merge
> > > > > > > the
> > > > > > > > > > > > > > > > GroupPrepareAssignment into HB given that we
> can
> > > > > make the
> > > > > > > > > > > > additional
> > > > > > > > > > > > > > fields
> > > > > > > > > > > > > > > > encoding the full member (subscription)
> metadata
> > > and
> > > > > > > topic
> > > > > > > > > > metadata
> > > > > > > > > > > > > as
> > > > > > > > > > > > > > > > optional fields.
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >> Hi all,
> > > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> During an offline conversation, someone
> asked
> > > why we
> > > > > > > need
> > > > > > > > > the
> > > > > > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and
> suggested
> > > > > that we
> > > > > > > > > could
> > > > > > > > > > > > > instead
> > > > > > > > > > > > > > > >> provide the group state in the heartbeat
> > > response.
> > > > > This
> > > > > > > has
> > > > > > > > > a
> > > > > > > > > > few
> > > > > > > > > > > > > > > >> advantages: 1) it does not require using a
> > > special
> > > > > error
> > > > > > > > > code
> > > > > > > > > > to
> > > > > > > > > > > > > > > >> signal that a new assignment is required as
> the
> > > > > signal
> > > > > > > would
> > > > > > > > > > be
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> provided group state; 2) it removes one
> round
> > > trip
> > > > > when
> > > > > > > a
> > > > > > > > > > client
> > > > > > > > > > > > > side
> > > > > > > > > > > > > > > >> assignor is used. The downside is that it
> makes
> > > the
> > > > > > > > > heartbeat
> > > > > > > > > > > > > > > >> response's definition quite large. I recall
> > > that I
> > > > > went
> > > > > > > with
> > > > > > > > > > the
> > > > > > > > > > > > > > > >> current approach due to this.
> > > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> Providing the group state in the heartbeat
> > > response
> > > > > is
> > > > > > > > > > appealing.
> > > > > > > > > > > > > What
> > > > > > > > > > > > > > > >> do you guys think?
> > > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> Best,
> > > > > > > > > > > > > > > >> David
> > > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot
> <
> > > > > > > > > > djacot@confluent.io>
> > > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > 1. I have added a reference to the
> relevant
> > > > > chapter
> > > > > > > > > instead
> > > > > > > > > > of
> > > > > > > > > > > > > > > >> > repeating the whole thing. Does that work
> for
> > > you?
> > > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you
> are
> > > > > referring
> > > > > > > to
> > > > > > > > > is
> > > > > > > > > > > > about
> > > > > > > > > > > > > > when
> > > > > > > > > > > > > > > >> > a rebalance should be triggered for the
> > > > > non-upgraded
> > > > > > > > > members
> > > > > > > > > > > > using
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> > old protocol. The section mentions that a
> > > > > rebalance
> > > > > > > must
> > > > > > > > > be
> > > > > > > > > > > > > > triggered
> > > > > > > > > > > > > > > >> > when a new assignment is installed. This
> > > implies
> > > > > that
> > > > > > > the
> > > > > > > > > > group
> > > > > > > > > > > > > > epoch
> > > > > > > > > > > > > > > >> > was updated either by a native member or a
> > > > > > > non-upgraded
> > > > > > > > > > member.
> > > > > > > > > > > > > For
> > > > > > > > > > > > > > > >> > the latter, the JoinGroup request would
> be the
> > > > > > > trigger. I
> > > > > > > > > > have
> > > > > > > > > > > > > > added a
> > > > > > > > > > > > > > > >> > reference to the relevant chapter in the
> > > > > "JoinGroup
> > > > > > > > > > Handling"
> > > > > > > > > > > > > > section
> > > > > > > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > Thanks,
> > > > > > > > > > > > > > > >> > David
> > > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang
> Wang
> > > <
> > > > > > > > > > > > wangguoz@gmail.com
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > Hello David,
> > > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > Alright I think that's sufficient. Just
> to
> > > make
> > > > > that
> > > > > > > > > > clear in
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > doc,
> > > > > > > > > > > > > > > >> > > could we update:
> > > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > 1) the heartbeat request handling
> section,
> > > > > stating
> > > > > > > when
> > > > > > > > > > > > > > coordinator
> > > > > > > > > > > > > > > >> will
> > > > > > > > > > > > > > > >> > > trigger rebalance based on the HB's
> member
> > > > > metadata
> > > > > > > /
> > > > > > > > > > reason?
> > > > > > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to
> > > include
> > > > > what
> > > > > > > we
> > > > > > > > > > > > described
> > > > > > > > > > > > > > in
> > > > > > > > > > > > > > > >> "Group
> > > > > > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as
> > > well?
> > > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > Guozhang
> > > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David
> Jacot
> > > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > > >> > > wrote:
> > > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > I thought that the assignor will
> always be
> > > > > > > consulted
> > > > > > > > > > when
> > > > > > > > > > > > the
> > > > > > > > > > > > > > next
> > > > > > > > > > > > > > > >> > > > heartbeat request is constructed. In
> other
> > > > > words,
> > > > > > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be
> > > called
> > > > > for
> > > > > > > every
> > > > > > > > > > > > > heartbeat.
> > > > > > > > > > > > > > > >> This
> > > > > > > > > > > > > > > >> > > > gives the opportunity for the
> assignor to
> > > > > enforce
> > > > > > > a
> > > > > > > > > > > > rebalance
> > > > > > > > > > > > > by
> > > > > > > > > > > > > > > >> > > > setting the reason to a non-zero
> value or
> > > by
> > > > > > > changing
> > > > > > > > > > the
> > > > > > > > > > > > > > bytes. Do
> > > > > > > > > > > > > > > >> > > > you think that this is not sufficient?
> > > Are you
> > > > > > > > > > concerned by
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> delay?
> > > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > Best,
> > > > > > > > > > > > > > > >> > > > David
> > > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM
> Guozhang
> > > Wang <
> > > > > > > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > One of Jun's comments make me
> thinking:
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > > > > >> > > > > In this case, a new assignment is
> > > triggered
> > > > > by
> > > > > > > the
> > > > > > > > > > client
> > > > > > > > > > > > > side
> > > > > > > > > > > > > > > >> > > > > assignor. When constructing the HB,
> the
> > > > > consumer
> > > > > > > > > will
> > > > > > > > > > > > always
> > > > > > > > > > > > > > > >> consult
> > > > > > > > > > > > > > > >> > > > > the client side assignor and
> propagate
> > > the
> > > > > > > > > > information to
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > coordinator. In other words, we
> don't
> > > expect
> > > > > > > users
> > > > > > > > > to
> > > > > > > > > > call
> > > > > > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > As I looked at the current
> > > > > PartitionAssignor's
> > > > > > > > > > interface,
> > > > > > > > > > > > we
> > > > > > > > > > > > > > > >> actually do
> > > > > > > > > > > > > > > >> > > > > not have a way yet to instruct how
> to
> > > > > construct
> > > > > > > the
> > > > > > > > > > next
> > > > > > > > > > > > HB
> > > > > > > > > > > > > > > >> request, e.g.
> > > > > > > > > > > > > > > >> > > > > when the assignor wants to enforce
> a new
> > > > > > > rebalance
> > > > > > > > > > with a
> > > > > > > > > > > > > new
> > > > > > > > > > > > > > > >> assignment,
> > > > > > > > > > > > > > > >> > > > > we'd need some customizable APIs
> inside
> > > the
> > > > > > > > > > > > > PartitionAssignor
> > > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> indicate
> > > > > > > > > > > > > > > >> > > > > the next HB telling broker about so.
> > > WDYT
> > > > > about
> > > > > > > > > adding
> > > > > > > > > > > > such
> > > > > > > > > > > > > an
> > > > > > > > > > > > > > > >> API on the
> > > > > > > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David
> > > Jacot
> > > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > I have updated the KIP to include
> your
> > > > > > > feedback. I
> > > > > > > > > > have
> > > > > > > > > > > > > also
> > > > > > > > > > > > > > > >> tried to
> > > > > > > > > > > > > > > >> > > > > > clarify the parts which were not
> > > cleared.
> > > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM
> David
> > > > > Jacot <
> > > > > > > > > > > > > > djacot@confluent.io
> > > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me
> > > start
> > > > > by
> > > > > > > > > > answering
> > > > > > > > > > > > your
> > > > > > > > > > > > > > > >> questions
> > > > > > > > > > > > > > > >> > > > > > > inline and I will update the KIP
> > > next
> > > > > week.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall,
> the
> > > main
> > > > > > > benefits
> > > > > > > > > > of
> > > > > > > > > > > > the
> > > > > > > > > > > > > > KIP
> > > > > > > > > > > > > > > >> seem to
> > > > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > > > > efficient
> > > > > > > > > > support of
> > > > > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > I would also add that the KIP
> > > removes
> > > > > the
> > > > > > > global
> > > > > > > > > > sync
> > > > > > > > > > > > > > barrier
> > > > > > > > > > > > > > > >> in the
> > > > > > > > > > > > > > > >> > > > > > > protocol which is essential to
> > > improve
> > > > > group
> > > > > > > > > > stability
> > > > > > > > > > > > > and
> > > > > > > > > > > > > > > >> > > > > > > scalability, and the KIP also
> > > > > simplifies the
> > > > > > > > > > client by
> > > > > > > > > > > > > > moving
> > > > > > > > > > > > > > > >> most of
> > > > > > > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 30.
> ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> > > singleton.
> > > > > Do we
> > > > > > > plan
> > > > > > > > > > to
> > > > > > > > > > > > > > support
> > > > > > > > > > > > > > > >> rolling
> > > > > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in
> the
> > > > > > > consumers?
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Definitely. The group
> coordinator
> > > will
> > > > > use
> > > > > > > the
> > > > > > > > > > > > assignor
> > > > > > > > > > > > > > used
> > > > > > > > > > > > > > > >> by a
> > > > > > > > > > > > > > > >> > > > > > > majority of the members. This
> > > allows the
> > > > > > > group
> > > > > > > > > to
> > > > > > > > > > move
> > > > > > > > > > > > > > from
> > > > > > > > > > > > > > > >> one
> > > > > > > > > > > > > > > >> > > > > > > assignor to another by a roll.
> This
> > > is
> > > > > > > explained
> > > > > > > > > > in
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> Assignor
> > > > > > > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you
> > > explain
> > > > > > > whether
> > > > > > > > > > it's
> > > > > > > > > > > > > > > >> required in
> > > > > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > > > > >> > > > > > > > request or the scenarios when
> it
> > > > > needs to
> > > > > > > be
> > > > > > > > > > filled?
> > > > > > > > > > > > > For
> > > > > > > > > > > > > > > >> example,
> > > > > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > > > > >> > > > > > > > clear to me when
> TopicPartitions
> > > > > needs to
> > > > > > > be
> > > > > > > > > > filled.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > The client is expected to set
> those
> > > > > fields
> > > > > > > in
> > > > > > > > > > case of
> > > > > > > > > > > > a
> > > > > > > > > > > > > > > >> connection
> > > > > > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the
> > > fields
> > > > > have
> > > > > > > > > > changed
> > > > > > > > > > > > > since
> > > > > > > > > > > > > > > >> the last
> > > > > > > > > > > > > > > >> > > > > > > HB. The server populates those
> > > fields as
> > > > > > > long as
> > > > > > > > > > the
> > > > > > > > > > > > > > member
> > > > > > > > > > > > > > > >> is not
> > > > > > > > > > > > > > > >> > > > > > > fully reconciled - the member
> should
> > > > > > > acknowledge
> > > > > > > > > > that
> > > > > > > > > > > > it
> > > > > > > > > > > > > > has
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > expected epoch and assignment. I
> > > will
> > > > > > > clarify
> > > > > > > > > > this in
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > KIP.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> > > protocol,
> > > > > the
> > > > > > > rack
> > > > > > > > > > > > > affinity
> > > > > > > > > > > > > > > >> between the
> > > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > > >> > > > > > > > and the broker is only
> considered
> > > > > during
> > > > > > > > > > fetching,
> > > > > > > > > > > > but
> > > > > > > > > > > > > > not
> > > > > > > > > > > > > > > >> during
> > > > > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> > > Sometimes,
> > > > > once
> > > > > > > the
> > > > > > > > > > > > > assignment
> > > > > > > > > > > > > > is
> > > > > > > > > > > > > > > >> made,
> > > > > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > > > > >> > > > > > > > no opportunity for read
> affinity
> > > > > because
> > > > > > > no
> > > > > > > > > > replicas
> > > > > > > > > > > > > of
> > > > > > > > > > > > > > > >> assigned
> > > > > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > > > > wondering
> > > > > > > if we
> > > > > > > > > > should
> > > > > > > > > > > > > use
> > > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > > > > >> > > > > > > > to address this by including
> rack
> > > in
> > > > > > > > > > GroupMember.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > That's an interesting idea. I
> don't
> > > see
> > > > > any
> > > > > > > > > issue
> > > > > > > > > > with
> > > > > > > > > > > > > > adding
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > rack
> > > > > > > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often,
> > > it's
> > > > > > > useful to
> > > > > > > > > > know
> > > > > > > > > > > > how
> > > > > > > > > > > > > > busy
> > > > > > > > > > > > > > > >> a group
> > > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> > > event
> > > > > loop
> > > > > > > > > model,
> > > > > > > > > > it
> > > > > > > > > > > > > seems
> > > > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the
> fraction
> > > of
> > > > > the
> > > > > > > time
> > > > > > > > > > the
> > > > > > > > > > > > > event
> > > > > > > > > > > > > > > >> loop is
> > > > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > That's a great idea. I will add
> it.
> > > > > Thanks.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > > > > coordinator
> > > > > > > > > > failover
> > > > > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the
> check
> > > if
> > > > > any
> > > > > > > group
> > > > > > > > > > with
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> > > matching
> > > > > topic?
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Sure. When the new group
> coordinator
> > > > > takes
> > > > > > > over,
> > > > > > > > > > it
> > > > > > > > > > > > has
> > > > > > > > > > > > > > to:
> > > > > > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a
> > > client
> > > > > side
> > > > > > > > > > assignor
> > > > > > > > > > > > is
> > > > > > > > > > > > > > used.
> > > > > > > > > > > > > > > >> We
> > > > > > > > > > > > > > > >> > > > don't
> > > > > > > > > > > > > > > >> > > > > > > store the information about the
> > > member
> > > > > > > selected
> > > > > > > > > > to run
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > > > > > > >> > > > > > > * Update the topics metadata,
> > > verify the
> > > > > > > > > wildcard
> > > > > > > > > > > > > > > >> subscriptions, and
> > > > > > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 34.
> ConsumerGroupMetadataValue,
> > > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > > > > >> > > > > > > >
> ConsumerGroupMemberMetadataValue:
> > > > > Could we
> > > > > > > > > > document
> > > > > > > > > > > > > what
> > > > > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does
> the
> > > epoch
> > > > > in
> > > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> > > about the
> > > > > > > one in
> > > > > > > > > > > > > > > >> > > > > > > >
> > > ConsumerGroupPartitionMetadataValue
> > > > > and
> > > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Sure. I will clarify that but
> it is
> > > > > always
> > > > > > > the
> > > > > > > > > > latest
> > > > > > > > > > > > > > group
> > > > > > > > > > > > > > > >> epoch.
> > > > > > > > > > > > > > > >> > > > > > > When the group state is
> updated, the
> > > > > group
> > > > > > > epoch
> > > > > > > > > > is
> > > > > > > > > > > > > > bumped so
> > > > > > > > > > > > > > > >> we use
> > > > > > > > > > > > > > > >> > > > > > > that one for all the change
> records
> > > > > related
> > > > > > > to
> > > > > > > > > the
> > > > > > > > > > > > > update.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator
> will
> > > ensure
> > > > > > > that
> > > > > > > > > the
> > > > > > > > > > > > > > following
> > > > > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > > > > >> > > > > > > > met: ... All members exists."
> It's
> > > > > > > possible
> > > > > > > > > for
> > > > > > > > > > a
> > > > > > > > > > > > > member
> > > > > > > > > > > > > > > >> not to
> > > > > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > That's right. Here I meant that
> the
> > > > > members
> > > > > > > > > > provided
> > > > > > > > > > > > by
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > > > >> > > > > > > in the assignment must exist in
> the
> > > > > group.
> > > > > > > The
> > > > > > > > > > > > assignor
> > > > > > > > > > > > > > can
> > > > > > > > > > > > > > > >> not make
> > > > > > > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> > > with a
> > > > > > > member
> > > > > > > > > > epoch
> > > > > > > > > > > > > > equals to
> > > > > > > > > > > > > > > >> 0":
> > > > > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what
> member
> > > id
> > > > > > > would be
> > > > > > > > > > used?
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > A member is expected to abandon
> all
> > > its
> > > > > > > > > > partitions and
> > > > > > > > > > > > > > > >> rejoins when
> > > > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH
> > > error.
> > > > > In
> > > > > > > this
> > > > > > > > > > case,
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > > coordinator will have removed
> the
> > > member
> > > > > > > from
> > > > > > > > > the
> > > > > > > > > > > > group.
> > > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> member
> > > > > > > > > > > > > > > >> > > > > > > can rejoin the group with the
> same
> > > > > member
> > > > > > > id but
> > > > > > > > > > with
> > > > > > > > > > > > 0
> > > > > > > > > > > > > as
> > > > > > > > > > > > > > > >> epoch. Let
> > > > > > > > > > > > > > > >> > > > > > > me see if I can clarify this in
> the
> > > KIP.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will
> > > have
> > > > > the
> > > > > > > > > ability
> > > > > > > > > > to
> > > > > > > > > > > > > > trigger a
> > > > > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > > > > >> > > > > > > > by either providing a non-zero
> > > reason
> > > > > or
> > > > > > > by
> > > > > > > > > > updating
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to
> be
> > > > > > > conflicting
> > > > > > > > > > with
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > In this case, a new assignment
> is
> > > > > triggered
> > > > > > > by
> > > > > > > > > the
> > > > > > > > > > > > > client
> > > > > > > > > > > > > > side
> > > > > > > > > > > > > > > >> > > > > > > assignor. When constructing the
> HB,
> > > the
> > > > > > > consumer
> > > > > > > > > > will
> > > > > > > > > > > > > > always
> > > > > > > > > > > > > > > >> consult
> > > > > > > > > > > > > > > >> > > > > > > the client side assignor and
> > > propagate
> > > > > the
> > > > > > > > > > information
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > > coordinator. In other words, we
> > > don't
> > > > > expect
> > > > > > > > > > users to
> > > > > > > > > > > > > call
> > > > > > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance
> anymore.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples
> are
> > > > > nice.
> > > > > > > But
> > > > > > > > > the
> > > > > > > > > > > > > section
> > > > > > > > > > > > > > > >> seems to
> > > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group
> transitions to
> > > > > epoch
> > > > > > > 2, B
> > > > > > > > > > > > > > immediately
> > > > > > > > > > > > > > > >> gets into
> > > > > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]",
> > > which
> > > > > seems
> > > > > > > > > > > > incorrect.
> > > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group
> transitions to
> > > > > epoch
> > > > > > > 3, C
> > > > > > > > > > seems
> > > > > > > > > > > > to
> > > > > > > > > > > > > > get
> > > > > > > > > > > > > > > >> into
> > > > > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to
> epoch
> > > 3, C
> > > > > > > still
> > > > > > > > > > has A -
> > > > > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Sorry for that! I will revise
> them.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of
> consumers:
> > > Do
> > > > > we
> > > > > > > > > support
> > > > > > > > > > the
> > > > > > > > > > > > > > upgrade
> > > > > > > > > > > > > > > >> from
> > > > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > We will support upgrading from
> the
> > > > > consumer
> > > > > > > > > > protocol
> > > > > > > > > > > > > > version
> > > > > > > > > > > > > > > >> 3,
> > > > > > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792
> is
> > > not
> > > > > > > > > implemented
> > > > > > > > > > yet
> > > > > > > > > > > > so
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> earliest
> > > > > > > > > > > > > > > >> > > > > > > version is unknown at the
> moment.
> > > This
> > > > > is
> > > > > > > > > > explained in
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> migration
> > > > > > > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Thanks again for your feedback,
> > > Jun. I
> > > > > will
> > > > > > > > > > update the
> > > > > > > > > > > > > KIP
> > > > > > > > > > > > > > > >> based on
> > > > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > > > > > > >> > > > > > > David
> > > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM
> Jun
> > > Rao
> > > > > > > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall,
> the
> > > main
> > > > > > > benefits
> > > > > > > > > > of
> > > > > > > > > > > > the
> > > > > > > > > > > > > > KIP
> > > > > > > > > > > > > > > >> seem to
> > > > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > > > > efficient
> > > > > > > > > > support of
> > > > > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 30.
> ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> > > singleton.
> > > > > Do we
> > > > > > > plan
> > > > > > > > > > to
> > > > > > > > > > > > > > support
> > > > > > > > > > > > > > > >> rolling
> > > > > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in
> the
> > > > > > > consumers?
> > > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you
> > > explain
> > > > > > > whether
> > > > > > > > > > it's
> > > > > > > > > > > > > > > >> required in
> > > > > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > > > > >> > > > > > > > request or the scenarios when
> it
> > > > > needs to
> > > > > > > be
> > > > > > > > > > filled?
> > > > > > > > > > > > > For
> > > > > > > > > > > > > > > >> example,
> > > > > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > > > > >> > > > > > > > clear to me when
> TopicPartitions
> > > > > needs to
> > > > > > > be
> > > > > > > > > > filled.
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> > > protocol,
> > > > > the
> > > > > > > rack
> > > > > > > > > > > > > affinity
> > > > > > > > > > > > > > > >> between the
> > > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > > >> > > > > > > > and the broker is only
> considered
> > > > > during
> > > > > > > > > > fetching,
> > > > > > > > > > > > but
> > > > > > > > > > > > > > not
> > > > > > > > > > > > > > > >> during
> > > > > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> > > Sometimes,
> > > > > once
> > > > > > > the
> > > > > > > > > > > > > assignment
> > > > > > > > > > > > > > is
> > > > > > > > > > > > > > > >> made,
> > > > > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > > > > >> > > > > > > > no opportunity for read
> affinity
> > > > > because
> > > > > > > no
> > > > > > > > > > replicas
> > > > > > > > > > > > > of
> > > > > > > > > > > > > > > >> assigned
> > > > > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > > > > wondering
> > > > > > > if we
> > > > > > > > > > should
> > > > > > > > > > > > > use
> > > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > > > > >> > > > > > > > to address this by including
> rack
> > > in
> > > > > > > > > > GroupMember.
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often,
> > > it's
> > > > > > > useful to
> > > > > > > > > > know
> > > > > > > > > > > > how
> > > > > > > > > > > > > > busy
> > > > > > > > > > > > > > > >> a group
> > > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> > > event
> > > > > loop
> > > > > > > > > model,
> > > > > > > > > > it
> > > > > > > > > > > > > seems
> > > > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the
> fraction
> > > of
> > > > > the
> > > > > > > time
> > > > > > > > > > the
> > > > > > > > > > > > > event
> > > > > > > > > > > > > > > >> loop is
> > > > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > > > > coordinator
> > > > > > > > > > failover
> > > > > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the
> check
> > > if
> > > > > any
> > > > > > > group
> > > > > > > > > > with
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> > > matching
> > > > > topic?
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 34.
> ConsumerGroupMetadataValue,
> > > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > > > > >> > > > > > > >
> ConsumerGroupMemberMetadataValue:
> > > > > Could we
> > > > > > > > > > document
> > > > > > > > > > > > > what
> > > > > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does
> the
> > > epoch
> > > > > in
> > > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> > > about the
> > > > > > > one in
> > > > > > > > > > > > > > > >> > > > > > > >
> > > ConsumerGroupPartitionMetadataValue
> > > > > and
> > > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator
> will
> > > ensure
> > > > > > > that
> > > > > > > > > the
> > > > > > > > > > > > > > following
> > > > > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > > > > >> > > > > > > > met: ... All members exists."
> It's
> > > > > > > possible
> > > > > > > > > for
> > > > > > > > > > a
> > > > > > > > > > > > > member
> > > > > > > > > > > > > > > >> not to
> > > > > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> > > with a
> > > > > > > member
> > > > > > > > > > epoch
> > > > > > > > > > > > > > equals to
> > > > > > > > > > > > > > > >> 0":
> > > > > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what
> member
> > > id
> > > > > > > would be
> > > > > > > > > > used?
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will
> > > have
> > > > > the
> > > > > > > > > ability
> > > > > > > > > > to
> > > > > > > > > > > > > > trigger a
> > > > > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > > > > >> > > > > > > > by either providing a non-zero
> > > reason
> > > > > or
> > > > > > > by
> > > > > > > > > > updating
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to
> be
> > > > > > > conflicting
> > > > > > > > > > with
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples
> are
> > > > > nice.
> > > > > > > But
> > > > > > > > > the
> > > > > > > > > > > > > section
> > > > > > > > > > > > > > > >> seems to
> > > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group
> transitions to
> > > > > epoch
> > > > > > > 2, B
> > > > > > > > > > > > > > immediately
> > > > > > > > > > > > > > > >> gets into
> > > > > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]",
> > > which
> > > > > seems
> > > > > > > > > > > > incorrect.
> > > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group
> transitions to
> > > > > epoch
> > > > > > > 3, C
> > > > > > > > > > seems
> > > > > > > > > > > > to
> > > > > > > > > > > > > > get
> > > > > > > > > > > > > > > >> into
> > > > > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to
> epoch
> > > 3, C
> > > > > > > still
> > > > > > > > > > has A -
> > > > > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of
> consumers:
> > > Do
> > > > > we
> > > > > > > > > support
> > > > > > > > > > the
> > > > > > > > > > > > > > upgrade
> > > > > > > > > > > > > > > >> from
> > > > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20
> AM
> > > David
> > > > > > > Jacot
> > > > > > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > > > > > > re-implement the
> > > > > > > > > > > > > > coordinator
> > > > > > > > > > > > > > > >> in
> > > > > > > > > > > > > > > >> > > > Java. I
> > > > > > > > > > > > > > > >> > > > > > > > > discussed this offline with
> a
> > > few
> > > > > folks
> > > > > > > and
> > > > > > > > > > folks
> > > > > > > > > > > > > are
> > > > > > > > > > > > > > > >> concerned
> > > > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > > > >> > > > > > > > > we could introduce many
> > > regressions
> > > > > in
> > > > > > > the
> > > > > > > > > old
> > > > > > > > > > > > > > protocol
> > > > > > > > > > > > > > > >> if we do
> > > > > > > > > > > > > > > >> > > > so.
> > > > > > > > > > > > > > > >> > > > > > > > > Therefore, I am going to
> remove
> > > this
> > > > > > > > > statement
> > > > > > > > > > > > from
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> KIP. It
> > > > > > > > > > > > > > > >> > > > is an
> > > > > > > > > > > > > > > >> > > > > > > > > implementation detail after
> all
> > > so
> > > > > it
> > > > > > > does
> > > > > > > > > not
> > > > > > > > > > > > have
> > > > > > > > > > > > > > to be
> > > > > > > > > > > > > > > >> > > > decided at
> > > > > > > > > > > > > > > >> > > > > > > > > this stage. We will likely
> > > start by
> > > > > > > trying
> > > > > > > > > to
> > > > > > > > > > > > > > refactor the
> > > > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > > > >> > > > > > > > > implementation as a first
> step.
> > > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at
> 3:52 PM
> > > > > David
> > > > > > > Jacot
> > > > > > > > > <
> > > > > > > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> > > machine
> > > > > are:
> > > > > > > > > > "Empty,
> > > > > > > > > > > > > > assigning,
> > > > > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in
> Consumer
> > > > > Group
> > > > > > > States
> > > > > > > > > > > > > section,
> > > > > > > > > > > > > > > >> right?
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > This sentence does not
> refer
> > > to
> > > > > those
> > > > > > > > > group
> > > > > > > > > > > > states
> > > > > > > > > > > > > > but
> > > > > > > > > > > > > > > >> rather
> > > > > > > > > > > > > > > >> > > > to a
> > > > > > > > > > > > > > > >> > > > > > > > > > state machine replication
> > > (SMR).
> > > > > This
> > > > > > > > > > refers to
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> entire
> > > > > > > > > > > > > > > >> > > > state of
> > > > > > > > > > > > > > > >> > > > > > > > > > group coordinator which is
> > > > > replicated
> > > > > > > via
> > > > > > > > > > the
> > > > > > > > > > > > log
> > > > > > > > > > > > > > > >> layer. I will
> > > > > > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean
> "each
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > is
> > > > > > > > > > > > > > modelled
> > > > > > > > > > > > > > > >> as an
> > > > > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > The idea is to follow a
> model
> > > > > similar
> > > > > > > to
> > > > > > > > > > the new
> > > > > > > > > > > > > > quorum
> > > > > > > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > > > > > > >> > > > > > > > > > will have N threads to
> process
> > > > > events.
> > > > > > > > > Each
> > > > > > > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > > > > > > >> > > > > > > > > > partition is assigned to a
> > > unique
> > > > > > > thread
> > > > > > > > > > and all
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> events
> > > > > > > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > > > > > > >> > > > > > > > > > requests, callbacks,
> etc.) are
> > > > > > > processed
> > > > > > > > > by
> > > > > > > > > > this
> > > > > > > > > > > > > > > >> thread. This
> > > > > > > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > > > > > > >> > > > > > > > > > concurrency and will
> enable
> > > us to
> > > > > do
> > > > > > > > > > simulation
> > > > > > > > > > > > > > testing
> > > > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a
> state
> > > > > machine
> > > > > > > per
> > > > > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > > > > consumer
> > > > > > > group"
> > > > > > > > > > owned
> > > > > > > > > > > > > by
> > > > > > > > > > > > > > a
> > > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > > > > coordinator
> > > > > > > owns 2
> > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> > > will we
> > > > > > > have 1
> > > > > > > > > > state
> > > > > > > > > > > > > > machine
> > > > > > > > > > > > > > > >> for it,
> > > > > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion
> comes
> > > from
> > > > > > > there, I
> > > > > > > > > > > > think.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > > > > "*group.coordinator.threads"
> > > > > > > > > > > > *is
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> number
> > > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > > > > >> > > > > > > > > > > to run the state
> machines.
> > > But
> > > > > I'm
> > > > > > > > > > wondering
> > > > > > > > > > > > if
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> purpose
> > > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the
> state of
> > > > > each
> > > > > > > > > consumer
> > > > > > > > > > > > group
> > > > > > > > > > > > > > (or
> > > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no
> heavy
> > > > > > > computation,
> > > > > > > > > > why
> > > > > > > > > > > > > > should we
> > > > > > > > > > > > > > > >> need
> > > > > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to
> have
> > > an
> > > > > > > ability to
> > > > > > > > > > shard
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > processing as
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > computation could be
> heavy.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer
> session
> > > > > timeout,
> > > > > > > why
> > > > > > > > > > does
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> default
> > > > > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > > > > >> > > > > > > > > > > locate between min
> (45s) and
> > > > > > > max(60s)? I
> > > > > > > > > > > > thought
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> min/max
> > > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> > > lower/upper
> > > > > > > bound
> > > > > > > > > of
> > > > > > > > > > it,
> > > > > > > > > > > > > no?
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > group.consumer.session.timeout.ms
> > > > > > > int
> > > > > > > > > > 30s The
> > > > > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> > > consumer
> > > > > > > group
> > > > > > > > > > > > protocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > group.consumer.min.session.timeout.ms
> > > > > > > > > > int 45s
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> minimum
> > > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > group.consumer.max.session.timeout.ms
> > > > > > > > > > int 60s
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> maximum
> > > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > This is indeed a mistake.
> The
> > > > > default
> > > > > > > > > > session
> > > > > > > > > > > > > > timeout
> > > > > > > > > > > > > > > >> should
> > > > > > > > > > > > > > > >> > > > be 45s
> > > > > > > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server
> side
> > > > > > > assignor
> > > > > > > > > are
> > > > > > > > > > > > > [range,
> > > > > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > > > > assignor.
> > > > > > > I'd
> > > > > > > > > > like to
> > > > > > > > > > > > > > know
> > > > > > > > > > > > > > > >> why not
> > > > > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users
> will
> > > > > choose
> > > > > > > > > uniform
> > > > > > > > > > > > > assignor
> > > > > > > > > > > > > > > >> (former
> > > > > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> > > distribution.
> > > > > Any
> > > > > > > > > other
> > > > > > > > > > > > reason
> > > > > > > > > > > > > > we
> > > > > > > > > > > > > > > >> choose
> > > > > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors
> > > List
> > > > > range,
> > > > > > > > > > uniform
> > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> server side
> > > > > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > The order on the server
> side
> > > has
> > > > > no
> > > > > > > > > > influence
> > > > > > > > > > > > > > because
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > > > > >> > > > > > must
> > > > > > > > > > > > > > > >> > > > > > > > > > chose the selector that he
> > > wants
> > > > > to
> > > > > > > use.
> > > > > > > > > > There
> > > > > > > > > > > > is
> > > > > > > > > > > > > no
> > > > > > > > > > > > > > > >> default
> > > > > > > > > > > > > > > >> > > > in the
> > > > > > > > > > > > > > > >> > > > > > > > > > current proposal. If the
> > > assignor
> > > > > is
> > > > > > > not
> > > > > > > > > > > > specified
> > > > > > > > > > > > > > by
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > client,
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > request is rejected. The
> > > default
> > > > > > > client
> > > > > > > > > > value
> > > > > > > > > > > > for
> > > > > > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is
> > > > > `uniform`
> > > > > > > > > though.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > Thanks for your very good
> > > > > comments,
> > > > > > > Luke.
> > > > > > > > > I
> > > > > > > > > > hope
> > > > > > > > > > > > > > that my
> > > > > > > > > > > > > > > >> > > > answers
> > > > > > > > > > > > > > > >> > > > > > help
> > > > > > > > > > > > > > > >> > > > > > > > > > to clarify things. I will
> > > update
> > > > > the
> > > > > > > KIP
> > > > > > > > > as
> > > > > > > > > > well
> > > > > > > > > > > > > > based
> > > > > > > > > > > > > > > >> on your
> > > > > > > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at
> 9:29
> > > AM
> > > > > Luke
> > > > > > > Chen
> > > > > > > > > <
> > > > > > > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator
> > > > > section, you
> > > > > > > > > > > > mentioned:
> > > > > > > > > > > > > > > >> > > > > > > > > > > > The new group
> coordinator
> > > will
> > > > > > > have a
> > > > > > > > > > state
> > > > > > > > > > > > > > machine
> > > > > > > > > > > > > > > >> per
> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets*
> > > partitions,
> > > > > > > where
> > > > > > > > > > each
> > > > > > > > > > > > > state
> > > > > > > > > > > > > > > >> machine is
> > > > > > > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > > > > > > >> > > > > > > > > > > event loop. Those state
> > > machines
> > > > > > > will be
> > > > > > > > > > > > > executed
> > > > > > > > > > > > > > in
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> *group.coordinator.threads*
> > > > > threads.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> > > machine
> > > > > are:
> > > > > > > > > > "Empty,
> > > > > > > > > > > > > > assigning,
> > > > > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in
> Consumer
> > > > > Group
> > > > > > > States
> > > > > > > > > > > > > section,
> > > > > > > > > > > > > > > >> right?
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean
> "each
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > is
> > > > > > > > > > > > > > modelled
> > > > > > > > > > > > > > > >> as an
> > > > > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a
> state
> > > > > machine
> > > > > > > per
> > > > > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > > > > consumer
> > > > > > > group"
> > > > > > > > > > owned
> > > > > > > > > > > > > by
> > > > > > > > > > > > > > a
> > > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > > > > coordinator
> > > > > > > owns 2
> > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> > > will we
> > > > > > > have 1
> > > > > > > > > > state
> > > > > > > > > > > > > > machine
> > > > > > > > > > > > > > > >> for it,
> > > > > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > > > > "*group.coordinator.threads"
> > > > > > > > > > > > *is
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> number
> > > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > > > > >> > > > > > > > > > > to run the state
> machines.
> > > But
> > > > > I'm
> > > > > > > > > > wondering
> > > > > > > > > > > > if
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> purpose
> > > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the
> state of
> > > > > each
> > > > > > > > > consumer
> > > > > > > > > > > > group
> > > > > > > > > > > > > > (or
> > > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no
> heavy
> > > > > > > computation,
> > > > > > > > > > why
> > > > > > > > > > > > > > should we
> > > > > > > > > > > > > > > >> need
> > > > > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 2. For the default
> value in
> > > the
> > > > > new
> > > > > > > > > > configs:
> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer
> session
> > > > > timeout,
> > > > > > > why
> > > > > > > > > > does
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> default
> > > > > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > > > > >> > > > > > > > > > > locate between min
> (45s) and
> > > > > > > max(60s)? I
> > > > > > > > > > > > thought
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> min/max
> > > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> > > lower/upper
> > > > > > > bound
> > > > > > > > > of
> > > > > > > > > > it,
> > > > > > > > > > > > > no?
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > group.consumer.session.timeout.ms
> > > > > > > int
> > > > > > > > > > 30s The
> > > > > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> > > consumer
> > > > > > > group
> > > > > > > > > > > > protocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > group.consumer.min.session.timeout.ms
> > > > > > > > > > int 45s
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> minimum
> > > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > group.consumer.max.session.timeout.ms
> > > > > > > > > > int 60s
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> maximum
> > > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server
> side
> > > > > > > assignor
> > > > > > > > > are
> > > > > > > > > > > > > [range,
> > > > > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > > > > assignor.
> > > > > > > I'd
> > > > > > > > > > like to
> > > > > > > > > > > > > > know
> > > > > > > > > > > > > > > >> why not
> > > > > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users
> will
> > > > > choose
> > > > > > > > > uniform
> > > > > > > > > > > > > assignor
> > > > > > > > > > > > > > > >> (former
> > > > > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> > > distribution.
> > > > > Any
> > > > > > > > > other
> > > > > > > > > > > > reason
> > > > > > > > > > > > > > we
> > > > > > > > > > > > > > > >> choose
> > > > > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors
> > > List
> > > > > range,
> > > > > > > > > > uniform
> > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> server side
> > > > > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at
> > > 2:10 PM
> > > > > Luke
> > > > > > > > > Chen
> > > > > > > > > > <
> > > > > > > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts
> about
> > > > > Kafka
> > > > > > > > > Connect
> > > > > > > > > > > > > > > >> integrating with
> > > > > > > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > > > > > > >> > > > > > > > > > > > think we should have a
> > > > > separate
> > > > > > > > > > discussion
> > > > > > > > > > > > > > thread
> > > > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > > > >> > > > Kafka
> > > > > > > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka
> > > Connect
> > > > > > > With
> > > > > > > > > New
> > > > > > > > > > > > > Consumer
> > > > > > > > > > > > > > > >> Rebalance
> > > > > > > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > > > > > > >> > > > > > > > > > > > and let this
> discussion
> > > thread
> > > > > > > focus
> > > > > > > > > on
> > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > > >> rebalance
> > > > > > > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >>
> > > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022
> at
> > > 9:31
> > > > > PM
> > > > > > > Sagar
> > > > > > > > > <
> > > > > > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Thank you
> Guozhang/David
> > > for
> > > > > the
> > > > > > > > > > feedback.
> > > > > > > > > > > > > > Looks
> > > > > > > > > > > > > > > >> like
> > > > > > > > > > > > > > > >> > > > there's
> > > > > > > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs
> for
> > > > > Connect.
> > > > > > > I
> > > > > > > > > > would
> > > > > > > > > > > > > > revisit
> > > > > > > > > > > > > > > >> the doc
> > > > > > > > > > > > > > > >> > > > and
> > > > > > > > > > > > > > > >> > > > > > see
> > > > > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> changes are to be
> made.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022
> at
> > > 7:11
> > > > > PM
> > > > > > > David
> > > > > > > > > > Jacot
> > > > > > > > > > > > > > > >> > > > > > > > > <djacot@confluent.io.invalid
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the
> feedback
> > > > > and the
> > > > > > > > > > document.
> > > > > > > > > > > > > > That's
> > > > > > > > > > > > > > > >> really
> > > > > > > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will take a look
> at it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems
> to me
> > > > > that
> > > > > > > both
> > > > > > > > > > Connect
> > > > > > > > > > > > > > and the
> > > > > > > > > > > > > > > >> > > > Consumer
> > > > > > > > > > > > > > > >> > > > > > could
> > > > > > > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the same underlying
> > > > > "engine".
> > > > > > > The
> > > > > > > > > > main
> > > > > > > > > > > > > > > >> difference is
> > > > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigns
> > > topic-partitions to
> > > > > > > members
> > > > > > > > > > > > whereas
> > > > > > > > > > > > > > > >> Connect
> > > > > > > > > > > > > > > >> > > > assigns
> > > > > > > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two
> > > ways to
> > > > > move
> > > > > > > > > > forward:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the
> new
> > > > > proposed
> > > > > > > APIs
> > > > > > > > > to
> > > > > > > > > > > > > support
> > > > > > > > > > > > > > > >> different
> > > > > > > > > > > > > > > >> > > > > > resource
> > > > > > > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions,
> > > tasks,
> > > > > > > etc.); or
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new
> dedicated
> > > > > APIs
> > > > > > > for
> > > > > > > > > > Connect.
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> dedicated
> > > > > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new
> > > ones but
> > > > > > > > > > different on
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > they would rely on
> the
> > > same
> > > > > > > engine
> > > > > > > > > > on the
> > > > > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > > > > >> > > > side.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean
> > > towards
> > > > > 2)
> > > > > > > > > because
> > > > > > > > > > I am
> > > > > > > > > > > > > > not a
> > > > > > > > > > > > > > > >> fan of
> > > > > > > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve
> different
> > > > > > > purposes.
> > > > > > > > > > That
> > > > > > > > > > > > > being
> > > > > > > > > > > > > > > >> said, I am
> > > > > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find
> an
> > > > > elegant
> > > > > > > way to
> > > > > > > > > > do
> > > > > > > > > > > > it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I think that we can
> > > > > continue to
> > > > > > > > > > discuss
> > > > > > > > > > > > it
> > > > > > > > > > > > > > here
> > > > > > > > > > > > > > > >> for now
> > > > > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > > > > >> > > > > > order
> > > > > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this
> KIP is
> > > > > > > compatible
> > > > > > > > > > with
> > > > > > > > > > > > > what
> > > > > > > > > > > > > > we
> > > > > > > > > > > > > > > >> will do
> > > > > > > > > > > > > > > >> > > > for
> > > > > > > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8,
> 2022 at
> > > > > 2:41 PM
> > > > > > > > > David
> > > > > > > > > > > > Jacot
> > > > > > > > > > > > > <
> > > > > > > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from
> > > vacation.
> > > > > I
> > > > > > > will
> > > > > > > > > go
> > > > > > > > > > > > > through
> > > > > > > > > > > > > > and
> > > > > > > > > > > > > > > >> address
> > > > > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming
> days.
> > > > > Thanks
> > > > > > > for
> > > > > > > > > your
> > > > > > > > > > > > > > feedback.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3,
> 2022
> > > at
> > > > > 10:05
> > > > > > > PM
> > > > > > > > > > Gregory
> > > > > > > > > > > > > > Harris
> > > > > > > > > > > > > > > >> <
> > > > > > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the
> KIP,
> > > > > it's
> > > > > > > > > > wonderful to
> > > > > > > > > > > > > see
> > > > > > > > > > > > > > > >> > > > cooperative
> > > > > > > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few
> > > questions:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The
> 'Rejected
> > > > > > > Alternatives'
> > > > > > > > > > > > section
> > > > > > > > > > > > > > > >> describes how
> > > > > > > > > > > > > > > >> > > > > > member
> > > > > > > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step
> > > with
> > > > > the
> > > > > > > group
> > > > > > > > > > epoch
> > > > > > > > > > > > > and
> > > > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a
> good
> > > > > idea
> > > > > > > for
> > > > > > > > > the
> > > > > > > > > > > > > reasons
> > > > > > > > > > > > > > > >> described
> > > > > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is
> > > > > incrementally
> > > > > > > > > > assigning
> > > > > > > > > > > > > > > >> partitions to a
> > > > > > > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each
> > > > > incremental
> > > > > > > > > > > > assignment
> > > > > > > > > > > > > > use?
> > > > > > > > > > > > > > > >> Are
> > > > > > > > > > > > > > > >> > > > member
> > > > > > > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > a single member
> > > epoch
> > > > > can
> > > > > > > > > > correspond
> > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> multiple
> > > > > > > > > > > > > > > >> > > > > > different
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > larger)
> > > assignments?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the
> > > Assignor's
> > > > > > > 'Reason'
> > > > > > > > > > field
> > > > > > > > > > > > > > opaque
> > > > > > > > > > > > > > > >> to the
> > > > > > > > > > > > > > > >> > > > group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should
> custom
> > > > > > > client-side
> > > > > > > > > > > > assignor
> > > > > > > > > > > > > > > >> > > > implementations
> > > > > > > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field,
> and
> > > how
> > > > > is
> > > > > > > its
> > > > > > > > > > common
> > > > > > > > > > > > > > meaning
> > > > > > > > > > > > > > > >> agreed
> > > > > > > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a
> > > distinct
> > > > > > > Reason
> > > > > > > > > > field
> > > > > > > > > > > > over
> > > > > > > > > > > > > > > >> including
> > > > > > > > > > > > > > > >> > > > such
> > > > > > > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque
> metadata?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The
> following is
> > > > > > > included in
> > > > > > > > > > the
> > > > > > > > > > > > > KIP:
> > > > > > > > > > > > > > > >> "Thanks to
> > > > > > > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client side
> > > assignor is
> > > > > > > > > entirely
> > > > > > > > > > > > driven
> > > > > > > > > > > > > > by
> > > > > > > > > > > > > > > >> the group
> > > > > > > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no
> > > longer
> > > > > > > > > > responsible for
> > > > > > > > > > > > > > > >> maintaining
> > > > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > > > >> > > > > > state
> > > > > > > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned
> > > partitions."
> > > > > Does
> > > > > > > this
> > > > > > > > > > mean
> > > > > > > > > > > > > > that the
> > > > > > > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate
> > > additional
> > > > > > > > > > non-Metadata
> > > > > > > > > > > > > state
> > > > > > > > > > > > > > > >> (such as
> > > > > > > > > > > > > > > >> > > > > > partition
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory
> metrics,
> > > > > config
> > > > > > > > > > topics,
> > > > > > > > > > > > > etc),
> > > > > > > > > > > > > > or
> > > > > > > > > > > > > > > >> that
> > > > > > > > > > > > > > > >> > > > > > additional
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD
> NOT be
> > > > > used?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that
> > > there are
> > > > > > > > > separate
> > > > > > > > > > > > > classes
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > > > > > > >> > > >
> > > > > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap
> > > significantly.
> > > > > Is
> > > > > > > it
> > > > > > > > > > possible
> > > > > > > > > > > > > for
> > > > > > > > > > > > > > > >> these two
> > > > > > > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This
> would
> > > > > serve
> > > > > > > to
> > > > > > > > > > promote
> > > > > > > > > > > > > > feature
> > > > > > > > > > > > > > > >> parity
> > > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side
> > > assignors,
> > > > > and
> > > > > > > > > would
> > > > > > > > > > also
> > > > > > > > > > > > > > > >> facilitate
> > > > > > > > > > > > > > > >> > > > > > operational
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > certain
> > > situations. For
> > > > > > > > > example,
> > > > > > > > > > if a
> > > > > > > > > > > > > > > >> server-side
> > > > > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and
> needs
> > > a
> > > > > patch,
> > > > > > > > > > deploying
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> patched
> > > > > > > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching
> one
> > > > > consumer
> > > > > > > > > group
> > > > > > > > > > to a
> > > > > > > > > > > > > > > >> client-side
> > > > > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky
> than
> > > > > > > patching
> > > > > > > > > all
> > > > > > > > > > of
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> brokers.
> > > > > > > > > > > > > > > >> > > > With
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed
> distinct
> > > > > APIs, a
> > > > > > > > > > non-trivial
> > > > > > > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and
> if
> > > the
> > > > > two
> > > > > > > APIs
> > > > > > > > > > have
> > > > > > > > > > > > > > diverged
> > > > > > > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > > > > > > reimplementation
> > > > > > > > > > > > would
> > > > > > > > > > > > > > not be
> > > > > > > > > > > > > > > >> > > > possible.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > gharris1727@gmail.com
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > github.com/gharris1727
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3,
> > > 2022 at
> > > > > > > 8:39 AM
> > > > > > > > > > Sagar
> > > > > > > > > > > > <
> > > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi
> > > Guozhang/David,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a
> > > > > confluence
> > > > > > > page
> > > > > > > > > to
> > > > > > > > > > > > > discuss
> > > > > > > > > > > > > > how
> > > > > > > > > > > > > > > >> Connect
> > > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the
> new
> > > > > > > rebalance
> > > > > > > > > > > > protocol.
> > > > > > > > > > > > > > > >> Here's the
> > > > > > > > > > > > > > > >> > > > page:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >>
> > > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also
> pretty
> > > > > longish
> > > > > > > and
> > > > > > > > > I
> > > > > > > > > > have
> > > > > > > > > > > > > > tried
> > > > > > > > > > > > > > > >> to keep
> > > > > > > > > > > > > > > >> > > > a
> > > > > > > > > > > > > > > >> > > > > > format
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let
> me
> > > know
> > > > > > > what you
> > > > > > > > > > > > think.
> > > > > > > > > > > > > > Also,
> > > > > > > > > > > > > > > >> do you
> > > > > > > > > > > > > > > >> > > > > > think this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a
> > > separate
> > > > > > > > > discussion
> > > > > > > > > > > > thread
> > > > > > > > > > > > > > or
> > > > > > > > > > > > > > > >> is this
> > > > > > > > > > > > > > > >> > > > one
> > > > > > > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul
> 26,
> > > 2022
> > > > > at
> > > > > > > 7:37
> > > > > > > > > AM
> > > > > > > > > > > > > Sagar <
> > > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello
> Guozhang,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you
> so
> > > much
> > > > > for
> > > > > > > the
> > > > > > > > > > doc on
> > > > > > > > > > > > > > Kafka
> > > > > > > > > > > > > > > >> Streams.
> > > > > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis
> and
> > > come
> > > > > up
> > > > > > > with
> > > > > > > > > > such a
> > > > > > > > > > > > > > > >> document.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue,
> Jul 26,
> > > > > 2022 at
> > > > > > > > > 4:47
> > > > > > > > > > AM
> > > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > > > >> Wang <
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello
> Sagar,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would
> be
> > > great
> > > > > if
> > > > > > > you
> > > > > > > > > > could
> > > > > > > > > > > > > come
> > > > > > > > > > > > > > > >> back with
> > > > > > > > > > > > > > > >> > > > some
> > > > > > > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement
> the
> > > > > Connect
> > > > > > > side
> > > > > > > > > > > > > > integration
> > > > > > > > > > > > > > > >> with
> > > > > > > > > > > > > > > >> > > > the new
> > > > > > > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides
> > > > > leveraging on
> > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > > > > "protocol
> > > > > > > > > > > > > > > >> type"
> > > > > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > > > > >> > > > > > did not
> > > > > > > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> Connect
> > > side
> > > > > > > > > > > > implementations.
> > > > > > > > > > > > > > For
> > > > > > > > > > > > > > > >> Streams
> > > > > > > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> integration
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > > >>
> > > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI
> for
> > > your
> > > > > > > analysis
> > > > > > > > > > on
> > > > > > > > > > > > > > Connect.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue,
> Jul
> > > 19,
> > > > > 2022
> > > > > > > at
> > > > > > > > > > 10:48 PM
> > > > > > > > > > > > > > Sagar <
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > sagarmeansocean@gmail.com
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi
> David,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank
> you
> > > for
> > > > > your
> > > > > > > > > > response.
> > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> reason I
> > > > > > > > > > > > > > > >> > > > thought
> > > > > > > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new
> > > scheme
> > > > > is
> > > > > > > that
> > > > > > > > > > even
> > > > > > > > > > > > > > today the
> > > > > > > > > > > > > > > >> > > > connect
> > > > > > > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> WorkerCoordinator
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> extending
> > > from
> > > > > > > > > > > > > > AbstractCoordinator to
> > > > > > > > > > > > > > > >> empower
> > > > > > > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > tasks/connectors.
> > > > > > > The
> > > > > > > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and
> uses the
> > > > > > > metadata()
> > > > > > > > > > method
> > > > > > > > > > > > > by
> > > > > > > > > > > > > > > >> plumbing
> > > > > > > > > > > > > > > >> > > > into
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > JoinGroupRequestProtocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think
> the
> > > > > changes
> > > > > > > to
> > > > > > > > > > support
> > > > > > > > > > > > > > > >> connect would
> > > > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> changes
> > > in
> > > > > > > streams
> > > > > > > > > > mainly
> > > > > > > > > > > > > > because
> > > > > > > > > > > > > > > >> of the
> > > > > > > > > > > > > > > >> > > > > > Client
> > > > > > > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in
> > > both.
> > > > > At an
> > > > > > > > > > > > > implementation
> > > > > > > > > > > > > > > >> level, we
> > > > > > > > > > > > > > > >> > > > > > might
> > > > > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes
> to
> > > get
> > > > > onto
> > > > > > > this
> > > > > > > > > > new
> > > > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > > > >> > > > protocol
> > > > > > > > > > > > > > > >> > > > > > like
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> JoinGroup
> > > > > > > > > > request/response and
> > > > > > > > > > > > > > > >> SyncGroup and
> > > > > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > ConsumerGroupHeartbeat
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc
> > > again on
> > > > > > > similar
> > > > > > > > > > lines
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> streams (or
> > > > > > > > > > > > > > > >> > > > > > there
> > > > > > > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> deviations). I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would
> try to
> > > > > > > perform a
> > > > > > > > > > > > detailed
> > > > > > > > > > > > > > > >> analysis of
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > same
> > > > > > > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate
> > > > > discussion
> > > > > > > > > > thread for
> > > > > > > > > > > > > > that
> > > > > > > > > > > > > > > >> as that
> > > > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread.
> Let
> > > me
> > > > > know
> > > > > > > if
> > > > > > > > > > that
> > > > > > > > > > > > > sounds
> > > > > > > > > > > > > > > >> good to
> > > > > > > > > > > > > > > >> > > > you.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri,
> Jul
> > > 15,
> > > > > > > 2022 at
> > > > > > > > > > 5:47
> > > > > > > > > > > > PM
> > > > > > > > > > > > > > David
> > > > > > > > > > > > > > > >> Jacot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi
> Sagar,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> Thanks for
> > > > > your
> > > > > > > > > > comments.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1)
> Yes.
> > > That
> > > > > > > refers to
> > > > > > > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The
> > > idea
> > > > > is to
> > > > > > > > > > > > transition C
> > > > > > > > > > > > > > from
> > > > > > > > > > > > > > > >> his
> > > > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target
> > > > > assignment
> > > > > > > when
> > > > > > > > > > he
> > > > > > > > > > > > can
> > > > > > > > > > > > > > move
> > > > > > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > > > > > > >> > > > > > When
> > > > > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > > > > assignment
> > > > > > > is
> > > > > > > > > > updated
> > > > > > > > > > > > > and
> > > > > > > > > > > > > > > >> persisted
> > > > > > > > > > > > > > > >> > > > > > with all
> > > > > > > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> partitions
> > > > > even if
> > > > > > > > > they
> > > > > > > > > > are
> > > > > > > > > > > > > not
> > > > > > > > > > > > > > all
> > > > > > > > > > > > > > > >> revoked
> > > > > > > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > > > > assignment
> > > > > > > > > > becomes
> > > > > > > > > > > > the
> > > > > > > > > > > > > > target
> > > > > > > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > optimization
> > > > > to
> > > > > > > avoid
> > > > > > > > > > having
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> write all
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> examples
> > > are
> > > > > > > based on
> > > > > > > > > > the
> > > > > > > > > > > > > > persisted
> > > > > > > > > > > > > > > >> state
> > > > > > > > > > > > > > > >> > > > so I
> > > > > > > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> confusion.
> > > > > Let me
> > > > > > > see
> > > > > > > > > > if I
> > > > > > > > > > > > can
> > > > > > > > > > > > > > > >> improve
> > > > > > > > > > > > > > > >> > > > this in
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3)
> > > Regarding
> > > > > > > Connect,
> > > > > > > > > it
> > > > > > > > > > > > could
> > > > > > > > > > > > > > > >> reuse the
> > > > > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> assignor
> > > if it
> > > > > > > fits in
> > > > > > > > > > the
> > > > > > > > > > > > > > > >> protocol. The
> > > > > > > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > topicid-partitions +
> > > > > > > > > > > > metadata,
> > > > > > > > > > > > > > could
> > > > > > > > > > > > > > > >> > > > Connect
> > > > > > > > > > > > > > > >> > > > > > fit
> > > > > > > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On
> Fri,
> > > Jul
> > > > > 15,
> > > > > > > 2022
> > > > > > > > > at
> > > > > > > > > > 1:55
> > > > > > > > > > > > > PM
> > > > > > > > > > > > > > > >> Sagar <
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi
> > > David,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> Thanks
> > > for
> > > > > the
> > > > > > > KIP.
> > > > > > > > > I
> > > > > > > > > > just
> > > > > > > > > > > > > had
> > > > > > > > > > > > > > > >> minor
> > > > > > > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1)
> In
> > > the
> > > > > > > Assignment
> > > > > > > > > > Error
> > > > > > > > > > > > > > > >> section in
> > > > > > > > > > > > > > > >> > > > Client
> > > > > > > > > > > > > > > >> > > > > > Side
> > > > > > > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> process,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you
> > > > > mentioned
> > > > > > > => `In
> > > > > > > > > > this
> > > > > > > > > > > > > > case,
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > > > > >> > > > > > side
> > > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> error
> > > to the
> > > > > > > group
> > > > > > > > > > > > > > coordinator`.
> > > > > > > > > > > > > > > >> In this
> > > > > > > > > > > > > > > >> > > > > > case are
> > > > > > > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> Assignor
> > > > > > > returning
> > > > > > > > > an
> > > > > > > > > > > > > > > >> AssignmentError
> > > > > > > > > > > > > > > >> > > > that's
> > > > > > > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If
> yes,
> > > do
> > > > > you
> > > > > > > think
> > > > > > > > > > it
> > > > > > > > > > > > > would
> > > > > > > > > > > > > > > >> make sense
> > > > > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2)
> In
> > > the
> > > > > Case
> > > > > > > > > Studies
> > > > > > > > > > > > > > section, I
> > > > > > > > > > > > > > > >> have a
> > > > > > > > > > > > > > > >> > > > > > slight
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> others
> > > have
> > > > > the
> > > > > > > > > same.
> > > > > > > > > > > > > Consider
> > > > > > > > > > > > > > > >> this step:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> When B
> > > > > > > heartbeats,
> > > > > > > > > the
> > > > > > > > > > > > group
> > > > > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> because
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B
> has no
> > > > > > > partitions
> > > > > > > > > to
> > > > > > > > > > > > > > revoke. It
> > > > > > > > > > > > > > > >> > > > persists
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> > > Group
> > > > > > > (epoch=3)
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > A
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > B
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > C
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> > > Target
> > > > > > > > > Assignment
> > > > > > > > > > > > > > (epoch=3)
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > A -
> > > > > > > > > > > > > partitions=[foo-0]
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > B -
> > > > > > > > > > > > > partitions=[foo-2]
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > C -
> > > > > > > > > > > > > partitions=[foo-1]
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> > > Member
> > > > > > > > > Assignment
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > A -
> > > > > > > epoch=2,
> > > > > > > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > B -
> > > > > > > epoch=3,
> > > > > > > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
>    -
> > > C -
> > > > > > > epoch=3,
> > > > > > > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> When C
> > > > > > > heartbeats,
> > > > > > > > > it
> > > > > > > > > > > > > > transitions
> > > > > > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > > > > > > >> > > > > > but
> > > > > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > Here,it's
> > > > > > > mentioned
> > > > > > > > > > that
> > > > > > > > > > > > > > member C
> > > > > > > > > > > > > > > >> can't
> > > > > > > > > > > > > > > >> > > > get
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> based
> > > on the
> > > > > > > > > > description
> > > > > > > > > > > > > > above,
> > > > > > > > > > > > > > > >> it seems
> > > > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> would be
> > > > > better
> > > > > > > to
> > > > > > > > > > remove
> > > > > > > > > > > > it
> > > > > > > > > > > > > > and
> > > > > > > > > > > > > > > >> > > > populate it
> > > > > > > > > > > > > > > >> > > > > > only
> > > > > > > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I
> see
> > > this
> > > > > in a
> > > > > > > lot
> > > > > > > > > of
> > > > > > > > > > > > other
> > > > > > > > > > > > > > > >> places, so
> > > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > > >> > > > > > I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> incorrectly
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > Regarding
> > > > > > > connect ,
> > > > > > > > > it
> > > > > > > > > > > > might
> > > > > > > > > > > > > > be
> > > > > > > > > > > > > > > >> out of
> > > > > > > > > > > > > > > >> > > > scope
> > > > > > > > > > > > > > > >> > > > > > of
> > > > > > > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> what I
> > > > > > > understood it
> > > > > > > > > > would
> > > > > > > > > > > > > > > >> probably be
> > > > > > > > > > > > > > > >> > > > > > running in
> > > > > > > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> even on
> > > the
> > > > > new
> > > > > > > > > > rebalance
> > > > > > > > > > > > > > > >> protocol as it
> > > > > > > > > > > > > > > >> > > > has
> > > > > > > > > > > > > > > >> > > > > > its
> > > > > > > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > Assignors(Eager
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > > > > > > IncrementalCooperative).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> Thanks!
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> Sagar.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On
> Fri,
> > > Jul
> > > > > 15,
> > > > > > > 2022
> > > > > > > > > > at
> > > > > > > > > > > > 5:00
> > > > > > > > > > > > > > PM
> > > > > > > > > > > > > > > >> David
> > > > > > > > > > > > > > > >> > > > Jacot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> Thanks
> > > > > > > Hector! Our
> > > > > > > > > > goal
> > > > > > > > > > > > is
> > > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> move
> > > > > > > > > > > > > > > >> > > > forward
> > > > > > > > > > > > > > > >> > > > > > with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > instead of
> > > > > > > relying
> > > > > > > > > > on
> > > > > > > > > > > > one
> > > > > > > > > > > > > > > >> generic API.
> > > > > > > > > > > > > > > >> > > > For
> > > > > > > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> exact
> > > same
> > > > > > > pattern
> > > > > > > > > > and
> > > > > > > > > > > > > > > >> reuse/share the
> > > > > > > > > > > > > > > >> > > > core
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> server
> > > > > side.
> > > > > > > For
> > > > > > > > > the
> > > > > > > > > > > > > schema
> > > > > > > > > > > > > > > >> registry, I
> > > > > > > > > > > > > > > >> > > > > > think
> > > > > > > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > having a
> > > > > > > tailored
> > > > > > > > > > API to
> > > > > > > > > > > > > do
> > > > > > > > > > > > > > > >> simple
> > > > > > > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> Best,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> David
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On
> > > Fri,
> > > > > Jul
> > > > > > > 15,
> > > > > > > > > > 2022 at
> > > > > > > > > > > > > > 10:22
> > > > > > > > > > > > > > > >> AM Ismael
> > > > > > > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > Three
> > > > > quick
> > > > > > > > > > comments:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> 1.
> > > > > Regarding
> > > > > > > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > > > > > > >> > > > vs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > com.google.re2j.Pattern,
> > > > > > > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > document the
> > > > > > > > > > > > differences
> > > > > > > > > > > > > > in
> > > > > > > > > > > > > > > >> more
> > > > > > > > > > > > > > > >> > > > detail
> > > > > > > > > > > > > > > >> > > > > > before
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > another.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> That
> > > > > said,
> > > > > > > if
> > > > > > > > > > people
> > > > > > > > > > > > > pass
> > > > > > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > semantics
> > > > > > > to be
> > > > > > > > > > > > honored.
> > > > > > > > > > > > > > If
> > > > > > > > > > > > > > > >> we are
> > > > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > should
> > > > > > > consider
> > > > > > > > > > adding
> > > > > > > > > > > > > an
> > > > > > > > > > > > > > > >> overload
> > > > > > > > > > > > > > > >> > > > with
> > > > > > > > > > > > > > > >> > > > > > our own
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > think
> > > > > we'd
> > > > > > > want
> > > > > > > > > to
> > > > > > > > > > > > > expose
> > > > > > > > > > > > > > > >> re2j's at
> > > > > > > > > > > > > > > >> > > > this
> > > > > > > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> 2.
> > > > > Regarding
> > > > > > > > > topic
> > > > > > > > > > > > ids,
> > > > > > > > > > > > > > any
> > > > > > > > > > > > > > > >> major new
> > > > > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> it
> > > and
> > > > > > > should
> > > > > > > > > > handle
> > > > > > > > > > > > the
> > > > > > > > > > > > > > topic
> > > > > > > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > part we
> > > > > > > need to
> > > > > > > > > > > > handle.
> > > > > > > > > > > > > I
> > > > > > > > > > > > > > > >> agree with
> > > > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> ids
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> to
> > > the
> > > > > > > relevant
> > > > > > > > > > > > > protocols
> > > > > > > > > > > > > > > >> that don't
> > > > > > > > > > > > > > > >> > > > > > have it
> > > > > > > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> probably
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > focus
> > > > > on the
> > > > > > > > > > internals
> > > > > > > > > > > > > > versus
> > > > > > > > > > > > > > > >> adding
> > > > > > > > > > > > > > > >> > > > new
> > > > > > > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > (unless
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> we
> > > find
> > > > > that
> > > > > > > > > > adding
> > > > > > > > > > > > new
> > > > > > > > > > > > > > APIs
> > > > > > > > > > > > > > > >> is
> > > > > > > > > > > > > > > >> > > > required
> > > > > > > > > > > > > > > >> > > > > > for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> semantics).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> 3.
> > > I am
> > > > > > > still
> > > > > > > > > not
> > > > > > > > > > sure
> > > > > > > > > > > > > > about
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > powerful for
> > > > > > > > > > configs
> > > > > > > > > > > > to
> > > > > > > > > > > > > be
> > > > > > > > > > > > > > > >> > > > centralized
> > > > > > > > > > > > > > > >> > > > > > in the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> various
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > reasons
> > > > > > > > > > (auditability,
> > > > > > > > > > > > > > > >> visibility,
> > > > > > > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> Similarly, I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> sure
> > > > > about
> > > > > > > > > > > > automatically
> > > > > > > > > > > > > > > >> deleting
> > > > > > > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > recovered. A
> > > > > > > > > good
> > > > > > > > > > > > > property
> > > > > > > > > > > > > > > >> for modern
> > > > > > > > > > > > > > > >> > > > > > systems
> > > > > > > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > unrecoverable
> > > > > > > > > data
> > > > > > > > > > > > loss
> > > > > > > > > > > > > > > >> scenarios.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > Ismael
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> On
> > > Wed,
> > > > > Jul
> > > > > > > 13,
> > > > > > > > > > 2022
> > > > > > > > > > > > at
> > > > > > > > > > > > > > 3:47
> > > > > > > > > > > > > > > >> PM David
> > > > > > > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > Thanks
> > > > > > > > > > Guozhang. My
> > > > > > > > > > > > > > answers
> > > > > > > > > > > > > > > >> are
> > > > > > > > > > > > > > > >> > > > below:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > 1)
> > > > > the
> > > > > > > > > > migration
> > > > > > > > > > > > > path,
> > > > > > > > > > > > > > > >> especially
> > > > > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > flag
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > to
> > > > > > > enable
> > > > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> both
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > protocols /
> > > > > > > > > > rpcs
> > > > > > > > > > > > and
> > > > > > > > > > > > > > old
> > > > > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> same
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > group.
> > > > > > > How
> > > > > > > > > the
> > > > > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> using
> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > new
> > > > > > > protocol
> > > > > > > > > > is
> > > > > > > > > > > > > > something
> > > > > > > > > > > > > > > >> we
> > > > > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > Noted. I
> > > > > > > just
> > > > > > > > > > > > > published
> > > > > > > > > > > > > > a
> > > > > > > > > > > > > > > >> new
> > > > > > > > > > > > > > > >> > > > version
> > > > > > > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > details
> > > > > > > about
> > > > > > > > > > this.
> > > > > > > > > > > > > See
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> Upgrade"
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > and
> > > > > the
> > > > > > > > > > > > > "Compatibility,
> > > > > > > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > > > > > > >> > > > > > and
> > > > > > > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > I
> > > > > have to
> > > > > > > > > think
> > > > > > > > > > > > > through
> > > > > > > > > > > > > > a
> > > > > > > > > > > > > > > >> few cases
> > > > > > > > > > > > > > > >> > > > > > now but
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > mechanism
> > > > > > > > > > should be
> > > > > > > > > > > > > > > >> understandable.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > 2)
> > > > > the
> > > > > > > usage
> > > > > > > > > > of
> > > > > > > > > > > > > topic
> > > > > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> used
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > as
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > part of
> > > > > > > RPCs
> > > > > > > > > > and
> > > > > > > > > > > > > admin
> > > > > > > > > > > > > > > >> client,
> > > > > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> public
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > APIs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > to
> > > > > > > consumers
> > > > > > > > > > yet.
> > > > > > > > > > > > I
> > > > > > > > > > > > > > think
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > client
> > > > > > > to be
> > > > > > > > > > > > > > maintaining
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > leverage
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > on
> > > > > all
> > > > > > > the
> > > > > > > > > > > > augmented
> > > > > > > > > > > > > > > >> existing
> > > > > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> ids;
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > secondly,
> > > > > > > > > > should
> > > > > > > > > > > > we
> > > > > > > > > > > > > > ever
> > > > > > > > > > > > > > > >> consider
> > > > > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > public
> > > > > > > APIs
> > > > > > > > > as
> > > > > > > > > > > > well
> > > > > > > > > > > > > > (both
> > > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > rebalance
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > listener for
> > > > > > > > > > cases
> > > > > > > > > > > > > > like
> > > > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > deletion-and-recreation).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > a)
> > > > > > > Assuming
> > > > > > > > > > that we
> > > > > > > > > > > > > > would
> > > > > > > > > > > > > > > >> include
> > > > > > > > > > > > > > > >> > > > > > converting
> > > > > > > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> related
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > RPCs
> > > > > to
> > > > > > > using
> > > > > > > > > > topic
> > > > > > > > > > > > > ids
> > > > > > > > > > > > > > in
> > > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > > >> > > > KIP,
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > fully
> > > > > > > operate
> > > > > > > > > > with
> > > > > > > > > > > > > topic
> > > > > > > > > > > > > > > >> ids. That
> > > > > > > > > > > > > > > >> > > > > > being
> > > > > > > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> provide
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > the
> > > > > topics
> > > > > > > > > > names in
> > > > > > > > > > > > > > various
> > > > > > > > > > > > > > > >> APIs so
> > > > > > > > > > > > > > > >> > > > > > having a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > seems
> > > > > > > > > > inevitable to
> > > > > > > > > > > > > me.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > b) I
> > > > > don't
> > > > > > > > > have
> > > > > > > > > > a
> > > > > > > > > > > > > strong
> > > > > > > > > > > > > > > >> opinion on
> > > > > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > beyond the
> > > > > > > > > > scope of
> > > > > > > > > > > > > this
> > > > > > > > > > > > > > > >> KIP. I
> > > > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > > > >> > > > > > rather
> > > > > > > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> internals
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > here
> > > > > and
> > > > > > > we
> > > > > > > > > can
> > > > > > > > > > > > > consider
> > > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > Coming
> > > > > > > back to
> > > > > > > > > > > > > Ismael's
> > > > > > > > > > > > > > > >> point about
> > > > > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > > > > > > >> > > > think
> > > > > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> advantage
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > favour of
> > > > > > > it.
> > > > > > > > > > The
> > > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > > >> will
> > > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > the
> > > > > topics
> > > > > > > > > > exists
> > > > > > > > > > > > > before
> > > > > > > > > > > > > > > >> passing
> > > > > > > > > > > > > > > >> > > > them
> > > > > > > > > > > > > > > >> > > > > > into
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > protocol.
> > > > > > > > > > Obviously,
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > coordinator
> > > > > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > not
> > > > > really
> > > > > > > > > have
> > > > > > > > > > a
> > > > > > > > > > > > way
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> reject an
> > > > > > > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > I'm
> > > > > > > agreeing
> > > > > > > > > > with
> > > > > > > > > > > > > > David
> > > > > > > > > > > > > > > >> on all
> > > > > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > > > > >> question:
> > > > > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> necessary
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > deprecate
> > > > > > > > > the
> > > > > > > > > > > > > > subscribe
> > > > > > > > > > > > > > > >> API with
> > > > > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > Pattern
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > while
> > > > > > > just
> > > > > > > > > > > > > documenting
> > > > > > > > > > > > > > > >> that our
> > > > > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected
> by
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > server.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > Since
> > > > > > > the
> > > > > > > > > > > > > incompatible
> > > > > > > > > > > > > > > >> case is a
> > > > > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > overloaded
> > > > > > > > > > > > `String`
> > > > > > > > > > > > > > based
> > > > > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > various
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > invalid
> > > > > > > > > > regexes.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > That
> > > > > could
> > > > > > > > > > work. I
> > > > > > > > > > > > > have
> > > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> look at
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > engines to
> > > > > > > > > > better
> > > > > > > > > > > > > > > >> understand the
> > > > > > > > > > > > > > > >> > > > > > potential
> > > > > > > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > understanding
> > > > > is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > that
> > > > > would
> > > > > > > > > work
> > > > > > > > > > for
> > > > > > > > > > > > > all
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> basic
> > > > > > > > > > > > > > > >> > > > > > regular
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > differences
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > between
> > > > > > > the
> > > > > > > > > two
> > > > > > > > > > are
> > > > > > > > > > > > > > mainly
> > > > > > > > > > > > > > > >> about
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > various
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > wonder
> > > > > > > what
> > > > > > > > > > other
> > > > > > > > > > > > > people
> > > > > > > > > > > > > > > >> think
> > > > > > > > > > > > > > > >> > > > about
> > > > > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > Best,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > David
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > On
> > > > > Tue,
> > > > > > > Jul
> > > > > > > > > 12,
> > > > > > > > > > 2022
> > > > > > > > > > > > > at
> > > > > > > > > > > > > > > >> 11:28 PM
> > > > > > > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > wangguoz@gmail.com
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > Thanks
> > > > > > > > > David!
> > > > > > > > > > I
> > > > > > > > > > > > > think
> > > > > > > > > > > > > > on
> > > > > > > > > > > > > > > >> the high
> > > > > > > > > > > > > > > >> > > > > > level
> > > > > > > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> points we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> need
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > to
> > > > > > > > > concretize
> > > > > > > > > > a
> > > > > > > > > > > > bit
> > > > > > > > > > > > > > more:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > 1)
> > > > > the
> > > > > > > > > > migration
> > > > > > > > > > > > > path,
> > > > > > > > > > > > > > > >> especially
> > > > > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > flag
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > to
> > > > > > > enable
> > > > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> both
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > protocols /
> > > > > > > > > > rpcs
> > > > > > > > > > > > and
> > > > > > > > > > > > > > old
> > > > > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> same
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > group.
> > > > > > > How
> > > > > > > > > the
> > > > > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> using
> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > new
> > > > > > > protocol
> > > > > > > > > > is
> > > > > > > > > > > > > > something
> > > > > > > > > > > > > > > >> we
> > > > > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > 2)
> > > > > the
> > > > > > > usage
> > > > > > > > > > of
> > > > > > > > > > > > > topic
> > > > > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> used
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > as
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > part of
> > > > > > > RPCs
> > > > > > > > > > and
> > > > > > > > > > > > > admin
> > > > > > > > > > > > > > > >> client,
> > > > > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> public
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > APIs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > to
> > > > > > > consumers
> > > > > > > > > > yet.
> > > > > > > > > > > > I
> > > > > > > > > > > > > > think
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > client
> > > > > > > to be
> > > > > > > > > > > > > > maintaining
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > leverage
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > on
> > > > > all
> > > > > > > the
> > > > > > > > > > > > augmented
> > > > > > > > > > > > > > > >> existing
> > > > > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> ids;
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > secondly,
> > > > > > > > > > should
> > > > > > > > > > > > we
> > > > > > > > > > > > > > ever
> > > > > > > > > > > > > > > >> consider
> > > > > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > public
> > > > > > > APIs
> > > > > > > > > as
> > > > > > > > > > > > well
> > > > > > > > > > > > > > (both
> > > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > rebalance
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > listener for
> > > > > > > > > > cases
> > > > > > > > > > > > > > like
> > > > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > deletion-and-recreation).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > I'm
> > > > > > > agreeing
> > > > > > > > > > with
> > > > > > > > > > > > > > David
> > > > > > > > > > > > > > > >> on all
> > > > > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > > > > >> question:
> > > > > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> necessary
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > deprecate
> > > > > > > > > the
> > > > > > > > > > > > > > subscribe
> > > > > > > > > > > > > > > >> API with
> > > > > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > Pattern
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > while
> > > > > > > just
> > > > > > > > > > > > > documenting
> > > > > > > > > > > > > > > >> that our
> > > > > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected
> by
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > server.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > Since
> > > > > > > the
> > > > > > > > > > > > > incompatible
> > > > > > > > > > > > > > > >> case is a
> > > > > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > overloaded
> > > > > > > > > > > > `String`
> > > > > > > > > > > > > > based
> > > > > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > various
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > invalid
> > > > > > > > > > regexes.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > Guozhang
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > On
> > > > > Tue,
> > > > > > > Jul
> > > > > > > > > > 12,
> > > > > > > > > > > > 2022
> > > > > > > > > > > > > > at
> > > > > > > > > > > > > > > >> 5:23 AM
> > > > > > > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > Hi
> > > > > > > Ismael,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > Thanks for
> > > > > > > > > > your
> > > > > > > > > > > > > > > >> feedback. Let
> > > > > > > > > > > > > > > >> > > > me
> > > > > > > > > > > > > > > >> > > > > > answer
> > > > > > > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 1. I
> > > > > > > > > think
> > > > > > > > > > > > it's
> > > > > > > > > > > > > > > >> premature to
> > > > > > > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > deprecation
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > removal
> > > > > > > > > > of the
> > > > > > > > > > > > > > > >> existing group
> > > > > > > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > affects a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > core
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > client
> > > > > > > > > > > > protocol
> > > > > > > > > > > > > > and
> > > > > > > > > > > > > > > >> hence
> > > > > > > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > dependent
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > how
> > > > > > > > > > quickly
> > > > > > > > > > > > > > > >> applications
> > > > > > > > > > > > > > > >> > > > migrate
> > > > > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > That
> > > > > > > makes
> > > > > > > > > > > > sense.
> > > > > > > > > > > > > I
> > > > > > > > > > > > > > will
> > > > > > > > > > > > > > > >> > > > remove it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 2.
> > > > > > > The
> > > > > > > > > KIP
> > > > > > > > > > > > says
> > > > > > > > > > > > > we
> > > > > > > > > > > > > > > >> intend to
> > > > > > > > > > > > > > > >> > > > > > release
> > > > > > > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> made
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > clear
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > why. If
> > > > > > > > > we
> > > > > > > > > > > > added
> > > > > > > > > > > > > > that
> > > > > > > > > > > > > > > >> as a
> > > > > > > > > > > > > > > >> > > > way to
> > > > > > > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> deprecate
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > remove
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > the
> > > > > > > > > group
> > > > > > > > > > > > > > protocol, I
> > > > > > > > > > > > > > > >> also
> > > > > > > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Let me
> > > > > > > > > > explain
> > > > > > > > > > > > my
> > > > > > > > > > > > > > > >> reasoning. As
> > > > > > > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > coordinator
> > > > > > > > > > in
> > > > > > > > > > > > > Java
> > > > > > > > > > > > > > > >> while we
> > > > > > > > > > > > > > > >> > > > > > implement
> > > > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > that
> > > > > > > the
> > > > > > > > > > > > internals
> > > > > > > > > > > > > > will
> > > > > > > > > > > > > > > >> be
> > > > > > > > > > > > > > > >> > > > slightly
> > > > > > > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > model).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > Therefore, I
> > > > > > > > > > > > > wanted
> > > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> tighten
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > coordinator
> > > > > > > > > > to
> > > > > > > > > > > > the
> > > > > > > > > > > > > > new
> > > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > alternative
> > > > > > > > > > > > would
> > > > > > > > > > > > > > be to
> > > > > > > > > > > > > > > >> use a
> > > > > > > > > > > > > > > >> > > > flag
> > > > > > > > > > > > > > > >> > > > > > to do
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > relying
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > on the
> > > > > > > > > > software
> > > > > > > > > > > > > > upgrade.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 3.
> > > > > > > We
> > > > > > > > > > need to
> > > > > > > > > > > > > > flesh
> > > > > > > > > > > > > > > >> out the
> > > > > > > > > > > > > > > >> > > > > > details of
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> sounds
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > like
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > we're
> > > > > > > > > > saying
> > > > > > > > > > > > we
> > > > > > > > > > > > > > will
> > > > > > > > > > > > > > > >> support
> > > > > > > > > > > > > > > >> > > > > > online
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> correct?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > explain
> > > > > > > > > > this
> > > > > > > > > > > > in
> > > > > > > > > > > > > > > >> detail. It
> > > > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > > > >> > > > > > also
> > > > > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > it's
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > easier.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Yes,
> > > > > > > we
> > > > > > > > > will
> > > > > > > > > > > > > support
> > > > > > > > > > > > > > > >> online
> > > > > > > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > a
> > > > > > > group
> > > > > > > > > > using
> > > > > > > > > > > > the
> > > > > > > > > > > > > > old
> > > > > > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > protocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Let me
> > > > > > > > > > briefly
> > > > > > > > > > > > > > explain
> > > > > > > > > > > > > > > >> how that
> > > > > > > > > > > > > > > >> > > > > > will work
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> basically
> > > a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > four
> > > > > > > step
> > > > > > > > > > > > process:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > 1. The
> > > > > > > > > > cluster
> > > > > > > > > > > > > must
> > > > > > > > > > > > > > be
> > > > > > > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> supporting
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > new
> > > > > > > group
> > > > > > > > > > > > > > coordinator.
> > > > > > > > > > > > > > > >> Both the
> > > > > > > > > > > > > > > >> > > > > > old and
> > > > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > support
> > > > > > > > > the
> > > > > > > > > > old
> > > > > > > > > > > > > > > >> protocol and
> > > > > > > > > > > > > > > >> > > > rely
> > > > > > > > > > > > > > > >> > > > > > on the
> > > > > > > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> metadata
> > > so
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > they
> > > > > > > can
> > > > > > > > > > work
> > > > > > > > > > > > > > together.
> > > > > > > > > > > > > > > >> This
> > > > > > > > > > > > > > > >> > > > point
> > > > > > > > > > > > > > > >> > > > > > is an
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> cannot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > do
> > > > > > > this
> > > > > > > > > one
> > > > > > > > > > live
> > > > > > > > > > > > > > > >> because it
> > > > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > > > >> > > > > > require
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> current
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > coordinator
> > > > > > > > > > and
> > > > > > > > > > > > > > > >> starting up the
> > > > > > > > > > > > > > > >> > > > > > new one
> > > > > > > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > unavailabilities.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > 2. The
> > > > > > > > > > cluster's
> > > > > > > > > > > > > > > >> metadata
> > > > > > > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> order
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > to
> > > > > > > enable
> > > > > > > > > > the
> > > > > > > > > > > > new
> > > > > > > > > > > > > > > >> protocol.
> > > > > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > terminated
> > > > > > > > > > > > because
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> old
> > > > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > protocol.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > 3. The
> > > > > > > > > > consumers
> > > > > > > > > > > > > > must be
> > > > > > > > > > > > > > > >> > > > upgraded
> > > > > > > > > > > > > > > >> > > > > > to a
> > > > > > > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> online
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > migration
> > > > > > > > > > (must
> > > > > > > > > > > > > have
> > > > > > > > > > > > > > > >> KIP-792).
> > > > > > > > > > > > > > > >> > > > If
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > Nothing
> > > > > > > > > > must be
> > > > > > > > > > > > > > done at
> > > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > > >> > > > point.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > 4. The
> > > > > > > > > > consumers
> > > > > > > > > > > > > > must
> > > > > > > > > > > > > > > >> be rolled
> > > > > > > > > > > > > > > >> > > > > > with the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> The
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > consumer
> > > > > > > > > > group
> > > > > > > > > > > > is
> > > > > > > > > > > > > > > >> automatically
> > > > > > > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > using
> > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > > > > protocol
> > > > > > > > > > > > > > > >> joins
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > group.
> > > > > > > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > old
> > > > > > > > > > protocol are
> > > > > > > > > > > > > > being
> > > > > > > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > > > > > > >> > > > > > the old
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > new
> > > > > > > one.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Let me
> > > > > > > > > > clarify
> > > > > > > > > > > > all
> > > > > > > > > > > > > > of
> > > > > > > > > > > > > > > >> this in
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 4.
> > > > > > > I am
> > > > > > > > > > happy
> > > > > > > > > > > > > > that we
> > > > > > > > > > > > > > > >> are
> > > > > > > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> subscriptions
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > server,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > but
> > > > > > > it
> > > > > > > > > > seems
> > > > > > > > > > > > > like
> > > > > > > > > > > > > > > >> there
> > > > > > > > > > > > > > > >> > > > could be
> > > > > > > > > > > > > > > >> > > > > > some
> > > > > > > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> compatibility
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > issues.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > Will
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > have a
> > > > > > > > > > > > mechanism
> > > > > > > > > > > > > > for
> > > > > > > > > > > > > > > >> users to
> > > > > > > > > > > > > > > >> > > > > > detect
> > > > > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> their
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > regex
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > before
> > > > > > > > > > > > switching
> > > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> the new
> > > > > > > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > I
> > > > > > > think
> > > > > > > > > > that I
> > > > > > > > > > > > am
> > > > > > > > > > > > > a
> > > > > > > > > > > > > > bit
> > > > > > > > > > > > > > > >> more
> > > > > > > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> point. I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > believe
> > > > > > > > > > that the
> > > > > > > > > > > > > > > >> majority of
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > cases
> > > > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > work
> > > > > > > with
> > > > > > > > > > the
> > > > > > > > > > > > new
> > > > > > > > > > > > > > > >> engine. The
> > > > > > > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> anyway
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > and
> > > > > > > reject
> > > > > > > > > > the
> > > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > > >> if the
> > > > > > > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > migration
> > > > > > > > > > path,
> > > > > > > > > > > > in
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> worst
> > > > > > > > > > > > > > > >> > > > case,
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > joining
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > the
> > > > > > > group
> > > > > > > > > > will
> > > > > > > > > > > > be
> > > > > > > > > > > > > > > >> rejected.
> > > > > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > > > > >> > > > > > should
> > > > > > > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > defence, I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > would
> > > > > > > say.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > One
> > > > > > > way
> > > > > > > > > for
> > > > > > > > > > > > > > customers to
> > > > > > > > > > > > > > > >> > > > validate
> > > > > > > > > > > > > > > >> > > > > > their
> > > > > > > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> upgrading
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> their
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > prod
> > > > > > > would
> > > > > > > > > > be to
> > > > > > > > > > > > > > test
> > > > > > > > > > > > > > > >> them with
> > > > > > > > > > > > > > > >> > > > > > another
> > > > > > > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > could
> > > > > > > be
> > > > > > > > > > done
> > > > > > > > > > > > in a
> > > > > > > > > > > > > > > >> pre-prod
> > > > > > > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > extend the
> > > > > > > > > > > > > > > >> consumer-group tool
> > > > > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > mechanism.
> > > > > > > > > > Would
> > > > > > > > > > > > > > this be
> > > > > > > > > > > > > > > >> > > > enough in
> > > > > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 5.
> > > > > > > > > > Related to
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > last
> > > > > > > > > > > > > > > >> > > > question,
> > > > > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > users
> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > stick
> > > > > > > > > > with the
> > > > > > > > > > > > > > > >> current regex
> > > > > > > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> reasons?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> For
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > example,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > it
> > > > > > > may
> > > > > > > > > be
> > > > > > > > > > > > handy
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> keep using
> > > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > migrations
> > > > > > > > > > > > > simple
> > > > > > > > > > > > > > and
> > > > > > > > > > > > > > > >> then
> > > > > > > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > second
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > step.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > I
> > > > > > > > > understand
> > > > > > > > > > > > your
> > > > > > > > > > > > > > point
> > > > > > > > > > > > > > > >> but I
> > > > > > > > > > > > > > > >> > > > am
> > > > > > > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> users
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > to
> > > > > > > > > actually
> > > > > > > > > > stay
> > > > > > > > > > > > > in
> > > > > > > > > > > > > > > >> this mode.
> > > > > > > > > > > > > > > >> > > > That
> > > > > > > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > simplifying
> > > > > > > > > > the
> > > > > > > > > > > > > > client
> > > > > > > > > > > > > > > >> because
> > > > > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> monitoring
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > the
> > > > > > > > > > metadata on
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> client
> > > > > > > > > > > > > > > >> > > > side. I
> > > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 6.
> > > > > > > When
> > > > > > > > > > we say
> > > > > > > > > > > > > > that
> > > > > > > > > > > > > > > >> the group
> > > > > > > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > responsible
> > > > > for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > storing
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > the
> > > > > > > > > > > > > configurations
> > > > > > > > > > > > > > > >> and that
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> deleted
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> when
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > group is
> > > > > > > > > > > > > deleted.
> > > > > > > > > > > > > > > >> Will a
> > > > > > > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > > > configurations?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > That's
> > > > > > > > > > right.
> > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> > > > configurations
> > > > > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > deleted.
> > > > > > > > > > They go
> > > > > > > > > > > > > > > >> together.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 7.
> > > > > > > Will
> > > > > > > > > > the
> > > > > > > > > > > > > > choice to
> > > > > > > > > > > > > > > >> store
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > make
> > > > > it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > harder
> > > > > > > > > to
> > > > > > > > > > list
> > > > > > > > > > > > > all
> > > > > > > > > > > > > > > >> cluster
> > > > > > > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > I
> > > > > > > don't
> > > > > > > > > > think
> > > > > > > > > > > > so.
> > > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> cluster
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > configs.
> > > > > > > > > If
> > > > > > > > > > you
> > > > > > > > > > > > > > want to
> > > > > > > > > > > > > > > >> know
> > > > > > > > > > > > > > > >> > > > all
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> have
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > to
> > > > > > > ask all
> > > > > > > > > > the
> > > > > > > > > > > > > group
> > > > > > > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > > > > > > >> > > > > > You
> > > > > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> metadata
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> log
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > for
> > > > > > > > > > instance.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 8.
> > > > > > > How
> > > > > > > > > > would
> > > > > > > > > > > > > > someone
> > > > > > > > > > > > > > > >> > > > configure a
> > > > > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > consumers?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Have
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > we
> > > > > > > > > > considered
> > > > > > > > > > > > > > > >> allowing the
> > > > > > > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > > > > > Alternatively,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > configs
> > > > > > > > > > could
> > > > > > > > > > > > be
> > > > > > > > > > > > > > > >> decoupled
> > > > > > > > > > > > > > > >> > > > from
> > > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Yes.
> > > > > > > The
> > > > > > > > > > group
> > > > > > > > > > > > > will
> > > > > > > > > > > > > > be
> > > > > > > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> However,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > the
> > > > > > > > > configs
> > > > > > > > > > will
> > > > > > > > > > > > > be
> > > > > > > > > > > > > > > >> lost after
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > passes.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 9.
> > > > > > > Will
> > > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > > > > > > >> > > > > > method
> > > > > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> take
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > > > > > > >> > > > do
> > > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> overload?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > That's a
> > > > > > > > > > very
> > > > > > > > > > > > > group
> > > > > > > > > > > > > > > >> question. I
> > > > > > > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > > > > >> > > > > > fully
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > we
> > > > > > > plan to
> > > > > > > > > > use,
> > > > > > > > > > > > it
> > > > > > > > > > > > > > > >> might be
> > > > > > > > > > > > > > > >> > > > better
> > > > > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > overload
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > which
> > > > > > > > > takes
> > > > > > > > > > a
> > > > > > > > > > > > > > string.
> > > > > > > > > > > > > > > >> We would
> > > > > > > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > validation.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > During the
> > > > > > > > > > > > > > migration, I
> > > > > > > > > > > > > > > >> think
> > > > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> toString
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > regex
> > > > > > > and
> > > > > > > > > > use
> > > > > > > > > > > > it.
> > > > > > > > > > > > > > That
> > > > > > > > > > > > > > > >> should
> > > > > > > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > cases.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 10.
> > > > > > > I
> > > > > > > > > > agree
> > > > > > > > > > > > with
> > > > > > > > > > > > > > > >> Justine
> > > > > > > > > > > > > > > >> > > > that we
> > > > > > > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> reason
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > switch
> > > > > > > > > to
> > > > > > > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > > > > > > >> > > > > > from the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> mean
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > can
> > > > > > > > > > switch the
> > > > > > > > > > > > > > > >> default for
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > feature
> > > > > > > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > > > metadata.version
> > > > > > > > > > > > > > once
> > > > > > > > > > > > > > > >> we
> > > > > > > > > > > > > > > >> > > > want to
> > > > > > > > > > > > > > > >> > > > > > make
> > > > > > > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > My
> > > > > > > plan
> > > > > > > > > was
> > > > > > > > > > to
> > > > > > > > > > > > use
> > > > > > > > > > > > > > that
> > > > > > > > > > > > > > > >> feature
> > > > > > > > > > > > > > > >> > > > > > flag
> > > > > > > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > development
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > phase. I
> > > > > > > > > > should
> > > > > > > > > > > > > not
> > > > > > > > > > > > > > have
> > > > > > > > > > > > > > > >> > > > mentioned
> > > > > > > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> use
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > an
> > > > > > > > > internal
> > > > > > > > > > > > config
> > > > > > > > > > > > > > for
> > > > > > > > > > > > > > > >> it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 11.
> > > > > > > Some
> > > > > > > > > > of
> > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> protocol APIs
> > > > > > > > > > > > > > > >> > > > > > don't
> > > > > > > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > would
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > good to
> > > > > > > > > > add
> > > > > > > > > > > > that
> > > > > > > > > > > > > > for
> > > > > > > > > > > > > > > >> > > > consistency.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Noted.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 12.
> > > > > > > It
> > > > > > > > > is
> > > > > > > > > > a
> > > > > > > > > > > > bit
> > > > > > > > > > > > > > odd
> > > > > > > > > > > > > > > >> that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> ConsumerGroupHeartbeat
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> Group"
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > even
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > though
> > > > > > > > > it
> > > > > > > > > > > > seems
> > > > > > > > > > > > > > to do
> > > > > > > > > > > > > > > >> more
> > > > > > > > > > > > > > > >> > > > than
> > > > > > > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > I
> > > > > > > agree.
> > > > > > > > > > This is
> > > > > > > > > > > > > > how the
> > > > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > require
> > > > > > > > > > "Read
> > > > > > > > > > > > > > Group" to
> > > > > > > > > > > > > > > >> join a
> > > > > > > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> changing
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> this
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > but I
> > > > > > > am
> > > > > > > > > not
> > > > > > > > > > > > sure
> > > > > > > > > > > > > > that
> > > > > > > > > > > > > > > >> it is
> > > > > > > > > > > > > > > >> > > > worth
> > > > > > > > > > > > > > > >> > > > > > it.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 13.
> > > > > > > How
> > > > > > > > > is
> > > > > > > > > > > > topic
> > > > > > > > > > > > > > > >> recreation
> > > > > > > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > group
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > protocol?
> > > > > > > > > > It
> > > > > > > > > > > > > > would be
> > > > > > > > > > > > > > > >> good to
> > > > > > > > > > > > > > > >> > > > > > have a
> > > > > > > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Noted.
> > > > > > > > > From
> > > > > > > > > > a
> > > > > > > > > > > > > > protocol
> > > > > > > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > topic
> > > > > > > id
> > > > > > > > > so
> > > > > > > > > > it
> > > > > > > > > > > > > will
> > > > > > > > > > > > > > > >> treat it
> > > > > > > > > > > > > > > >> > > > like a
> > > > > > > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> The
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > only
> > > > > > > issue
> > > > > > > > > > is
> > > > > > > > > > > > that
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > IDs
> > > > > > > so the
> > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > > >> would
> > > > > > > > > > > > > > > >> > > > reuse the
> > > > > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > that
> > > > > > > we
> > > > > > > > > > should
> > > > > > > > > > > > > > update
> > > > > > > > > > > > > > > >> those
> > > > > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > > > > >> > > > > > as well
> > > > > > > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> consistent
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> end
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > to
> > > > > > > end.
> > > > > > > > > That
> > > > > > > > > > > > would
> > > > > > > > > > > > > > > >> strengthen
> > > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 14.
> > > > > > > The
> > > > > > > > > > KIP
> > > > > > > > > > > > > > mentions
> > > > > > > > > > > > > > > >> we will
> > > > > > > > > > > > > > > >> > > > > > write the
> > > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> Even
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > though
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > this is
> > > > > > > > > an
> > > > > > > > > > > > > > > >> implementation
> > > > > > > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > module
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > for
> > > > > > > it?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Yes.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 15.
> > > > > > > Do
> > > > > > > > > we
> > > > > > > > > > > > have a
> > > > > > > > > > > > > > > >> scalability
> > > > > > > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > members
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > new
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > group
> > > > > > > > > > protocol
> > > > > > > > > > > > > can
> > > > > > > > > > > > > > > >> support?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > We
> > > > > > > don't
> > > > > > > > > > have
> > > > > > > > > > > > > > numbers
> > > > > > > > > > > > > > > >> at the
> > > > > > > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> support
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> 1000s
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > of
> > > > > > > members
> > > > > > > > > > per
> > > > > > > > > > > > > > group.
> > > > > > > > > > > > > > > >> We will
> > > > > > > > > > > > > > > >> > > > > > measure
> > > > > > > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > implementation.
> > > > > > > > > > > > > Note
> > > > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > > > >> > > > might
> > > > > > > > > > > > > > > >> > > > > > have
> > > > > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > road
> > > > > > > (e.g.
> > > > > > > > > > > > offset
> > > > > > > > > > > > > > > >> commits).
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > 16.
> > > > > > > Did
> > > > > > > > > we
> > > > > > > > > > > > > > consider
> > > > > > > > > > > > > > > >> having
> > > > > > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > > > > > SubscribedTopicNames
> > > > > > > > > > > > > > > >> in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> idea
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > since we
> > > > > > > > > > have
> > > > > > > > > > > > to
> > > > > > > > > > > > > > > >> resolve the
> > > > > > > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> same
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > the
> > > > > > > > > topic
> > > > > > > > > > > > name?
> > > > > > > > > > > > > > The
> > > > > > > > > > > > > > > >> > > > difference
> > > > > > > > > > > > > > > >> > > > > > is that
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > efficient
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > whereas
> > > > > > > > > > > > sending
> > > > > > > > > > > > > > the
> > > > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > > > >> > > > > > is less
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > Furthermore,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > delete
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > recreation is
> > > > > > > > > > > > > > easier
> > > > > > > > > > > > > > > >> to
> > > > > > > > > > > > > > > >> > > > handle
> > > > > > > > > > > > > > > >> > > > > > if we
> > > > > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > The
> > > > > > > idea
> > > > > > > > > > was to
> > > > > > > > > > > > > > > >> consolidate the
> > > > > > > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> both
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > paths
> > > > > > > but
> > > > > > > > > I
> > > > > > > > > > do
> > > > > > > > > > > > > agree
> > > > > > > > > > > > > > > >> with your
> > > > > > > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > ids
> > > > > > > may be
> > > > > > > > > > > > better
> > > > > > > > > > > > > > here
> > > > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also,
> I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > suppose
> > > > > > > > > > that we
> > > > > > > > > > > > > may
> > > > > > > > > > > > > > > >> allow
> > > > > > > > > > > > > > > >> > > > users to
> > > > > > > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > future
> > > > > > > > > > because
> > > > > > > > > > > > > that
> > > > > > > > > > > > > > is
> > > > > > > > > > > > > > > >> the only
> > > > > > > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > re-creation.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > Best,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > David
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > On
> > > > > > > Tue,
> > > > > > > > > Jul
> > > > > > > > > > 12,
> > > > > > > > > > > > > > 2022 at
> > > > > > > > > > > > > > > >> 1:38 PM
> > > > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > djacot@confluent.io>
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> >
> > > wrote:
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > Hi
> > > > > > > > > > Justine,
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > Thanks
> > > > > > > > > for
> > > > > > > > > > > > your
> > > > > > > > > > > > > > > >> comments.
> > > > > > > > > > > > > > > >> > > > Please
> > > > > > > > > > > > > > > >> > > > > > find
> > > > > > > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > -
> > > > > > > Yes,
> > > > > > > > > > the new
> > > > > > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > > >> > > > relies on
> > > > > > > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> exception
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> of the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > > > topic
> > > > > > > > > > names
> > > > > > > > > > > > > based
> > > > > > > > > > > > > > in
> > > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> sure
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > if
> > > > > > > using
> > > > > > > > > > topic
> > > > > > > > > > > > > > names
> > > > > > > > > > > > > > > >> is the
> > > > > > > > > > > > > > > >> > > > > > right call
> > > > > > > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> about
> > > it
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >
> > > > >
> > > > > a
> > > > > > > little
> > > > > > > > > > more.
> > > > > > > > > > > > > > > >> Obviously, the
> > > > > > > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > -- Guozhang
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

3. ConsumerGroupMetadata is already defined in the consumer [1]. I
realized that we pass it in the current assignor so I thought that we
could reuse it. The alternative would be to add the same information
in a class which includes the error and the member assignment as well.

[1] https://github.com/apache/kafka/blob/99b9b3e84f4e98c3f07714e1de6a139a004cbc5b/clients/src/main/java/org/apache/kafka/clients/consumer/ConsumerGroupMetadata.java

Best,
David

On Mon, Oct 3, 2022 at 6:15 AM Guozhang Wang <wa...@gmail.com> wrote:
>
> 1. Yeah that makes sense. I originally concerned if there's any consistency
> issues if the HB request with encoded topic IDs and the GPA/GIA RPCs with
> encoded topic IDs use mappings from two different brokers, and one of them
> is stale, but after thinking twice about it I think as long as we handle
> e.g. `unknown topic ids` from HB responses then we should be fine.
>
> 3. Ack, in that case I think we also need to define "ConsumerGroupMetadata"
> which is passed in `onAssignment`. We have a schema defined in the
> persistent storage for different purposes, but not in the consumer APIs
> yet. I think that would just include a single "Map<String, Metadata>"?
>
> On Fri, Sep 30, 2022 at 5:51 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hey Guozhang,
> >
> > Thanks for your comments.
> >
> > 1. That's a very good question. I think that it is better to separate
> > the two concerns here. In my opinion, the GPA should only be used to
> > feed the assignor. I see the GPA as a way to call a remote function so
> > the GPA provides all its inputs to ensure that the assignor does not
> > depend on any local state. Knowing this, I would not use the GPA to
> > update the local state either. That is not its purpose. The consumer
> > is responsible to update its local metadata cache either when it gets
> > an unknown topic in the user-provided subscription or in the
> > coordinator-provided assignment. The consumer has to do this anyway to
> > get all topic metadata (e.g. leaders).
> >
> > 2. Interesting thought. Handling the translation is definitely
> > possible. I wonder if there are cases where having the topic ids is
> > beneficial. I am thinking about when a topic is deleted and recreated
> > right away. In this case, the current member assignment would still
> > have the old topic id and we would not be able to resolve it anymore.
> > I suppose that we should directly remove it from the current
> > assignment on the server side in this case and only add the new one.
> > As I first thought, that seems OK. The major downside is that the user
> > of the assignor would not know that the topic is different from what
> > it may have cached locally.
> >
> > We could say that we don't expose topic ids externally for now in
> > order to stay consistent with the other interfaces but it would be
> > annoying to add it back if we want to expose topic ids in the future.
> > I personally believe that we will have to expose them at some point.
> > For instance, in the rebalance listener, it would be useful to know
> > that a topic has been deleted and recreated. The only way to do this
> > is to know the topic id. Let me think a little more about this and
> > come back to you.
> >
> > 3. Sorry for this one. I was a typo. PartitionAssingor#assign should
> > return GroupAssignment.
> >
> > Best,
> > David
> >
> > On Thu, Sep 29, 2022 at 6:06 PM Guozhang Wang <wa...@gmail.com> wrote:
> > >
> > > Thanks David, I read through the updated wiki, and I have a couple
> > > follow-up questions:
> > >
> > > 1. It seems we would have two potential sources to update the consumer's
> > > internally maintained topic name -> topic ID mapping, which is used for
> > > encoding subscribed topic Ids in the HB request, and offset commit / list
> > > offset requests, etc. One is from metadata responses, and another one is
> > > from the GPA (group prepare assignment). On the other hand, the assignor
> > > APIs directly take parsed AssignmentMemberSpec and
> > AssignmentTopicMetadata
> > > from GPA responses directly instead of trying to consult the consumer's
> > own
> > > cached mapping. So just to clarify, consumers would also update its
> > > internal mapping from the GPA responses as well, right?
> > >
> > > 2. I'm wondering if it's necessary to expose the topic IDs to the
> > assignor
> > > developers. What they need to do at the moment is to take the mapping
> > from
> > > AssignmentTopicMetadata, and translate their decided assignment into
> > topic
> > > IDs when returning the MemberAssignment. We can instead just let consumer
> > > to do such translation for them internally and still let the assignor
> > > developers to only deal with the topic names. Of course that means the
> > > client side assignor APIs would be a bit further diverged from the
> > > broker-side assignor such that:
> > >
> > > a) Topic metadata would be encoded just as String -> num.partitions;
> > > b) AssignmentMemberSpec contains subscribedTopicNames and topicPartitions
> > > as target assignments;
> > > c) Returned MemberAssignment contains list of topicPartitions instead of
> > > topicIdPartitions.
> > >
> > > Internally, consumers would convert the IDs to names when calling the
> > > assign() function, and convert them back when sending the request; and
> > > similarly it would convert the IDs to names when calling onAssignment()
> > as
> > > well.
> > >
> > > In short, I'm wondering if we should bite the bullet for the users,
> > eating
> > > more complexities internally than pushing that to users.
> > >
> > > 3. This is a nit one: the return type of `Assignment
> > assign(AssignmentSpec
> > > assignmentSpec);` seems not defined in the doc. I think it would just
> > > include two fields, as:
> > >
> > > class Assignment {
> > >         GroupAssignment groupAssignment;
> > >
> > >         ConsumerGroupMetadata consumerGroupMetadata;
> > >     }
> > >
> > > Is that right?
> > >
> > > Guozhang
> > >
> > > On Thu, Sep 29, 2022 at 1:26 AM David Jacot <djacot@confluent.io.invalid
> > >
> > > wrote:
> > >
> > > > Hi Guozhang,
> > > >
> > > > Thanks for your input.
> > > >
> > > > > 1) in HB request, when clients send the subscription, they are still
> > > > > sending them in terms of topic names. That is, the consumer would
> > not try
> > > > > to keep all the mappings from topic names to topic ids for the
> > subscribed
> > > > > topics.
> > > >
> > > > This is a mistake of mine. The heartbeat request should use Topic Ids.
> > > > I forgot to change it. The idea is that the client will get the
> > > > metadata first to translate the topic names to topic ids and thus also
> > > > have the mapping in its local metadata cache.
> > > >
> > > > > 2) in HB responses, the assignment's topics are encoded as IDs
> > already.
> > > > And
> > > > > fetch requests can directly rely on that since they just need topic
> > IDs
> > > > and
> > > > > do not care what actual topic names are.
> > > >
> > > > That's right.
> > > >
> > > > > 3) in prepare-assignment responses, we return subscriptions as topic
> > IDs
> > > > as
> > > > > well.
> > > >
> > > > That's right.
> > > >
> > > > > In a word, the consumer does not try to maintain the mapping of topic
> > > > names
> > > > > to IDs at all, and even if they do, they will not expose it in e.g.
> > > > > assignor APIs. But for client-side assignors like Streams, if they
> > do not
> > > > > maintain the mapping from topic IDs to topic names, certain logic
> > such as
> > > > > co-partitioning indeed cannot be executed.
> > > >
> > > > As I said, the consumer will have the mapping in its local metadata
> > > > cache. It needs to get the partition metadata anyway and the topic id
> > > > is part of it. However, I do agree that this mapping is not exposed.
> > > >
> > > > > So I think we'd need to consider 1) use topic names in subscriptions
> > in
> > > > the
> > > > > prepare-assignment responses, 2) use topic Names in client-side
> > > > > PartitionAssignor#GroupMember#subscribedTopics and
> > > > > PartitionAssignor#TopicMetadata#topicName.
> > > >
> > > > 1) makes sense to me. I think that we could actually provide both
> > > > names and ids. Regarding 2), I think that I am slightly in favor of
> > > > using both names and ids as well. Let me try to update the protocol
> > > > and the interface towards this.
> > > >
> > > > Thanks,
> > > > David
> > > >
> > > > On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > > >
> > > > > Hello Jun/David,
> > > > >
> > > > > Just to chime in here for Jun's comment #50, today Streams does not
> > have
> > > > a
> > > > > way to map topic names to topic IDs, and in its topology description
> > > > topics
> > > > > are encoded with names.
> > > > >
> > > > > What I originally thought was that, on the consumer side, we'd need
> > to
> > > > > maintain this mapping from topic names to topic IDs anyways, and
> > Streams
> > > > > should leverage on the consumer's functionality to get the
> > corresponding
> > > > > topic names for the given topic IDs. But when I read through the RPC
> > > > again
> > > > > I realized it may not be the case, since:
> > > > >
> > > > > 1) in HB request, when clients send the subscription, they are still
> > > > > sending them in terms of topic names. That is, the consumer would
> > not try
> > > > > to keep all the mappings from topic names to topic ids for the
> > subscribed
> > > > > topics.
> > > > > 2) in HB responses, the assignment's topics are encoded as IDs
> > already.
> > > > And
> > > > > fetch requests can directly rely on that since they just need topic
> > IDs
> > > > and
> > > > > do not care what actual topic names are.
> > > > > 3) in prepare-assignment responses, we return subscriptions as topic
> > IDs
> > > > as
> > > > > well.
> > > > >
> > > > > In a word, the consumer does not try to maintain the mapping of topic
> > > > names
> > > > > to IDs at all, and even if they do, they will not expose it in e.g.
> > > > > assignor APIs. But for client-side assignors like Streams, if they
> > do not
> > > > > maintain the mapping from topic IDs to topic names, certain logic
> > such as
> > > > > co-partitioning indeed cannot be executed.
> > > > >
> > > > > So I think we'd need to consider 1) use topic names in subscriptions
> > in
> > > > the
> > > > > prepare-assignment responses, 2) use topic Names in client-side
> > > > > PartitionAssignor#GroupMember#subscribedTopics and
> > > > > PartitionAssignor#TopicMetadata#topicName.
> > > > >
> > > > > On Wed, Sep 28, 2022 at 1:26 AM David Jacot
> > <djacot@confluent.io.invalid
> > > > >
> > > > > wrote:
> > > > >
> > > > > > Hi Jun,
> > > > > >
> > > > > > Thanks for your comments.
> > > > > >
> > > > > > 48. Okay. There is a typo in the text:
> > > > > > > Note that the persisted target assignment of B already includes
> > foo-2
> > > > > > here
> > > > > > but B does not get it yet because foo-2 is not free.
> > > > > >
> > > > > > It should be:
> > > > > > > Note that the persisted member assignment of B already includes
> > foo-2
> > > > > > here
> > > > > > but B does not get it yet because foo-2 is not free.
> > > > > >
> > > > > > The examples are showing the persisted state in the
> > __consumer_offsets
> > > > > > topics. When a member transitions to the next epoch, we directly
> > > > > > persist the whole target assignment in order to reduce the number
> > of
> > > > > > writes and we manage dependencies in memory. So "B - epoch=2,
> > > > > > partitions=[foo-2]" is actually correct from a persisted record
> > > > > > perspective. However, the in-memory state would be different
> > because
> > > > > > foo-2 is not yet available as you pointed out. The in-memory state
> > > > > > would like as follow:
> > > > > >
> > > > > > Group (epoch=2)
> > > > > > A
> > > > > > B
> > > > > > Target Assignment (epoch=2)
> > > > > > A - partitions=[foo-0, foo-1]
> > > > > > B - partitions=[foo-2]
> > > > > > Member Assignment
> > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending
> > partitions=[]
> > > > > > B - epoch=2, partitions=[], pending partitions=[foo-2]
> > > > > >
> > > > > > I suppose that the confusion comes from this. I will go and convert
> > > > > > all the examples to their in-memory representation in order to
> > reduce
> > > > > > the confusion.
> > > > > >
> > > > > > 50. Okay. I have to admit that I am not so familiar with the
> > internals
> > > > > > of KStreams. If I understand correctly, you're saying that KStreams
> > > > > > may not have topic ids in its topology so topic names are missing
> > here
> > > > > > to match the topics. I was hoping that KStreams would have topic
> > ids
> > > > > > in the topology as well. If that is not the case, we can also add
> > > > > > topic names on our side. I will check this.
> > > > > >
> > > > > > 53. It seems that I was not clear. Sorry for that. I did not mean
> > to
> > > > > > say that MemberAssignment.metadata is not used by the group
> > > > > > coordinator. Let me re-explain the flow.
> > > > > >
> > > > > > PartitionAssignor.assign() returns an Assignment. Assignment
> > contains
> > > > > > MemberAssignments where each member has MemberAssignment.metadata.
> > The
> > > > > > Assignment is installed by the client with the
> > > > > > ConsumerGroupInstallAssignment API and this overrides the current
> > > > > > target assignment in the coordinator. Then the group coordinator
> > uses
> > > > > > it to reconcile members. Therefore the target assignment will
> > > > > > eventually become the current member assignment and it will be
> > > > > > delivered to the member, including the original
> > > > > > MemberAssignment.metadata defines by the assignor.
> > > > > >
> > > > > > I hope that this helps to clarify how MemberAssignment.metadata is
> > > > > > used afterwards.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <jun@confluent.io.invalid
> > >
> > > > wrote:
> > > > > > > Hi, David,
> > > > > > >
> > > > > > > Thanks for the reply. A few more follow up questions.
> > > > > > >
> > > > > > > 48. Hmm, so member assignment is what the client sees. Then, I
> > am not
> > > > > > sure
> > > > > > > about the following. The text says B cannot get foo-2 yet. But in
> > > > Member
> > > > > > > Assignment, B has foo-2.
> > > > > > >
> > > > > > > At this point B can transition to epoch 2 but cannot get foo-2
> > until
> > > > A
> > > > > > > revokes it.
> > > > > > >
> > > > > > > Note that the persisted target assignment of B already includes
> > foo-2
> > > > > > here
> > > > > > > but B does not get it yet because foo-2 is not free.
> > > > > > >
> > > > > > > Group (epoch=2)
> > > > > > > A
> > > > > > > B
> > > > > > > Target Assignment (epoch=2)
> > > > > > > A - partitions=[foo-0, foo-1]
> > > > > > > B - partitions=[foo-2]
> > > > > > > Member Assignment
> > > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > > > > > B - epoch=2, partitions=[foo-2]
> > > > > > >
> > > > > > >
> > > > > > > 50. Hmm, I thought KStreams doesn't use range assignor all the
> > time.
> > > > It
> > > > > > > does the assignment based on the topology. For example, if you
> > are
> > > > > > joining
> > > > > > > two topics A and B and deriving the join result in a 3rd topic C,
> > > > which
> > > > > > > will be aggregated. You want to use range to co-locate the
> > partitions
> > > > > > from
> > > > > > > topic A and B. However, for topic C, the partition could
> > potentially
> > > > be
> > > > > > > assigned in round robin. Without the topic name, KStreams won't
> > know
> > > > > > which
> > > > > > > topics need to be co-located.
> > > > > > >
> > > > > > > 53. If MemberAssignment.metadata is not used by the group
> > > > coordinator,
> > > > > > > where is it being used?
> > > > > > >
> > > > > > > Thanks,
> > > > > > >
> > > > > > > Jun
> > > > > > >
> > > > > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > > > > > <ja...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hi David,
> > > > > > > >
> > > > > > > > Thanks for the response. Replied below:
> > > > > > > >
> > > > > > > > > I don't have a strong opinion on this one. However, it seems
> > that
> > > > > > this
> > > > > > > > could be a pandora box if our subset turns out to not be
> > sufficient
> > > > > > > > for some existing users. We will have to add more to it and the
> > > > > > > > compatibility story between versions could be a real
> > nightmare. One
> > > > > > > > thing that we could do for sure is disabling all unicode
> > groups in
> > > > the
> > > > > > > > library. That seems possible and would already reduce the
> > scope.
> > > > > > > >
> > > > > > > > Changes to the regex syntax seems not a major problem to me. We
> > > > would
> > > > > > bump
> > > > > > > > the protocol version as we usually do. Actually that begs the
> > same
> > > > > > question
> > > > > > > > about RE2. How stable is the syntax and does it differ
> > depending
> > > > on the
> > > > > > > > version of the library? In general, I do agree that we'd be
> > better
> > > > off
> > > > > > not
> > > > > > > > having to manage the syntax ourselves if we can convince
> > ourselves
> > > > that
> > > > > > > > compatibility won't be a problem and the client won't
> > eventually
> > > > need
> > > > > > to
> > > > > > > > have access to it.
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > > Jason
> > > > > > > >
> > > > > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi Luke,
> > > > > > > > >
> > > > > > > > > Thanks for your comments.
> > > > > > > > >
> > > > > > > > > Your suggestion makes sense. Jun's made the same one. We are
> > also
> > > > > > > > > discussing whether we should directly return the group state
> > in
> > > > the
> > > > > > > > > response and drop the prepare assignment api. It seems that
> > the
> > > > > > > > > current consensus is to keep it though. If that stays like
> > this,
> > > > I
> > > > > > > > > will apply your suggestion.
> > > > > > > > >
> > > > > > > > > Thanks,
> > > > > > > > > David
> > > > > > > > >
> > > > > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <
> > showuon@gmail.com>
> > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Hi David,
> > > > > > > > > >
> > > > > > > > > > Thanks for your patience for answering all these questions!
> > > > > > > > > >
> > > > > > > > > > One more question from me:
> > > > > > > > > > 1. When the selected consumer is going to be in client-side
> > > > > > assigning,
> > > > > > > > we
> > > > > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> > > > > > wondering if
> > > > > > > > we
> > > > > > > > > > could use a flag to indicate the assignment is required.
> > The
> > > > > > rationale
> > > > > > > > is
> > > > > > > > > > when returning error, we lose a chance to catch up with the
> > > > latest
> > > > > > > > target
> > > > > > > > > > assignment or revoke partitions. Since we expected the
> > record
> > > > > > fetching
> > > > > > > > is
> > > > > > > > > > not blocked by the rebalance/assignment computing, we might
> > > > want to
> > > > > > > > allow
> > > > > > > > > > the members to update the member assignment in each
> > heartbeat.
> > > > > > > > > >
> > > > > > > > > > Luke
> > > > > > > > > >
> > > > > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > > > > > <ja...@confluent.io.invalid>
> > > > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hey David,
> > > > > > > > > > >
> > > > > > > > > > > Thanks for the response.
> > > > > > > > > > >
> > > > > > > > > > > > Do you think that we could have a case where two
> > members
> > > > with
> > > > > > the
> > > > > > > > > same
> > > > > > > > > > > static id are active?
> > > > > > > > > > >
> > > > > > > > > > > I was just thinking about the handoff when we spin up a
> > new
> > > > node
> > > > > > > > > before the
> > > > > > > > > > > old one completes shutdown. It's probably ok to fence
> > the old
> > > > > > member
> > > > > > > > > > > abruptly. I do recall we had a few issues with this
> > handoff
> > > > in
> > > > > > the
> > > > > > > > > past. It
> > > > > > > > > > > was due to the fact that the replaced node might have
> > been
> > > > the
> > > > > > > > leader.
> > > > > > > > > > > Since we've removed the leader notion, perhaps this kind
> > of
> > > > issue
> > > > > > > > won't
> > > > > > > > > > > resurface.
> > > > > > > > > > >
> > > > > > > > > > > In regard to the RE2 library. I understand the intent to
> > pass
> > > > > > this
> > > > > > > > > through
> > > > > > > > > > > as a string without any validation so that we don't need
> > the
> > > > > > > > > dependence on
> > > > > > > > > > > the client. I do wonder if that will be restrictive in
> > > > practice.
> > > > > > For
> > > > > > > > > > > example, if a regex changes on the client side, would we
> > > > need to
> > > > > > > > > validate
> > > > > > > > > > > against the current assignment to see which partitions to
> > > > > > continue
> > > > > > > > > > > fetching? Or would we instead invalidate the complete
> > current
> > > > > > > > > assignment?
> > > > > > > > > > > Perhaps this is a rare case and we can do the latter.
> > There
> > > > > > could be
> > > > > > > > > more
> > > > > > > > > > > situations though.
> > > > > > > > > > >
> > > > > > > > > > > As an alternative, I wonder if we could define a
> > minimalistic
> > > > > > regex
> > > > > > > > > syntax
> > > > > > > > > > > (a subset of RE2 say) that could map more easily to regex
> > > > support
> > > > > > > > that
> > > > > > > > > > > already exists in clients? For example, in this
> > scenario, we
> > > > are
> > > > > > > > > matching
> > > > > > > > > > > topics, so we only need to be able to represent topic
> > > > characters
> > > > > > and
> > > > > > > > > not
> > > > > > > > > > > the whole smorgasbord of whitespace and unicode options.
> > > > Would it
> > > > > > > > make
> > > > > > > > > > > sense to have a reduced syntax which includes only
> > standard
> > > > topic
> > > > > > > > > > > characters, wildcards, character classes, and
> > repetitions?
> > > > > > > > > > >
> > > > > > > > > > > Best,
> > > > > > > > > > > Jason
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > > > > > wangguoz@gmail.com>
> > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > > Hi David, thanks for all the detailed explanations. I
> > think
> > > > > > they
> > > > > > > > all
> > > > > > > > > make
> > > > > > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > > > > > >
> > > > > > > > > > > > > I don't really see the benefits here because server
> > side
> > > > > > > > assignors
> > > > > > > > > > > > don't have metadata at all. They only assign
> > > > topic-partitions.
> > > > > > They
> > > > > > > > > > > > are not supposed to generate metadata nor to receive
> > > > metadata
> > > > > > from
> > > > > > > > > the
> > > > > > > > > > > > members.
> > > > > > > > > > > >
> > > > > > > > > > > > Just to clarify I was asking about the `version` of the
> > > > > > assignor
> > > > > > > > > (i.e. up
> > > > > > > > > > > > to what version that the client would support), and I
> > do
> > > > agree
> > > > > > we
> > > > > > > > > would
> > > > > > > > > > > not
> > > > > > > > > > > > need metadata. What I have in mind is that, for some
> > > > specific
> > > > > > > > > built-in
> > > > > > > > > > > > broker-assignors, e.g. rack-aware assignors, if it's
> > > > possible
> > > > > > that
> > > > > > > > > in a
> > > > > > > > > > > > newer version we would have a hierarchical rack ID
> > string
> > > > > > format,
> > > > > > > > > like
> > > > > > > > > > > > "tier1-tier2" etc, but if some client has not upgraded
> > > > their
> > > > > > rack
> > > > > > > > ID
> > > > > > > > > > > > would still be in old format. In this case, the broker
> > then
> > > > > > needs
> > > > > > > > to
> > > > > > > > > > > choose
> > > > > > > > > > > > the old versioned assignor. I'm probably making
> > something
> > > > up
> > > > > > here
> > > > > > > > for
> > > > > > > > > > > rack
> > > > > > > > > > > > aware assignors, but I'm wondering if in general such
> > an
> > > > > > > > > "auto-downgrade"
> > > > > > > > > > > > behavior would be needed still for broker-side
> > assignor,
> > > > and
> > > > > > if yes
> > > > > > > > > would
> > > > > > > > > > > > "version" still be useful.
> > > > > > > > > > > >
> > > > > > > > > > > > > Yeah, that's right. Within a rebalance,
> > `onAssignment` is
> > > > > > called
> > > > > > > > > once
> > > > > > > > > > > > when the member transitions to a new epoch. This one
> > > > contains
> > > > > > the
> > > > > > > > > full
> > > > > > > > > > > > metadata provided by the client side assignor. Then,
> > > > > > `onAssignment`
> > > > > > > > > > > > can be called max N times where N is the number of
> > > > partitions
> > > > > > > > pending
> > > > > > > > > > > > revocation by other members. Let me try to clarify
> > this in
> > > > the
> > > > > > KIP.
> > > > > > > > > > > >
> > > > > > > > > > > > Okay, my understanding is that the calling ordering of
> > > > these
> > > > > > > > > callbacks
> > > > > > > > > > > > would be like the following:
> > > > > > > > > > > >
> > > > > > > > > > > > ----------------------------------------
> > > > > > > > > > > > onPartitionsRevoked();   // just once, since we do not
> > > > really
> > > > > > need
> > > > > > > > > > > > to revoke incrementally.
> > > > > > > > > > > >
> > > > > > > > > > > > onAssignment();    // the first call, with epoch
> > > > incremented
> > > > > > > > > > > > onPartitionsAssigned();   // paired with the
> > onAssignment
> > > > > > > > > > > >
> > > > > > > > > > > > onAssignment();              // the first onAssignment
> > > > would
> > > > > > bump
> > > > > > > > up
> > > > > > > > > the
> > > > > > > > > > > > epoch, and the metadata reflected.
> > > > > > > > > > > > onPartitionsAssigned();   // each time we get an
> > additional
> > > > > > > > > assignment,
> > > > > > > > > > > we
> > > > > > > > > > > > call onAssignment and then paired with an
> > > > onPartitionsAssigned
> > > > > > > > > > > > ...
> > > > > > > > > > > > onAssignment();
> > > > > > > > > > > > onPartitionsAssigned();   // on each of the
> > onAssignment
> > > > > > calls, the
> > > > > > > > > > > encoded
> > > > > > > > > > > > metadata would not change, only the incrementally added
> > > > > > partitions
> > > > > > > > be
> > > > > > > > > > > > reflected
> > > > > > > > > > > >
> > > > > > > > > > > > Is that the case?
> > > > > > > > > > > >
> > > > > > > > > > > > I'm wondering if we would still call onAssignment just
> > > > once,
> > > > > > that
> > > > > > > > > encodes
> > > > > > > > > > > > all the assignment for this rebalance, including all
> > the
> > > > > > partitions
> > > > > > > > > that
> > > > > > > > > > > > should be assigned to the member but not yet assigned
> > since
> > > > > > they
> > > > > > > > > have not
> > > > > > > > > > > > been revoked by others. In that case the call ordering
> > > > would
> > > > > > be:
> > > > > > > > > > > >
> > > > > > > > > > > > ----------------------------------------
> > > > > > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > > > > > onAssignment();    // just once, with epoch
> > incremented,
> > > > and
> > > > > > > > metadata
> > > > > > > > > > > > encoded changed, the "assignment" field also reflect
> > the
> > > > final
> > > > > > > > target
> > > > > > > > > > > > assignment
> > > > > > > > > > > > onPartitionsAssigned();   // multiple times, which
> > > > represent
> > > > > > > > > > > incrementally
> > > > > > > > > > > > added partitions
> > > > > > > > > > > > ...
> > > > > > > > > > > > onPartitionsAssigned();
> > > > > > > > > > > >
> > > > > > > > > > > > The motivation from this is that, most users would only
> > > > > > implement
> > > > > > > > the
> > > > > > > > > > > > rebalance callback listeners and hence we'd definitely
> > > > need to
> > > > > > make
> > > > > > > > > sure
> > > > > > > > > > > > the semantics of that does not change much, and the
> > time
> > > > > > > > > > > > `onPartitionsAssigned` indicate the time when the
> > > > partitions
> > > > > > are
> > > > > > > > > actually
> > > > > > > > > > > > assigned to it; while for assignors, the
> > `onAssignment` is
> > > > > > used to
> > > > > > > > > > > indicate
> > > > > > > > > > > > what decision is made regarding for this member, i.e.
> > when
> > > > the
> > > > > > > > > partitions
> > > > > > > > > > > > are decided to be given to it, but not necessarily
> > meant
> > > > that
> > > > > > it
> > > > > > > > has
> > > > > > > > > been
> > > > > > > > > > > > given, since that time should be determined by the
> > time of
> > > > > > > > > > > > `onPartitionsAssigned`. The benefits though, would be
> > that
> > > > > > assignor
> > > > > > > > > > > > implementers would not need to reason which
> > `onAssignment`
> > > > > > would be
> > > > > > > > > the
> > > > > > > > > > > > last one for this epoch.
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > Guozhang
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > Guozhang
> > > > > > > > > > > >
> > > > > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >
> > > > > > > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > > Hi Guozhang,
> > > > > > > > > > > > >
> > > > > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment"
> > has
> > > > an
> > > > > > Error
> > > > > > > > > byte
> > > > > > > > > > > > > field,
> > > > > > > > > > > > > > while the broker-side
> > "PartitionAssignor#Assignment"
> > > > does
> > > > > > not.
> > > > > > > > > And
> > > > > > > > > > > the
> > > > > > > > > > > > > > rationale seems to be assuming that we should
> > always be
> > > > > > able to
> > > > > > > > > do
> > > > > > > > > > > the
> > > > > > > > > > > > > > assignment at the broker-side assignor without
> > errors.
> > > > > > > > > Personally I
> > > > > > > > > > > > think
> > > > > > > > > > > > > > it's still potentially beneficial to add the Error
> > > > field
> > > > > > even
> > > > > > > > for
> > > > > > > > > > > > > > broker-side assignors, e.g. for some edge cases
> > where
> > > > some
> > > > > > > > > subscribed
> > > > > > > > > > > > > > topics are not recognized with the current broker's
> > > > > > metadata.
> > > > > > > > > What do
> > > > > > > > > > > > you
> > > > > > > > > > > > > > think?
> > > > > > > > > > > > >
> > > > > > > > > > > > > Yeah, that seems reasonable. However, I wonder if it
> > > > would be
> > > > > > > > > better
> > > > > > > > > > > > > to use an exception on the server side. This is what
> > we
> > > > > > usually
> > > > > > > > do
> > > > > > > > > for
> > > > > > > > > > > > > server side plugins. On the client side, we use a
> > field
> > > > > > because
> > > > > > > > the
> > > > > > > > > > > > > errors are not defined in advance.
> > > > > > > > > > > > >
> > > > > > > > > > > > > Your comment also makes me think about what we
> > should do
> > > > > > when the
> > > > > > > > > > > > > server side assignor fails. I suppose that we have to
> > > > keep
> > > > > > the
> > > > > > > > > current
> > > > > > > > > > > > > assignment until a new event occurs. For instance, in
> > > > your
> > > > > > > > example,
> > > > > > > > > > > > > the coordinator would have to trigger a rebalance
> > when
> > > > > > > > unrecognized
> > > > > > > > > > > > > topics are available. This would be part of the
> > metadata
> > > > > > > > > monitoring.
> > > > > > > > > > > > >
> > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> > additional
> > > > > > fields
> > > > > > > > > > > > > > reason/version/metadata compared with the
> > broker-side
> > > > > > > > > GroupMember. I
> > > > > > > > > > > > > agree
> > > > > > > > > > > > > > that broker-side assignor would not need
> > > > reason/metadata
> > > > > > since
> > > > > > > > > they
> > > > > > > > > > > are
> > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but what
> > about
> > > > > > version?
> > > > > > > > > E.g.
> > > > > > > > > > > is
> > > > > > > > > > > > > it
> > > > > > > > > > > > > > possible that we evolve our broker-side built-in
> > > > assignor
> > > > > > but
> > > > > > > > > the old
> > > > > > > > > > > > > > versioned clients would not be able to work with
> > the
> > > > new
> > > > > > > > > version, in
> > > > > > > > > > > > > which
> > > > > > > > > > > > > > case we need to let the broker being aware of this
> > and
> > > > > > upgrade
> > > > > > > > > its
> > > > > > > > > > > > > behavior
> > > > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > > > >
> > > > > > > > > > > > > I don't really see the benefits here because server
> > side
> > > > > > > > assignors
> > > > > > > > > > > > > don't have metadata at all. They only assign
> > > > > > topic-partitions.
> > > > > > > > They
> > > > > > > > > > > > > are not supposed to generate metadata nor to receive
> > > > metadata
> > > > > > > > from
> > > > > > > > > the
> > > > > > > > > > > > > members.
> > > > > > > > > > > > >
> > > > > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > > > > "GroupMember",
> > > > > > > > > > > instead
> > > > > > > > > > > > > of
> > > > > > > > > > > > > > including these three fields, what about just
> > adding
> > > > the
> > > > > > > > > "Metadata"
> > > > > > > > > > > > field
> > > > > > > > > > > > > > class which has these three fields? Also, there
> > are two
> > > > > > > > > "Metadata"
> > > > > > > > > > > > > > currently in the APIs, the first is a class that
> > > > encodes
> > > > > > > > > > > > > > reason/version/metadata, and the second is just the
> > > > encoded
> > > > > > > > > metadata
> > > > > > > > > > > > > bytes.
> > > > > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > > > > memberMetadata,
> > > > > > > > > > > which
> > > > > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> > > > naming the
> > > > > > > > > second
> > > > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > > > >
> > > > > > > > > > > > > That's a good point. Let me try to rationalize this
> > > > interface
> > > > > > > > > based on
> > > > > > > > > > > > > your suggestions.
> > > > > > > > > > > > >
> > > > > > > > > > > > > Best,
> > > > > > > > > > > > > David
> > > > > > > > > > > > >
> > > > > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > wrote:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Hello David,
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Just had a few more nit questions about the public
> > > > APIs:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment"
> > has
> > > > an
> > > > > > Error
> > > > > > > > > byte
> > > > > > > > > > > > > field,
> > > > > > > > > > > > > > while the broker-side
> > "PartitionAssignor#Assignment"
> > > > does
> > > > > > not.
> > > > > > > > > And
> > > > > > > > > > > the
> > > > > > > > > > > > > > rationale seems to be assuming that we should
> > always be
> > > > > > able to
> > > > > > > > > do
> > > > > > > > > > > the
> > > > > > > > > > > > > > assignment at the broker-side assignor without
> > errors.
> > > > > > > > > Personally I
> > > > > > > > > > > > think
> > > > > > > > > > > > > > it's still potentially beneficial to add the Error
> > > > field
> > > > > > even
> > > > > > > > for
> > > > > > > > > > > > > > broker-side assignors, e.g. for some edge cases
> > where
> > > > some
> > > > > > > > > subscribed
> > > > > > > > > > > > > > topics are not recognized with the current broker's
> > > > > > metadata.
> > > > > > > > > What do
> > > > > > > > > > > > you
> > > > > > > > > > > > > > think?
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> > additional
> > > > > > fields
> > > > > > > > > > > > > > reason/version/metadata compared with the
> > broker-side
> > > > > > > > > GroupMember. I
> > > > > > > > > > > > > agree
> > > > > > > > > > > > > > that broker-side assignor would not need
> > > > reason/metadata
> > > > > > since
> > > > > > > > > they
> > > > > > > > > > > are
> > > > > > > > > > > > > > blackbox strings/bytes to the assignor, but what
> > about
> > > > > > version?
> > > > > > > > > E.g.
> > > > > > > > > > > is
> > > > > > > > > > > > > it
> > > > > > > > > > > > > > possible that we evolve our broker-side built-in
> > > > assignor
> > > > > > but
> > > > > > > > > the old
> > > > > > > > > > > > > > versioned clients would not be able to work with
> > the
> > > > new
> > > > > > > > > version, in
> > > > > > > > > > > > > which
> > > > > > > > > > > > > > case we need to let the broker being aware of this
> > and
> > > > > > upgrade
> > > > > > > > > its
> > > > > > > > > > > > > behavior
> > > > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > > > > "GroupMember",
> > > > > > > > > > > instead
> > > > > > > > > > > > > of
> > > > > > > > > > > > > > including these three fields, what about just
> > adding
> > > > the
> > > > > > > > > "Metadata"
> > > > > > > > > > > > field
> > > > > > > > > > > > > > class which has these three fields? Also, there
> > are two
> > > > > > > > > "Metadata"
> > > > > > > > > > > > > > currently in the APIs, the first is a class that
> > > > encodes
> > > > > > > > > > > > > > reason/version/metadata, and the second is just the
> > > > encoded
> > > > > > > > > metadata
> > > > > > > > > > > > > bytes.
> > > > > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > > > > memberMetadata,
> > > > > > > > > > > which
> > > > > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> > > > naming the
> > > > > > > > > second
> > > > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Hello David,
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Thanks for bringing this question up. I think the
> > > > main
> > > > > > > > > benefits as
> > > > > > > > > > > > you
> > > > > > > > > > > > > > > listed is 2) above if it stays; just to clarify,
> > we
> > > > would
> > > > > > > > only
> > > > > > > > > be
> > > > > > > > > > > > able
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > > save one round trip if the rebalance is still
> > > > triggered
> > > > > > by
> > > > > > > > the
> > > > > > > > > > > > broker;
> > > > > > > > > > > > > if
> > > > > > > > > > > > > > > the rebalance is triggered by the client then the
> > > > > > > > > num.round.trips
> > > > > > > > > > > are
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > same:
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > T0: client decides to do a new assignment,
> > suppose
> > > > it has
> > > > > > > > > already
> > > > > > > > > > > > sent
> > > > > > > > > > > > > a
> > > > > > > > > > > > > > > HB and hence has to wait for it to return first
> > since
> > > > > > only
> > > > > > > > one
> > > > > > > > > > > > request
> > > > > > > > > > > > > /
> > > > > > > > > > > > > > > response can be inflight with the coordinator's
> > > > socket.
> > > > > > > > > > > > > > > T1: client receives the HB response, and then
> > sends
> > > > the
> > > > > > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > > > > > T2: the GroupPrepareAssignment response is
> > returned.
> > > > > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > > > request.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > In total, two round trips.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > T0: client decides to do a new assignment,
> > suppose
> > > > it has
> > > > > > > > > already
> > > > > > > > > > > > sent
> > > > > > > > > > > > > a
> > > > > > > > > > > > > > > HB and hence has to wait for it to return first
> > since
> > > > > > only
> > > > > > > > one
> > > > > > > > > > > > request
> > > > > > > > > > > > > /
> > > > > > > > > > > > > > > response can be inflight with the coordinator's
> > > > socket.
> > > > > > > > > > > > > > > T1: client receives the HB response, and then
> > sends
> > > > the
> > > > > > new
> > > > > > > > HB
> > > > > > > > > > > > request
> > > > > > > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > > > > > > T2: the HB response with the optional member
> > metadata
> > > > > > map is
> > > > > > > > > > > > returned.
> > > > > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > > > request.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > -----------------------------
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > So to complete the full picture here, we'd need
> > to
> > > > modify
> > > > > > > > both
> > > > > > > > > HB
> > > > > > > > > > > > > request
> > > > > > > > > > > > > > > and response so that the client can also
> > indicate a
> > > > new
> > > > > > > > > rebalance
> > > > > > > > > > > via
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > > HB request as well, right?
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Assuming all above is true, I think it's okay to
> > > > merge
> > > > > > the
> > > > > > > > > > > > > > > GroupPrepareAssignment into HB given that we can
> > > > make the
> > > > > > > > > > > additional
> > > > > > > > > > > > > fields
> > > > > > > > > > > > > > > encoding the full member (subscription) metadata
> > and
> > > > > > topic
> > > > > > > > > metadata
> > > > > > > > > > > > as
> > > > > > > > > > > > > > > optional fields.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > >> Hi all,
> > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > >> During an offline conversation, someone asked
> > why we
> > > > > > need
> > > > > > > > the
> > > > > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested
> > > > that we
> > > > > > > > could
> > > > > > > > > > > > instead
> > > > > > > > > > > > > > >> provide the group state in the heartbeat
> > response.
> > > > This
> > > > > > has
> > > > > > > > a
> > > > > > > > > few
> > > > > > > > > > > > > > >> advantages: 1) it does not require using a
> > special
> > > > error
> > > > > > > > code
> > > > > > > > > to
> > > > > > > > > > > > > > >> signal that a new assignment is required as the
> > > > signal
> > > > > > would
> > > > > > > > > be
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> provided group state; 2) it removes one round
> > trip
> > > > when
> > > > > > a
> > > > > > > > > client
> > > > > > > > > > > > side
> > > > > > > > > > > > > > >> assignor is used. The downside is that it makes
> > the
> > > > > > > > heartbeat
> > > > > > > > > > > > > > >> response's definition quite large. I recall
> > that I
> > > > went
> > > > > > with
> > > > > > > > > the
> > > > > > > > > > > > > > >> current approach due to this.
> > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > >> Providing the group state in the heartbeat
> > response
> > > > is
> > > > > > > > > appealing.
> > > > > > > > > > > > What
> > > > > > > > > > > > > > >> do you guys think?
> > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > >> Best,
> > > > > > > > > > > > > > >> David
> > > > > > > > > > > > > > >>
> > > > > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > > > > > > djacot@confluent.io>
> > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > 1. I have added a reference to the relevant
> > > > chapter
> > > > > > > > instead
> > > > > > > > > of
> > > > > > > > > > > > > > >> > repeating the whole thing. Does that work for
> > you?
> > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you are
> > > > referring
> > > > > > to
> > > > > > > > is
> > > > > > > > > > > about
> > > > > > > > > > > > > when
> > > > > > > > > > > > > > >> > a rebalance should be triggered for the
> > > > non-upgraded
> > > > > > > > members
> > > > > > > > > > > using
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> > old protocol. The section mentions that a
> > > > rebalance
> > > > > > must
> > > > > > > > be
> > > > > > > > > > > > > triggered
> > > > > > > > > > > > > > >> > when a new assignment is installed. This
> > implies
> > > > that
> > > > > > the
> > > > > > > > > group
> > > > > > > > > > > > > epoch
> > > > > > > > > > > > > > >> > was updated either by a native member or a
> > > > > > non-upgraded
> > > > > > > > > member.
> > > > > > > > > > > > For
> > > > > > > > > > > > > > >> > the latter, the JoinGroup request would be the
> > > > > > trigger. I
> > > > > > > > > have
> > > > > > > > > > > > > added a
> > > > > > > > > > > > > > >> > reference to the relevant chapter in the
> > > > "JoinGroup
> > > > > > > > > Handling"
> > > > > > > > > > > > > section
> > > > > > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > Thanks,
> > > > > > > > > > > > > > >> > David
> > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang
> > <
> > > > > > > > > > > wangguoz@gmail.com
> > > > > > > > > > > > >
> > > > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > Hello David,
> > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > Alright I think that's sufficient. Just to
> > make
> > > > that
> > > > > > > > > clear in
> > > > > > > > > > > > the
> > > > > > > > > > > > > doc,
> > > > > > > > > > > > > > >> > > could we update:
> > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > 1) the heartbeat request handling section,
> > > > stating
> > > > > > when
> > > > > > > > > > > > > coordinator
> > > > > > > > > > > > > > >> will
> > > > > > > > > > > > > > >> > > trigger rebalance based on the HB's member
> > > > metadata
> > > > > > /
> > > > > > > > > reason?
> > > > > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to
> > include
> > > > what
> > > > > > we
> > > > > > > > > > > described
> > > > > > > > > > > > > in
> > > > > > > > > > > > > > >> "Group
> > > > > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as
> > well?
> > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > Guozhang
> > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > >> > > wrote:
> > > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > I thought that the assignor will always be
> > > > > > consulted
> > > > > > > > > when
> > > > > > > > > > > the
> > > > > > > > > > > > > next
> > > > > > > > > > > > > > >> > > > heartbeat request is constructed. In other
> > > > words,
> > > > > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be
> > called
> > > > for
> > > > > > every
> > > > > > > > > > > > heartbeat.
> > > > > > > > > > > > > > >> This
> > > > > > > > > > > > > > >> > > > gives the opportunity for the assignor to
> > > > enforce
> > > > > > a
> > > > > > > > > > > rebalance
> > > > > > > > > > > > by
> > > > > > > > > > > > > > >> > > > setting the reason to a non-zero value or
> > by
> > > > > > changing
> > > > > > > > > the
> > > > > > > > > > > > > bytes. Do
> > > > > > > > > > > > > > >> > > > you think that this is not sufficient?
> > Are you
> > > > > > > > > concerned by
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> delay?
> > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > Best,
> > > > > > > > > > > > > > >> > > > David
> > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang
> > Wang <
> > > > > > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > > > >> > > > > In this case, a new assignment is
> > triggered
> > > > by
> > > > > > the
> > > > > > > > > client
> > > > > > > > > > > > side
> > > > > > > > > > > > > > >> > > > > assignor. When constructing the HB, the
> > > > consumer
> > > > > > > > will
> > > > > > > > > > > always
> > > > > > > > > > > > > > >> consult
> > > > > > > > > > > > > > >> > > > > the client side assignor and propagate
> > the
> > > > > > > > > information to
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > coordinator. In other words, we don't
> > expect
> > > > > > users
> > > > > > > > to
> > > > > > > > > call
> > > > > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > As I looked at the current
> > > > PartitionAssignor's
> > > > > > > > > interface,
> > > > > > > > > > > we
> > > > > > > > > > > > > > >> actually do
> > > > > > > > > > > > > > >> > > > > not have a way yet to instruct how to
> > > > construct
> > > > > > the
> > > > > > > > > next
> > > > > > > > > > > HB
> > > > > > > > > > > > > > >> request, e.g.
> > > > > > > > > > > > > > >> > > > > when the assignor wants to enforce a new
> > > > > > rebalance
> > > > > > > > > with a
> > > > > > > > > > > > new
> > > > > > > > > > > > > > >> assignment,
> > > > > > > > > > > > > > >> > > > > we'd need some customizable APIs inside
> > the
> > > > > > > > > > > > PartitionAssignor
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > >> indicate
> > > > > > > > > > > > > > >> > > > > the next HB telling broker about so.
> > WDYT
> > > > about
> > > > > > > > adding
> > > > > > > > > > > such
> > > > > > > > > > > > an
> > > > > > > > > > > > > > >> API on the
> > > > > > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David
> > Jacot
> > > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > I have updated the KIP to include your
> > > > > > feedback. I
> > > > > > > > > have
> > > > > > > > > > > > also
> > > > > > > > > > > > > > >> tried to
> > > > > > > > > > > > > > >> > > > > > clarify the parts which were not
> > cleared.
> > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David
> > > > Jacot <
> > > > > > > > > > > > > djacot@confluent.io
> > > > > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me
> > start
> > > > by
> > > > > > > > > answering
> > > > > > > > > > > your
> > > > > > > > > > > > > > >> questions
> > > > > > > > > > > > > > >> > > > > > > inline and I will update the KIP
> > next
> > > > week.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the
> > main
> > > > > > benefits
> > > > > > > > > of
> > > > > > > > > > > the
> > > > > > > > > > > > > KIP
> > > > > > > > > > > > > > >> seem to
> > > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > > > efficient
> > > > > > > > > support of
> > > > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > I would also add that the KIP
> > removes
> > > > the
> > > > > > global
> > > > > > > > > sync
> > > > > > > > > > > > > barrier
> > > > > > > > > > > > > > >> in the
> > > > > > > > > > > > > > >> > > > > > > protocol which is essential to
> > improve
> > > > group
> > > > > > > > > stability
> > > > > > > > > > > > and
> > > > > > > > > > > > > > >> > > > > > > scalability, and the KIP also
> > > > simplifies the
> > > > > > > > > client by
> > > > > > > > > > > > > moving
> > > > > > > > > > > > > > >> most of
> > > > > > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> > singleton.
> > > > Do we
> > > > > > plan
> > > > > > > > > to
> > > > > > > > > > > > > support
> > > > > > > > > > > > > > >> rolling
> > > > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > > > > consumers?
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Definitely. The group coordinator
> > will
> > > > use
> > > > > > the
> > > > > > > > > > > assignor
> > > > > > > > > > > > > used
> > > > > > > > > > > > > > >> by a
> > > > > > > > > > > > > > >> > > > > > > majority of the members. This
> > allows the
> > > > > > group
> > > > > > > > to
> > > > > > > > > move
> > > > > > > > > > > > > from
> > > > > > > > > > > > > > >> one
> > > > > > > > > > > > > > >> > > > > > > assignor to another by a roll. This
> > is
> > > > > > explained
> > > > > > > > > in
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> Assignor
> > > > > > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you
> > explain
> > > > > > whether
> > > > > > > > > it's
> > > > > > > > > > > > > > >> required in
> > > > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> > > > needs to
> > > > > > be
> > > > > > > > > filled?
> > > > > > > > > > > > For
> > > > > > > > > > > > > > >> example,
> > > > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> > > > needs to
> > > > > > be
> > > > > > > > > filled.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > The client is expected to set those
> > > > fields
> > > > > > in
> > > > > > > > > case of
> > > > > > > > > > > a
> > > > > > > > > > > > > > >> connection
> > > > > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the
> > fields
> > > > have
> > > > > > > > > changed
> > > > > > > > > > > > since
> > > > > > > > > > > > > > >> the last
> > > > > > > > > > > > > > >> > > > > > > HB. The server populates those
> > fields as
> > > > > > long as
> > > > > > > > > the
> > > > > > > > > > > > > member
> > > > > > > > > > > > > > >> is not
> > > > > > > > > > > > > > >> > > > > > > fully reconciled - the member should
> > > > > > acknowledge
> > > > > > > > > that
> > > > > > > > > > > it
> > > > > > > > > > > > > has
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > expected epoch and assignment. I
> > will
> > > > > > clarify
> > > > > > > > > this in
> > > > > > > > > > > > the
> > > > > > > > > > > > > KIP.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> > protocol,
> > > > the
> > > > > > rack
> > > > > > > > > > > > affinity
> > > > > > > > > > > > > > >> between the
> > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > >> > > > > > > > and the broker is only considered
> > > > during
> > > > > > > > > fetching,
> > > > > > > > > > > but
> > > > > > > > > > > > > not
> > > > > > > > > > > > > > >> during
> > > > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> > Sometimes,
> > > > once
> > > > > > the
> > > > > > > > > > > > assignment
> > > > > > > > > > > > > is
> > > > > > > > > > > > > > >> made,
> > > > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> > > > because
> > > > > > no
> > > > > > > > > replicas
> > > > > > > > > > > > of
> > > > > > > > > > > > > > >> assigned
> > > > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > > > wondering
> > > > > > if we
> > > > > > > > > should
> > > > > > > > > > > > use
> > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > > > >> > > > > > > > to address this by including rack
> > in
> > > > > > > > > GroupMember.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > That's an interesting idea. I don't
> > see
> > > > any
> > > > > > > > issue
> > > > > > > > > with
> > > > > > > > > > > > > adding
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > rack
> > > > > > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often,
> > it's
> > > > > > useful to
> > > > > > > > > know
> > > > > > > > > > > how
> > > > > > > > > > > > > busy
> > > > > > > > > > > > > > >> a group
> > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> > event
> > > > loop
> > > > > > > > model,
> > > > > > > > > it
> > > > > > > > > > > > seems
> > > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction
> > of
> > > > the
> > > > > > time
> > > > > > > > > the
> > > > > > > > > > > > event
> > > > > > > > > > > > > > >> loop is
> > > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > That's a great idea. I will add it.
> > > > Thanks.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > > > coordinator
> > > > > > > > > failover
> > > > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the check
> > if
> > > > any
> > > > > > group
> > > > > > > > > with
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> > matching
> > > > topic?
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Sure. When the new group coordinator
> > > > takes
> > > > > > over,
> > > > > > > > > it
> > > > > > > > > > > has
> > > > > > > > > > > > > to:
> > > > > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a
> > client
> > > > side
> > > > > > > > > assignor
> > > > > > > > > > > is
> > > > > > > > > > > > > used.
> > > > > > > > > > > > > > >> We
> > > > > > > > > > > > > > >> > > > don't
> > > > > > > > > > > > > > >> > > > > > > store the information about the
> > member
> > > > > > selected
> > > > > > > > > to run
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > > > > > >> > > > > > > * Update the topics metadata,
> > verify the
> > > > > > > > wildcard
> > > > > > > > > > > > > > >> subscriptions, and
> > > > > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> > > > Could we
> > > > > > > > > document
> > > > > > > > > > > > what
> > > > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does the
> > epoch
> > > > in
> > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> > about the
> > > > > > one in
> > > > > > > > > > > > > > >> > > > > > > >
> > ConsumerGroupPartitionMetadataValue
> > > > and
> > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Sure. I will clarify that but it is
> > > > always
> > > > > > the
> > > > > > > > > latest
> > > > > > > > > > > > > group
> > > > > > > > > > > > > > >> epoch.
> > > > > > > > > > > > > > >> > > > > > > When the group state is updated, the
> > > > group
> > > > > > epoch
> > > > > > > > > is
> > > > > > > > > > > > > bumped so
> > > > > > > > > > > > > > >> we use
> > > > > > > > > > > > > > >> > > > > > > that one for all the change records
> > > > related
> > > > > > to
> > > > > > > > the
> > > > > > > > > > > > update.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will
> > ensure
> > > > > > that
> > > > > > > > the
> > > > > > > > > > > > > following
> > > > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > > > > possible
> > > > > > > > for
> > > > > > > > > a
> > > > > > > > > > > > member
> > > > > > > > > > > > > > >> not to
> > > > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > That's right. Here I meant that the
> > > > members
> > > > > > > > > provided
> > > > > > > > > > > by
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > > >> > > > > > > in the assignment must exist in the
> > > > group.
> > > > > > The
> > > > > > > > > > > assignor
> > > > > > > > > > > > > can
> > > > > > > > > > > > > > >> not make
> > > > > > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> > with a
> > > > > > member
> > > > > > > > > epoch
> > > > > > > > > > > > > equals to
> > > > > > > > > > > > > > >> 0":
> > > > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member
> > id
> > > > > > would be
> > > > > > > > > used?
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > A member is expected to abandon all
> > its
> > > > > > > > > partitions and
> > > > > > > > > > > > > > >> rejoins when
> > > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH
> > error.
> > > > In
> > > > > > this
> > > > > > > > > case,
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > > > coordinator will have removed the
> > member
> > > > > > from
> > > > > > > > the
> > > > > > > > > > > group.
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > >> member
> > > > > > > > > > > > > > >> > > > > > > can rejoin the group with the same
> > > > member
> > > > > > id but
> > > > > > > > > with
> > > > > > > > > > > 0
> > > > > > > > > > > > as
> > > > > > > > > > > > > > >> epoch. Let
> > > > > > > > > > > > > > >> > > > > > > me see if I can clarify this in the
> > KIP.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will
> > have
> > > > the
> > > > > > > > ability
> > > > > > > > > to
> > > > > > > > > > > > > trigger a
> > > > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > > > >> > > > > > > > by either providing a non-zero
> > reason
> > > > or
> > > > > > by
> > > > > > > > > updating
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > > > > conflicting
> > > > > > > > > with
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > In this case, a new assignment is
> > > > triggered
> > > > > > by
> > > > > > > > the
> > > > > > > > > > > > client
> > > > > > > > > > > > > side
> > > > > > > > > > > > > > >> > > > > > > assignor. When constructing the HB,
> > the
> > > > > > consumer
> > > > > > > > > will
> > > > > > > > > > > > > always
> > > > > > > > > > > > > > >> consult
> > > > > > > > > > > > > > >> > > > > > > the client side assignor and
> > propagate
> > > > the
> > > > > > > > > information
> > > > > > > > > > > > to
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > > > coordinator. In other words, we
> > don't
> > > > expect
> > > > > > > > > users to
> > > > > > > > > > > > call
> > > > > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> > > > nice.
> > > > > > But
> > > > > > > > the
> > > > > > > > > > > > section
> > > > > > > > > > > > > > >> seems to
> > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> > > > epoch
> > > > > > 2, B
> > > > > > > > > > > > > immediately
> > > > > > > > > > > > > > >> gets into
> > > > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]",
> > which
> > > > seems
> > > > > > > > > > > incorrect.
> > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> > > > epoch
> > > > > > 3, C
> > > > > > > > > seems
> > > > > > > > > > > to
> > > > > > > > > > > > > get
> > > > > > > > > > > > > > >> into
> > > > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch
> > 3, C
> > > > > > still
> > > > > > > > > has A -
> > > > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers:
> > Do
> > > > we
> > > > > > > > support
> > > > > > > > > the
> > > > > > > > > > > > > upgrade
> > > > > > > > > > > > > > >> from
> > > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > We will support upgrading from the
> > > > consumer
> > > > > > > > > protocol
> > > > > > > > > > > > > version
> > > > > > > > > > > > > > >> 3,
> > > > > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is
> > not
> > > > > > > > implemented
> > > > > > > > > yet
> > > > > > > > > > > so
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> earliest
> > > > > > > > > > > > > > >> > > > > > > version is unknown at the moment.
> > This
> > > > is
> > > > > > > > > explained in
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> migration
> > > > > > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Thanks again for your feedback,
> > Jun. I
> > > > will
> > > > > > > > > update the
> > > > > > > > > > > > KIP
> > > > > > > > > > > > > > >> based on
> > > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > > > > > >> > > > > > > David
> > > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun
> > Rao
> > > > > > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the
> > main
> > > > > > benefits
> > > > > > > > > of
> > > > > > > > > > > the
> > > > > > > > > > > > > KIP
> > > > > > > > > > > > > > >> seem to
> > > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > > > efficient
> > > > > > > > > support of
> > > > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> > singleton.
> > > > Do we
> > > > > > plan
> > > > > > > > > to
> > > > > > > > > > > > > support
> > > > > > > > > > > > > > >> rolling
> > > > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > > > > consumers?
> > > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you
> > explain
> > > > > > whether
> > > > > > > > > it's
> > > > > > > > > > > > > > >> required in
> > > > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> > > > needs to
> > > > > > be
> > > > > > > > > filled?
> > > > > > > > > > > > For
> > > > > > > > > > > > > > >> example,
> > > > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> > > > needs to
> > > > > > be
> > > > > > > > > filled.
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> > protocol,
> > > > the
> > > > > > rack
> > > > > > > > > > > > affinity
> > > > > > > > > > > > > > >> between the
> > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > >> > > > > > > > and the broker is only considered
> > > > during
> > > > > > > > > fetching,
> > > > > > > > > > > but
> > > > > > > > > > > > > not
> > > > > > > > > > > > > > >> during
> > > > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> > Sometimes,
> > > > once
> > > > > > the
> > > > > > > > > > > > assignment
> > > > > > > > > > > > > is
> > > > > > > > > > > > > > >> made,
> > > > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> > > > because
> > > > > > no
> > > > > > > > > replicas
> > > > > > > > > > > > of
> > > > > > > > > > > > > > >> assigned
> > > > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > > > wondering
> > > > > > if we
> > > > > > > > > should
> > > > > > > > > > > > use
> > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > > > >> > > > > > > > to address this by including rack
> > in
> > > > > > > > > GroupMember.
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often,
> > it's
> > > > > > useful to
> > > > > > > > > know
> > > > > > > > > > > how
> > > > > > > > > > > > > busy
> > > > > > > > > > > > > > >> a group
> > > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> > event
> > > > loop
> > > > > > > > model,
> > > > > > > > > it
> > > > > > > > > > > > seems
> > > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction
> > of
> > > > the
> > > > > > time
> > > > > > > > > the
> > > > > > > > > > > > event
> > > > > > > > > > > > > > >> loop is
> > > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > > > coordinator
> > > > > > > > > failover
> > > > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > > > >> > > > > > > > does it need to trigger the check
> > if
> > > > any
> > > > > > group
> > > > > > > > > with
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> > matching
> > > > topic?
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> > > > Could we
> > > > > > > > > document
> > > > > > > > > > > > what
> > > > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > > > >> > > > > > > > reflects? For example, does the
> > epoch
> > > > in
> > > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> > about the
> > > > > > one in
> > > > > > > > > > > > > > >> > > > > > > >
> > ConsumerGroupPartitionMetadataValue
> > > > and
> > > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will
> > ensure
> > > > > > that
> > > > > > > > the
> > > > > > > > > > > > > following
> > > > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > > > > possible
> > > > > > > > for
> > > > > > > > > a
> > > > > > > > > > > > member
> > > > > > > > > > > > > > >> not to
> > > > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> > with a
> > > > > > member
> > > > > > > > > epoch
> > > > > > > > > > > > > equals to
> > > > > > > > > > > > > > >> 0":
> > > > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member
> > id
> > > > > > would be
> > > > > > > > > used?
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will
> > have
> > > > the
> > > > > > > > ability
> > > > > > > > > to
> > > > > > > > > > > > > trigger a
> > > > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > > > >> > > > > > > > by either providing a non-zero
> > reason
> > > > or
> > > > > > by
> > > > > > > > > updating
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > > > > conflicting
> > > > > > > > > with
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> > > > nice.
> > > > > > But
> > > > > > > > the
> > > > > > > > > > > > section
> > > > > > > > > > > > > > >> seems to
> > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> > > > epoch
> > > > > > 2, B
> > > > > > > > > > > > > immediately
> > > > > > > > > > > > > > >> gets into
> > > > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]",
> > which
> > > > seems
> > > > > > > > > > > incorrect.
> > > > > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> > > > epoch
> > > > > > 3, C
> > > > > > > > > seems
> > > > > > > > > > > to
> > > > > > > > > > > > > get
> > > > > > > > > > > > > > >> into
> > > > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch
> > 3, C
> > > > > > still
> > > > > > > > > has A -
> > > > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers:
> > Do
> > > > we
> > > > > > > > support
> > > > > > > > > the
> > > > > > > > > > > > > upgrade
> > > > > > > > > > > > > > >> from
> > > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM
> > David
> > > > > > Jacot
> > > > > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > > > > > re-implement the
> > > > > > > > > > > > > coordinator
> > > > > > > > > > > > > > >> in
> > > > > > > > > > > > > > >> > > > Java. I
> > > > > > > > > > > > > > >> > > > > > > > > discussed this offline with a
> > few
> > > > folks
> > > > > > and
> > > > > > > > > folks
> > > > > > > > > > > > are
> > > > > > > > > > > > > > >> concerned
> > > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > > >> > > > > > > > > we could introduce many
> > regressions
> > > > in
> > > > > > the
> > > > > > > > old
> > > > > > > > > > > > > protocol
> > > > > > > > > > > > > > >> if we do
> > > > > > > > > > > > > > >> > > > so.
> > > > > > > > > > > > > > >> > > > > > > > > Therefore, I am going to remove
> > this
> > > > > > > > statement
> > > > > > > > > > > from
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> KIP. It
> > > > > > > > > > > > > > >> > > > is an
> > > > > > > > > > > > > > >> > > > > > > > > implementation detail after all
> > so
> > > > it
> > > > > > does
> > > > > > > > not
> > > > > > > > > > > have
> > > > > > > > > > > > > to be
> > > > > > > > > > > > > > >> > > > decided at
> > > > > > > > > > > > > > >> > > > > > > > > this stage. We will likely
> > start by
> > > > > > trying
> > > > > > > > to
> > > > > > > > > > > > > refactor the
> > > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM
> > > > David
> > > > > > Jacot
> > > > > > > > <
> > > > > > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> > machine
> > > > are:
> > > > > > > > > "Empty,
> > > > > > > > > > > > > assigning,
> > > > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> > > > Group
> > > > > > States
> > > > > > > > > > > > section,
> > > > > > > > > > > > > > >> right?
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > This sentence does not refer
> > to
> > > > those
> > > > > > > > group
> > > > > > > > > > > states
> > > > > > > > > > > > > but
> > > > > > > > > > > > > > >> rather
> > > > > > > > > > > > > > >> > > > to a
> > > > > > > > > > > > > > >> > > > > > > > > > state machine replication
> > (SMR).
> > > > This
> > > > > > > > > refers to
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> entire
> > > > > > > > > > > > > > >> > > > state of
> > > > > > > > > > > > > > >> > > > > > > > > > group coordinator which is
> > > > replicated
> > > > > > via
> > > > > > > > > the
> > > > > > > > > > > log
> > > > > > > > > > > > > > >> layer. I will
> > > > > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> > > > state
> > > > > > > > machine
> > > > > > > > > is
> > > > > > > > > > > > > modelled
> > > > > > > > > > > > > > >> as an
> > > > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > The idea is to follow a model
> > > > similar
> > > > > > to
> > > > > > > > > the new
> > > > > > > > > > > > > quorum
> > > > > > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > > > > > >> > > > > > > > > > will have N threads to process
> > > > events.
> > > > > > > > Each
> > > > > > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > > > > > >> > > > > > > > > > partition is assigned to a
> > unique
> > > > > > thread
> > > > > > > > > and all
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> events
> > > > > > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> > > > > > processed
> > > > > > > > by
> > > > > > > > > this
> > > > > > > > > > > > > > >> thread. This
> > > > > > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > > > > > >> > > > > > > > > > concurrency and will enable
> > us to
> > > > do
> > > > > > > > > simulation
> > > > > > > > > > > > > testing
> > > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> > > > machine
> > > > > > per
> > > > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > > > consumer
> > > > > > group"
> > > > > > > > > owned
> > > > > > > > > > > > by
> > > > > > > > > > > > > a
> > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > > > coordinator
> > > > > > owns 2
> > > > > > > > > > > > consumer
> > > > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> > will we
> > > > > > have 1
> > > > > > > > > state
> > > > > > > > > > > > > machine
> > > > > > > > > > > > > > >> for it,
> > > > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes
> > from
> > > > > > there, I
> > > > > > > > > > > think.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > > > "*group.coordinator.threads"
> > > > > > > > > > > *is
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> number
> > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > > > >> > > > > > > > > > > to run the state machines.
> > But
> > > > I'm
> > > > > > > > > wondering
> > > > > > > > > > > if
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> purpose
> > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> > > > each
> > > > > > > > consumer
> > > > > > > > > > > group
> > > > > > > > > > > > > (or
> > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > > > > computation,
> > > > > > > > > why
> > > > > > > > > > > > > should we
> > > > > > > > > > > > > > >> need
> > > > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have
> > an
> > > > > > ability to
> > > > > > > > > shard
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> > > > processing as
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> > > > timeout,
> > > > > > why
> > > > > > > > > does
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> default
> > > > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > > > > max(60s)? I
> > > > > > > > > > > thought
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> min/max
> > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> > lower/upper
> > > > > > bound
> > > > > > > > of
> > > > > > > > > it,
> > > > > > > > > > > > no?
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > group.consumer.session.timeout.ms
> > > > > > int
> > > > > > > > > 30s The
> > > > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> > consumer
> > > > > > group
> > > > > > > > > > > protocol.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > group.consumer.min.session.timeout.ms
> > > > > > > > > int 45s
> > > > > > > > > > > > The
> > > > > > > > > > > > > > >> minimum
> > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > group.consumer.max.session.timeout.ms
> > > > > > > > > int 60s
> > > > > > > > > > > > The
> > > > > > > > > > > > > > >> maximum
> > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The
> > > > default
> > > > > > > > > session
> > > > > > > > > > > > > timeout
> > > > > > > > > > > > > > >> should
> > > > > > > > > > > > > > >> > > > be 45s
> > > > > > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > > > > assignor
> > > > > > > > are
> > > > > > > > > > > > [range,
> > > > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > > > assignor.
> > > > > > I'd
> > > > > > > > > like to
> > > > > > > > > > > > > know
> > > > > > > > > > > > > > >> why not
> > > > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> > > > choose
> > > > > > > > uniform
> > > > > > > > > > > > assignor
> > > > > > > > > > > > > > >> (former
> > > > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> > distribution.
> > > > Any
> > > > > > > > other
> > > > > > > > > > > reason
> > > > > > > > > > > > > we
> > > > > > > > > > > > > > >> choose
> > > > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors
> > List
> > > > range,
> > > > > > > > > uniform
> > > > > > > > > > > The
> > > > > > > > > > > > > > >> server side
> > > > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > The order on the server side
> > has
> > > > no
> > > > > > > > > influence
> > > > > > > > > > > > > because
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > > > >> > > > > > must
> > > > > > > > > > > > > > >> > > > > > > > > > chose the selector that he
> > wants
> > > > to
> > > > > > use.
> > > > > > > > > There
> > > > > > > > > > > is
> > > > > > > > > > > > no
> > > > > > > > > > > > > > >> default
> > > > > > > > > > > > > > >> > > > in the
> > > > > > > > > > > > > > >> > > > > > > > > > current proposal. If the
> > assignor
> > > > is
> > > > > > not
> > > > > > > > > > > specified
> > > > > > > > > > > > > by
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > client,
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > request is rejected. The
> > default
> > > > > > client
> > > > > > > > > value
> > > > > > > > > > > for
> > > > > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is
> > > > `uniform`
> > > > > > > > though.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > Thanks for your very good
> > > > comments,
> > > > > > Luke.
> > > > > > > > I
> > > > > > > > > hope
> > > > > > > > > > > > > that my
> > > > > > > > > > > > > > >> > > > answers
> > > > > > > > > > > > > > >> > > > > > help
> > > > > > > > > > > > > > >> > > > > > > > > > to clarify things. I will
> > update
> > > > the
> > > > > > KIP
> > > > > > > > as
> > > > > > > > > well
> > > > > > > > > > > > > based
> > > > > > > > > > > > > > >> on your
> > > > > > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29
> > AM
> > > > Luke
> > > > > > Chen
> > > > > > > > <
> > > > > > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator
> > > > section, you
> > > > > > > > > > > mentioned:
> > > > > > > > > > > > > > >> > > > > > > > > > > > The new group coordinator
> > will
> > > > > > have a
> > > > > > > > > state
> > > > > > > > > > > > > machine
> > > > > > > > > > > > > > >> per
> > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets*
> > partitions,
> > > > > > where
> > > > > > > > > each
> > > > > > > > > > > > state
> > > > > > > > > > > > > > >> machine is
> > > > > > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > > > > > >> > > > > > > > > > > event loop. Those state
> > machines
> > > > > > will be
> > > > > > > > > > > > executed
> > > > > > > > > > > > > in
> > > > > > > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads*
> > > > threads.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> > machine
> > > > are:
> > > > > > > > > "Empty,
> > > > > > > > > > > > > assigning,
> > > > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> > > > Group
> > > > > > States
> > > > > > > > > > > > section,
> > > > > > > > > > > > > > >> right?
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> > > > state
> > > > > > > > machine
> > > > > > > > > is
> > > > > > > > > > > > > modelled
> > > > > > > > > > > > > > >> as an
> > > > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> > > > machine
> > > > > > per
> > > > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > > > consumer
> > > > > > group"
> > > > > > > > > owned
> > > > > > > > > > > > by
> > > > > > > > > > > > > a
> > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > > > coordinator
> > > > > > owns 2
> > > > > > > > > > > > consumer
> > > > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> > will we
> > > > > > have 1
> > > > > > > > > state
> > > > > > > > > > > > > machine
> > > > > > > > > > > > > > >> for it,
> > > > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > > > "*group.coordinator.threads"
> > > > > > > > > > > *is
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> number
> > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > > > >> > > > > > > > > > > to run the state machines.
> > But
> > > > I'm
> > > > > > > > > wondering
> > > > > > > > > > > if
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> purpose
> > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> > > > each
> > > > > > > > consumer
> > > > > > > > > > > group
> > > > > > > > > > > > > (or
> > > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > > > > computation,
> > > > > > > > > why
> > > > > > > > > > > > > should we
> > > > > > > > > > > > > > >> need
> > > > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 2. For the default value in
> > the
> > > > new
> > > > > > > > > configs:
> > > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> > > > timeout,
> > > > > > why
> > > > > > > > > does
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> default
> > > > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > > > > max(60s)? I
> > > > > > > > > > > thought
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> min/max
> > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> > lower/upper
> > > > > > bound
> > > > > > > > of
> > > > > > > > > it,
> > > > > > > > > > > > no?
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > group.consumer.session.timeout.ms
> > > > > > int
> > > > > > > > > 30s The
> > > > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> > consumer
> > > > > > group
> > > > > > > > > > > protocol.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > group.consumer.min.session.timeout.ms
> > > > > > > > > int 45s
> > > > > > > > > > > > The
> > > > > > > > > > > > > > >> minimum
> > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > group.consumer.max.session.timeout.ms
> > > > > > > > > int 60s
> > > > > > > > > > > > The
> > > > > > > > > > > > > > >> maximum
> > > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > > > > assignor
> > > > > > > > are
> > > > > > > > > > > > [range,
> > > > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > > > assignor.
> > > > > > I'd
> > > > > > > > > like to
> > > > > > > > > > > > > know
> > > > > > > > > > > > > > >> why not
> > > > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> > > > choose
> > > > > > > > uniform
> > > > > > > > > > > > assignor
> > > > > > > > > > > > > > >> (former
> > > > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> > distribution.
> > > > Any
> > > > > > > > other
> > > > > > > > > > > reason
> > > > > > > > > > > > > we
> > > > > > > > > > > > > > >> choose
> > > > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors
> > List
> > > > range,
> > > > > > > > > uniform
> > > > > > > > > > > The
> > > > > > > > > > > > > > >> server side
> > > > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at
> > 2:10 PM
> > > > Luke
> > > > > > > > Chen
> > > > > > > > > <
> > > > > > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts about
> > > > Kafka
> > > > > > > > Connect
> > > > > > > > > > > > > > >> integrating with
> > > > > > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > > > > > >> > > > > > > > > > > > think we should have a
> > > > separate
> > > > > > > > > discussion
> > > > > > > > > > > > > thread
> > > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > > >> > > > Kafka
> > > > > > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka
> > Connect
> > > > > > With
> > > > > > > > New
> > > > > > > > > > > > Consumer
> > > > > > > > > > > > > > >> Rebalance
> > > > > > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > > > > > >> > > > > > > > > > > > and let this discussion
> > thread
> > > > > > focus
> > > > > > > > on
> > > > > > > > > > > > consumer
> > > > > > > > > > > > > > >> rebalance
> > > > > > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >>
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at
> > 9:31
> > > > PM
> > > > > > Sagar
> > > > > > > > <
> > > > > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David
> > for
> > > > the
> > > > > > > > > feedback.
> > > > > > > > > > > > > Looks
> > > > > > > > > > > > > > >> like
> > > > > > > > > > > > > > >> > > > there's
> > > > > > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs for
> > > > Connect.
> > > > > > I
> > > > > > > > > would
> > > > > > > > > > > > > revisit
> > > > > > > > > > > > > > >> the doc
> > > > > > > > > > > > > > >> > > > and
> > > > > > > > > > > > > > >> > > > > > see
> > > > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at
> > 7:11
> > > > PM
> > > > > > David
> > > > > > > > > Jacot
> > > > > > > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback
> > > > and the
> > > > > > > > > document.
> > > > > > > > > > > > > That's
> > > > > > > > > > > > > > >> really
> > > > > > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me
> > > > that
> > > > > > both
> > > > > > > > > Connect
> > > > > > > > > > > > > and the
> > > > > > > > > > > > > > >> > > > Consumer
> > > > > > > > > > > > > > >> > > > > > could
> > > > > > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the same underlying
> > > > "engine".
> > > > > > The
> > > > > > > > > main
> > > > > > > > > > > > > > >> difference is
> > > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigns
> > topic-partitions to
> > > > > > members
> > > > > > > > > > > whereas
> > > > > > > > > > > > > > >> Connect
> > > > > > > > > > > > > > >> > > > assigns
> > > > > > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two
> > ways to
> > > > move
> > > > > > > > > forward:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new
> > > > proposed
> > > > > > APIs
> > > > > > > > to
> > > > > > > > > > > > support
> > > > > > > > > > > > > > >> different
> > > > > > > > > > > > > > >> > > > > > resource
> > > > > > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions,
> > tasks,
> > > > > > etc.); or
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated
> > > > APIs
> > > > > > for
> > > > > > > > > Connect.
> > > > > > > > > > > > The
> > > > > > > > > > > > > > >> dedicated
> > > > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new
> > ones but
> > > > > > > > > different on
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > they would rely on the
> > same
> > > > > > engine
> > > > > > > > > on the
> > > > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > > > >> > > > side.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean
> > towards
> > > > 2)
> > > > > > > > because
> > > > > > > > > I am
> > > > > > > > > > > > > not a
> > > > > > > > > > > > > > >> fan of
> > > > > > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> > > > > > purposes.
> > > > > > > > > That
> > > > > > > > > > > > being
> > > > > > > > > > > > > > >> said, I am
> > > > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an
> > > > elegant
> > > > > > way to
> > > > > > > > > do
> > > > > > > > > > > it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > I think that we can
> > > > continue to
> > > > > > > > > discuss
> > > > > > > > > > > it
> > > > > > > > > > > > > here
> > > > > > > > > > > > > > >> for now
> > > > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > > > >> > > > > > order
> > > > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> > > > > > compatible
> > > > > > > > > with
> > > > > > > > > > > > what
> > > > > > > > > > > > > we
> > > > > > > > > > > > > > >> will do
> > > > > > > > > > > > > > >> > > > for
> > > > > > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at
> > > > 2:41 PM
> > > > > > > > David
> > > > > > > > > > > Jacot
> > > > > > > > > > > > <
> > > > > > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from
> > vacation.
> > > > I
> > > > > > will
> > > > > > > > go
> > > > > > > > > > > > through
> > > > > > > > > > > > > and
> > > > > > > > > > > > > > >> address
> > > > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming days.
> > > > Thanks
> > > > > > for
> > > > > > > > your
> > > > > > > > > > > > > feedback.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022
> > at
> > > > 10:05
> > > > > > PM
> > > > > > > > > Gregory
> > > > > > > > > > > > > Harris
> > > > > > > > > > > > > > >> <
> > > > > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP,
> > > > it's
> > > > > > > > > wonderful to
> > > > > > > > > > > > see
> > > > > > > > > > > > > > >> > > > cooperative
> > > > > > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few
> > questions:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> > > > > > Alternatives'
> > > > > > > > > > > section
> > > > > > > > > > > > > > >> describes how
> > > > > > > > > > > > > > >> > > > > > member
> > > > > > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step
> > with
> > > > the
> > > > > > group
> > > > > > > > > epoch
> > > > > > > > > > > > and
> > > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good
> > > > idea
> > > > > > for
> > > > > > > > the
> > > > > > > > > > > > reasons
> > > > > > > > > > > > > > >> described
> > > > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is
> > > > incrementally
> > > > > > > > > assigning
> > > > > > > > > > > > > > >> partitions to a
> > > > > > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each
> > > > incremental
> > > > > > > > > > > assignment
> > > > > > > > > > > > > use?
> > > > > > > > > > > > > > >> Are
> > > > > > > > > > > > > > >> > > > member
> > > > > > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > a single member
> > epoch
> > > > can
> > > > > > > > > correspond
> > > > > > > > > > > to
> > > > > > > > > > > > > > >> multiple
> > > > > > > > > > > > > > >> > > > > > different
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > larger)
> > assignments?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the
> > Assignor's
> > > > > > 'Reason'
> > > > > > > > > field
> > > > > > > > > > > > > opaque
> > > > > > > > > > > > > > >> to the
> > > > > > > > > > > > > > >> > > > group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> > > > > > client-side
> > > > > > > > > > > assignor
> > > > > > > > > > > > > > >> > > > implementations
> > > > > > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and
> > how
> > > > is
> > > > > > its
> > > > > > > > > common
> > > > > > > > > > > > > meaning
> > > > > > > > > > > > > > >> agreed
> > > > > > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a
> > distinct
> > > > > > Reason
> > > > > > > > > field
> > > > > > > > > > > over
> > > > > > > > > > > > > > >> including
> > > > > > > > > > > > > > >> > > > such
> > > > > > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> > > > > > included in
> > > > > > > > > the
> > > > > > > > > > > > KIP:
> > > > > > > > > > > > > > >> "Thanks to
> > > > > > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client side
> > assignor is
> > > > > > > > entirely
> > > > > > > > > > > driven
> > > > > > > > > > > > > by
> > > > > > > > > > > > > > >> the group
> > > > > > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no
> > longer
> > > > > > > > > responsible for
> > > > > > > > > > > > > > >> maintaining
> > > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > > >> > > > > > state
> > > > > > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned
> > partitions."
> > > > Does
> > > > > > this
> > > > > > > > > mean
> > > > > > > > > > > > > that the
> > > > > > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate
> > additional
> > > > > > > > > non-Metadata
> > > > > > > > > > > > state
> > > > > > > > > > > > > > >> (such as
> > > > > > > > > > > > > > >> > > > > > partition
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics,
> > > > config
> > > > > > > > > topics,
> > > > > > > > > > > > etc),
> > > > > > > > > > > > > or
> > > > > > > > > > > > > > >> that
> > > > > > > > > > > > > > >> > > > > > additional
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be
> > > > used?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that
> > there are
> > > > > > > > separate
> > > > > > > > > > > > classes
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > > > > > >> > > >
> > > > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap
> > significantly.
> > > > Is
> > > > > > it
> > > > > > > > > possible
> > > > > > > > > > > > for
> > > > > > > > > > > > > > >> these two
> > > > > > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would
> > > > serve
> > > > > > to
> > > > > > > > > promote
> > > > > > > > > > > > > feature
> > > > > > > > > > > > > > >> parity
> > > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side
> > assignors,
> > > > and
> > > > > > > > would
> > > > > > > > > also
> > > > > > > > > > > > > > >> facilitate
> > > > > > > > > > > > > > >> > > > > > operational
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > certain
> > situations. For
> > > > > > > > example,
> > > > > > > > > if a
> > > > > > > > > > > > > > >> server-side
> > > > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs
> > a
> > > > patch,
> > > > > > > > > deploying
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> patched
> > > > > > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching one
> > > > consumer
> > > > > > > > group
> > > > > > > > > to a
> > > > > > > > > > > > > > >> client-side
> > > > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> > > > > > patching
> > > > > > > > all
> > > > > > > > > of
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> brokers.
> > > > > > > > > > > > > > >> > > > With
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct
> > > > APIs, a
> > > > > > > > > non-trivial
> > > > > > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if
> > the
> > > > two
> > > > > > APIs
> > > > > > > > > have
> > > > > > > > > > > > > diverged
> > > > > > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > > > > > reimplementation
> > > > > > > > > > > would
> > > > > > > > > > > > > not be
> > > > > > > > > > > > > > >> > > > possible.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > gharris1727@gmail.com
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > github.com/gharris1727
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3,
> > 2022 at
> > > > > > 8:39 AM
> > > > > > > > > Sagar
> > > > > > > > > > > <
> > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi
> > Guozhang/David,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a
> > > > confluence
> > > > > > page
> > > > > > > > to
> > > > > > > > > > > > discuss
> > > > > > > > > > > > > how
> > > > > > > > > > > > > > >> Connect
> > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> > > > > > rebalance
> > > > > > > > > > > protocol.
> > > > > > > > > > > > > > >> Here's the
> > > > > > > > > > > > > > >> > > > page:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >>
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty
> > > > longish
> > > > > > and
> > > > > > > > I
> > > > > > > > > have
> > > > > > > > > > > > > tried
> > > > > > > > > > > > > > >> to keep
> > > > > > > > > > > > > > >> > > > a
> > > > > > > > > > > > > > >> > > > > > format
> > > > > > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me
> > know
> > > > > > what you
> > > > > > > > > > > think.
> > > > > > > > > > > > > Also,
> > > > > > > > > > > > > > >> do you
> > > > > > > > > > > > > > >> > > > > > think this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a
> > separate
> > > > > > > > discussion
> > > > > > > > > > > thread
> > > > > > > > > > > > > or
> > > > > > > > > > > > > > >> is this
> > > > > > > > > > > > > > >> > > > one
> > > > > > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26,
> > 2022
> > > > at
> > > > > > 7:37
> > > > > > > > AM
> > > > > > > > > > > > Sagar <
> > > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so
> > much
> > > > for
> > > > > > the
> > > > > > > > > doc on
> > > > > > > > > > > > > Kafka
> > > > > > > > > > > > > > >> Streams.
> > > > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and
> > come
> > > > up
> > > > > > with
> > > > > > > > > such a
> > > > > > > > > > > > > > >> document.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26,
> > > > 2022 at
> > > > > > > > 4:47
> > > > > > > > > AM
> > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > > >> Wang <
> > > > > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be
> > great
> > > > if
> > > > > > you
> > > > > > > > > could
> > > > > > > > > > > > come
> > > > > > > > > > > > > > >> back with
> > > > > > > > > > > > > > >> > > > some
> > > > > > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the
> > > > Connect
> > > > > > side
> > > > > > > > > > > > > integration
> > > > > > > > > > > > > > >> with
> > > > > > > > > > > > > > >> > > > the new
> > > > > > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides
> > > > leveraging on
> > > > > > the
> > > > > > > > > new
> > > > > > > > > > > > > "protocol
> > > > > > > > > > > > > > >> type"
> > > > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > > > >> > > > > > did not
> > > > > > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect
> > side
> > > > > > > > > > > implementations.
> > > > > > > > > > > > > For
> > > > > > > > > > > > > > >> Streams
> > > > > > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > > >>
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for
> > your
> > > > > > analysis
> > > > > > > > > on
> > > > > > > > > > > > > Connect.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul
> > 19,
> > > > 2022
> > > > > > at
> > > > > > > > > 10:48 PM
> > > > > > > > > > > > > Sagar <
> > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > sagarmeansocean@gmail.com
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you
> > for
> > > > your
> > > > > > > > > response.
> > > > > > > > > > > The
> > > > > > > > > > > > > > >> reason I
> > > > > > > > > > > > > > >> > > > thought
> > > > > > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new
> > scheme
> > > > is
> > > > > > that
> > > > > > > > > even
> > > > > > > > > > > > > today the
> > > > > > > > > > > > > > >> > > > connect
> > > > > > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending
> > from
> > > > > > > > > > > > > AbstractCoordinator to
> > > > > > > > > > > > > > >> empower
> > > > > > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > tasks/connectors.
> > > > > > The
> > > > > > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> > > > > > metadata()
> > > > > > > > > method
> > > > > > > > > > > > by
> > > > > > > > > > > > > > >> plumbing
> > > > > > > > > > > > > > >> > > > into
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > JoinGroupRequestProtocol.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the
> > > > changes
> > > > > > to
> > > > > > > > > support
> > > > > > > > > > > > > > >> connect would
> > > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes
> > in
> > > > > > streams
> > > > > > > > > mainly
> > > > > > > > > > > > > because
> > > > > > > > > > > > > > >> of the
> > > > > > > > > > > > > > >> > > > > > Client
> > > > > > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in
> > both.
> > > > At an
> > > > > > > > > > > > implementation
> > > > > > > > > > > > > > >> level, we
> > > > > > > > > > > > > > >> > > > > > might
> > > > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to
> > get
> > > > onto
> > > > > > this
> > > > > > > > > new
> > > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > > >> > > > protocol
> > > > > > > > > > > > > > >> > > > > > like
> > > > > > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > > > > > > request/response and
> > > > > > > > > > > > > > >> SyncGroup and
> > > > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > ConsumerGroupHeartbeat
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc
> > again on
> > > > > > similar
> > > > > > > > > lines
> > > > > > > > > > > > to
> > > > > > > > > > > > > > >> streams (or
> > > > > > > > > > > > > > >> > > > > > there
> > > > > > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> > > > > > perform a
> > > > > > > > > > > detailed
> > > > > > > > > > > > > > >> analysis of
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > same
> > > > > > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate
> > > > discussion
> > > > > > > > > thread for
> > > > > > > > > > > > > that
> > > > > > > > > > > > > > >> as that
> > > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let
> > me
> > > > know
> > > > > > if
> > > > > > > > > that
> > > > > > > > > > > > sounds
> > > > > > > > > > > > > > >> good to
> > > > > > > > > > > > > > >> > > > you.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul
> > 15,
> > > > > > 2022 at
> > > > > > > > > 5:47
> > > > > > > > > > > PM
> > > > > > > > > > > > > David
> > > > > > > > > > > > > > >> Jacot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for
> > > > your
> > > > > > > > > comments.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes.
> > That
> > > > > > refers to
> > > > > > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The
> > idea
> > > > is to
> > > > > > > > > > > transition C
> > > > > > > > > > > > > from
> > > > > > > > > > > > > > >> his
> > > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target
> > > > assignment
> > > > > > when
> > > > > > > > > he
> > > > > > > > > > > can
> > > > > > > > > > > > > move
> > > > > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > > > > > >> > > > > > When
> > > > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > > > assignment
> > > > > > is
> > > > > > > > > updated
> > > > > > > > > > > > and
> > > > > > > > > > > > > > >> persisted
> > > > > > > > > > > > > > >> > > > > > with all
> > > > > > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions
> > > > even if
> > > > > > > > they
> > > > > > > > > are
> > > > > > > > > > > > not
> > > > > > > > > > > > > all
> > > > > > > > > > > > > > >> revoked
> > > > > > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > > > assignment
> > > > > > > > > becomes
> > > > > > > > > > > the
> > > > > > > > > > > > > target
> > > > > > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > optimization
> > > > to
> > > > > > avoid
> > > > > > > > > having
> > > > > > > > > > > > to
> > > > > > > > > > > > > > >> write all
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples
> > are
> > > > > > based on
> > > > > > > > > the
> > > > > > > > > > > > > persisted
> > > > > > > > > > > > > > >> state
> > > > > > > > > > > > > > >> > > > so I
> > > > > > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion.
> > > > Let me
> > > > > > see
> > > > > > > > > if I
> > > > > > > > > > > can
> > > > > > > > > > > > > > >> improve
> > > > > > > > > > > > > > >> > > > this in
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3)
> > Regarding
> > > > > > Connect,
> > > > > > > > it
> > > > > > > > > > > could
> > > > > > > > > > > > > > >> reuse the
> > > > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor
> > if it
> > > > > > fits in
> > > > > > > > > the
> > > > > > > > > > > > > > >> protocol. The
> > > > > > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > topicid-partitions +
> > > > > > > > > > > metadata,
> > > > > > > > > > > > > could
> > > > > > > > > > > > > > >> > > > Connect
> > > > > > > > > > > > > > >> > > > > > fit
> > > > > > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri,
> > Jul
> > > > 15,
> > > > > > 2022
> > > > > > > > at
> > > > > > > > > 1:55
> > > > > > > > > > > > PM
> > > > > > > > > > > > > > >> Sagar <
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi
> > David,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks
> > for
> > > > the
> > > > > > KIP.
> > > > > > > > I
> > > > > > > > > just
> > > > > > > > > > > > had
> > > > > > > > > > > > > > >> minor
> > > > > > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In
> > the
> > > > > > Assignment
> > > > > > > > > Error
> > > > > > > > > > > > > > >> section in
> > > > > > > > > > > > > > >> > > > Client
> > > > > > > > > > > > > > >> > > > > > Side
> > > > > > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you
> > > > mentioned
> > > > > > => `In
> > > > > > > > > this
> > > > > > > > > > > > > case,
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > > > >> > > > > > side
> > > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error
> > to the
> > > > > > group
> > > > > > > > > > > > > coordinator`.
> > > > > > > > > > > > > > >> In this
> > > > > > > > > > > > > > >> > > > > > case are
> > > > > > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> > > > > > returning
> > > > > > > > an
> > > > > > > > > > > > > > >> AssignmentError
> > > > > > > > > > > > > > >> > > > that's
> > > > > > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes,
> > do
> > > > you
> > > > > > think
> > > > > > > > > it
> > > > > > > > > > > > would
> > > > > > > > > > > > > > >> make sense
> > > > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In
> > the
> > > > Case
> > > > > > > > Studies
> > > > > > > > > > > > > section, I
> > > > > > > > > > > > > > >> have a
> > > > > > > > > > > > > > >> > > > > > slight
> > > > > > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others
> > have
> > > > the
> > > > > > > > same.
> > > > > > > > > > > > Consider
> > > > > > > > > > > > > > >> this step:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> > > > > > heartbeats,
> > > > > > > > the
> > > > > > > > > > > group
> > > > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> > > > > > partitions
> > > > > > > > to
> > > > > > > > > > > > > revoke. It
> > > > > > > > > > > > > > >> > > > persists
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> > Group
> > > > > > (epoch=3)
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > A
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > B
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > C
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> > Target
> > > > > > > > Assignment
> > > > > > > > > > > > > (epoch=3)
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > A -
> > > > > > > > > > > > partitions=[foo-0]
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > B -
> > > > > > > > > > > > partitions=[foo-2]
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > C -
> > > > > > > > > > > > partitions=[foo-1]
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> > Member
> > > > > > > > Assignment
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > A -
> > > > > > epoch=2,
> > > > > > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > B -
> > > > > > epoch=3,
> > > > > > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> > C -
> > > > > > epoch=3,
> > > > > > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> > > > > > heartbeats,
> > > > > > > > it
> > > > > > > > > > > > > transitions
> > > > > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > > > > > >> > > > > > but
> > > > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > Here,it's
> > > > > > mentioned
> > > > > > > > > that
> > > > > > > > > > > > > member C
> > > > > > > > > > > > > > >> can't
> > > > > > > > > > > > > > >> > > > get
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based
> > on the
> > > > > > > > > description
> > > > > > > > > > > > > above,
> > > > > > > > > > > > > > >> it seems
> > > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be
> > > > better
> > > > > > to
> > > > > > > > > remove
> > > > > > > > > > > it
> > > > > > > > > > > > > and
> > > > > > > > > > > > > > >> > > > populate it
> > > > > > > > > > > > > > >> > > > > > only
> > > > > > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see
> > this
> > > > in a
> > > > > > lot
> > > > > > > > of
> > > > > > > > > > > other
> > > > > > > > > > > > > > >> places, so
> > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > >> > > > > > I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > Regarding
> > > > > > connect ,
> > > > > > > > it
> > > > > > > > > > > might
> > > > > > > > > > > > > be
> > > > > > > > > > > > > > >> out of
> > > > > > > > > > > > > > >> > > > scope
> > > > > > > > > > > > > > >> > > > > > of
> > > > > > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> > > > > > understood it
> > > > > > > > > would
> > > > > > > > > > > > > > >> probably be
> > > > > > > > > > > > > > >> > > > > > running in
> > > > > > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on
> > the
> > > > new
> > > > > > > > > rebalance
> > > > > > > > > > > > > > >> protocol as it
> > > > > > > > > > > > > > >> > > > has
> > > > > > > > > > > > > > >> > > > > > its
> > > > > > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > Assignors(Eager
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > > > > > IncrementalCooperative).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri,
> > Jul
> > > > 15,
> > > > > > 2022
> > > > > > > > > at
> > > > > > > > > > > 5:00
> > > > > > > > > > > > > PM
> > > > > > > > > > > > > > >> David
> > > > > > > > > > > > > > >> > > > Jacot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> > > > > > Hector! Our
> > > > > > > > > goal
> > > > > > > > > > > is
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > >> move
> > > > > > > > > > > > > > >> > > > forward
> > > > > > > > > > > > > > >> > > > > > with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > instead of
> > > > > > relying
> > > > > > > > > on
> > > > > > > > > > > one
> > > > > > > > > > > > > > >> generic API.
> > > > > > > > > > > > > > >> > > > For
> > > > > > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact
> > same
> > > > > > pattern
> > > > > > > > > and
> > > > > > > > > > > > > > >> reuse/share the
> > > > > > > > > > > > > > >> > > > core
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server
> > > > side.
> > > > > > For
> > > > > > > > the
> > > > > > > > > > > > schema
> > > > > > > > > > > > > > >> registry, I
> > > > > > > > > > > > > > >> > > > > > think
> > > > > > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > having a
> > > > > > tailored
> > > > > > > > > API to
> > > > > > > > > > > > do
> > > > > > > > > > > > > > >> simple
> > > > > > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On
> > Fri,
> > > > Jul
> > > > > > 15,
> > > > > > > > > 2022 at
> > > > > > > > > > > > > 10:22
> > > > > > > > > > > > > > >> AM Ismael
> > > > > > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > Three
> > > > quick
> > > > > > > > > comments:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1.
> > > > Regarding
> > > > > > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > > > > > >> > > > vs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > com.google.re2j.Pattern,
> > > > > > > > we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > document the
> > > > > > > > > > > differences
> > > > > > > > > > > > > in
> > > > > > > > > > > > > > >> more
> > > > > > > > > > > > > > >> > > > detail
> > > > > > > > > > > > > > >> > > > > > before
> > > > > > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > another.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That
> > > > said,
> > > > > > if
> > > > > > > > > people
> > > > > > > > > > > > pass
> > > > > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > semantics
> > > > > > to be
> > > > > > > > > > > honored.
> > > > > > > > > > > > > If
> > > > > > > > > > > > > > >> we are
> > > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > should
> > > > > > consider
> > > > > > > > > adding
> > > > > > > > > > > > an
> > > > > > > > > > > > > > >> overload
> > > > > > > > > > > > > > >> > > > with
> > > > > > > > > > > > > > >> > > > > > our own
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > think
> > > > we'd
> > > > > > want
> > > > > > > > to
> > > > > > > > > > > > expose
> > > > > > > > > > > > > > >> re2j's at
> > > > > > > > > > > > > > >> > > > this
> > > > > > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2.
> > > > Regarding
> > > > > > > > topic
> > > > > > > > > > > ids,
> > > > > > > > > > > > > any
> > > > > > > > > > > > > > >> major new
> > > > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it
> > and
> > > > > > should
> > > > > > > > > handle
> > > > > > > > > > > the
> > > > > > > > > > > > > topic
> > > > > > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > part we
> > > > > > need to
> > > > > > > > > > > handle.
> > > > > > > > > > > > I
> > > > > > > > > > > > > > >> agree with
> > > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to
> > the
> > > > > > relevant
> > > > > > > > > > > > protocols
> > > > > > > > > > > > > > >> that don't
> > > > > > > > > > > > > > >> > > > > > have it
> > > > > > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > focus
> > > > on the
> > > > > > > > > internals
> > > > > > > > > > > > > versus
> > > > > > > > > > > > > > >> adding
> > > > > > > > > > > > > > >> > > > new
> > > > > > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > (unless
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we
> > find
> > > > that
> > > > > > > > > adding
> > > > > > > > > > > new
> > > > > > > > > > > > > APIs
> > > > > > > > > > > > > > >> is
> > > > > > > > > > > > > > >> > > > required
> > > > > > > > > > > > > > >> > > > > > for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3.
> > I am
> > > > > > still
> > > > > > > > not
> > > > > > > > > sure
> > > > > > > > > > > > > about
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > powerful for
> > > > > > > > > configs
> > > > > > > > > > > to
> > > > > > > > > > > > be
> > > > > > > > > > > > > > >> > > > centralized
> > > > > > > > > > > > > > >> > > > > > in the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > reasons
> > > > > > > > > (auditability,
> > > > > > > > > > > > > > >> visibility,
> > > > > > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure
> > > > about
> > > > > > > > > > > automatically
> > > > > > > > > > > > > > >> deleting
> > > > > > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > recovered. A
> > > > > > > > good
> > > > > > > > > > > > property
> > > > > > > > > > > > > > >> for modern
> > > > > > > > > > > > > > >> > > > > > systems
> > > > > > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > unrecoverable
> > > > > > > > data
> > > > > > > > > > > loss
> > > > > > > > > > > > > > >> scenarios.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > Ismael
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On
> > Wed,
> > > > Jul
> > > > > > 13,
> > > > > > > > > 2022
> > > > > > > > > > > at
> > > > > > > > > > > > > 3:47
> > > > > > > > > > > > > > >> PM David
> > > > > > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Thanks
> > > > > > > > > Guozhang. My
> > > > > > > > > > > > > answers
> > > > > > > > > > > > > > >> are
> > > > > > > > > > > > > > >> > > > below:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > 1)
> > > > the
> > > > > > > > > migration
> > > > > > > > > > > > path,
> > > > > > > > > > > > > > >> especially
> > > > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > flag
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > > enable
> > > > > > > > the
> > > > > > > > > new
> > > > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > protocols /
> > > > > > > > > rpcs
> > > > > > > > > > > and
> > > > > > > > > > > > > old
> > > > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > group.
> > > > > > How
> > > > > > > > the
> > > > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using
> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > new
> > > > > > protocol
> > > > > > > > > is
> > > > > > > > > > > > > something
> > > > > > > > > > > > > > >> we
> > > > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > Noted. I
> > > > > > just
> > > > > > > > > > > > published
> > > > > > > > > > > > > a
> > > > > > > > > > > > > > >> new
> > > > > > > > > > > > > > >> > > > version
> > > > > > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > details
> > > > > > about
> > > > > > > > > this.
> > > > > > > > > > > > See
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > and
> > > > the
> > > > > > > > > > > > "Compatibility,
> > > > > > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > > > > > >> > > > > > and
> > > > > > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I
> > > > have to
> > > > > > > > think
> > > > > > > > > > > > through
> > > > > > > > > > > > > a
> > > > > > > > > > > > > > >> few cases
> > > > > > > > > > > > > > >> > > > > > now but
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > mechanism
> > > > > > > > > should be
> > > > > > > > > > > > > > >> understandable.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > 2)
> > > > the
> > > > > > usage
> > > > > > > > > of
> > > > > > > > > > > > topic
> > > > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > part of
> > > > > > RPCs
> > > > > > > > > and
> > > > > > > > > > > > admin
> > > > > > > > > > > > > > >> client,
> > > > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > APIs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > > consumers
> > > > > > > > > yet.
> > > > > > > > > > > I
> > > > > > > > > > > > > think
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > client
> > > > > > to be
> > > > > > > > > > > > > maintaining
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > leverage
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > on
> > > > all
> > > > > > the
> > > > > > > > > > > augmented
> > > > > > > > > > > > > > >> existing
> > > > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > secondly,
> > > > > > > > > should
> > > > > > > > > > > we
> > > > > > > > > > > > > ever
> > > > > > > > > > > > > > >> consider
> > > > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > public
> > > > > > APIs
> > > > > > > > as
> > > > > > > > > > > well
> > > > > > > > > > > > > (both
> > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > rebalance
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > listener for
> > > > > > > > > cases
> > > > > > > > > > > > > like
> > > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > deletion-and-recreation).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> > > > > > Assuming
> > > > > > > > > that we
> > > > > > > > > > > > > would
> > > > > > > > > > > > > > >> include
> > > > > > > > > > > > > > >> > > > > > converting
> > > > > > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > RPCs
> > > > to
> > > > > > using
> > > > > > > > > topic
> > > > > > > > > > > > ids
> > > > > > > > > > > > > in
> > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > >> > > > KIP,
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > fully
> > > > > > operate
> > > > > > > > > with
> > > > > > > > > > > > topic
> > > > > > > > > > > > > > >> ids. That
> > > > > > > > > > > > > > >> > > > > > being
> > > > > > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > the
> > > > topics
> > > > > > > > > names in
> > > > > > > > > > > > > various
> > > > > > > > > > > > > > >> APIs so
> > > > > > > > > > > > > > >> > > > > > having a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > seems
> > > > > > > > > inevitable to
> > > > > > > > > > > > me.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > b) I
> > > > don't
> > > > > > > > have
> > > > > > > > > a
> > > > > > > > > > > > strong
> > > > > > > > > > > > > > >> opinion on
> > > > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > beyond the
> > > > > > > > > scope of
> > > > > > > > > > > > this
> > > > > > > > > > > > > > >> KIP. I
> > > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > > >> > > > > > rather
> > > > > > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > here
> > > > and
> > > > > > we
> > > > > > > > can
> > > > > > > > > > > > consider
> > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Coming
> > > > > > back to
> > > > > > > > > > > > Ismael's
> > > > > > > > > > > > > > >> point about
> > > > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > > > > > >> > > > think
> > > > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > favour of
> > > > > > it.
> > > > > > > > > The
> > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > >> will
> > > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > the
> > > > topics
> > > > > > > > > exists
> > > > > > > > > > > > before
> > > > > > > > > > > > > > >> passing
> > > > > > > > > > > > > > >> > > > them
> > > > > > > > > > > > > > >> > > > > > into
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > protocol.
> > > > > > > > > Obviously,
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> > > > coordinator
> > > > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > not
> > > > really
> > > > > > > > have
> > > > > > > > > a
> > > > > > > > > > > way
> > > > > > > > > > > > to
> > > > > > > > > > > > > > >> reject an
> > > > > > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > I'm
> > > > > > agreeing
> > > > > > > > > with
> > > > > > > > > > > > > David
> > > > > > > > > > > > > > >> on all
> > > > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > > > >> question:
> > > > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > deprecate
> > > > > > > > the
> > > > > > > > > > > > > subscribe
> > > > > > > > > > > > > > >> API with
> > > > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > Pattern
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > while
> > > > > > just
> > > > > > > > > > > > documenting
> > > > > > > > > > > > > > >> that our
> > > > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > server.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > Since
> > > > > > the
> > > > > > > > > > > > incompatible
> > > > > > > > > > > > > > >> case is a
> > > > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > overloaded
> > > > > > > > > > > `String`
> > > > > > > > > > > > > based
> > > > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > various
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > invalid
> > > > > > > > > regexes.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > That
> > > > could
> > > > > > > > > work. I
> > > > > > > > > > > > have
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > >> look at
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > engines to
> > > > > > > > > better
> > > > > > > > > > > > > > >> understand the
> > > > > > > > > > > > > > >> > > > > > potential
> > > > > > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > understanding
> > > > is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > that
> > > > would
> > > > > > > > work
> > > > > > > > > for
> > > > > > > > > > > > all
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> basic
> > > > > > > > > > > > > > >> > > > > > regular
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > differences
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > between
> > > > > > the
> > > > > > > > two
> > > > > > > > > are
> > > > > > > > > > > > > mainly
> > > > > > > > > > > > > > >> about
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > various
> > > > > > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > wonder
> > > > > > what
> > > > > > > > > other
> > > > > > > > > > > > people
> > > > > > > > > > > > > > >> think
> > > > > > > > > > > > > > >> > > > about
> > > > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Best,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > David
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On
> > > > Tue,
> > > > > > Jul
> > > > > > > > 12,
> > > > > > > > > 2022
> > > > > > > > > > > > at
> > > > > > > > > > > > > > >> 11:28 PM
> > > > > > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > wangguoz@gmail.com
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > Thanks
> > > > > > > > David!
> > > > > > > > > I
> > > > > > > > > > > > think
> > > > > > > > > > > > > on
> > > > > > > > > > > > > > >> the high
> > > > > > > > > > > > > > >> > > > > > level
> > > > > > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > > > > concretize
> > > > > > > > > a
> > > > > > > > > > > bit
> > > > > > > > > > > > > more:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > 1)
> > > > the
> > > > > > > > > migration
> > > > > > > > > > > > path,
> > > > > > > > > > > > > > >> especially
> > > > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > flag
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > > enable
> > > > > > > > the
> > > > > > > > > new
> > > > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > protocols /
> > > > > > > > > rpcs
> > > > > > > > > > > and
> > > > > > > > > > > > > old
> > > > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > group.
> > > > > > How
> > > > > > > > the
> > > > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using
> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > new
> > > > > > protocol
> > > > > > > > > is
> > > > > > > > > > > > > something
> > > > > > > > > > > > > > >> we
> > > > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > 2)
> > > > the
> > > > > > usage
> > > > > > > > > of
> > > > > > > > > > > > topic
> > > > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > part of
> > > > > > RPCs
> > > > > > > > > and
> > > > > > > > > > > > admin
> > > > > > > > > > > > > > >> client,
> > > > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > APIs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > > consumers
> > > > > > > > > yet.
> > > > > > > > > > > I
> > > > > > > > > > > > > think
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > client
> > > > > > to be
> > > > > > > > > > > > > maintaining
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > leverage
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > on
> > > > all
> > > > > > the
> > > > > > > > > > > augmented
> > > > > > > > > > > > > > >> existing
> > > > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > secondly,
> > > > > > > > > should
> > > > > > > > > > > we
> > > > > > > > > > > > > ever
> > > > > > > > > > > > > > >> consider
> > > > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > public
> > > > > > APIs
> > > > > > > > as
> > > > > > > > > > > well
> > > > > > > > > > > > > (both
> > > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > rebalance
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > listener for
> > > > > > > > > cases
> > > > > > > > > > > > > like
> > > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > deletion-and-recreation).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > I'm
> > > > > > agreeing
> > > > > > > > > with
> > > > > > > > > > > > > David
> > > > > > > > > > > > > > >> on all
> > > > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > > > >> question:
> > > > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > deprecate
> > > > > > > > the
> > > > > > > > > > > > > subscribe
> > > > > > > > > > > > > > >> API with
> > > > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > Pattern
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > while
> > > > > > just
> > > > > > > > > > > > documenting
> > > > > > > > > > > > > > >> that our
> > > > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > server.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > Since
> > > > > > the
> > > > > > > > > > > > incompatible
> > > > > > > > > > > > > > >> case is a
> > > > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > overloaded
> > > > > > > > > > > `String`
> > > > > > > > > > > > > based
> > > > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > various
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > invalid
> > > > > > > > > regexes.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > Guozhang
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > On
> > > > Tue,
> > > > > > Jul
> > > > > > > > > 12,
> > > > > > > > > > > 2022
> > > > > > > > > > > > > at
> > > > > > > > > > > > > > >> 5:23 AM
> > > > > > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > Hi
> > > > > > Ismael,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > Thanks for
> > > > > > > > > your
> > > > > > > > > > > > > > >> feedback. Let
> > > > > > > > > > > > > > >> > > > me
> > > > > > > > > > > > > > >> > > > > > answer
> > > > > > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 1. I
> > > > > > > > think
> > > > > > > > > > > it's
> > > > > > > > > > > > > > >> premature to
> > > > > > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > deprecation
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > removal
> > > > > > > > > of the
> > > > > > > > > > > > > > >> existing group
> > > > > > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > affects a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > core
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > client
> > > > > > > > > > > protocol
> > > > > > > > > > > > > and
> > > > > > > > > > > > > > >> hence
> > > > > > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > dependent
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > how
> > > > > > > > > quickly
> > > > > > > > > > > > > > >> applications
> > > > > > > > > > > > > > >> > > > migrate
> > > > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > That
> > > > > > makes
> > > > > > > > > > > sense.
> > > > > > > > > > > > I
> > > > > > > > > > > > > will
> > > > > > > > > > > > > > >> > > > remove it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 2.
> > > > > > The
> > > > > > > > KIP
> > > > > > > > > > > says
> > > > > > > > > > > > we
> > > > > > > > > > > > > > >> intend to
> > > > > > > > > > > > > > >> > > > > > release
> > > > > > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > clear
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > why. If
> > > > > > > > we
> > > > > > > > > > > added
> > > > > > > > > > > > > that
> > > > > > > > > > > > > > >> as a
> > > > > > > > > > > > > > >> > > > way to
> > > > > > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > remove
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > the
> > > > > > > > group
> > > > > > > > > > > > > protocol, I
> > > > > > > > > > > > > > >> also
> > > > > > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Let me
> > > > > > > > > explain
> > > > > > > > > > > my
> > > > > > > > > > > > > > >> reasoning. As
> > > > > > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > coordinator
> > > > > > > > > in
> > > > > > > > > > > > Java
> > > > > > > > > > > > > > >> while we
> > > > > > > > > > > > > > >> > > > > > implement
> > > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > that
> > > > > > the
> > > > > > > > > > > internals
> > > > > > > > > > > > > will
> > > > > > > > > > > > > > >> be
> > > > > > > > > > > > > > >> > > > slightly
> > > > > > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > model).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > Therefore, I
> > > > > > > > > > > > wanted
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > >> tighten
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > coordinator
> > > > > > > > > to
> > > > > > > > > > > the
> > > > > > > > > > > > > new
> > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > alternative
> > > > > > > > > > > would
> > > > > > > > > > > > > be to
> > > > > > > > > > > > > > >> use a
> > > > > > > > > > > > > > >> > > > flag
> > > > > > > > > > > > > > >> > > > > > to do
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > relying
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > on the
> > > > > > > > > software
> > > > > > > > > > > > > upgrade.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 3.
> > > > > > We
> > > > > > > > > need to
> > > > > > > > > > > > > flesh
> > > > > > > > > > > > > > >> out the
> > > > > > > > > > > > > > >> > > > > > details of
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > like
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > we're
> > > > > > > > > saying
> > > > > > > > > > > we
> > > > > > > > > > > > > will
> > > > > > > > > > > > > > >> support
> > > > > > > > > > > > > > >> > > > > > online
> > > > > > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > explain
> > > > > > > > > this
> > > > > > > > > > > in
> > > > > > > > > > > > > > >> detail. It
> > > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > > >> > > > > > also
> > > > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > it's
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > easier.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Yes,
> > > > > > we
> > > > > > > > will
> > > > > > > > > > > > support
> > > > > > > > > > > > > > >> online
> > > > > > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > a
> > > > > > group
> > > > > > > > > using
> > > > > > > > > > > the
> > > > > > > > > > > > > old
> > > > > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > protocol.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Let me
> > > > > > > > > briefly
> > > > > > > > > > > > > explain
> > > > > > > > > > > > > > >> how that
> > > > > > > > > > > > > > >> > > > > > will work
> > > > > > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically
> > a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > four
> > > > > > step
> > > > > > > > > > > process:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > 1. The
> > > > > > > > > cluster
> > > > > > > > > > > > must
> > > > > > > > > > > > > be
> > > > > > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > new
> > > > > > group
> > > > > > > > > > > > > coordinator.
> > > > > > > > > > > > > > >> Both the
> > > > > > > > > > > > > > >> > > > > > old and
> > > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > support
> > > > > > > > the
> > > > > > > > > old
> > > > > > > > > > > > > > >> protocol and
> > > > > > > > > > > > > > >> > > > rely
> > > > > > > > > > > > > > >> > > > > > on the
> > > > > > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > so
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > they
> > > > > > can
> > > > > > > > > work
> > > > > > > > > > > > > together.
> > > > > > > > > > > > > > >> This
> > > > > > > > > > > > > > >> > > > point
> > > > > > > > > > > > > > >> > > > > > is an
> > > > > > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > do
> > > > > > this
> > > > > > > > one
> > > > > > > > > live
> > > > > > > > > > > > > > >> because it
> > > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > > >> > > > > > require
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > coordinator
> > > > > > > > > and
> > > > > > > > > > > > > > >> starting up the
> > > > > > > > > > > > > > >> > > > > > new one
> > > > > > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > unavailabilities.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > 2. The
> > > > > > > > > cluster's
> > > > > > > > > > > > > > >> metadata
> > > > > > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > to
> > > > > > enable
> > > > > > > > > the
> > > > > > > > > > > new
> > > > > > > > > > > > > > >> protocol.
> > > > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > terminated
> > > > > > > > > > > because
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> old
> > > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > protocol.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > 3. The
> > > > > > > > > consumers
> > > > > > > > > > > > > must be
> > > > > > > > > > > > > > >> > > > upgraded
> > > > > > > > > > > > > > >> > > > > > to a
> > > > > > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > migration
> > > > > > > > > (must
> > > > > > > > > > > > have
> > > > > > > > > > > > > > >> KIP-792).
> > > > > > > > > > > > > > >> > > > If
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > Nothing
> > > > > > > > > must be
> > > > > > > > > > > > > done at
> > > > > > > > > > > > > > >> this
> > > > > > > > > > > > > > >> > > > point.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > 4. The
> > > > > > > > > consumers
> > > > > > > > > > > > > must
> > > > > > > > > > > > > > >> be rolled
> > > > > > > > > > > > > > >> > > > > > with the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > consumer
> > > > > > > > > group
> > > > > > > > > > > is
> > > > > > > > > > > > > > >> automatically
> > > > > > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > using
> > > > > > the
> > > > > > > > > new
> > > > > > > > > > > > > protocol
> > > > > > > > > > > > > > >> joins
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > group.
> > > > > > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > old
> > > > > > > > > protocol are
> > > > > > > > > > > > > being
> > > > > > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > > > > > >> > > > > > the old
> > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > new
> > > > > > one.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Let me
> > > > > > > > > clarify
> > > > > > > > > > > all
> > > > > > > > > > > > > of
> > > > > > > > > > > > > > >> this in
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 4.
> > > > > > I am
> > > > > > > > > happy
> > > > > > > > > > > > > that we
> > > > > > > > > > > > > > >> are
> > > > > > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > server,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > but
> > > > > > it
> > > > > > > > > seems
> > > > > > > > > > > > like
> > > > > > > > > > > > > > >> there
> > > > > > > > > > > > > > >> > > > could be
> > > > > > > > > > > > > > >> > > > > > some
> > > > > > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > issues.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Will
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > have a
> > > > > > > > > > > mechanism
> > > > > > > > > > > > > for
> > > > > > > > > > > > > > >> users to
> > > > > > > > > > > > > > >> > > > > > detect
> > > > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > regex
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > before
> > > > > > > > > > > switching
> > > > > > > > > > > > > to
> > > > > > > > > > > > > > >> the new
> > > > > > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > I
> > > > > > think
> > > > > > > > > that I
> > > > > > > > > > > am
> > > > > > > > > > > > a
> > > > > > > > > > > > > bit
> > > > > > > > > > > > > > >> more
> > > > > > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > believe
> > > > > > > > > that the
> > > > > > > > > > > > > > >> majority of
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > cases
> > > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > work
> > > > > > with
> > > > > > > > > the
> > > > > > > > > > > new
> > > > > > > > > > > > > > >> engine. The
> > > > > > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > and
> > > > > > reject
> > > > > > > > > the
> > > > > > > > > > > > > consumer
> > > > > > > > > > > > > > >> if the
> > > > > > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > migration
> > > > > > > > > path,
> > > > > > > > > > > in
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> worst
> > > > > > > > > > > > > > >> > > > case,
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > joining
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > the
> > > > > > group
> > > > > > > > > will
> > > > > > > > > > > be
> > > > > > > > > > > > > > >> rejected.
> > > > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > > > >> > > > > > should
> > > > > > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > defence, I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > would
> > > > > > say.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > One
> > > > > > way
> > > > > > > > for
> > > > > > > > > > > > > customers to
> > > > > > > > > > > > > > >> > > > validate
> > > > > > > > > > > > > > >> > > > > > their
> > > > > > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > prod
> > > > > > would
> > > > > > > > > be to
> > > > > > > > > > > > > test
> > > > > > > > > > > > > > >> them with
> > > > > > > > > > > > > > >> > > > > > another
> > > > > > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > could
> > > > > > be
> > > > > > > > > done
> > > > > > > > > > > in a
> > > > > > > > > > > > > > >> pre-prod
> > > > > > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > extend the
> > > > > > > > > > > > > > >> consumer-group tool
> > > > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > mechanism.
> > > > > > > > > Would
> > > > > > > > > > > > > this be
> > > > > > > > > > > > > > >> > > > enough in
> > > > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 5.
> > > > > > > > > Related to
> > > > > > > > > > > > the
> > > > > > > > > > > > > last
> > > > > > > > > > > > > > >> > > > question,
> > > > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > users
> > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > stick
> > > > > > > > > with the
> > > > > > > > > > > > > > >> current regex
> > > > > > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > example,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > it
> > > > > > may
> > > > > > > > be
> > > > > > > > > > > handy
> > > > > > > > > > > > to
> > > > > > > > > > > > > > >> keep using
> > > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > migrations
> > > > > > > > > > > > simple
> > > > > > > > > > > > > and
> > > > > > > > > > > > > > >> then
> > > > > > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > second
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > step.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > I
> > > > > > > > understand
> > > > > > > > > > > your
> > > > > > > > > > > > > point
> > > > > > > > > > > > > > >> but I
> > > > > > > > > > > > > > >> > > > am
> > > > > > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > to
> > > > > > > > actually
> > > > > > > > > stay
> > > > > > > > > > > > in
> > > > > > > > > > > > > > >> this mode.
> > > > > > > > > > > > > > >> > > > That
> > > > > > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > simplifying
> > > > > > > > > the
> > > > > > > > > > > > > client
> > > > > > > > > > > > > > >> because
> > > > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > the
> > > > > > > > > metadata on
> > > > > > > > > > > > the
> > > > > > > > > > > > > > >> client
> > > > > > > > > > > > > > >> > > > side. I
> > > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 6.
> > > > > > When
> > > > > > > > > we say
> > > > > > > > > > > > > that
> > > > > > > > > > > > > > >> the group
> > > > > > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > responsible
> > > > for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > storing
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > the
> > > > > > > > > > > > configurations
> > > > > > > > > > > > > > >> and that
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > group is
> > > > > > > > > > > > deleted.
> > > > > > > > > > > > > > >> Will a
> > > > > > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > > > configurations?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > That's
> > > > > > > > > right.
> > > > > > > > > > > The
> > > > > > > > > > > > > > >> > > > configurations
> > > > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > deleted.
> > > > > > > > > They go
> > > > > > > > > > > > > > >> together.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 7.
> > > > > > Will
> > > > > > > > > the
> > > > > > > > > > > > > choice to
> > > > > > > > > > > > > > >> store
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > coordinator
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > make
> > > > it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > harder
> > > > > > > > to
> > > > > > > > > list
> > > > > > > > > > > > all
> > > > > > > > > > > > > > >> cluster
> > > > > > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > I
> > > > > > don't
> > > > > > > > > think
> > > > > > > > > > > so.
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > >> group
> > > > > > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > configs.
> > > > > > > > If
> > > > > > > > > you
> > > > > > > > > > > > > want to
> > > > > > > > > > > > > > >> know
> > > > > > > > > > > > > > >> > > > all
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > to
> > > > > > ask all
> > > > > > > > > the
> > > > > > > > > > > > group
> > > > > > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > > > > > >> > > > > > You
> > > > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > for
> > > > > > > > > instance.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 8.
> > > > > > How
> > > > > > > > > would
> > > > > > > > > > > > > someone
> > > > > > > > > > > > > > >> > > > configure a
> > > > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > consumers?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Have
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > we
> > > > > > > > > considered
> > > > > > > > > > > > > > >> allowing the
> > > > > > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > Alternatively,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > configs
> > > > > > > > > could
> > > > > > > > > > > be
> > > > > > > > > > > > > > >> decoupled
> > > > > > > > > > > > > > >> > > > from
> > > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Yes.
> > > > > > The
> > > > > > > > > group
> > > > > > > > > > > > will
> > > > > > > > > > > > > be
> > > > > > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > the
> > > > > > > > configs
> > > > > > > > > will
> > > > > > > > > > > > be
> > > > > > > > > > > > > > >> lost after
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > passes.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 9.
> > > > > > Will
> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > > > > > >> > > > > > method
> > > > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > > > > > >> > > > do
> > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > That's a
> > > > > > > > > very
> > > > > > > > > > > > group
> > > > > > > > > > > > > > >> question. I
> > > > > > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > > > >> > > > > > fully
> > > > > > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > we
> > > > > > plan to
> > > > > > > > > use,
> > > > > > > > > > > it
> > > > > > > > > > > > > > >> might be
> > > > > > > > > > > > > > >> > > > better
> > > > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > overload
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > which
> > > > > > > > takes
> > > > > > > > > a
> > > > > > > > > > > > > string.
> > > > > > > > > > > > > > >> We would
> > > > > > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > validation.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > During the
> > > > > > > > > > > > > migration, I
> > > > > > > > > > > > > > >> think
> > > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > regex
> > > > > > and
> > > > > > > > > use
> > > > > > > > > > > it.
> > > > > > > > > > > > > That
> > > > > > > > > > > > > > >> should
> > > > > > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > cases.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 10.
> > > > > > I
> > > > > > > > > agree
> > > > > > > > > > > with
> > > > > > > > > > > > > > >> Justine
> > > > > > > > > > > > > > >> > > > that we
> > > > > > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > switch
> > > > > > > > to
> > > > > > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > > > > > >> > > > > > from the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > can
> > > > > > > > > switch the
> > > > > > > > > > > > > > >> default for
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > feature
> > > > > > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > > > metadata.version
> > > > > > > > > > > > > once
> > > > > > > > > > > > > > >> we
> > > > > > > > > > > > > > >> > > > want to
> > > > > > > > > > > > > > >> > > > > > make
> > > > > > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > My
> > > > > > plan
> > > > > > > > was
> > > > > > > > > to
> > > > > > > > > > > use
> > > > > > > > > > > > > that
> > > > > > > > > > > > > > >> feature
> > > > > > > > > > > > > > >> > > > > > flag
> > > > > > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > development
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > phase. I
> > > > > > > > > should
> > > > > > > > > > > > not
> > > > > > > > > > > > > have
> > > > > > > > > > > > > > >> > > > mentioned
> > > > > > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > an
> > > > > > > > internal
> > > > > > > > > > > config
> > > > > > > > > > > > > for
> > > > > > > > > > > > > > >> it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 11.
> > > > > > Some
> > > > > > > > > of
> > > > > > > > > > > the
> > > > > > > > > > > > > > >> protocol APIs
> > > > > > > > > > > > > > >> > > > > > don't
> > > > > > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > would
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > good to
> > > > > > > > > add
> > > > > > > > > > > that
> > > > > > > > > > > > > for
> > > > > > > > > > > > > > >> > > > consistency.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Noted.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 12.
> > > > > > It
> > > > > > > > is
> > > > > > > > > a
> > > > > > > > > > > bit
> > > > > > > > > > > > > odd
> > > > > > > > > > > > > > >> that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > even
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > though
> > > > > > > > it
> > > > > > > > > > > seems
> > > > > > > > > > > > > to do
> > > > > > > > > > > > > > >> more
> > > > > > > > > > > > > > >> > > > than
> > > > > > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > I
> > > > > > agree.
> > > > > > > > > This is
> > > > > > > > > > > > > how the
> > > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > require
> > > > > > > > > "Read
> > > > > > > > > > > > > Group" to
> > > > > > > > > > > > > > >> join a
> > > > > > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > but I
> > > > > > am
> > > > > > > > not
> > > > > > > > > > > sure
> > > > > > > > > > > > > that
> > > > > > > > > > > > > > >> it is
> > > > > > > > > > > > > > >> > > > worth
> > > > > > > > > > > > > > >> > > > > > it.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 13.
> > > > > > How
> > > > > > > > is
> > > > > > > > > > > topic
> > > > > > > > > > > > > > >> recreation
> > > > > > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > group
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > protocol?
> > > > > > > > > It
> > > > > > > > > > > > > would be
> > > > > > > > > > > > > > >> good to
> > > > > > > > > > > > > > >> > > > > > have a
> > > > > > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Noted.
> > > > > > > > From
> > > > > > > > > a
> > > > > > > > > > > > > protocol
> > > > > > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > topic
> > > > > > id
> > > > > > > > so
> > > > > > > > > it
> > > > > > > > > > > > will
> > > > > > > > > > > > > > >> treat it
> > > > > > > > > > > > > > >> > > > like a
> > > > > > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > only
> > > > > > issue
> > > > > > > > > is
> > > > > > > > > > > that
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > IDs
> > > > > > so the
> > > > > > > > > > > > consumer
> > > > > > > > > > > > > > >> would
> > > > > > > > > > > > > > >> > > > reuse the
> > > > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > that
> > > > > > we
> > > > > > > > > should
> > > > > > > > > > > > > update
> > > > > > > > > > > > > > >> those
> > > > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > > > >> > > > > > as well
> > > > > > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > to
> > > > > > end.
> > > > > > > > That
> > > > > > > > > > > would
> > > > > > > > > > > > > > >> strengthen
> > > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 14.
> > > > > > The
> > > > > > > > > KIP
> > > > > > > > > > > > > mentions
> > > > > > > > > > > > > > >> we will
> > > > > > > > > > > > > > >> > > > > > write the
> > > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > though
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > this is
> > > > > > > > an
> > > > > > > > > > > > > > >> implementation
> > > > > > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > module
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > for
> > > > > > it?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Yes.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 15.
> > > > > > Do
> > > > > > > > we
> > > > > > > > > > > have a
> > > > > > > > > > > > > > >> scalability
> > > > > > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > members
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > new
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > group
> > > > > > > > > protocol
> > > > > > > > > > > > can
> > > > > > > > > > > > > > >> support?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > We
> > > > > > don't
> > > > > > > > > have
> > > > > > > > > > > > > numbers
> > > > > > > > > > > > > > >> at the
> > > > > > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > of
> > > > > > members
> > > > > > > > > per
> > > > > > > > > > > > > group.
> > > > > > > > > > > > > > >> We will
> > > > > > > > > > > > > > >> > > > > > measure
> > > > > > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > implementation.
> > > > > > > > > > > > Note
> > > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > > >> > > > might
> > > > > > > > > > > > > > >> > > > > > have
> > > > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > road
> > > > > > (e.g.
> > > > > > > > > > > offset
> > > > > > > > > > > > > > >> commits).
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > 16.
> > > > > > Did
> > > > > > > > we
> > > > > > > > > > > > > consider
> > > > > > > > > > > > > > >> having
> > > > > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > > > > > SubscribedTopicNames
> > > > > > > > > > > > > > >> in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > since we
> > > > > > > > > have
> > > > > > > > > > > to
> > > > > > > > > > > > > > >> resolve the
> > > > > > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > the
> > > > > > > > topic
> > > > > > > > > > > name?
> > > > > > > > > > > > > The
> > > > > > > > > > > > > > >> > > > difference
> > > > > > > > > > > > > > >> > > > > > is that
> > > > > > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > efficient
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > whereas
> > > > > > > > > > > sending
> > > > > > > > > > > > > the
> > > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > > >> > > > > > is less
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > Furthermore,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > delete
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > recreation is
> > > > > > > > > > > > > easier
> > > > > > > > > > > > > > >> to
> > > > > > > > > > > > > > >> > > > handle
> > > > > > > > > > > > > > >> > > > > > if we
> > > > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > The
> > > > > > idea
> > > > > > > > > was to
> > > > > > > > > > > > > > >> consolidate the
> > > > > > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > paths
> > > > > > but
> > > > > > > > I
> > > > > > > > > do
> > > > > > > > > > > > agree
> > > > > > > > > > > > > > >> with your
> > > > > > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > ids
> > > > > > may be
> > > > > > > > > > > better
> > > > > > > > > > > > > here
> > > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > suppose
> > > > > > > > > that we
> > > > > > > > > > > > may
> > > > > > > > > > > > > > >> allow
> > > > > > > > > > > > > > >> > > > users to
> > > > > > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > future
> > > > > > > > > because
> > > > > > > > > > > > that
> > > > > > > > > > > > > is
> > > > > > > > > > > > > > >> the only
> > > > > > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > re-creation.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > Best,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > David
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > On
> > > > > > Tue,
> > > > > > > > Jul
> > > > > > > > > 12,
> > > > > > > > > > > > > 2022 at
> > > > > > > > > > > > > > >> 1:38 PM
> > > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > djacot@confluent.io>
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > wrote:
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > Hi
> > > > > > > > > Justine,
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > Thanks
> > > > > > > > for
> > > > > > > > > > > your
> > > > > > > > > > > > > > >> comments.
> > > > > > > > > > > > > > >> > > > Please
> > > > > > > > > > > > > > >> > > > > > find
> > > > > > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > -
> > > > > > Yes,
> > > > > > > > > the new
> > > > > > > > > > > > > > >> protocol
> > > > > > > > > > > > > > >> > > > relies on
> > > > > > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > > > topic
> > > > > > > > > names
> > > > > > > > > > > > based
> > > > > > > > > > > > > in
> > > > > > > > > > > > > > >> the
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > if
> > > > > > using
> > > > > > > > > topic
> > > > > > > > > > > > > names
> > > > > > > > > > > > > > >> is the
> > > > > > > > > > > > > > >> > > > > > right call
> > > > > > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about
> > it
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >
> > > > a
> > > > > > little
> > > > > > > > > more.
> > > > > > > > > > > > > > >> Obviously, the
> > > > > > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > > -- Guozhang
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

1. Yeah that makes sense. I originally concerned if there's any consistency
issues if the HB request with encoded topic IDs and the GPA/GIA RPCs with
encoded topic IDs use mappings from two different brokers, and one of them
is stale, but after thinking twice about it I think as long as we handle
e.g. `unknown topic ids` from HB responses then we should be fine.

3. Ack, in that case I think we also need to define "ConsumerGroupMetadata"
which is passed in `onAssignment`. We have a schema defined in the
persistent storage for different purposes, but not in the consumer APIs
yet. I think that would just include a single "Map<String, Metadata>"?

On Fri, Sep 30, 2022 at 5:51 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hey Guozhang,
>
> Thanks for your comments.
>
> 1. That's a very good question. I think that it is better to separate
> the two concerns here. In my opinion, the GPA should only be used to
> feed the assignor. I see the GPA as a way to call a remote function so
> the GPA provides all its inputs to ensure that the assignor does not
> depend on any local state. Knowing this, I would not use the GPA to
> update the local state either. That is not its purpose. The consumer
> is responsible to update its local metadata cache either when it gets
> an unknown topic in the user-provided subscription or in the
> coordinator-provided assignment. The consumer has to do this anyway to
> get all topic metadata (e.g. leaders).
>
> 2. Interesting thought. Handling the translation is definitely
> possible. I wonder if there are cases where having the topic ids is
> beneficial. I am thinking about when a topic is deleted and recreated
> right away. In this case, the current member assignment would still
> have the old topic id and we would not be able to resolve it anymore.
> I suppose that we should directly remove it from the current
> assignment on the server side in this case and only add the new one.
> As I first thought, that seems OK. The major downside is that the user
> of the assignor would not know that the topic is different from what
> it may have cached locally.
>
> We could say that we don't expose topic ids externally for now in
> order to stay consistent with the other interfaces but it would be
> annoying to add it back if we want to expose topic ids in the future.
> I personally believe that we will have to expose them at some point.
> For instance, in the rebalance listener, it would be useful to know
> that a topic has been deleted and recreated. The only way to do this
> is to know the topic id. Let me think a little more about this and
> come back to you.
>
> 3. Sorry for this one. I was a typo. PartitionAssingor#assign should
> return GroupAssignment.
>
> Best,
> David
>
> On Thu, Sep 29, 2022 at 6:06 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Thanks David, I read through the updated wiki, and I have a couple
> > follow-up questions:
> >
> > 1. It seems we would have two potential sources to update the consumer's
> > internally maintained topic name -> topic ID mapping, which is used for
> > encoding subscribed topic Ids in the HB request, and offset commit / list
> > offset requests, etc. One is from metadata responses, and another one is
> > from the GPA (group prepare assignment). On the other hand, the assignor
> > APIs directly take parsed AssignmentMemberSpec and
> AssignmentTopicMetadata
> > from GPA responses directly instead of trying to consult the consumer's
> own
> > cached mapping. So just to clarify, consumers would also update its
> > internal mapping from the GPA responses as well, right?
> >
> > 2. I'm wondering if it's necessary to expose the topic IDs to the
> assignor
> > developers. What they need to do at the moment is to take the mapping
> from
> > AssignmentTopicMetadata, and translate their decided assignment into
> topic
> > IDs when returning the MemberAssignment. We can instead just let consumer
> > to do such translation for them internally and still let the assignor
> > developers to only deal with the topic names. Of course that means the
> > client side assignor APIs would be a bit further diverged from the
> > broker-side assignor such that:
> >
> > a) Topic metadata would be encoded just as String -> num.partitions;
> > b) AssignmentMemberSpec contains subscribedTopicNames and topicPartitions
> > as target assignments;
> > c) Returned MemberAssignment contains list of topicPartitions instead of
> > topicIdPartitions.
> >
> > Internally, consumers would convert the IDs to names when calling the
> > assign() function, and convert them back when sending the request; and
> > similarly it would convert the IDs to names when calling onAssignment()
> as
> > well.
> >
> > In short, I'm wondering if we should bite the bullet for the users,
> eating
> > more complexities internally than pushing that to users.
> >
> > 3. This is a nit one: the return type of `Assignment
> assign(AssignmentSpec
> > assignmentSpec);` seems not defined in the doc. I think it would just
> > include two fields, as:
> >
> > class Assignment {
> >         GroupAssignment groupAssignment;
> >
> >         ConsumerGroupMetadata consumerGroupMetadata;
> >     }
> >
> > Is that right?
> >
> > Guozhang
> >
> > On Thu, Sep 29, 2022 at 1:26 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > Thanks for your input.
> > >
> > > > 1) in HB request, when clients send the subscription, they are still
> > > > sending them in terms of topic names. That is, the consumer would
> not try
> > > > to keep all the mappings from topic names to topic ids for the
> subscribed
> > > > topics.
> > >
> > > This is a mistake of mine. The heartbeat request should use Topic Ids.
> > > I forgot to change it. The idea is that the client will get the
> > > metadata first to translate the topic names to topic ids and thus also
> > > have the mapping in its local metadata cache.
> > >
> > > > 2) in HB responses, the assignment's topics are encoded as IDs
> already.
> > > And
> > > > fetch requests can directly rely on that since they just need topic
> IDs
> > > and
> > > > do not care what actual topic names are.
> > >
> > > That's right.
> > >
> > > > 3) in prepare-assignment responses, we return subscriptions as topic
> IDs
> > > as
> > > > well.
> > >
> > > That's right.
> > >
> > > > In a word, the consumer does not try to maintain the mapping of topic
> > > names
> > > > to IDs at all, and even if they do, they will not expose it in e.g.
> > > > assignor APIs. But for client-side assignors like Streams, if they
> do not
> > > > maintain the mapping from topic IDs to topic names, certain logic
> such as
> > > > co-partitioning indeed cannot be executed.
> > >
> > > As I said, the consumer will have the mapping in its local metadata
> > > cache. It needs to get the partition metadata anyway and the topic id
> > > is part of it. However, I do agree that this mapping is not exposed.
> > >
> > > > So I think we'd need to consider 1) use topic names in subscriptions
> in
> > > the
> > > > prepare-assignment responses, 2) use topic Names in client-side
> > > > PartitionAssignor#GroupMember#subscribedTopics and
> > > > PartitionAssignor#TopicMetadata#topicName.
> > >
> > > 1) makes sense to me. I think that we could actually provide both
> > > names and ids. Regarding 2), I think that I am slightly in favor of
> > > using both names and ids as well. Let me try to update the protocol
> > > and the interface towards this.
> > >
> > > Thanks,
> > > David
> > >
> > > On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > >
> > > > Hello Jun/David,
> > > >
> > > > Just to chime in here for Jun's comment #50, today Streams does not
> have
> > > a
> > > > way to map topic names to topic IDs, and in its topology description
> > > topics
> > > > are encoded with names.
> > > >
> > > > What I originally thought was that, on the consumer side, we'd need
> to
> > > > maintain this mapping from topic names to topic IDs anyways, and
> Streams
> > > > should leverage on the consumer's functionality to get the
> corresponding
> > > > topic names for the given topic IDs. But when I read through the RPC
> > > again
> > > > I realized it may not be the case, since:
> > > >
> > > > 1) in HB request, when clients send the subscription, they are still
> > > > sending them in terms of topic names. That is, the consumer would
> not try
> > > > to keep all the mappings from topic names to topic ids for the
> subscribed
> > > > topics.
> > > > 2) in HB responses, the assignment's topics are encoded as IDs
> already.
> > > And
> > > > fetch requests can directly rely on that since they just need topic
> IDs
> > > and
> > > > do not care what actual topic names are.
> > > > 3) in prepare-assignment responses, we return subscriptions as topic
> IDs
> > > as
> > > > well.
> > > >
> > > > In a word, the consumer does not try to maintain the mapping of topic
> > > names
> > > > to IDs at all, and even if they do, they will not expose it in e.g.
> > > > assignor APIs. But for client-side assignors like Streams, if they
> do not
> > > > maintain the mapping from topic IDs to topic names, certain logic
> such as
> > > > co-partitioning indeed cannot be executed.
> > > >
> > > > So I think we'd need to consider 1) use topic names in subscriptions
> in
> > > the
> > > > prepare-assignment responses, 2) use topic Names in client-side
> > > > PartitionAssignor#GroupMember#subscribedTopics and
> > > > PartitionAssignor#TopicMetadata#topicName.
> > > >
> > > > On Wed, Sep 28, 2022 at 1:26 AM David Jacot
> <djacot@confluent.io.invalid
> > > >
> > > > wrote:
> > > >
> > > > > Hi Jun,
> > > > >
> > > > > Thanks for your comments.
> > > > >
> > > > > 48. Okay. There is a typo in the text:
> > > > > > Note that the persisted target assignment of B already includes
> foo-2
> > > > > here
> > > > > but B does not get it yet because foo-2 is not free.
> > > > >
> > > > > It should be:
> > > > > > Note that the persisted member assignment of B already includes
> foo-2
> > > > > here
> > > > > but B does not get it yet because foo-2 is not free.
> > > > >
> > > > > The examples are showing the persisted state in the
> __consumer_offsets
> > > > > topics. When a member transitions to the next epoch, we directly
> > > > > persist the whole target assignment in order to reduce the number
> of
> > > > > writes and we manage dependencies in memory. So "B - epoch=2,
> > > > > partitions=[foo-2]" is actually correct from a persisted record
> > > > > perspective. However, the in-memory state would be different
> because
> > > > > foo-2 is not yet available as you pointed out. The in-memory state
> > > > > would like as follow:
> > > > >
> > > > > Group (epoch=2)
> > > > > A
> > > > > B
> > > > > Target Assignment (epoch=2)
> > > > > A - partitions=[foo-0, foo-1]
> > > > > B - partitions=[foo-2]
> > > > > Member Assignment
> > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending
> partitions=[]
> > > > > B - epoch=2, partitions=[], pending partitions=[foo-2]
> > > > >
> > > > > I suppose that the confusion comes from this. I will go and convert
> > > > > all the examples to their in-memory representation in order to
> reduce
> > > > > the confusion.
> > > > >
> > > > > 50. Okay. I have to admit that I am not so familiar with the
> internals
> > > > > of KStreams. If I understand correctly, you're saying that KStreams
> > > > > may not have topic ids in its topology so topic names are missing
> here
> > > > > to match the topics. I was hoping that KStreams would have topic
> ids
> > > > > in the topology as well. If that is not the case, we can also add
> > > > > topic names on our side. I will check this.
> > > > >
> > > > > 53. It seems that I was not clear. Sorry for that. I did not mean
> to
> > > > > say that MemberAssignment.metadata is not used by the group
> > > > > coordinator. Let me re-explain the flow.
> > > > >
> > > > > PartitionAssignor.assign() returns an Assignment. Assignment
> contains
> > > > > MemberAssignments where each member has MemberAssignment.metadata.
> The
> > > > > Assignment is installed by the client with the
> > > > > ConsumerGroupInstallAssignment API and this overrides the current
> > > > > target assignment in the coordinator. Then the group coordinator
> uses
> > > > > it to reconcile members. Therefore the target assignment will
> > > > > eventually become the current member assignment and it will be
> > > > > delivered to the member, including the original
> > > > > MemberAssignment.metadata defines by the assignor.
> > > > >
> > > > > I hope that this helps to clarify how MemberAssignment.metadata is
> > > > > used afterwards.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <jun@confluent.io.invalid
> >
> > > wrote:
> > > > > > Hi, David,
> > > > > >
> > > > > > Thanks for the reply. A few more follow up questions.
> > > > > >
> > > > > > 48. Hmm, so member assignment is what the client sees. Then, I
> am not
> > > > > sure
> > > > > > about the following. The text says B cannot get foo-2 yet. But in
> > > Member
> > > > > > Assignment, B has foo-2.
> > > > > >
> > > > > > At this point B can transition to epoch 2 but cannot get foo-2
> until
> > > A
> > > > > > revokes it.
> > > > > >
> > > > > > Note that the persisted target assignment of B already includes
> foo-2
> > > > > here
> > > > > > but B does not get it yet because foo-2 is not free.
> > > > > >
> > > > > > Group (epoch=2)
> > > > > > A
> > > > > > B
> > > > > > Target Assignment (epoch=2)
> > > > > > A - partitions=[foo-0, foo-1]
> > > > > > B - partitions=[foo-2]
> > > > > > Member Assignment
> > > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > > > > B - epoch=2, partitions=[foo-2]
> > > > > >
> > > > > >
> > > > > > 50. Hmm, I thought KStreams doesn't use range assignor all the
> time.
> > > It
> > > > > > does the assignment based on the topology. For example, if you
> are
> > > > > joining
> > > > > > two topics A and B and deriving the join result in a 3rd topic C,
> > > which
> > > > > > will be aggregated. You want to use range to co-locate the
> partitions
> > > > > from
> > > > > > topic A and B. However, for topic C, the partition could
> potentially
> > > be
> > > > > > assigned in round robin. Without the topic name, KStreams won't
> know
> > > > > which
> > > > > > topics need to be co-located.
> > > > > >
> > > > > > 53. If MemberAssignment.metadata is not used by the group
> > > coordinator,
> > > > > > where is it being used?
> > > > > >
> > > > > > Thanks,
> > > > > >
> > > > > > Jun
> > > > > >
> > > > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > > > > <ja...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > > > Hi David,
> > > > > > >
> > > > > > > Thanks for the response. Replied below:
> > > > > > >
> > > > > > > > I don't have a strong opinion on this one. However, it seems
> that
> > > > > this
> > > > > > > could be a pandora box if our subset turns out to not be
> sufficient
> > > > > > > for some existing users. We will have to add more to it and the
> > > > > > > compatibility story between versions could be a real
> nightmare. One
> > > > > > > thing that we could do for sure is disabling all unicode
> groups in
> > > the
> > > > > > > library. That seems possible and would already reduce the
> scope.
> > > > > > >
> > > > > > > Changes to the regex syntax seems not a major problem to me. We
> > > would
> > > > > bump
> > > > > > > the protocol version as we usually do. Actually that begs the
> same
> > > > > question
> > > > > > > about RE2. How stable is the syntax and does it differ
> depending
> > > on the
> > > > > > > version of the library? In general, I do agree that we'd be
> better
> > > off
> > > > > not
> > > > > > > having to manage the syntax ourselves if we can convince
> ourselves
> > > that
> > > > > > > compatibility won't be a problem and the client won't
> eventually
> > > need
> > > > > to
> > > > > > > have access to it.
> > > > > > >
> > > > > > > Thanks,
> > > > > > > Jason
> > > > > > >
> > > > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hi Luke,
> > > > > > > >
> > > > > > > > Thanks for your comments.
> > > > > > > >
> > > > > > > > Your suggestion makes sense. Jun's made the same one. We are
> also
> > > > > > > > discussing whether we should directly return the group state
> in
> > > the
> > > > > > > > response and drop the prepare assignment api. It seems that
> the
> > > > > > > > current consensus is to keep it though. If that stays like
> this,
> > > I
> > > > > > > > will apply your suggestion.
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > > David
> > > > > > > >
> > > > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <
> showuon@gmail.com>
> > > > > wrote:
> > > > > > > > >
> > > > > > > > > Hi David,
> > > > > > > > >
> > > > > > > > > Thanks for your patience for answering all these questions!
> > > > > > > > >
> > > > > > > > > One more question from me:
> > > > > > > > > 1. When the selected consumer is going to be in client-side
> > > > > assigning,
> > > > > > > we
> > > > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> > > > > wondering if
> > > > > > > we
> > > > > > > > > could use a flag to indicate the assignment is required.
> The
> > > > > rationale
> > > > > > > is
> > > > > > > > > when returning error, we lose a chance to catch up with the
> > > latest
> > > > > > > target
> > > > > > > > > assignment or revoke partitions. Since we expected the
> record
> > > > > fetching
> > > > > > > is
> > > > > > > > > not blocked by the rebalance/assignment computing, we might
> > > want to
> > > > > > > allow
> > > > > > > > > the members to update the member assignment in each
> heartbeat.
> > > > > > > > >
> > > > > > > > > Luke
> > > > > > > > >
> > > > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > > > > <ja...@confluent.io.invalid>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hey David,
> > > > > > > > > >
> > > > > > > > > > Thanks for the response.
> > > > > > > > > >
> > > > > > > > > > > Do you think that we could have a case where two
> members
> > > with
> > > > > the
> > > > > > > > same
> > > > > > > > > > static id are active?
> > > > > > > > > >
> > > > > > > > > > I was just thinking about the handoff when we spin up a
> new
> > > node
> > > > > > > > before the
> > > > > > > > > > old one completes shutdown. It's probably ok to fence
> the old
> > > > > member
> > > > > > > > > > abruptly. I do recall we had a few issues with this
> handoff
> > > in
> > > > > the
> > > > > > > > past. It
> > > > > > > > > > was due to the fact that the replaced node might have
> been
> > > the
> > > > > > > leader.
> > > > > > > > > > Since we've removed the leader notion, perhaps this kind
> of
> > > issue
> > > > > > > won't
> > > > > > > > > > resurface.
> > > > > > > > > >
> > > > > > > > > > In regard to the RE2 library. I understand the intent to
> pass
> > > > > this
> > > > > > > > through
> > > > > > > > > > as a string without any validation so that we don't need
> the
> > > > > > > > dependence on
> > > > > > > > > > the client. I do wonder if that will be restrictive in
> > > practice.
> > > > > For
> > > > > > > > > > example, if a regex changes on the client side, would we
> > > need to
> > > > > > > > validate
> > > > > > > > > > against the current assignment to see which partitions to
> > > > > continue
> > > > > > > > > > fetching? Or would we instead invalidate the complete
> current
> > > > > > > > assignment?
> > > > > > > > > > Perhaps this is a rare case and we can do the latter.
> There
> > > > > could be
> > > > > > > > more
> > > > > > > > > > situations though.
> > > > > > > > > >
> > > > > > > > > > As an alternative, I wonder if we could define a
> minimalistic
> > > > > regex
> > > > > > > > syntax
> > > > > > > > > > (a subset of RE2 say) that could map more easily to regex
> > > support
> > > > > > > that
> > > > > > > > > > already exists in clients? For example, in this
> scenario, we
> > > are
> > > > > > > > matching
> > > > > > > > > > topics, so we only need to be able to represent topic
> > > characters
> > > > > and
> > > > > > > > not
> > > > > > > > > > the whole smorgasbord of whitespace and unicode options.
> > > Would it
> > > > > > > make
> > > > > > > > > > sense to have a reduced syntax which includes only
> standard
> > > topic
> > > > > > > > > > characters, wildcards, character classes, and
> repetitions?
> > > > > > > > > >
> > > > > > > > > > Best,
> > > > > > > > > > Jason
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hi David, thanks for all the detailed explanations. I
> think
> > > > > they
> > > > > > > all
> > > > > > > > make
> > > > > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > > > > >
> > > > > > > > > > > > I don't really see the benefits here because server
> side
> > > > > > > assignors
> > > > > > > > > > > don't have metadata at all. They only assign
> > > topic-partitions.
> > > > > They
> > > > > > > > > > > are not supposed to generate metadata nor to receive
> > > metadata
> > > > > from
> > > > > > > > the
> > > > > > > > > > > members.
> > > > > > > > > > >
> > > > > > > > > > > Just to clarify I was asking about the `version` of the
> > > > > assignor
> > > > > > > > (i.e. up
> > > > > > > > > > > to what version that the client would support), and I
> do
> > > agree
> > > > > we
> > > > > > > > would
> > > > > > > > > > not
> > > > > > > > > > > need metadata. What I have in mind is that, for some
> > > specific
> > > > > > > > built-in
> > > > > > > > > > > broker-assignors, e.g. rack-aware assignors, if it's
> > > possible
> > > > > that
> > > > > > > > in a
> > > > > > > > > > > newer version we would have a hierarchical rack ID
> string
> > > > > format,
> > > > > > > > like
> > > > > > > > > > > "tier1-tier2" etc, but if some client has not upgraded
> > > their
> > > > > rack
> > > > > > > ID
> > > > > > > > > > > would still be in old format. In this case, the broker
> then
> > > > > needs
> > > > > > > to
> > > > > > > > > > choose
> > > > > > > > > > > the old versioned assignor. I'm probably making
> something
> > > up
> > > > > here
> > > > > > > for
> > > > > > > > > > rack
> > > > > > > > > > > aware assignors, but I'm wondering if in general such
> an
> > > > > > > > "auto-downgrade"
> > > > > > > > > > > behavior would be needed still for broker-side
> assignor,
> > > and
> > > > > if yes
> > > > > > > > would
> > > > > > > > > > > "version" still be useful.
> > > > > > > > > > >
> > > > > > > > > > > > Yeah, that's right. Within a rebalance,
> `onAssignment` is
> > > > > called
> > > > > > > > once
> > > > > > > > > > > when the member transitions to a new epoch. This one
> > > contains
> > > > > the
> > > > > > > > full
> > > > > > > > > > > metadata provided by the client side assignor. Then,
> > > > > `onAssignment`
> > > > > > > > > > > can be called max N times where N is the number of
> > > partitions
> > > > > > > pending
> > > > > > > > > > > revocation by other members. Let me try to clarify
> this in
> > > the
> > > > > KIP.
> > > > > > > > > > >
> > > > > > > > > > > Okay, my understanding is that the calling ordering of
> > > these
> > > > > > > > callbacks
> > > > > > > > > > > would be like the following:
> > > > > > > > > > >
> > > > > > > > > > > ----------------------------------------
> > > > > > > > > > > onPartitionsRevoked();   // just once, since we do not
> > > really
> > > > > need
> > > > > > > > > > > to revoke incrementally.
> > > > > > > > > > >
> > > > > > > > > > > onAssignment();    // the first call, with epoch
> > > incremented
> > > > > > > > > > > onPartitionsAssigned();   // paired with the
> onAssignment
> > > > > > > > > > >
> > > > > > > > > > > onAssignment();              // the first onAssignment
> > > would
> > > > > bump
> > > > > > > up
> > > > > > > > the
> > > > > > > > > > > epoch, and the metadata reflected.
> > > > > > > > > > > onPartitionsAssigned();   // each time we get an
> additional
> > > > > > > > assignment,
> > > > > > > > > > we
> > > > > > > > > > > call onAssignment and then paired with an
> > > onPartitionsAssigned
> > > > > > > > > > > ...
> > > > > > > > > > > onAssignment();
> > > > > > > > > > > onPartitionsAssigned();   // on each of the
> onAssignment
> > > > > calls, the
> > > > > > > > > > encoded
> > > > > > > > > > > metadata would not change, only the incrementally added
> > > > > partitions
> > > > > > > be
> > > > > > > > > > > reflected
> > > > > > > > > > >
> > > > > > > > > > > Is that the case?
> > > > > > > > > > >
> > > > > > > > > > > I'm wondering if we would still call onAssignment just
> > > once,
> > > > > that
> > > > > > > > encodes
> > > > > > > > > > > all the assignment for this rebalance, including all
> the
> > > > > partitions
> > > > > > > > that
> > > > > > > > > > > should be assigned to the member but not yet assigned
> since
> > > > > they
> > > > > > > > have not
> > > > > > > > > > > been revoked by others. In that case the call ordering
> > > would
> > > > > be:
> > > > > > > > > > >
> > > > > > > > > > > ----------------------------------------
> > > > > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > > > > onAssignment();    // just once, with epoch
> incremented,
> > > and
> > > > > > > metadata
> > > > > > > > > > > encoded changed, the "assignment" field also reflect
> the
> > > final
> > > > > > > target
> > > > > > > > > > > assignment
> > > > > > > > > > > onPartitionsAssigned();   // multiple times, which
> > > represent
> > > > > > > > > > incrementally
> > > > > > > > > > > added partitions
> > > > > > > > > > > ...
> > > > > > > > > > > onPartitionsAssigned();
> > > > > > > > > > >
> > > > > > > > > > > The motivation from this is that, most users would only
> > > > > implement
> > > > > > > the
> > > > > > > > > > > rebalance callback listeners and hence we'd definitely
> > > need to
> > > > > make
> > > > > > > > sure
> > > > > > > > > > > the semantics of that does not change much, and the
> time
> > > > > > > > > > > `onPartitionsAssigned` indicate the time when the
> > > partitions
> > > > > are
> > > > > > > > actually
> > > > > > > > > > > assigned to it; while for assignors, the
> `onAssignment` is
> > > > > used to
> > > > > > > > > > indicate
> > > > > > > > > > > what decision is made regarding for this member, i.e.
> when
> > > the
> > > > > > > > partitions
> > > > > > > > > > > are decided to be given to it, but not necessarily
> meant
> > > that
> > > > > it
> > > > > > > has
> > > > > > > > been
> > > > > > > > > > > given, since that time should be determined by the
> time of
> > > > > > > > > > > `onPartitionsAssigned`. The benefits though, would be
> that
> > > > > assignor
> > > > > > > > > > > implementers would not need to reason which
> `onAssignment`
> > > > > would be
> > > > > > > > the
> > > > > > > > > > > last one for this epoch.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Guozhang
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Guozhang
> > > > > > > > > > >
> > > > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >
> > > > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > > Hi Guozhang,
> > > > > > > > > > > >
> > > > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment"
> has
> > > an
> > > > > Error
> > > > > > > > byte
> > > > > > > > > > > > field,
> > > > > > > > > > > > > while the broker-side
> "PartitionAssignor#Assignment"
> > > does
> > > > > not.
> > > > > > > > And
> > > > > > > > > > the
> > > > > > > > > > > > > rationale seems to be assuming that we should
> always be
> > > > > able to
> > > > > > > > do
> > > > > > > > > > the
> > > > > > > > > > > > > assignment at the broker-side assignor without
> errors.
> > > > > > > > Personally I
> > > > > > > > > > > think
> > > > > > > > > > > > > it's still potentially beneficial to add the Error
> > > field
> > > > > even
> > > > > > > for
> > > > > > > > > > > > > broker-side assignors, e.g. for some edge cases
> where
> > > some
> > > > > > > > subscribed
> > > > > > > > > > > > > topics are not recognized with the current broker's
> > > > > metadata.
> > > > > > > > What do
> > > > > > > > > > > you
> > > > > > > > > > > > > think?
> > > > > > > > > > > >
> > > > > > > > > > > > Yeah, that seems reasonable. However, I wonder if it
> > > would be
> > > > > > > > better
> > > > > > > > > > > > to use an exception on the server side. This is what
> we
> > > > > usually
> > > > > > > do
> > > > > > > > for
> > > > > > > > > > > > server side plugins. On the client side, we use a
> field
> > > > > because
> > > > > > > the
> > > > > > > > > > > > errors are not defined in advance.
> > > > > > > > > > > >
> > > > > > > > > > > > Your comment also makes me think about what we
> should do
> > > > > when the
> > > > > > > > > > > > server side assignor fails. I suppose that we have to
> > > keep
> > > > > the
> > > > > > > > current
> > > > > > > > > > > > assignment until a new event occurs. For instance, in
> > > your
> > > > > > > example,
> > > > > > > > > > > > the coordinator would have to trigger a rebalance
> when
> > > > > > > unrecognized
> > > > > > > > > > > > topics are available. This would be part of the
> metadata
> > > > > > > > monitoring.
> > > > > > > > > > > >
> > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> additional
> > > > > fields
> > > > > > > > > > > > > reason/version/metadata compared with the
> broker-side
> > > > > > > > GroupMember. I
> > > > > > > > > > > > agree
> > > > > > > > > > > > > that broker-side assignor would not need
> > > reason/metadata
> > > > > since
> > > > > > > > they
> > > > > > > > > > are
> > > > > > > > > > > > > blackbox strings/bytes to the assignor, but what
> about
> > > > > version?
> > > > > > > > E.g.
> > > > > > > > > > is
> > > > > > > > > > > > it
> > > > > > > > > > > > > possible that we evolve our broker-side built-in
> > > assignor
> > > > > but
> > > > > > > > the old
> > > > > > > > > > > > > versioned clients would not be able to work with
> the
> > > new
> > > > > > > > version, in
> > > > > > > > > > > > which
> > > > > > > > > > > > > case we need to let the broker being aware of this
> and
> > > > > upgrade
> > > > > > > > its
> > > > > > > > > > > > behavior
> > > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > > >
> > > > > > > > > > > > I don't really see the benefits here because server
> side
> > > > > > > assignors
> > > > > > > > > > > > don't have metadata at all. They only assign
> > > > > topic-partitions.
> > > > > > > They
> > > > > > > > > > > > are not supposed to generate metadata nor to receive
> > > metadata
> > > > > > > from
> > > > > > > > the
> > > > > > > > > > > > members.
> > > > > > > > > > > >
> > > > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > > > "GroupMember",
> > > > > > > > > > instead
> > > > > > > > > > > > of
> > > > > > > > > > > > > including these three fields, what about just
> adding
> > > the
> > > > > > > > "Metadata"
> > > > > > > > > > > field
> > > > > > > > > > > > > class which has these three fields? Also, there
> are two
> > > > > > > > "Metadata"
> > > > > > > > > > > > > currently in the APIs, the first is a class that
> > > encodes
> > > > > > > > > > > > > reason/version/metadata, and the second is just the
> > > encoded
> > > > > > > > metadata
> > > > > > > > > > > > bytes.
> > > > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > > > memberMetadata,
> > > > > > > > > > which
> > > > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> > > naming the
> > > > > > > > second
> > > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > > >
> > > > > > > > > > > > That's a good point. Let me try to rationalize this
> > > interface
> > > > > > > > based on
> > > > > > > > > > > > your suggestions.
> > > > > > > > > > > >
> > > > > > > > > > > > Best,
> > > > > > > > > > > > David
> > > > > > > > > > > >
> > > > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > wrote:
> > > > > > > > > > > > >
> > > > > > > > > > > > > Hello David,
> > > > > > > > > > > > >
> > > > > > > > > > > > > Just had a few more nit questions about the public
> > > APIs:
> > > > > > > > > > > > >
> > > > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment"
> has
> > > an
> > > > > Error
> > > > > > > > byte
> > > > > > > > > > > > field,
> > > > > > > > > > > > > while the broker-side
> "PartitionAssignor#Assignment"
> > > does
> > > > > not.
> > > > > > > > And
> > > > > > > > > > the
> > > > > > > > > > > > > rationale seems to be assuming that we should
> always be
> > > > > able to
> > > > > > > > do
> > > > > > > > > > the
> > > > > > > > > > > > > assignment at the broker-side assignor without
> errors.
> > > > > > > > Personally I
> > > > > > > > > > > think
> > > > > > > > > > > > > it's still potentially beneficial to add the Error
> > > field
> > > > > even
> > > > > > > for
> > > > > > > > > > > > > broker-side assignors, e.g. for some edge cases
> where
> > > some
> > > > > > > > subscribed
> > > > > > > > > > > > > topics are not recognized with the current broker's
> > > > > metadata.
> > > > > > > > What do
> > > > > > > > > > > you
> > > > > > > > > > > > > think?
> > > > > > > > > > > > >
> > > > > > > > > > > > > 2) The client-side "GroupMember" has three
> additional
> > > > > fields
> > > > > > > > > > > > > reason/version/metadata compared with the
> broker-side
> > > > > > > > GroupMember. I
> > > > > > > > > > > > agree
> > > > > > > > > > > > > that broker-side assignor would not need
> > > reason/metadata
> > > > > since
> > > > > > > > they
> > > > > > > > > > are
> > > > > > > > > > > > > blackbox strings/bytes to the assignor, but what
> about
> > > > > version?
> > > > > > > > E.g.
> > > > > > > > > > is
> > > > > > > > > > > > it
> > > > > > > > > > > > > possible that we evolve our broker-side built-in
> > > assignor
> > > > > but
> > > > > > > > the old
> > > > > > > > > > > > > versioned clients would not be able to work with
> the
> > > new
> > > > > > > > version, in
> > > > > > > > > > > > which
> > > > > > > > > > > > > case we need to let the broker being aware of this
> and
> > > > > upgrade
> > > > > > > > its
> > > > > > > > > > > > behavior
> > > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > > > >
> > > > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > > > "GroupMember",
> > > > > > > > > > instead
> > > > > > > > > > > > of
> > > > > > > > > > > > > including these three fields, what about just
> adding
> > > the
> > > > > > > > "Metadata"
> > > > > > > > > > > field
> > > > > > > > > > > > > class which has these three fields? Also, there
> are two
> > > > > > > > "Metadata"
> > > > > > > > > > > > > currently in the APIs, the first is a class that
> > > encodes
> > > > > > > > > > > > > reason/version/metadata, and the second is just the
> > > encoded
> > > > > > > > metadata
> > > > > > > > > > > > bytes.
> > > > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > > > memberMetadata,
> > > > > > > > > > which
> > > > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> > > naming the
> > > > > > > > second
> > > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > > > >
> > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > wrote:
> > > > > > > > > > > > >
> > > > > > > > > > > > > > Hello David,
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Thanks for bringing this question up. I think the
> > > main
> > > > > > > > benefits as
> > > > > > > > > > > you
> > > > > > > > > > > > > > listed is 2) above if it stays; just to clarify,
> we
> > > would
> > > > > > > only
> > > > > > > > be
> > > > > > > > > > > able
> > > > > > > > > > > > to
> > > > > > > > > > > > > > save one round trip if the rebalance is still
> > > triggered
> > > > > by
> > > > > > > the
> > > > > > > > > > > broker;
> > > > > > > > > > > > if
> > > > > > > > > > > > > > the rebalance is triggered by the client then the
> > > > > > > > num.round.trips
> > > > > > > > > > are
> > > > > > > > > > > > the
> > > > > > > > > > > > > > same:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > T0: client decides to do a new assignment,
> suppose
> > > it has
> > > > > > > > already
> > > > > > > > > > > sent
> > > > > > > > > > > > a
> > > > > > > > > > > > > > HB and hence has to wait for it to return first
> since
> > > > > only
> > > > > > > one
> > > > > > > > > > > request
> > > > > > > > > > > > /
> > > > > > > > > > > > > > response can be inflight with the coordinator's
> > > socket.
> > > > > > > > > > > > > > T1: client receives the HB response, and then
> sends
> > > the
> > > > > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > > > > T2: the GroupPrepareAssignment response is
> returned.
> > > > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > > request.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > In total, two round trips.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > T0: client decides to do a new assignment,
> suppose
> > > it has
> > > > > > > > already
> > > > > > > > > > > sent
> > > > > > > > > > > > a
> > > > > > > > > > > > > > HB and hence has to wait for it to return first
> since
> > > > > only
> > > > > > > one
> > > > > > > > > > > request
> > > > > > > > > > > > /
> > > > > > > > > > > > > > response can be inflight with the coordinator's
> > > socket.
> > > > > > > > > > > > > > T1: client receives the HB response, and then
> sends
> > > the
> > > > > new
> > > > > > > HB
> > > > > > > > > > > request
> > > > > > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > > > > > T2: the HB response with the optional member
> metadata
> > > > > map is
> > > > > > > > > > > returned.
> > > > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > > request.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > -----------------------------
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > So to complete the full picture here, we'd need
> to
> > > modify
> > > > > > > both
> > > > > > > > HB
> > > > > > > > > > > > request
> > > > > > > > > > > > > > and response so that the client can also
> indicate a
> > > new
> > > > > > > > rebalance
> > > > > > > > > > via
> > > > > > > > > > > > the
> > > > > > > > > > > > > > HB request as well, right?
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Assuming all above is true, I think it's okay to
> > > merge
> > > > > the
> > > > > > > > > > > > > > GroupPrepareAssignment into HB given that we can
> > > make the
> > > > > > > > > > additional
> > > > > > > > > > > > fields
> > > > > > > > > > > > > > encoding the full member (subscription) metadata
> and
> > > > > topic
> > > > > > > > metadata
> > > > > > > > > > > as
> > > > > > > > > > > > > > optional fields.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > > wrote:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > >> Hi all,
> > > > > > > > > > > > > >>
> > > > > > > > > > > > > >> During an offline conversation, someone asked
> why we
> > > > > need
> > > > > > > the
> > > > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested
> > > that we
> > > > > > > could
> > > > > > > > > > > instead
> > > > > > > > > > > > > >> provide the group state in the heartbeat
> response.
> > > This
> > > > > has
> > > > > > > a
> > > > > > > > few
> > > > > > > > > > > > > >> advantages: 1) it does not require using a
> special
> > > error
> > > > > > > code
> > > > > > > > to
> > > > > > > > > > > > > >> signal that a new assignment is required as the
> > > signal
> > > > > would
> > > > > > > > be
> > > > > > > > > > the
> > > > > > > > > > > > > >> provided group state; 2) it removes one round
> trip
> > > when
> > > > > a
> > > > > > > > client
> > > > > > > > > > > side
> > > > > > > > > > > > > >> assignor is used. The downside is that it makes
> the
> > > > > > > heartbeat
> > > > > > > > > > > > > >> response's definition quite large. I recall
> that I
> > > went
> > > > > with
> > > > > > > > the
> > > > > > > > > > > > > >> current approach due to this.
> > > > > > > > > > > > > >>
> > > > > > > > > > > > > >> Providing the group state in the heartbeat
> response
> > > is
> > > > > > > > appealing.
> > > > > > > > > > > What
> > > > > > > > > > > > > >> do you guys think?
> > > > > > > > > > > > > >>
> > > > > > > > > > > > > >> Best,
> > > > > > > > > > > > > >> David
> > > > > > > > > > > > > >>
> > > > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > > > > > djacot@confluent.io>
> > > > > > > > > > > > wrote:
> > > > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > 1. I have added a reference to the relevant
> > > chapter
> > > > > > > instead
> > > > > > > > of
> > > > > > > > > > > > > >> > repeating the whole thing. Does that work for
> you?
> > > > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you are
> > > referring
> > > > > to
> > > > > > > is
> > > > > > > > > > about
> > > > > > > > > > > > when
> > > > > > > > > > > > > >> > a rebalance should be triggered for the
> > > non-upgraded
> > > > > > > members
> > > > > > > > > > using
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> > old protocol. The section mentions that a
> > > rebalance
> > > > > must
> > > > > > > be
> > > > > > > > > > > > triggered
> > > > > > > > > > > > > >> > when a new assignment is installed. This
> implies
> > > that
> > > > > the
> > > > > > > > group
> > > > > > > > > > > > epoch
> > > > > > > > > > > > > >> > was updated either by a native member or a
> > > > > non-upgraded
> > > > > > > > member.
> > > > > > > > > > > For
> > > > > > > > > > > > > >> > the latter, the JoinGroup request would be the
> > > > > trigger. I
> > > > > > > > have
> > > > > > > > > > > > added a
> > > > > > > > > > > > > >> > reference to the relevant chapter in the
> > > "JoinGroup
> > > > > > > > Handling"
> > > > > > > > > > > > section
> > > > > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > Thanks,
> > > > > > > > > > > > > >> > David
> > > > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang
> <
> > > > > > > > > > wangguoz@gmail.com
> > > > > > > > > > > >
> > > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > Hello David,
> > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > Alright I think that's sufficient. Just to
> make
> > > that
> > > > > > > > clear in
> > > > > > > > > > > the
> > > > > > > > > > > > doc,
> > > > > > > > > > > > > >> > > could we update:
> > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > 1) the heartbeat request handling section,
> > > stating
> > > > > when
> > > > > > > > > > > > coordinator
> > > > > > > > > > > > > >> will
> > > > > > > > > > > > > >> > > trigger rebalance based on the HB's member
> > > metadata
> > > > > /
> > > > > > > > reason?
> > > > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to
> include
> > > what
> > > > > we
> > > > > > > > > > described
> > > > > > > > > > > > in
> > > > > > > > > > > > > >> "Group
> > > > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as
> well?
> > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > Guozhang
> > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > > >> > > wrote:
> > > > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > I thought that the assignor will always be
> > > > > consulted
> > > > > > > > when
> > > > > > > > > > the
> > > > > > > > > > > > next
> > > > > > > > > > > > > >> > > > heartbeat request is constructed. In other
> > > words,
> > > > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be
> called
> > > for
> > > > > every
> > > > > > > > > > > heartbeat.
> > > > > > > > > > > > > >> This
> > > > > > > > > > > > > >> > > > gives the opportunity for the assignor to
> > > enforce
> > > > > a
> > > > > > > > > > rebalance
> > > > > > > > > > > by
> > > > > > > > > > > > > >> > > > setting the reason to a non-zero value or
> by
> > > > > changing
> > > > > > > > the
> > > > > > > > > > > > bytes. Do
> > > > > > > > > > > > > >> > > > you think that this is not sufficient?
> Are you
> > > > > > > > concerned by
> > > > > > > > > > > the
> > > > > > > > > > > > > >> delay?
> > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > Best,
> > > > > > > > > > > > > >> > > > David
> > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang
> Wang <
> > > > > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > > >> > > > > In this case, a new assignment is
> triggered
> > > by
> > > > > the
> > > > > > > > client
> > > > > > > > > > > side
> > > > > > > > > > > > > >> > > > > assignor. When constructing the HB, the
> > > consumer
> > > > > > > will
> > > > > > > > > > always
> > > > > > > > > > > > > >> consult
> > > > > > > > > > > > > >> > > > > the client side assignor and propagate
> the
> > > > > > > > information to
> > > > > > > > > > > the
> > > > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > coordinator. In other words, we don't
> expect
> > > > > users
> > > > > > > to
> > > > > > > > call
> > > > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > As I looked at the current
> > > PartitionAssignor's
> > > > > > > > interface,
> > > > > > > > > > we
> > > > > > > > > > > > > >> actually do
> > > > > > > > > > > > > >> > > > > not have a way yet to instruct how to
> > > construct
> > > > > the
> > > > > > > > next
> > > > > > > > > > HB
> > > > > > > > > > > > > >> request, e.g.
> > > > > > > > > > > > > >> > > > > when the assignor wants to enforce a new
> > > > > rebalance
> > > > > > > > with a
> > > > > > > > > > > new
> > > > > > > > > > > > > >> assignment,
> > > > > > > > > > > > > >> > > > > we'd need some customizable APIs inside
> the
> > > > > > > > > > > PartitionAssignor
> > > > > > > > > > > > to
> > > > > > > > > > > > > >> indicate
> > > > > > > > > > > > > >> > > > > the next HB telling broker about so.
> WDYT
> > > about
> > > > > > > adding
> > > > > > > > > > such
> > > > > > > > > > > an
> > > > > > > > > > > > > >> API on the
> > > > > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David
> Jacot
> > > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > I have updated the KIP to include your
> > > > > feedback. I
> > > > > > > > have
> > > > > > > > > > > also
> > > > > > > > > > > > > >> tried to
> > > > > > > > > > > > > >> > > > > > clarify the parts which were not
> cleared.
> > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David
> > > Jacot <
> > > > > > > > > > > > djacot@confluent.io
> > > > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me
> start
> > > by
> > > > > > > > answering
> > > > > > > > > > your
> > > > > > > > > > > > > >> questions
> > > > > > > > > > > > > >> > > > > > > inline and I will update the KIP
> next
> > > week.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the
> main
> > > > > benefits
> > > > > > > > of
> > > > > > > > > > the
> > > > > > > > > > > > KIP
> > > > > > > > > > > > > >> seem to
> > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > > efficient
> > > > > > > > support of
> > > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > I would also add that the KIP
> removes
> > > the
> > > > > global
> > > > > > > > sync
> > > > > > > > > > > > barrier
> > > > > > > > > > > > > >> in the
> > > > > > > > > > > > > >> > > > > > > protocol which is essential to
> improve
> > > group
> > > > > > > > stability
> > > > > > > > > > > and
> > > > > > > > > > > > > >> > > > > > > scalability, and the KIP also
> > > simplifies the
> > > > > > > > client by
> > > > > > > > > > > > moving
> > > > > > > > > > > > > >> most of
> > > > > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> singleton.
> > > Do we
> > > > > plan
> > > > > > > > to
> > > > > > > > > > > > support
> > > > > > > > > > > > > >> rolling
> > > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > > > consumers?
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Definitely. The group coordinator
> will
> > > use
> > > > > the
> > > > > > > > > > assignor
> > > > > > > > > > > > used
> > > > > > > > > > > > > >> by a
> > > > > > > > > > > > > >> > > > > > > majority of the members. This
> allows the
> > > > > group
> > > > > > > to
> > > > > > > > move
> > > > > > > > > > > > from
> > > > > > > > > > > > > >> one
> > > > > > > > > > > > > >> > > > > > > assignor to another by a roll. This
> is
> > > > > explained
> > > > > > > > in
> > > > > > > > > > the
> > > > > > > > > > > > > >> Assignor
> > > > > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you
> explain
> > > > > whether
> > > > > > > > it's
> > > > > > > > > > > > > >> required in
> > > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> > > needs to
> > > > > be
> > > > > > > > filled?
> > > > > > > > > > > For
> > > > > > > > > > > > > >> example,
> > > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> > > needs to
> > > > > be
> > > > > > > > filled.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > The client is expected to set those
> > > fields
> > > > > in
> > > > > > > > case of
> > > > > > > > > > a
> > > > > > > > > > > > > >> connection
> > > > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the
> fields
> > > have
> > > > > > > > changed
> > > > > > > > > > > since
> > > > > > > > > > > > > >> the last
> > > > > > > > > > > > > >> > > > > > > HB. The server populates those
> fields as
> > > > > long as
> > > > > > > > the
> > > > > > > > > > > > member
> > > > > > > > > > > > > >> is not
> > > > > > > > > > > > > >> > > > > > > fully reconciled - the member should
> > > > > acknowledge
> > > > > > > > that
> > > > > > > > > > it
> > > > > > > > > > > > has
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > expected epoch and assignment. I
> will
> > > > > clarify
> > > > > > > > this in
> > > > > > > > > > > the
> > > > > > > > > > > > KIP.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> protocol,
> > > the
> > > > > rack
> > > > > > > > > > > affinity
> > > > > > > > > > > > > >> between the
> > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > >> > > > > > > > and the broker is only considered
> > > during
> > > > > > > > fetching,
> > > > > > > > > > but
> > > > > > > > > > > > not
> > > > > > > > > > > > > >> during
> > > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> Sometimes,
> > > once
> > > > > the
> > > > > > > > > > > assignment
> > > > > > > > > > > > is
> > > > > > > > > > > > > >> made,
> > > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> > > because
> > > > > no
> > > > > > > > replicas
> > > > > > > > > > > of
> > > > > > > > > > > > > >> assigned
> > > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > > wondering
> > > > > if we
> > > > > > > > should
> > > > > > > > > > > use
> > > > > > > > > > > > > >> this
> > > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > > >> > > > > > > > to address this by including rack
> in
> > > > > > > > GroupMember.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > That's an interesting idea. I don't
> see
> > > any
> > > > > > > issue
> > > > > > > > with
> > > > > > > > > > > > adding
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > rack
> > > > > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often,
> it's
> > > > > useful to
> > > > > > > > know
> > > > > > > > > > how
> > > > > > > > > > > > busy
> > > > > > > > > > > > > >> a group
> > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> event
> > > loop
> > > > > > > model,
> > > > > > > > it
> > > > > > > > > > > seems
> > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction
> of
> > > the
> > > > > time
> > > > > > > > the
> > > > > > > > > > > event
> > > > > > > > > > > > > >> loop is
> > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > That's a great idea. I will add it.
> > > Thanks.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > > coordinator
> > > > > > > > failover
> > > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > > >> > > > > > > > does it need to trigger the check
> if
> > > any
> > > > > group
> > > > > > > > with
> > > > > > > > > > > the
> > > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> matching
> > > topic?
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Sure. When the new group coordinator
> > > takes
> > > > > over,
> > > > > > > > it
> > > > > > > > > > has
> > > > > > > > > > > > to:
> > > > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a
> client
> > > side
> > > > > > > > assignor
> > > > > > > > > > is
> > > > > > > > > > > > used.
> > > > > > > > > > > > > >> We
> > > > > > > > > > > > > >> > > > don't
> > > > > > > > > > > > > >> > > > > > > store the information about the
> member
> > > > > selected
> > > > > > > > to run
> > > > > > > > > > > the
> > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > > > > >> > > > > > > * Update the topics metadata,
> verify the
> > > > > > > wildcard
> > > > > > > > > > > > > >> subscriptions, and
> > > > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> > > Could we
> > > > > > > > document
> > > > > > > > > > > what
> > > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > > >> > > > > > > > reflects? For example, does the
> epoch
> > > in
> > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> about the
> > > > > one in
> > > > > > > > > > > > > >> > > > > > > >
> ConsumerGroupPartitionMetadataValue
> > > and
> > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Sure. I will clarify that but it is
> > > always
> > > > > the
> > > > > > > > latest
> > > > > > > > > > > > group
> > > > > > > > > > > > > >> epoch.
> > > > > > > > > > > > > >> > > > > > > When the group state is updated, the
> > > group
> > > > > epoch
> > > > > > > > is
> > > > > > > > > > > > bumped so
> > > > > > > > > > > > > >> we use
> > > > > > > > > > > > > >> > > > > > > that one for all the change records
> > > related
> > > > > to
> > > > > > > the
> > > > > > > > > > > update.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will
> ensure
> > > > > that
> > > > > > > the
> > > > > > > > > > > > following
> > > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > > > possible
> > > > > > > for
> > > > > > > > a
> > > > > > > > > > > member
> > > > > > > > > > > > > >> not to
> > > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > That's right. Here I meant that the
> > > members
> > > > > > > > provided
> > > > > > > > > > by
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > >> > > > > > > in the assignment must exist in the
> > > group.
> > > > > The
> > > > > > > > > > assignor
> > > > > > > > > > > > can
> > > > > > > > > > > > > >> not make
> > > > > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> with a
> > > > > member
> > > > > > > > epoch
> > > > > > > > > > > > equals to
> > > > > > > > > > > > > >> 0":
> > > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member
> id
> > > > > would be
> > > > > > > > used?
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > A member is expected to abandon all
> its
> > > > > > > > partitions and
> > > > > > > > > > > > > >> rejoins when
> > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH
> error.
> > > In
> > > > > this
> > > > > > > > case,
> > > > > > > > > > > the
> > > > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > > > coordinator will have removed the
> member
> > > > > from
> > > > > > > the
> > > > > > > > > > group.
> > > > > > > > > > > > The
> > > > > > > > > > > > > >> member
> > > > > > > > > > > > > >> > > > > > > can rejoin the group with the same
> > > member
> > > > > id but
> > > > > > > > with
> > > > > > > > > > 0
> > > > > > > > > > > as
> > > > > > > > > > > > > >> epoch. Let
> > > > > > > > > > > > > >> > > > > > > me see if I can clarify this in the
> KIP.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will
> have
> > > the
> > > > > > > ability
> > > > > > > > to
> > > > > > > > > > > > trigger a
> > > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > > >> > > > > > > > by either providing a non-zero
> reason
> > > or
> > > > > by
> > > > > > > > updating
> > > > > > > > > > > the
> > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > > > conflicting
> > > > > > > > with
> > > > > > > > > > the
> > > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > In this case, a new assignment is
> > > triggered
> > > > > by
> > > > > > > the
> > > > > > > > > > > client
> > > > > > > > > > > > side
> > > > > > > > > > > > > >> > > > > > > assignor. When constructing the HB,
> the
> > > > > consumer
> > > > > > > > will
> > > > > > > > > > > > always
> > > > > > > > > > > > > >> consult
> > > > > > > > > > > > > >> > > > > > > the client side assignor and
> propagate
> > > the
> > > > > > > > information
> > > > > > > > > > > to
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > > > coordinator. In other words, we
> don't
> > > expect
> > > > > > > > users to
> > > > > > > > > > > call
> > > > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> > > nice.
> > > > > But
> > > > > > > the
> > > > > > > > > > > section
> > > > > > > > > > > > > >> seems to
> > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> > > epoch
> > > > > 2, B
> > > > > > > > > > > > immediately
> > > > > > > > > > > > > >> gets into
> > > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]",
> which
> > > seems
> > > > > > > > > > incorrect.
> > > > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> > > epoch
> > > > > 3, C
> > > > > > > > seems
> > > > > > > > > > to
> > > > > > > > > > > > get
> > > > > > > > > > > > > >> into
> > > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch
> 3, C
> > > > > still
> > > > > > > > has A -
> > > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers:
> Do
> > > we
> > > > > > > support
> > > > > > > > the
> > > > > > > > > > > > upgrade
> > > > > > > > > > > > > >> from
> > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > We will support upgrading from the
> > > consumer
> > > > > > > > protocol
> > > > > > > > > > > > version
> > > > > > > > > > > > > >> 3,
> > > > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is
> not
> > > > > > > implemented
> > > > > > > > yet
> > > > > > > > > > so
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> earliest
> > > > > > > > > > > > > >> > > > > > > version is unknown at the moment.
> This
> > > is
> > > > > > > > explained in
> > > > > > > > > > > the
> > > > > > > > > > > > > >> migration
> > > > > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Thanks again for your feedback,
> Jun. I
> > > will
> > > > > > > > update the
> > > > > > > > > > > KIP
> > > > > > > > > > > > > >> based on
> > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > > > > >> > > > > > > David
> > > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun
> Rao
> > > > > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the
> main
> > > > > benefits
> > > > > > > > of
> > > > > > > > > > the
> > > > > > > > > > > > KIP
> > > > > > > > > > > > > >> seem to
> > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > > efficient
> > > > > > > > support of
> > > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a
> singleton.
> > > Do we
> > > > > plan
> > > > > > > > to
> > > > > > > > > > > > support
> > > > > > > > > > > > > >> rolling
> > > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > > > consumers?
> > > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you
> explain
> > > > > whether
> > > > > > > > it's
> > > > > > > > > > > > > >> required in
> > > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> > > needs to
> > > > > be
> > > > > > > > filled?
> > > > > > > > > > > For
> > > > > > > > > > > > > >> example,
> > > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> > > needs to
> > > > > be
> > > > > > > > filled.
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 31. In the current consumer
> protocol,
> > > the
> > > > > rack
> > > > > > > > > > > affinity
> > > > > > > > > > > > > >> between the
> > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > >> > > > > > > > and the broker is only considered
> > > during
> > > > > > > > fetching,
> > > > > > > > > > but
> > > > > > > > > > > > not
> > > > > > > > > > > > > >> during
> > > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > > >> > > > > > > > partitions to consumers.
> Sometimes,
> > > once
> > > > > the
> > > > > > > > > > > assignment
> > > > > > > > > > > > is
> > > > > > > > > > > > > >> made,
> > > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> > > because
> > > > > no
> > > > > > > > replicas
> > > > > > > > > > > of
> > > > > > > > > > > > > >> assigned
> > > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > > wondering
> > > > > if we
> > > > > > > > should
> > > > > > > > > > > use
> > > > > > > > > > > > > >> this
> > > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > > >> > > > > > > > to address this by including rack
> in
> > > > > > > > GroupMember.
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often,
> it's
> > > > > useful to
> > > > > > > > know
> > > > > > > > > > how
> > > > > > > > > > > > busy
> > > > > > > > > > > > > >> a group
> > > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the
> event
> > > loop
> > > > > > > model,
> > > > > > > > it
> > > > > > > > > > > seems
> > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction
> of
> > > the
> > > > > time
> > > > > > > > the
> > > > > > > > > > > event
> > > > > > > > > > > > > >> loop is
> > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > > coordinator
> > > > > > > > failover
> > > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > > >> > > > > > > > does it need to trigger the check
> if
> > > any
> > > > > group
> > > > > > > > with
> > > > > > > > > > > the
> > > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > > >> > > > > > > > subscription now has a new
> matching
> > > topic?
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> > > Could we
> > > > > > > > document
> > > > > > > > > > > what
> > > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > > >> > > > > > > > reflects? For example, does the
> epoch
> > > in
> > > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > > >> > > > > > > > the latest group epoch? What
> about the
> > > > > one in
> > > > > > > > > > > > > >> > > > > > > >
> ConsumerGroupPartitionMetadataValue
> > > and
> > > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will
> ensure
> > > > > that
> > > > > > > the
> > > > > > > > > > > > following
> > > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > > > possible
> > > > > > > for
> > > > > > > > a
> > > > > > > > > > > member
> > > > > > > > > > > > > >> not to
> > > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group
> with a
> > > > > member
> > > > > > > > epoch
> > > > > > > > > > > > equals to
> > > > > > > > > > > > > >> 0":
> > > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member
> id
> > > > > would be
> > > > > > > > used?
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will
> have
> > > the
> > > > > > > ability
> > > > > > > > to
> > > > > > > > > > > > trigger a
> > > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > > >> > > > > > > > by either providing a non-zero
> reason
> > > or
> > > > > by
> > > > > > > > updating
> > > > > > > > > > > the
> > > > > > > > > > > > > >> assignor
> > > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > > > conflicting
> > > > > > > > with
> > > > > > > > > > the
> > > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> > > nice.
> > > > > But
> > > > > > > the
> > > > > > > > > > > section
> > > > > > > > > > > > > >> seems to
> > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> > > epoch
> > > > > 2, B
> > > > > > > > > > > > immediately
> > > > > > > > > > > > > >> gets into
> > > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]",
> which
> > > seems
> > > > > > > > > > incorrect.
> > > > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> > > epoch
> > > > > 3, C
> > > > > > > > seems
> > > > > > > > > > to
> > > > > > > > > > > > get
> > > > > > > > > > > > > >> into
> > > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch
> 3, C
> > > > > still
> > > > > > > > has A -
> > > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers:
> Do
> > > we
> > > > > > > support
> > > > > > > > the
> > > > > > > > > > > > upgrade
> > > > > > > > > > > > > >> from
> > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM
> David
> > > > > Jacot
> > > > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > > > > re-implement the
> > > > > > > > > > > > coordinator
> > > > > > > > > > > > > >> in
> > > > > > > > > > > > > >> > > > Java. I
> > > > > > > > > > > > > >> > > > > > > > > discussed this offline with a
> few
> > > folks
> > > > > and
> > > > > > > > folks
> > > > > > > > > > > are
> > > > > > > > > > > > > >> concerned
> > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > >> > > > > > > > > we could introduce many
> regressions
> > > in
> > > > > the
> > > > > > > old
> > > > > > > > > > > > protocol
> > > > > > > > > > > > > >> if we do
> > > > > > > > > > > > > >> > > > so.
> > > > > > > > > > > > > >> > > > > > > > > Therefore, I am going to remove
> this
> > > > > > > statement
> > > > > > > > > > from
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> KIP. It
> > > > > > > > > > > > > >> > > > is an
> > > > > > > > > > > > > >> > > > > > > > > implementation detail after all
> so
> > > it
> > > > > does
> > > > > > > not
> > > > > > > > > > have
> > > > > > > > > > > > to be
> > > > > > > > > > > > > >> > > > decided at
> > > > > > > > > > > > > >> > > > > > > > > this stage. We will likely
> start by
> > > > > trying
> > > > > > > to
> > > > > > > > > > > > refactor the
> > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM
> > > David
> > > > > Jacot
> > > > > > > <
> > > > > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> machine
> > > are:
> > > > > > > > "Empty,
> > > > > > > > > > > > assigning,
> > > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> > > Group
> > > > > States
> > > > > > > > > > > section,
> > > > > > > > > > > > > >> right?
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > This sentence does not refer
> to
> > > those
> > > > > > > group
> > > > > > > > > > states
> > > > > > > > > > > > but
> > > > > > > > > > > > > >> rather
> > > > > > > > > > > > > >> > > > to a
> > > > > > > > > > > > > >> > > > > > > > > > state machine replication
> (SMR).
> > > This
> > > > > > > > refers to
> > > > > > > > > > > the
> > > > > > > > > > > > > >> entire
> > > > > > > > > > > > > >> > > > state of
> > > > > > > > > > > > > >> > > > > > > > > > group coordinator which is
> > > replicated
> > > > > via
> > > > > > > > the
> > > > > > > > > > log
> > > > > > > > > > > > > >> layer. I will
> > > > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> > > state
> > > > > > > machine
> > > > > > > > is
> > > > > > > > > > > > modelled
> > > > > > > > > > > > > >> as an
> > > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > The idea is to follow a model
> > > similar
> > > > > to
> > > > > > > > the new
> > > > > > > > > > > > quorum
> > > > > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > > > > >> > > > > > > > > > will have N threads to process
> > > events.
> > > > > > > Each
> > > > > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > > > > >> > > > > > > > > > partition is assigned to a
> unique
> > > > > thread
> > > > > > > > and all
> > > > > > > > > > > the
> > > > > > > > > > > > > >> events
> > > > > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> > > > > processed
> > > > > > > by
> > > > > > > > this
> > > > > > > > > > > > > >> thread. This
> > > > > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > > > > >> > > > > > > > > > concurrency and will enable
> us to
> > > do
> > > > > > > > simulation
> > > > > > > > > > > > testing
> > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> > > machine
> > > > > per
> > > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > > consumer
> > > > > group"
> > > > > > > > owned
> > > > > > > > > > > by
> > > > > > > > > > > > a
> > > > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > > coordinator
> > > > > owns 2
> > > > > > > > > > > consumer
> > > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> will we
> > > > > have 1
> > > > > > > > state
> > > > > > > > > > > > machine
> > > > > > > > > > > > > >> for it,
> > > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes
> from
> > > > > there, I
> > > > > > > > > > think.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > > "*group.coordinator.threads"
> > > > > > > > > > *is
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> number
> > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > > >> > > > > > > > > > > to run the state machines.
> But
> > > I'm
> > > > > > > > wondering
> > > > > > > > > > if
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> purpose
> > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> > > each
> > > > > > > consumer
> > > > > > > > > > group
> > > > > > > > > > > > (or
> > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > > > computation,
> > > > > > > > why
> > > > > > > > > > > > should we
> > > > > > > > > > > > > >> need
> > > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have
> an
> > > > > ability to
> > > > > > > > shard
> > > > > > > > > > > the
> > > > > > > > > > > > > >> > > > processing as
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> > > timeout,
> > > > > why
> > > > > > > > does
> > > > > > > > > > the
> > > > > > > > > > > > > >> default
> > > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > > > max(60s)? I
> > > > > > > > > > thought
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> min/max
> > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> lower/upper
> > > > > bound
> > > > > > > of
> > > > > > > > it,
> > > > > > > > > > > no?
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.session.timeout.ms
> > > > > int
> > > > > > > > 30s The
> > > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> consumer
> > > > > group
> > > > > > > > > > protocol.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > group.consumer.min.session.timeout.ms
> > > > > > > > int 45s
> > > > > > > > > > > The
> > > > > > > > > > > > > >> minimum
> > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > group.consumer.max.session.timeout.ms
> > > > > > > > int 60s
> > > > > > > > > > > The
> > > > > > > > > > > > > >> maximum
> > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The
> > > default
> > > > > > > > session
> > > > > > > > > > > > timeout
> > > > > > > > > > > > > >> should
> > > > > > > > > > > > > >> > > > be 45s
> > > > > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > > > assignor
> > > > > > > are
> > > > > > > > > > > [range,
> > > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > > assignor.
> > > > > I'd
> > > > > > > > like to
> > > > > > > > > > > > know
> > > > > > > > > > > > > >> why not
> > > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> > > choose
> > > > > > > uniform
> > > > > > > > > > > assignor
> > > > > > > > > > > > > >> (former
> > > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> distribution.
> > > Any
> > > > > > > other
> > > > > > > > > > reason
> > > > > > > > > > > > we
> > > > > > > > > > > > > >> choose
> > > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors
> List
> > > range,
> > > > > > > > uniform
> > > > > > > > > > The
> > > > > > > > > > > > > >> server side
> > > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > The order on the server side
> has
> > > no
> > > > > > > > influence
> > > > > > > > > > > > because
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > > >> > > > > > must
> > > > > > > > > > > > > >> > > > > > > > > > chose the selector that he
> wants
> > > to
> > > > > use.
> > > > > > > > There
> > > > > > > > > > is
> > > > > > > > > > > no
> > > > > > > > > > > > > >> default
> > > > > > > > > > > > > >> > > > in the
> > > > > > > > > > > > > >> > > > > > > > > > current proposal. If the
> assignor
> > > is
> > > > > not
> > > > > > > > > > specified
> > > > > > > > > > > > by
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > client,
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > request is rejected. The
> default
> > > > > client
> > > > > > > > value
> > > > > > > > > > for
> > > > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is
> > > `uniform`
> > > > > > > though.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > Thanks for your very good
> > > comments,
> > > > > Luke.
> > > > > > > I
> > > > > > > > hope
> > > > > > > > > > > > that my
> > > > > > > > > > > > > >> > > > answers
> > > > > > > > > > > > > >> > > > > > help
> > > > > > > > > > > > > >> > > > > > > > > > to clarify things. I will
> update
> > > the
> > > > > KIP
> > > > > > > as
> > > > > > > > well
> > > > > > > > > > > > based
> > > > > > > > > > > > > >> on your
> > > > > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29
> AM
> > > Luke
> > > > > Chen
> > > > > > > <
> > > > > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator
> > > section, you
> > > > > > > > > > mentioned:
> > > > > > > > > > > > > >> > > > > > > > > > > > The new group coordinator
> will
> > > > > have a
> > > > > > > > state
> > > > > > > > > > > > machine
> > > > > > > > > > > > > >> per
> > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets*
> partitions,
> > > > > where
> > > > > > > > each
> > > > > > > > > > > state
> > > > > > > > > > > > > >> machine is
> > > > > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > > > > >> > > > > > > > > > > event loop. Those state
> machines
> > > > > will be
> > > > > > > > > > > executed
> > > > > > > > > > > > in
> > > > > > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads*
> > > threads.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state
> machine
> > > are:
> > > > > > > > "Empty,
> > > > > > > > > > > > assigning,
> > > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> > > Group
> > > > > States
> > > > > > > > > > > section,
> > > > > > > > > > > > > >> right?
> > > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> > > state
> > > > > > > machine
> > > > > > > > is
> > > > > > > > > > > > modelled
> > > > > > > > > > > > > >> as an
> > > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> > > machine
> > > > > per
> > > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > > consumer
> > > > > group"
> > > > > > > > owned
> > > > > > > > > > > by
> > > > > > > > > > > > a
> > > > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > > coordinator
> > > > > owns 2
> > > > > > > > > > > consumer
> > > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*,
> will we
> > > > > have 1
> > > > > > > > state
> > > > > > > > > > > > machine
> > > > > > > > > > > > > >> for it,
> > > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > > "*group.coordinator.threads"
> > > > > > > > > > *is
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> number
> > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > > >> > > > > > > > > > > to run the state machines.
> But
> > > I'm
> > > > > > > > wondering
> > > > > > > > > > if
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> purpose
> > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> > > each
> > > > > > > consumer
> > > > > > > > > > group
> > > > > > > > > > > > (or
> > > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > > > computation,
> > > > > > > > why
> > > > > > > > > > > > should we
> > > > > > > > > > > > > >> need
> > > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 2. For the default value in
> the
> > > new
> > > > > > > > configs:
> > > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> > > timeout,
> > > > > why
> > > > > > > > does
> > > > > > > > > > the
> > > > > > > > > > > > > >> default
> > > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > > > max(60s)? I
> > > > > > > > > > thought
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> min/max
> > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > >> > > > > > > > > > > timeout is to define
> lower/upper
> > > > > bound
> > > > > > > of
> > > > > > > > it,
> > > > > > > > > > > no?
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.session.timeout.ms
> > > > > int
> > > > > > > > 30s The
> > > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > >> > > > > > > > > > > failures when using the
> consumer
> > > > > group
> > > > > > > > > > protocol.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > group.consumer.min.session.timeout.ms
> > > > > > > > int 45s
> > > > > > > > > > > The
> > > > > > > > > > > > > >> minimum
> > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > group.consumer.max.session.timeout.ms
> > > > > > > > int 60s
> > > > > > > > > > > The
> > > > > > > > > > > > > >> maximum
> > > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > > > assignor
> > > > > > > are
> > > > > > > > > > > [range,
> > > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > > assignor.
> > > > > I'd
> > > > > > > > like to
> > > > > > > > > > > > know
> > > > > > > > > > > > > >> why not
> > > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> > > choose
> > > > > > > uniform
> > > > > > > > > > > assignor
> > > > > > > > > > > > > >> (former
> > > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > > >> > > > > > > > > > > for better evenly
> distribution.
> > > Any
> > > > > > > other
> > > > > > > > > > reason
> > > > > > > > > > > > we
> > > > > > > > > > > > > >> choose
> > > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors
> List
> > > range,
> > > > > > > > uniform
> > > > > > > > > > The
> > > > > > > > > > > > > >> server side
> > > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at
> 2:10 PM
> > > Luke
> > > > > > > Chen
> > > > > > > > <
> > > > > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts about
> > > Kafka
> > > > > > > Connect
> > > > > > > > > > > > > >> integrating with
> > > > > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > > > > >> > > > > > > > > > > > think we should have a
> > > separate
> > > > > > > > discussion
> > > > > > > > > > > > thread
> > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > >> > > > Kafka
> > > > > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka
> Connect
> > > > > With
> > > > > > > New
> > > > > > > > > > > Consumer
> > > > > > > > > > > > > >> Rebalance
> > > > > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > > > > >> > > > > > > > > > > > and let this discussion
> thread
> > > > > focus
> > > > > > > on
> > > > > > > > > > > consumer
> > > > > > > > > > > > > >> rebalance
> > > > > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >>
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at
> 9:31
> > > PM
> > > > > Sagar
> > > > > > > <
> > > > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David
> for
> > > the
> > > > > > > > feedback.
> > > > > > > > > > > > Looks
> > > > > > > > > > > > > >> like
> > > > > > > > > > > > > >> > > > there's
> > > > > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs for
> > > Connect.
> > > > > I
> > > > > > > > would
> > > > > > > > > > > > revisit
> > > > > > > > > > > > > >> the doc
> > > > > > > > > > > > > >> > > > and
> > > > > > > > > > > > > >> > > > > > see
> > > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at
> 7:11
> > > PM
> > > > > David
> > > > > > > > Jacot
> > > > > > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback
> > > and the
> > > > > > > > document.
> > > > > > > > > > > > That's
> > > > > > > > > > > > > >> really
> > > > > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me
> > > that
> > > > > both
> > > > > > > > Connect
> > > > > > > > > > > > and the
> > > > > > > > > > > > > >> > > > Consumer
> > > > > > > > > > > > > >> > > > > > could
> > > > > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the same underlying
> > > "engine".
> > > > > The
> > > > > > > > main
> > > > > > > > > > > > > >> difference is
> > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > assigns
> topic-partitions to
> > > > > members
> > > > > > > > > > whereas
> > > > > > > > > > > > > >> Connect
> > > > > > > > > > > > > >> > > > assigns
> > > > > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two
> ways to
> > > move
> > > > > > > > forward:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new
> > > proposed
> > > > > APIs
> > > > > > > to
> > > > > > > > > > > support
> > > > > > > > > > > > > >> different
> > > > > > > > > > > > > >> > > > > > resource
> > > > > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions,
> tasks,
> > > > > etc.); or
> > > > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated
> > > APIs
> > > > > for
> > > > > > > > Connect.
> > > > > > > > > > > The
> > > > > > > > > > > > > >> dedicated
> > > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new
> ones but
> > > > > > > > different on
> > > > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > they would rely on the
> same
> > > > > engine
> > > > > > > > on the
> > > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > > >> > > > side.
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean
> towards
> > > 2)
> > > > > > > because
> > > > > > > > I am
> > > > > > > > > > > > not a
> > > > > > > > > > > > > >> fan of
> > > > > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> > > > > purposes.
> > > > > > > > That
> > > > > > > > > > > being
> > > > > > > > > > > > > >> said, I am
> > > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an
> > > elegant
> > > > > way to
> > > > > > > > do
> > > > > > > > > > it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > I think that we can
> > > continue to
> > > > > > > > discuss
> > > > > > > > > > it
> > > > > > > > > > > > here
> > > > > > > > > > > > > >> for now
> > > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > > >> > > > > > order
> > > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> > > > > compatible
> > > > > > > > with
> > > > > > > > > > > what
> > > > > > > > > > > > we
> > > > > > > > > > > > > >> will do
> > > > > > > > > > > > > >> > > > for
> > > > > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at
> > > 2:41 PM
> > > > > > > David
> > > > > > > > > > Jacot
> > > > > > > > > > > <
> > > > > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from
> vacation.
> > > I
> > > > > will
> > > > > > > go
> > > > > > > > > > > through
> > > > > > > > > > > > and
> > > > > > > > > > > > > >> address
> > > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming days.
> > > Thanks
> > > > > for
> > > > > > > your
> > > > > > > > > > > > feedback.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022
> at
> > > 10:05
> > > > > PM
> > > > > > > > Gregory
> > > > > > > > > > > > Harris
> > > > > > > > > > > > > >> <
> > > > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP,
> > > it's
> > > > > > > > wonderful to
> > > > > > > > > > > see
> > > > > > > > > > > > > >> > > > cooperative
> > > > > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few
> questions:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> > > > > Alternatives'
> > > > > > > > > > section
> > > > > > > > > > > > > >> describes how
> > > > > > > > > > > > > >> > > > > > member
> > > > > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step
> with
> > > the
> > > > > group
> > > > > > > > epoch
> > > > > > > > > > > and
> > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good
> > > idea
> > > > > for
> > > > > > > the
> > > > > > > > > > > reasons
> > > > > > > > > > > > > >> described
> > > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is
> > > incrementally
> > > > > > > > assigning
> > > > > > > > > > > > > >> partitions to a
> > > > > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each
> > > incremental
> > > > > > > > > > assignment
> > > > > > > > > > > > use?
> > > > > > > > > > > > > >> Are
> > > > > > > > > > > > > >> > > > member
> > > > > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > a single member
> epoch
> > > can
> > > > > > > > correspond
> > > > > > > > > > to
> > > > > > > > > > > > > >> multiple
> > > > > > > > > > > > > >> > > > > > different
> > > > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > larger)
> assignments?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the
> Assignor's
> > > > > 'Reason'
> > > > > > > > field
> > > > > > > > > > > > opaque
> > > > > > > > > > > > > >> to the
> > > > > > > > > > > > > >> > > > group
> > > > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> > > > > client-side
> > > > > > > > > > assignor
> > > > > > > > > > > > > >> > > > implementations
> > > > > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and
> how
> > > is
> > > > > its
> > > > > > > > common
> > > > > > > > > > > > meaning
> > > > > > > > > > > > > >> agreed
> > > > > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a
> distinct
> > > > > Reason
> > > > > > > > field
> > > > > > > > > > over
> > > > > > > > > > > > > >> including
> > > > > > > > > > > > > >> > > > such
> > > > > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> > > > > included in
> > > > > > > > the
> > > > > > > > > > > KIP:
> > > > > > > > > > > > > >> "Thanks to
> > > > > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client side
> assignor is
> > > > > > > entirely
> > > > > > > > > > driven
> > > > > > > > > > > > by
> > > > > > > > > > > > > >> the group
> > > > > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no
> longer
> > > > > > > > responsible for
> > > > > > > > > > > > > >> maintaining
> > > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > > >> > > > > > state
> > > > > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned
> partitions."
> > > Does
> > > > > this
> > > > > > > > mean
> > > > > > > > > > > > that the
> > > > > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate
> additional
> > > > > > > > non-Metadata
> > > > > > > > > > > state
> > > > > > > > > > > > > >> (such as
> > > > > > > > > > > > > >> > > > > > partition
> > > > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics,
> > > config
> > > > > > > > topics,
> > > > > > > > > > > etc),
> > > > > > > > > > > > or
> > > > > > > > > > > > > >> that
> > > > > > > > > > > > > >> > > > > > additional
> > > > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be
> > > used?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that
> there are
> > > > > > > separate
> > > > > > > > > > > classes
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > > > > >> > > > > >
> > > > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > > > > >> > > >
> > > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap
> significantly.
> > > Is
> > > > > it
> > > > > > > > possible
> > > > > > > > > > > for
> > > > > > > > > > > > > >> these two
> > > > > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would
> > > serve
> > > > > to
> > > > > > > > promote
> > > > > > > > > > > > feature
> > > > > > > > > > > > > >> parity
> > > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side
> assignors,
> > > and
> > > > > > > would
> > > > > > > > also
> > > > > > > > > > > > > >> facilitate
> > > > > > > > > > > > > >> > > > > > operational
> > > > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > certain
> situations. For
> > > > > > > example,
> > > > > > > > if a
> > > > > > > > > > > > > >> server-side
> > > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs
> a
> > > patch,
> > > > > > > > deploying
> > > > > > > > > > > the
> > > > > > > > > > > > > >> patched
> > > > > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching one
> > > consumer
> > > > > > > group
> > > > > > > > to a
> > > > > > > > > > > > > >> client-side
> > > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> > > > > patching
> > > > > > > all
> > > > > > > > of
> > > > > > > > > > the
> > > > > > > > > > > > > >> brokers.
> > > > > > > > > > > > > >> > > > With
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct
> > > APIs, a
> > > > > > > > non-trivial
> > > > > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if
> the
> > > two
> > > > > APIs
> > > > > > > > have
> > > > > > > > > > > > diverged
> > > > > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > > > > reimplementation
> > > > > > > > > > would
> > > > > > > > > > > > not be
> > > > > > > > > > > > > >> > > > possible.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> gharris1727@gmail.com
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> github.com/gharris1727
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3,
> 2022 at
> > > > > 8:39 AM
> > > > > > > > Sagar
> > > > > > > > > > <
> > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi
> Guozhang/David,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a
> > > confluence
> > > > > page
> > > > > > > to
> > > > > > > > > > > discuss
> > > > > > > > > > > > how
> > > > > > > > > > > > > >> Connect
> > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> > > > > rebalance
> > > > > > > > > > protocol.
> > > > > > > > > > > > > >> Here's the
> > > > > > > > > > > > > >> > > > page:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >>
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty
> > > longish
> > > > > and
> > > > > > > I
> > > > > > > > have
> > > > > > > > > > > > tried
> > > > > > > > > > > > > >> to keep
> > > > > > > > > > > > > >> > > > a
> > > > > > > > > > > > > >> > > > > > format
> > > > > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me
> know
> > > > > what you
> > > > > > > > > > think.
> > > > > > > > > > > > Also,
> > > > > > > > > > > > > >> do you
> > > > > > > > > > > > > >> > > > > > think this
> > > > > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a
> separate
> > > > > > > discussion
> > > > > > > > > > thread
> > > > > > > > > > > > or
> > > > > > > > > > > > > >> is this
> > > > > > > > > > > > > >> > > > one
> > > > > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26,
> 2022
> > > at
> > > > > 7:37
> > > > > > > AM
> > > > > > > > > > > Sagar <
> > > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so
> much
> > > for
> > > > > the
> > > > > > > > doc on
> > > > > > > > > > > > Kafka
> > > > > > > > > > > > > >> Streams.
> > > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and
> come
> > > up
> > > > > with
> > > > > > > > such a
> > > > > > > > > > > > > >> document.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26,
> > > 2022 at
> > > > > > > 4:47
> > > > > > > > AM
> > > > > > > > > > > > Guozhang
> > > > > > > > > > > > > >> Wang <
> > > > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be
> great
> > > if
> > > > > you
> > > > > > > > could
> > > > > > > > > > > come
> > > > > > > > > > > > > >> back with
> > > > > > > > > > > > > >> > > > some
> > > > > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the
> > > Connect
> > > > > side
> > > > > > > > > > > > integration
> > > > > > > > > > > > > >> with
> > > > > > > > > > > > > >> > > > the new
> > > > > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides
> > > leveraging on
> > > > > the
> > > > > > > > new
> > > > > > > > > > > > "protocol
> > > > > > > > > > > > > >> type"
> > > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > > >> > > > > > did not
> > > > > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect
> side
> > > > > > > > > > implementations.
> > > > > > > > > > > > For
> > > > > > > > > > > > > >> Streams
> > > > > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > >
> > > > > > > > > > > > > >>
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for
> your
> > > > > analysis
> > > > > > > > on
> > > > > > > > > > > > Connect.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul
> 19,
> > > 2022
> > > > > at
> > > > > > > > 10:48 PM
> > > > > > > > > > > > Sagar <
> > > > > > > > > > > > > >> > > > > > > > > > > >>
> sagarmeansocean@gmail.com
> > > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you
> for
> > > your
> > > > > > > > response.
> > > > > > > > > > The
> > > > > > > > > > > > > >> reason I
> > > > > > > > > > > > > >> > > > thought
> > > > > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new
> scheme
> > > is
> > > > > that
> > > > > > > > even
> > > > > > > > > > > > today the
> > > > > > > > > > > > > >> > > > connect
> > > > > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending
> from
> > > > > > > > > > > > AbstractCoordinator to
> > > > > > > > > > > > > >> empower
> > > > > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > tasks/connectors.
> > > > > The
> > > > > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> > > > > metadata()
> > > > > > > > method
> > > > > > > > > > > by
> > > > > > > > > > > > > >> plumbing
> > > > > > > > > > > > > >> > > > into
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > JoinGroupRequestProtocol.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the
> > > changes
> > > > > to
> > > > > > > > support
> > > > > > > > > > > > > >> connect would
> > > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes
> in
> > > > > streams
> > > > > > > > mainly
> > > > > > > > > > > > because
> > > > > > > > > > > > > >> of the
> > > > > > > > > > > > > >> > > > > > Client
> > > > > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in
> both.
> > > At an
> > > > > > > > > > > implementation
> > > > > > > > > > > > > >> level, we
> > > > > > > > > > > > > >> > > > > > might
> > > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to
> get
> > > onto
> > > > > this
> > > > > > > > new
> > > > > > > > > > > > > >> assignment
> > > > > > > > > > > > > >> > > > protocol
> > > > > > > > > > > > > >> > > > > > like
> > > > > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > > > > > request/response and
> > > > > > > > > > > > > >> SyncGroup and
> > > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > ConsumerGroupHeartbeat
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc
> again on
> > > > > similar
> > > > > > > > lines
> > > > > > > > > > > to
> > > > > > > > > > > > > >> streams (or
> > > > > > > > > > > > > >> > > > > > there
> > > > > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> > > > > perform a
> > > > > > > > > > detailed
> > > > > > > > > > > > > >> analysis of
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > same
> > > > > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate
> > > discussion
> > > > > > > > thread for
> > > > > > > > > > > > that
> > > > > > > > > > > > > >> as that
> > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let
> me
> > > know
> > > > > if
> > > > > > > > that
> > > > > > > > > > > sounds
> > > > > > > > > > > > > >> good to
> > > > > > > > > > > > > >> > > > you.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul
> 15,
> > > > > 2022 at
> > > > > > > > 5:47
> > > > > > > > > > PM
> > > > > > > > > > > > David
> > > > > > > > > > > > > >> Jacot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for
> > > your
> > > > > > > > comments.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes.
> That
> > > > > refers to
> > > > > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The
> idea
> > > is to
> > > > > > > > > > transition C
> > > > > > > > > > > > from
> > > > > > > > > > > > > >> his
> > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target
> > > assignment
> > > > > when
> > > > > > > > he
> > > > > > > > > > can
> > > > > > > > > > > > move
> > > > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > > > > >> > > > > > When
> > > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > > assignment
> > > > > is
> > > > > > > > updated
> > > > > > > > > > > and
> > > > > > > > > > > > > >> persisted
> > > > > > > > > > > > > >> > > > > > with all
> > > > > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions
> > > even if
> > > > > > > they
> > > > > > > > are
> > > > > > > > > > > not
> > > > > > > > > > > > all
> > > > > > > > > > > > > >> revoked
> > > > > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > > assignment
> > > > > > > > becomes
> > > > > > > > > > the
> > > > > > > > > > > > target
> > > > > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> optimization
> > > to
> > > > > avoid
> > > > > > > > having
> > > > > > > > > > > to
> > > > > > > > > > > > > >> write all
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples
> are
> > > > > based on
> > > > > > > > the
> > > > > > > > > > > > persisted
> > > > > > > > > > > > > >> state
> > > > > > > > > > > > > >> > > > so I
> > > > > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion.
> > > Let me
> > > > > see
> > > > > > > > if I
> > > > > > > > > > can
> > > > > > > > > > > > > >> improve
> > > > > > > > > > > > > >> > > > this in
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3)
> Regarding
> > > > > Connect,
> > > > > > > it
> > > > > > > > > > could
> > > > > > > > > > > > > >> reuse the
> > > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor
> if it
> > > > > fits in
> > > > > > > > the
> > > > > > > > > > > > > >> protocol. The
> > > > > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > topicid-partitions +
> > > > > > > > > > metadata,
> > > > > > > > > > > > could
> > > > > > > > > > > > > >> > > > Connect
> > > > > > > > > > > > > >> > > > > > fit
> > > > > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri,
> Jul
> > > 15,
> > > > > 2022
> > > > > > > at
> > > > > > > > 1:55
> > > > > > > > > > > PM
> > > > > > > > > > > > > >> Sagar <
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> sagarmeansocean@gmail.com>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi
> David,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks
> for
> > > the
> > > > > KIP.
> > > > > > > I
> > > > > > > > just
> > > > > > > > > > > had
> > > > > > > > > > > > > >> minor
> > > > > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In
> the
> > > > > Assignment
> > > > > > > > Error
> > > > > > > > > > > > > >> section in
> > > > > > > > > > > > > >> > > > Client
> > > > > > > > > > > > > >> > > > > > Side
> > > > > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you
> > > mentioned
> > > > > => `In
> > > > > > > > this
> > > > > > > > > > > > case,
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > > >> > > > > > side
> > > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error
> to the
> > > > > group
> > > > > > > > > > > > coordinator`.
> > > > > > > > > > > > > >> In this
> > > > > > > > > > > > > >> > > > > > case are
> > > > > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> > > > > returning
> > > > > > > an
> > > > > > > > > > > > > >> AssignmentError
> > > > > > > > > > > > > >> > > > that's
> > > > > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes,
> do
> > > you
> > > > > think
> > > > > > > > it
> > > > > > > > > > > would
> > > > > > > > > > > > > >> make sense
> > > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In
> the
> > > Case
> > > > > > > Studies
> > > > > > > > > > > > section, I
> > > > > > > > > > > > > >> have a
> > > > > > > > > > > > > >> > > > > > slight
> > > > > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others
> have
> > > the
> > > > > > > same.
> > > > > > > > > > > Consider
> > > > > > > > > > > > > >> this step:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> > > > > heartbeats,
> > > > > > > the
> > > > > > > > > > group
> > > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> > > > > partitions
> > > > > > > to
> > > > > > > > > > > > revoke. It
> > > > > > > > > > > > > >> > > > persists
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> Group
> > > > > (epoch=3)
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> A
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> B
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> C
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> Target
> > > > > > > Assignment
> > > > > > > > > > > > (epoch=3)
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> A -
> > > > > > > > > > > partitions=[foo-0]
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> B -
> > > > > > > > > > > partitions=[foo-2]
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> C -
> > > > > > > > > > > partitions=[foo-1]
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    -
> Member
> > > > > > > Assignment
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> A -
> > > > > epoch=2,
> > > > > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> B -
> > > > > epoch=3,
> > > > > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       -
> C -
> > > > > epoch=3,
> > > > > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> > > > > heartbeats,
> > > > > > > it
> > > > > > > > > > > > transitions
> > > > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > > > > >> > > > > > but
> > > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> Here,it's
> > > > > mentioned
> > > > > > > > that
> > > > > > > > > > > > member C
> > > > > > > > > > > > > >> can't
> > > > > > > > > > > > > >> > > > get
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based
> on the
> > > > > > > > description
> > > > > > > > > > > > above,
> > > > > > > > > > > > > >> it seems
> > > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be
> > > better
> > > > > to
> > > > > > > > remove
> > > > > > > > > > it
> > > > > > > > > > > > and
> > > > > > > > > > > > > >> > > > populate it
> > > > > > > > > > > > > >> > > > > > only
> > > > > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see
> this
> > > in a
> > > > > lot
> > > > > > > of
> > > > > > > > > > other
> > > > > > > > > > > > > >> places, so
> > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > >> > > > > > I
> > > > > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> Regarding
> > > > > connect ,
> > > > > > > it
> > > > > > > > > > might
> > > > > > > > > > > > be
> > > > > > > > > > > > > >> out of
> > > > > > > > > > > > > >> > > > scope
> > > > > > > > > > > > > >> > > > > > of
> > > > > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> > > > > understood it
> > > > > > > > would
> > > > > > > > > > > > > >> probably be
> > > > > > > > > > > > > >> > > > > > running in
> > > > > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on
> the
> > > new
> > > > > > > > rebalance
> > > > > > > > > > > > > >> protocol as it
> > > > > > > > > > > > > >> > > > has
> > > > > > > > > > > > > >> > > > > > its
> > > > > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > Assignors(Eager
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > > > > IncrementalCooperative).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri,
> Jul
> > > 15,
> > > > > 2022
> > > > > > > > at
> > > > > > > > > > 5:00
> > > > > > > > > > > > PM
> > > > > > > > > > > > > >> David
> > > > > > > > > > > > > >> > > > Jacot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> > > > > Hector! Our
> > > > > > > > goal
> > > > > > > > > > is
> > > > > > > > > > > > to
> > > > > > > > > > > > > >> move
> > > > > > > > > > > > > >> > > > forward
> > > > > > > > > > > > > >> > > > > > with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> instead of
> > > > > relying
> > > > > > > > on
> > > > > > > > > > one
> > > > > > > > > > > > > >> generic API.
> > > > > > > > > > > > > >> > > > For
> > > > > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact
> same
> > > > > pattern
> > > > > > > > and
> > > > > > > > > > > > > >> reuse/share the
> > > > > > > > > > > > > >> > > > core
> > > > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server
> > > side.
> > > > > For
> > > > > > > the
> > > > > > > > > > > schema
> > > > > > > > > > > > > >> registry, I
> > > > > > > > > > > > > >> > > > > > think
> > > > > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> having a
> > > > > tailored
> > > > > > > > API to
> > > > > > > > > > > do
> > > > > > > > > > > > > >> simple
> > > > > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On
> Fri,
> > > Jul
> > > > > 15,
> > > > > > > > 2022 at
> > > > > > > > > > > > 10:22
> > > > > > > > > > > > > >> AM Ismael
> > > > > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> Three
> > > quick
> > > > > > > > comments:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1.
> > > Regarding
> > > > > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > > > > >> > > > vs
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > com.google.re2j.Pattern,
> > > > > > > we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > document the
> > > > > > > > > > differences
> > > > > > > > > > > > in
> > > > > > > > > > > > > >> more
> > > > > > > > > > > > > >> > > > detail
> > > > > > > > > > > > > >> > > > > > before
> > > > > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> another.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That
> > > said,
> > > > > if
> > > > > > > > people
> > > > > > > > > > > pass
> > > > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > semantics
> > > > > to be
> > > > > > > > > > honored.
> > > > > > > > > > > > If
> > > > > > > > > > > > > >> we are
> > > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> should
> > > > > consider
> > > > > > > > adding
> > > > > > > > > > > an
> > > > > > > > > > > > > >> overload
> > > > > > > > > > > > > >> > > > with
> > > > > > > > > > > > > >> > > > > > our own
> > > > > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> think
> > > we'd
> > > > > want
> > > > > > > to
> > > > > > > > > > > expose
> > > > > > > > > > > > > >> re2j's at
> > > > > > > > > > > > > >> > > > this
> > > > > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2.
> > > Regarding
> > > > > > > topic
> > > > > > > > > > ids,
> > > > > > > > > > > > any
> > > > > > > > > > > > > >> major new
> > > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it
> and
> > > > > should
> > > > > > > > handle
> > > > > > > > > > the
> > > > > > > > > > > > topic
> > > > > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> part we
> > > > > need to
> > > > > > > > > > handle.
> > > > > > > > > > > I
> > > > > > > > > > > > > >> agree with
> > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to
> the
> > > > > relevant
> > > > > > > > > > > protocols
> > > > > > > > > > > > > >> that don't
> > > > > > > > > > > > > >> > > > > > have it
> > > > > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> focus
> > > on the
> > > > > > > > internals
> > > > > > > > > > > > versus
> > > > > > > > > > > > > >> adding
> > > > > > > > > > > > > >> > > > new
> > > > > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> (unless
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we
> find
> > > that
> > > > > > > > adding
> > > > > > > > > > new
> > > > > > > > > > > > APIs
> > > > > > > > > > > > > >> is
> > > > > > > > > > > > > >> > > > required
> > > > > > > > > > > > > >> > > > > > for
> > > > > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3.
> I am
> > > > > still
> > > > > > > not
> > > > > > > > sure
> > > > > > > > > > > > about
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > powerful for
> > > > > > > > configs
> > > > > > > > > > to
> > > > > > > > > > > be
> > > > > > > > > > > > > >> > > > centralized
> > > > > > > > > > > > > >> > > > > > in the
> > > > > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> reasons
> > > > > > > > (auditability,
> > > > > > > > > > > > > >> visibility,
> > > > > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure
> > > about
> > > > > > > > > > automatically
> > > > > > > > > > > > > >> deleting
> > > > > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > recovered. A
> > > > > > > good
> > > > > > > > > > > property
> > > > > > > > > > > > > >> for modern
> > > > > > > > > > > > > >> > > > > > systems
> > > > > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > unrecoverable
> > > > > > > data
> > > > > > > > > > loss
> > > > > > > > > > > > > >> scenarios.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> Ismael
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On
> Wed,
> > > Jul
> > > > > 13,
> > > > > > > > 2022
> > > > > > > > > > at
> > > > > > > > > > > > 3:47
> > > > > > > > > > > > > >> PM David
> > > > > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Thanks
> > > > > > > > Guozhang. My
> > > > > > > > > > > > answers
> > > > > > > > > > > > > >> are
> > > > > > > > > > > > > >> > > > below:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> 1)
> > > the
> > > > > > > > migration
> > > > > > > > > > > path,
> > > > > > > > > > > > > >> especially
> > > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> flag
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> to
> > > > > enable
> > > > > > > the
> > > > > > > > new
> > > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > protocols /
> > > > > > > > rpcs
> > > > > > > > > > and
> > > > > > > > > > > > old
> > > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > group.
> > > > > How
> > > > > > > the
> > > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using
> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> new
> > > > > protocol
> > > > > > > > is
> > > > > > > > > > > > something
> > > > > > > > > > > > > >> we
> > > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > Noted. I
> > > > > just
> > > > > > > > > > > published
> > > > > > > > > > > > a
> > > > > > > > > > > > > >> new
> > > > > > > > > > > > > >> > > > version
> > > > > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > details
> > > > > about
> > > > > > > > this.
> > > > > > > > > > > See
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> and
> > > the
> > > > > > > > > > > "Compatibility,
> > > > > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > > > > >> > > > > > and
> > > > > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I
> > > have to
> > > > > > > think
> > > > > > > > > > > through
> > > > > > > > > > > > a
> > > > > > > > > > > > > >> few cases
> > > > > > > > > > > > > >> > > > > > now but
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > mechanism
> > > > > > > > should be
> > > > > > > > > > > > > >> understandable.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> 2)
> > > the
> > > > > usage
> > > > > > > > of
> > > > > > > > > > > topic
> > > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > part of
> > > > > RPCs
> > > > > > > > and
> > > > > > > > > > > admin
> > > > > > > > > > > > > >> client,
> > > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> APIs
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> to
> > > > > consumers
> > > > > > > > yet.
> > > > > > > > > > I
> > > > > > > > > > > > think
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > client
> > > > > to be
> > > > > > > > > > > > maintaining
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > leverage
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> on
> > > all
> > > > > the
> > > > > > > > > > augmented
> > > > > > > > > > > > > >> existing
> > > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > secondly,
> > > > > > > > should
> > > > > > > > > > we
> > > > > > > > > > > > ever
> > > > > > > > > > > > > >> consider
> > > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > public
> > > > > APIs
> > > > > > > as
> > > > > > > > > > well
> > > > > > > > > > > > (both
> > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> rebalance
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > listener for
> > > > > > > > cases
> > > > > > > > > > > > like
> > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> deletion-and-recreation).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> > > > > Assuming
> > > > > > > > that we
> > > > > > > > > > > > would
> > > > > > > > > > > > > >> include
> > > > > > > > > > > > > >> > > > > > converting
> > > > > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> RPCs
> > > to
> > > > > using
> > > > > > > > topic
> > > > > > > > > > > ids
> > > > > > > > > > > > in
> > > > > > > > > > > > > >> this
> > > > > > > > > > > > > >> > > > KIP,
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> fully
> > > > > operate
> > > > > > > > with
> > > > > > > > > > > topic
> > > > > > > > > > > > > >> ids. That
> > > > > > > > > > > > > >> > > > > > being
> > > > > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> the
> > > topics
> > > > > > > > names in
> > > > > > > > > > > > various
> > > > > > > > > > > > > >> APIs so
> > > > > > > > > > > > > >> > > > > > having a
> > > > > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> seems
> > > > > > > > inevitable to
> > > > > > > > > > > me.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> b) I
> > > don't
> > > > > > > have
> > > > > > > > a
> > > > > > > > > > > strong
> > > > > > > > > > > > > >> opinion on
> > > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > beyond the
> > > > > > > > scope of
> > > > > > > > > > > this
> > > > > > > > > > > > > >> KIP. I
> > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > >> > > > > > rather
> > > > > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> here
> > > and
> > > > > we
> > > > > > > can
> > > > > > > > > > > consider
> > > > > > > > > > > > > >> this
> > > > > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Coming
> > > > > back to
> > > > > > > > > > > Ismael's
> > > > > > > > > > > > > >> point about
> > > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > > > > >> > > > think
> > > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > favour of
> > > > > it.
> > > > > > > > The
> > > > > > > > > > > > consumer
> > > > > > > > > > > > > >> will
> > > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> the
> > > topics
> > > > > > > > exists
> > > > > > > > > > > before
> > > > > > > > > > > > > >> passing
> > > > > > > > > > > > > >> > > > them
> > > > > > > > > > > > > >> > > > > > into
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > protocol.
> > > > > > > > Obviously,
> > > > > > > > > > > the
> > > > > > > > > > > > > >> > > > coordinator
> > > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> not
> > > really
> > > > > > > have
> > > > > > > > a
> > > > > > > > > > way
> > > > > > > > > > > to
> > > > > > > > > > > > > >> reject an
> > > > > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> I'm
> > > > > agreeing
> > > > > > > > with
> > > > > > > > > > > > David
> > > > > > > > > > > > > >> on all
> > > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > > >> question:
> > > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > deprecate
> > > > > > > the
> > > > > > > > > > > > subscribe
> > > > > > > > > > > > > >> API with
> > > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > Pattern
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > while
> > > > > just
> > > > > > > > > > > documenting
> > > > > > > > > > > > > >> that our
> > > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > server.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > Since
> > > > > the
> > > > > > > > > > > incompatible
> > > > > > > > > > > > > >> case is a
> > > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > overloaded
> > > > > > > > > > `String`
> > > > > > > > > > > > based
> > > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> various
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > invalid
> > > > > > > > regexes.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> That
> > > could
> > > > > > > > work. I
> > > > > > > > > > > have
> > > > > > > > > > > > to
> > > > > > > > > > > > > >> look at
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > engines to
> > > > > > > > better
> > > > > > > > > > > > > >> understand the
> > > > > > > > > > > > > >> > > > > > potential
> > > > > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> understanding
> > > is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> that
> > > would
> > > > > > > work
> > > > > > > > for
> > > > > > > > > > > all
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> basic
> > > > > > > > > > > > > >> > > > > > regular
> > > > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> differences
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > between
> > > > > the
> > > > > > > two
> > > > > > > > are
> > > > > > > > > > > > mainly
> > > > > > > > > > > > > >> about
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > various
> > > > > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> wonder
> > > > > what
> > > > > > > > other
> > > > > > > > > > > people
> > > > > > > > > > > > > >> think
> > > > > > > > > > > > > >> > > > about
> > > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Best,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> David
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On
> > > Tue,
> > > > > Jul
> > > > > > > 12,
> > > > > > > > 2022
> > > > > > > > > > > at
> > > > > > > > > > > > > >> 11:28 PM
> > > > > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > wangguoz@gmail.com
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > Thanks
> > > > > > > David!
> > > > > > > > I
> > > > > > > > > > > think
> > > > > > > > > > > > on
> > > > > > > > > > > > > >> the high
> > > > > > > > > > > > > >> > > > > > level
> > > > > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> to
> > > > > > > concretize
> > > > > > > > a
> > > > > > > > > > bit
> > > > > > > > > > > > more:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> 1)
> > > the
> > > > > > > > migration
> > > > > > > > > > > path,
> > > > > > > > > > > > > >> especially
> > > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> flag
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> to
> > > > > enable
> > > > > > > the
> > > > > > > > new
> > > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > protocols /
> > > > > > > > rpcs
> > > > > > > > > > and
> > > > > > > > > > > > old
> > > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > group.
> > > > > How
> > > > > > > the
> > > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using
> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> new
> > > > > protocol
> > > > > > > > is
> > > > > > > > > > > > something
> > > > > > > > > > > > > >> we
> > > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> 2)
> > > the
> > > > > usage
> > > > > > > > of
> > > > > > > > > > > topic
> > > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > part of
> > > > > RPCs
> > > > > > > > and
> > > > > > > > > > > admin
> > > > > > > > > > > > > >> client,
> > > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> APIs
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> to
> > > > > consumers
> > > > > > > > yet.
> > > > > > > > > > I
> > > > > > > > > > > > think
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > client
> > > > > to be
> > > > > > > > > > > > maintaining
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > leverage
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> on
> > > all
> > > > > the
> > > > > > > > > > augmented
> > > > > > > > > > > > > >> existing
> > > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > secondly,
> > > > > > > > should
> > > > > > > > > > we
> > > > > > > > > > > > ever
> > > > > > > > > > > > > >> consider
> > > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > public
> > > > > APIs
> > > > > > > as
> > > > > > > > > > well
> > > > > > > > > > > > (both
> > > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> rebalance
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > listener for
> > > > > > > > cases
> > > > > > > > > > > > like
> > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> deletion-and-recreation).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> I'm
> > > > > agreeing
> > > > > > > > with
> > > > > > > > > > > > David
> > > > > > > > > > > > > >> on all
> > > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > > >> question:
> > > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > deprecate
> > > > > > > the
> > > > > > > > > > > > subscribe
> > > > > > > > > > > > > >> API with
> > > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > Pattern
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > while
> > > > > just
> > > > > > > > > > > documenting
> > > > > > > > > > > > > >> that our
> > > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > server.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > Since
> > > > > the
> > > > > > > > > > > incompatible
> > > > > > > > > > > > > >> case is a
> > > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > overloaded
> > > > > > > > > > `String`
> > > > > > > > > > > > based
> > > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> various
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > invalid
> > > > > > > > regexes.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > Guozhang
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> On
> > > Tue,
> > > > > Jul
> > > > > > > > 12,
> > > > > > > > > > 2022
> > > > > > > > > > > > at
> > > > > > > > > > > > > >> 5:23 AM
> > > > > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > Hi
> > > > > Ismael,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > Thanks for
> > > > > > > > your
> > > > > > > > > > > > > >> feedback. Let
> > > > > > > > > > > > > >> > > > me
> > > > > > > > > > > > > >> > > > > > answer
> > > > > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 1. I
> > > > > > > think
> > > > > > > > > > it's
> > > > > > > > > > > > > >> premature to
> > > > > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > deprecation
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > removal
> > > > > > > > of the
> > > > > > > > > > > > > >> existing group
> > > > > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> affects a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> core
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > client
> > > > > > > > > > protocol
> > > > > > > > > > > > and
> > > > > > > > > > > > > >> hence
> > > > > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > dependent
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > on
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > how
> > > > > > > > quickly
> > > > > > > > > > > > > >> applications
> > > > > > > > > > > > > >> > > > migrate
> > > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > That
> > > > > makes
> > > > > > > > > > sense.
> > > > > > > > > > > I
> > > > > > > > > > > > will
> > > > > > > > > > > > > >> > > > remove it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 2.
> > > > > The
> > > > > > > KIP
> > > > > > > > > > says
> > > > > > > > > > > we
> > > > > > > > > > > > > >> intend to
> > > > > > > > > > > > > >> > > > > > release
> > > > > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > clear
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > why. If
> > > > > > > we
> > > > > > > > > > added
> > > > > > > > > > > > that
> > > > > > > > > > > > > >> as a
> > > > > > > > > > > > > >> > > > way to
> > > > > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> remove
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > the
> > > > > > > group
> > > > > > > > > > > > protocol, I
> > > > > > > > > > > > > >> also
> > > > > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Let me
> > > > > > > > explain
> > > > > > > > > > my
> > > > > > > > > > > > > >> reasoning. As
> > > > > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > coordinator
> > > > > > > > in
> > > > > > > > > > > Java
> > > > > > > > > > > > > >> while we
> > > > > > > > > > > > > >> > > > > > implement
> > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > that
> > > > > the
> > > > > > > > > > internals
> > > > > > > > > > > > will
> > > > > > > > > > > > > >> be
> > > > > > > > > > > > > >> > > > slightly
> > > > > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> model).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > Therefore, I
> > > > > > > > > > > wanted
> > > > > > > > > > > > to
> > > > > > > > > > > > > >> tighten
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > coordinator
> > > > > > > > to
> > > > > > > > > > the
> > > > > > > > > > > > new
> > > > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > alternative
> > > > > > > > > > would
> > > > > > > > > > > > be to
> > > > > > > > > > > > > >> use a
> > > > > > > > > > > > > >> > > > flag
> > > > > > > > > > > > > >> > > > > > to do
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> relying
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > on the
> > > > > > > > software
> > > > > > > > > > > > upgrade.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 3.
> > > > > We
> > > > > > > > need to
> > > > > > > > > > > > flesh
> > > > > > > > > > > > > >> out the
> > > > > > > > > > > > > >> > > > > > details of
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > like
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > we're
> > > > > > > > saying
> > > > > > > > > > we
> > > > > > > > > > > > will
> > > > > > > > > > > > > >> support
> > > > > > > > > > > > > >> > > > > > online
> > > > > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > explain
> > > > > > > > this
> > > > > > > > > > in
> > > > > > > > > > > > > >> detail. It
> > > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > > >> > > > > > also
> > > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> it's
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > easier.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Yes,
> > > > > we
> > > > > > > will
> > > > > > > > > > > support
> > > > > > > > > > > > > >> online
> > > > > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > a
> > > > > group
> > > > > > > > using
> > > > > > > > > > the
> > > > > > > > > > > > old
> > > > > > > > > > > > > >> protocol
> > > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > protocol.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Let me
> > > > > > > > briefly
> > > > > > > > > > > > explain
> > > > > > > > > > > > > >> how that
> > > > > > > > > > > > > >> > > > > > will work
> > > > > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically
> a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > four
> > > > > step
> > > > > > > > > > process:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > 1. The
> > > > > > > > cluster
> > > > > > > > > > > must
> > > > > > > > > > > > be
> > > > > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > new
> > > > > group
> > > > > > > > > > > > coordinator.
> > > > > > > > > > > > > >> Both the
> > > > > > > > > > > > > >> > > > > > old and
> > > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > support
> > > > > > > the
> > > > > > > > old
> > > > > > > > > > > > > >> protocol and
> > > > > > > > > > > > > >> > > > rely
> > > > > > > > > > > > > >> > > > > > on the
> > > > > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> so
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > they
> > > > > can
> > > > > > > > work
> > > > > > > > > > > > together.
> > > > > > > > > > > > > >> This
> > > > > > > > > > > > > >> > > > point
> > > > > > > > > > > > > >> > > > > > is an
> > > > > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > do
> > > > > this
> > > > > > > one
> > > > > > > > live
> > > > > > > > > > > > > >> because it
> > > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > > >> > > > > > require
> > > > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > coordinator
> > > > > > > > and
> > > > > > > > > > > > > >> starting up the
> > > > > > > > > > > > > >> > > > > > new one
> > > > > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > unavailabilities.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > 2. The
> > > > > > > > cluster's
> > > > > > > > > > > > > >> metadata
> > > > > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > enable
> > > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > > > >> protocol.
> > > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > terminated
> > > > > > > > > > because
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> old
> > > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > protocol.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > 3. The
> > > > > > > > consumers
> > > > > > > > > > > > must be
> > > > > > > > > > > > > >> > > > upgraded
> > > > > > > > > > > > > >> > > > > > to a
> > > > > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > migration
> > > > > > > > (must
> > > > > > > > > > > have
> > > > > > > > > > > > > >> KIP-792).
> > > > > > > > > > > > > >> > > > If
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > Nothing
> > > > > > > > must be
> > > > > > > > > > > > done at
> > > > > > > > > > > > > >> this
> > > > > > > > > > > > > >> > > > point.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > 4. The
> > > > > > > > consumers
> > > > > > > > > > > > must
> > > > > > > > > > > > > >> be rolled
> > > > > > > > > > > > > >> > > > > > with the
> > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > consumer
> > > > > > > > group
> > > > > > > > > > is
> > > > > > > > > > > > > >> automatically
> > > > > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > using
> > > > > the
> > > > > > > > new
> > > > > > > > > > > > protocol
> > > > > > > > > > > > > >> joins
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > group.
> > > > > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > old
> > > > > > > > protocol are
> > > > > > > > > > > > being
> > > > > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > > > > >> > > > > > the old
> > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > new
> > > > > one.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Let me
> > > > > > > > clarify
> > > > > > > > > > all
> > > > > > > > > > > > of
> > > > > > > > > > > > > >> this in
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 4.
> > > > > I am
> > > > > > > > happy
> > > > > > > > > > > > that we
> > > > > > > > > > > > > >> are
> > > > > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > server,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > but
> > > > > it
> > > > > > > > seems
> > > > > > > > > > > like
> > > > > > > > > > > > > >> there
> > > > > > > > > > > > > >> > > > could be
> > > > > > > > > > > > > >> > > > > > some
> > > > > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> issues.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Will
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > have a
> > > > > > > > > > mechanism
> > > > > > > > > > > > for
> > > > > > > > > > > > > >> users to
> > > > > > > > > > > > > >> > > > > > detect
> > > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> regex
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > before
> > > > > > > > > > switching
> > > > > > > > > > > > to
> > > > > > > > > > > > > >> the new
> > > > > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > I
> > > > > think
> > > > > > > > that I
> > > > > > > > > > am
> > > > > > > > > > > a
> > > > > > > > > > > > bit
> > > > > > > > > > > > > >> more
> > > > > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > believe
> > > > > > > > that the
> > > > > > > > > > > > > >> majority of
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > cases
> > > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > work
> > > > > with
> > > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > > > >> engine. The
> > > > > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > and
> > > > > reject
> > > > > > > > the
> > > > > > > > > > > > consumer
> > > > > > > > > > > > > >> if the
> > > > > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > migration
> > > > > > > > path,
> > > > > > > > > > in
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> worst
> > > > > > > > > > > > > >> > > > case,
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> joining
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > the
> > > > > group
> > > > > > > > will
> > > > > > > > > > be
> > > > > > > > > > > > > >> rejected.
> > > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > > >> > > > > > should
> > > > > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> defence, I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > would
> > > > > say.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > One
> > > > > way
> > > > > > > for
> > > > > > > > > > > > customers to
> > > > > > > > > > > > > >> > > > validate
> > > > > > > > > > > > > >> > > > > > their
> > > > > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > prod
> > > > > would
> > > > > > > > be to
> > > > > > > > > > > > test
> > > > > > > > > > > > > >> them with
> > > > > > > > > > > > > >> > > > > > another
> > > > > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > could
> > > > > be
> > > > > > > > done
> > > > > > > > > > in a
> > > > > > > > > > > > > >> pre-prod
> > > > > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > extend the
> > > > > > > > > > > > > >> consumer-group tool
> > > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > mechanism.
> > > > > > > > Would
> > > > > > > > > > > > this be
> > > > > > > > > > > > > >> > > > enough in
> > > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 5.
> > > > > > > > Related to
> > > > > > > > > > > the
> > > > > > > > > > > > last
> > > > > > > > > > > > > >> > > > question,
> > > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> users
> > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > stick
> > > > > > > > with the
> > > > > > > > > > > > > >> current regex
> > > > > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > example,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > it
> > > > > may
> > > > > > > be
> > > > > > > > > > handy
> > > > > > > > > > > to
> > > > > > > > > > > > > >> keep using
> > > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > migrations
> > > > > > > > > > > simple
> > > > > > > > > > > > and
> > > > > > > > > > > > > >> then
> > > > > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> second
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > step.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > I
> > > > > > > understand
> > > > > > > > > > your
> > > > > > > > > > > > point
> > > > > > > > > > > > > >> but I
> > > > > > > > > > > > > >> > > > am
> > > > > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > > > actually
> > > > > > > > stay
> > > > > > > > > > > in
> > > > > > > > > > > > > >> this mode.
> > > > > > > > > > > > > >> > > > That
> > > > > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > simplifying
> > > > > > > > the
> > > > > > > > > > > > client
> > > > > > > > > > > > > >> because
> > > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > the
> > > > > > > > metadata on
> > > > > > > > > > > the
> > > > > > > > > > > > > >> client
> > > > > > > > > > > > > >> > > > side. I
> > > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 6.
> > > > > When
> > > > > > > > we say
> > > > > > > > > > > > that
> > > > > > > > > > > > > >> the group
> > > > > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> responsible
> > > for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > storing
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > the
> > > > > > > > > > > configurations
> > > > > > > > > > > > > >> and that
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > group is
> > > > > > > > > > > deleted.
> > > > > > > > > > > > > >> Will a
> > > > > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > configurations?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > That's
> > > > > > > > right.
> > > > > > > > > > The
> > > > > > > > > > > > > >> > > > configurations
> > > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > deleted.
> > > > > > > > They go
> > > > > > > > > > > > > >> together.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 7.
> > > > > Will
> > > > > > > > the
> > > > > > > > > > > > choice to
> > > > > > > > > > > > > >> store
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> coordinator
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> make
> > > it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > harder
> > > > > > > to
> > > > > > > > list
> > > > > > > > > > > all
> > > > > > > > > > > > > >> cluster
> > > > > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > I
> > > > > don't
> > > > > > > > think
> > > > > > > > > > so.
> > > > > > > > > > > > The
> > > > > > > > > > > > > >> group
> > > > > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > configs.
> > > > > > > If
> > > > > > > > you
> > > > > > > > > > > > want to
> > > > > > > > > > > > > >> know
> > > > > > > > > > > > > >> > > > all
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > ask all
> > > > > > > > the
> > > > > > > > > > > group
> > > > > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > > > > >> > > > > > You
> > > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > for
> > > > > > > > instance.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 8.
> > > > > How
> > > > > > > > would
> > > > > > > > > > > > someone
> > > > > > > > > > > > > >> > > > configure a
> > > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> consumers?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Have
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > we
> > > > > > > > considered
> > > > > > > > > > > > > >> allowing the
> > > > > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > Alternatively,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > configs
> > > > > > > > could
> > > > > > > > > > be
> > > > > > > > > > > > > >> decoupled
> > > > > > > > > > > > > >> > > > from
> > > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Yes.
> > > > > The
> > > > > > > > group
> > > > > > > > > > > will
> > > > > > > > > > > > be
> > > > > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > the
> > > > > > > configs
> > > > > > > > will
> > > > > > > > > > > be
> > > > > > > > > > > > > >> lost after
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > passes.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 9.
> > > > > Will
> > > > > > > > the
> > > > > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > > > > >> > > > > > method
> > > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > > > > >> > > > do
> > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > That's a
> > > > > > > > very
> > > > > > > > > > > group
> > > > > > > > > > > > > >> question. I
> > > > > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > > >> > > > > > fully
> > > > > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > we
> > > > > plan to
> > > > > > > > use,
> > > > > > > > > > it
> > > > > > > > > > > > > >> might be
> > > > > > > > > > > > > >> > > > better
> > > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> overload
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > which
> > > > > > > takes
> > > > > > > > a
> > > > > > > > > > > > string.
> > > > > > > > > > > > > >> We would
> > > > > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> validation.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > During the
> > > > > > > > > > > > migration, I
> > > > > > > > > > > > > >> think
> > > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > regex
> > > > > and
> > > > > > > > use
> > > > > > > > > > it.
> > > > > > > > > > > > That
> > > > > > > > > > > > > >> should
> > > > > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > cases.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 10.
> > > > > I
> > > > > > > > agree
> > > > > > > > > > with
> > > > > > > > > > > > > >> Justine
> > > > > > > > > > > > > >> > > > that we
> > > > > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > switch
> > > > > > > to
> > > > > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > > > > >> > > > > > from the
> > > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > can
> > > > > > > > switch the
> > > > > > > > > > > > > >> default for
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > feature
> > > > > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > metadata.version
> > > > > > > > > > > > once
> > > > > > > > > > > > > >> we
> > > > > > > > > > > > > >> > > > want to
> > > > > > > > > > > > > >> > > > > > make
> > > > > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > My
> > > > > plan
> > > > > > > was
> > > > > > > > to
> > > > > > > > > > use
> > > > > > > > > > > > that
> > > > > > > > > > > > > >> feature
> > > > > > > > > > > > > >> > > > > > flag
> > > > > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> development
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > phase. I
> > > > > > > > should
> > > > > > > > > > > not
> > > > > > > > > > > > have
> > > > > > > > > > > > > >> > > > mentioned
> > > > > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > an
> > > > > > > internal
> > > > > > > > > > config
> > > > > > > > > > > > for
> > > > > > > > > > > > > >> it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 11.
> > > > > Some
> > > > > > > > of
> > > > > > > > > > the
> > > > > > > > > > > > > >> protocol APIs
> > > > > > > > > > > > > >> > > > > > don't
> > > > > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> would
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > good to
> > > > > > > > add
> > > > > > > > > > that
> > > > > > > > > > > > for
> > > > > > > > > > > > > >> > > > consistency.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Noted.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 12.
> > > > > It
> > > > > > > is
> > > > > > > > a
> > > > > > > > > > bit
> > > > > > > > > > > > odd
> > > > > > > > > > > > > >> that
> > > > > > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > even
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > though
> > > > > > > it
> > > > > > > > > > seems
> > > > > > > > > > > > to do
> > > > > > > > > > > > > >> more
> > > > > > > > > > > > > >> > > > than
> > > > > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > I
> > > > > agree.
> > > > > > > > This is
> > > > > > > > > > > > how the
> > > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > require
> > > > > > > > "Read
> > > > > > > > > > > > Group" to
> > > > > > > > > > > > > >> join a
> > > > > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > but I
> > > > > am
> > > > > > > not
> > > > > > > > > > sure
> > > > > > > > > > > > that
> > > > > > > > > > > > > >> it is
> > > > > > > > > > > > > >> > > > worth
> > > > > > > > > > > > > >> > > > > > it.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 13.
> > > > > How
> > > > > > > is
> > > > > > > > > > topic
> > > > > > > > > > > > > >> recreation
> > > > > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> group
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > protocol?
> > > > > > > > It
> > > > > > > > > > > > would be
> > > > > > > > > > > > > >> good to
> > > > > > > > > > > > > >> > > > > > have a
> > > > > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Noted.
> > > > > > > From
> > > > > > > > a
> > > > > > > > > > > > protocol
> > > > > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > topic
> > > > > id
> > > > > > > so
> > > > > > > > it
> > > > > > > > > > > will
> > > > > > > > > > > > > >> treat it
> > > > > > > > > > > > > >> > > > like a
> > > > > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > only
> > > > > issue
> > > > > > > > is
> > > > > > > > > > that
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > IDs
> > > > > so the
> > > > > > > > > > > consumer
> > > > > > > > > > > > > >> would
> > > > > > > > > > > > > >> > > > reuse the
> > > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > that
> > > > > we
> > > > > > > > should
> > > > > > > > > > > > update
> > > > > > > > > > > > > >> those
> > > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > > >> > > > > > as well
> > > > > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > to
> > > > > end.
> > > > > > > That
> > > > > > > > > > would
> > > > > > > > > > > > > >> strengthen
> > > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 14.
> > > > > The
> > > > > > > > KIP
> > > > > > > > > > > > mentions
> > > > > > > > > > > > > >> we will
> > > > > > > > > > > > > >> > > > > > write the
> > > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > though
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > this is
> > > > > > > an
> > > > > > > > > > > > > >> implementation
> > > > > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> module
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > for
> > > > > it?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Yes.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 15.
> > > > > Do
> > > > > > > we
> > > > > > > > > > have a
> > > > > > > > > > > > > >> scalability
> > > > > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> members
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > new
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > group
> > > > > > > > protocol
> > > > > > > > > > > can
> > > > > > > > > > > > > >> support?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > We
> > > > > don't
> > > > > > > > have
> > > > > > > > > > > > numbers
> > > > > > > > > > > > > >> at the
> > > > > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > of
> > > > > members
> > > > > > > > per
> > > > > > > > > > > > group.
> > > > > > > > > > > > > >> We will
> > > > > > > > > > > > > >> > > > > > measure
> > > > > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > implementation.
> > > > > > > > > > > Note
> > > > > > > > > > > > > >> that we
> > > > > > > > > > > > > >> > > > might
> > > > > > > > > > > > > >> > > > > > have
> > > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > road
> > > > > (e.g.
> > > > > > > > > > offset
> > > > > > > > > > > > > >> commits).
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > 16.
> > > > > Did
> > > > > > > we
> > > > > > > > > > > > consider
> > > > > > > > > > > > > >> having
> > > > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > SubscribedTopicNames
> > > > > > > > > > > > > >> in
> > > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> that
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > since we
> > > > > > > > have
> > > > > > > > > > to
> > > > > > > > > > > > > >> resolve the
> > > > > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > the
> > > > > > > topic
> > > > > > > > > > name?
> > > > > > > > > > > > The
> > > > > > > > > > > > > >> > > > difference
> > > > > > > > > > > > > >> > > > > > is that
> > > > > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > efficient
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > whereas
> > > > > > > > > > sending
> > > > > > > > > > > > the
> > > > > > > > > > > > > >> topic
> > > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > > >> > > > > > is less
> > > > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> Furthermore,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> delete
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > recreation is
> > > > > > > > > > > > easier
> > > > > > > > > > > > > >> to
> > > > > > > > > > > > > >> > > > handle
> > > > > > > > > > > > > >> > > > > > if we
> > > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > The
> > > > > idea
> > > > > > > > was to
> > > > > > > > > > > > > >> consolidate the
> > > > > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > paths
> > > > > but
> > > > > > > I
> > > > > > > > do
> > > > > > > > > > > agree
> > > > > > > > > > > > > >> with your
> > > > > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > ids
> > > > > may be
> > > > > > > > > > better
> > > > > > > > > > > > here
> > > > > > > > > > > > > >> for the
> > > > > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > suppose
> > > > > > > > that we
> > > > > > > > > > > may
> > > > > > > > > > > > > >> allow
> > > > > > > > > > > > > >> > > > users to
> > > > > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > future
> > > > > > > > because
> > > > > > > > > > > that
> > > > > > > > > > > > is
> > > > > > > > > > > > > >> the only
> > > > > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > re-creation.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > Best,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > David
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > On
> > > > > Tue,
> > > > > > > Jul
> > > > > > > > 12,
> > > > > > > > > > > > 2022 at
> > > > > > > > > > > > > >> 1:38 PM
> > > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > djacot@confluent.io>
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> wrote:
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > Hi
> > > > > > > > Justine,
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > Thanks
> > > > > > > for
> > > > > > > > > > your
> > > > > > > > > > > > > >> comments.
> > > > > > > > > > > > > >> > > > Please
> > > > > > > > > > > > > >> > > > > > find
> > > > > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > -
> > > > > Yes,
> > > > > > > > the new
> > > > > > > > > > > > > >> protocol
> > > > > > > > > > > > > >> > > > relies on
> > > > > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > > > topic
> > > > > > > > names
> > > > > > > > > > > based
> > > > > > > > > > > > in
> > > > > > > > > > > > > >> the
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > if
> > > > > using
> > > > > > > > topic
> > > > > > > > > > > > names
> > > > > > > > > > > > > >> is the
> > > > > > > > > > > > > >> > > > > > right call
> > > > > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about
> it
> > > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >
> > > a
> > > > > little
> > > > > > > > more.
> > > > > > > > > > > > > >> Obviously, the
> > > > > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hey Guozhang,

Thanks for your comments.

1. That's a very good question. I think that it is better to separate
the two concerns here. In my opinion, the GPA should only be used to
feed the assignor. I see the GPA as a way to call a remote function so
the GPA provides all its inputs to ensure that the assignor does not
depend on any local state. Knowing this, I would not use the GPA to
update the local state either. That is not its purpose. The consumer
is responsible to update its local metadata cache either when it gets
an unknown topic in the user-provided subscription or in the
coordinator-provided assignment. The consumer has to do this anyway to
get all topic metadata (e.g. leaders).

2. Interesting thought. Handling the translation is definitely
possible. I wonder if there are cases where having the topic ids is
beneficial. I am thinking about when a topic is deleted and recreated
right away. In this case, the current member assignment would still
have the old topic id and we would not be able to resolve it anymore.
I suppose that we should directly remove it from the current
assignment on the server side in this case and only add the new one.
As I first thought, that seems OK. The major downside is that the user
of the assignor would not know that the topic is different from what
it may have cached locally.

We could say that we don't expose topic ids externally for now in
order to stay consistent with the other interfaces but it would be
annoying to add it back if we want to expose topic ids in the future.
I personally believe that we will have to expose them at some point.
For instance, in the rebalance listener, it would be useful to know
that a topic has been deleted and recreated. The only way to do this
is to know the topic id. Let me think a little more about this and
come back to you.

3. Sorry for this one. I was a typo. PartitionAssingor#assign should
return GroupAssignment.

Best,
David

On Thu, Sep 29, 2022 at 6:06 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Thanks David, I read through the updated wiki, and I have a couple
> follow-up questions:
>
> 1. It seems we would have two potential sources to update the consumer's
> internally maintained topic name -> topic ID mapping, which is used for
> encoding subscribed topic Ids in the HB request, and offset commit / list
> offset requests, etc. One is from metadata responses, and another one is
> from the GPA (group prepare assignment). On the other hand, the assignor
> APIs directly take parsed AssignmentMemberSpec and AssignmentTopicMetadata
> from GPA responses directly instead of trying to consult the consumer's own
> cached mapping. So just to clarify, consumers would also update its
> internal mapping from the GPA responses as well, right?
>
> 2. I'm wondering if it's necessary to expose the topic IDs to the assignor
> developers. What they need to do at the moment is to take the mapping from
> AssignmentTopicMetadata, and translate their decided assignment into topic
> IDs when returning the MemberAssignment. We can instead just let consumer
> to do such translation for them internally and still let the assignor
> developers to only deal with the topic names. Of course that means the
> client side assignor APIs would be a bit further diverged from the
> broker-side assignor such that:
>
> a) Topic metadata would be encoded just as String -> num.partitions;
> b) AssignmentMemberSpec contains subscribedTopicNames and topicPartitions
> as target assignments;
> c) Returned MemberAssignment contains list of topicPartitions instead of
> topicIdPartitions.
>
> Internally, consumers would convert the IDs to names when calling the
> assign() function, and convert them back when sending the request; and
> similarly it would convert the IDs to names when calling onAssignment() as
> well.
>
> In short, I'm wondering if we should bite the bullet for the users, eating
> more complexities internally than pushing that to users.
>
> 3. This is a nit one: the return type of `Assignment assign(AssignmentSpec
> assignmentSpec);` seems not defined in the doc. I think it would just
> include two fields, as:
>
> class Assignment {
>         GroupAssignment groupAssignment;
>
>         ConsumerGroupMetadata consumerGroupMetadata;
>     }
>
> Is that right?
>
> Guozhang
>
> On Thu, Sep 29, 2022 at 1:26 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Guozhang,
> >
> > Thanks for your input.
> >
> > > 1) in HB request, when clients send the subscription, they are still
> > > sending them in terms of topic names. That is, the consumer would not try
> > > to keep all the mappings from topic names to topic ids for the subscribed
> > > topics.
> >
> > This is a mistake of mine. The heartbeat request should use Topic Ids.
> > I forgot to change it. The idea is that the client will get the
> > metadata first to translate the topic names to topic ids and thus also
> > have the mapping in its local metadata cache.
> >
> > > 2) in HB responses, the assignment's topics are encoded as IDs already.
> > And
> > > fetch requests can directly rely on that since they just need topic IDs
> > and
> > > do not care what actual topic names are.
> >
> > That's right.
> >
> > > 3) in prepare-assignment responses, we return subscriptions as topic IDs
> > as
> > > well.
> >
> > That's right.
> >
> > > In a word, the consumer does not try to maintain the mapping of topic
> > names
> > > to IDs at all, and even if they do, they will not expose it in e.g.
> > > assignor APIs. But for client-side assignors like Streams, if they do not
> > > maintain the mapping from topic IDs to topic names, certain logic such as
> > > co-partitioning indeed cannot be executed.
> >
> > As I said, the consumer will have the mapping in its local metadata
> > cache. It needs to get the partition metadata anyway and the topic id
> > is part of it. However, I do agree that this mapping is not exposed.
> >
> > > So I think we'd need to consider 1) use topic names in subscriptions in
> > the
> > > prepare-assignment responses, 2) use topic Names in client-side
> > > PartitionAssignor#GroupMember#subscribedTopics and
> > > PartitionAssignor#TopicMetadata#topicName.
> >
> > 1) makes sense to me. I think that we could actually provide both
> > names and ids. Regarding 2), I think that I am slightly in favor of
> > using both names and ids as well. Let me try to update the protocol
> > and the interface towards this.
> >
> > Thanks,
> > David
> >
> > On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wa...@gmail.com> wrote:
> > >
> > > Hello Jun/David,
> > >
> > > Just to chime in here for Jun's comment #50, today Streams does not have
> > a
> > > way to map topic names to topic IDs, and in its topology description
> > topics
> > > are encoded with names.
> > >
> > > What I originally thought was that, on the consumer side, we'd need to
> > > maintain this mapping from topic names to topic IDs anyways, and Streams
> > > should leverage on the consumer's functionality to get the corresponding
> > > topic names for the given topic IDs. But when I read through the RPC
> > again
> > > I realized it may not be the case, since:
> > >
> > > 1) in HB request, when clients send the subscription, they are still
> > > sending them in terms of topic names. That is, the consumer would not try
> > > to keep all the mappings from topic names to topic ids for the subscribed
> > > topics.
> > > 2) in HB responses, the assignment's topics are encoded as IDs already.
> > And
> > > fetch requests can directly rely on that since they just need topic IDs
> > and
> > > do not care what actual topic names are.
> > > 3) in prepare-assignment responses, we return subscriptions as topic IDs
> > as
> > > well.
> > >
> > > In a word, the consumer does not try to maintain the mapping of topic
> > names
> > > to IDs at all, and even if they do, they will not expose it in e.g.
> > > assignor APIs. But for client-side assignors like Streams, if they do not
> > > maintain the mapping from topic IDs to topic names, certain logic such as
> > > co-partitioning indeed cannot be executed.
> > >
> > > So I think we'd need to consider 1) use topic names in subscriptions in
> > the
> > > prepare-assignment responses, 2) use topic Names in client-side
> > > PartitionAssignor#GroupMember#subscribedTopics and
> > > PartitionAssignor#TopicMetadata#topicName.
> > >
> > > On Wed, Sep 28, 2022 at 1:26 AM David Jacot <djacot@confluent.io.invalid
> > >
> > > wrote:
> > >
> > > > Hi Jun,
> > > >
> > > > Thanks for your comments.
> > > >
> > > > 48. Okay. There is a typo in the text:
> > > > > Note that the persisted target assignment of B already includes foo-2
> > > > here
> > > > but B does not get it yet because foo-2 is not free.
> > > >
> > > > It should be:
> > > > > Note that the persisted member assignment of B already includes foo-2
> > > > here
> > > > but B does not get it yet because foo-2 is not free.
> > > >
> > > > The examples are showing the persisted state in the __consumer_offsets
> > > > topics. When a member transitions to the next epoch, we directly
> > > > persist the whole target assignment in order to reduce the number of
> > > > writes and we manage dependencies in memory. So "B - epoch=2,
> > > > partitions=[foo-2]" is actually correct from a persisted record
> > > > perspective. However, the in-memory state would be different because
> > > > foo-2 is not yet available as you pointed out. The in-memory state
> > > > would like as follow:
> > > >
> > > > Group (epoch=2)
> > > > A
> > > > B
> > > > Target Assignment (epoch=2)
> > > > A - partitions=[foo-0, foo-1]
> > > > B - partitions=[foo-2]
> > > > Member Assignment
> > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending partitions=[]
> > > > B - epoch=2, partitions=[], pending partitions=[foo-2]
> > > >
> > > > I suppose that the confusion comes from this. I will go and convert
> > > > all the examples to their in-memory representation in order to reduce
> > > > the confusion.
> > > >
> > > > 50. Okay. I have to admit that I am not so familiar with the internals
> > > > of KStreams. If I understand correctly, you're saying that KStreams
> > > > may not have topic ids in its topology so topic names are missing here
> > > > to match the topics. I was hoping that KStreams would have topic ids
> > > > in the topology as well. If that is not the case, we can also add
> > > > topic names on our side. I will check this.
> > > >
> > > > 53. It seems that I was not clear. Sorry for that. I did not mean to
> > > > say that MemberAssignment.metadata is not used by the group
> > > > coordinator. Let me re-explain the flow.
> > > >
> > > > PartitionAssignor.assign() returns an Assignment. Assignment contains
> > > > MemberAssignments where each member has MemberAssignment.metadata. The
> > > > Assignment is installed by the client with the
> > > > ConsumerGroupInstallAssignment API and this overrides the current
> > > > target assignment in the coordinator. Then the group coordinator uses
> > > > it to reconcile members. Therefore the target assignment will
> > > > eventually become the current member assignment and it will be
> > > > delivered to the member, including the original
> > > > MemberAssignment.metadata defines by the assignor.
> > > >
> > > > I hope that this helps to clarify how MemberAssignment.metadata is
> > > > used afterwards.
> > > >
> > > > Best,
> > > > David
> > > >
> > > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <ju...@confluent.io.invalid>
> > wrote:
> > > > > Hi, David,
> > > > >
> > > > > Thanks for the reply. A few more follow up questions.
> > > > >
> > > > > 48. Hmm, so member assignment is what the client sees. Then, I am not
> > > > sure
> > > > > about the following. The text says B cannot get foo-2 yet. But in
> > Member
> > > > > Assignment, B has foo-2.
> > > > >
> > > > > At this point B can transition to epoch 2 but cannot get foo-2 until
> > A
> > > > > revokes it.
> > > > >
> > > > > Note that the persisted target assignment of B already includes foo-2
> > > > here
> > > > > but B does not get it yet because foo-2 is not free.
> > > > >
> > > > > Group (epoch=2)
> > > > > A
> > > > > B
> > > > > Target Assignment (epoch=2)
> > > > > A - partitions=[foo-0, foo-1]
> > > > > B - partitions=[foo-2]
> > > > > Member Assignment
> > > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > > > B - epoch=2, partitions=[foo-2]
> > > > >
> > > > >
> > > > > 50. Hmm, I thought KStreams doesn't use range assignor all the time.
> > It
> > > > > does the assignment based on the topology. For example, if you are
> > > > joining
> > > > > two topics A and B and deriving the join result in a 3rd topic C,
> > which
> > > > > will be aggregated. You want to use range to co-locate the partitions
> > > > from
> > > > > topic A and B. However, for topic C, the partition could potentially
> > be
> > > > > assigned in round robin. Without the topic name, KStreams won't know
> > > > which
> > > > > topics need to be co-located.
> > > > >
> > > > > 53. If MemberAssignment.metadata is not used by the group
> > coordinator,
> > > > > where is it being used?
> > > > >
> > > > > Thanks,
> > > > >
> > > > > Jun
> > > > >
> > > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > > > <ja...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > > > Hi David,
> > > > > >
> > > > > > Thanks for the response. Replied below:
> > > > > >
> > > > > > > I don't have a strong opinion on this one. However, it seems that
> > > > this
> > > > > > could be a pandora box if our subset turns out to not be sufficient
> > > > > > for some existing users. We will have to add more to it and the
> > > > > > compatibility story between versions could be a real nightmare. One
> > > > > > thing that we could do for sure is disabling all unicode groups in
> > the
> > > > > > library. That seems possible and would already reduce the scope.
> > > > > >
> > > > > > Changes to the regex syntax seems not a major problem to me. We
> > would
> > > > bump
> > > > > > the protocol version as we usually do. Actually that begs the same
> > > > question
> > > > > > about RE2. How stable is the syntax and does it differ depending
> > on the
> > > > > > version of the library? In general, I do agree that we'd be better
> > off
> > > > not
> > > > > > having to manage the syntax ourselves if we can convince ourselves
> > that
> > > > > > compatibility won't be a problem and the client won't eventually
> > need
> > > > to
> > > > > > have access to it.
> > > > > >
> > > > > > Thanks,
> > > > > > Jason
> > > > > >
> > > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > > > <dj...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > > > Hi Luke,
> > > > > > >
> > > > > > > Thanks for your comments.
> > > > > > >
> > > > > > > Your suggestion makes sense. Jun's made the same one. We are also
> > > > > > > discussing whether we should directly return the group state in
> > the
> > > > > > > response and drop the prepare assignment api. It seems that the
> > > > > > > current consensus is to keep it though. If that stays like this,
> > I
> > > > > > > will apply your suggestion.
> > > > > > >
> > > > > > > Thanks,
> > > > > > > David
> > > > > > >
> > > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com>
> > > > wrote:
> > > > > > > >
> > > > > > > > Hi David,
> > > > > > > >
> > > > > > > > Thanks for your patience for answering all these questions!
> > > > > > > >
> > > > > > > > One more question from me:
> > > > > > > > 1. When the selected consumer is going to be in client-side
> > > > assigning,
> > > > > > we
> > > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> > > > wondering if
> > > > > > we
> > > > > > > > could use a flag to indicate the assignment is required. The
> > > > rationale
> > > > > > is
> > > > > > > > when returning error, we lose a chance to catch up with the
> > latest
> > > > > > target
> > > > > > > > assignment or revoke partitions. Since we expected the record
> > > > fetching
> > > > > > is
> > > > > > > > not blocked by the rebalance/assignment computing, we might
> > want to
> > > > > > allow
> > > > > > > > the members to update the member assignment in each heartbeat.
> > > > > > > >
> > > > > > > > Luke
> > > > > > > >
> > > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > > > <ja...@confluent.io.invalid>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hey David,
> > > > > > > > >
> > > > > > > > > Thanks for the response.
> > > > > > > > >
> > > > > > > > > > Do you think that we could have a case where two members
> > with
> > > > the
> > > > > > > same
> > > > > > > > > static id are active?
> > > > > > > > >
> > > > > > > > > I was just thinking about the handoff when we spin up a new
> > node
> > > > > > > before the
> > > > > > > > > old one completes shutdown. It's probably ok to fence the old
> > > > member
> > > > > > > > > abruptly. I do recall we had a few issues with this handoff
> > in
> > > > the
> > > > > > > past. It
> > > > > > > > > was due to the fact that the replaced node might have been
> > the
> > > > > > leader.
> > > > > > > > > Since we've removed the leader notion, perhaps this kind of
> > issue
> > > > > > won't
> > > > > > > > > resurface.
> > > > > > > > >
> > > > > > > > > In regard to the RE2 library. I understand the intent to pass
> > > > this
> > > > > > > through
> > > > > > > > > as a string without any validation so that we don't need the
> > > > > > > dependence on
> > > > > > > > > the client. I do wonder if that will be restrictive in
> > practice.
> > > > For
> > > > > > > > > example, if a regex changes on the client side, would we
> > need to
> > > > > > > validate
> > > > > > > > > against the current assignment to see which partitions to
> > > > continue
> > > > > > > > > fetching? Or would we instead invalidate the complete current
> > > > > > > assignment?
> > > > > > > > > Perhaps this is a rare case and we can do the latter. There
> > > > could be
> > > > > > > more
> > > > > > > > > situations though.
> > > > > > > > >
> > > > > > > > > As an alternative, I wonder if we could define a minimalistic
> > > > regex
> > > > > > > syntax
> > > > > > > > > (a subset of RE2 say) that could map more easily to regex
> > support
> > > > > > that
> > > > > > > > > already exists in clients? For example, in this scenario, we
> > are
> > > > > > > matching
> > > > > > > > > topics, so we only need to be able to represent topic
> > characters
> > > > and
> > > > > > > not
> > > > > > > > > the whole smorgasbord of whitespace and unicode options.
> > Would it
> > > > > > make
> > > > > > > > > sense to have a reduced syntax which includes only standard
> > topic
> > > > > > > > > characters, wildcards, character classes, and repetitions?
> > > > > > > > >
> > > > > > > > > Best,
> > > > > > > > > Jason
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > > > wangguoz@gmail.com>
> > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hi David, thanks for all the detailed explanations. I think
> > > > they
> > > > > > all
> > > > > > > make
> > > > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > > > >
> > > > > > > > > > > I don't really see the benefits here because server side
> > > > > > assignors
> > > > > > > > > > don't have metadata at all. They only assign
> > topic-partitions.
> > > > They
> > > > > > > > > > are not supposed to generate metadata nor to receive
> > metadata
> > > > from
> > > > > > > the
> > > > > > > > > > members.
> > > > > > > > > >
> > > > > > > > > > Just to clarify I was asking about the `version` of the
> > > > assignor
> > > > > > > (i.e. up
> > > > > > > > > > to what version that the client would support), and I do
> > agree
> > > > we
> > > > > > > would
> > > > > > > > > not
> > > > > > > > > > need metadata. What I have in mind is that, for some
> > specific
> > > > > > > built-in
> > > > > > > > > > broker-assignors, e.g. rack-aware assignors, if it's
> > possible
> > > > that
> > > > > > > in a
> > > > > > > > > > newer version we would have a hierarchical rack ID string
> > > > format,
> > > > > > > like
> > > > > > > > > > "tier1-tier2" etc, but if some client has not upgraded
> > their
> > > > rack
> > > > > > ID
> > > > > > > > > > would still be in old format. In this case, the broker then
> > > > needs
> > > > > > to
> > > > > > > > > choose
> > > > > > > > > > the old versioned assignor. I'm probably making something
> > up
> > > > here
> > > > > > for
> > > > > > > > > rack
> > > > > > > > > > aware assignors, but I'm wondering if in general such an
> > > > > > > "auto-downgrade"
> > > > > > > > > > behavior would be needed still for broker-side assignor,
> > and
> > > > if yes
> > > > > > > would
> > > > > > > > > > "version" still be useful.
> > > > > > > > > >
> > > > > > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> > > > called
> > > > > > > once
> > > > > > > > > > when the member transitions to a new epoch. This one
> > contains
> > > > the
> > > > > > > full
> > > > > > > > > > metadata provided by the client side assignor. Then,
> > > > `onAssignment`
> > > > > > > > > > can be called max N times where N is the number of
> > partitions
> > > > > > pending
> > > > > > > > > > revocation by other members. Let me try to clarify this in
> > the
> > > > KIP.
> > > > > > > > > >
> > > > > > > > > > Okay, my understanding is that the calling ordering of
> > these
> > > > > > > callbacks
> > > > > > > > > > would be like the following:
> > > > > > > > > >
> > > > > > > > > > ----------------------------------------
> > > > > > > > > > onPartitionsRevoked();   // just once, since we do not
> > really
> > > > need
> > > > > > > > > > to revoke incrementally.
> > > > > > > > > >
> > > > > > > > > > onAssignment();    // the first call, with epoch
> > incremented
> > > > > > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > > > > > >
> > > > > > > > > > onAssignment();              // the first onAssignment
> > would
> > > > bump
> > > > > > up
> > > > > > > the
> > > > > > > > > > epoch, and the metadata reflected.
> > > > > > > > > > onPartitionsAssigned();   // each time we get an additional
> > > > > > > assignment,
> > > > > > > > > we
> > > > > > > > > > call onAssignment and then paired with an
> > onPartitionsAssigned
> > > > > > > > > > ...
> > > > > > > > > > onAssignment();
> > > > > > > > > > onPartitionsAssigned();   // on each of the onAssignment
> > > > calls, the
> > > > > > > > > encoded
> > > > > > > > > > metadata would not change, only the incrementally added
> > > > partitions
> > > > > > be
> > > > > > > > > > reflected
> > > > > > > > > >
> > > > > > > > > > Is that the case?
> > > > > > > > > >
> > > > > > > > > > I'm wondering if we would still call onAssignment just
> > once,
> > > > that
> > > > > > > encodes
> > > > > > > > > > all the assignment for this rebalance, including all the
> > > > partitions
> > > > > > > that
> > > > > > > > > > should be assigned to the member but not yet assigned since
> > > > they
> > > > > > > have not
> > > > > > > > > > been revoked by others. In that case the call ordering
> > would
> > > > be:
> > > > > > > > > >
> > > > > > > > > > ----------------------------------------
> > > > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > > > onAssignment();    // just once, with epoch incremented,
> > and
> > > > > > metadata
> > > > > > > > > > encoded changed, the "assignment" field also reflect the
> > final
> > > > > > target
> > > > > > > > > > assignment
> > > > > > > > > > onPartitionsAssigned();   // multiple times, which
> > represent
> > > > > > > > > incrementally
> > > > > > > > > > added partitions
> > > > > > > > > > ...
> > > > > > > > > > onPartitionsAssigned();
> > > > > > > > > >
> > > > > > > > > > The motivation from this is that, most users would only
> > > > implement
> > > > > > the
> > > > > > > > > > rebalance callback listeners and hence we'd definitely
> > need to
> > > > make
> > > > > > > sure
> > > > > > > > > > the semantics of that does not change much, and the time
> > > > > > > > > > `onPartitionsAssigned` indicate the time when the
> > partitions
> > > > are
> > > > > > > actually
> > > > > > > > > > assigned to it; while for assignors, the `onAssignment` is
> > > > used to
> > > > > > > > > indicate
> > > > > > > > > > what decision is made regarding for this member, i.e. when
> > the
> > > > > > > partitions
> > > > > > > > > > are decided to be given to it, but not necessarily meant
> > that
> > > > it
> > > > > > has
> > > > > > > been
> > > > > > > > > > given, since that time should be determined by the time of
> > > > > > > > > > `onPartitionsAssigned`. The benefits though, would be that
> > > > assignor
> > > > > > > > > > implementers would not need to reason which `onAssignment`
> > > > would be
> > > > > > > the
> > > > > > > > > > last one for this epoch.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Guozhang
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Guozhang
> > > > > > > > > >
> > > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > >
> > > > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hi Guozhang,
> > > > > > > > > > >
> > > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has
> > an
> > > > Error
> > > > > > > byte
> > > > > > > > > > > field,
> > > > > > > > > > > > while the broker-side "PartitionAssignor#Assignment"
> > does
> > > > not.
> > > > > > > And
> > > > > > > > > the
> > > > > > > > > > > > rationale seems to be assuming that we should always be
> > > > able to
> > > > > > > do
> > > > > > > > > the
> > > > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > > > Personally I
> > > > > > > > > > think
> > > > > > > > > > > > it's still potentially beneficial to add the Error
> > field
> > > > even
> > > > > > for
> > > > > > > > > > > > broker-side assignors, e.g. for some edge cases where
> > some
> > > > > > > subscribed
> > > > > > > > > > > > topics are not recognized with the current broker's
> > > > metadata.
> > > > > > > What do
> > > > > > > > > > you
> > > > > > > > > > > > think?
> > > > > > > > > > >
> > > > > > > > > > > Yeah, that seems reasonable. However, I wonder if it
> > would be
> > > > > > > better
> > > > > > > > > > > to use an exception on the server side. This is what we
> > > > usually
> > > > > > do
> > > > > > > for
> > > > > > > > > > > server side plugins. On the client side, we use a field
> > > > because
> > > > > > the
> > > > > > > > > > > errors are not defined in advance.
> > > > > > > > > > >
> > > > > > > > > > > Your comment also makes me think about what we should do
> > > > when the
> > > > > > > > > > > server side assignor fails. I suppose that we have to
> > keep
> > > > the
> > > > > > > current
> > > > > > > > > > > assignment until a new event occurs. For instance, in
> > your
> > > > > > example,
> > > > > > > > > > > the coordinator would have to trigger a rebalance when
> > > > > > unrecognized
> > > > > > > > > > > topics are available. This would be part of the metadata
> > > > > > > monitoring.
> > > > > > > > > > >
> > > > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > > > fields
> > > > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > > > GroupMember. I
> > > > > > > > > > > agree
> > > > > > > > > > > > that broker-side assignor would not need
> > reason/metadata
> > > > since
> > > > > > > they
> > > > > > > > > are
> > > > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > > > version?
> > > > > > > E.g.
> > > > > > > > > is
> > > > > > > > > > > it
> > > > > > > > > > > > possible that we evolve our broker-side built-in
> > assignor
> > > > but
> > > > > > > the old
> > > > > > > > > > > > versioned clients would not be able to work with the
> > new
> > > > > > > version, in
> > > > > > > > > > > which
> > > > > > > > > > > > case we need to let the broker being aware of this and
> > > > upgrade
> > > > > > > its
> > > > > > > > > > > behavior
> > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > >
> > > > > > > > > > > I don't really see the benefits here because server side
> > > > > > assignors
> > > > > > > > > > > don't have metadata at all. They only assign
> > > > topic-partitions.
> > > > > > They
> > > > > > > > > > > are not supposed to generate metadata nor to receive
> > metadata
> > > > > > from
> > > > > > > the
> > > > > > > > > > > members.
> > > > > > > > > > >
> > > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > > "GroupMember",
> > > > > > > > > instead
> > > > > > > > > > > of
> > > > > > > > > > > > including these three fields, what about just adding
> > the
> > > > > > > "Metadata"
> > > > > > > > > > field
> > > > > > > > > > > > class which has these three fields? Also, there are two
> > > > > > > "Metadata"
> > > > > > > > > > > > currently in the APIs, the first is a class that
> > encodes
> > > > > > > > > > > > reason/version/metadata, and the second is just the
> > encoded
> > > > > > > metadata
> > > > > > > > > > > bytes.
> > > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > > memberMetadata,
> > > > > > > > > which
> > > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> > naming the
> > > > > > > second
> > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > >
> > > > > > > > > > > That's a good point. Let me try to rationalize this
> > interface
> > > > > > > based on
> > > > > > > > > > > your suggestions.
> > > > > > > > > > >
> > > > > > > > > > > Best,
> > > > > > > > > > > David
> > > > > > > > > > >
> > > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > > > wangguoz@gmail.com>
> > > > > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > Hello David,
> > > > > > > > > > > >
> > > > > > > > > > > > Just had a few more nit questions about the public
> > APIs:
> > > > > > > > > > > >
> > > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has
> > an
> > > > Error
> > > > > > > byte
> > > > > > > > > > > field,
> > > > > > > > > > > > while the broker-side "PartitionAssignor#Assignment"
> > does
> > > > not.
> > > > > > > And
> > > > > > > > > the
> > > > > > > > > > > > rationale seems to be assuming that we should always be
> > > > able to
> > > > > > > do
> > > > > > > > > the
> > > > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > > > Personally I
> > > > > > > > > > think
> > > > > > > > > > > > it's still potentially beneficial to add the Error
> > field
> > > > even
> > > > > > for
> > > > > > > > > > > > broker-side assignors, e.g. for some edge cases where
> > some
> > > > > > > subscribed
> > > > > > > > > > > > topics are not recognized with the current broker's
> > > > metadata.
> > > > > > > What do
> > > > > > > > > > you
> > > > > > > > > > > > think?
> > > > > > > > > > > >
> > > > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > > > fields
> > > > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > > > GroupMember. I
> > > > > > > > > > > agree
> > > > > > > > > > > > that broker-side assignor would not need
> > reason/metadata
> > > > since
> > > > > > > they
> > > > > > > > > are
> > > > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > > > version?
> > > > > > > E.g.
> > > > > > > > > is
> > > > > > > > > > > it
> > > > > > > > > > > > possible that we evolve our broker-side built-in
> > assignor
> > > > but
> > > > > > > the old
> > > > > > > > > > > > versioned clients would not be able to work with the
> > new
> > > > > > > version, in
> > > > > > > > > > > which
> > > > > > > > > > > > case we need to let the broker being aware of this and
> > > > upgrade
> > > > > > > its
> > > > > > > > > > > behavior
> > > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > > >
> > > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > > "GroupMember",
> > > > > > > > > instead
> > > > > > > > > > > of
> > > > > > > > > > > > including these three fields, what about just adding
> > the
> > > > > > > "Metadata"
> > > > > > > > > > field
> > > > > > > > > > > > class which has these three fields? Also, there are two
> > > > > > > "Metadata"
> > > > > > > > > > > > currently in the APIs, the first is a class that
> > encodes
> > > > > > > > > > > > reason/version/metadata, and the second is just the
> > encoded
> > > > > > > metadata
> > > > > > > > > > > bytes.
> > > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > > memberMetadata,
> > > > > > > > > which
> > > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> > naming the
> > > > > > > second
> > > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > > >
> > > > > > > > > > > > Guozhang
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > > Hello David,
> > > > > > > > > > > > >
> > > > > > > > > > > > > Thanks for bringing this question up. I think the
> > main
> > > > > > > benefits as
> > > > > > > > > > you
> > > > > > > > > > > > > listed is 2) above if it stays; just to clarify, we
> > would
> > > > > > only
> > > > > > > be
> > > > > > > > > > able
> > > > > > > > > > > to
> > > > > > > > > > > > > save one round trip if the rebalance is still
> > triggered
> > > > by
> > > > > > the
> > > > > > > > > > broker;
> > > > > > > > > > > if
> > > > > > > > > > > > > the rebalance is triggered by the client then the
> > > > > > > num.round.trips
> > > > > > > > > are
> > > > > > > > > > > the
> > > > > > > > > > > > > same:
> > > > > > > > > > > > >
> > > > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > > > >
> > > > > > > > > > > > > T0: client decides to do a new assignment, suppose
> > it has
> > > > > > > already
> > > > > > > > > > sent
> > > > > > > > > > > a
> > > > > > > > > > > > > HB and hence has to wait for it to return first since
> > > > only
> > > > > > one
> > > > > > > > > > request
> > > > > > > > > > > /
> > > > > > > > > > > > > response can be inflight with the coordinator's
> > socket.
> > > > > > > > > > > > > T1: client receives the HB response, and then sends
> > the
> > > > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > request.
> > > > > > > > > > > > >
> > > > > > > > > > > > > In total, two round trips.
> > > > > > > > > > > > >
> > > > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > > > >
> > > > > > > > > > > > > T0: client decides to do a new assignment, suppose
> > it has
> > > > > > > already
> > > > > > > > > > sent
> > > > > > > > > > > a
> > > > > > > > > > > > > HB and hence has to wait for it to return first since
> > > > only
> > > > > > one
> > > > > > > > > > request
> > > > > > > > > > > /
> > > > > > > > > > > > > response can be inflight with the coordinator's
> > socket.
> > > > > > > > > > > > > T1: client receives the HB response, and then sends
> > the
> > > > new
> > > > > > HB
> > > > > > > > > > request
> > > > > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > > > > T2: the HB response with the optional member metadata
> > > > map is
> > > > > > > > > > returned.
> > > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > > request.
> > > > > > > > > > > > >
> > > > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > > > >
> > > > > > > > > > > > > -----------------------------
> > > > > > > > > > > > >
> > > > > > > > > > > > > So to complete the full picture here, we'd need to
> > modify
> > > > > > both
> > > > > > > HB
> > > > > > > > > > > request
> > > > > > > > > > > > > and response so that the client can also indicate a
> > new
> > > > > > > rebalance
> > > > > > > > > via
> > > > > > > > > > > the
> > > > > > > > > > > > > HB request as well, right?
> > > > > > > > > > > > >
> > > > > > > > > > > > > Assuming all above is true, I think it's okay to
> > merge
> > > > the
> > > > > > > > > > > > > GroupPrepareAssignment into HB given that we can
> > make the
> > > > > > > > > additional
> > > > > > > > > > > fields
> > > > > > > > > > > > > encoding the full member (subscription) metadata and
> > > > topic
> > > > > > > metadata
> > > > > > > > > > as
> > > > > > > > > > > > > optional fields.
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > > Guozhang
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > > wrote:
> > > > > > > > > > > > >
> > > > > > > > > > > > >> Hi all,
> > > > > > > > > > > > >>
> > > > > > > > > > > > >> During an offline conversation, someone asked why we
> > > > need
> > > > > > the
> > > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested
> > that we
> > > > > > could
> > > > > > > > > > instead
> > > > > > > > > > > > >> provide the group state in the heartbeat response.
> > This
> > > > has
> > > > > > a
> > > > > > > few
> > > > > > > > > > > > >> advantages: 1) it does not require using a special
> > error
> > > > > > code
> > > > > > > to
> > > > > > > > > > > > >> signal that a new assignment is required as the
> > signal
> > > > would
> > > > > > > be
> > > > > > > > > the
> > > > > > > > > > > > >> provided group state; 2) it removes one round trip
> > when
> > > > a
> > > > > > > client
> > > > > > > > > > side
> > > > > > > > > > > > >> assignor is used. The downside is that it makes the
> > > > > > heartbeat
> > > > > > > > > > > > >> response's definition quite large. I recall that I
> > went
> > > > with
> > > > > > > the
> > > > > > > > > > > > >> current approach due to this.
> > > > > > > > > > > > >>
> > > > > > > > > > > > >> Providing the group state in the heartbeat response
> > is
> > > > > > > appealing.
> > > > > > > > > > What
> > > > > > > > > > > > >> do you guys think?
> > > > > > > > > > > > >>
> > > > > > > > > > > > >> Best,
> > > > > > > > > > > > >> David
> > > > > > > > > > > > >>
> > > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > > > > djacot@confluent.io>
> > > > > > > > > > > wrote:
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > 1. I have added a reference to the relevant
> > chapter
> > > > > > instead
> > > > > > > of
> > > > > > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you are
> > referring
> > > > to
> > > > > > is
> > > > > > > > > about
> > > > > > > > > > > when
> > > > > > > > > > > > >> > a rebalance should be triggered for the
> > non-upgraded
> > > > > > members
> > > > > > > > > using
> > > > > > > > > > > the
> > > > > > > > > > > > >> > old protocol. The section mentions that a
> > rebalance
> > > > must
> > > > > > be
> > > > > > > > > > > triggered
> > > > > > > > > > > > >> > when a new assignment is installed. This implies
> > that
> > > > the
> > > > > > > group
> > > > > > > > > > > epoch
> > > > > > > > > > > > >> > was updated either by a native member or a
> > > > non-upgraded
> > > > > > > member.
> > > > > > > > > > For
> > > > > > > > > > > > >> > the latter, the JoinGroup request would be the
> > > > trigger. I
> > > > > > > have
> > > > > > > > > > > added a
> > > > > > > > > > > > >> > reference to the relevant chapter in the
> > "JoinGroup
> > > > > > > Handling"
> > > > > > > > > > > section
> > > > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > Thanks,
> > > > > > > > > > > > >> > David
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > > > > > wangguoz@gmail.com
> > > > > > > > > > >
> > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > Hello David,
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > Alright I think that's sufficient. Just to make
> > that
> > > > > > > clear in
> > > > > > > > > > the
> > > > > > > > > > > doc,
> > > > > > > > > > > > >> > > could we update:
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > 1) the heartbeat request handling section,
> > stating
> > > > when
> > > > > > > > > > > coordinator
> > > > > > > > > > > > >> will
> > > > > > > > > > > > >> > > trigger rebalance based on the HB's member
> > metadata
> > > > /
> > > > > > > reason?
> > > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to include
> > what
> > > > we
> > > > > > > > > described
> > > > > > > > > > > in
> > > > > > > > > > > > >> "Group
> > > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > Guozhang
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > >> > > wrote:
> > > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > I thought that the assignor will always be
> > > > consulted
> > > > > > > when
> > > > > > > > > the
> > > > > > > > > > > next
> > > > > > > > > > > > >> > > > heartbeat request is constructed. In other
> > words,
> > > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be called
> > for
> > > > every
> > > > > > > > > > heartbeat.
> > > > > > > > > > > > >> This
> > > > > > > > > > > > >> > > > gives the opportunity for the assignor to
> > enforce
> > > > a
> > > > > > > > > rebalance
> > > > > > > > > > by
> > > > > > > > > > > > >> > > > setting the reason to a non-zero value or by
> > > > changing
> > > > > > > the
> > > > > > > > > > > bytes. Do
> > > > > > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > > > > > concerned by
> > > > > > > > > > the
> > > > > > > > > > > > >> delay?
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > Best,
> > > > > > > > > > > > >> > > > David
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > > >> wrote:
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > >> > > > > In this case, a new assignment is triggered
> > by
> > > > the
> > > > > > > client
> > > > > > > > > > side
> > > > > > > > > > > > >> > > > > assignor. When constructing the HB, the
> > consumer
> > > > > > will
> > > > > > > > > always
> > > > > > > > > > > > >> consult
> > > > > > > > > > > > >> > > > > the client side assignor and propagate the
> > > > > > > information to
> > > > > > > > > > the
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > coordinator. In other words, we don't expect
> > > > users
> > > > > > to
> > > > > > > call
> > > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > As I looked at the current
> > PartitionAssignor's
> > > > > > > interface,
> > > > > > > > > we
> > > > > > > > > > > > >> actually do
> > > > > > > > > > > > >> > > > > not have a way yet to instruct how to
> > construct
> > > > the
> > > > > > > next
> > > > > > > > > HB
> > > > > > > > > > > > >> request, e.g.
> > > > > > > > > > > > >> > > > > when the assignor wants to enforce a new
> > > > rebalance
> > > > > > > with a
> > > > > > > > > > new
> > > > > > > > > > > > >> assignment,
> > > > > > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > > > > > PartitionAssignor
> > > > > > > > > > > to
> > > > > > > > > > > > >> indicate
> > > > > > > > > > > > >> > > > > the next HB telling broker about so. WDYT
> > about
> > > > > > adding
> > > > > > > > > such
> > > > > > > > > > an
> > > > > > > > > > > > >> API on the
> > > > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > I have updated the KIP to include your
> > > > feedback. I
> > > > > > > have
> > > > > > > > > > also
> > > > > > > > > > > > >> tried to
> > > > > > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David
> > Jacot <
> > > > > > > > > > > djacot@confluent.io
> > > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me start
> > by
> > > > > > > answering
> > > > > > > > > your
> > > > > > > > > > > > >> questions
> > > > > > > > > > > > >> > > > > > > inline and I will update the KIP next
> > week.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > > > benefits
> > > > > > > of
> > > > > > > > > the
> > > > > > > > > > > KIP
> > > > > > > > > > > > >> seem to
> > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > efficient
> > > > > > > support of
> > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > I would also add that the KIP removes
> > the
> > > > global
> > > > > > > sync
> > > > > > > > > > > barrier
> > > > > > > > > > > > >> in the
> > > > > > > > > > > > >> > > > > > > protocol which is essential to improve
> > group
> > > > > > > stability
> > > > > > > > > > and
> > > > > > > > > > > > >> > > > > > > scalability, and the KIP also
> > simplifies the
> > > > > > > client by
> > > > > > > > > > > moving
> > > > > > > > > > > > >> most of
> > > > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton.
> > Do we
> > > > plan
> > > > > > > to
> > > > > > > > > > > support
> > > > > > > > > > > > >> rolling
> > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > > consumers?
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Definitely. The group coordinator will
> > use
> > > > the
> > > > > > > > > assignor
> > > > > > > > > > > used
> > > > > > > > > > > > >> by a
> > > > > > > > > > > > >> > > > > > > majority of the members. This allows the
> > > > group
> > > > > > to
> > > > > > > move
> > > > > > > > > > > from
> > > > > > > > > > > > >> one
> > > > > > > > > > > > >> > > > > > > assignor to another by a roll. This is
> > > > explained
> > > > > > > in
> > > > > > > > > the
> > > > > > > > > > > > >> Assignor
> > > > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > > > whether
> > > > > > > it's
> > > > > > > > > > > > >> required in
> > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> > needs to
> > > > be
> > > > > > > filled?
> > > > > > > > > > For
> > > > > > > > > > > > >> example,
> > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> > needs to
> > > > be
> > > > > > > filled.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > The client is expected to set those
> > fields
> > > > in
> > > > > > > case of
> > > > > > > > > a
> > > > > > > > > > > > >> connection
> > > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields
> > have
> > > > > > > changed
> > > > > > > > > > since
> > > > > > > > > > > > >> the last
> > > > > > > > > > > > >> > > > > > > HB. The server populates those fields as
> > > > long as
> > > > > > > the
> > > > > > > > > > > member
> > > > > > > > > > > > >> is not
> > > > > > > > > > > > >> > > > > > > fully reconciled - the member should
> > > > acknowledge
> > > > > > > that
> > > > > > > > > it
> > > > > > > > > > > has
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > expected epoch and assignment. I will
> > > > clarify
> > > > > > > this in
> > > > > > > > > > the
> > > > > > > > > > > KIP.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol,
> > the
> > > > rack
> > > > > > > > > > affinity
> > > > > > > > > > > > >> between the
> > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > >> > > > > > > > and the broker is only considered
> > during
> > > > > > > fetching,
> > > > > > > > > but
> > > > > > > > > > > not
> > > > > > > > > > > > >> during
> > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes,
> > once
> > > > the
> > > > > > > > > > assignment
> > > > > > > > > > > is
> > > > > > > > > > > > >> made,
> > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> > because
> > > > no
> > > > > > > replicas
> > > > > > > > > > of
> > > > > > > > > > > > >> assigned
> > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > wondering
> > > > if we
> > > > > > > should
> > > > > > > > > > use
> > > > > > > > > > > > >> this
> > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > > > GroupMember.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > That's an interesting idea. I don't see
> > any
> > > > > > issue
> > > > > > > with
> > > > > > > > > > > adding
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > rack
> > > > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > > > useful to
> > > > > > > know
> > > > > > > > > how
> > > > > > > > > > > busy
> > > > > > > > > > > > >> a group
> > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the event
> > loop
> > > > > > model,
> > > > > > > it
> > > > > > > > > > seems
> > > > > > > > > > > > >> that we
> > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of
> > the
> > > > time
> > > > > > > the
> > > > > > > > > > event
> > > > > > > > > > > > >> loop is
> > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > That's a great idea. I will add it.
> > Thanks.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > coordinator
> > > > > > > failover
> > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > >> > > > > > > > does it need to trigger the check if
> > any
> > > > group
> > > > > > > with
> > > > > > > > > > the
> > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > >> > > > > > > > subscription now has a new matching
> > topic?
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Sure. When the new group coordinator
> > takes
> > > > over,
> > > > > > > it
> > > > > > > > > has
> > > > > > > > > > > to:
> > > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a client
> > side
> > > > > > > assignor
> > > > > > > > > is
> > > > > > > > > > > used.
> > > > > > > > > > > > >> We
> > > > > > > > > > > > >> > > > don't
> > > > > > > > > > > > >> > > > > > > store the information about the member
> > > > selected
> > > > > > > to run
> > > > > > > > > > the
> > > > > > > > > > > > >> assignment
> > > > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > > > > > wildcard
> > > > > > > > > > > > >> subscriptions, and
> > > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> > Could we
> > > > > > > document
> > > > > > > > > > what
> > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch
> > in
> > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > > > one in
> > > > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue
> > and
> > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Sure. I will clarify that but it is
> > always
> > > > the
> > > > > > > latest
> > > > > > > > > > > group
> > > > > > > > > > > > >> epoch.
> > > > > > > > > > > > >> > > > > > > When the group state is updated, the
> > group
> > > > epoch
> > > > > > > is
> > > > > > > > > > > bumped so
> > > > > > > > > > > > >> we use
> > > > > > > > > > > > >> > > > > > > that one for all the change records
> > related
> > > > to
> > > > > > the
> > > > > > > > > > update.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > > > that
> > > > > > the
> > > > > > > > > > > following
> > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > > possible
> > > > > > for
> > > > > > > a
> > > > > > > > > > member
> > > > > > > > > > > > >> not to
> > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > That's right. Here I meant that the
> > members
> > > > > > > provided
> > > > > > > > > by
> > > > > > > > > > > the
> > > > > > > > > > > > >> assignor
> > > > > > > > > > > > >> > > > > > > in the assignment must exist in the
> > group.
> > > > The
> > > > > > > > > assignor
> > > > > > > > > > > can
> > > > > > > > > > > > >> not make
> > > > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > > > member
> > > > > > > epoch
> > > > > > > > > > > equals to
> > > > > > > > > > > > >> 0":
> > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > > > would be
> > > > > > > used?
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > > > > > partitions and
> > > > > > > > > > > > >> rejoins when
> > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error.
> > In
> > > > this
> > > > > > > case,
> > > > > > > > > > the
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > > > coordinator will have removed the member
> > > > from
> > > > > > the
> > > > > > > > > group.
> > > > > > > > > > > The
> > > > > > > > > > > > >> member
> > > > > > > > > > > > >> > > > > > > can rejoin the group with the same
> > member
> > > > id but
> > > > > > > with
> > > > > > > > > 0
> > > > > > > > > > as
> > > > > > > > > > > > >> epoch. Let
> > > > > > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have
> > the
> > > > > > ability
> > > > > > > to
> > > > > > > > > > > trigger a
> > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > >> > > > > > > > by either providing a non-zero reason
> > or
> > > > by
> > > > > > > updating
> > > > > > > > > > the
> > > > > > > > > > > > >> assignor
> > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > > conflicting
> > > > > > > with
> > > > > > > > > the
> > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > In this case, a new assignment is
> > triggered
> > > > by
> > > > > > the
> > > > > > > > > > client
> > > > > > > > > > > side
> > > > > > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> > > > consumer
> > > > > > > will
> > > > > > > > > > > always
> > > > > > > > > > > > >> consult
> > > > > > > > > > > > >> > > > > > > the client side assignor and propagate
> > the
> > > > > > > information
> > > > > > > > > > to
> > > > > > > > > > > the
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > > > coordinator. In other words, we don't
> > expect
> > > > > > > users to
> > > > > > > > > > call
> > > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> > nice.
> > > > But
> > > > > > the
> > > > > > > > > > section
> > > > > > > > > > > > >> seems to
> > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> > epoch
> > > > 2, B
> > > > > > > > > > > immediately
> > > > > > > > > > > > >> gets into
> > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which
> > seems
> > > > > > > > > incorrect.
> > > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> > epoch
> > > > 3, C
> > > > > > > seems
> > > > > > > > > to
> > > > > > > > > > > get
> > > > > > > > > > > > >> into
> > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > > > still
> > > > > > > has A -
> > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do
> > we
> > > > > > support
> > > > > > > the
> > > > > > > > > > > upgrade
> > > > > > > > > > > > >> from
> > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > We will support upgrading from the
> > consumer
> > > > > > > protocol
> > > > > > > > > > > version
> > > > > > > > > > > > >> 3,
> > > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > > > > > implemented
> > > > > > > yet
> > > > > > > > > so
> > > > > > > > > > > the
> > > > > > > > > > > > >> earliest
> > > > > > > > > > > > >> > > > > > > version is unknown at the moment. This
> > is
> > > > > > > explained in
> > > > > > > > > > the
> > > > > > > > > > > > >> migration
> > > > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I
> > will
> > > > > > > update the
> > > > > > > > > > KIP
> > > > > > > > > > > > >> based on
> > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > > > >> > > > > > > David
> > > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > > > benefits
> > > > > > > of
> > > > > > > > > the
> > > > > > > > > > > KIP
> > > > > > > > > > > > >> seem to
> > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> > efficient
> > > > > > > support of
> > > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton.
> > Do we
> > > > plan
> > > > > > > to
> > > > > > > > > > > support
> > > > > > > > > > > > >> rolling
> > > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > > consumers?
> > > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > > > whether
> > > > > > > it's
> > > > > > > > > > > > >> required in
> > > > > > > > > > > > >> > > > every
> > > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> > needs to
> > > > be
> > > > > > > filled?
> > > > > > > > > > For
> > > > > > > > > > > > >> example,
> > > > > > > > > > > > >> > > > it's
> > > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> > needs to
> > > > be
> > > > > > > filled.
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol,
> > the
> > > > rack
> > > > > > > > > > affinity
> > > > > > > > > > > > >> between the
> > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > >> > > > > > > > and the broker is only considered
> > during
> > > > > > > fetching,
> > > > > > > > > but
> > > > > > > > > > > not
> > > > > > > > > > > > >> during
> > > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes,
> > once
> > > > the
> > > > > > > > > > assignment
> > > > > > > > > > > is
> > > > > > > > > > > > >> made,
> > > > > > > > > > > > >> > > > there
> > > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> > because
> > > > no
> > > > > > > replicas
> > > > > > > > > > of
> > > > > > > > > > > > >> assigned
> > > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > > >> > > > > > > > are close to the member. I am
> > wondering
> > > > if we
> > > > > > > should
> > > > > > > > > > use
> > > > > > > > > > > > >> this
> > > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > > > GroupMember.
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > > > useful to
> > > > > > > know
> > > > > > > > > how
> > > > > > > > > > > busy
> > > > > > > > > > > > >> a group
> > > > > > > > > > > > >> > > > > > > > coordinator is. By moving the event
> > loop
> > > > > > model,
> > > > > > > it
> > > > > > > > > > seems
> > > > > > > > > > > > >> that we
> > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of
> > the
> > > > time
> > > > > > > the
> > > > > > > > > > event
> > > > > > > > > > > > >> loop is
> > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> > coordinator
> > > > > > > failover
> > > > > > > > > > > > >> handling? For
> > > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > > >> > > > > > > > does it need to trigger the check if
> > any
> > > > group
> > > > > > > with
> > > > > > > > > > the
> > > > > > > > > > > > >> wildcard
> > > > > > > > > > > > >> > > > > > > > subscription now has a new matching
> > topic?
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> > Could we
> > > > > > > document
> > > > > > > > > > what
> > > > > > > > > > > > >> the epoch
> > > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch
> > in
> > > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > > > one in
> > > > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue
> > and
> > > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > > > that
> > > > > > the
> > > > > > > > > > > following
> > > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > > possible
> > > > > > for
> > > > > > > a
> > > > > > > > > > member
> > > > > > > > > > > > >> not to
> > > > > > > > > > > > >> > > > get any
> > > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > > > member
> > > > > > > epoch
> > > > > > > > > > > equals to
> > > > > > > > > > > > >> 0":
> > > > > > > > > > > > >> > > > When
> > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > > > would be
> > > > > > > used?
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have
> > the
> > > > > > ability
> > > > > > > to
> > > > > > > > > > > trigger a
> > > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > > >> > > > > > > > by either providing a non-zero reason
> > or
> > > > by
> > > > > > > updating
> > > > > > > > > > the
> > > > > > > > > > > > >> assignor
> > > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > > conflicting
> > > > > > > with
> > > > > > > > > the
> > > > > > > > > > > > >> deprecation
> > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> > nice.
> > > > But
> > > > > > the
> > > > > > > > > > section
> > > > > > > > > > > > >> seems to
> > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> > epoch
> > > > 2, B
> > > > > > > > > > > immediately
> > > > > > > > > > > > >> gets into
> > > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which
> > seems
> > > > > > > > > incorrect.
> > > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> > epoch
> > > > 3, C
> > > > > > > seems
> > > > > > > > > to
> > > > > > > > > > > get
> > > > > > > > > > > > >> into
> > > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > > > still
> > > > > > > has A -
> > > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do
> > we
> > > > > > support
> > > > > > > the
> > > > > > > > > > > upgrade
> > > > > > > > > > > > >> from
> > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David
> > > > Jacot
> > > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > > > re-implement the
> > > > > > > > > > > coordinator
> > > > > > > > > > > > >> in
> > > > > > > > > > > > >> > > > Java. I
> > > > > > > > > > > > >> > > > > > > > > discussed this offline with a few
> > folks
> > > > and
> > > > > > > folks
> > > > > > > > > > are
> > > > > > > > > > > > >> concerned
> > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > >> > > > > > > > > we could introduce many regressions
> > in
> > > > the
> > > > > > old
> > > > > > > > > > > protocol
> > > > > > > > > > > > >> if we do
> > > > > > > > > > > > >> > > > so.
> > > > > > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > > > > > statement
> > > > > > > > > from
> > > > > > > > > > > the
> > > > > > > > > > > > >> KIP. It
> > > > > > > > > > > > >> > > > is an
> > > > > > > > > > > > >> > > > > > > > > implementation detail after all so
> > it
> > > > does
> > > > > > not
> > > > > > > > > have
> > > > > > > > > > > to be
> > > > > > > > > > > > >> > > > decided at
> > > > > > > > > > > > >> > > > > > > > > this stage. We will likely start by
> > > > trying
> > > > > > to
> > > > > > > > > > > refactor the
> > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM
> > David
> > > > Jacot
> > > > > > <
> > > > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine
> > are:
> > > > > > > "Empty,
> > > > > > > > > > > assigning,
> > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> > Group
> > > > States
> > > > > > > > > > section,
> > > > > > > > > > > > >> right?
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > This sentence does not refer to
> > those
> > > > > > group
> > > > > > > > > states
> > > > > > > > > > > but
> > > > > > > > > > > > >> rather
> > > > > > > > > > > > >> > > > to a
> > > > > > > > > > > > >> > > > > > > > > > state machine replication (SMR).
> > This
> > > > > > > refers to
> > > > > > > > > > the
> > > > > > > > > > > > >> entire
> > > > > > > > > > > > >> > > > state of
> > > > > > > > > > > > >> > > > > > > > > > group coordinator which is
> > replicated
> > > > via
> > > > > > > the
> > > > > > > > > log
> > > > > > > > > > > > >> layer. I will
> > > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> > state
> > > > > > machine
> > > > > > > is
> > > > > > > > > > > modelled
> > > > > > > > > > > > >> as an
> > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > The idea is to follow a model
> > similar
> > > > to
> > > > > > > the new
> > > > > > > > > > > quorum
> > > > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > > > >> > > > > > > > > > will have N threads to process
> > events.
> > > > > > Each
> > > > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > > > >> > > > > > > > > > partition is assigned to a unique
> > > > thread
> > > > > > > and all
> > > > > > > > > > the
> > > > > > > > > > > > >> events
> > > > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> > > > processed
> > > > > > by
> > > > > > > this
> > > > > > > > > > > > >> thread. This
> > > > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > > > >> > > > > > > > > > concurrency and will enable us to
> > do
> > > > > > > simulation
> > > > > > > > > > > testing
> > > > > > > > > > > > >> for the
> > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> > machine
> > > > per
> > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > consumer
> > > > group"
> > > > > > > owned
> > > > > > > > > > by
> > > > > > > > > > > a
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > coordinator
> > > > owns 2
> > > > > > > > > > consumer
> > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > > > have 1
> > > > > > > state
> > > > > > > > > > > machine
> > > > > > > > > > > > >> for it,
> > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> > > > there, I
> > > > > > > > > think.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > "*group.coordinator.threads"
> > > > > > > > > *is
> > > > > > > > > > > the
> > > > > > > > > > > > >> number
> > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > >> > > > > > > > > > > to run the state machines. But
> > I'm
> > > > > > > wondering
> > > > > > > > > if
> > > > > > > > > > > the
> > > > > > > > > > > > >> purpose
> > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> > each
> > > > > > consumer
> > > > > > > > > group
> > > > > > > > > > > (or
> > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > > computation,
> > > > > > > why
> > > > > > > > > > > should we
> > > > > > > > > > > > >> need
> > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an
> > > > ability to
> > > > > > > shard
> > > > > > > > > > the
> > > > > > > > > > > > >> > > > processing as
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> > timeout,
> > > > why
> > > > > > > does
> > > > > > > > > the
> > > > > > > > > > > > >> default
> > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > > max(60s)? I
> > > > > > > > > thought
> > > > > > > > > > > the
> > > > > > > > > > > > >> min/max
> > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > > > bound
> > > > > > of
> > > > > > > it,
> > > > > > > > > > no?
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.session.timeout.ms
> > > > int
> > > > > > > 30s The
> > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > > > group
> > > > > > > > > protocol.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > group.consumer.min.session.timeout.ms
> > > > > > > int 45s
> > > > > > > > > > The
> > > > > > > > > > > > >> minimum
> > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > group.consumer.max.session.timeout.ms
> > > > > > > int 60s
> > > > > > > > > > The
> > > > > > > > > > > > >> maximum
> > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The
> > default
> > > > > > > session
> > > > > > > > > > > timeout
> > > > > > > > > > > > >> should
> > > > > > > > > > > > >> > > > be 45s
> > > > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > > assignor
> > > > > > are
> > > > > > > > > > [range,
> > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > assignor.
> > > > I'd
> > > > > > > like to
> > > > > > > > > > > know
> > > > > > > > > > > > >> why not
> > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> > choose
> > > > > > uniform
> > > > > > > > > > assignor
> > > > > > > > > > > > >> (former
> > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > >> > > > > > > > > > > for better evenly distribution.
> > Any
> > > > > > other
> > > > > > > > > reason
> > > > > > > > > > > we
> > > > > > > > > > > > >> choose
> > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List
> > range,
> > > > > > > uniform
> > > > > > > > > The
> > > > > > > > > > > > >> server side
> > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > The order on the server side has
> > no
> > > > > > > influence
> > > > > > > > > > > because
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > >> > > > > > must
> > > > > > > > > > > > >> > > > > > > > > > chose the selector that he wants
> > to
> > > > use.
> > > > > > > There
> > > > > > > > > is
> > > > > > > > > > no
> > > > > > > > > > > > >> default
> > > > > > > > > > > > >> > > > in the
> > > > > > > > > > > > >> > > > > > > > > > current proposal. If the assignor
> > is
> > > > not
> > > > > > > > > specified
> > > > > > > > > > > by
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > client,
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > request is rejected. The default
> > > > client
> > > > > > > value
> > > > > > > > > for
> > > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is
> > `uniform`
> > > > > > though.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > Thanks for your very good
> > comments,
> > > > Luke.
> > > > > > I
> > > > > > > hope
> > > > > > > > > > > that my
> > > > > > > > > > > > >> > > > answers
> > > > > > > > > > > > >> > > > > > help
> > > > > > > > > > > > >> > > > > > > > > > to clarify things. I will update
> > the
> > > > KIP
> > > > > > as
> > > > > > > well
> > > > > > > > > > > based
> > > > > > > > > > > > >> on your
> > > > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM
> > Luke
> > > > Chen
> > > > > > <
> > > > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator
> > section, you
> > > > > > > > > mentioned:
> > > > > > > > > > > > >> > > > > > > > > > > > The new group coordinator will
> > > > have a
> > > > > > > state
> > > > > > > > > > > machine
> > > > > > > > > > > > >> per
> > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions,
> > > > where
> > > > > > > each
> > > > > > > > > > state
> > > > > > > > > > > > >> machine is
> > > > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > > > >> > > > > > > > > > > event loop. Those state machines
> > > > will be
> > > > > > > > > > executed
> > > > > > > > > > > in
> > > > > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads*
> > threads.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine
> > are:
> > > > > > > "Empty,
> > > > > > > > > > > assigning,
> > > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> > Group
> > > > States
> > > > > > > > > > section,
> > > > > > > > > > > > >> right?
> > > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> > state
> > > > > > machine
> > > > > > > is
> > > > > > > > > > > modelled
> > > > > > > > > > > > >> as an
> > > > > > > > > > > > >> > > > event
> > > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> > machine
> > > > per
> > > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> > consumer
> > > > group"
> > > > > > > owned
> > > > > > > > > > by
> > > > > > > > > > > a
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > > >> > > > > > > > > > > example, if one group
> > coordinator
> > > > owns 2
> > > > > > > > > > consumer
> > > > > > > > > > > > >> groups, and
> > > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > > > have 1
> > > > > > > state
> > > > > > > > > > > machine
> > > > > > > > > > > > >> for it,
> > > > > > > > > > > > >> > > > or
> > > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > > "*group.coordinator.threads"
> > > > > > > > > *is
> > > > > > > > > > > the
> > > > > > > > > > > > >> number
> > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > > >> > > > > > > > > > > to run the state machines. But
> > I'm
> > > > > > > wondering
> > > > > > > > > if
> > > > > > > > > > > the
> > > > > > > > > > > > >> purpose
> > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> > each
> > > > > > consumer
> > > > > > > > > group
> > > > > > > > > > > (or
> > > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > > computation,
> > > > > > > why
> > > > > > > > > > > should we
> > > > > > > > > > > > >> need
> > > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 2. For the default value in the
> > new
> > > > > > > configs:
> > > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> > timeout,
> > > > why
> > > > > > > does
> > > > > > > > > the
> > > > > > > > > > > > >> default
> > > > > > > > > > > > >> > > > session
> > > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > > max(60s)? I
> > > > > > > > > thought
> > > > > > > > > > > the
> > > > > > > > > > > > >> min/max
> > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > > > bound
> > > > > > of
> > > > > > > it,
> > > > > > > > > > no?
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.session.timeout.ms
> > > > int
> > > > > > > 30s The
> > > > > > > > > > > > >> timeout to
> > > > > > > > > > > > >> > > > detect
> > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > > > group
> > > > > > > > > protocol.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > group.consumer.min.session.timeout.ms
> > > > > > > int 45s
> > > > > > > > > > The
> > > > > > > > > > > > >> minimum
> > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > group.consumer.max.session.timeout.ms
> > > > > > > int 60s
> > > > > > > > > > The
> > > > > > > > > > > > >> maximum
> > > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > > assignor
> > > > > > are
> > > > > > > > > > [range,
> > > > > > > > > > > > >> uniform],
> > > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> > assignor.
> > > > I'd
> > > > > > > like to
> > > > > > > > > > > know
> > > > > > > > > > > > >> why not
> > > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> > choose
> > > > > > uniform
> > > > > > > > > > assignor
> > > > > > > > > > > > >> (former
> > > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > > >> > > > > > > > > > > for better evenly distribution.
> > Any
> > > > > > other
> > > > > > > > > reason
> > > > > > > > > > > we
> > > > > > > > > > > > >> choose
> > > > > > > > > > > > >> > > > range
> > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List
> > range,
> > > > > > > uniform
> > > > > > > > > The
> > > > > > > > > > > > >> server side
> > > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM
> > Luke
> > > > > > Chen
> > > > > > > <
> > > > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts about
> > Kafka
> > > > > > Connect
> > > > > > > > > > > > >> integrating with
> > > > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > > > >> > > > > > > > > > > > think we should have a
> > separate
> > > > > > > discussion
> > > > > > > > > > > thread
> > > > > > > > > > > > >> for the
> > > > > > > > > > > > >> > > > Kafka
> > > > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect
> > > > With
> > > > > > New
> > > > > > > > > > Consumer
> > > > > > > > > > > > >> Rebalance
> > > > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > > > >> > > > > > > > > > > > and let this discussion thread
> > > > focus
> > > > > > on
> > > > > > > > > > consumer
> > > > > > > > > > > > >> rebalance
> > > > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >>
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31
> > PM
> > > > Sagar
> > > > > > <
> > > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for
> > the
> > > > > > > feedback.
> > > > > > > > > > > Looks
> > > > > > > > > > > > >> like
> > > > > > > > > > > > >> > > > there's
> > > > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs for
> > Connect.
> > > > I
> > > > > > > would
> > > > > > > > > > > revisit
> > > > > > > > > > > > >> the doc
> > > > > > > > > > > > >> > > > and
> > > > > > > > > > > > >> > > > > > see
> > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11
> > PM
> > > > David
> > > > > > > Jacot
> > > > > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback
> > and the
> > > > > > > document.
> > > > > > > > > > > That's
> > > > > > > > > > > > >> really
> > > > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me
> > that
> > > > both
> > > > > > > Connect
> > > > > > > > > > > and the
> > > > > > > > > > > > >> > > > Consumer
> > > > > > > > > > > > >> > > > > > could
> > > > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > > > >> > > > > > > > > > > >> > the same underlying
> > "engine".
> > > > The
> > > > > > > main
> > > > > > > > > > > > >> difference is
> > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> > > > members
> > > > > > > > > whereas
> > > > > > > > > > > > >> Connect
> > > > > > > > > > > > >> > > > assigns
> > > > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to
> > move
> > > > > > > forward:
> > > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new
> > proposed
> > > > APIs
> > > > > > to
> > > > > > > > > > support
> > > > > > > > > > > > >> different
> > > > > > > > > > > > >> > > > > > resource
> > > > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks,
> > > > etc.); or
> > > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated
> > APIs
> > > > for
> > > > > > > Connect.
> > > > > > > > > > The
> > > > > > > > > > > > >> dedicated
> > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > > > > > different on
> > > > > > > > > > the
> > > > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> > > > engine
> > > > > > > on the
> > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > >> > > > side.
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean towards
> > 2)
> > > > > > because
> > > > > > > I am
> > > > > > > > > > > not a
> > > > > > > > > > > > >> fan of
> > > > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> > > > purposes.
> > > > > > > That
> > > > > > > > > > being
> > > > > > > > > > > > >> said, I am
> > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an
> > elegant
> > > > way to
> > > > > > > do
> > > > > > > > > it.
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > I think that we can
> > continue to
> > > > > > > discuss
> > > > > > > > > it
> > > > > > > > > > > here
> > > > > > > > > > > > >> for now
> > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > >> > > > > > order
> > > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> > > > compatible
> > > > > > > with
> > > > > > > > > > what
> > > > > > > > > > > we
> > > > > > > > > > > > >> will do
> > > > > > > > > > > > >> > > > for
> > > > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at
> > 2:41 PM
> > > > > > David
> > > > > > > > > Jacot
> > > > > > > > > > <
> > > > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation.
> > I
> > > > will
> > > > > > go
> > > > > > > > > > through
> > > > > > > > > > > and
> > > > > > > > > > > > >> address
> > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming days.
> > Thanks
> > > > for
> > > > > > your
> > > > > > > > > > > feedback.
> > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at
> > 10:05
> > > > PM
> > > > > > > Gregory
> > > > > > > > > > > Harris
> > > > > > > > > > > > >> <
> > > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP,
> > it's
> > > > > > > wonderful to
> > > > > > > > > > see
> > > > > > > > > > > > >> > > > cooperative
> > > > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> > > > Alternatives'
> > > > > > > > > section
> > > > > > > > > > > > >> describes how
> > > > > > > > > > > > >> > > > > > member
> > > > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step with
> > the
> > > > group
> > > > > > > epoch
> > > > > > > > > > and
> > > > > > > > > > > > >> assignment
> > > > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good
> > idea
> > > > for
> > > > > > the
> > > > > > > > > > reasons
> > > > > > > > > > > > >> described
> > > > > > > > > > > > >> > > > in
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is
> > incrementally
> > > > > > > assigning
> > > > > > > > > > > > >> partitions to a
> > > > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each
> > incremental
> > > > > > > > > assignment
> > > > > > > > > > > use?
> > > > > > > > > > > > >> Are
> > > > > > > > > > > > >> > > > member
> > > > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch
> > can
> > > > > > > correspond
> > > > > > > > > to
> > > > > > > > > > > > >> multiple
> > > > > > > > > > > > >> > > > > > different
> > > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's
> > > > 'Reason'
> > > > > > > field
> > > > > > > > > > > opaque
> > > > > > > > > > > > >> to the
> > > > > > > > > > > > >> > > > group
> > > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> > > > client-side
> > > > > > > > > assignor
> > > > > > > > > > > > >> > > > implementations
> > > > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how
> > is
> > > > its
> > > > > > > common
> > > > > > > > > > > meaning
> > > > > > > > > > > > >> agreed
> > > > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct
> > > > Reason
> > > > > > > field
> > > > > > > > > over
> > > > > > > > > > > > >> including
> > > > > > > > > > > > >> > > > such
> > > > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> > > > included in
> > > > > > > the
> > > > > > > > > > KIP:
> > > > > > > > > > > > >> "Thanks to
> > > > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > > > > > entirely
> > > > > > > > > driven
> > > > > > > > > > > by
> > > > > > > > > > > > >> the group
> > > > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > > > > > responsible for
> > > > > > > > > > > > >> maintaining
> > > > > > > > > > > > >> > > > any
> > > > > > > > > > > > >> > > > > > state
> > > > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions."
> > Does
> > > > this
> > > > > > > mean
> > > > > > > > > > > that the
> > > > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > > > > > non-Metadata
> > > > > > > > > > state
> > > > > > > > > > > > >> (such as
> > > > > > > > > > > > >> > > > > > partition
> > > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics,
> > config
> > > > > > > topics,
> > > > > > > > > > etc),
> > > > > > > > > > > or
> > > > > > > > > > > > >> that
> > > > > > > > > > > > >> > > > > > additional
> > > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be
> > used?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > > > > > separate
> > > > > > > > > > classes
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > > > >> > > > > >
> > > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > > > >> > > >
> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly.
> > Is
> > > > it
> > > > > > > possible
> > > > > > > > > > for
> > > > > > > > > > > > >> these two
> > > > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would
> > serve
> > > > to
> > > > > > > promote
> > > > > > > > > > > feature
> > > > > > > > > > > > >> parity
> > > > > > > > > > > > >> > > > of
> > > > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors,
> > and
> > > > > > would
> > > > > > > also
> > > > > > > > > > > > >> facilitate
> > > > > > > > > > > > >> > > > > > operational
> > > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > > > > > example,
> > > > > > > if a
> > > > > > > > > > > > >> server-side
> > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a
> > patch,
> > > > > > > deploying
> > > > > > > > > > the
> > > > > > > > > > > > >> patched
> > > > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching one
> > consumer
> > > > > > group
> > > > > > > to a
> > > > > > > > > > > > >> client-side
> > > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> > > > patching
> > > > > > all
> > > > > > > of
> > > > > > > > > the
> > > > > > > > > > > > >> brokers.
> > > > > > > > > > > > >> > > > With
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct
> > APIs, a
> > > > > > > non-trivial
> > > > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the
> > two
> > > > APIs
> > > > > > > have
> > > > > > > > > > > diverged
> > > > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > > > reimplementation
> > > > > > > > > would
> > > > > > > > > > > not be
> > > > > > > > > > > > >> > > > possible.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at
> > > > 8:39 AM
> > > > > > > Sagar
> > > > > > > > > <
> > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a
> > confluence
> > > > page
> > > > > > to
> > > > > > > > > > discuss
> > > > > > > > > > > how
> > > > > > > > > > > > >> Connect
> > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> > > > rebalance
> > > > > > > > > protocol.
> > > > > > > > > > > > >> Here's the
> > > > > > > > > > > > >> > > > page:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >>
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty
> > longish
> > > > and
> > > > > > I
> > > > > > > have
> > > > > > > > > > > tried
> > > > > > > > > > > > >> to keep
> > > > > > > > > > > > >> > > > a
> > > > > > > > > > > > >> > > > > > format
> > > > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know
> > > > what you
> > > > > > > > > think.
> > > > > > > > > > > Also,
> > > > > > > > > > > > >> do you
> > > > > > > > > > > > >> > > > > > think this
> > > > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > > > > > discussion
> > > > > > > > > thread
> > > > > > > > > > > or
> > > > > > > > > > > > >> is this
> > > > > > > > > > > > >> > > > one
> > > > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022
> > at
> > > > 7:37
> > > > > > AM
> > > > > > > > > > Sagar <
> > > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much
> > for
> > > > the
> > > > > > > doc on
> > > > > > > > > > > Kafka
> > > > > > > > > > > > >> Streams.
> > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come
> > up
> > > > with
> > > > > > > such a
> > > > > > > > > > > > >> document.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26,
> > 2022 at
> > > > > > 4:47
> > > > > > > AM
> > > > > > > > > > > Guozhang
> > > > > > > > > > > > >> Wang <
> > > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great
> > if
> > > > you
> > > > > > > could
> > > > > > > > > > come
> > > > > > > > > > > > >> back with
> > > > > > > > > > > > >> > > > some
> > > > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the
> > Connect
> > > > side
> > > > > > > > > > > integration
> > > > > > > > > > > > >> with
> > > > > > > > > > > > >> > > > the new
> > > > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides
> > leveraging on
> > > > the
> > > > > > > new
> > > > > > > > > > > "protocol
> > > > > > > > > > > > >> type"
> > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > >> > > > > > did not
> > > > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > > > > > implementations.
> > > > > > > > > > > For
> > > > > > > > > > > > >> Streams
> > > > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > >
> > > > > > > > > > > > >>
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your
> > > > analysis
> > > > > > > on
> > > > > > > > > > > Connect.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19,
> > 2022
> > > > at
> > > > > > > 10:48 PM
> > > > > > > > > > > Sagar <
> > > > > > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for
> > your
> > > > > > > response.
> > > > > > > > > The
> > > > > > > > > > > > >> reason I
> > > > > > > > > > > > >> > > > thought
> > > > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme
> > is
> > > > that
> > > > > > > even
> > > > > > > > > > > today the
> > > > > > > > > > > > >> > > > connect
> > > > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > > > > > AbstractCoordinator to
> > > > > > > > > > > > >> empower
> > > > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > tasks/connectors.
> > > > The
> > > > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> > > > metadata()
> > > > > > > method
> > > > > > > > > > by
> > > > > > > > > > > > >> plumbing
> > > > > > > > > > > > >> > > > into
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > JoinGroupRequestProtocol.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the
> > changes
> > > > to
> > > > > > > support
> > > > > > > > > > > > >> connect would
> > > > > > > > > > > > >> > > > be
> > > > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in
> > > > streams
> > > > > > > mainly
> > > > > > > > > > > because
> > > > > > > > > > > > >> of the
> > > > > > > > > > > > >> > > > > > Client
> > > > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both.
> > At an
> > > > > > > > > > implementation
> > > > > > > > > > > > >> level, we
> > > > > > > > > > > > >> > > > > > might
> > > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get
> > onto
> > > > this
> > > > > > > new
> > > > > > > > > > > > >> assignment
> > > > > > > > > > > > >> > > > protocol
> > > > > > > > > > > > >> > > > > > like
> > > > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > > > > request/response and
> > > > > > > > > > > > >> SyncGroup and
> > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > ConsumerGroupHeartbeat
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on
> > > > similar
> > > > > > > lines
> > > > > > > > > > to
> > > > > > > > > > > > >> streams (or
> > > > > > > > > > > > >> > > > > > there
> > > > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> > > > perform a
> > > > > > > > > detailed
> > > > > > > > > > > > >> analysis of
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > same
> > > > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate
> > discussion
> > > > > > > thread for
> > > > > > > > > > > that
> > > > > > > > > > > > >> as that
> > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me
> > know
> > > > if
> > > > > > > that
> > > > > > > > > > sounds
> > > > > > > > > > > > >> good to
> > > > > > > > > > > > >> > > > you.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15,
> > > > 2022 at
> > > > > > > 5:47
> > > > > > > > > PM
> > > > > > > > > > > David
> > > > > > > > > > > > >> Jacot
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for
> > your
> > > > > > > comments.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That
> > > > refers to
> > > > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea
> > is to
> > > > > > > > > transition C
> > > > > > > > > > > from
> > > > > > > > > > > > >> his
> > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target
> > assignment
> > > > when
> > > > > > > he
> > > > > > > > > can
> > > > > > > > > > > move
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > > > >> > > > > > When
> > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > assignment
> > > > is
> > > > > > > updated
> > > > > > > > > > and
> > > > > > > > > > > > >> persisted
> > > > > > > > > > > > >> > > > > > with all
> > > > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions
> > even if
> > > > > > they
> > > > > > > are
> > > > > > > > > > not
> > > > > > > > > > > all
> > > > > > > > > > > > >> revoked
> > > > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> > assignment
> > > > > > > becomes
> > > > > > > > > the
> > > > > > > > > > > target
> > > > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization
> > to
> > > > avoid
> > > > > > > having
> > > > > > > > > > to
> > > > > > > > > > > > >> write all
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are
> > > > based on
> > > > > > > the
> > > > > > > > > > > persisted
> > > > > > > > > > > > >> state
> > > > > > > > > > > > >> > > > so I
> > > > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion.
> > Let me
> > > > see
> > > > > > > if I
> > > > > > > > > can
> > > > > > > > > > > > >> improve
> > > > > > > > > > > > >> > > > this in
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding
> > > > Connect,
> > > > > > it
> > > > > > > > > could
> > > > > > > > > > > > >> reuse the
> > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it
> > > > fits in
> > > > > > > the
> > > > > > > > > > > > >> protocol. The
> > > > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > topicid-partitions +
> > > > > > > > > metadata,
> > > > > > > > > > > could
> > > > > > > > > > > > >> > > > Connect
> > > > > > > > > > > > >> > > > > > fit
> > > > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul
> > 15,
> > > > 2022
> > > > > > at
> > > > > > > 1:55
> > > > > > > > > > PM
> > > > > > > > > > > > >> Sagar <
> > > > > > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for
> > the
> > > > KIP.
> > > > > > I
> > > > > > > just
> > > > > > > > > > had
> > > > > > > > > > > > >> minor
> > > > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the
> > > > Assignment
> > > > > > > Error
> > > > > > > > > > > > >> section in
> > > > > > > > > > > > >> > > > Client
> > > > > > > > > > > > >> > > > > > Side
> > > > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you
> > mentioned
> > > > => `In
> > > > > > > this
> > > > > > > > > > > case,
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > client
> > > > > > > > > > > > >> > > > > > side
> > > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the
> > > > group
> > > > > > > > > > > coordinator`.
> > > > > > > > > > > > >> In this
> > > > > > > > > > > > >> > > > > > case are
> > > > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> > > > returning
> > > > > > an
> > > > > > > > > > > > >> AssignmentError
> > > > > > > > > > > > >> > > > that's
> > > > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do
> > you
> > > > think
> > > > > > > it
> > > > > > > > > > would
> > > > > > > > > > > > >> make sense
> > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the
> > Case
> > > > > > Studies
> > > > > > > > > > > section, I
> > > > > > > > > > > > >> have a
> > > > > > > > > > > > >> > > > > > slight
> > > > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have
> > the
> > > > > > same.
> > > > > > > > > > Consider
> > > > > > > > > > > > >> this step:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> > > > heartbeats,
> > > > > > the
> > > > > > > > > group
> > > > > > > > > > > > >> coordinator
> > > > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> > > > partitions
> > > > > > to
> > > > > > > > > > > revoke. It
> > > > > > > > > > > > >> > > > persists
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group
> > > > (epoch=3)
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > > > > > Assignment
> > > > > > > > > > > (epoch=3)
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > > > > > partitions=[foo-0]
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > > > > > partitions=[foo-2]
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > > > > > partitions=[foo-1]
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > > > > > Assignment
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > epoch=2,
> > > > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > epoch=3,
> > > > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > epoch=3,
> > > > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> > > > heartbeats,
> > > > > > it
> > > > > > > > > > > transitions
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > > > >> > > > > > but
> > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's
> > > > mentioned
> > > > > > > that
> > > > > > > > > > > member C
> > > > > > > > > > > > >> can't
> > > > > > > > > > > > >> > > > get
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > > > > > description
> > > > > > > > > > > above,
> > > > > > > > > > > > >> it seems
> > > > > > > > > > > > >> > > > it
> > > > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be
> > better
> > > > to
> > > > > > > remove
> > > > > > > > > it
> > > > > > > > > > > and
> > > > > > > > > > > > >> > > > populate it
> > > > > > > > > > > > >> > > > > > only
> > > > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this
> > in a
> > > > lot
> > > > > > of
> > > > > > > > > other
> > > > > > > > > > > > >> places, so
> > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > >> > > > > > I
> > > > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding
> > > > connect ,
> > > > > > it
> > > > > > > > > might
> > > > > > > > > > > be
> > > > > > > > > > > > >> out of
> > > > > > > > > > > > >> > > > scope
> > > > > > > > > > > > >> > > > > > of
> > > > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> > > > understood it
> > > > > > > would
> > > > > > > > > > > > >> probably be
> > > > > > > > > > > > >> > > > > > running in
> > > > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the
> > new
> > > > > > > rebalance
> > > > > > > > > > > > >> protocol as it
> > > > > > > > > > > > >> > > > has
> > > > > > > > > > > > >> > > > > > its
> > > > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > Assignors(Eager
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > > > IncrementalCooperative).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul
> > 15,
> > > > 2022
> > > > > > > at
> > > > > > > > > 5:00
> > > > > > > > > > > PM
> > > > > > > > > > > > >> David
> > > > > > > > > > > > >> > > > Jacot
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> > > > Hector! Our
> > > > > > > goal
> > > > > > > > > is
> > > > > > > > > > > to
> > > > > > > > > > > > >> move
> > > > > > > > > > > > >> > > > forward
> > > > > > > > > > > > >> > > > > > with
> > > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of
> > > > relying
> > > > > > > on
> > > > > > > > > one
> > > > > > > > > > > > >> generic API.
> > > > > > > > > > > > >> > > > For
> > > > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same
> > > > pattern
> > > > > > > and
> > > > > > > > > > > > >> reuse/share the
> > > > > > > > > > > > >> > > > core
> > > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server
> > side.
> > > > For
> > > > > > the
> > > > > > > > > > schema
> > > > > > > > > > > > >> registry, I
> > > > > > > > > > > > >> > > > > > think
> > > > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a
> > > > tailored
> > > > > > > API to
> > > > > > > > > > do
> > > > > > > > > > > > >> simple
> > > > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri,
> > Jul
> > > > 15,
> > > > > > > 2022 at
> > > > > > > > > > > 10:22
> > > > > > > > > > > > >> AM Ismael
> > > > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three
> > quick
> > > > > > > comments:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1.
> > Regarding
> > > > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > > > >> > > > vs
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > com.google.re2j.Pattern,
> > > > > > we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > document the
> > > > > > > > > differences
> > > > > > > > > > > in
> > > > > > > > > > > > >> more
> > > > > > > > > > > > >> > > > detail
> > > > > > > > > > > > >> > > > > > before
> > > > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That
> > said,
> > > > if
> > > > > > > people
> > > > > > > > > > pass
> > > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > semantics
> > > > to be
> > > > > > > > > honored.
> > > > > > > > > > > If
> > > > > > > > > > > > >> we are
> > > > > > > > > > > > >> > > > doing
> > > > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should
> > > > consider
> > > > > > > adding
> > > > > > > > > > an
> > > > > > > > > > > > >> overload
> > > > > > > > > > > > >> > > > with
> > > > > > > > > > > > >> > > > > > our own
> > > > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think
> > we'd
> > > > want
> > > > > > to
> > > > > > > > > > expose
> > > > > > > > > > > > >> re2j's at
> > > > > > > > > > > > >> > > > this
> > > > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2.
> > Regarding
> > > > > > topic
> > > > > > > > > ids,
> > > > > > > > > > > any
> > > > > > > > > > > > >> major new
> > > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and
> > > > should
> > > > > > > handle
> > > > > > > > > the
> > > > > > > > > > > topic
> > > > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we
> > > > need to
> > > > > > > > > handle.
> > > > > > > > > > I
> > > > > > > > > > > > >> agree with
> > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the
> > > > relevant
> > > > > > > > > > protocols
> > > > > > > > > > > > >> that don't
> > > > > > > > > > > > >> > > > > > have it
> > > > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus
> > on the
> > > > > > > internals
> > > > > > > > > > > versus
> > > > > > > > > > > > >> adding
> > > > > > > > > > > > >> > > > new
> > > > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find
> > that
> > > > > > > adding
> > > > > > > > > new
> > > > > > > > > > > APIs
> > > > > > > > > > > > >> is
> > > > > > > > > > > > >> > > > required
> > > > > > > > > > > > >> > > > > > for
> > > > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am
> > > > still
> > > > > > not
> > > > > > > sure
> > > > > > > > > > > about
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > powerful for
> > > > > > > configs
> > > > > > > > > to
> > > > > > > > > > be
> > > > > > > > > > > > >> > > > centralized
> > > > > > > > > > > > >> > > > > > in the
> > > > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > > > > > (auditability,
> > > > > > > > > > > > >> visibility,
> > > > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure
> > about
> > > > > > > > > automatically
> > > > > > > > > > > > >> deleting
> > > > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > recovered. A
> > > > > > good
> > > > > > > > > > property
> > > > > > > > > > > > >> for modern
> > > > > > > > > > > > >> > > > > > systems
> > > > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > unrecoverable
> > > > > > data
> > > > > > > > > loss
> > > > > > > > > > > > >> scenarios.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed,
> > Jul
> > > > 13,
> > > > > > > 2022
> > > > > > > > > at
> > > > > > > > > > > 3:47
> > > > > > > > > > > > >> PM David
> > > > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > > > > > Guozhang. My
> > > > > > > > > > > answers
> > > > > > > > > > > > >> are
> > > > > > > > > > > > >> > > > below:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1)
> > the
> > > > > > > migration
> > > > > > > > > > path,
> > > > > > > > > > > > >> especially
> > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > enable
> > > > > > the
> > > > > > > new
> > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > protocols /
> > > > > > > rpcs
> > > > > > > > > and
> > > > > > > > > > > old
> > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > group.
> > > > How
> > > > > > the
> > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > > > protocol
> > > > > > > is
> > > > > > > > > > > something
> > > > > > > > > > > > >> we
> > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Noted. I
> > > > just
> > > > > > > > > > published
> > > > > > > > > > > a
> > > > > > > > > > > > >> new
> > > > > > > > > > > > >> > > > version
> > > > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > details
> > > > about
> > > > > > > this.
> > > > > > > > > > See
> > > > > > > > > > > the
> > > > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > the
> > > > > > > > > > "Compatibility,
> > > > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > > > >> > > > > > and
> > > > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I
> > have to
> > > > > > think
> > > > > > > > > > through
> > > > > > > > > > > a
> > > > > > > > > > > > >> few cases
> > > > > > > > > > > > >> > > > > > now but
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > mechanism
> > > > > > > should be
> > > > > > > > > > > > >> understandable.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2)
> > the
> > > > usage
> > > > > > > of
> > > > > > > > > > topic
> > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > part of
> > > > RPCs
> > > > > > > and
> > > > > > > > > > admin
> > > > > > > > > > > > >> client,
> > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > consumers
> > > > > > > yet.
> > > > > > > > > I
> > > > > > > > > > > think
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > client
> > > > to be
> > > > > > > > > > > maintaining
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > leverage
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on
> > all
> > > > the
> > > > > > > > > augmented
> > > > > > > > > > > > >> existing
> > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > secondly,
> > > > > > > should
> > > > > > > > > we
> > > > > > > > > > > ever
> > > > > > > > > > > > >> consider
> > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > public
> > > > APIs
> > > > > > as
> > > > > > > > > well
> > > > > > > > > > > (both
> > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > listener for
> > > > > > > cases
> > > > > > > > > > > like
> > > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> > > > Assuming
> > > > > > > that we
> > > > > > > > > > > would
> > > > > > > > > > > > >> include
> > > > > > > > > > > > >> > > > > > converting
> > > > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs
> > to
> > > > using
> > > > > > > topic
> > > > > > > > > > ids
> > > > > > > > > > > in
> > > > > > > > > > > > >> this
> > > > > > > > > > > > >> > > > KIP,
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully
> > > > operate
> > > > > > > with
> > > > > > > > > > topic
> > > > > > > > > > > > >> ids. That
> > > > > > > > > > > > >> > > > > > being
> > > > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > topics
> > > > > > > names in
> > > > > > > > > > > various
> > > > > > > > > > > > >> APIs so
> > > > > > > > > > > > >> > > > > > having a
> > > > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > > > > > inevitable to
> > > > > > > > > > me.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I
> > don't
> > > > > > have
> > > > > > > a
> > > > > > > > > > strong
> > > > > > > > > > > > >> opinion on
> > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > beyond the
> > > > > > > scope of
> > > > > > > > > > this
> > > > > > > > > > > > >> KIP. I
> > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > >> > > > > > rather
> > > > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here
> > and
> > > > we
> > > > > > can
> > > > > > > > > > consider
> > > > > > > > > > > > >> this
> > > > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming
> > > > back to
> > > > > > > > > > Ismael's
> > > > > > > > > > > > >> point about
> > > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > > > >> > > > think
> > > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > favour of
> > > > it.
> > > > > > > The
> > > > > > > > > > > consumer
> > > > > > > > > > > > >> will
> > > > > > > > > > > > >> > > > have
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > topics
> > > > > > > exists
> > > > > > > > > > before
> > > > > > > > > > > > >> passing
> > > > > > > > > > > > >> > > > them
> > > > > > > > > > > > >> > > > > > into
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > protocol.
> > > > > > > Obviously,
> > > > > > > > > > the
> > > > > > > > > > > > >> > > > coordinator
> > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not
> > really
> > > > > > have
> > > > > > > a
> > > > > > > > > way
> > > > > > > > > > to
> > > > > > > > > > > > >> reject an
> > > > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > > > agreeing
> > > > > > > with
> > > > > > > > > > > David
> > > > > > > > > > > > >> on all
> > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > >> question:
> > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > deprecate
> > > > > > the
> > > > > > > > > > > subscribe
> > > > > > > > > > > > >> API with
> > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Pattern
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > while
> > > > just
> > > > > > > > > > documenting
> > > > > > > > > > > > >> that our
> > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > server.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > Since
> > > > the
> > > > > > > > > > incompatible
> > > > > > > > > > > > >> case is a
> > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > overloaded
> > > > > > > > > `String`
> > > > > > > > > > > based
> > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > invalid
> > > > > > > regexes.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That
> > could
> > > > > > > work. I
> > > > > > > > > > have
> > > > > > > > > > > to
> > > > > > > > > > > > >> look at
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > engines to
> > > > > > > better
> > > > > > > > > > > > >> understand the
> > > > > > > > > > > > >> > > > > > potential
> > > > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding
> > is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > would
> > > > > > work
> > > > > > > for
> > > > > > > > > > all
> > > > > > > > > > > the
> > > > > > > > > > > > >> basic
> > > > > > > > > > > > >> > > > > > regular
> > > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > between
> > > > the
> > > > > > two
> > > > > > > are
> > > > > > > > > > > mainly
> > > > > > > > > > > > >> about
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > various
> > > > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder
> > > > what
> > > > > > > other
> > > > > > > > > > people
> > > > > > > > > > > > >> think
> > > > > > > > > > > > >> > > > about
> > > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On
> > Tue,
> > > > Jul
> > > > > > 12,
> > > > > > > 2022
> > > > > > > > > > at
> > > > > > > > > > > > >> 11:28 PM
> > > > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > wangguoz@gmail.com
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > Thanks
> > > > > > David!
> > > > > > > I
> > > > > > > > > > think
> > > > > > > > > > > on
> > > > > > > > > > > > >> the high
> > > > > > > > > > > > >> > > > > > level
> > > > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > > > concretize
> > > > > > > a
> > > > > > > > > bit
> > > > > > > > > > > more:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1)
> > the
> > > > > > > migration
> > > > > > > > > > path,
> > > > > > > > > > > > >> especially
> > > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > enable
> > > > > > the
> > > > > > > new
> > > > > > > > > > > > >> protocol, in
> > > > > > > > > > > > >> > > > which
> > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > protocols /
> > > > > > > rpcs
> > > > > > > > > and
> > > > > > > > > > > old
> > > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > group.
> > > > How
> > > > > > the
> > > > > > > > > > > > >> coordinator could
> > > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > > > protocol
> > > > > > > is
> > > > > > > > > > > something
> > > > > > > > > > > > >> we
> > > > > > > > > > > > >> > > > need to
> > > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2)
> > the
> > > > usage
> > > > > > > of
> > > > > > > > > > topic
> > > > > > > > > > > > >> ids. So
> > > > > > > > > > > > >> > > > far as
> > > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > part of
> > > > RPCs
> > > > > > > and
> > > > > > > > > > admin
> > > > > > > > > > > > >> client,
> > > > > > > > > > > > >> > > > but
> > > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > consumers
> > > > > > > yet.
> > > > > > > > > I
> > > > > > > > > > > think
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > client
> > > > to be
> > > > > > > > > > > maintaining
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > leverage
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on
> > all
> > > > the
> > > > > > > > > augmented
> > > > > > > > > > > > >> existing
> > > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > secondly,
> > > > > > > should
> > > > > > > > > we
> > > > > > > > > > > ever
> > > > > > > > > > > > >> consider
> > > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > public
> > > > APIs
> > > > > > as
> > > > > > > > > well
> > > > > > > > > > > (both
> > > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > listener for
> > > > > > > cases
> > > > > > > > > > > like
> > > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > > > agreeing
> > > > > > > with
> > > > > > > > > > > David
> > > > > > > > > > > > >> on all
> > > > > > > > > > > > >> > > > other
> > > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > > >> question:
> > > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > deprecate
> > > > > > the
> > > > > > > > > > > subscribe
> > > > > > > > > > > > >> API with
> > > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Pattern
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > while
> > > > just
> > > > > > > > > > documenting
> > > > > > > > > > > > >> that our
> > > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > server.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > Since
> > > > the
> > > > > > > > > > incompatible
> > > > > > > > > > > > >> case is a
> > > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > overloaded
> > > > > > > > > `String`
> > > > > > > > > > > based
> > > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > invalid
> > > > > > > regexes.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > Guozhang
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On
> > Tue,
> > > > Jul
> > > > > > > 12,
> > > > > > > > > 2022
> > > > > > > > > > > at
> > > > > > > > > > > > >> 5:23 AM
> > > > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi
> > > > Ismael,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > Thanks for
> > > > > > > your
> > > > > > > > > > > > >> feedback. Let
> > > > > > > > > > > > >> > > > me
> > > > > > > > > > > > >> > > > > > answer
> > > > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 1. I
> > > > > > think
> > > > > > > > > it's
> > > > > > > > > > > > >> premature to
> > > > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > deprecation
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > removal
> > > > > > > of the
> > > > > > > > > > > > >> existing group
> > > > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > client
> > > > > > > > > protocol
> > > > > > > > > > > and
> > > > > > > > > > > > >> hence
> > > > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > dependent
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > how
> > > > > > > quickly
> > > > > > > > > > > > >> applications
> > > > > > > > > > > > >> > > > migrate
> > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > That
> > > > makes
> > > > > > > > > sense.
> > > > > > > > > > I
> > > > > > > > > > > will
> > > > > > > > > > > > >> > > > remove it.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 2.
> > > > The
> > > > > > KIP
> > > > > > > > > says
> > > > > > > > > > we
> > > > > > > > > > > > >> intend to
> > > > > > > > > > > > >> > > > > > release
> > > > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > clear
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > why. If
> > > > > > we
> > > > > > > > > added
> > > > > > > > > > > that
> > > > > > > > > > > > >> as a
> > > > > > > > > > > > >> > > > way to
> > > > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > the
> > > > > > group
> > > > > > > > > > > protocol, I
> > > > > > > > > > > > >> also
> > > > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Let me
> > > > > > > explain
> > > > > > > > > my
> > > > > > > > > > > > >> reasoning. As
> > > > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > coordinator
> > > > > > > in
> > > > > > > > > > Java
> > > > > > > > > > > > >> while we
> > > > > > > > > > > > >> > > > > > implement
> > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > that
> > > > the
> > > > > > > > > internals
> > > > > > > > > > > will
> > > > > > > > > > > > >> be
> > > > > > > > > > > > >> > > > slightly
> > > > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > Therefore, I
> > > > > > > > > > wanted
> > > > > > > > > > > to
> > > > > > > > > > > > >> tighten
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > coordinator
> > > > > > > to
> > > > > > > > > the
> > > > > > > > > > > new
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > alternative
> > > > > > > > > would
> > > > > > > > > > > be to
> > > > > > > > > > > > >> use a
> > > > > > > > > > > > >> > > > flag
> > > > > > > > > > > > >> > > > > > to do
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > on the
> > > > > > > software
> > > > > > > > > > > upgrade.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 3.
> > > > We
> > > > > > > need to
> > > > > > > > > > > flesh
> > > > > > > > > > > > >> out the
> > > > > > > > > > > > >> > > > > > details of
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > like
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > we're
> > > > > > > saying
> > > > > > > > > we
> > > > > > > > > > > will
> > > > > > > > > > > > >> support
> > > > > > > > > > > > >> > > > > > online
> > > > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > should
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > explain
> > > > > > > this
> > > > > > > > > in
> > > > > > > > > > > > >> detail. It
> > > > > > > > > > > > >> > > > could
> > > > > > > > > > > > >> > > > > > also
> > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > easier.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Yes,
> > > > we
> > > > > > will
> > > > > > > > > > support
> > > > > > > > > > > > >> online
> > > > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > group
> > > > > > > using
> > > > > > > > > the
> > > > > > > > > > > old
> > > > > > > > > > > > >> protocol
> > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > protocol.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Let me
> > > > > > > briefly
> > > > > > > > > > > explain
> > > > > > > > > > > > >> how that
> > > > > > > > > > > > >> > > > > > will work
> > > > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > four
> > > > step
> > > > > > > > > process:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > 1. The
> > > > > > > cluster
> > > > > > > > > > must
> > > > > > > > > > > be
> > > > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > new
> > > > group
> > > > > > > > > > > coordinator.
> > > > > > > > > > > > >> Both the
> > > > > > > > > > > > >> > > > > > old and
> > > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > support
> > > > > > the
> > > > > > > old
> > > > > > > > > > > > >> protocol and
> > > > > > > > > > > > >> > > > rely
> > > > > > > > > > > > >> > > > > > on the
> > > > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > they
> > > > can
> > > > > > > work
> > > > > > > > > > > together.
> > > > > > > > > > > > >> This
> > > > > > > > > > > > >> > > > point
> > > > > > > > > > > > >> > > > > > is an
> > > > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > this
> > > > > > one
> > > > > > > live
> > > > > > > > > > > > >> because it
> > > > > > > > > > > > >> > > > would
> > > > > > > > > > > > >> > > > > > require
> > > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > coordinator
> > > > > > > and
> > > > > > > > > > > > >> starting up the
> > > > > > > > > > > > >> > > > > > new one
> > > > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > unavailabilities.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > 2. The
> > > > > > > cluster's
> > > > > > > > > > > > >> metadata
> > > > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > enable
> > > > > > > the
> > > > > > > > > new
> > > > > > > > > > > > >> protocol.
> > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > terminated
> > > > > > > > > because
> > > > > > > > > > > the
> > > > > > > > > > > > >> old
> > > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > protocol.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > 3. The
> > > > > > > consumers
> > > > > > > > > > > must be
> > > > > > > > > > > > >> > > > upgraded
> > > > > > > > > > > > >> > > > > > to a
> > > > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > migration
> > > > > > > (must
> > > > > > > > > > have
> > > > > > > > > > > > >> KIP-792).
> > > > > > > > > > > > >> > > > If
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > Nothing
> > > > > > > must be
> > > > > > > > > > > done at
> > > > > > > > > > > > >> this
> > > > > > > > > > > > >> > > > point.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > 4. The
> > > > > > > consumers
> > > > > > > > > > > must
> > > > > > > > > > > > >> be rolled
> > > > > > > > > > > > >> > > > > > with the
> > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > consumer
> > > > > > > group
> > > > > > > > > is
> > > > > > > > > > > > >> automatically
> > > > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > using
> > > > the
> > > > > > > new
> > > > > > > > > > > protocol
> > > > > > > > > > > > >> joins
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > group.
> > > > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > old
> > > > > > > protocol are
> > > > > > > > > > > being
> > > > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > > > >> > > > > > the old
> > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > new
> > > > one.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Let me
> > > > > > > clarify
> > > > > > > > > all
> > > > > > > > > > > of
> > > > > > > > > > > > >> this in
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 4.
> > > > I am
> > > > > > > happy
> > > > > > > > > > > that we
> > > > > > > > > > > > >> are
> > > > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > server,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > but
> > > > it
> > > > > > > seems
> > > > > > > > > > like
> > > > > > > > > > > > >> there
> > > > > > > > > > > > >> > > > could be
> > > > > > > > > > > > >> > > > > > some
> > > > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > have a
> > > > > > > > > mechanism
> > > > > > > > > > > for
> > > > > > > > > > > > >> users to
> > > > > > > > > > > > >> > > > > > detect
> > > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > before
> > > > > > > > > switching
> > > > > > > > > > > to
> > > > > > > > > > > > >> the new
> > > > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > think
> > > > > > > that I
> > > > > > > > > am
> > > > > > > > > > a
> > > > > > > > > > > bit
> > > > > > > > > > > > >> more
> > > > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > believe
> > > > > > > that the
> > > > > > > > > > > > >> majority of
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > cases
> > > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > work
> > > > with
> > > > > > > the
> > > > > > > > > new
> > > > > > > > > > > > >> engine. The
> > > > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > and
> > > > reject
> > > > > > > the
> > > > > > > > > > > consumer
> > > > > > > > > > > > >> if the
> > > > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > migration
> > > > > > > path,
> > > > > > > > > in
> > > > > > > > > > > the
> > > > > > > > > > > > >> worst
> > > > > > > > > > > > >> > > > case,
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > the
> > > > group
> > > > > > > will
> > > > > > > > > be
> > > > > > > > > > > > >> rejected.
> > > > > > > > > > > > >> > > > This
> > > > > > > > > > > > >> > > > > > should
> > > > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > would
> > > > say.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > One
> > > > way
> > > > > > for
> > > > > > > > > > > customers to
> > > > > > > > > > > > >> > > > validate
> > > > > > > > > > > > >> > > > > > their
> > > > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > prod
> > > > would
> > > > > > > be to
> > > > > > > > > > > test
> > > > > > > > > > > > >> them with
> > > > > > > > > > > > >> > > > > > another
> > > > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > could
> > > > be
> > > > > > > done
> > > > > > > > > in a
> > > > > > > > > > > > >> pre-prod
> > > > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > extend the
> > > > > > > > > > > > >> consumer-group tool
> > > > > > > > > > > > >> > > > to
> > > > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > mechanism.
> > > > > > > Would
> > > > > > > > > > > this be
> > > > > > > > > > > > >> > > > enough in
> > > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 5.
> > > > > > > Related to
> > > > > > > > > > the
> > > > > > > > > > > last
> > > > > > > > > > > > >> > > > question,
> > > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users
> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > stick
> > > > > > > with the
> > > > > > > > > > > > >> current regex
> > > > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > example,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > it
> > > > may
> > > > > > be
> > > > > > > > > handy
> > > > > > > > > > to
> > > > > > > > > > > > >> keep using
> > > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > migrations
> > > > > > > > > > simple
> > > > > > > > > > > and
> > > > > > > > > > > > >> then
> > > > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > step.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > > > understand
> > > > > > > > > your
> > > > > > > > > > > point
> > > > > > > > > > > > >> but I
> > > > > > > > > > > > >> > > > am
> > > > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > actually
> > > > > > > stay
> > > > > > > > > > in
> > > > > > > > > > > > >> this mode.
> > > > > > > > > > > > >> > > > That
> > > > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > simplifying
> > > > > > > the
> > > > > > > > > > > client
> > > > > > > > > > > > >> because
> > > > > > > > > > > > >> > > > we
> > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > the
> > > > > > > metadata on
> > > > > > > > > > the
> > > > > > > > > > > > >> client
> > > > > > > > > > > > >> > > > side. I
> > > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 6.
> > > > When
> > > > > > > we say
> > > > > > > > > > > that
> > > > > > > > > > > > >> the group
> > > > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible
> > for
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > storing
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > the
> > > > > > > > > > configurations
> > > > > > > > > > > > >> and that
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > group is
> > > > > > > > > > deleted.
> > > > > > > > > > > > >> Will a
> > > > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > configurations?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > That's
> > > > > > > right.
> > > > > > > > > The
> > > > > > > > > > > > >> > > > configurations
> > > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > deleted.
> > > > > > > They go
> > > > > > > > > > > > >> together.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 7.
> > > > Will
> > > > > > > the
> > > > > > > > > > > choice to
> > > > > > > > > > > > >> store
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make
> > it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > harder
> > > > > > to
> > > > > > > list
> > > > > > > > > > all
> > > > > > > > > > > > >> cluster
> > > > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > don't
> > > > > > > think
> > > > > > > > > so.
> > > > > > > > > > > The
> > > > > > > > > > > > >> group
> > > > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > configs.
> > > > > > If
> > > > > > > you
> > > > > > > > > > > want to
> > > > > > > > > > > > >> know
> > > > > > > > > > > > >> > > > all
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > ask all
> > > > > > > the
> > > > > > > > > > group
> > > > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > > > >> > > > > > You
> > > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > for
> > > > > > > instance.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 8.
> > > > How
> > > > > > > would
> > > > > > > > > > > someone
> > > > > > > > > > > > >> > > > configure a
> > > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Have
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > we
> > > > > > > considered
> > > > > > > > > > > > >> allowing the
> > > > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > Alternatively,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > configs
> > > > > > > could
> > > > > > > > > be
> > > > > > > > > > > > >> decoupled
> > > > > > > > > > > > >> > > > from
> > > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Yes.
> > > > The
> > > > > > > group
> > > > > > > > > > will
> > > > > > > > > > > be
> > > > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > the
> > > > > > configs
> > > > > > > will
> > > > > > > > > > be
> > > > > > > > > > > > >> lost after
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > passes.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 9.
> > > > Will
> > > > > > > the
> > > > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > > > >> > > > > > method
> > > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > > > >> > > > do
> > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > That's a
> > > > > > > very
> > > > > > > > > > group
> > > > > > > > > > > > >> question. I
> > > > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > > > >> > > > not
> > > > > > > > > > > > >> > > > > > fully
> > > > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > plan to
> > > > > > > use,
> > > > > > > > > it
> > > > > > > > > > > > >> might be
> > > > > > > > > > > > >> > > > better
> > > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > which
> > > > > > takes
> > > > > > > a
> > > > > > > > > > > string.
> > > > > > > > > > > > >> We would
> > > > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > During the
> > > > > > > > > > > migration, I
> > > > > > > > > > > > >> think
> > > > > > > > > > > > >> > > > that
> > > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > regex
> > > > and
> > > > > > > use
> > > > > > > > > it.
> > > > > > > > > > > That
> > > > > > > > > > > > >> should
> > > > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > cases.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 10.
> > > > I
> > > > > > > agree
> > > > > > > > > with
> > > > > > > > > > > > >> Justine
> > > > > > > > > > > > >> > > > that we
> > > > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > switch
> > > > > > to
> > > > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > > > >> > > > > > from the
> > > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > can
> > > > > > > switch the
> > > > > > > > > > > > >> default for
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > feature
> > > > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > metadata.version
> > > > > > > > > > > once
> > > > > > > > > > > > >> we
> > > > > > > > > > > > >> > > > want to
> > > > > > > > > > > > >> > > > > > make
> > > > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My
> > > > plan
> > > > > > was
> > > > > > > to
> > > > > > > > > use
> > > > > > > > > > > that
> > > > > > > > > > > > >> feature
> > > > > > > > > > > > >> > > > > > flag
> > > > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > phase. I
> > > > > > > should
> > > > > > > > > > not
> > > > > > > > > > > have
> > > > > > > > > > > > >> > > > mentioned
> > > > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > > > > > internal
> > > > > > > > > config
> > > > > > > > > > > for
> > > > > > > > > > > > >> it.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 11.
> > > > Some
> > > > > > > of
> > > > > > > > > the
> > > > > > > > > > > > >> protocol APIs
> > > > > > > > > > > > >> > > > > > don't
> > > > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > good to
> > > > > > > add
> > > > > > > > > that
> > > > > > > > > > > for
> > > > > > > > > > > > >> > > > consistency.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Noted.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 12.
> > > > It
> > > > > > is
> > > > > > > a
> > > > > > > > > bit
> > > > > > > > > > > odd
> > > > > > > > > > > > >> that
> > > > > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > even
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > though
> > > > > > it
> > > > > > > > > seems
> > > > > > > > > > > to do
> > > > > > > > > > > > >> more
> > > > > > > > > > > > >> > > > than
> > > > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > agree.
> > > > > > > This is
> > > > > > > > > > > how the
> > > > > > > > > > > > >> > > > current
> > > > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > require
> > > > > > > "Read
> > > > > > > > > > > Group" to
> > > > > > > > > > > > >> join a
> > > > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > but I
> > > > am
> > > > > > not
> > > > > > > > > sure
> > > > > > > > > > > that
> > > > > > > > > > > > >> it is
> > > > > > > > > > > > >> > > > worth
> > > > > > > > > > > > >> > > > > > it.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 13.
> > > > How
> > > > > > is
> > > > > > > > > topic
> > > > > > > > > > > > >> recreation
> > > > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > protocol?
> > > > > > > It
> > > > > > > > > > > would be
> > > > > > > > > > > > >> good to
> > > > > > > > > > > > >> > > > > > have a
> > > > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Noted.
> > > > > > From
> > > > > > > a
> > > > > > > > > > > protocol
> > > > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > topic
> > > > id
> > > > > > so
> > > > > > > it
> > > > > > > > > > will
> > > > > > > > > > > > >> treat it
> > > > > > > > > > > > >> > > > like a
> > > > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > only
> > > > issue
> > > > > > > is
> > > > > > > > > that
> > > > > > > > > > > the
> > > > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > IDs
> > > > so the
> > > > > > > > > > consumer
> > > > > > > > > > > > >> would
> > > > > > > > > > > > >> > > > reuse the
> > > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > that
> > > > we
> > > > > > > should
> > > > > > > > > > > update
> > > > > > > > > > > > >> those
> > > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > > >> > > > > > as well
> > > > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > end.
> > > > > > That
> > > > > > > > > would
> > > > > > > > > > > > >> strengthen
> > > > > > > > > > > > >> > > > the
> > > > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 14.
> > > > The
> > > > > > > KIP
> > > > > > > > > > > mentions
> > > > > > > > > > > > >> we will
> > > > > > > > > > > > >> > > > > > write the
> > > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > though
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > this is
> > > > > > an
> > > > > > > > > > > > >> implementation
> > > > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > for
> > > > it?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Yes.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 15.
> > > > Do
> > > > > > we
> > > > > > > > > have a
> > > > > > > > > > > > >> scalability
> > > > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > new
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > group
> > > > > > > protocol
> > > > > > > > > > can
> > > > > > > > > > > > >> support?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We
> > > > don't
> > > > > > > have
> > > > > > > > > > > numbers
> > > > > > > > > > > > >> at the
> > > > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of
> > > > members
> > > > > > > per
> > > > > > > > > > > group.
> > > > > > > > > > > > >> We will
> > > > > > > > > > > > >> > > > > > measure
> > > > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > implementation.
> > > > > > > > > > Note
> > > > > > > > > > > > >> that we
> > > > > > > > > > > > >> > > > might
> > > > > > > > > > > > >> > > > > > have
> > > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > road
> > > > (e.g.
> > > > > > > > > offset
> > > > > > > > > > > > >> commits).
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > 16.
> > > > Did
> > > > > > we
> > > > > > > > > > > consider
> > > > > > > > > > > > >> having
> > > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > SubscribedTopicNames
> > > > > > > > > > > > >> in
> > > > > > > > > > > > >> > > > > > > > > > > >> >
> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > since we
> > > > > > > have
> > > > > > > > > to
> > > > > > > > > > > > >> resolve the
> > > > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > the
> > > > > > topic
> > > > > > > > > name?
> > > > > > > > > > > The
> > > > > > > > > > > > >> > > > difference
> > > > > > > > > > > > >> > > > > > is that
> > > > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > efficient
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > whereas
> > > > > > > > > sending
> > > > > > > > > > > the
> > > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > > > names
> > > > > > > > > > > > >> > > > > > is less
> > > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > and
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > recreation is
> > > > > > > > > > > easier
> > > > > > > > > > > > >> to
> > > > > > > > > > > > >> > > > handle
> > > > > > > > > > > > >> > > > > > if we
> > > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > The
> > > > idea
> > > > > > > was to
> > > > > > > > > > > > >> consolidate the
> > > > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > paths
> > > > but
> > > > > > I
> > > > > > > do
> > > > > > > > > > agree
> > > > > > > > > > > > >> with your
> > > > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > ids
> > > > may be
> > > > > > > > > better
> > > > > > > > > > > here
> > > > > > > > > > > > >> for the
> > > > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > suppose
> > > > > > > that we
> > > > > > > > > > may
> > > > > > > > > > > > >> allow
> > > > > > > > > > > > >> > > > users to
> > > > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > future
> > > > > > > because
> > > > > > > > > > that
> > > > > > > > > > > is
> > > > > > > > > > > > >> the only
> > > > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > re-creation.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Best,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > David
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On
> > > > Tue,
> > > > > > Jul
> > > > > > > 12,
> > > > > > > > > > > 2022 at
> > > > > > > > > > > > >> 1:38 PM
> > > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > djacot@confluent.io>
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > Hi
> > > > > > > Justine,
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > Thanks
> > > > > > for
> > > > > > > > > your
> > > > > > > > > > > > >> comments.
> > > > > > > > > > > > >> > > > Please
> > > > > > > > > > > > >> > > > > > find
> > > > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > -
> > > > Yes,
> > > > > > > the new
> > > > > > > > > > > > >> protocol
> > > > > > > > > > > > >> > > > relies on
> > > > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > topic
> > > > > > > names
> > > > > > > > > > based
> > > > > > > > > > > in
> > > > > > > > > > > > >> the
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > if
> > > > using
> > > > > > > topic
> > > > > > > > > > > names
> > > > > > > > > > > > >> is the
> > > > > > > > > > > > >> > > > > > right call
> > > > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > a
> > > > little
> > > > > > > more.
> > > > > > > > > > > > >> Obviously, the
> > > > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > > > >> > > > > > > > > > > >>
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Thanks David, I read through the updated wiki, and I have a couple
follow-up questions:

1. It seems we would have two potential sources to update the consumer's
internally maintained topic name -> topic ID mapping, which is used for
encoding subscribed topic Ids in the HB request, and offset commit / list
offset requests, etc. One is from metadata responses, and another one is
from the GPA (group prepare assignment). On the other hand, the assignor
APIs directly take parsed AssignmentMemberSpec and AssignmentTopicMetadata
from GPA responses directly instead of trying to consult the consumer's own
cached mapping. So just to clarify, consumers would also update its
internal mapping from the GPA responses as well, right?

2. I'm wondering if it's necessary to expose the topic IDs to the assignor
developers. What they need to do at the moment is to take the mapping from
AssignmentTopicMetadata, and translate their decided assignment into topic
IDs when returning the MemberAssignment. We can instead just let consumer
to do such translation for them internally and still let the assignor
developers to only deal with the topic names. Of course that means the
client side assignor APIs would be a bit further diverged from the
broker-side assignor such that:

a) Topic metadata would be encoded just as String -> num.partitions;
b) AssignmentMemberSpec contains subscribedTopicNames and topicPartitions
as target assignments;
c) Returned MemberAssignment contains list of topicPartitions instead of
topicIdPartitions.

Internally, consumers would convert the IDs to names when calling the
assign() function, and convert them back when sending the request; and
similarly it would convert the IDs to names when calling onAssignment() as
well.

In short, I'm wondering if we should bite the bullet for the users, eating
more complexities internally than pushing that to users.

3. This is a nit one: the return type of `Assignment assign(AssignmentSpec
assignmentSpec);` seems not defined in the doc. I think it would just
include two fields, as:

class Assignment {
        GroupAssignment groupAssignment;

        ConsumerGroupMetadata consumerGroupMetadata;
    }

Is that right?

Guozhang

On Thu, Sep 29, 2022 at 1:26 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Guozhang,
>
> Thanks for your input.
>
> > 1) in HB request, when clients send the subscription, they are still
> > sending them in terms of topic names. That is, the consumer would not try
> > to keep all the mappings from topic names to topic ids for the subscribed
> > topics.
>
> This is a mistake of mine. The heartbeat request should use Topic Ids.
> I forgot to change it. The idea is that the client will get the
> metadata first to translate the topic names to topic ids and thus also
> have the mapping in its local metadata cache.
>
> > 2) in HB responses, the assignment's topics are encoded as IDs already.
> And
> > fetch requests can directly rely on that since they just need topic IDs
> and
> > do not care what actual topic names are.
>
> That's right.
>
> > 3) in prepare-assignment responses, we return subscriptions as topic IDs
> as
> > well.
>
> That's right.
>
> > In a word, the consumer does not try to maintain the mapping of topic
> names
> > to IDs at all, and even if they do, they will not expose it in e.g.
> > assignor APIs. But for client-side assignors like Streams, if they do not
> > maintain the mapping from topic IDs to topic names, certain logic such as
> > co-partitioning indeed cannot be executed.
>
> As I said, the consumer will have the mapping in its local metadata
> cache. It needs to get the partition metadata anyway and the topic id
> is part of it. However, I do agree that this mapping is not exposed.
>
> > So I think we'd need to consider 1) use topic names in subscriptions in
> the
> > prepare-assignment responses, 2) use topic Names in client-side
> > PartitionAssignor#GroupMember#subscribedTopics and
> > PartitionAssignor#TopicMetadata#topicName.
>
> 1) makes sense to me. I think that we could actually provide both
> names and ids. Regarding 2), I think that I am slightly in favor of
> using both names and ids as well. Let me try to update the protocol
> and the interface towards this.
>
> Thanks,
> David
>
> On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello Jun/David,
> >
> > Just to chime in here for Jun's comment #50, today Streams does not have
> a
> > way to map topic names to topic IDs, and in its topology description
> topics
> > are encoded with names.
> >
> > What I originally thought was that, on the consumer side, we'd need to
> > maintain this mapping from topic names to topic IDs anyways, and Streams
> > should leverage on the consumer's functionality to get the corresponding
> > topic names for the given topic IDs. But when I read through the RPC
> again
> > I realized it may not be the case, since:
> >
> > 1) in HB request, when clients send the subscription, they are still
> > sending them in terms of topic names. That is, the consumer would not try
> > to keep all the mappings from topic names to topic ids for the subscribed
> > topics.
> > 2) in HB responses, the assignment's topics are encoded as IDs already.
> And
> > fetch requests can directly rely on that since they just need topic IDs
> and
> > do not care what actual topic names are.
> > 3) in prepare-assignment responses, we return subscriptions as topic IDs
> as
> > well.
> >
> > In a word, the consumer does not try to maintain the mapping of topic
> names
> > to IDs at all, and even if they do, they will not expose it in e.g.
> > assignor APIs. But for client-side assignors like Streams, if they do not
> > maintain the mapping from topic IDs to topic names, certain logic such as
> > co-partitioning indeed cannot be executed.
> >
> > So I think we'd need to consider 1) use topic names in subscriptions in
> the
> > prepare-assignment responses, 2) use topic Names in client-side
> > PartitionAssignor#GroupMember#subscribedTopics and
> > PartitionAssignor#TopicMetadata#topicName.
> >
> > On Wed, Sep 28, 2022 at 1:26 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hi Jun,
> > >
> > > Thanks for your comments.
> > >
> > > 48. Okay. There is a typo in the text:
> > > > Note that the persisted target assignment of B already includes foo-2
> > > here
> > > but B does not get it yet because foo-2 is not free.
> > >
> > > It should be:
> > > > Note that the persisted member assignment of B already includes foo-2
> > > here
> > > but B does not get it yet because foo-2 is not free.
> > >
> > > The examples are showing the persisted state in the __consumer_offsets
> > > topics. When a member transitions to the next epoch, we directly
> > > persist the whole target assignment in order to reduce the number of
> > > writes and we manage dependencies in memory. So "B - epoch=2,
> > > partitions=[foo-2]" is actually correct from a persisted record
> > > perspective. However, the in-memory state would be different because
> > > foo-2 is not yet available as you pointed out. The in-memory state
> > > would like as follow:
> > >
> > > Group (epoch=2)
> > > A
> > > B
> > > Target Assignment (epoch=2)
> > > A - partitions=[foo-0, foo-1]
> > > B - partitions=[foo-2]
> > > Member Assignment
> > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending partitions=[]
> > > B - epoch=2, partitions=[], pending partitions=[foo-2]
> > >
> > > I suppose that the confusion comes from this. I will go and convert
> > > all the examples to their in-memory representation in order to reduce
> > > the confusion.
> > >
> > > 50. Okay. I have to admit that I am not so familiar with the internals
> > > of KStreams. If I understand correctly, you're saying that KStreams
> > > may not have topic ids in its topology so topic names are missing here
> > > to match the topics. I was hoping that KStreams would have topic ids
> > > in the topology as well. If that is not the case, we can also add
> > > topic names on our side. I will check this.
> > >
> > > 53. It seems that I was not clear. Sorry for that. I did not mean to
> > > say that MemberAssignment.metadata is not used by the group
> > > coordinator. Let me re-explain the flow.
> > >
> > > PartitionAssignor.assign() returns an Assignment. Assignment contains
> > > MemberAssignments where each member has MemberAssignment.metadata. The
> > > Assignment is installed by the client with the
> > > ConsumerGroupInstallAssignment API and this overrides the current
> > > target assignment in the coordinator. Then the group coordinator uses
> > > it to reconcile members. Therefore the target assignment will
> > > eventually become the current member assignment and it will be
> > > delivered to the member, including the original
> > > MemberAssignment.metadata defines by the assignor.
> > >
> > > I hope that this helps to clarify how MemberAssignment.metadata is
> > > used afterwards.
> > >
> > > Best,
> > > David
> > >
> > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <ju...@confluent.io.invalid>
> wrote:
> > > > Hi, David,
> > > >
> > > > Thanks for the reply. A few more follow up questions.
> > > >
> > > > 48. Hmm, so member assignment is what the client sees. Then, I am not
> > > sure
> > > > about the following. The text says B cannot get foo-2 yet. But in
> Member
> > > > Assignment, B has foo-2.
> > > >
> > > > At this point B can transition to epoch 2 but cannot get foo-2 until
> A
> > > > revokes it.
> > > >
> > > > Note that the persisted target assignment of B already includes foo-2
> > > here
> > > > but B does not get it yet because foo-2 is not free.
> > > >
> > > > Group (epoch=2)
> > > > A
> > > > B
> > > > Target Assignment (epoch=2)
> > > > A - partitions=[foo-0, foo-1]
> > > > B - partitions=[foo-2]
> > > > Member Assignment
> > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > > B - epoch=2, partitions=[foo-2]
> > > >
> > > >
> > > > 50. Hmm, I thought KStreams doesn't use range assignor all the time.
> It
> > > > does the assignment based on the topology. For example, if you are
> > > joining
> > > > two topics A and B and deriving the join result in a 3rd topic C,
> which
> > > > will be aggregated. You want to use range to co-locate the partitions
> > > from
> > > > topic A and B. However, for topic C, the partition could potentially
> be
> > > > assigned in round robin. Without the topic name, KStreams won't know
> > > which
> > > > topics need to be co-located.
> > > >
> > > > 53. If MemberAssignment.metadata is not used by the group
> coordinator,
> > > > where is it being used?
> > > >
> > > > Thanks,
> > > >
> > > > Jun
> > > >
> > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > > <ja...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi David,
> > > > >
> > > > > Thanks for the response. Replied below:
> > > > >
> > > > > > I don't have a strong opinion on this one. However, it seems that
> > > this
> > > > > could be a pandora box if our subset turns out to not be sufficient
> > > > > for some existing users. We will have to add more to it and the
> > > > > compatibility story between versions could be a real nightmare. One
> > > > > thing that we could do for sure is disabling all unicode groups in
> the
> > > > > library. That seems possible and would already reduce the scope.
> > > > >
> > > > > Changes to the regex syntax seems not a major problem to me. We
> would
> > > bump
> > > > > the protocol version as we usually do. Actually that begs the same
> > > question
> > > > > about RE2. How stable is the syntax and does it differ depending
> on the
> > > > > version of the library? In general, I do agree that we'd be better
> off
> > > not
> > > > > having to manage the syntax ourselves if we can convince ourselves
> that
> > > > > compatibility won't be a problem and the client won't eventually
> need
> > > to
> > > > > have access to it.
> > > > >
> > > > > Thanks,
> > > > > Jason
> > > > >
> > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > > <dj...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > > > Hi Luke,
> > > > > >
> > > > > > Thanks for your comments.
> > > > > >
> > > > > > Your suggestion makes sense. Jun's made the same one. We are also
> > > > > > discussing whether we should directly return the group state in
> the
> > > > > > response and drop the prepare assignment api. It seems that the
> > > > > > current consensus is to keep it though. If that stays like this,
> I
> > > > > > will apply your suggestion.
> > > > > >
> > > > > > Thanks,
> > > > > > David
> > > > > >
> > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com>
> > > wrote:
> > > > > > >
> > > > > > > Hi David,
> > > > > > >
> > > > > > > Thanks for your patience for answering all these questions!
> > > > > > >
> > > > > > > One more question from me:
> > > > > > > 1. When the selected consumer is going to be in client-side
> > > assigning,
> > > > > we
> > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> > > wondering if
> > > > > we
> > > > > > > could use a flag to indicate the assignment is required. The
> > > rationale
> > > > > is
> > > > > > > when returning error, we lose a chance to catch up with the
> latest
> > > > > target
> > > > > > > assignment or revoke partitions. Since we expected the record
> > > fetching
> > > > > is
> > > > > > > not blocked by the rebalance/assignment computing, we might
> want to
> > > > > allow
> > > > > > > the members to update the member assignment in each heartbeat.
> > > > > > >
> > > > > > > Luke
> > > > > > >
> > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > > <ja...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hey David,
> > > > > > > >
> > > > > > > > Thanks for the response.
> > > > > > > >
> > > > > > > > > Do you think that we could have a case where two members
> with
> > > the
> > > > > > same
> > > > > > > > static id are active?
> > > > > > > >
> > > > > > > > I was just thinking about the handoff when we spin up a new
> node
> > > > > > before the
> > > > > > > > old one completes shutdown. It's probably ok to fence the old
> > > member
> > > > > > > > abruptly. I do recall we had a few issues with this handoff
> in
> > > the
> > > > > > past. It
> > > > > > > > was due to the fact that the replaced node might have been
> the
> > > > > leader.
> > > > > > > > Since we've removed the leader notion, perhaps this kind of
> issue
> > > > > won't
> > > > > > > > resurface.
> > > > > > > >
> > > > > > > > In regard to the RE2 library. I understand the intent to pass
> > > this
> > > > > > through
> > > > > > > > as a string without any validation so that we don't need the
> > > > > > dependence on
> > > > > > > > the client. I do wonder if that will be restrictive in
> practice.
> > > For
> > > > > > > > example, if a regex changes on the client side, would we
> need to
> > > > > > validate
> > > > > > > > against the current assignment to see which partitions to
> > > continue
> > > > > > > > fetching? Or would we instead invalidate the complete current
> > > > > > assignment?
> > > > > > > > Perhaps this is a rare case and we can do the latter. There
> > > could be
> > > > > > more
> > > > > > > > situations though.
> > > > > > > >
> > > > > > > > As an alternative, I wonder if we could define a minimalistic
> > > regex
> > > > > > syntax
> > > > > > > > (a subset of RE2 say) that could map more easily to regex
> support
> > > > > that
> > > > > > > > already exists in clients? For example, in this scenario, we
> are
> > > > > > matching
> > > > > > > > topics, so we only need to be able to represent topic
> characters
> > > and
> > > > > > not
> > > > > > > > the whole smorgasbord of whitespace and unicode options.
> Would it
> > > > > make
> > > > > > > > sense to have a reduced syntax which includes only standard
> topic
> > > > > > > > characters, wildcards, character classes, and repetitions?
> > > > > > > >
> > > > > > > > Best,
> > > > > > > > Jason
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi David, thanks for all the detailed explanations. I think
> > > they
> > > > > all
> > > > > > make
> > > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > > >
> > > > > > > > > > I don't really see the benefits here because server side
> > > > > assignors
> > > > > > > > > don't have metadata at all. They only assign
> topic-partitions.
> > > They
> > > > > > > > > are not supposed to generate metadata nor to receive
> metadata
> > > from
> > > > > > the
> > > > > > > > > members.
> > > > > > > > >
> > > > > > > > > Just to clarify I was asking about the `version` of the
> > > assignor
> > > > > > (i.e. up
> > > > > > > > > to what version that the client would support), and I do
> agree
> > > we
> > > > > > would
> > > > > > > > not
> > > > > > > > > need metadata. What I have in mind is that, for some
> specific
> > > > > > built-in
> > > > > > > > > broker-assignors, e.g. rack-aware assignors, if it's
> possible
> > > that
> > > > > > in a
> > > > > > > > > newer version we would have a hierarchical rack ID string
> > > format,
> > > > > > like
> > > > > > > > > "tier1-tier2" etc, but if some client has not upgraded
> their
> > > rack
> > > > > ID
> > > > > > > > > would still be in old format. In this case, the broker then
> > > needs
> > > > > to
> > > > > > > > choose
> > > > > > > > > the old versioned assignor. I'm probably making something
> up
> > > here
> > > > > for
> > > > > > > > rack
> > > > > > > > > aware assignors, but I'm wondering if in general such an
> > > > > > "auto-downgrade"
> > > > > > > > > behavior would be needed still for broker-side assignor,
> and
> > > if yes
> > > > > > would
> > > > > > > > > "version" still be useful.
> > > > > > > > >
> > > > > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> > > called
> > > > > > once
> > > > > > > > > when the member transitions to a new epoch. This one
> contains
> > > the
> > > > > > full
> > > > > > > > > metadata provided by the client side assignor. Then,
> > > `onAssignment`
> > > > > > > > > can be called max N times where N is the number of
> partitions
> > > > > pending
> > > > > > > > > revocation by other members. Let me try to clarify this in
> the
> > > KIP.
> > > > > > > > >
> > > > > > > > > Okay, my understanding is that the calling ordering of
> these
> > > > > > callbacks
> > > > > > > > > would be like the following:
> > > > > > > > >
> > > > > > > > > ----------------------------------------
> > > > > > > > > onPartitionsRevoked();   // just once, since we do not
> really
> > > need
> > > > > > > > > to revoke incrementally.
> > > > > > > > >
> > > > > > > > > onAssignment();    // the first call, with epoch
> incremented
> > > > > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > > > > >
> > > > > > > > > onAssignment();              // the first onAssignment
> would
> > > bump
> > > > > up
> > > > > > the
> > > > > > > > > epoch, and the metadata reflected.
> > > > > > > > > onPartitionsAssigned();   // each time we get an additional
> > > > > > assignment,
> > > > > > > > we
> > > > > > > > > call onAssignment and then paired with an
> onPartitionsAssigned
> > > > > > > > > ...
> > > > > > > > > onAssignment();
> > > > > > > > > onPartitionsAssigned();   // on each of the onAssignment
> > > calls, the
> > > > > > > > encoded
> > > > > > > > > metadata would not change, only the incrementally added
> > > partitions
> > > > > be
> > > > > > > > > reflected
> > > > > > > > >
> > > > > > > > > Is that the case?
> > > > > > > > >
> > > > > > > > > I'm wondering if we would still call onAssignment just
> once,
> > > that
> > > > > > encodes
> > > > > > > > > all the assignment for this rebalance, including all the
> > > partitions
> > > > > > that
> > > > > > > > > should be assigned to the member but not yet assigned since
> > > they
> > > > > > have not
> > > > > > > > > been revoked by others. In that case the call ordering
> would
> > > be:
> > > > > > > > >
> > > > > > > > > ----------------------------------------
> > > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > > onAssignment();    // just once, with epoch incremented,
> and
> > > > > metadata
> > > > > > > > > encoded changed, the "assignment" field also reflect the
> final
> > > > > target
> > > > > > > > > assignment
> > > > > > > > > onPartitionsAssigned();   // multiple times, which
> represent
> > > > > > > > incrementally
> > > > > > > > > added partitions
> > > > > > > > > ...
> > > > > > > > > onPartitionsAssigned();
> > > > > > > > >
> > > > > > > > > The motivation from this is that, most users would only
> > > implement
> > > > > the
> > > > > > > > > rebalance callback listeners and hence we'd definitely
> need to
> > > make
> > > > > > sure
> > > > > > > > > the semantics of that does not change much, and the time
> > > > > > > > > `onPartitionsAssigned` indicate the time when the
> partitions
> > > are
> > > > > > actually
> > > > > > > > > assigned to it; while for assignors, the `onAssignment` is
> > > used to
> > > > > > > > indicate
> > > > > > > > > what decision is made regarding for this member, i.e. when
> the
> > > > > > partitions
> > > > > > > > > are decided to be given to it, but not necessarily meant
> that
> > > it
> > > > > has
> > > > > > been
> > > > > > > > > given, since that time should be determined by the time of
> > > > > > > > > `onPartitionsAssigned`. The benefits though, would be that
> > > assignor
> > > > > > > > > implementers would not need to reason which `onAssignment`
> > > would be
> > > > > > the
> > > > > > > > > last one for this epoch.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > >
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hi Guozhang,
> > > > > > > > > >
> > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has
> an
> > > Error
> > > > > > byte
> > > > > > > > > > field,
> > > > > > > > > > > while the broker-side "PartitionAssignor#Assignment"
> does
> > > not.
> > > > > > And
> > > > > > > > the
> > > > > > > > > > > rationale seems to be assuming that we should always be
> > > able to
> > > > > > do
> > > > > > > > the
> > > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > > Personally I
> > > > > > > > > think
> > > > > > > > > > > it's still potentially beneficial to add the Error
> field
> > > even
> > > > > for
> > > > > > > > > > > broker-side assignors, e.g. for some edge cases where
> some
> > > > > > subscribed
> > > > > > > > > > > topics are not recognized with the current broker's
> > > metadata.
> > > > > > What do
> > > > > > > > > you
> > > > > > > > > > > think?
> > > > > > > > > >
> > > > > > > > > > Yeah, that seems reasonable. However, I wonder if it
> would be
> > > > > > better
> > > > > > > > > > to use an exception on the server side. This is what we
> > > usually
> > > > > do
> > > > > > for
> > > > > > > > > > server side plugins. On the client side, we use a field
> > > because
> > > > > the
> > > > > > > > > > errors are not defined in advance.
> > > > > > > > > >
> > > > > > > > > > Your comment also makes me think about what we should do
> > > when the
> > > > > > > > > > server side assignor fails. I suppose that we have to
> keep
> > > the
> > > > > > current
> > > > > > > > > > assignment until a new event occurs. For instance, in
> your
> > > > > example,
> > > > > > > > > > the coordinator would have to trigger a rebalance when
> > > > > unrecognized
> > > > > > > > > > topics are available. This would be part of the metadata
> > > > > > monitoring.
> > > > > > > > > >
> > > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > > fields
> > > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > > GroupMember. I
> > > > > > > > > > agree
> > > > > > > > > > > that broker-side assignor would not need
> reason/metadata
> > > since
> > > > > > they
> > > > > > > > are
> > > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > > version?
> > > > > > E.g.
> > > > > > > > is
> > > > > > > > > > it
> > > > > > > > > > > possible that we evolve our broker-side built-in
> assignor
> > > but
> > > > > > the old
> > > > > > > > > > > versioned clients would not be able to work with the
> new
> > > > > > version, in
> > > > > > > > > > which
> > > > > > > > > > > case we need to let the broker being aware of this and
> > > upgrade
> > > > > > its
> > > > > > > > > > behavior
> > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > >
> > > > > > > > > > I don't really see the benefits here because server side
> > > > > assignors
> > > > > > > > > > don't have metadata at all. They only assign
> > > topic-partitions.
> > > > > They
> > > > > > > > > > are not supposed to generate metadata nor to receive
> metadata
> > > > > from
> > > > > > the
> > > > > > > > > > members.
> > > > > > > > > >
> > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > "GroupMember",
> > > > > > > > instead
> > > > > > > > > > of
> > > > > > > > > > > including these three fields, what about just adding
> the
> > > > > > "Metadata"
> > > > > > > > > field
> > > > > > > > > > > class which has these three fields? Also, there are two
> > > > > > "Metadata"
> > > > > > > > > > > currently in the APIs, the first is a class that
> encodes
> > > > > > > > > > > reason/version/metadata, and the second is just the
> encoded
> > > > > > metadata
> > > > > > > > > > bytes.
> > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > memberMetadata,
> > > > > > > > which
> > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> naming the
> > > > > > second
> > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > >
> > > > > > > > > > That's a good point. Let me try to rationalize this
> interface
> > > > > > based on
> > > > > > > > > > your suggestions.
> > > > > > > > > >
> > > > > > > > > > Best,
> > > > > > > > > > David
> > > > > > > > > >
> > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > Hello David,
> > > > > > > > > > >
> > > > > > > > > > > Just had a few more nit questions about the public
> APIs:
> > > > > > > > > > >
> > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has
> an
> > > Error
> > > > > > byte
> > > > > > > > > > field,
> > > > > > > > > > > while the broker-side "PartitionAssignor#Assignment"
> does
> > > not.
> > > > > > And
> > > > > > > > the
> > > > > > > > > > > rationale seems to be assuming that we should always be
> > > able to
> > > > > > do
> > > > > > > > the
> > > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > > Personally I
> > > > > > > > > think
> > > > > > > > > > > it's still potentially beneficial to add the Error
> field
> > > even
> > > > > for
> > > > > > > > > > > broker-side assignors, e.g. for some edge cases where
> some
> > > > > > subscribed
> > > > > > > > > > > topics are not recognized with the current broker's
> > > metadata.
> > > > > > What do
> > > > > > > > > you
> > > > > > > > > > > think?
> > > > > > > > > > >
> > > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > > fields
> > > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > > GroupMember. I
> > > > > > > > > > agree
> > > > > > > > > > > that broker-side assignor would not need
> reason/metadata
> > > since
> > > > > > they
> > > > > > > > are
> > > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > > version?
> > > > > > E.g.
> > > > > > > > is
> > > > > > > > > > it
> > > > > > > > > > > possible that we evolve our broker-side built-in
> assignor
> > > but
> > > > > > the old
> > > > > > > > > > > versioned clients would not be able to work with the
> new
> > > > > > version, in
> > > > > > > > > > which
> > > > > > > > > > > case we need to let the broker being aware of this and
> > > upgrade
> > > > > > its
> > > > > > > > > > behavior
> > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > >
> > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > "GroupMember",
> > > > > > > > instead
> > > > > > > > > > of
> > > > > > > > > > > including these three fields, what about just adding
> the
> > > > > > "Metadata"
> > > > > > > > > field
> > > > > > > > > > > class which has these three fields? Also, there are two
> > > > > > "Metadata"
> > > > > > > > > > > currently in the APIs, the first is a class that
> encodes
> > > > > > > > > > > reason/version/metadata, and the second is just the
> encoded
> > > > > > metadata
> > > > > > > > > > bytes.
> > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > memberMetadata,
> > > > > > > > which
> > > > > > > > > > > then has a bytebuffer field Metadata, or instead
> naming the
> > > > > > second
> > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > >
> > > > > > > > > > > Guozhang
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > > > wangguoz@gmail.com>
> > > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > > Hello David,
> > > > > > > > > > > >
> > > > > > > > > > > > Thanks for bringing this question up. I think the
> main
> > > > > > benefits as
> > > > > > > > > you
> > > > > > > > > > > > listed is 2) above if it stays; just to clarify, we
> would
> > > > > only
> > > > > > be
> > > > > > > > > able
> > > > > > > > > > to
> > > > > > > > > > > > save one round trip if the rebalance is still
> triggered
> > > by
> > > > > the
> > > > > > > > > broker;
> > > > > > > > > > if
> > > > > > > > > > > > the rebalance is triggered by the client then the
> > > > > > num.round.trips
> > > > > > > > are
> > > > > > > > > > the
> > > > > > > > > > > > same:
> > > > > > > > > > > >
> > > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > > >
> > > > > > > > > > > > T0: client decides to do a new assignment, suppose
> it has
> > > > > > already
> > > > > > > > > sent
> > > > > > > > > > a
> > > > > > > > > > > > HB and hence has to wait for it to return first since
> > > only
> > > > > one
> > > > > > > > > request
> > > > > > > > > > /
> > > > > > > > > > > > response can be inflight with the coordinator's
> socket.
> > > > > > > > > > > > T1: client receives the HB response, and then sends
> the
> > > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > request.
> > > > > > > > > > > >
> > > > > > > > > > > > In total, two round trips.
> > > > > > > > > > > >
> > > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > > >
> > > > > > > > > > > > T0: client decides to do a new assignment, suppose
> it has
> > > > > > already
> > > > > > > > > sent
> > > > > > > > > > a
> > > > > > > > > > > > HB and hence has to wait for it to return first since
> > > only
> > > > > one
> > > > > > > > > request
> > > > > > > > > > /
> > > > > > > > > > > > response can be inflight with the coordinator's
> socket.
> > > > > > > > > > > > T1: client receives the HB response, and then sends
> the
> > > new
> > > > > HB
> > > > > > > > > request
> > > > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > > > T2: the HB response with the optional member metadata
> > > map is
> > > > > > > > > returned.
> > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > request.
> > > > > > > > > > > >
> > > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > > >
> > > > > > > > > > > > -----------------------------
> > > > > > > > > > > >
> > > > > > > > > > > > So to complete the full picture here, we'd need to
> modify
> > > > > both
> > > > > > HB
> > > > > > > > > > request
> > > > > > > > > > > > and response so that the client can also indicate a
> new
> > > > > > rebalance
> > > > > > > > via
> > > > > > > > > > the
> > > > > > > > > > > > HB request as well, right?
> > > > > > > > > > > >
> > > > > > > > > > > > Assuming all above is true, I think it's okay to
> merge
> > > the
> > > > > > > > > > > > GroupPrepareAssignment into HB given that we can
> make the
> > > > > > > > additional
> > > > > > > > > > fields
> > > > > > > > > > > > encoding the full member (subscription) metadata and
> > > topic
> > > > > > metadata
> > > > > > > > > as
> > > > > > > > > > > > optional fields.
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > Guozhang
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > >> Hi all,
> > > > > > > > > > > >>
> > > > > > > > > > > >> During an offline conversation, someone asked why we
> > > need
> > > > > the
> > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested
> that we
> > > > > could
> > > > > > > > > instead
> > > > > > > > > > > >> provide the group state in the heartbeat response.
> This
> > > has
> > > > > a
> > > > > > few
> > > > > > > > > > > >> advantages: 1) it does not require using a special
> error
> > > > > code
> > > > > > to
> > > > > > > > > > > >> signal that a new assignment is required as the
> signal
> > > would
> > > > > > be
> > > > > > > > the
> > > > > > > > > > > >> provided group state; 2) it removes one round trip
> when
> > > a
> > > > > > client
> > > > > > > > > side
> > > > > > > > > > > >> assignor is used. The downside is that it makes the
> > > > > heartbeat
> > > > > > > > > > > >> response's definition quite large. I recall that I
> went
> > > with
> > > > > > the
> > > > > > > > > > > >> current approach due to this.
> > > > > > > > > > > >>
> > > > > > > > > > > >> Providing the group state in the heartbeat response
> is
> > > > > > appealing.
> > > > > > > > > What
> > > > > > > > > > > >> do you guys think?
> > > > > > > > > > > >>
> > > > > > > > > > > >> Best,
> > > > > > > > > > > >> David
> > > > > > > > > > > >>
> > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > > > djacot@confluent.io>
> > > > > > > > > > wrote:
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > 1. I have added a reference to the relevant
> chapter
> > > > > instead
> > > > > > of
> > > > > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you are
> referring
> > > to
> > > > > is
> > > > > > > > about
> > > > > > > > > > when
> > > > > > > > > > > >> > a rebalance should be triggered for the
> non-upgraded
> > > > > members
> > > > > > > > using
> > > > > > > > > > the
> > > > > > > > > > > >> > old protocol. The section mentions that a
> rebalance
> > > must
> > > > > be
> > > > > > > > > > triggered
> > > > > > > > > > > >> > when a new assignment is installed. This implies
> that
> > > the
> > > > > > group
> > > > > > > > > > epoch
> > > > > > > > > > > >> > was updated either by a native member or a
> > > non-upgraded
> > > > > > member.
> > > > > > > > > For
> > > > > > > > > > > >> > the latter, the JoinGroup request would be the
> > > trigger. I
> > > > > > have
> > > > > > > > > > added a
> > > > > > > > > > > >> > reference to the relevant chapter in the
> "JoinGroup
> > > > > > Handling"
> > > > > > > > > > section
> > > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > Thanks,
> > > > > > > > > > > >> > David
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > > > > wangguoz@gmail.com
> > > > > > > > > >
> > > > > > > > > > > >> wrote:
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Hello David,
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Alright I think that's sufficient. Just to make
> that
> > > > > > clear in
> > > > > > > > > the
> > > > > > > > > > doc,
> > > > > > > > > > > >> > > could we update:
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > 1) the heartbeat request handling section,
> stating
> > > when
> > > > > > > > > > coordinator
> > > > > > > > > > > >> will
> > > > > > > > > > > >> > > trigger rebalance based on the HB's member
> metadata
> > > /
> > > > > > reason?
> > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to include
> what
> > > we
> > > > > > > > described
> > > > > > > > > > in
> > > > > > > > > > > >> "Group
> > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Guozhang
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > wrote:
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > I thought that the assignor will always be
> > > consulted
> > > > > > when
> > > > > > > > the
> > > > > > > > > > next
> > > > > > > > > > > >> > > > heartbeat request is constructed. In other
> words,
> > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be called
> for
> > > every
> > > > > > > > > heartbeat.
> > > > > > > > > > > >> This
> > > > > > > > > > > >> > > > gives the opportunity for the assignor to
> enforce
> > > a
> > > > > > > > rebalance
> > > > > > > > > by
> > > > > > > > > > > >> > > > setting the reason to a non-zero value or by
> > > changing
> > > > > > the
> > > > > > > > > > bytes. Do
> > > > > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > > > > concerned by
> > > > > > > > > the
> > > > > > > > > > > >> delay?
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > Best,
> > > > > > > > > > > >> > > > David
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > >> wrote:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > >> > > > > In this case, a new assignment is triggered
> by
> > > the
> > > > > > client
> > > > > > > > > side
> > > > > > > > > > > >> > > > > assignor. When constructing the HB, the
> consumer
> > > > > will
> > > > > > > > always
> > > > > > > > > > > >> consult
> > > > > > > > > > > >> > > > > the client side assignor and propagate the
> > > > > > information to
> > > > > > > > > the
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > coordinator. In other words, we don't expect
> > > users
> > > > > to
> > > > > > call
> > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > As I looked at the current
> PartitionAssignor's
> > > > > > interface,
> > > > > > > > we
> > > > > > > > > > > >> actually do
> > > > > > > > > > > >> > > > > not have a way yet to instruct how to
> construct
> > > the
> > > > > > next
> > > > > > > > HB
> > > > > > > > > > > >> request, e.g.
> > > > > > > > > > > >> > > > > when the assignor wants to enforce a new
> > > rebalance
> > > > > > with a
> > > > > > > > > new
> > > > > > > > > > > >> assignment,
> > > > > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > > > > PartitionAssignor
> > > > > > > > > > to
> > > > > > > > > > > >> indicate
> > > > > > > > > > > >> > > > > the next HB telling broker about so. WDYT
> about
> > > > > adding
> > > > > > > > such
> > > > > > > > > an
> > > > > > > > > > > >> API on the
> > > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > I have updated the KIP to include your
> > > feedback. I
> > > > > > have
> > > > > > > > > also
> > > > > > > > > > > >> tried to
> > > > > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David
> Jacot <
> > > > > > > > > > djacot@confluent.io
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me start
> by
> > > > > > answering
> > > > > > > > your
> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > > > inline and I will update the KIP next
> week.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > > benefits
> > > > > > of
> > > > > > > > the
> > > > > > > > > > KIP
> > > > > > > > > > > >> seem to
> > > > > > > > > > > >> > > > be
> > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> efficient
> > > > > > support of
> > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > I would also add that the KIP removes
> the
> > > global
> > > > > > sync
> > > > > > > > > > barrier
> > > > > > > > > > > >> in the
> > > > > > > > > > > >> > > > > > > protocol which is essential to improve
> group
> > > > > > stability
> > > > > > > > > and
> > > > > > > > > > > >> > > > > > > scalability, and the KIP also
> simplifies the
> > > > > > client by
> > > > > > > > > > moving
> > > > > > > > > > > >> most of
> > > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton.
> Do we
> > > plan
> > > > > > to
> > > > > > > > > > support
> > > > > > > > > > > >> rolling
> > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > consumers?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Definitely. The group coordinator will
> use
> > > the
> > > > > > > > assignor
> > > > > > > > > > used
> > > > > > > > > > > >> by a
> > > > > > > > > > > >> > > > > > > majority of the members. This allows the
> > > group
> > > > > to
> > > > > > move
> > > > > > > > > > from
> > > > > > > > > > > >> one
> > > > > > > > > > > >> > > > > > > assignor to another by a roll. This is
> > > explained
> > > > > > in
> > > > > > > > the
> > > > > > > > > > > >> Assignor
> > > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > > whether
> > > > > > it's
> > > > > > > > > > > >> required in
> > > > > > > > > > > >> > > > every
> > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> needs to
> > > be
> > > > > > filled?
> > > > > > > > > For
> > > > > > > > > > > >> example,
> > > > > > > > > > > >> > > > it's
> > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> needs to
> > > be
> > > > > > filled.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > The client is expected to set those
> fields
> > > in
> > > > > > case of
> > > > > > > > a
> > > > > > > > > > > >> connection
> > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields
> have
> > > > > > changed
> > > > > > > > > since
> > > > > > > > > > > >> the last
> > > > > > > > > > > >> > > > > > > HB. The server populates those fields as
> > > long as
> > > > > > the
> > > > > > > > > > member
> > > > > > > > > > > >> is not
> > > > > > > > > > > >> > > > > > > fully reconciled - the member should
> > > acknowledge
> > > > > > that
> > > > > > > > it
> > > > > > > > > > has
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > expected epoch and assignment. I will
> > > clarify
> > > > > > this in
> > > > > > > > > the
> > > > > > > > > > KIP.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol,
> the
> > > rack
> > > > > > > > > affinity
> > > > > > > > > > > >> between the
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > and the broker is only considered
> during
> > > > > > fetching,
> > > > > > > > but
> > > > > > > > > > not
> > > > > > > > > > > >> during
> > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes,
> once
> > > the
> > > > > > > > > assignment
> > > > > > > > > > is
> > > > > > > > > > > >> made,
> > > > > > > > > > > >> > > > there
> > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> because
> > > no
> > > > > > replicas
> > > > > > > > > of
> > > > > > > > > > > >> assigned
> > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > >> > > > > > > > are close to the member. I am
> wondering
> > > if we
> > > > > > should
> > > > > > > > > use
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > > GroupMember.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > That's an interesting idea. I don't see
> any
> > > > > issue
> > > > > > with
> > > > > > > > > > adding
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > rack
> > > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > > useful to
> > > > > > know
> > > > > > > > how
> > > > > > > > > > busy
> > > > > > > > > > > >> a group
> > > > > > > > > > > >> > > > > > > > coordinator is. By moving the event
> loop
> > > > > model,
> > > > > > it
> > > > > > > > > seems
> > > > > > > > > > > >> that we
> > > > > > > > > > > >> > > > could
> > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of
> the
> > > time
> > > > > > the
> > > > > > > > > event
> > > > > > > > > > > >> loop is
> > > > > > > > > > > >> > > > doing
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > That's a great idea. I will add it.
> Thanks.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> coordinator
> > > > > > failover
> > > > > > > > > > > >> handling? For
> > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > >> > > > > > > > does it need to trigger the check if
> any
> > > group
> > > > > > with
> > > > > > > > > the
> > > > > > > > > > > >> wildcard
> > > > > > > > > > > >> > > > > > > > subscription now has a new matching
> topic?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Sure. When the new group coordinator
> takes
> > > over,
> > > > > > it
> > > > > > > > has
> > > > > > > > > > to:
> > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a client
> side
> > > > > > assignor
> > > > > > > > is
> > > > > > > > > > used.
> > > > > > > > > > > >> We
> > > > > > > > > > > >> > > > don't
> > > > > > > > > > > >> > > > > > > store the information about the member
> > > selected
> > > > > > to run
> > > > > > > > > the
> > > > > > > > > > > >> assignment
> > > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > > > > wildcard
> > > > > > > > > > > >> subscriptions, and
> > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> Could we
> > > > > > document
> > > > > > > > > what
> > > > > > > > > > > >> the epoch
> > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch
> in
> > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > > one in
> > > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue
> and
> > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Sure. I will clarify that but it is
> always
> > > the
> > > > > > latest
> > > > > > > > > > group
> > > > > > > > > > > >> epoch.
> > > > > > > > > > > >> > > > > > > When the group state is updated, the
> group
> > > epoch
> > > > > > is
> > > > > > > > > > bumped so
> > > > > > > > > > > >> we use
> > > > > > > > > > > >> > > > > > > that one for all the change records
> related
> > > to
> > > > > the
> > > > > > > > > update.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > > that
> > > > > the
> > > > > > > > > > following
> > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > possible
> > > > > for
> > > > > > a
> > > > > > > > > member
> > > > > > > > > > > >> not to
> > > > > > > > > > > >> > > > get any
> > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > That's right. Here I meant that the
> members
> > > > > > provided
> > > > > > > > by
> > > > > > > > > > the
> > > > > > > > > > > >> assignor
> > > > > > > > > > > >> > > > > > > in the assignment must exist in the
> group.
> > > The
> > > > > > > > assignor
> > > > > > > > > > can
> > > > > > > > > > > >> not make
> > > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > > member
> > > > > > epoch
> > > > > > > > > > equals to
> > > > > > > > > > > >> 0":
> > > > > > > > > > > >> > > > When
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > > would be
> > > > > > used?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > > > > partitions and
> > > > > > > > > > > >> rejoins when
> > > > > > > > > > > >> > > > it
> > > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error.
> In
> > > this
> > > > > > case,
> > > > > > > > > the
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > coordinator will have removed the member
> > > from
> > > > > the
> > > > > > > > group.
> > > > > > > > > > The
> > > > > > > > > > > >> member
> > > > > > > > > > > >> > > > > > > can rejoin the group with the same
> member
> > > id but
> > > > > > with
> > > > > > > > 0
> > > > > > > > > as
> > > > > > > > > > > >> epoch. Let
> > > > > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have
> the
> > > > > ability
> > > > > > to
> > > > > > > > > > trigger a
> > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > >> > > > > > > > by either providing a non-zero reason
> or
> > > by
> > > > > > updating
> > > > > > > > > the
> > > > > > > > > > > >> assignor
> > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > conflicting
> > > > > > with
> > > > > > > > the
> > > > > > > > > > > >> deprecation
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > In this case, a new assignment is
> triggered
> > > by
> > > > > the
> > > > > > > > > client
> > > > > > > > > > side
> > > > > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> > > consumer
> > > > > > will
> > > > > > > > > > always
> > > > > > > > > > > >> consult
> > > > > > > > > > > >> > > > > > > the client side assignor and propagate
> the
> > > > > > information
> > > > > > > > > to
> > > > > > > > > > the
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > coordinator. In other words, we don't
> expect
> > > > > > users to
> > > > > > > > > call
> > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> nice.
> > > But
> > > > > the
> > > > > > > > > section
> > > > > > > > > > > >> seems to
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> epoch
> > > 2, B
> > > > > > > > > > immediately
> > > > > > > > > > > >> gets into
> > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which
> seems
> > > > > > > > incorrect.
> > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> epoch
> > > 3, C
> > > > > > seems
> > > > > > > > to
> > > > > > > > > > get
> > > > > > > > > > > >> into
> > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > > still
> > > > > > has A -
> > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do
> we
> > > > > support
> > > > > > the
> > > > > > > > > > upgrade
> > > > > > > > > > > >> from
> > > > > > > > > > > >> > > > any
> > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > We will support upgrading from the
> consumer
> > > > > > protocol
> > > > > > > > > > version
> > > > > > > > > > > >> 3,
> > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > > > > implemented
> > > > > > yet
> > > > > > > > so
> > > > > > > > > > the
> > > > > > > > > > > >> earliest
> > > > > > > > > > > >> > > > > > > version is unknown at the moment. This
> is
> > > > > > explained in
> > > > > > > > > the
> > > > > > > > > > > >> migration
> > > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I
> will
> > > > > > update the
> > > > > > > > > KIP
> > > > > > > > > > > >> based on
> > > > > > > > > > > >> > > > it
> > > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > > >> > > > > > > David
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > > benefits
> > > > > > of
> > > > > > > > the
> > > > > > > > > > KIP
> > > > > > > > > > > >> seem to
> > > > > > > > > > > >> > > > be
> > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more
> efficient
> > > > > > support of
> > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton.
> Do we
> > > plan
> > > > > > to
> > > > > > > > > > support
> > > > > > > > > > > >> rolling
> > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > consumers?
> > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > > whether
> > > > > > it's
> > > > > > > > > > > >> required in
> > > > > > > > > > > >> > > > every
> > > > > > > > > > > >> > > > > > > > request or the scenarios when it
> needs to
> > > be
> > > > > > filled?
> > > > > > > > > For
> > > > > > > > > > > >> example,
> > > > > > > > > > > >> > > > it's
> > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions
> needs to
> > > be
> > > > > > filled.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol,
> the
> > > rack
> > > > > > > > > affinity
> > > > > > > > > > > >> between the
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > and the broker is only considered
> during
> > > > > > fetching,
> > > > > > > > but
> > > > > > > > > > not
> > > > > > > > > > > >> during
> > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes,
> once
> > > the
> > > > > > > > > assignment
> > > > > > > > > > is
> > > > > > > > > > > >> made,
> > > > > > > > > > > >> > > > there
> > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > >> > > > > > > > no opportunity for read affinity
> because
> > > no
> > > > > > replicas
> > > > > > > > > of
> > > > > > > > > > > >> assigned
> > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > >> > > > > > > > are close to the member. I am
> wondering
> > > if we
> > > > > > should
> > > > > > > > > use
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > > GroupMember.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > > useful to
> > > > > > know
> > > > > > > > how
> > > > > > > > > > busy
> > > > > > > > > > > >> a group
> > > > > > > > > > > >> > > > > > > > coordinator is. By moving the event
> loop
> > > > > model,
> > > > > > it
> > > > > > > > > seems
> > > > > > > > > > > >> that we
> > > > > > > > > > > >> > > > could
> > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of
> the
> > > time
> > > > > > the
> > > > > > > > > event
> > > > > > > > > > > >> loop is
> > > > > > > > > > > >> > > > doing
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 33. Could we add a section on
> coordinator
> > > > > > failover
> > > > > > > > > > > >> handling? For
> > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > >> > > > > > > > does it need to trigger the check if
> any
> > > group
> > > > > > with
> > > > > > > > > the
> > > > > > > > > > > >> wildcard
> > > > > > > > > > > >> > > > > > > > subscription now has a new matching
> topic?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue:
> Could we
> > > > > > document
> > > > > > > > > what
> > > > > > > > > > > >> the epoch
> > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch
> in
> > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > > one in
> > > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue
> and
> > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > > that
> > > > > the
> > > > > > > > > > following
> > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > possible
> > > > > for
> > > > > > a
> > > > > > > > > member
> > > > > > > > > > > >> not to
> > > > > > > > > > > >> > > > get any
> > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > > member
> > > > > > epoch
> > > > > > > > > > equals to
> > > > > > > > > > > >> 0":
> > > > > > > > > > > >> > > > When
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > > would be
> > > > > > used?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have
> the
> > > > > ability
> > > > > > to
> > > > > > > > > > trigger a
> > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > >> > > > > > > > by either providing a non-zero reason
> or
> > > by
> > > > > > updating
> > > > > > > > > the
> > > > > > > > > > > >> assignor
> > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > conflicting
> > > > > > with
> > > > > > > > the
> > > > > > > > > > > >> deprecation
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are
> nice.
> > > But
> > > > > the
> > > > > > > > > section
> > > > > > > > > > > >> seems to
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to
> epoch
> > > 2, B
> > > > > > > > > > immediately
> > > > > > > > > > > >> gets into
> > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which
> seems
> > > > > > > > incorrect.
> > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to
> epoch
> > > 3, C
> > > > > > seems
> > > > > > > > to
> > > > > > > > > > get
> > > > > > > > > > > >> into
> > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > > still
> > > > > > has A -
> > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do
> we
> > > > > support
> > > > > > the
> > > > > > > > > > upgrade
> > > > > > > > > > > >> from
> > > > > > > > > > > >> > > > any
> > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David
> > > Jacot
> > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > > re-implement the
> > > > > > > > > > coordinator
> > > > > > > > > > > >> in
> > > > > > > > > > > >> > > > Java. I
> > > > > > > > > > > >> > > > > > > > > discussed this offline with a few
> folks
> > > and
> > > > > > folks
> > > > > > > > > are
> > > > > > > > > > > >> concerned
> > > > > > > > > > > >> > > > that
> > > > > > > > > > > >> > > > > > > > > we could introduce many regressions
> in
> > > the
> > > > > old
> > > > > > > > > > protocol
> > > > > > > > > > > >> if we do
> > > > > > > > > > > >> > > > so.
> > > > > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > > > > statement
> > > > > > > > from
> > > > > > > > > > the
> > > > > > > > > > > >> KIP. It
> > > > > > > > > > > >> > > > is an
> > > > > > > > > > > >> > > > > > > > > implementation detail after all so
> it
> > > does
> > > > > not
> > > > > > > > have
> > > > > > > > > > to be
> > > > > > > > > > > >> > > > decided at
> > > > > > > > > > > >> > > > > > > > > this stage. We will likely start by
> > > trying
> > > > > to
> > > > > > > > > > refactor the
> > > > > > > > > > > >> > > > current
> > > > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM
> David
> > > Jacot
> > > > > <
> > > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine
> are:
> > > > > > "Empty,
> > > > > > > > > > assigning,
> > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> Group
> > > States
> > > > > > > > > section,
> > > > > > > > > > > >> right?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > This sentence does not refer to
> those
> > > > > group
> > > > > > > > states
> > > > > > > > > > but
> > > > > > > > > > > >> rather
> > > > > > > > > > > >> > > > to a
> > > > > > > > > > > >> > > > > > > > > > state machine replication (SMR).
> This
> > > > > > refers to
> > > > > > > > > the
> > > > > > > > > > > >> entire
> > > > > > > > > > > >> > > > state of
> > > > > > > > > > > >> > > > > > > > > > group coordinator which is
> replicated
> > > via
> > > > > > the
> > > > > > > > log
> > > > > > > > > > > >> layer. I will
> > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> state
> > > > > machine
> > > > > > is
> > > > > > > > > > modelled
> > > > > > > > > > > >> as an
> > > > > > > > > > > >> > > > event
> > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > The idea is to follow a model
> similar
> > > to
> > > > > > the new
> > > > > > > > > > quorum
> > > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > > >> > > > > > > > > > will have N threads to process
> events.
> > > > > Each
> > > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > > >> > > > > > > > > > partition is assigned to a unique
> > > thread
> > > > > > and all
> > > > > > > > > the
> > > > > > > > > > > >> events
> > > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> > > processed
> > > > > by
> > > > > > this
> > > > > > > > > > > >> thread. This
> > > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > > >> > > > > > > > > > concurrency and will enable us to
> do
> > > > > > simulation
> > > > > > > > > > testing
> > > > > > > > > > > >> for the
> > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> machine
> > > per
> > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> consumer
> > > group"
> > > > > > owned
> > > > > > > > > by
> > > > > > > > > > a
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > >> > > > > > > > > > > example, if one group
> coordinator
> > > owns 2
> > > > > > > > > consumer
> > > > > > > > > > > >> groups, and
> > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > > have 1
> > > > > > state
> > > > > > > > > > machine
> > > > > > > > > > > >> for it,
> > > > > > > > > > > >> > > > or
> > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> > > there, I
> > > > > > > > think.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > "*group.coordinator.threads"
> > > > > > > > *is
> > > > > > > > > > the
> > > > > > > > > > > >> number
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > >> > > > > > > > > > > to run the state machines. But
> I'm
> > > > > > wondering
> > > > > > > > if
> > > > > > > > > > the
> > > > > > > > > > > >> purpose
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> each
> > > > > consumer
> > > > > > > > group
> > > > > > > > > > (or
> > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > computation,
> > > > > > why
> > > > > > > > > > should we
> > > > > > > > > > > >> need
> > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an
> > > ability to
> > > > > > shard
> > > > > > > > > the
> > > > > > > > > > > >> > > > processing as
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> timeout,
> > > why
> > > > > > does
> > > > > > > > the
> > > > > > > > > > > >> default
> > > > > > > > > > > >> > > > session
> > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > max(60s)? I
> > > > > > > > thought
> > > > > > > > > > the
> > > > > > > > > > > >> min/max
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > > bound
> > > > > of
> > > > > > it,
> > > > > > > > > no?
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> group.consumer.session.timeout.ms
> > > int
> > > > > > 30s The
> > > > > > > > > > > >> timeout to
> > > > > > > > > > > >> > > > detect
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > > group
> > > > > > > > protocol.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.min.session.timeout.ms
> > > > > > int 45s
> > > > > > > > > The
> > > > > > > > > > > >> minimum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.max.session.timeout.ms
> > > > > > int 60s
> > > > > > > > > The
> > > > > > > > > > > >> maximum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The
> default
> > > > > > session
> > > > > > > > > > timeout
> > > > > > > > > > > >> should
> > > > > > > > > > > >> > > > be 45s
> > > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > assignor
> > > > > are
> > > > > > > > > [range,
> > > > > > > > > > > >> uniform],
> > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> assignor.
> > > I'd
> > > > > > like to
> > > > > > > > > > know
> > > > > > > > > > > >> why not
> > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> choose
> > > > > uniform
> > > > > > > > > assignor
> > > > > > > > > > > >> (former
> > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > >> > > > > > > > > > > for better evenly distribution.
> Any
> > > > > other
> > > > > > > > reason
> > > > > > > > > > we
> > > > > > > > > > > >> choose
> > > > > > > > > > > >> > > > range
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List
> range,
> > > > > > uniform
> > > > > > > > The
> > > > > > > > > > > >> server side
> > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > The order on the server side has
> no
> > > > > > influence
> > > > > > > > > > because
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > client
> > > > > > > > > > > >> > > > > > must
> > > > > > > > > > > >> > > > > > > > > > chose the selector that he wants
> to
> > > use.
> > > > > > There
> > > > > > > > is
> > > > > > > > > no
> > > > > > > > > > > >> default
> > > > > > > > > > > >> > > > in the
> > > > > > > > > > > >> > > > > > > > > > current proposal. If the assignor
> is
> > > not
> > > > > > > > specified
> > > > > > > > > > by
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > client,
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > request is rejected. The default
> > > client
> > > > > > value
> > > > > > > > for
> > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is
> `uniform`
> > > > > though.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > Thanks for your very good
> comments,
> > > Luke.
> > > > > I
> > > > > > hope
> > > > > > > > > > that my
> > > > > > > > > > > >> > > > answers
> > > > > > > > > > > >> > > > > > help
> > > > > > > > > > > >> > > > > > > > > > to clarify things. I will update
> the
> > > KIP
> > > > > as
> > > > > > well
> > > > > > > > > > based
> > > > > > > > > > > >> on your
> > > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM
> Luke
> > > Chen
> > > > > <
> > > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator
> section, you
> > > > > > > > mentioned:
> > > > > > > > > > > >> > > > > > > > > > > > The new group coordinator will
> > > have a
> > > > > > state
> > > > > > > > > > machine
> > > > > > > > > > > >> per
> > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions,
> > > where
> > > > > > each
> > > > > > > > > state
> > > > > > > > > > > >> machine is
> > > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > > >> > > > > > > > > > > event loop. Those state machines
> > > will be
> > > > > > > > > executed
> > > > > > > > > > in
> > > > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads*
> threads.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine
> are:
> > > > > > "Empty,
> > > > > > > > > > assigning,
> > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer
> Group
> > > States
> > > > > > > > > section,
> > > > > > > > > > > >> right?
> > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each
> state
> > > > > machine
> > > > > > is
> > > > > > > > > > modelled
> > > > > > > > > > > >> as an
> > > > > > > > > > > >> > > > event
> > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state
> machine
> > > per
> > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per
> consumer
> > > group"
> > > > > > owned
> > > > > > > > > by
> > > > > > > > > > a
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > >> > > > > > > > > > > example, if one group
> coordinator
> > > owns 2
> > > > > > > > > consumer
> > > > > > > > > > > >> groups, and
> > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > > have 1
> > > > > > state
> > > > > > > > > > machine
> > > > > > > > > > > >> for it,
> > > > > > > > > > > >> > > > or
> > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > "*group.coordinator.threads"
> > > > > > > > *is
> > > > > > > > > > the
> > > > > > > > > > > >> number
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > >> > > > > > > > > > > to run the state machines. But
> I'm
> > > > > > wondering
> > > > > > > > if
> > > > > > > > > > the
> > > > > > > > > > > >> purpose
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of
> each
> > > > > consumer
> > > > > > > > group
> > > > > > > > > > (or
> > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > computation,
> > > > > > why
> > > > > > > > > > should we
> > > > > > > > > > > >> need
> > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2. For the default value in the
> new
> > > > > > configs:
> > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session
> timeout,
> > > why
> > > > > > does
> > > > > > > > the
> > > > > > > > > > > >> default
> > > > > > > > > > > >> > > > session
> > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > max(60s)? I
> > > > > > > > thought
> > > > > > > > > > the
> > > > > > > > > > > >> min/max
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > > bound
> > > > > of
> > > > > > it,
> > > > > > > > > no?
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> group.consumer.session.timeout.ms
> > > int
> > > > > > 30s The
> > > > > > > > > > > >> timeout to
> > > > > > > > > > > >> > > > detect
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > > group
> > > > > > > > protocol.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.min.session.timeout.ms
> > > > > > int 45s
> > > > > > > > > The
> > > > > > > > > > > >> minimum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.max.session.timeout.ms
> > > > > > int 60s
> > > > > > > > > The
> > > > > > > > > > > >> maximum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > assignor
> > > > > are
> > > > > > > > > [range,
> > > > > > > > > > > >> uniform],
> > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > >> > > > > > > > > > > we'll default to "range"
> assignor.
> > > I'd
> > > > > > like to
> > > > > > > > > > know
> > > > > > > > > > > >> why not
> > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > >> > > > > > > > > > > thought usually users will
> choose
> > > > > uniform
> > > > > > > > > assignor
> > > > > > > > > > > >> (former
> > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > >> > > > > > > > > > > for better evenly distribution.
> Any
> > > > > other
> > > > > > > > reason
> > > > > > > > > > we
> > > > > > > > > > > >> choose
> > > > > > > > > > > >> > > > range
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List
> range,
> > > > > > uniform
> > > > > > > > The
> > > > > > > > > > > >> server side
> > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM
> Luke
> > > > > Chen
> > > > > > <
> > > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts about
> Kafka
> > > > > Connect
> > > > > > > > > > > >> integrating with
> > > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > > >> > > > > > > > > > > > think we should have a
> separate
> > > > > > discussion
> > > > > > > > > > thread
> > > > > > > > > > > >> for the
> > > > > > > > > > > >> > > > Kafka
> > > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect
> > > With
> > > > > New
> > > > > > > > > Consumer
> > > > > > > > > > > >> Rebalance
> > > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > > >> > > > > > > > > > > > and let this discussion thread
> > > focus
> > > > > on
> > > > > > > > > consumer
> > > > > > > > > > > >> rebalance
> > > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31
> PM
> > > Sagar
> > > > > <
> > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for
> the
> > > > > > feedback.
> > > > > > > > > > Looks
> > > > > > > > > > > >> like
> > > > > > > > > > > >> > > > there's
> > > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs for
> Connect.
> > > I
> > > > > > would
> > > > > > > > > > revisit
> > > > > > > > > > > >> the doc
> > > > > > > > > > > >> > > > and
> > > > > > > > > > > >> > > > > > see
> > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11
> PM
> > > David
> > > > > > Jacot
> > > > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback
> and the
> > > > > > document.
> > > > > > > > > > That's
> > > > > > > > > > > >> really
> > > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me
> that
> > > both
> > > > > > Connect
> > > > > > > > > > and the
> > > > > > > > > > > >> > > > Consumer
> > > > > > > > > > > >> > > > > > could
> > > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > > >> > > > > > > > > > > >> > the same underlying
> "engine".
> > > The
> > > > > > main
> > > > > > > > > > > >> difference is
> > > > > > > > > > > >> > > > that
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> > > members
> > > > > > > > whereas
> > > > > > > > > > > >> Connect
> > > > > > > > > > > >> > > > assigns
> > > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to
> move
> > > > > > forward:
> > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new
> proposed
> > > APIs
> > > > > to
> > > > > > > > > support
> > > > > > > > > > > >> different
> > > > > > > > > > > >> > > > > > resource
> > > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks,
> > > etc.); or
> > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated
> APIs
> > > for
> > > > > > Connect.
> > > > > > > > > The
> > > > > > > > > > > >> dedicated
> > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > > > > different on
> > > > > > > > > the
> > > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> > > engine
> > > > > > on the
> > > > > > > > > > > >> coordinator
> > > > > > > > > > > >> > > > side.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean towards
> 2)
> > > > > because
> > > > > > I am
> > > > > > > > > > not a
> > > > > > > > > > > >> fan of
> > > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> > > purposes.
> > > > > > That
> > > > > > > > > being
> > > > > > > > > > > >> said, I am
> > > > > > > > > > > >> > > > not
> > > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an
> elegant
> > > way to
> > > > > > do
> > > > > > > > it.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > I think that we can
> continue to
> > > > > > discuss
> > > > > > > > it
> > > > > > > > > > here
> > > > > > > > > > > >> for now
> > > > > > > > > > > >> > > > in
> > > > > > > > > > > >> > > > > > order
> > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> > > compatible
> > > > > > with
> > > > > > > > > what
> > > > > > > > > > we
> > > > > > > > > > > >> will do
> > > > > > > > > > > >> > > > for
> > > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at
> 2:41 PM
> > > > > David
> > > > > > > > Jacot
> > > > > > > > > <
> > > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation.
> I
> > > will
> > > > > go
> > > > > > > > > through
> > > > > > > > > > and
> > > > > > > > > > > >> address
> > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming days.
> Thanks
> > > for
> > > > > your
> > > > > > > > > > feedback.
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at
> 10:05
> > > PM
> > > > > > Gregory
> > > > > > > > > > Harris
> > > > > > > > > > > >> <
> > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP,
> it's
> > > > > > wonderful to
> > > > > > > > > see
> > > > > > > > > > > >> > > > cooperative
> > > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> > > Alternatives'
> > > > > > > > section
> > > > > > > > > > > >> describes how
> > > > > > > > > > > >> > > > > > member
> > > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step with
> the
> > > group
> > > > > > epoch
> > > > > > > > > and
> > > > > > > > > > > >> assignment
> > > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good
> idea
> > > for
> > > > > the
> > > > > > > > > reasons
> > > > > > > > > > > >> described
> > > > > > > > > > > >> > > > in
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is
> incrementally
> > > > > > assigning
> > > > > > > > > > > >> partitions to a
> > > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each
> incremental
> > > > > > > > assignment
> > > > > > > > > > use?
> > > > > > > > > > > >> Are
> > > > > > > > > > > >> > > > member
> > > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch
> can
> > > > > > correspond
> > > > > > > > to
> > > > > > > > > > > >> multiple
> > > > > > > > > > > >> > > > > > different
> > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's
> > > 'Reason'
> > > > > > field
> > > > > > > > > > opaque
> > > > > > > > > > > >> to the
> > > > > > > > > > > >> > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> > > client-side
> > > > > > > > assignor
> > > > > > > > > > > >> > > > implementations
> > > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how
> is
> > > its
> > > > > > common
> > > > > > > > > > meaning
> > > > > > > > > > > >> agreed
> > > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct
> > > Reason
> > > > > > field
> > > > > > > > over
> > > > > > > > > > > >> including
> > > > > > > > > > > >> > > > such
> > > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> > > included in
> > > > > > the
> > > > > > > > > KIP:
> > > > > > > > > > > >> "Thanks to
> > > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > > > > entirely
> > > > > > > > driven
> > > > > > > > > > by
> > > > > > > > > > > >> the group
> > > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > > > > responsible for
> > > > > > > > > > > >> maintaining
> > > > > > > > > > > >> > > > any
> > > > > > > > > > > >> > > > > > state
> > > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions."
> Does
> > > this
> > > > > > mean
> > > > > > > > > > that the
> > > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > > > > non-Metadata
> > > > > > > > > state
> > > > > > > > > > > >> (such as
> > > > > > > > > > > >> > > > > > partition
> > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics,
> config
> > > > > > topics,
> > > > > > > > > etc),
> > > > > > > > > > or
> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > additional
> > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be
> used?
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > > > > separate
> > > > > > > > > classes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > > >> > > > > >
> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > > >> > > >
> > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly.
> Is
> > > it
> > > > > > possible
> > > > > > > > > for
> > > > > > > > > > > >> these two
> > > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would
> serve
> > > to
> > > > > > promote
> > > > > > > > > > feature
> > > > > > > > > > > >> parity
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors,
> and
> > > > > would
> > > > > > also
> > > > > > > > > > > >> facilitate
> > > > > > > > > > > >> > > > > > operational
> > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > > > > example,
> > > > > > if a
> > > > > > > > > > > >> server-side
> > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a
> patch,
> > > > > > deploying
> > > > > > > > > the
> > > > > > > > > > > >> patched
> > > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching one
> consumer
> > > > > group
> > > > > > to a
> > > > > > > > > > > >> client-side
> > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> > > patching
> > > > > all
> > > > > > of
> > > > > > > > the
> > > > > > > > > > > >> brokers.
> > > > > > > > > > > >> > > > With
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct
> APIs, a
> > > > > > non-trivial
> > > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the
> two
> > > APIs
> > > > > > have
> > > > > > > > > > diverged
> > > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > > reimplementation
> > > > > > > > would
> > > > > > > > > > not be
> > > > > > > > > > > >> > > > possible.
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at
> > > 8:39 AM
> > > > > > Sagar
> > > > > > > > <
> > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a
> confluence
> > > page
> > > > > to
> > > > > > > > > discuss
> > > > > > > > > > how
> > > > > > > > > > > >> Connect
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> > > rebalance
> > > > > > > > protocol.
> > > > > > > > > > > >> Here's the
> > > > > > > > > > > >> > > > page:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty
> longish
> > > and
> > > > > I
> > > > > > have
> > > > > > > > > > tried
> > > > > > > > > > > >> to keep
> > > > > > > > > > > >> > > > a
> > > > > > > > > > > >> > > > > > format
> > > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know
> > > what you
> > > > > > > > think.
> > > > > > > > > > Also,
> > > > > > > > > > > >> do you
> > > > > > > > > > > >> > > > > > think this
> > > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > > > > discussion
> > > > > > > > thread
> > > > > > > > > > or
> > > > > > > > > > > >> is this
> > > > > > > > > > > >> > > > one
> > > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022
> at
> > > 7:37
> > > > > AM
> > > > > > > > > Sagar <
> > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much
> for
> > > the
> > > > > > doc on
> > > > > > > > > > Kafka
> > > > > > > > > > > >> Streams.
> > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come
> up
> > > with
> > > > > > such a
> > > > > > > > > > > >> document.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26,
> 2022 at
> > > > > 4:47
> > > > > > AM
> > > > > > > > > > Guozhang
> > > > > > > > > > > >> Wang <
> > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great
> if
> > > you
> > > > > > could
> > > > > > > > > come
> > > > > > > > > > > >> back with
> > > > > > > > > > > >> > > > some
> > > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the
> Connect
> > > side
> > > > > > > > > > integration
> > > > > > > > > > > >> with
> > > > > > > > > > > >> > > > the new
> > > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides
> leveraging on
> > > the
> > > > > > new
> > > > > > > > > > "protocol
> > > > > > > > > > > >> type"
> > > > > > > > > > > >> > > > we
> > > > > > > > > > > >> > > > > > did not
> > > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > > > > implementations.
> > > > > > > > > > For
> > > > > > > > > > > >> Streams
> > > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your
> > > analysis
> > > > > > on
> > > > > > > > > > Connect.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19,
> 2022
> > > at
> > > > > > 10:48 PM
> > > > > > > > > > Sagar <
> > > > > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for
> your
> > > > > > response.
> > > > > > > > The
> > > > > > > > > > > >> reason I
> > > > > > > > > > > >> > > > thought
> > > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme
> is
> > > that
> > > > > > even
> > > > > > > > > > today the
> > > > > > > > > > > >> > > > connect
> > > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > > > > AbstractCoordinator to
> > > > > > > > > > > >> empower
> > > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> tasks/connectors.
> > > The
> > > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> > > metadata()
> > > > > > method
> > > > > > > > > by
> > > > > > > > > > > >> plumbing
> > > > > > > > > > > >> > > > into
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > JoinGroupRequestProtocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the
> changes
> > > to
> > > > > > support
> > > > > > > > > > > >> connect would
> > > > > > > > > > > >> > > > be
> > > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in
> > > streams
> > > > > > mainly
> > > > > > > > > > because
> > > > > > > > > > > >> of the
> > > > > > > > > > > >> > > > > > Client
> > > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both.
> At an
> > > > > > > > > implementation
> > > > > > > > > > > >> level, we
> > > > > > > > > > > >> > > > > > might
> > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get
> onto
> > > this
> > > > > > new
> > > > > > > > > > > >> assignment
> > > > > > > > > > > >> > > > protocol
> > > > > > > > > > > >> > > > > > like
> > > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > > > request/response and
> > > > > > > > > > > >> SyncGroup and
> > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> ConsumerGroupHeartbeat
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on
> > > similar
> > > > > > lines
> > > > > > > > > to
> > > > > > > > > > > >> streams (or
> > > > > > > > > > > >> > > > > > there
> > > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> > > perform a
> > > > > > > > detailed
> > > > > > > > > > > >> analysis of
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > same
> > > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate
> discussion
> > > > > > thread for
> > > > > > > > > > that
> > > > > > > > > > > >> as that
> > > > > > > > > > > >> > > > would
> > > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me
> know
> > > if
> > > > > > that
> > > > > > > > > sounds
> > > > > > > > > > > >> good to
> > > > > > > > > > > >> > > > you.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15,
> > > 2022 at
> > > > > > 5:47
> > > > > > > > PM
> > > > > > > > > > David
> > > > > > > > > > > >> Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for
> your
> > > > > > comments.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That
> > > refers to
> > > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea
> is to
> > > > > > > > transition C
> > > > > > > > > > from
> > > > > > > > > > > >> his
> > > > > > > > > > > >> > > > current
> > > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target
> assignment
> > > when
> > > > > > he
> > > > > > > > can
> > > > > > > > > > move
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > > >> > > > > > When
> > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> assignment
> > > is
> > > > > > updated
> > > > > > > > > and
> > > > > > > > > > > >> persisted
> > > > > > > > > > > >> > > > > > with all
> > > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions
> even if
> > > > > they
> > > > > > are
> > > > > > > > > not
> > > > > > > > > > all
> > > > > > > > > > > >> revoked
> > > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member
> assignment
> > > > > > becomes
> > > > > > > > the
> > > > > > > > > > target
> > > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization
> to
> > > avoid
> > > > > > having
> > > > > > > > > to
> > > > > > > > > > > >> write all
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are
> > > based on
> > > > > > the
> > > > > > > > > > persisted
> > > > > > > > > > > >> state
> > > > > > > > > > > >> > > > so I
> > > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion.
> Let me
> > > see
> > > > > > if I
> > > > > > > > can
> > > > > > > > > > > >> improve
> > > > > > > > > > > >> > > > this in
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding
> > > Connect,
> > > > > it
> > > > > > > > could
> > > > > > > > > > > >> reuse the
> > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it
> > > fits in
> > > > > > the
> > > > > > > > > > > >> protocol. The
> > > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > topicid-partitions +
> > > > > > > > metadata,
> > > > > > > > > > could
> > > > > > > > > > > >> > > > Connect
> > > > > > > > > > > >> > > > > > fit
> > > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul
> 15,
> > > 2022
> > > > > at
> > > > > > 1:55
> > > > > > > > > PM
> > > > > > > > > > > >> Sagar <
> > > > > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for
> the
> > > KIP.
> > > > > I
> > > > > > just
> > > > > > > > > had
> > > > > > > > > > > >> minor
> > > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the
> > > Assignment
> > > > > > Error
> > > > > > > > > > > >> section in
> > > > > > > > > > > >> > > > Client
> > > > > > > > > > > >> > > > > > Side
> > > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you
> mentioned
> > > => `In
> > > > > > this
> > > > > > > > > > case,
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > client
> > > > > > > > > > > >> > > > > > side
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the
> > > group
> > > > > > > > > > coordinator`.
> > > > > > > > > > > >> In this
> > > > > > > > > > > >> > > > > > case are
> > > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> > > returning
> > > > > an
> > > > > > > > > > > >> AssignmentError
> > > > > > > > > > > >> > > > that's
> > > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do
> you
> > > think
> > > > > > it
> > > > > > > > > would
> > > > > > > > > > > >> make sense
> > > > > > > > > > > >> > > > to
> > > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the
> Case
> > > > > Studies
> > > > > > > > > > section, I
> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > > slight
> > > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have
> the
> > > > > same.
> > > > > > > > > Consider
> > > > > > > > > > > >> this step:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> > > heartbeats,
> > > > > the
> > > > > > > > group
> > > > > > > > > > > >> coordinator
> > > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> > > partitions
> > > > > to
> > > > > > > > > > revoke. It
> > > > > > > > > > > >> > > > persists
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group
> > > (epoch=3)
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > > > > Assignment
> > > > > > > > > > (epoch=3)
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > > > > partitions=[foo-0]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > > > > partitions=[foo-2]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > > > > partitions=[foo-1]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > > > > Assignment
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > epoch=2,
> > > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > epoch=3,
> > > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > epoch=3,
> > > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> > > heartbeats,
> > > > > it
> > > > > > > > > > transitions
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > > >> > > > > > but
> > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's
> > > mentioned
> > > > > > that
> > > > > > > > > > member C
> > > > > > > > > > > >> can't
> > > > > > > > > > > >> > > > get
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > > > > description
> > > > > > > > > > above,
> > > > > > > > > > > >> it seems
> > > > > > > > > > > >> > > > it
> > > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be
> better
> > > to
> > > > > > remove
> > > > > > > > it
> > > > > > > > > > and
> > > > > > > > > > > >> > > > populate it
> > > > > > > > > > > >> > > > > > only
> > > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this
> in a
> > > lot
> > > > > of
> > > > > > > > other
> > > > > > > > > > > >> places, so
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > I
> > > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding
> > > connect ,
> > > > > it
> > > > > > > > might
> > > > > > > > > > be
> > > > > > > > > > > >> out of
> > > > > > > > > > > >> > > > scope
> > > > > > > > > > > >> > > > > > of
> > > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> > > understood it
> > > > > > would
> > > > > > > > > > > >> probably be
> > > > > > > > > > > >> > > > > > running in
> > > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the
> new
> > > > > > rebalance
> > > > > > > > > > > >> protocol as it
> > > > > > > > > > > >> > > > has
> > > > > > > > > > > >> > > > > > its
> > > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> Assignors(Eager
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > > IncrementalCooperative).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul
> 15,
> > > 2022
> > > > > > at
> > > > > > > > 5:00
> > > > > > > > > > PM
> > > > > > > > > > > >> David
> > > > > > > > > > > >> > > > Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> > > Hector! Our
> > > > > > goal
> > > > > > > > is
> > > > > > > > > > to
> > > > > > > > > > > >> move
> > > > > > > > > > > >> > > > forward
> > > > > > > > > > > >> > > > > > with
> > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of
> > > relying
> > > > > > on
> > > > > > > > one
> > > > > > > > > > > >> generic API.
> > > > > > > > > > > >> > > > For
> > > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same
> > > pattern
> > > > > > and
> > > > > > > > > > > >> reuse/share the
> > > > > > > > > > > >> > > > core
> > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server
> side.
> > > For
> > > > > the
> > > > > > > > > schema
> > > > > > > > > > > >> registry, I
> > > > > > > > > > > >> > > > > > think
> > > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a
> > > tailored
> > > > > > API to
> > > > > > > > > do
> > > > > > > > > > > >> simple
> > > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri,
> Jul
> > > 15,
> > > > > > 2022 at
> > > > > > > > > > 10:22
> > > > > > > > > > > >> AM Ismael
> > > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three
> quick
> > > > > > comments:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1.
> Regarding
> > > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > > >> > > > vs
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > com.google.re2j.Pattern,
> > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> document the
> > > > > > > > differences
> > > > > > > > > > in
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > detail
> > > > > > > > > > > >> > > > > > before
> > > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That
> said,
> > > if
> > > > > > people
> > > > > > > > > pass
> > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> semantics
> > > to be
> > > > > > > > honored.
> > > > > > > > > > If
> > > > > > > > > > > >> we are
> > > > > > > > > > > >> > > > doing
> > > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should
> > > consider
> > > > > > adding
> > > > > > > > > an
> > > > > > > > > > > >> overload
> > > > > > > > > > > >> > > > with
> > > > > > > > > > > >> > > > > > our own
> > > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think
> we'd
> > > want
> > > > > to
> > > > > > > > > expose
> > > > > > > > > > > >> re2j's at
> > > > > > > > > > > >> > > > this
> > > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2.
> Regarding
> > > > > topic
> > > > > > > > ids,
> > > > > > > > > > any
> > > > > > > > > > > >> major new
> > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and
> > > should
> > > > > > handle
> > > > > > > > the
> > > > > > > > > > topic
> > > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we
> > > need to
> > > > > > > > handle.
> > > > > > > > > I
> > > > > > > > > > > >> agree with
> > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the
> > > relevant
> > > > > > > > > protocols
> > > > > > > > > > > >> that don't
> > > > > > > > > > > >> > > > > > have it
> > > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus
> on the
> > > > > > internals
> > > > > > > > > > versus
> > > > > > > > > > > >> adding
> > > > > > > > > > > >> > > > new
> > > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find
> that
> > > > > > adding
> > > > > > > > new
> > > > > > > > > > APIs
> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > required
> > > > > > > > > > > >> > > > > > for
> > > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am
> > > still
> > > > > not
> > > > > > sure
> > > > > > > > > > about
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> powerful for
> > > > > > configs
> > > > > > > > to
> > > > > > > > > be
> > > > > > > > > > > >> > > > centralized
> > > > > > > > > > > >> > > > > > in the
> > > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > > > > (auditability,
> > > > > > > > > > > >> visibility,
> > > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure
> about
> > > > > > > > automatically
> > > > > > > > > > > >> deleting
> > > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> recovered. A
> > > > > good
> > > > > > > > > property
> > > > > > > > > > > >> for modern
> > > > > > > > > > > >> > > > > > systems
> > > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > unrecoverable
> > > > > data
> > > > > > > > loss
> > > > > > > > > > > >> scenarios.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed,
> Jul
> > > 13,
> > > > > > 2022
> > > > > > > > at
> > > > > > > > > > 3:47
> > > > > > > > > > > >> PM David
> > > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > > > > Guozhang. My
> > > > > > > > > > answers
> > > > > > > > > > > >> are
> > > > > > > > > > > >> > > > below:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1)
> the
> > > > > > migration
> > > > > > > > > path,
> > > > > > > > > > > >> especially
> > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > enable
> > > > > the
> > > > > > new
> > > > > > > > > > > >> protocol, in
> > > > > > > > > > > >> > > > which
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > protocols /
> > > > > > rpcs
> > > > > > > > and
> > > > > > > > > > old
> > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> group.
> > > How
> > > > > the
> > > > > > > > > > > >> coordinator could
> > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > > protocol
> > > > > > is
> > > > > > > > > > something
> > > > > > > > > > > >> we
> > > > > > > > > > > >> > > > need to
> > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Noted. I
> > > just
> > > > > > > > > published
> > > > > > > > > > a
> > > > > > > > > > > >> new
> > > > > > > > > > > >> > > > version
> > > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> details
> > > about
> > > > > > this.
> > > > > > > > > See
> > > > > > > > > > the
> > > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> the
> > > > > > > > > "Compatibility,
> > > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > > >> > > > > > and
> > > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I
> have to
> > > > > think
> > > > > > > > > through
> > > > > > > > > > a
> > > > > > > > > > > >> few cases
> > > > > > > > > > > >> > > > > > now but
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> mechanism
> > > > > > should be
> > > > > > > > > > > >> understandable.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2)
> the
> > > usage
> > > > > > of
> > > > > > > > > topic
> > > > > > > > > > > >> ids. So
> > > > > > > > > > > >> > > > far as
> > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> part of
> > > RPCs
> > > > > > and
> > > > > > > > > admin
> > > > > > > > > > > >> client,
> > > > > > > > > > > >> > > > but
> > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > consumers
> > > > > > yet.
> > > > > > > > I
> > > > > > > > > > think
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> client
> > > to be
> > > > > > > > > > maintaining
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > names
> > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> leverage
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on
> all
> > > the
> > > > > > > > augmented
> > > > > > > > > > > >> existing
> > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > secondly,
> > > > > > should
> > > > > > > > we
> > > > > > > > > > ever
> > > > > > > > > > > >> consider
> > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> public
> > > APIs
> > > > > as
> > > > > > > > well
> > > > > > > > > > (both
> > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > listener for
> > > > > > cases
> > > > > > > > > > like
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> > > Assuming
> > > > > > that we
> > > > > > > > > > would
> > > > > > > > > > > >> include
> > > > > > > > > > > >> > > > > > converting
> > > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs
> to
> > > using
> > > > > > topic
> > > > > > > > > ids
> > > > > > > > > > in
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > KIP,
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully
> > > operate
> > > > > > with
> > > > > > > > > topic
> > > > > > > > > > > >> ids. That
> > > > > > > > > > > >> > > > > > being
> > > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> topics
> > > > > > names in
> > > > > > > > > > various
> > > > > > > > > > > >> APIs so
> > > > > > > > > > > >> > > > > > having a
> > > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > > > > inevitable to
> > > > > > > > > me.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I
> don't
> > > > > have
> > > > > > a
> > > > > > > > > strong
> > > > > > > > > > > >> opinion on
> > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> beyond the
> > > > > > scope of
> > > > > > > > > this
> > > > > > > > > > > >> KIP. I
> > > > > > > > > > > >> > > > would
> > > > > > > > > > > >> > > > > > rather
> > > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here
> and
> > > we
> > > > > can
> > > > > > > > > consider
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming
> > > back to
> > > > > > > > > Ismael's
> > > > > > > > > > > >> point about
> > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > > >> > > > think
> > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> favour of
> > > it.
> > > > > > The
> > > > > > > > > > consumer
> > > > > > > > > > > >> will
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> topics
> > > > > > exists
> > > > > > > > > before
> > > > > > > > > > > >> passing
> > > > > > > > > > > >> > > > them
> > > > > > > > > > > >> > > > > > into
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> protocol.
> > > > > > Obviously,
> > > > > > > > > the
> > > > > > > > > > > >> > > > coordinator
> > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not
> really
> > > > > have
> > > > > > a
> > > > > > > > way
> > > > > > > > > to
> > > > > > > > > > > >> reject an
> > > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > > agreeing
> > > > > > with
> > > > > > > > > > David
> > > > > > > > > > > >> on all
> > > > > > > > > > > >> > > > other
> > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > >> question:
> > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > deprecate
> > > > > the
> > > > > > > > > > subscribe
> > > > > > > > > > > >> API with
> > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> while
> > > just
> > > > > > > > > documenting
> > > > > > > > > > > >> that our
> > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> server.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> Since
> > > the
> > > > > > > > > incompatible
> > > > > > > > > > > >> case is a
> > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > overloaded
> > > > > > > > `String`
> > > > > > > > > > based
> > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> invalid
> > > > > > regexes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That
> could
> > > > > > work. I
> > > > > > > > > have
> > > > > > > > > > to
> > > > > > > > > > > >> look at
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> engines to
> > > > > > better
> > > > > > > > > > > >> understand the
> > > > > > > > > > > >> > > > > > potential
> > > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding
> is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> would
> > > > > work
> > > > > > for
> > > > > > > > > all
> > > > > > > > > > the
> > > > > > > > > > > >> basic
> > > > > > > > > > > >> > > > > > regular
> > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> between
> > > the
> > > > > two
> > > > > > are
> > > > > > > > > > mainly
> > > > > > > > > > > >> about
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > various
> > > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder
> > > what
> > > > > > other
> > > > > > > > > people
> > > > > > > > > > > >> think
> > > > > > > > > > > >> > > > about
> > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On
> Tue,
> > > Jul
> > > > > 12,
> > > > > > 2022
> > > > > > > > > at
> > > > > > > > > > > >> 11:28 PM
> > > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> wangguoz@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> Thanks
> > > > > David!
> > > > > > I
> > > > > > > > > think
> > > > > > > > > > on
> > > > > > > > > > > >> the high
> > > > > > > > > > > >> > > > > > level
> > > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > > concretize
> > > > > > a
> > > > > > > > bit
> > > > > > > > > > more:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1)
> the
> > > > > > migration
> > > > > > > > > path,
> > > > > > > > > > > >> especially
> > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > enable
> > > > > the
> > > > > > new
> > > > > > > > > > > >> protocol, in
> > > > > > > > > > > >> > > > which
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > protocols /
> > > > > > rpcs
> > > > > > > > and
> > > > > > > > > > old
> > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> group.
> > > How
> > > > > the
> > > > > > > > > > > >> coordinator could
> > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > > protocol
> > > > > > is
> > > > > > > > > > something
> > > > > > > > > > > >> we
> > > > > > > > > > > >> > > > need to
> > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2)
> the
> > > usage
> > > > > > of
> > > > > > > > > topic
> > > > > > > > > > > >> ids. So
> > > > > > > > > > > >> > > > far as
> > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> part of
> > > RPCs
> > > > > > and
> > > > > > > > > admin
> > > > > > > > > > > >> client,
> > > > > > > > > > > >> > > > but
> > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > consumers
> > > > > > yet.
> > > > > > > > I
> > > > > > > > > > think
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> client
> > > to be
> > > > > > > > > > maintaining
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > names
> > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> leverage
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on
> all
> > > the
> > > > > > > > augmented
> > > > > > > > > > > >> existing
> > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > secondly,
> > > > > > should
> > > > > > > > we
> > > > > > > > > > ever
> > > > > > > > > > > >> consider
> > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> public
> > > APIs
> > > > > as
> > > > > > > > well
> > > > > > > > > > (both
> > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > listener for
> > > > > > cases
> > > > > > > > > > like
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > > agreeing
> > > > > > with
> > > > > > > > > > David
> > > > > > > > > > > >> on all
> > > > > > > > > > > >> > > > other
> > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > >> question:
> > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > deprecate
> > > > > the
> > > > > > > > > > subscribe
> > > > > > > > > > > >> API with
> > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> while
> > > just
> > > > > > > > > documenting
> > > > > > > > > > > >> that our
> > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> server.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> Since
> > > the
> > > > > > > > > incompatible
> > > > > > > > > > > >> case is a
> > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > overloaded
> > > > > > > > `String`
> > > > > > > > > > based
> > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> invalid
> > > > > > regexes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> Guozhang
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On
> Tue,
> > > Jul
> > > > > > 12,
> > > > > > > > 2022
> > > > > > > > > > at
> > > > > > > > > > > >> 5:23 AM
> > > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi
> > > Ismael,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > Thanks for
> > > > > > your
> > > > > > > > > > > >> feedback. Let
> > > > > > > > > > > >> > > > me
> > > > > > > > > > > >> > > > > > answer
> > > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 1. I
> > > > > think
> > > > > > > > it's
> > > > > > > > > > > >> premature to
> > > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > deprecation
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > removal
> > > > > > of the
> > > > > > > > > > > >> existing group
> > > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > client
> > > > > > > > protocol
> > > > > > > > > > and
> > > > > > > > > > > >> hence
> > > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> dependent
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> how
> > > > > > quickly
> > > > > > > > > > > >> applications
> > > > > > > > > > > >> > > > migrate
> > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> That
> > > makes
> > > > > > > > sense.
> > > > > > > > > I
> > > > > > > > > > will
> > > > > > > > > > > >> > > > remove it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 2.
> > > The
> > > > > KIP
> > > > > > > > says
> > > > > > > > > we
> > > > > > > > > > > >> intend to
> > > > > > > > > > > >> > > > > > release
> > > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> clear
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > why. If
> > > > > we
> > > > > > > > added
> > > > > > > > > > that
> > > > > > > > > > > >> as a
> > > > > > > > > > > >> > > > way to
> > > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> the
> > > > > group
> > > > > > > > > > protocol, I
> > > > > > > > > > > >> also
> > > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Let me
> > > > > > explain
> > > > > > > > my
> > > > > > > > > > > >> reasoning. As
> > > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > coordinator
> > > > > > in
> > > > > > > > > Java
> > > > > > > > > > > >> while we
> > > > > > > > > > > >> > > > > > implement
> > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> that
> > > the
> > > > > > > > internals
> > > > > > > > > > will
> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > slightly
> > > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > Therefore, I
> > > > > > > > > wanted
> > > > > > > > > > to
> > > > > > > > > > > >> tighten
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > coordinator
> > > > > > to
> > > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > alternative
> > > > > > > > would
> > > > > > > > > > be to
> > > > > > > > > > > >> use a
> > > > > > > > > > > >> > > > flag
> > > > > > > > > > > >> > > > > > to do
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> on the
> > > > > > software
> > > > > > > > > > upgrade.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 3.
> > > We
> > > > > > need to
> > > > > > > > > > flesh
> > > > > > > > > > > >> out the
> > > > > > > > > > > >> > > > > > details of
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> like
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > we're
> > > > > > saying
> > > > > > > > we
> > > > > > > > > > will
> > > > > > > > > > > >> support
> > > > > > > > > > > >> > > > > > online
> > > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > explain
> > > > > > this
> > > > > > > > in
> > > > > > > > > > > >> detail. It
> > > > > > > > > > > >> > > > could
> > > > > > > > > > > >> > > > > > also
> > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > easier.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Yes,
> > > we
> > > > > will
> > > > > > > > > support
> > > > > > > > > > > >> online
> > > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > group
> > > > > > using
> > > > > > > > the
> > > > > > > > > > old
> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Let me
> > > > > > briefly
> > > > > > > > > > explain
> > > > > > > > > > > >> how that
> > > > > > > > > > > >> > > > > > will work
> > > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> four
> > > step
> > > > > > > > process:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> 1. The
> > > > > > cluster
> > > > > > > > > must
> > > > > > > > > > be
> > > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> new
> > > group
> > > > > > > > > > coordinator.
> > > > > > > > > > > >> Both the
> > > > > > > > > > > >> > > > > > old and
> > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > support
> > > > > the
> > > > > > old
> > > > > > > > > > > >> protocol and
> > > > > > > > > > > >> > > > rely
> > > > > > > > > > > >> > > > > > on the
> > > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> they
> > > can
> > > > > > work
> > > > > > > > > > together.
> > > > > > > > > > > >> This
> > > > > > > > > > > >> > > > point
> > > > > > > > > > > >> > > > > > is an
> > > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > this
> > > > > one
> > > > > > live
> > > > > > > > > > > >> because it
> > > > > > > > > > > >> > > > would
> > > > > > > > > > > >> > > > > > require
> > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > coordinator
> > > > > > and
> > > > > > > > > > > >> starting up the
> > > > > > > > > > > >> > > > > > new one
> > > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > unavailabilities.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> 2. The
> > > > > > cluster's
> > > > > > > > > > > >> metadata
> > > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > enable
> > > > > > the
> > > > > > > > new
> > > > > > > > > > > >> protocol.
> > > > > > > > > > > >> > > > This
> > > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > terminated
> > > > > > > > because
> > > > > > > > > > the
> > > > > > > > > > > >> old
> > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> 3. The
> > > > > > consumers
> > > > > > > > > > must be
> > > > > > > > > > > >> > > > upgraded
> > > > > > > > > > > >> > > > > > to a
> > > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > migration
> > > > > > (must
> > > > > > > > > have
> > > > > > > > > > > >> KIP-792).
> > > > > > > > > > > >> > > > If
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > Nothing
> > > > > > must be
> > > > > > > > > > done at
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > point.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> 4. The
> > > > > > consumers
> > > > > > > > > > must
> > > > > > > > > > > >> be rolled
> > > > > > > > > > > >> > > > > > with the
> > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > consumer
> > > > > > group
> > > > > > > > is
> > > > > > > > > > > >> automatically
> > > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> using
> > > the
> > > > > > new
> > > > > > > > > > protocol
> > > > > > > > > > > >> joins
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > group.
> > > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> old
> > > > > > protocol are
> > > > > > > > > > being
> > > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > > >> > > > > > the old
> > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> new
> > > one.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Let me
> > > > > > clarify
> > > > > > > > all
> > > > > > > > > > of
> > > > > > > > > > > >> this in
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 4.
> > > I am
> > > > > > happy
> > > > > > > > > > that we
> > > > > > > > > > > >> are
> > > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > server,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> but
> > > it
> > > > > > seems
> > > > > > > > > like
> > > > > > > > > > > >> there
> > > > > > > > > > > >> > > > could be
> > > > > > > > > > > >> > > > > > some
> > > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > have a
> > > > > > > > mechanism
> > > > > > > > > > for
> > > > > > > > > > > >> users to
> > > > > > > > > > > >> > > > > > detect
> > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > before
> > > > > > > > switching
> > > > > > > > > > to
> > > > > > > > > > > >> the new
> > > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > think
> > > > > > that I
> > > > > > > > am
> > > > > > > > > a
> > > > > > > > > > bit
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > believe
> > > > > > that the
> > > > > > > > > > > >> majority of
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > cases
> > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> work
> > > with
> > > > > > the
> > > > > > > > new
> > > > > > > > > > > >> engine. The
> > > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> and
> > > reject
> > > > > > the
> > > > > > > > > > consumer
> > > > > > > > > > > >> if the
> > > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > migration
> > > > > > path,
> > > > > > > > in
> > > > > > > > > > the
> > > > > > > > > > > >> worst
> > > > > > > > > > > >> > > > case,
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> the
> > > group
> > > > > > will
> > > > > > > > be
> > > > > > > > > > > >> rejected.
> > > > > > > > > > > >> > > > This
> > > > > > > > > > > >> > > > > > should
> > > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> would
> > > say.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> One
> > > way
> > > > > for
> > > > > > > > > > customers to
> > > > > > > > > > > >> > > > validate
> > > > > > > > > > > >> > > > > > their
> > > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> prod
> > > would
> > > > > > be to
> > > > > > > > > > test
> > > > > > > > > > > >> them with
> > > > > > > > > > > >> > > > > > another
> > > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> could
> > > be
> > > > > > done
> > > > > > > > in a
> > > > > > > > > > > >> pre-prod
> > > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > extend the
> > > > > > > > > > > >> consumer-group tool
> > > > > > > > > > > >> > > > to
> > > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > mechanism.
> > > > > > Would
> > > > > > > > > > this be
> > > > > > > > > > > >> > > > enough in
> > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 5.
> > > > > > Related to
> > > > > > > > > the
> > > > > > > > > > last
> > > > > > > > > > > >> > > > question,
> > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users
> to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > stick
> > > > > > with the
> > > > > > > > > > > >> current regex
> > > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > example,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> it
> > > may
> > > > > be
> > > > > > > > handy
> > > > > > > > > to
> > > > > > > > > > > >> keep using
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > migrations
> > > > > > > > > simple
> > > > > > > > > > and
> > > > > > > > > > > >> then
> > > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> step.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > > understand
> > > > > > > > your
> > > > > > > > > > point
> > > > > > > > > > > >> but I
> > > > > > > > > > > >> > > > am
> > > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > actually
> > > > > > stay
> > > > > > > > > in
> > > > > > > > > > > >> this mode.
> > > > > > > > > > > >> > > > That
> > > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > simplifying
> > > > > > the
> > > > > > > > > > client
> > > > > > > > > > > >> because
> > > > > > > > > > > >> > > > we
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> the
> > > > > > metadata on
> > > > > > > > > the
> > > > > > > > > > > >> client
> > > > > > > > > > > >> > > > side. I
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 6.
> > > When
> > > > > > we say
> > > > > > > > > > that
> > > > > > > > > > > >> the group
> > > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible
> for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> storing
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> the
> > > > > > > > > configurations
> > > > > > > > > > > >> and that
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > group is
> > > > > > > > > deleted.
> > > > > > > > > > > >> Will a
> > > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > configurations?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> That's
> > > > > > right.
> > > > > > > > The
> > > > > > > > > > > >> > > > configurations
> > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > deleted.
> > > > > > They go
> > > > > > > > > > > >> together.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 7.
> > > Will
> > > > > > the
> > > > > > > > > > choice to
> > > > > > > > > > > >> store
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make
> it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > harder
> > > > > to
> > > > > > list
> > > > > > > > > all
> > > > > > > > > > > >> cluster
> > > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > don't
> > > > > > think
> > > > > > > > so.
> > > > > > > > > > The
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > configs.
> > > > > If
> > > > > > you
> > > > > > > > > > want to
> > > > > > > > > > > >> know
> > > > > > > > > > > >> > > > all
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > ask all
> > > > > > the
> > > > > > > > > group
> > > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > > >> > > > > > You
> > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> for
> > > > > > instance.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 8.
> > > How
> > > > > > would
> > > > > > > > > > someone
> > > > > > > > > > > >> > > > configure a
> > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> we
> > > > > > considered
> > > > > > > > > > > >> allowing the
> > > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > Alternatively,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > configs
> > > > > > could
> > > > > > > > be
> > > > > > > > > > > >> decoupled
> > > > > > > > > > > >> > > > from
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Yes.
> > > The
> > > > > > group
> > > > > > > > > will
> > > > > > > > > > be
> > > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> the
> > > > > configs
> > > > > > will
> > > > > > > > > be
> > > > > > > > > > > >> lost after
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > passes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 9.
> > > Will
> > > > > > the
> > > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > > >> > > > > > method
> > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > > >> > > > do
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > That's a
> > > > > > very
> > > > > > > > > group
> > > > > > > > > > > >> question. I
> > > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > > >> > > > not
> > > > > > > > > > > >> > > > > > fully
> > > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > plan to
> > > > > > use,
> > > > > > > > it
> > > > > > > > > > > >> might be
> > > > > > > > > > > >> > > > better
> > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> which
> > > > > takes
> > > > > > a
> > > > > > > > > > string.
> > > > > > > > > > > >> We would
> > > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > During the
> > > > > > > > > > migration, I
> > > > > > > > > > > >> think
> > > > > > > > > > > >> > > > that
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> regex
> > > and
> > > > > > use
> > > > > > > > it.
> > > > > > > > > > That
> > > > > > > > > > > >> should
> > > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> cases.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 10.
> > > I
> > > > > > agree
> > > > > > > > with
> > > > > > > > > > > >> Justine
> > > > > > > > > > > >> > > > that we
> > > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > switch
> > > > > to
> > > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > > >> > > > > > from the
> > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> can
> > > > > > switch the
> > > > > > > > > > > >> default for
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > feature
> > > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > metadata.version
> > > > > > > > > > once
> > > > > > > > > > > >> we
> > > > > > > > > > > >> > > > want to
> > > > > > > > > > > >> > > > > > make
> > > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My
> > > plan
> > > > > was
> > > > > > to
> > > > > > > > use
> > > > > > > > > > that
> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > > > > > flag
> > > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > phase. I
> > > > > > should
> > > > > > > > > not
> > > > > > > > > > have
> > > > > > > > > > > >> > > > mentioned
> > > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > > > > internal
> > > > > > > > config
> > > > > > > > > > for
> > > > > > > > > > > >> it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 11.
> > > Some
> > > > > > of
> > > > > > > > the
> > > > > > > > > > > >> protocol APIs
> > > > > > > > > > > >> > > > > > don't
> > > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > good to
> > > > > > add
> > > > > > > > that
> > > > > > > > > > for
> > > > > > > > > > > >> > > > consistency.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Noted.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 12.
> > > It
> > > > > is
> > > > > > a
> > > > > > > > bit
> > > > > > > > > > odd
> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> even
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > though
> > > > > it
> > > > > > > > seems
> > > > > > > > > > to do
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > than
> > > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > agree.
> > > > > > This is
> > > > > > > > > > how the
> > > > > > > > > > > >> > > > current
> > > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > require
> > > > > > "Read
> > > > > > > > > > Group" to
> > > > > > > > > > > >> join a
> > > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> but I
> > > am
> > > > > not
> > > > > > > > sure
> > > > > > > > > > that
> > > > > > > > > > > >> it is
> > > > > > > > > > > >> > > > worth
> > > > > > > > > > > >> > > > > > it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 13.
> > > How
> > > > > is
> > > > > > > > topic
> > > > > > > > > > > >> recreation
> > > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > protocol?
> > > > > > It
> > > > > > > > > > would be
> > > > > > > > > > > >> good to
> > > > > > > > > > > >> > > > > > have a
> > > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Noted.
> > > > > From
> > > > > > a
> > > > > > > > > > protocol
> > > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> topic
> > > id
> > > > > so
> > > > > > it
> > > > > > > > > will
> > > > > > > > > > > >> treat it
> > > > > > > > > > > >> > > > like a
> > > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> only
> > > issue
> > > > > > is
> > > > > > > > that
> > > > > > > > > > the
> > > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> IDs
> > > so the
> > > > > > > > > consumer
> > > > > > > > > > > >> would
> > > > > > > > > > > >> > > > reuse the
> > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> that
> > > we
> > > > > > should
> > > > > > > > > > update
> > > > > > > > > > > >> those
> > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > >> > > > > > as well
> > > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > end.
> > > > > That
> > > > > > > > would
> > > > > > > > > > > >> strengthen
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 14.
> > > The
> > > > > > KIP
> > > > > > > > > > mentions
> > > > > > > > > > > >> we will
> > > > > > > > > > > >> > > > > > write the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> though
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > this is
> > > > > an
> > > > > > > > > > > >> implementation
> > > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> for
> > > it?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Yes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 15.
> > > Do
> > > > > we
> > > > > > > > have a
> > > > > > > > > > > >> scalability
> > > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > group
> > > > > > protocol
> > > > > > > > > can
> > > > > > > > > > > >> support?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We
> > > don't
> > > > > > have
> > > > > > > > > > numbers
> > > > > > > > > > > >> at the
> > > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of
> > > members
> > > > > > per
> > > > > > > > > > group.
> > > > > > > > > > > >> We will
> > > > > > > > > > > >> > > > > > measure
> > > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > implementation.
> > > > > > > > > Note
> > > > > > > > > > > >> that we
> > > > > > > > > > > >> > > > might
> > > > > > > > > > > >> > > > > > have
> > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> road
> > > (e.g.
> > > > > > > > offset
> > > > > > > > > > > >> commits).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> 16.
> > > Did
> > > > > we
> > > > > > > > > > consider
> > > > > > > > > > > >> having
> > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > SubscribedTopicNames
> > > > > > > > > > > >> in
> > > > > > > > > > > >> > > > > > > > > > > >> >
> ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > since we
> > > > > > have
> > > > > > > > to
> > > > > > > > > > > >> resolve the
> > > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> the
> > > > > topic
> > > > > > > > name?
> > > > > > > > > > The
> > > > > > > > > > > >> > > > difference
> > > > > > > > > > > >> > > > > > is that
> > > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > efficient
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > whereas
> > > > > > > > sending
> > > > > > > > > > the
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > names
> > > > > > > > > > > >> > > > > > is less
> > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > recreation is
> > > > > > > > > > easier
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > handle
> > > > > > > > > > > >> > > > > > if we
> > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> The
> > > idea
> > > > > > was to
> > > > > > > > > > > >> consolidate the
> > > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> paths
> > > but
> > > > > I
> > > > > > do
> > > > > > > > > agree
> > > > > > > > > > > >> with your
> > > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> ids
> > > may be
> > > > > > > > better
> > > > > > > > > > here
> > > > > > > > > > > >> for the
> > > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > suppose
> > > > > > that we
> > > > > > > > > may
> > > > > > > > > > > >> allow
> > > > > > > > > > > >> > > > users to
> > > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> future
> > > > > > because
> > > > > > > > > that
> > > > > > > > > > is
> > > > > > > > > > > >> the only
> > > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > re-creation.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On
> > > Tue,
> > > > > Jul
> > > > > > 12,
> > > > > > > > > > 2022 at
> > > > > > > > > > > >> 1:38 PM
> > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > djacot@confluent.io>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> Hi
> > > > > > Justine,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > Thanks
> > > > > for
> > > > > > > > your
> > > > > > > > > > > >> comments.
> > > > > > > > > > > >> > > > Please
> > > > > > > > > > > >> > > > > > find
> > > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> -
> > > Yes,
> > > > > > the new
> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > relies on
> > > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > topic
> > > > > > names
> > > > > > > > > based
> > > > > > > > > > in
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> if
> > > using
> > > > > > topic
> > > > > > > > > > names
> > > > > > > > > > > >> is the
> > > > > > > > > > > >> > > > > > right call
> > > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> a
> > > little
> > > > > > more.
> > > > > > > > > > > >> Obviously, the
> > > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > > >> > > > > > > > > > > >>
> > >
> >
> >
> > --
> > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi all,

I have updated the KIP to incorporate all the recent feedback.

Change log:
* I switched all the examples to use the in-memory representation
instead of the on-disk representation. That was too confusing.
* Reworked the PartitionAssignor interfaces.
* Clarified how and when the ConsumerRebalanceListener and
PartitionAssignor#onAssignment are called. This is in the
Consumer/Rebalance Process chapter.
* Updated the ConsumerGroupHeartbeat API to include the pending
partitions. This is needed for onAssignment.
* Updated ConsumerGroupPrepareAssignment API to include the topic
names. Streams needs topic names as well. This also ensures that we
can provide TopicIdPartitions to the assignor without needing to use
the local cache. The assignor's inputs remain entirely driven by the
coordinator.
* Minor updates here and there.

Thank you all for the very good feedback.

Cheers,
David

On Thu, Sep 29, 2022 at 10:25 AM David Jacot <dj...@confluent.io> wrote:
>
> Hi Guozhang,
>
> Thanks for your input.
>
> > 1) in HB request, when clients send the subscription, they are still
> > sending them in terms of topic names. That is, the consumer would not try
> > to keep all the mappings from topic names to topic ids for the subscribed
> > topics.
>
> This is a mistake of mine. The heartbeat request should use Topic Ids.
> I forgot to change it. The idea is that the client will get the
> metadata first to translate the topic names to topic ids and thus also
> have the mapping in its local metadata cache.
>
> > 2) in HB responses, the assignment's topics are encoded as IDs already. And
> > fetch requests can directly rely on that since they just need topic IDs and
> > do not care what actual topic names are.
>
> That's right.
>
> > 3) in prepare-assignment responses, we return subscriptions as topic IDs as
> > well.
>
> That's right.
>
> > In a word, the consumer does not try to maintain the mapping of topic names
> > to IDs at all, and even if they do, they will not expose it in e.g.
> > assignor APIs. But for client-side assignors like Streams, if they do not
> > maintain the mapping from topic IDs to topic names, certain logic such as
> > co-partitioning indeed cannot be executed.
>
> As I said, the consumer will have the mapping in its local metadata
> cache. It needs to get the partition metadata anyway and the topic id
> is part of it. However, I do agree that this mapping is not exposed.
>
> > So I think we'd need to consider 1) use topic names in subscriptions in the
> > prepare-assignment responses, 2) use topic Names in client-side
> > PartitionAssignor#GroupMember#subscribedTopics and
> > PartitionAssignor#TopicMetadata#topicName.
>
> 1) makes sense to me. I think that we could actually provide both
> names and ids. Regarding 2), I think that I am slightly in favor of
> using both names and ids as well. Let me try to update the protocol
> and the interface towards this.
>
> Thanks,
> David
>
> On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello Jun/David,
> >
> > Just to chime in here for Jun's comment #50, today Streams does not have a
> > way to map topic names to topic IDs, and in its topology description topics
> > are encoded with names.
> >
> > What I originally thought was that, on the consumer side, we'd need to
> > maintain this mapping from topic names to topic IDs anyways, and Streams
> > should leverage on the consumer's functionality to get the corresponding
> > topic names for the given topic IDs. But when I read through the RPC again
> > I realized it may not be the case, since:
> >
> > 1) in HB request, when clients send the subscription, they are still
> > sending them in terms of topic names. That is, the consumer would not try
> > to keep all the mappings from topic names to topic ids for the subscribed
> > topics.
> > 2) in HB responses, the assignment's topics are encoded as IDs already. And
> > fetch requests can directly rely on that since they just need topic IDs and
> > do not care what actual topic names are.
> > 3) in prepare-assignment responses, we return subscriptions as topic IDs as
> > well.
> >
> > In a word, the consumer does not try to maintain the mapping of topic names
> > to IDs at all, and even if they do, they will not expose it in e.g.
> > assignor APIs. But for client-side assignors like Streams, if they do not
> > maintain the mapping from topic IDs to topic names, certain logic such as
> > co-partitioning indeed cannot be executed.
> >
> > So I think we'd need to consider 1) use topic names in subscriptions in the
> > prepare-assignment responses, 2) use topic Names in client-side
> > PartitionAssignor#GroupMember#subscribedTopics and
> > PartitionAssignor#TopicMetadata#topicName.
> >
> > On Wed, Sep 28, 2022 at 1:26 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi Jun,
> > >
> > > Thanks for your comments.
> > >
> > > 48. Okay. There is a typo in the text:
> > > > Note that the persisted target assignment of B already includes foo-2
> > > here
> > > but B does not get it yet because foo-2 is not free.
> > >
> > > It should be:
> > > > Note that the persisted member assignment of B already includes foo-2
> > > here
> > > but B does not get it yet because foo-2 is not free.
> > >
> > > The examples are showing the persisted state in the __consumer_offsets
> > > topics. When a member transitions to the next epoch, we directly
> > > persist the whole target assignment in order to reduce the number of
> > > writes and we manage dependencies in memory. So "B - epoch=2,
> > > partitions=[foo-2]" is actually correct from a persisted record
> > > perspective. However, the in-memory state would be different because
> > > foo-2 is not yet available as you pointed out. The in-memory state
> > > would like as follow:
> > >
> > > Group (epoch=2)
> > > A
> > > B
> > > Target Assignment (epoch=2)
> > > A - partitions=[foo-0, foo-1]
> > > B - partitions=[foo-2]
> > > Member Assignment
> > > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending partitions=[]
> > > B - epoch=2, partitions=[], pending partitions=[foo-2]
> > >
> > > I suppose that the confusion comes from this. I will go and convert
> > > all the examples to their in-memory representation in order to reduce
> > > the confusion.
> > >
> > > 50. Okay. I have to admit that I am not so familiar with the internals
> > > of KStreams. If I understand correctly, you're saying that KStreams
> > > may not have topic ids in its topology so topic names are missing here
> > > to match the topics. I was hoping that KStreams would have topic ids
> > > in the topology as well. If that is not the case, we can also add
> > > topic names on our side. I will check this.
> > >
> > > 53. It seems that I was not clear. Sorry for that. I did not mean to
> > > say that MemberAssignment.metadata is not used by the group
> > > coordinator. Let me re-explain the flow.
> > >
> > > PartitionAssignor.assign() returns an Assignment. Assignment contains
> > > MemberAssignments where each member has MemberAssignment.metadata. The
> > > Assignment is installed by the client with the
> > > ConsumerGroupInstallAssignment API and this overrides the current
> > > target assignment in the coordinator. Then the group coordinator uses
> > > it to reconcile members. Therefore the target assignment will
> > > eventually become the current member assignment and it will be
> > > delivered to the member, including the original
> > > MemberAssignment.metadata defines by the assignor.
> > >
> > > I hope that this helps to clarify how MemberAssignment.metadata is
> > > used afterwards.
> > >
> > > Best,
> > > David
> > >
> > > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <ju...@confluent.io.invalid> wrote:
> > > > Hi, David,
> > > >
> > > > Thanks for the reply. A few more follow up questions.
> > > >
> > > > 48. Hmm, so member assignment is what the client sees. Then, I am not
> > > sure
> > > > about the following. The text says B cannot get foo-2 yet. But in Member
> > > > Assignment, B has foo-2.
> > > >
> > > > At this point B can transition to epoch 2 but cannot get foo-2 until A
> > > > revokes it.
> > > >
> > > > Note that the persisted target assignment of B already includes foo-2
> > > here
> > > > but B does not get it yet because foo-2 is not free.
> > > >
> > > > Group (epoch=2)
> > > > A
> > > > B
> > > > Target Assignment (epoch=2)
> > > > A - partitions=[foo-0, foo-1]
> > > > B - partitions=[foo-2]
> > > > Member Assignment
> > > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > > B - epoch=2, partitions=[foo-2]
> > > >
> > > >
> > > > 50. Hmm, I thought KStreams doesn't use range assignor all the time. It
> > > > does the assignment based on the topology. For example, if you are
> > > joining
> > > > two topics A and B and deriving the join result in a 3rd topic C, which
> > > > will be aggregated. You want to use range to co-locate the partitions
> > > from
> > > > topic A and B. However, for topic C, the partition could potentially be
> > > > assigned in round robin. Without the topic name, KStreams won't know
> > > which
> > > > topics need to be co-located.
> > > >
> > > > 53. If MemberAssignment.metadata is not used by the group coordinator,
> > > > where is it being used?
> > > >
> > > > Thanks,
> > > >
> > > > Jun
> > > >
> > > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > > <ja...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi David,
> > > > >
> > > > > Thanks for the response. Replied below:
> > > > >
> > > > > > I don't have a strong opinion on this one. However, it seems that
> > > this
> > > > > could be a pandora box if our subset turns out to not be sufficient
> > > > > for some existing users. We will have to add more to it and the
> > > > > compatibility story between versions could be a real nightmare. One
> > > > > thing that we could do for sure is disabling all unicode groups in the
> > > > > library. That seems possible and would already reduce the scope.
> > > > >
> > > > > Changes to the regex syntax seems not a major problem to me. We would
> > > bump
> > > > > the protocol version as we usually do. Actually that begs the same
> > > question
> > > > > about RE2. How stable is the syntax and does it differ depending on the
> > > > > version of the library? In general, I do agree that we'd be better off
> > > not
> > > > > having to manage the syntax ourselves if we can convince ourselves that
> > > > > compatibility won't be a problem and the client won't eventually need
> > > to
> > > > > have access to it.
> > > > >
> > > > > Thanks,
> > > > > Jason
> > > > >
> > > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > > <dj...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > > > Hi Luke,
> > > > > >
> > > > > > Thanks for your comments.
> > > > > >
> > > > > > Your suggestion makes sense. Jun's made the same one. We are also
> > > > > > discussing whether we should directly return the group state in the
> > > > > > response and drop the prepare assignment api. It seems that the
> > > > > > current consensus is to keep it though. If that stays like this, I
> > > > > > will apply your suggestion.
> > > > > >
> > > > > > Thanks,
> > > > > > David
> > > > > >
> > > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com>
> > > wrote:
> > > > > > >
> > > > > > > Hi David,
> > > > > > >
> > > > > > > Thanks for your patience for answering all these questions!
> > > > > > >
> > > > > > > One more question from me:
> > > > > > > 1. When the selected consumer is going to be in client-side
> > > assigning,
> > > > > we
> > > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> > > wondering if
> > > > > we
> > > > > > > could use a flag to indicate the assignment is required. The
> > > rationale
> > > > > is
> > > > > > > when returning error, we lose a chance to catch up with the latest
> > > > > target
> > > > > > > assignment or revoke partitions. Since we expected the record
> > > fetching
> > > > > is
> > > > > > > not blocked by the rebalance/assignment computing, we might want to
> > > > > allow
> > > > > > > the members to update the member assignment in each heartbeat.
> > > > > > >
> > > > > > > Luke
> > > > > > >
> > > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > > <ja...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hey David,
> > > > > > > >
> > > > > > > > Thanks for the response.
> > > > > > > >
> > > > > > > > > Do you think that we could have a case where two members with
> > > the
> > > > > > same
> > > > > > > > static id are active?
> > > > > > > >
> > > > > > > > I was just thinking about the handoff when we spin up a new node
> > > > > > before the
> > > > > > > > old one completes shutdown. It's probably ok to fence the old
> > > member
> > > > > > > > abruptly. I do recall we had a few issues with this handoff in
> > > the
> > > > > > past. It
> > > > > > > > was due to the fact that the replaced node might have been the
> > > > > leader.
> > > > > > > > Since we've removed the leader notion, perhaps this kind of issue
> > > > > won't
> > > > > > > > resurface.
> > > > > > > >
> > > > > > > > In regard to the RE2 library. I understand the intent to pass
> > > this
> > > > > > through
> > > > > > > > as a string without any validation so that we don't need the
> > > > > > dependence on
> > > > > > > > the client. I do wonder if that will be restrictive in practice.
> > > For
> > > > > > > > example, if a regex changes on the client side, would we need to
> > > > > > validate
> > > > > > > > against the current assignment to see which partitions to
> > > continue
> > > > > > > > fetching? Or would we instead invalidate the complete current
> > > > > > assignment?
> > > > > > > > Perhaps this is a rare case and we can do the latter. There
> > > could be
> > > > > > more
> > > > > > > > situations though.
> > > > > > > >
> > > > > > > > As an alternative, I wonder if we could define a minimalistic
> > > regex
> > > > > > syntax
> > > > > > > > (a subset of RE2 say) that could map more easily to regex support
> > > > > that
> > > > > > > > already exists in clients? For example, in this scenario, we are
> > > > > > matching
> > > > > > > > topics, so we only need to be able to represent topic characters
> > > and
> > > > > > not
> > > > > > > > the whole smorgasbord of whitespace and unicode options. Would it
> > > > > make
> > > > > > > > sense to have a reduced syntax which includes only standard topic
> > > > > > > > characters, wildcards, character classes, and repetitions?
> > > > > > > >
> > > > > > > > Best,
> > > > > > > > Jason
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi David, thanks for all the detailed explanations. I think
> > > they
> > > > > all
> > > > > > make
> > > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > > >
> > > > > > > > > > I don't really see the benefits here because server side
> > > > > assignors
> > > > > > > > > don't have metadata at all. They only assign topic-partitions.
> > > They
> > > > > > > > > are not supposed to generate metadata nor to receive metadata
> > > from
> > > > > > the
> > > > > > > > > members.
> > > > > > > > >
> > > > > > > > > Just to clarify I was asking about the `version` of the
> > > assignor
> > > > > > (i.e. up
> > > > > > > > > to what version that the client would support), and I do agree
> > > we
> > > > > > would
> > > > > > > > not
> > > > > > > > > need metadata. What I have in mind is that, for some specific
> > > > > > built-in
> > > > > > > > > broker-assignors, e.g. rack-aware assignors, if it's possible
> > > that
> > > > > > in a
> > > > > > > > > newer version we would have a hierarchical rack ID string
> > > format,
> > > > > > like
> > > > > > > > > "tier1-tier2" etc, but if some client has not upgraded their
> > > rack
> > > > > ID
> > > > > > > > > would still be in old format. In this case, the broker then
> > > needs
> > > > > to
> > > > > > > > choose
> > > > > > > > > the old versioned assignor. I'm probably making something up
> > > here
> > > > > for
> > > > > > > > rack
> > > > > > > > > aware assignors, but I'm wondering if in general such an
> > > > > > "auto-downgrade"
> > > > > > > > > behavior would be needed still for broker-side assignor, and
> > > if yes
> > > > > > would
> > > > > > > > > "version" still be useful.
> > > > > > > > >
> > > > > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> > > called
> > > > > > once
> > > > > > > > > when the member transitions to a new epoch. This one contains
> > > the
> > > > > > full
> > > > > > > > > metadata provided by the client side assignor. Then,
> > > `onAssignment`
> > > > > > > > > can be called max N times where N is the number of partitions
> > > > > pending
> > > > > > > > > revocation by other members. Let me try to clarify this in the
> > > KIP.
> > > > > > > > >
> > > > > > > > > Okay, my understanding is that the calling ordering of these
> > > > > > callbacks
> > > > > > > > > would be like the following:
> > > > > > > > >
> > > > > > > > > ----------------------------------------
> > > > > > > > > onPartitionsRevoked();   // just once, since we do not really
> > > need
> > > > > > > > > to revoke incrementally.
> > > > > > > > >
> > > > > > > > > onAssignment();    // the first call, with epoch incremented
> > > > > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > > > > >
> > > > > > > > > onAssignment();              // the first onAssignment would
> > > bump
> > > > > up
> > > > > > the
> > > > > > > > > epoch, and the metadata reflected.
> > > > > > > > > onPartitionsAssigned();   // each time we get an additional
> > > > > > assignment,
> > > > > > > > we
> > > > > > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > > > > > ...
> > > > > > > > > onAssignment();
> > > > > > > > > onPartitionsAssigned();   // on each of the onAssignment
> > > calls, the
> > > > > > > > encoded
> > > > > > > > > metadata would not change, only the incrementally added
> > > partitions
> > > > > be
> > > > > > > > > reflected
> > > > > > > > >
> > > > > > > > > Is that the case?
> > > > > > > > >
> > > > > > > > > I'm wondering if we would still call onAssignment just once,
> > > that
> > > > > > encodes
> > > > > > > > > all the assignment for this rebalance, including all the
> > > partitions
> > > > > > that
> > > > > > > > > should be assigned to the member but not yet assigned since
> > > they
> > > > > > have not
> > > > > > > > > been revoked by others. In that case the call ordering would
> > > be:
> > > > > > > > >
> > > > > > > > > ----------------------------------------
> > > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > > onAssignment();    // just once, with epoch incremented, and
> > > > > metadata
> > > > > > > > > encoded changed, the "assignment" field also reflect the final
> > > > > target
> > > > > > > > > assignment
> > > > > > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > > > > > incrementally
> > > > > > > > > added partitions
> > > > > > > > > ...
> > > > > > > > > onPartitionsAssigned();
> > > > > > > > >
> > > > > > > > > The motivation from this is that, most users would only
> > > implement
> > > > > the
> > > > > > > > > rebalance callback listeners and hence we'd definitely need to
> > > make
> > > > > > sure
> > > > > > > > > the semantics of that does not change much, and the time
> > > > > > > > > `onPartitionsAssigned` indicate the time when the partitions
> > > are
> > > > > > actually
> > > > > > > > > assigned to it; while for assignors, the `onAssignment` is
> > > used to
> > > > > > > > indicate
> > > > > > > > > what decision is made regarding for this member, i.e. when the
> > > > > > partitions
> > > > > > > > > are decided to be given to it, but not necessarily meant that
> > > it
> > > > > has
> > > > > > been
> > > > > > > > > given, since that time should be determined by the time of
> > > > > > > > > `onPartitionsAssigned`. The benefits though, would be that
> > > assignor
> > > > > > > > > implementers would not need to reason which `onAssignment`
> > > would be
> > > > > > the
> > > > > > > > > last one for this epoch.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > >
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hi Guozhang,
> > > > > > > > > >
> > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> > > Error
> > > > > > byte
> > > > > > > > > > field,
> > > > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> > > not.
> > > > > > And
> > > > > > > > the
> > > > > > > > > > > rationale seems to be assuming that we should always be
> > > able to
> > > > > > do
> > > > > > > > the
> > > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > > Personally I
> > > > > > > > > think
> > > > > > > > > > > it's still potentially beneficial to add the Error field
> > > even
> > > > > for
> > > > > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > > > > subscribed
> > > > > > > > > > > topics are not recognized with the current broker's
> > > metadata.
> > > > > > What do
> > > > > > > > > you
> > > > > > > > > > > think?
> > > > > > > > > >
> > > > > > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > > > > > better
> > > > > > > > > > to use an exception on the server side. This is what we
> > > usually
> > > > > do
> > > > > > for
> > > > > > > > > > server side plugins. On the client side, we use a field
> > > because
> > > > > the
> > > > > > > > > > errors are not defined in advance.
> > > > > > > > > >
> > > > > > > > > > Your comment also makes me think about what we should do
> > > when the
> > > > > > > > > > server side assignor fails. I suppose that we have to keep
> > > the
> > > > > > current
> > > > > > > > > > assignment until a new event occurs. For instance, in your
> > > > > example,
> > > > > > > > > > the coordinator would have to trigger a rebalance when
> > > > > unrecognized
> > > > > > > > > > topics are available. This would be part of the metadata
> > > > > > monitoring.
> > > > > > > > > >
> > > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > > fields
> > > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > > GroupMember. I
> > > > > > > > > > agree
> > > > > > > > > > > that broker-side assignor would not need reason/metadata
> > > since
> > > > > > they
> > > > > > > > are
> > > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > > version?
> > > > > > E.g.
> > > > > > > > is
> > > > > > > > > > it
> > > > > > > > > > > possible that we evolve our broker-side built-in assignor
> > > but
> > > > > > the old
> > > > > > > > > > > versioned clients would not be able to work with the new
> > > > > > version, in
> > > > > > > > > > which
> > > > > > > > > > > case we need to let the broker being aware of this and
> > > upgrade
> > > > > > its
> > > > > > > > > > behavior
> > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > >
> > > > > > > > > > I don't really see the benefits here because server side
> > > > > assignors
> > > > > > > > > > don't have metadata at all. They only assign
> > > topic-partitions.
> > > > > They
> > > > > > > > > > are not supposed to generate metadata nor to receive metadata
> > > > > from
> > > > > > the
> > > > > > > > > > members.
> > > > > > > > > >
> > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > "GroupMember",
> > > > > > > > instead
> > > > > > > > > > of
> > > > > > > > > > > including these three fields, what about just adding the
> > > > > > "Metadata"
> > > > > > > > > field
> > > > > > > > > > > class which has these three fields? Also, there are two
> > > > > > "Metadata"
> > > > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > > > reason/version/metadata, and the second is just the encoded
> > > > > > metadata
> > > > > > > > > > bytes.
> > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > memberMetadata,
> > > > > > > > which
> > > > > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > > > > second
> > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > >
> > > > > > > > > > That's a good point. Let me try to rationalize this interface
> > > > > > based on
> > > > > > > > > > your suggestions.
> > > > > > > > > >
> > > > > > > > > > Best,
> > > > > > > > > > David
> > > > > > > > > >
> > > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > Hello David,
> > > > > > > > > > >
> > > > > > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > > > > > >
> > > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> > > Error
> > > > > > byte
> > > > > > > > > > field,
> > > > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> > > not.
> > > > > > And
> > > > > > > > the
> > > > > > > > > > > rationale seems to be assuming that we should always be
> > > able to
> > > > > > do
> > > > > > > > the
> > > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > > Personally I
> > > > > > > > > think
> > > > > > > > > > > it's still potentially beneficial to add the Error field
> > > even
> > > > > for
> > > > > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > > > > subscribed
> > > > > > > > > > > topics are not recognized with the current broker's
> > > metadata.
> > > > > > What do
> > > > > > > > > you
> > > > > > > > > > > think?
> > > > > > > > > > >
> > > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > > fields
> > > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > > GroupMember. I
> > > > > > > > > > agree
> > > > > > > > > > > that broker-side assignor would not need reason/metadata
> > > since
> > > > > > they
> > > > > > > > are
> > > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > > version?
> > > > > > E.g.
> > > > > > > > is
> > > > > > > > > > it
> > > > > > > > > > > possible that we evolve our broker-side built-in assignor
> > > but
> > > > > > the old
> > > > > > > > > > > versioned clients would not be able to work with the new
> > > > > > version, in
> > > > > > > > > > which
> > > > > > > > > > > case we need to let the broker being aware of this and
> > > upgrade
> > > > > > its
> > > > > > > > > > behavior
> > > > > > > > > > > to cooperate with the clients?
> > > > > > > > > > >
> > > > > > > > > > > 3) Also related to 2) above, for the client-side
> > > "GroupMember",
> > > > > > > > instead
> > > > > > > > > > of
> > > > > > > > > > > including these three fields, what about just adding the
> > > > > > "Metadata"
> > > > > > > > > field
> > > > > > > > > > > class which has these three fields? Also, there are two
> > > > > > "Metadata"
> > > > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > > > reason/version/metadata, and the second is just the encoded
> > > > > > metadata
> > > > > > > > > > bytes.
> > > > > > > > > > > I'm wondering what about just naming the first as
> > > > > memberMetadata,
> > > > > > > > which
> > > > > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > > > > second
> > > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > > >
> > > > > > > > > > > Guozhang
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > > > wangguoz@gmail.com>
> > > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > > Hello David,
> > > > > > > > > > > >
> > > > > > > > > > > > Thanks for bringing this question up. I think the main
> > > > > > benefits as
> > > > > > > > > you
> > > > > > > > > > > > listed is 2) above if it stays; just to clarify, we would
> > > > > only
> > > > > > be
> > > > > > > > > able
> > > > > > > > > > to
> > > > > > > > > > > > save one round trip if the rebalance is still triggered
> > > by
> > > > > the
> > > > > > > > > broker;
> > > > > > > > > > if
> > > > > > > > > > > > the rebalance is triggered by the client then the
> > > > > > num.round.trips
> > > > > > > > are
> > > > > > > > > > the
> > > > > > > > > > > > same:
> > > > > > > > > > > >
> > > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > > >
> > > > > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > > > > already
> > > > > > > > > sent
> > > > > > > > > > a
> > > > > > > > > > > > HB and hence has to wait for it to return first since
> > > only
> > > > > one
> > > > > > > > > request
> > > > > > > > > > /
> > > > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > request.
> > > > > > > > > > > >
> > > > > > > > > > > > In total, two round trips.
> > > > > > > > > > > >
> > > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > > >
> > > > > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > > > > already
> > > > > > > > > sent
> > > > > > > > > > a
> > > > > > > > > > > > HB and hence has to wait for it to return first since
> > > only
> > > > > one
> > > > > > > > > request
> > > > > > > > > > /
> > > > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > > > T1: client receives the HB response, and then sends the
> > > new
> > > > > HB
> > > > > > > > > request
> > > > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > > > T2: the HB response with the optional member metadata
> > > map is
> > > > > > > > > returned.
> > > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > > request.
> > > > > > > > > > > >
> > > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > > >
> > > > > > > > > > > > -----------------------------
> > > > > > > > > > > >
> > > > > > > > > > > > So to complete the full picture here, we'd need to modify
> > > > > both
> > > > > > HB
> > > > > > > > > > request
> > > > > > > > > > > > and response so that the client can also indicate a new
> > > > > > rebalance
> > > > > > > > via
> > > > > > > > > > the
> > > > > > > > > > > > HB request as well, right?
> > > > > > > > > > > >
> > > > > > > > > > > > Assuming all above is true, I think it's okay to merge
> > > the
> > > > > > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > > > > > additional
> > > > > > > > > > fields
> > > > > > > > > > > > encoding the full member (subscription) metadata and
> > > topic
> > > > > > metadata
> > > > > > > > > as
> > > > > > > > > > > > optional fields.
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > Guozhang
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > >> Hi all,
> > > > > > > > > > > >>
> > > > > > > > > > > >> During an offline conversation, someone asked why we
> > > need
> > > > > the
> > > > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we
> > > > > could
> > > > > > > > > instead
> > > > > > > > > > > >> provide the group state in the heartbeat response. This
> > > has
> > > > > a
> > > > > > few
> > > > > > > > > > > >> advantages: 1) it does not require using a special error
> > > > > code
> > > > > > to
> > > > > > > > > > > >> signal that a new assignment is required as the signal
> > > would
> > > > > > be
> > > > > > > > the
> > > > > > > > > > > >> provided group state; 2) it removes one round trip when
> > > a
> > > > > > client
> > > > > > > > > side
> > > > > > > > > > > >> assignor is used. The downside is that it makes the
> > > > > heartbeat
> > > > > > > > > > > >> response's definition quite large. I recall that I went
> > > with
> > > > > > the
> > > > > > > > > > > >> current approach due to this.
> > > > > > > > > > > >>
> > > > > > > > > > > >> Providing the group state in the heartbeat response is
> > > > > > appealing.
> > > > > > > > > What
> > > > > > > > > > > >> do you guys think?
> > > > > > > > > > > >>
> > > > > > > > > > > >> Best,
> > > > > > > > > > > >> David
> > > > > > > > > > > >>
> > > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > > > djacot@confluent.io>
> > > > > > > > > > wrote:
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > 1. I have added a reference to the relevant chapter
> > > > > instead
> > > > > > of
> > > > > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > 2. The "Rebalance Triggers" section you are referring
> > > to
> > > > > is
> > > > > > > > about
> > > > > > > > > > when
> > > > > > > > > > > >> > a rebalance should be triggered for the non-upgraded
> > > > > members
> > > > > > > > using
> > > > > > > > > > the
> > > > > > > > > > > >> > old protocol. The section mentions that a rebalance
> > > must
> > > > > be
> > > > > > > > > > triggered
> > > > > > > > > > > >> > when a new assignment is installed. This implies that
> > > the
> > > > > > group
> > > > > > > > > > epoch
> > > > > > > > > > > >> > was updated either by a native member or a
> > > non-upgraded
> > > > > > member.
> > > > > > > > > For
> > > > > > > > > > > >> > the latter, the JoinGroup request would be the
> > > trigger. I
> > > > > > have
> > > > > > > > > > added a
> > > > > > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > > > > > Handling"
> > > > > > > > > > section
> > > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > Thanks,
> > > > > > > > > > > >> > David
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > > > > wangguoz@gmail.com
> > > > > > > > > >
> > > > > > > > > > > >> wrote:
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Hello David,
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > > > > > clear in
> > > > > > > > > the
> > > > > > > > > > doc,
> > > > > > > > > > > >> > > could we update:
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > 1) the heartbeat request handling section, stating
> > > when
> > > > > > > > > > coordinator
> > > > > > > > > > > >> will
> > > > > > > > > > > >> > > trigger rebalance based on the HB's member metadata
> > > /
> > > > > > reason?
> > > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to include what
> > > we
> > > > > > > > described
> > > > > > > > > > in
> > > > > > > > > > > >> "Group
> > > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Guozhang
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > wrote:
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > I thought that the assignor will always be
> > > consulted
> > > > > > when
> > > > > > > > the
> > > > > > > > > > next
> > > > > > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be called for
> > > every
> > > > > > > > > heartbeat.
> > > > > > > > > > > >> This
> > > > > > > > > > > >> > > > gives the opportunity for the assignor to enforce
> > > a
> > > > > > > > rebalance
> > > > > > > > > by
> > > > > > > > > > > >> > > > setting the reason to a non-zero value or by
> > > changing
> > > > > > the
> > > > > > > > > > bytes. Do
> > > > > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > > > > concerned by
> > > > > > > > > the
> > > > > > > > > > > >> delay?
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > Best,
> > > > > > > > > > > >> > > > David
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > > >> wrote:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > >> > > > > In this case, a new assignment is triggered by
> > > the
> > > > > > client
> > > > > > > > > side
> > > > > > > > > > > >> > > > > assignor. When constructing the HB, the consumer
> > > > > will
> > > > > > > > always
> > > > > > > > > > > >> consult
> > > > > > > > > > > >> > > > > the client side assignor and propagate the
> > > > > > information to
> > > > > > > > > the
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > coordinator. In other words, we don't expect
> > > users
> > > > > to
> > > > > > call
> > > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > >> > > > > ```
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > > > > > interface,
> > > > > > > > we
> > > > > > > > > > > >> actually do
> > > > > > > > > > > >> > > > > not have a way yet to instruct how to construct
> > > the
> > > > > > next
> > > > > > > > HB
> > > > > > > > > > > >> request, e.g.
> > > > > > > > > > > >> > > > > when the assignor wants to enforce a new
> > > rebalance
> > > > > > with a
> > > > > > > > > new
> > > > > > > > > > > >> assignment,
> > > > > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > > > > PartitionAssignor
> > > > > > > > > > to
> > > > > > > > > > > >> indicate
> > > > > > > > > > > >> > > > > the next HB telling broker about so. WDYT about
> > > > > adding
> > > > > > > > such
> > > > > > > > > an
> > > > > > > > > > > >> API on the
> > > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > I have updated the KIP to include your
> > > feedback. I
> > > > > > have
> > > > > > > > > also
> > > > > > > > > > > >> tried to
> > > > > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > > > > > djacot@confluent.io
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > > > > > answering
> > > > > > > > your
> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > > benefits
> > > > > > of
> > > > > > > > the
> > > > > > > > > > KIP
> > > > > > > > > > > >> seem to
> > > > > > > > > > > >> > > > be
> > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > > > support of
> > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > I would also add that the KIP removes the
> > > global
> > > > > > sync
> > > > > > > > > > barrier
> > > > > > > > > > > >> in the
> > > > > > > > > > > >> > > > > > > protocol which is essential to improve group
> > > > > > stability
> > > > > > > > > and
> > > > > > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > > > > > client by
> > > > > > > > > > moving
> > > > > > > > > > > >> most of
> > > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> > > plan
> > > > > > to
> > > > > > > > > > support
> > > > > > > > > > > >> rolling
> > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > consumers?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Definitely. The group coordinator will use
> > > the
> > > > > > > > assignor
> > > > > > > > > > used
> > > > > > > > > > > >> by a
> > > > > > > > > > > >> > > > > > > majority of the members. This allows the
> > > group
> > > > > to
> > > > > > move
> > > > > > > > > > from
> > > > > > > > > > > >> one
> > > > > > > > > > > >> > > > > > > assignor to another by a roll. This is
> > > explained
> > > > > > in
> > > > > > > > the
> > > > > > > > > > > >> Assignor
> > > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > > whether
> > > > > > it's
> > > > > > > > > > > >> required in
> > > > > > > > > > > >> > > > every
> > > > > > > > > > > >> > > > > > > > request or the scenarios when it needs to
> > > be
> > > > > > filled?
> > > > > > > > > For
> > > > > > > > > > > >> example,
> > > > > > > > > > > >> > > > it's
> > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to
> > > be
> > > > > > filled.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > The client is expected to set those fields
> > > in
> > > > > > case of
> > > > > > > > a
> > > > > > > > > > > >> connection
> > > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > > > > > changed
> > > > > > > > > since
> > > > > > > > > > > >> the last
> > > > > > > > > > > >> > > > > > > HB. The server populates those fields as
> > > long as
> > > > > > the
> > > > > > > > > > member
> > > > > > > > > > > >> is not
> > > > > > > > > > > >> > > > > > > fully reconciled - the member should
> > > acknowledge
> > > > > > that
> > > > > > > > it
> > > > > > > > > > has
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > expected epoch and assignment. I will
> > > clarify
> > > > > > this in
> > > > > > > > > the
> > > > > > > > > > KIP.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> > > rack
> > > > > > > > > affinity
> > > > > > > > > > > >> between the
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > > > fetching,
> > > > > > > > but
> > > > > > > > > > not
> > > > > > > > > > > >> during
> > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> > > the
> > > > > > > > > assignment
> > > > > > > > > > is
> > > > > > > > > > > >> made,
> > > > > > > > > > > >> > > > there
> > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> > > no
> > > > > > replicas
> > > > > > > > > of
> > > > > > > > > > > >> assigned
> > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> > > if we
> > > > > > should
> > > > > > > > > use
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > > GroupMember.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > That's an interesting idea. I don't see any
> > > > > issue
> > > > > > with
> > > > > > > > > > adding
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > rack
> > > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > > useful to
> > > > > > know
> > > > > > > > how
> > > > > > > > > > busy
> > > > > > > > > > > >> a group
> > > > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > > > model,
> > > > > > it
> > > > > > > > > seems
> > > > > > > > > > > >> that we
> > > > > > > > > > > >> > > > could
> > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> > > time
> > > > > > the
> > > > > > > > > event
> > > > > > > > > > > >> loop is
> > > > > > > > > > > >> > > > doing
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > > > > failover
> > > > > > > > > > > >> handling? For
> > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> > > group
> > > > > > with
> > > > > > > > > the
> > > > > > > > > > > >> wildcard
> > > > > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Sure. When the new group coordinator takes
> > > over,
> > > > > > it
> > > > > > > > has
> > > > > > > > > > to:
> > > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > > > > > assignor
> > > > > > > > is
> > > > > > > > > > used.
> > > > > > > > > > > >> We
> > > > > > > > > > > >> > > > don't
> > > > > > > > > > > >> > > > > > > store the information about the member
> > > selected
> > > > > > to run
> > > > > > > > > the
> > > > > > > > > > > >> assignment
> > > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > > > > wildcard
> > > > > > > > > > > >> subscriptions, and
> > > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > > > > document
> > > > > > > > > what
> > > > > > > > > > > >> the epoch
> > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > > one in
> > > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Sure. I will clarify that but it is always
> > > the
> > > > > > latest
> > > > > > > > > > group
> > > > > > > > > > > >> epoch.
> > > > > > > > > > > >> > > > > > > When the group state is updated, the group
> > > epoch
> > > > > > is
> > > > > > > > > > bumped so
> > > > > > > > > > > >> we use
> > > > > > > > > > > >> > > > > > > that one for all the change records related
> > > to
> > > > > the
> > > > > > > > > update.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > > that
> > > > > the
> > > > > > > > > > following
> > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > possible
> > > > > for
> > > > > > a
> > > > > > > > > member
> > > > > > > > > > > >> not to
> > > > > > > > > > > >> > > > get any
> > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > That's right. Here I meant that the members
> > > > > > provided
> > > > > > > > by
> > > > > > > > > > the
> > > > > > > > > > > >> assignor
> > > > > > > > > > > >> > > > > > > in the assignment must exist in the group.
> > > The
> > > > > > > > assignor
> > > > > > > > > > can
> > > > > > > > > > > >> not make
> > > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > > member
> > > > > > epoch
> > > > > > > > > > equals to
> > > > > > > > > > > >> 0":
> > > > > > > > > > > >> > > > When
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > > would be
> > > > > > used?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > > > > partitions and
> > > > > > > > > > > >> rejoins when
> > > > > > > > > > > >> > > > it
> > > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In
> > > this
> > > > > > case,
> > > > > > > > > the
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > coordinator will have removed the member
> > > from
> > > > > the
> > > > > > > > group.
> > > > > > > > > > The
> > > > > > > > > > > >> member
> > > > > > > > > > > >> > > > > > > can rejoin the group with the same member
> > > id but
> > > > > > with
> > > > > > > > 0
> > > > > > > > > as
> > > > > > > > > > > >> epoch. Let
> > > > > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > > > ability
> > > > > > to
> > > > > > > > > > trigger a
> > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> > > by
> > > > > > updating
> > > > > > > > > the
> > > > > > > > > > > >> assignor
> > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > conflicting
> > > > > > with
> > > > > > > > the
> > > > > > > > > > > >> deprecation
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > In this case, a new assignment is triggered
> > > by
> > > > > the
> > > > > > > > > client
> > > > > > > > > > side
> > > > > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> > > consumer
> > > > > > will
> > > > > > > > > > always
> > > > > > > > > > > >> consult
> > > > > > > > > > > >> > > > > > > the client side assignor and propagate the
> > > > > > information
> > > > > > > > > to
> > > > > > > > > > the
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > > > > > users to
> > > > > > > > > call
> > > > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> > > But
> > > > > the
> > > > > > > > > section
> > > > > > > > > > > >> seems to
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> > > 2, B
> > > > > > > > > > immediately
> > > > > > > > > > > >> gets into
> > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > > > > incorrect.
> > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> > > 3, C
> > > > > > seems
> > > > > > > > to
> > > > > > > > > > get
> > > > > > > > > > > >> into
> > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > > still
> > > > > > has A -
> > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > > > support
> > > > > > the
> > > > > > > > > > upgrade
> > > > > > > > > > > >> from
> > > > > > > > > > > >> > > > any
> > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > We will support upgrading from the consumer
> > > > > > protocol
> > > > > > > > > > version
> > > > > > > > > > > >> 3,
> > > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > > > > implemented
> > > > > > yet
> > > > > > > > so
> > > > > > > > > > the
> > > > > > > > > > > >> earliest
> > > > > > > > > > > >> > > > > > > version is unknown at the moment. This is
> > > > > > explained in
> > > > > > > > > the
> > > > > > > > > > > >> migration
> > > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> > > > > > update the
> > > > > > > > > KIP
> > > > > > > > > > > >> based on
> > > > > > > > > > > >> > > > it
> > > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > > >> > > > > > > David
> > > > > > > > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > > >> > > > wrote:
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > > benefits
> > > > > > of
> > > > > > > > the
> > > > > > > > > > KIP
> > > > > > > > > > > >> seem to
> > > > > > > > > > > >> > > > be
> > > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > > > support of
> > > > > > > > > > > >> wildcard. A few
> > > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> > > plan
> > > > > > to
> > > > > > > > > > support
> > > > > > > > > > > >> rolling
> > > > > > > > > > > >> > > > > > changing
> > > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > > consumers?
> > > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > > whether
> > > > > > it's
> > > > > > > > > > > >> required in
> > > > > > > > > > > >> > > > every
> > > > > > > > > > > >> > > > > > > > request or the scenarios when it needs to
> > > be
> > > > > > filled?
> > > > > > > > > For
> > > > > > > > > > > >> example,
> > > > > > > > > > > >> > > > it's
> > > > > > > > > > > >> > > > > > not
> > > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to
> > > be
> > > > > > filled.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> > > rack
> > > > > > > > > affinity
> > > > > > > > > > > >> between the
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > > > fetching,
> > > > > > > > but
> > > > > > > > > > not
> > > > > > > > > > > >> during
> > > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> > > the
> > > > > > > > > assignment
> > > > > > > > > > is
> > > > > > > > > > > >> made,
> > > > > > > > > > > >> > > > there
> > > > > > > > > > > >> > > > > > is
> > > > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> > > no
> > > > > > replicas
> > > > > > > > > of
> > > > > > > > > > > >> assigned
> > > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> > > if we
> > > > > > should
> > > > > > > > > use
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > > GroupMember.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > > useful to
> > > > > > know
> > > > > > > > how
> > > > > > > > > > busy
> > > > > > > > > > > >> a group
> > > > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > > > model,
> > > > > > it
> > > > > > > > > seems
> > > > > > > > > > > >> that we
> > > > > > > > > > > >> > > > could
> > > > > > > > > > > >> > > > > > add
> > > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> > > time
> > > > > > the
> > > > > > > > > event
> > > > > > > > > > > >> loop is
> > > > > > > > > > > >> > > > doing
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > > > > failover
> > > > > > > > > > > >> handling? For
> > > > > > > > > > > >> > > > > > example,
> > > > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> > > group
> > > > > > with
> > > > > > > > > the
> > > > > > > > > > > >> wildcard
> > > > > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > > > > document
> > > > > > > > > what
> > > > > > > > > > > >> the epoch
> > > > > > > > > > > >> > > > > > field
> > > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > > one in
> > > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > > that
> > > > > the
> > > > > > > > > > following
> > > > > > > > > > > >> > > > invariants
> > > > > > > > > > > >> > > > > > are
> > > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > > possible
> > > > > for
> > > > > > a
> > > > > > > > > member
> > > > > > > > > > > >> not to
> > > > > > > > > > > >> > > > get any
> > > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > > member
> > > > > > epoch
> > > > > > > > > > equals to
> > > > > > > > > > > >> 0":
> > > > > > > > > > > >> > > > When
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > > would be
> > > > > > used?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > > > ability
> > > > > > to
> > > > > > > > > > trigger a
> > > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> > > by
> > > > > > updating
> > > > > > > > > the
> > > > > > > > > > > >> assignor
> > > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > > conflicting
> > > > > > with
> > > > > > > > the
> > > > > > > > > > > >> deprecation
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> > > But
> > > > > the
> > > > > > > > > section
> > > > > > > > > > > >> seems to
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> > > 2, B
> > > > > > > > > > immediately
> > > > > > > > > > > >> gets into
> > > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > > > > incorrect.
> > > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> > > 3, C
> > > > > > seems
> > > > > > > > to
> > > > > > > > > > get
> > > > > > > > > > > >> into
> > > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > > still
> > > > > > has A -
> > > > > > > > > > > >> epoch=2,
> > > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > > > support
> > > > > > the
> > > > > > > > > > upgrade
> > > > > > > > > > > >> from
> > > > > > > > > > > >> > > > any
> > > > > > > > > > > >> > > > > > old
> > > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David
> > > Jacot
> > > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > > re-implement the
> > > > > > > > > > coordinator
> > > > > > > > > > > >> in
> > > > > > > > > > > >> > > > Java. I
> > > > > > > > > > > >> > > > > > > > > discussed this offline with a few folks
> > > and
> > > > > > folks
> > > > > > > > > are
> > > > > > > > > > > >> concerned
> > > > > > > > > > > >> > > > that
> > > > > > > > > > > >> > > > > > > > > we could introduce many regressions in
> > > the
> > > > > old
> > > > > > > > > > protocol
> > > > > > > > > > > >> if we do
> > > > > > > > > > > >> > > > so.
> > > > > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > > > > statement
> > > > > > > > from
> > > > > > > > > > the
> > > > > > > > > > > >> KIP. It
> > > > > > > > > > > >> > > > is an
> > > > > > > > > > > >> > > > > > > > > implementation detail after all so it
> > > does
> > > > > not
> > > > > > > > have
> > > > > > > > > > to be
> > > > > > > > > > > >> > > > decided at
> > > > > > > > > > > >> > > > > > > > > this stage. We will likely start by
> > > trying
> > > > > to
> > > > > > > > > > refactor the
> > > > > > > > > > > >> > > > current
> > > > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David
> > > Jacot
> > > > > <
> > > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > > > > "Empty,
> > > > > > > > > > assigning,
> > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> > > States
> > > > > > > > > section,
> > > > > > > > > > > >> right?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > This sentence does not refer to those
> > > > > group
> > > > > > > > states
> > > > > > > > > > but
> > > > > > > > > > > >> rather
> > > > > > > > > > > >> > > > to a
> > > > > > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > > > > > refers to
> > > > > > > > > the
> > > > > > > > > > > >> entire
> > > > > > > > > > > >> > > > state of
> > > > > > > > > > > >> > > > > > > > > > group coordinator which is replicated
> > > via
> > > > > > the
> > > > > > > > log
> > > > > > > > > > > >> layer. I will
> > > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > > > machine
> > > > > > is
> > > > > > > > > > modelled
> > > > > > > > > > > >> as an
> > > > > > > > > > > >> > > > event
> > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > The idea is to follow a model similar
> > > to
> > > > > > the new
> > > > > > > > > > quorum
> > > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > > >> > > > > > > > > > will have N threads to process events.
> > > > > Each
> > > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > > >> > > > > > > > > > partition is assigned to a unique
> > > thread
> > > > > > and all
> > > > > > > > > the
> > > > > > > > > > > >> events
> > > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> > > processed
> > > > > by
> > > > > > this
> > > > > > > > > > > >> thread. This
> > > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > > > > > simulation
> > > > > > > > > > testing
> > > > > > > > > > > >> for the
> > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> > > per
> > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> > > group"
> > > > > > owned
> > > > > > > > > by
> > > > > > > > > > a
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> > > owns 2
> > > > > > > > > consumer
> > > > > > > > > > > >> groups, and
> > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > > have 1
> > > > > > state
> > > > > > > > > > machine
> > > > > > > > > > > >> for it,
> > > > > > > > > > > >> > > > or
> > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> > > there, I
> > > > > > > > think.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > "*group.coordinator.threads"
> > > > > > > > *is
> > > > > > > > > > the
> > > > > > > > > > > >> number
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > > > wondering
> > > > > > > > if
> > > > > > > > > > the
> > > > > > > > > > > >> purpose
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > > > consumer
> > > > > > > > group
> > > > > > > > > > (or
> > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > computation,
> > > > > > why
> > > > > > > > > > should we
> > > > > > > > > > > >> need
> > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an
> > > ability to
> > > > > > shard
> > > > > > > > > the
> > > > > > > > > > > >> > > > processing as
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> > > why
> > > > > > does
> > > > > > > > the
> > > > > > > > > > > >> default
> > > > > > > > > > > >> > > > session
> > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > max(60s)? I
> > > > > > > > thought
> > > > > > > > > > the
> > > > > > > > > > > >> min/max
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > > bound
> > > > > of
> > > > > > it,
> > > > > > > > > no?
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> > > int
> > > > > > 30s The
> > > > > > > > > > > >> timeout to
> > > > > > > > > > > >> > > > detect
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > > group
> > > > > > > > protocol.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.min.session.timeout.ms
> > > > > > int 45s
> > > > > > > > > The
> > > > > > > > > > > >> minimum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.max.session.timeout.ms
> > > > > > int 60s
> > > > > > > > > The
> > > > > > > > > > > >> maximum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > > > > > session
> > > > > > > > > > timeout
> > > > > > > > > > > >> should
> > > > > > > > > > > >> > > > be 45s
> > > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > assignor
> > > > > are
> > > > > > > > > [range,
> > > > > > > > > > > >> uniform],
> > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> > > I'd
> > > > > > like to
> > > > > > > > > > know
> > > > > > > > > > > >> why not
> > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > > > uniform
> > > > > > > > > assignor
> > > > > > > > > > > >> (former
> > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > > > > other
> > > > > > > > reason
> > > > > > > > > > we
> > > > > > > > > > > >> choose
> > > > > > > > > > > >> > > > range
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > > > > uniform
> > > > > > > > The
> > > > > > > > > > > >> server side
> > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > The order on the server side has no
> > > > > > influence
> > > > > > > > > > because
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > client
> > > > > > > > > > > >> > > > > > must
> > > > > > > > > > > >> > > > > > > > > > chose the selector that he wants to
> > > use.
> > > > > > There
> > > > > > > > is
> > > > > > > > > no
> > > > > > > > > > > >> default
> > > > > > > > > > > >> > > > in the
> > > > > > > > > > > >> > > > > > > > > > current proposal. If the assignor is
> > > not
> > > > > > > > specified
> > > > > > > > > > by
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > client,
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > request is rejected. The default
> > > client
> > > > > > value
> > > > > > > > for
> > > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> > > > > though.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > Thanks for your very good comments,
> > > Luke.
> > > > > I
> > > > > > hope
> > > > > > > > > > that my
> > > > > > > > > > > >> > > > answers
> > > > > > > > > > > >> > > > > > help
> > > > > > > > > > > >> > > > > > > > > > to clarify things. I will update the
> > > KIP
> > > > > as
> > > > > > well
> > > > > > > > > > based
> > > > > > > > > > > >> on your
> > > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke
> > > Chen
> > > > > <
> > > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > > > > > mentioned:
> > > > > > > > > > > >> > > > > > > > > > > > The new group coordinator will
> > > have a
> > > > > > state
> > > > > > > > > > machine
> > > > > > > > > > > >> per
> > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions,
> > > where
> > > > > > each
> > > > > > > > > state
> > > > > > > > > > > >> machine is
> > > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > > >> > > > > > > > > > > event loop. Those state machines
> > > will be
> > > > > > > > > executed
> > > > > > > > > > in
> > > > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > > > > "Empty,
> > > > > > > > > > assigning,
> > > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> > > States
> > > > > > > > > section,
> > > > > > > > > > > >> right?
> > > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > > > machine
> > > > > > is
> > > > > > > > > > modelled
> > > > > > > > > > > >> as an
> > > > > > > > > > > >> > > > event
> > > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> > > per
> > > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> > > group"
> > > > > > owned
> > > > > > > > > by
> > > > > > > > > > a
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> > > owns 2
> > > > > > > > > consumer
> > > > > > > > > > > >> groups, and
> > > > > > > > > > > >> > > > > > both
> > > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > > have 1
> > > > > > state
> > > > > > > > > > machine
> > > > > > > > > > > >> for it,
> > > > > > > > > > > >> > > > or
> > > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > > "*group.coordinator.threads"
> > > > > > > > *is
> > > > > > > > > > the
> > > > > > > > > > > >> number
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > threads
> > > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > > > wondering
> > > > > > > > if
> > > > > > > > > > the
> > > > > > > > > > > >> purpose
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > > > consumer
> > > > > > > > group
> > > > > > > > > > (or
> > > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > > computation,
> > > > > > why
> > > > > > > > > > should we
> > > > > > > > > > > >> need
> > > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > > > > > configs:
> > > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> > > why
> > > > > > does
> > > > > > > > the
> > > > > > > > > > > >> default
> > > > > > > > > > > >> > > > session
> > > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > > max(60s)? I
> > > > > > > > thought
> > > > > > > > > > the
> > > > > > > > > > > >> min/max
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > > bound
> > > > > of
> > > > > > it,
> > > > > > > > > no?
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> > > int
> > > > > > 30s The
> > > > > > > > > > > >> timeout to
> > > > > > > > > > > >> > > > detect
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > > group
> > > > > > > > protocol.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.min.session.timeout.ms
> > > > > > int 45s
> > > > > > > > > The
> > > > > > > > > > > >> minimum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > > >
> > > group.consumer.max.session.timeout.ms
> > > > > > int 60s
> > > > > > > > > The
> > > > > > > > > > > >> maximum
> > > > > > > > > > > >> > > > > > session
> > > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > > assignor
> > > > > are
> > > > > > > > > [range,
> > > > > > > > > > > >> uniform],
> > > > > > > > > > > >> > > > > > which means
> > > > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> > > I'd
> > > > > > like to
> > > > > > > > > > know
> > > > > > > > > > > >> why not
> > > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > > > uniform
> > > > > > > > > assignor
> > > > > > > > > > > >> (former
> > > > > > > > > > > >> > > > sticky
> > > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > > > > other
> > > > > > > > reason
> > > > > > > > > > we
> > > > > > > > > > > >> choose
> > > > > > > > > > > >> > > > range
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > > > > uniform
> > > > > > > > The
> > > > > > > > > > > >> server side
> > > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke
> > > > > Chen
> > > > > > <
> > > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> > > > > Connect
> > > > > > > > > > > >> integrating with
> > > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > > >> > > > > > > > > > > > think we should have a separate
> > > > > > discussion
> > > > > > > > > > thread
> > > > > > > > > > > >> for the
> > > > > > > > > > > >> > > > Kafka
> > > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect
> > > With
> > > > > New
> > > > > > > > > Consumer
> > > > > > > > > > > >> Rebalance
> > > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > > >> > > > > > > > > > > > and let this discussion thread
> > > focus
> > > > > on
> > > > > > > > > consumer
> > > > > > > > > > > >> rebalance
> > > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > > >
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM
> > > Sagar
> > > > > <
> > > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > > > > > feedback.
> > > > > > > > > > Looks
> > > > > > > > > > > >> like
> > > > > > > > > > > >> > > > there's
> > > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect.
> > > I
> > > > > > would
> > > > > > > > > > revisit
> > > > > > > > > > > >> the doc
> > > > > > > > > > > >> > > > and
> > > > > > > > > > > >> > > > > > see
> > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM
> > > David
> > > > > > Jacot
> > > > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > > > > > document.
> > > > > > > > > > That's
> > > > > > > > > > > >> really
> > > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that
> > > both
> > > > > > Connect
> > > > > > > > > > and the
> > > > > > > > > > > >> > > > Consumer
> > > > > > > > > > > >> > > > > > could
> > > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > > >> > > > > > > > > > > >> > the same underlying "engine".
> > > The
> > > > > > main
> > > > > > > > > > > >> difference is
> > > > > > > > > > > >> > > > that
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> > > members
> > > > > > > > whereas
> > > > > > > > > > > >> Connect
> > > > > > > > > > > >> > > > assigns
> > > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > > > > > forward:
> > > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed
> > > APIs
> > > > > to
> > > > > > > > > support
> > > > > > > > > > > >> different
> > > > > > > > > > > >> > > > > > resource
> > > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks,
> > > etc.); or
> > > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs
> > > for
> > > > > > Connect.
> > > > > > > > > The
> > > > > > > > > > > >> dedicated
> > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > > > > different on
> > > > > > > > > the
> > > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> > > engine
> > > > > > on the
> > > > > > > > > > > >> coordinator
> > > > > > > > > > > >> > > > side.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> > > > > because
> > > > > > I am
> > > > > > > > > > not a
> > > > > > > > > > > >> fan of
> > > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> > > purposes.
> > > > > > That
> > > > > > > > > being
> > > > > > > > > > > >> said, I am
> > > > > > > > > > > >> > > > not
> > > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant
> > > way to
> > > > > > do
> > > > > > > > it.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > > > > > discuss
> > > > > > > > it
> > > > > > > > > > here
> > > > > > > > > > > >> for now
> > > > > > > > > > > >> > > > in
> > > > > > > > > > > >> > > > > > order
> > > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> > > compatible
> > > > > > with
> > > > > > > > > what
> > > > > > > > > > we
> > > > > > > > > > > >> will do
> > > > > > > > > > > >> > > > for
> > > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM
> > > > > David
> > > > > > > > Jacot
> > > > > > > > > <
> > > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I
> > > will
> > > > > go
> > > > > > > > > through
> > > > > > > > > > and
> > > > > > > > > > > >> address
> > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks
> > > for
> > > > > your
> > > > > > > > > > feedback.
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05
> > > PM
> > > > > > Gregory
> > > > > > > > > > Harris
> > > > > > > > > > > >> <
> > > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > > > > > wonderful to
> > > > > > > > > see
> > > > > > > > > > > >> > > > cooperative
> > > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> > > Alternatives'
> > > > > > > > section
> > > > > > > > > > > >> describes how
> > > > > > > > > > > >> > > > > > member
> > > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step with the
> > > group
> > > > > > epoch
> > > > > > > > > and
> > > > > > > > > > > >> assignment
> > > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea
> > > for
> > > > > the
> > > > > > > > > reasons
> > > > > > > > > > > >> described
> > > > > > > > > > > >> > > > in
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > > > > > assigning
> > > > > > > > > > > >> partitions to a
> > > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > > > > > assignment
> > > > > > > > > > use?
> > > > > > > > > > > >> Are
> > > > > > > > > > > >> > > > member
> > > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > > > > > correspond
> > > > > > > > to
> > > > > > > > > > > >> multiple
> > > > > > > > > > > >> > > > > > different
> > > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's
> > > 'Reason'
> > > > > > field
> > > > > > > > > > opaque
> > > > > > > > > > > >> to the
> > > > > > > > > > > >> > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> > > client-side
> > > > > > > > assignor
> > > > > > > > > > > >> > > > implementations
> > > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is
> > > its
> > > > > > common
> > > > > > > > > > meaning
> > > > > > > > > > > >> agreed
> > > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct
> > > Reason
> > > > > > field
> > > > > > > > over
> > > > > > > > > > > >> including
> > > > > > > > > > > >> > > > such
> > > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> > > included in
> > > > > > the
> > > > > > > > > KIP:
> > > > > > > > > > > >> "Thanks to
> > > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > > > > entirely
> > > > > > > > driven
> > > > > > > > > > by
> > > > > > > > > > > >> the group
> > > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > > > > responsible for
> > > > > > > > > > > >> maintaining
> > > > > > > > > > > >> > > > any
> > > > > > > > > > > >> > > > > > state
> > > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does
> > > this
> > > > > > mean
> > > > > > > > > > that the
> > > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > > > > non-Metadata
> > > > > > > > > state
> > > > > > > > > > > >> (such as
> > > > > > > > > > > >> > > > > > partition
> > > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > > > > > topics,
> > > > > > > > > etc),
> > > > > > > > > > or
> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > additional
> > > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > > > > separate
> > > > > > > > > classes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > > >> > > > > >
> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > > >> > > >
> > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is
> > > it
> > > > > > possible
> > > > > > > > > for
> > > > > > > > > > > >> these two
> > > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve
> > > to
> > > > > > promote
> > > > > > > > > > feature
> > > > > > > > > > > >> parity
> > > > > > > > > > > >> > > > of
> > > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and
> > > > > would
> > > > > > also
> > > > > > > > > > > >> facilitate
> > > > > > > > > > > >> > > > > > operational
> > > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > > > > example,
> > > > > > if a
> > > > > > > > > > > >> server-side
> > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > > > > deploying
> > > > > > > > > the
> > > > > > > > > > > >> patched
> > > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer
> > > > > group
> > > > > > to a
> > > > > > > > > > > >> client-side
> > > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> > > patching
> > > > > all
> > > > > > of
> > > > > > > > the
> > > > > > > > > > > >> brokers.
> > > > > > > > > > > >> > > > With
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > > > > > non-trivial
> > > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two
> > > APIs
> > > > > > have
> > > > > > > > > > diverged
> > > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > > reimplementation
> > > > > > > > would
> > > > > > > > > > not be
> > > > > > > > > > > >> > > > possible.
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at
> > > 8:39 AM
> > > > > > Sagar
> > > > > > > > <
> > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence
> > > page
> > > > > to
> > > > > > > > > discuss
> > > > > > > > > > how
> > > > > > > > > > > >> Connect
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> > > rebalance
> > > > > > > > protocol.
> > > > > > > > > > > >> Here's the
> > > > > > > > > > > >> > > > page:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > > >
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish
> > > and
> > > > > I
> > > > > > have
> > > > > > > > > > tried
> > > > > > > > > > > >> to keep
> > > > > > > > > > > >> > > > a
> > > > > > > > > > > >> > > > > > format
> > > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know
> > > what you
> > > > > > > > think.
> > > > > > > > > > Also,
> > > > > > > > > > > >> do you
> > > > > > > > > > > >> > > > > > think this
> > > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > > > > discussion
> > > > > > > > thread
> > > > > > > > > > or
> > > > > > > > > > > >> is this
> > > > > > > > > > > >> > > > one
> > > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at
> > > 7:37
> > > > > AM
> > > > > > > > > Sagar <
> > > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for
> > > the
> > > > > > doc on
> > > > > > > > > > Kafka
> > > > > > > > > > > >> Streams.
> > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up
> > > with
> > > > > > such a
> > > > > > > > > > > >> document.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at
> > > > > 4:47
> > > > > > AM
> > > > > > > > > > Guozhang
> > > > > > > > > > > >> Wang <
> > > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if
> > > you
> > > > > > could
> > > > > > > > > come
> > > > > > > > > > > >> back with
> > > > > > > > > > > >> > > > some
> > > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect
> > > side
> > > > > > > > > > integration
> > > > > > > > > > > >> with
> > > > > > > > > > > >> > > > the new
> > > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on
> > > the
> > > > > > new
> > > > > > > > > > "protocol
> > > > > > > > > > > >> type"
> > > > > > > > > > > >> > > > we
> > > > > > > > > > > >> > > > > > did not
> > > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > > > > implementations.
> > > > > > > > > > For
> > > > > > > > > > > >> Streams
> > > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >>
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > > >
> > > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your
> > > analysis
> > > > > > on
> > > > > > > > > > Connect.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022
> > > at
> > > > > > 10:48 PM
> > > > > > > > > > Sagar <
> > > > > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > > > > > response.
> > > > > > > > The
> > > > > > > > > > > >> reason I
> > > > > > > > > > > >> > > > thought
> > > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is
> > > that
> > > > > > even
> > > > > > > > > > today the
> > > > > > > > > > > >> > > > connect
> > > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > > > > AbstractCoordinator to
> > > > > > > > > > > >> empower
> > > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors.
> > > The
> > > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> > > metadata()
> > > > > > method
> > > > > > > > > by
> > > > > > > > > > > >> plumbing
> > > > > > > > > > > >> > > > into
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > JoinGroupRequestProtocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes
> > > to
> > > > > > support
> > > > > > > > > > > >> connect would
> > > > > > > > > > > >> > > > be
> > > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in
> > > streams
> > > > > > mainly
> > > > > > > > > > because
> > > > > > > > > > > >> of the
> > > > > > > > > > > >> > > > > > Client
> > > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > > > > > > implementation
> > > > > > > > > > > >> level, we
> > > > > > > > > > > >> > > > > > might
> > > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto
> > > this
> > > > > > new
> > > > > > > > > > > >> assignment
> > > > > > > > > > > >> > > > protocol
> > > > > > > > > > > >> > > > > > like
> > > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > > > request/response and
> > > > > > > > > > > >> SyncGroup and
> > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on
> > > similar
> > > > > > lines
> > > > > > > > > to
> > > > > > > > > > > >> streams (or
> > > > > > > > > > > >> > > > > > there
> > > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> > > perform a
> > > > > > > > detailed
> > > > > > > > > > > >> analysis of
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > same
> > > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> > > > > > thread for
> > > > > > > > > > that
> > > > > > > > > > > >> as that
> > > > > > > > > > > >> > > > would
> > > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know
> > > if
> > > > > > that
> > > > > > > > > sounds
> > > > > > > > > > > >> good to
> > > > > > > > > > > >> > > > you.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15,
> > > 2022 at
> > > > > > 5:47
> > > > > > > > PM
> > > > > > > > > > David
> > > > > > > > > > > >> Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > > > > > comments.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That
> > > refers to
> > > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > > > > > > transition C
> > > > > > > > > > from
> > > > > > > > > > > >> his
> > > > > > > > > > > >> > > > current
> > > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment
> > > when
> > > > > > he
> > > > > > > > can
> > > > > > > > > > move
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > > >> > > > > > When
> > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > > is
> > > > > > updated
> > > > > > > > > and
> > > > > > > > > > > >> persisted
> > > > > > > > > > > >> > > > > > with all
> > > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if
> > > > > they
> > > > > > are
> > > > > > > > > not
> > > > > > > > > > all
> > > > > > > > > > > >> revoked
> > > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > > > > > becomes
> > > > > > > > the
> > > > > > > > > > target
> > > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to
> > > avoid
> > > > > > having
> > > > > > > > > to
> > > > > > > > > > > >> write all
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are
> > > based on
> > > > > > the
> > > > > > > > > > persisted
> > > > > > > > > > > >> state
> > > > > > > > > > > >> > > > so I
> > > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me
> > > see
> > > > > > if I
> > > > > > > > can
> > > > > > > > > > > >> improve
> > > > > > > > > > > >> > > > this in
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding
> > > Connect,
> > > > > it
> > > > > > > > could
> > > > > > > > > > > >> reuse the
> > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it
> > > fits in
> > > > > > the
> > > > > > > > > > > >> protocol. The
> > > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > topicid-partitions +
> > > > > > > > metadata,
> > > > > > > > > > could
> > > > > > > > > > > >> > > > Connect
> > > > > > > > > > > >> > > > > > fit
> > > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15,
> > > 2022
> > > > > at
> > > > > > 1:55
> > > > > > > > > PM
> > > > > > > > > > > >> Sagar <
> > > > > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the
> > > KIP.
> > > > > I
> > > > > > just
> > > > > > > > > had
> > > > > > > > > > > >> minor
> > > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the
> > > Assignment
> > > > > > Error
> > > > > > > > > > > >> section in
> > > > > > > > > > > >> > > > Client
> > > > > > > > > > > >> > > > > > Side
> > > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned
> > > => `In
> > > > > > this
> > > > > > > > > > case,
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > client
> > > > > > > > > > > >> > > > > > side
> > > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the
> > > group
> > > > > > > > > > coordinator`.
> > > > > > > > > > > >> In this
> > > > > > > > > > > >> > > > > > case are
> > > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> > > returning
> > > > > an
> > > > > > > > > > > >> AssignmentError
> > > > > > > > > > > >> > > > that's
> > > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you
> > > think
> > > > > > it
> > > > > > > > > would
> > > > > > > > > > > >> make sense
> > > > > > > > > > > >> > > > to
> > > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case
> > > > > Studies
> > > > > > > > > > section, I
> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > > slight
> > > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the
> > > > > same.
> > > > > > > > > Consider
> > > > > > > > > > > >> this step:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> > > heartbeats,
> > > > > the
> > > > > > > > group
> > > > > > > > > > > >> coordinator
> > > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> > > partitions
> > > > > to
> > > > > > > > > > revoke. It
> > > > > > > > > > > >> > > > persists
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group
> > > (epoch=3)
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > > > > Assignment
> > > > > > > > > > (epoch=3)
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > > > > partitions=[foo-0]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > > > > partitions=[foo-2]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > > > > partitions=[foo-1]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > > > > Assignment
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > epoch=2,
> > > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > epoch=3,
> > > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > epoch=3,
> > > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> > > heartbeats,
> > > > > it
> > > > > > > > > > transitions
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > > >> > > > > > but
> > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's
> > > mentioned
> > > > > > that
> > > > > > > > > > member C
> > > > > > > > > > > >> can't
> > > > > > > > > > > >> > > > get
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > > > > description
> > > > > > > > > > above,
> > > > > > > > > > > >> it seems
> > > > > > > > > > > >> > > > it
> > > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better
> > > to
> > > > > > remove
> > > > > > > > it
> > > > > > > > > > and
> > > > > > > > > > > >> > > > populate it
> > > > > > > > > > > >> > > > > > only
> > > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a
> > > lot
> > > > > of
> > > > > > > > other
> > > > > > > > > > > >> places, so
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > I
> > > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding
> > > connect ,
> > > > > it
> > > > > > > > might
> > > > > > > > > > be
> > > > > > > > > > > >> out of
> > > > > > > > > > > >> > > > scope
> > > > > > > > > > > >> > > > > > of
> > > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> > > understood it
> > > > > > would
> > > > > > > > > > > >> probably be
> > > > > > > > > > > >> > > > > > running in
> > > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> > > > > > rebalance
> > > > > > > > > > > >> protocol as it
> > > > > > > > > > > >> > > > has
> > > > > > > > > > > >> > > > > > its
> > > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > > IncrementalCooperative).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15,
> > > 2022
> > > > > > at
> > > > > > > > 5:00
> > > > > > > > > > PM
> > > > > > > > > > > >> David
> > > > > > > > > > > >> > > > Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> > > Hector! Our
> > > > > > goal
> > > > > > > > is
> > > > > > > > > > to
> > > > > > > > > > > >> move
> > > > > > > > > > > >> > > > forward
> > > > > > > > > > > >> > > > > > with
> > > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of
> > > relying
> > > > > > on
> > > > > > > > one
> > > > > > > > > > > >> generic API.
> > > > > > > > > > > >> > > > For
> > > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same
> > > pattern
> > > > > > and
> > > > > > > > > > > >> reuse/share the
> > > > > > > > > > > >> > > > core
> > > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side.
> > > For
> > > > > the
> > > > > > > > > schema
> > > > > > > > > > > >> registry, I
> > > > > > > > > > > >> > > > > > think
> > > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a
> > > tailored
> > > > > > API to
> > > > > > > > > do
> > > > > > > > > > > >> simple
> > > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul
> > > 15,
> > > > > > 2022 at
> > > > > > > > > > 10:22
> > > > > > > > > > > >> AM Ismael
> > > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> > > > > > comments:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > > >> > > > vs
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > com.google.re2j.Pattern,
> > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > > > > > > differences
> > > > > > > > > > in
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > detail
> > > > > > > > > > > >> > > > > > before
> > > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said,
> > > if
> > > > > > people
> > > > > > > > > pass
> > > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics
> > > to be
> > > > > > > > honored.
> > > > > > > > > > If
> > > > > > > > > > > >> we are
> > > > > > > > > > > >> > > > doing
> > > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should
> > > consider
> > > > > > adding
> > > > > > > > > an
> > > > > > > > > > > >> overload
> > > > > > > > > > > >> > > > with
> > > > > > > > > > > >> > > > > > our own
> > > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd
> > > want
> > > > > to
> > > > > > > > > expose
> > > > > > > > > > > >> re2j's at
> > > > > > > > > > > >> > > > this
> > > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding
> > > > > topic
> > > > > > > > ids,
> > > > > > > > > > any
> > > > > > > > > > > >> major new
> > > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and
> > > should
> > > > > > handle
> > > > > > > > the
> > > > > > > > > > topic
> > > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we
> > > need to
> > > > > > > > handle.
> > > > > > > > > I
> > > > > > > > > > > >> agree with
> > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the
> > > relevant
> > > > > > > > > protocols
> > > > > > > > > > > >> that don't
> > > > > > > > > > > >> > > > > > have it
> > > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> > > > > > internals
> > > > > > > > > > versus
> > > > > > > > > > > >> adding
> > > > > > > > > > > >> > > > new
> > > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> > > > > > adding
> > > > > > > > new
> > > > > > > > > > APIs
> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > required
> > > > > > > > > > > >> > > > > > for
> > > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am
> > > still
> > > > > not
> > > > > > sure
> > > > > > > > > > about
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> > > > > > configs
> > > > > > > > to
> > > > > > > > > be
> > > > > > > > > > > >> > > > centralized
> > > > > > > > > > > >> > > > > > in the
> > > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > > > > (auditability,
> > > > > > > > > > > >> visibility,
> > > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > > > > > automatically
> > > > > > > > > > > >> deleting
> > > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A
> > > > > good
> > > > > > > > > property
> > > > > > > > > > > >> for modern
> > > > > > > > > > > >> > > > > > systems
> > > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > unrecoverable
> > > > > data
> > > > > > > > loss
> > > > > > > > > > > >> scenarios.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul
> > > 13,
> > > > > > 2022
> > > > > > > > at
> > > > > > > > > > 3:47
> > > > > > > > > > > >> PM David
> > > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > > > > Guozhang. My
> > > > > > > > > > answers
> > > > > > > > > > > >> are
> > > > > > > > > > > >> > > > below:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > > > migration
> > > > > > > > > path,
> > > > > > > > > > > >> especially
> > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > enable
> > > > > the
> > > > > > new
> > > > > > > > > > > >> protocol, in
> > > > > > > > > > > >> > > > which
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > protocols /
> > > > > > rpcs
> > > > > > > > and
> > > > > > > > > > old
> > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> > > How
> > > > > the
> > > > > > > > > > > >> coordinator could
> > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > > protocol
> > > > > > is
> > > > > > > > > > something
> > > > > > > > > > > >> we
> > > > > > > > > > > >> > > > need to
> > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I
> > > just
> > > > > > > > > published
> > > > > > > > > > a
> > > > > > > > > > > >> new
> > > > > > > > > > > >> > > > version
> > > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details
> > > about
> > > > > > this.
> > > > > > > > > See
> > > > > > > > > > the
> > > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > > > > > "Compatibility,
> > > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > > >> > > > > > and
> > > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to
> > > > > think
> > > > > > > > > through
> > > > > > > > > > a
> > > > > > > > > > > >> few cases
> > > > > > > > > > > >> > > > > > now but
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> > > > > > should be
> > > > > > > > > > > >> understandable.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> > > usage
> > > > > > of
> > > > > > > > > topic
> > > > > > > > > > > >> ids. So
> > > > > > > > > > > >> > > > far as
> > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> > > RPCs
> > > > > > and
> > > > > > > > > admin
> > > > > > > > > > > >> client,
> > > > > > > > > > > >> > > > but
> > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > consumers
> > > > > > yet.
> > > > > > > > I
> > > > > > > > > > think
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> > > to be
> > > > > > > > > > maintaining
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > names
> > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> > > the
> > > > > > > > augmented
> > > > > > > > > > > >> existing
> > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > secondly,
> > > > > > should
> > > > > > > > we
> > > > > > > > > > ever
> > > > > > > > > > > >> consider
> > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> > > APIs
> > > > > as
> > > > > > > > well
> > > > > > > > > > (both
> > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > listener for
> > > > > > cases
> > > > > > > > > > like
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> > > Assuming
> > > > > > that we
> > > > > > > > > > would
> > > > > > > > > > > >> include
> > > > > > > > > > > >> > > > > > converting
> > > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to
> > > using
> > > > > > topic
> > > > > > > > > ids
> > > > > > > > > > in
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > KIP,
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully
> > > operate
> > > > > > with
> > > > > > > > > topic
> > > > > > > > > > > >> ids. That
> > > > > > > > > > > >> > > > > > being
> > > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > > > > names in
> > > > > > > > > > various
> > > > > > > > > > > >> APIs so
> > > > > > > > > > > >> > > > > > having a
> > > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > > > > inevitable to
> > > > > > > > > me.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't
> > > > > have
> > > > > > a
> > > > > > > > > strong
> > > > > > > > > > > >> opinion on
> > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> > > > > > scope of
> > > > > > > > > this
> > > > > > > > > > > >> KIP. I
> > > > > > > > > > > >> > > > would
> > > > > > > > > > > >> > > > > > rather
> > > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and
> > > we
> > > > > can
> > > > > > > > > consider
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming
> > > back to
> > > > > > > > > Ismael's
> > > > > > > > > > > >> point about
> > > > > > > > > > > >> > > > > > using
> > > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > > >> > > > think
> > > > > > > > > > > >> > > > > > that
> > > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of
> > > it.
> > > > > > The
> > > > > > > > > > consumer
> > > > > > > > > > > >> will
> > > > > > > > > > > >> > > > have
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > > > > exists
> > > > > > > > > before
> > > > > > > > > > > >> passing
> > > > > > > > > > > >> > > > them
> > > > > > > > > > > >> > > > > > into
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> > > > > > Obviously,
> > > > > > > > > the
> > > > > > > > > > > >> > > > coordinator
> > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really
> > > > > have
> > > > > > a
> > > > > > > > way
> > > > > > > > > to
> > > > > > > > > > > >> reject an
> > > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > > agreeing
> > > > > > with
> > > > > > > > > > David
> > > > > > > > > > > >> on all
> > > > > > > > > > > >> > > > other
> > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > >> question:
> > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > deprecate
> > > > > the
> > > > > > > > > > subscribe
> > > > > > > > > > > >> API with
> > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> > > just
> > > > > > > > > documenting
> > > > > > > > > > > >> that our
> > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> > > the
> > > > > > > > > incompatible
> > > > > > > > > > > >> case is a
> > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > overloaded
> > > > > > > > `String`
> > > > > > > > > > based
> > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > > > > regexes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> > > > > > work. I
> > > > > > > > > have
> > > > > > > > > > to
> > > > > > > > > > > >> look at
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> > > > > > better
> > > > > > > > > > > >> understand the
> > > > > > > > > > > >> > > > > > potential
> > > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would
> > > > > work
> > > > > > for
> > > > > > > > > all
> > > > > > > > > > the
> > > > > > > > > > > >> basic
> > > > > > > > > > > >> > > > > > regular
> > > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between
> > > the
> > > > > two
> > > > > > are
> > > > > > > > > > mainly
> > > > > > > > > > > >> about
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > various
> > > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder
> > > what
> > > > > > other
> > > > > > > > > people
> > > > > > > > > > > >> think
> > > > > > > > > > > >> > > > about
> > > > > > > > > > > >> > > > > > this.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue,
> > > Jul
> > > > > 12,
> > > > > > 2022
> > > > > > > > > at
> > > > > > > > > > > >> 11:28 PM
> > > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks
> > > > > David!
> > > > > > I
> > > > > > > > > think
> > > > > > > > > > on
> > > > > > > > > > > >> the high
> > > > > > > > > > > >> > > > > > level
> > > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > > concretize
> > > > > > a
> > > > > > > > bit
> > > > > > > > > > more:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > > > migration
> > > > > > > > > path,
> > > > > > > > > > > >> especially
> > > > > > > > > > > >> > > > > > the last
> > > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > enable
> > > > > the
> > > > > > new
> > > > > > > > > > > >> protocol, in
> > > > > > > > > > > >> > > > which
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > protocols /
> > > > > > rpcs
> > > > > > > > and
> > > > > > > > > > old
> > > > > > > > > > > >> > > > protocols /
> > > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> > > How
> > > > > the
> > > > > > > > > > > >> coordinator could
> > > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > > protocol
> > > > > > is
> > > > > > > > > > something
> > > > > > > > > > > >> we
> > > > > > > > > > > >> > > > need to
> > > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> > > usage
> > > > > > of
> > > > > > > > > topic
> > > > > > > > > > > >> ids. So
> > > > > > > > > > > >> > > > far as
> > > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> > > RPCs
> > > > > > and
> > > > > > > > > admin
> > > > > > > > > > > >> client,
> > > > > > > > > > > >> > > > but
> > > > > > > > > > > >> > > > > > they
> > > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > consumers
> > > > > > yet.
> > > > > > > > I
> > > > > > > > > > think
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> > > to be
> > > > > > > > > > maintaining
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > names
> > > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> > > the
> > > > > > > > augmented
> > > > > > > > > > > >> existing
> > > > > > > > > > > >> > > > RPCs
> > > > > > > > > > > >> > > > > > and the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > secondly,
> > > > > > should
> > > > > > > > we
> > > > > > > > > > ever
> > > > > > > > > > > >> consider
> > > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> > > APIs
> > > > > as
> > > > > > > > well
> > > > > > > > > > (both
> > > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > listener for
> > > > > > cases
> > > > > > > > > > like
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > > agreeing
> > > > > > with
> > > > > > > > > > David
> > > > > > > > > > > >> on all
> > > > > > > > > > > >> > > > other
> > > > > > > > > > > >> > > > > > minor
> > > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > `subscribe(Pattern)`
> > > > > > > > > > > >> question:
> > > > > > > > > > > >> > > > > > personally
> > > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > deprecate
> > > > > the
> > > > > > > > > > subscribe
> > > > > > > > > > > >> API with
> > > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> > > just
> > > > > > > > > documenting
> > > > > > > > > > > >> that our
> > > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> > > the
> > > > > > > > > incompatible
> > > > > > > > > > > >> case is a
> > > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > overloaded
> > > > > > > > `String`
> > > > > > > > > > based
> > > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > > > > regexes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue,
> > > Jul
> > > > > > 12,
> > > > > > > > 2022
> > > > > > > > > > at
> > > > > > > > > > > >> 5:23 AM
> > > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi
> > > Ismael,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > Thanks for
> > > > > > your
> > > > > > > > > > > >> feedback. Let
> > > > > > > > > > > >> > > > me
> > > > > > > > > > > >> > > > > > answer
> > > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I
> > > > > think
> > > > > > > > it's
> > > > > > > > > > > >> premature to
> > > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > deprecation
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > removal
> > > > > > of the
> > > > > > > > > > > >> existing group
> > > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > client
> > > > > > > > protocol
> > > > > > > > > > and
> > > > > > > > > > > >> hence
> > > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> > > > > > quickly
> > > > > > > > > > > >> applications
> > > > > > > > > > > >> > > > migrate
> > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That
> > > makes
> > > > > > > > sense.
> > > > > > > > > I
> > > > > > > > > > will
> > > > > > > > > > > >> > > > remove it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2.
> > > The
> > > > > KIP
> > > > > > > > says
> > > > > > > > > we
> > > > > > > > > > > >> intend to
> > > > > > > > > > > >> > > > > > release
> > > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > why. If
> > > > > we
> > > > > > > > added
> > > > > > > > > > that
> > > > > > > > > > > >> as a
> > > > > > > > > > > >> > > > way to
> > > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > group
> > > > > > > > > > protocol, I
> > > > > > > > > > > >> also
> > > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > > > explain
> > > > > > > > my
> > > > > > > > > > > >> reasoning. As
> > > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > coordinator
> > > > > > in
> > > > > > > > > Java
> > > > > > > > > > > >> while we
> > > > > > > > > > > >> > > > > > implement
> > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > the
> > > > > > > > internals
> > > > > > > > > > will
> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > slightly
> > > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > Therefore, I
> > > > > > > > > wanted
> > > > > > > > > > to
> > > > > > > > > > > >> tighten
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > coordinator
> > > > > > to
> > > > > > > > the
> > > > > > > > > > new
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > alternative
> > > > > > > > would
> > > > > > > > > > be to
> > > > > > > > > > > >> use a
> > > > > > > > > > > >> > > > flag
> > > > > > > > > > > >> > > > > > to do
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> > > > > > software
> > > > > > > > > > upgrade.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3.
> > > We
> > > > > > need to
> > > > > > > > > > flesh
> > > > > > > > > > > >> out the
> > > > > > > > > > > >> > > > > > details of
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > we're
> > > > > > saying
> > > > > > > > we
> > > > > > > > > > will
> > > > > > > > > > > >> support
> > > > > > > > > > > >> > > > > > online
> > > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > explain
> > > > > > this
> > > > > > > > in
> > > > > > > > > > > >> detail. It
> > > > > > > > > > > >> > > > could
> > > > > > > > > > > >> > > > > > also
> > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > easier.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes,
> > > we
> > > > > will
> > > > > > > > > support
> > > > > > > > > > > >> online
> > > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > group
> > > > > > using
> > > > > > > > the
> > > > > > > > > > old
> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > > > briefly
> > > > > > > > > > explain
> > > > > > > > > > > >> how that
> > > > > > > > > > > >> > > > > > will work
> > > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four
> > > step
> > > > > > > > process:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> > > > > > cluster
> > > > > > > > > must
> > > > > > > > > > be
> > > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > group
> > > > > > > > > > coordinator.
> > > > > > > > > > > >> Both the
> > > > > > > > > > > >> > > > > > old and
> > > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > support
> > > > > the
> > > > > > old
> > > > > > > > > > > >> protocol and
> > > > > > > > > > > >> > > > rely
> > > > > > > > > > > >> > > > > > on the
> > > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they
> > > can
> > > > > > work
> > > > > > > > > > together.
> > > > > > > > > > > >> This
> > > > > > > > > > > >> > > > point
> > > > > > > > > > > >> > > > > > is an
> > > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > this
> > > > > one
> > > > > > live
> > > > > > > > > > > >> because it
> > > > > > > > > > > >> > > > would
> > > > > > > > > > > >> > > > > > require
> > > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > coordinator
> > > > > > and
> > > > > > > > > > > >> starting up the
> > > > > > > > > > > >> > > > > > new one
> > > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > unavailabilities.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> > > > > > cluster's
> > > > > > > > > > > >> metadata
> > > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > enable
> > > > > > the
> > > > > > > > new
> > > > > > > > > > > >> protocol.
> > > > > > > > > > > >> > > > This
> > > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > terminated
> > > > > > > > because
> > > > > > > > > > the
> > > > > > > > > > > >> old
> > > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > protocol.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> > > > > > consumers
> > > > > > > > > > must be
> > > > > > > > > > > >> > > > upgraded
> > > > > > > > > > > >> > > > > > to a
> > > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > migration
> > > > > > (must
> > > > > > > > > have
> > > > > > > > > > > >> KIP-792).
> > > > > > > > > > > >> > > > If
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > Nothing
> > > > > > must be
> > > > > > > > > > done at
> > > > > > > > > > > >> this
> > > > > > > > > > > >> > > > point.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> > > > > > consumers
> > > > > > > > > > must
> > > > > > > > > > > >> be rolled
> > > > > > > > > > > >> > > > > > with the
> > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > consumer
> > > > > > group
> > > > > > > > is
> > > > > > > > > > > >> automatically
> > > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using
> > > the
> > > > > > new
> > > > > > > > > > protocol
> > > > > > > > > > > >> joins
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > group.
> > > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > > > > > protocol are
> > > > > > > > > > being
> > > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > > >> > > > > > the old
> > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > one.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > > > clarify
> > > > > > > > all
> > > > > > > > > > of
> > > > > > > > > > > >> this in
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4.
> > > I am
> > > > > > happy
> > > > > > > > > > that we
> > > > > > > > > > > >> are
> > > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > server,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but
> > > it
> > > > > > seems
> > > > > > > > > like
> > > > > > > > > > > >> there
> > > > > > > > > > > >> > > > could be
> > > > > > > > > > > >> > > > > > some
> > > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > have a
> > > > > > > > mechanism
> > > > > > > > > > for
> > > > > > > > > > > >> users to
> > > > > > > > > > > >> > > > > > detect
> > > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > before
> > > > > > > > switching
> > > > > > > > > > to
> > > > > > > > > > > >> the new
> > > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > think
> > > > > > that I
> > > > > > > > am
> > > > > > > > > a
> > > > > > > > > > bit
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > believe
> > > > > > that the
> > > > > > > > > > > >> majority of
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > cases
> > > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work
> > > with
> > > > > > the
> > > > > > > > new
> > > > > > > > > > > >> engine. The
> > > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > reject
> > > > > > the
> > > > > > > > > > consumer
> > > > > > > > > > > >> if the
> > > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > migration
> > > > > > path,
> > > > > > > > in
> > > > > > > > > > the
> > > > > > > > > > > >> worst
> > > > > > > > > > > >> > > > case,
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > group
> > > > > > will
> > > > > > > > be
> > > > > > > > > > > >> rejected.
> > > > > > > > > > > >> > > > This
> > > > > > > > > > > >> > > > > > should
> > > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > say.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One
> > > way
> > > > > for
> > > > > > > > > > customers to
> > > > > > > > > > > >> > > > validate
> > > > > > > > > > > >> > > > > > their
> > > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod
> > > would
> > > > > > be to
> > > > > > > > > > test
> > > > > > > > > > > >> them with
> > > > > > > > > > > >> > > > > > another
> > > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could
> > > be
> > > > > > done
> > > > > > > > in a
> > > > > > > > > > > >> pre-prod
> > > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > extend the
> > > > > > > > > > > >> consumer-group tool
> > > > > > > > > > > >> > > > to
> > > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > mechanism.
> > > > > > Would
> > > > > > > > > > this be
> > > > > > > > > > > >> > > > enough in
> > > > > > > > > > > >> > > > > > your
> > > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > > > > > Related to
> > > > > > > > > the
> > > > > > > > > > last
> > > > > > > > > > > >> > > > question,
> > > > > > > > > > > >> > > > > > will
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > stick
> > > > > > with the
> > > > > > > > > > > >> current regex
> > > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > example,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it
> > > may
> > > > > be
> > > > > > > > handy
> > > > > > > > > to
> > > > > > > > > > > >> keep using
> > > > > > > > > > > >> > > > > > client
> > > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > migrations
> > > > > > > > > simple
> > > > > > > > > > and
> > > > > > > > > > > >> then
> > > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > > understand
> > > > > > > > your
> > > > > > > > > > point
> > > > > > > > > > > >> but I
> > > > > > > > > > > >> > > > am
> > > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > actually
> > > > > > stay
> > > > > > > > > in
> > > > > > > > > > > >> this mode.
> > > > > > > > > > > >> > > > That
> > > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > simplifying
> > > > > > the
> > > > > > > > > > client
> > > > > > > > > > > >> because
> > > > > > > > > > > >> > > > we
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > metadata on
> > > > > > > > > the
> > > > > > > > > > > >> client
> > > > > > > > > > > >> > > > side. I
> > > > > > > > > > > >> > > > > > would
> > > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6.
> > > When
> > > > > > we say
> > > > > > > > > > that
> > > > > > > > > > > >> the group
> > > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > > > > > configurations
> > > > > > > > > > > >> and that
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > group is
> > > > > > > > > deleted.
> > > > > > > > > > > >> Will a
> > > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > configurations?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> > > > > > right.
> > > > > > > > The
> > > > > > > > > > > >> > > > configurations
> > > > > > > > > > > >> > > > > > will be
> > > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > deleted.
> > > > > > They go
> > > > > > > > > > > >> together.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7.
> > > Will
> > > > > > the
> > > > > > > > > > choice to
> > > > > > > > > > > >> store
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > harder
> > > > > to
> > > > > > list
> > > > > > > > > all
> > > > > > > > > > > >> cluster
> > > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > don't
> > > > > > think
> > > > > > > > so.
> > > > > > > > > > The
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > configs.
> > > > > If
> > > > > > you
> > > > > > > > > > want to
> > > > > > > > > > > >> know
> > > > > > > > > > > >> > > > all
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > ask all
> > > > > > the
> > > > > > > > > group
> > > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > > >> > > > > > You
> > > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > instance.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8.
> > > How
> > > > > > would
> > > > > > > > > > someone
> > > > > > > > > > > >> > > > configure a
> > > > > > > > > > > >> > > > > > group
> > > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > > > > > considered
> > > > > > > > > > > >> allowing the
> > > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > Alternatively,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > configs
> > > > > > could
> > > > > > > > be
> > > > > > > > > > > >> decoupled
> > > > > > > > > > > >> > > > from
> > > > > > > > > > > >> > > > > > the
> > > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > The
> > > > > > group
> > > > > > > > > will
> > > > > > > > > > be
> > > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > configs
> > > > > > will
> > > > > > > > > be
> > > > > > > > > > > >> lost after
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > passes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9.
> > > Will
> > > > > > the
> > > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > > >> > > > > > method
> > > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > > >> > > > do
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > That's a
> > > > > > very
> > > > > > > > > group
> > > > > > > > > > > >> question. I
> > > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > > >> > > > not
> > > > > > > > > > > >> > > > > > fully
> > > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > plan to
> > > > > > use,
> > > > > > > > it
> > > > > > > > > > > >> might be
> > > > > > > > > > > >> > > > better
> > > > > > > > > > > >> > > > > > to
> > > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which
> > > > > takes
> > > > > > a
> > > > > > > > > > string.
> > > > > > > > > > > >> We would
> > > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > During the
> > > > > > > > > > migration, I
> > > > > > > > > > > >> think
> > > > > > > > > > > >> > > > that
> > > > > > > > > > > >> > > > > > we
> > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > and
> > > > > > use
> > > > > > > > it.
> > > > > > > > > > That
> > > > > > > > > > > >> should
> > > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10.
> > > I
> > > > > > agree
> > > > > > > > with
> > > > > > > > > > > >> Justine
> > > > > > > > > > > >> > > > that we
> > > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > switch
> > > > > to
> > > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > > >> > > > > > from the
> > > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> > > > > > switch the
> > > > > > > > > > > >> default for
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > feature
> > > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > metadata.version
> > > > > > > > > > once
> > > > > > > > > > > >> we
> > > > > > > > > > > >> > > > want to
> > > > > > > > > > > >> > > > > > make
> > > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My
> > > plan
> > > > > was
> > > > > > to
> > > > > > > > use
> > > > > > > > > > that
> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > > > > > flag
> > > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > phase. I
> > > > > > should
> > > > > > > > > not
> > > > > > > > > > have
> > > > > > > > > > > >> > > > mentioned
> > > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > > > > internal
> > > > > > > > config
> > > > > > > > > > for
> > > > > > > > > > > >> it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11.
> > > Some
> > > > > > of
> > > > > > > > the
> > > > > > > > > > > >> protocol APIs
> > > > > > > > > > > >> > > > > > don't
> > > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > good to
> > > > > > add
> > > > > > > > that
> > > > > > > > > > for
> > > > > > > > > > > >> > > > consistency.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12.
> > > It
> > > > > is
> > > > > > a
> > > > > > > > bit
> > > > > > > > > > odd
> > > > > > > > > > > >> that
> > > > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > though
> > > > > it
> > > > > > > > seems
> > > > > > > > > > to do
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > than
> > > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > agree.
> > > > > > This is
> > > > > > > > > > how the
> > > > > > > > > > > >> > > > current
> > > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > require
> > > > > > "Read
> > > > > > > > > > Group" to
> > > > > > > > > > > >> join a
> > > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I
> > > am
> > > > > not
> > > > > > > > sure
> > > > > > > > > > that
> > > > > > > > > > > >> it is
> > > > > > > > > > > >> > > > worth
> > > > > > > > > > > >> > > > > > it.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13.
> > > How
> > > > > is
> > > > > > > > topic
> > > > > > > > > > > >> recreation
> > > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > protocol?
> > > > > > It
> > > > > > > > > > would be
> > > > > > > > > > > >> good to
> > > > > > > > > > > >> > > > > > have a
> > > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > From
> > > > > > a
> > > > > > > > > > protocol
> > > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic
> > > id
> > > > > so
> > > > > > it
> > > > > > > > > will
> > > > > > > > > > > >> treat it
> > > > > > > > > > > >> > > > like a
> > > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only
> > > issue
> > > > > > is
> > > > > > > > that
> > > > > > > > > > the
> > > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs
> > > so the
> > > > > > > > > consumer
> > > > > > > > > > > >> would
> > > > > > > > > > > >> > > > reuse the
> > > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > we
> > > > > > should
> > > > > > > > > > update
> > > > > > > > > > > >> those
> > > > > > > > > > > >> > > > APIs
> > > > > > > > > > > >> > > > > > as well
> > > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > end.
> > > > > That
> > > > > > > > would
> > > > > > > > > > > >> strengthen
> > > > > > > > > > > >> > > > the
> > > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14.
> > > The
> > > > > > KIP
> > > > > > > > > > mentions
> > > > > > > > > > > >> we will
> > > > > > > > > > > >> > > > > > write the
> > > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > this is
> > > > > an
> > > > > > > > > > > >> implementation
> > > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for
> > > it?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15.
> > > Do
> > > > > we
> > > > > > > > have a
> > > > > > > > > > > >> scalability
> > > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > group
> > > > > > protocol
> > > > > > > > > can
> > > > > > > > > > > >> support?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We
> > > don't
> > > > > > have
> > > > > > > > > > numbers
> > > > > > > > > > > >> at the
> > > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of
> > > members
> > > > > > per
> > > > > > > > > > group.
> > > > > > > > > > > >> We will
> > > > > > > > > > > >> > > > > > measure
> > > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > implementation.
> > > > > > > > > Note
> > > > > > > > > > > >> that we
> > > > > > > > > > > >> > > > might
> > > > > > > > > > > >> > > > > > have
> > > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road
> > > (e.g.
> > > > > > > > offset
> > > > > > > > > > > >> commits).
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16.
> > > Did
> > > > > we
> > > > > > > > > > consider
> > > > > > > > > > > >> having
> > > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > SubscribedTopicNames
> > > > > > > > > > > >> in
> > > > > > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > since we
> > > > > > have
> > > > > > > > to
> > > > > > > > > > > >> resolve the
> > > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > topic
> > > > > > > > name?
> > > > > > > > > > The
> > > > > > > > > > > >> > > > difference
> > > > > > > > > > > >> > > > > > is that
> > > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > efficient
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > whereas
> > > > > > > > sending
> > > > > > > > > > the
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > names
> > > > > > > > > > > >> > > > > > is less
> > > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > recreation is
> > > > > > > > > > easier
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > handle
> > > > > > > > > > > >> > > > > > if we
> > > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The
> > > idea
> > > > > > was to
> > > > > > > > > > > >> consolidate the
> > > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths
> > > but
> > > > > I
> > > > > > do
> > > > > > > > > agree
> > > > > > > > > > > >> with your
> > > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids
> > > may be
> > > > > > > > better
> > > > > > > > > > here
> > > > > > > > > > > >> for the
> > > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > suppose
> > > > > > that we
> > > > > > > > > may
> > > > > > > > > > > >> allow
> > > > > > > > > > > >> > > > users to
> > > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> > > > > > because
> > > > > > > > > that
> > > > > > > > > > is
> > > > > > > > > > > >> the only
> > > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > re-creation.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On
> > > Tue,
> > > > > Jul
> > > > > > 12,
> > > > > > > > > > 2022 at
> > > > > > > > > > > >> 1:38 PM
> > > > > > > > > > > >> > > > > > David
> > > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > djacot@confluent.io>
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > > > > > Justine,
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > Thanks
> > > > > for
> > > > > > > > your
> > > > > > > > > > > >> comments.
> > > > > > > > > > > >> > > > Please
> > > > > > > > > > > >> > > > > > find
> > > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > -
> > > Yes,
> > > > > > the new
> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > > > relies on
> > > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > topic
> > > > > > names
> > > > > > > > > based
> > > > > > > > > > in
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if
> > > using
> > > > > > topic
> > > > > > > > > > names
> > > > > > > > > > > >> is the
> > > > > > > > > > > >> > > > > > right call
> > > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a
> > > little
> > > > > > more.
> > > > > > > > > > > >> Obviously, the
> > > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > > >> > > > > > > > > > > >>
> > >
> >
> >
> > --
> > -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Guozhang,

Thanks for your input.

> 1) in HB request, when clients send the subscription, they are still
> sending them in terms of topic names. That is, the consumer would not try
> to keep all the mappings from topic names to topic ids for the subscribed
> topics.

This is a mistake of mine. The heartbeat request should use Topic Ids.
I forgot to change it. The idea is that the client will get the
metadata first to translate the topic names to topic ids and thus also
have the mapping in its local metadata cache.

> 2) in HB responses, the assignment's topics are encoded as IDs already. And
> fetch requests can directly rely on that since they just need topic IDs and
> do not care what actual topic names are.

That's right.

> 3) in prepare-assignment responses, we return subscriptions as topic IDs as
> well.

That's right.

> In a word, the consumer does not try to maintain the mapping of topic names
> to IDs at all, and even if they do, they will not expose it in e.g.
> assignor APIs. But for client-side assignors like Streams, if they do not
> maintain the mapping from topic IDs to topic names, certain logic such as
> co-partitioning indeed cannot be executed.

As I said, the consumer will have the mapping in its local metadata
cache. It needs to get the partition metadata anyway and the topic id
is part of it. However, I do agree that this mapping is not exposed.

> So I think we'd need to consider 1) use topic names in subscriptions in the
> prepare-assignment responses, 2) use topic Names in client-side
> PartitionAssignor#GroupMember#subscribedTopics and
> PartitionAssignor#TopicMetadata#topicName.

1) makes sense to me. I think that we could actually provide both
names and ids. Regarding 2), I think that I am slightly in favor of
using both names and ids as well. Let me try to update the protocol
and the interface towards this.

Thanks,
David

On Wed, Sep 28, 2022 at 9:35 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hello Jun/David,
>
> Just to chime in here for Jun's comment #50, today Streams does not have a
> way to map topic names to topic IDs, and in its topology description topics
> are encoded with names.
>
> What I originally thought was that, on the consumer side, we'd need to
> maintain this mapping from topic names to topic IDs anyways, and Streams
> should leverage on the consumer's functionality to get the corresponding
> topic names for the given topic IDs. But when I read through the RPC again
> I realized it may not be the case, since:
>
> 1) in HB request, when clients send the subscription, they are still
> sending them in terms of topic names. That is, the consumer would not try
> to keep all the mappings from topic names to topic ids for the subscribed
> topics.
> 2) in HB responses, the assignment's topics are encoded as IDs already. And
> fetch requests can directly rely on that since they just need topic IDs and
> do not care what actual topic names are.
> 3) in prepare-assignment responses, we return subscriptions as topic IDs as
> well.
>
> In a word, the consumer does not try to maintain the mapping of topic names
> to IDs at all, and even if they do, they will not expose it in e.g.
> assignor APIs. But for client-side assignors like Streams, if they do not
> maintain the mapping from topic IDs to topic names, certain logic such as
> co-partitioning indeed cannot be executed.
>
> So I think we'd need to consider 1) use topic names in subscriptions in the
> prepare-assignment responses, 2) use topic Names in client-side
> PartitionAssignor#GroupMember#subscribedTopics and
> PartitionAssignor#TopicMetadata#topicName.
>
> On Wed, Sep 28, 2022 at 1:26 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Jun,
> >
> > Thanks for your comments.
> >
> > 48. Okay. There is a typo in the text:
> > > Note that the persisted target assignment of B already includes foo-2
> > here
> > but B does not get it yet because foo-2 is not free.
> >
> > It should be:
> > > Note that the persisted member assignment of B already includes foo-2
> > here
> > but B does not get it yet because foo-2 is not free.
> >
> > The examples are showing the persisted state in the __consumer_offsets
> > topics. When a member transitions to the next epoch, we directly
> > persist the whole target assignment in order to reduce the number of
> > writes and we manage dependencies in memory. So "B - epoch=2,
> > partitions=[foo-2]" is actually correct from a persisted record
> > perspective. However, the in-memory state would be different because
> > foo-2 is not yet available as you pointed out. The in-memory state
> > would like as follow:
> >
> > Group (epoch=2)
> > A
> > B
> > Target Assignment (epoch=2)
> > A - partitions=[foo-0, foo-1]
> > B - partitions=[foo-2]
> > Member Assignment
> > A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending partitions=[]
> > B - epoch=2, partitions=[], pending partitions=[foo-2]
> >
> > I suppose that the confusion comes from this. I will go and convert
> > all the examples to their in-memory representation in order to reduce
> > the confusion.
> >
> > 50. Okay. I have to admit that I am not so familiar with the internals
> > of KStreams. If I understand correctly, you're saying that KStreams
> > may not have topic ids in its topology so topic names are missing here
> > to match the topics. I was hoping that KStreams would have topic ids
> > in the topology as well. If that is not the case, we can also add
> > topic names on our side. I will check this.
> >
> > 53. It seems that I was not clear. Sorry for that. I did not mean to
> > say that MemberAssignment.metadata is not used by the group
> > coordinator. Let me re-explain the flow.
> >
> > PartitionAssignor.assign() returns an Assignment. Assignment contains
> > MemberAssignments where each member has MemberAssignment.metadata. The
> > Assignment is installed by the client with the
> > ConsumerGroupInstallAssignment API and this overrides the current
> > target assignment in the coordinator. Then the group coordinator uses
> > it to reconcile members. Therefore the target assignment will
> > eventually become the current member assignment and it will be
> > delivered to the member, including the original
> > MemberAssignment.metadata defines by the assignor.
> >
> > I hope that this helps to clarify how MemberAssignment.metadata is
> > used afterwards.
> >
> > Best,
> > David
> >
> > On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <ju...@confluent.io.invalid> wrote:
> > > Hi, David,
> > >
> > > Thanks for the reply. A few more follow up questions.
> > >
> > > 48. Hmm, so member assignment is what the client sees. Then, I am not
> > sure
> > > about the following. The text says B cannot get foo-2 yet. But in Member
> > > Assignment, B has foo-2.
> > >
> > > At this point B can transition to epoch 2 but cannot get foo-2 until A
> > > revokes it.
> > >
> > > Note that the persisted target assignment of B already includes foo-2
> > here
> > > but B does not get it yet because foo-2 is not free.
> > >
> > > Group (epoch=2)
> > > A
> > > B
> > > Target Assignment (epoch=2)
> > > A - partitions=[foo-0, foo-1]
> > > B - partitions=[foo-2]
> > > Member Assignment
> > > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > > B - epoch=2, partitions=[foo-2]
> > >
> > >
> > > 50. Hmm, I thought KStreams doesn't use range assignor all the time. It
> > > does the assignment based on the topology. For example, if you are
> > joining
> > > two topics A and B and deriving the join result in a 3rd topic C, which
> > > will be aggregated. You want to use range to co-locate the partitions
> > from
> > > topic A and B. However, for topic C, the partition could potentially be
> > > assigned in round robin. Without the topic name, KStreams won't know
> > which
> > > topics need to be co-located.
> > >
> > > 53. If MemberAssignment.metadata is not used by the group coordinator,
> > > where is it being used?
> > >
> > > Thanks,
> > >
> > > Jun
> > >
> > > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> > <ja...@confluent.io.invalid>
> > > wrote:
> > >
> > > > Hi David,
> > > >
> > > > Thanks for the response. Replied below:
> > > >
> > > > > I don't have a strong opinion on this one. However, it seems that
> > this
> > > > could be a pandora box if our subset turns out to not be sufficient
> > > > for some existing users. We will have to add more to it and the
> > > > compatibility story between versions could be a real nightmare. One
> > > > thing that we could do for sure is disabling all unicode groups in the
> > > > library. That seems possible and would already reduce the scope.
> > > >
> > > > Changes to the regex syntax seems not a major problem to me. We would
> > bump
> > > > the protocol version as we usually do. Actually that begs the same
> > question
> > > > about RE2. How stable is the syntax and does it differ depending on the
> > > > version of the library? In general, I do agree that we'd be better off
> > not
> > > > having to manage the syntax ourselves if we can convince ourselves that
> > > > compatibility won't be a problem and the client won't eventually need
> > to
> > > > have access to it.
> > > >
> > > > Thanks,
> > > > Jason
> > > >
> > > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> > <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi Luke,
> > > > >
> > > > > Thanks for your comments.
> > > > >
> > > > > Your suggestion makes sense. Jun's made the same one. We are also
> > > > > discussing whether we should directly return the group state in the
> > > > > response and drop the prepare assignment api. It seems that the
> > > > > current consensus is to keep it though. If that stays like this, I
> > > > > will apply your suggestion.
> > > > >
> > > > > Thanks,
> > > > > David
> > > > >
> > > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com>
> > wrote:
> > > > > >
> > > > > > Hi David,
> > > > > >
> > > > > > Thanks for your patience for answering all these questions!
> > > > > >
> > > > > > One more question from me:
> > > > > > 1. When the selected consumer is going to be in client-side
> > assigning,
> > > > we
> > > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> > wondering if
> > > > we
> > > > > > could use a flag to indicate the assignment is required. The
> > rationale
> > > > is
> > > > > > when returning error, we lose a chance to catch up with the latest
> > > > target
> > > > > > assignment or revoke partitions. Since we expected the record
> > fetching
> > > > is
> > > > > > not blocked by the rebalance/assignment computing, we might want to
> > > > allow
> > > > > > the members to update the member assignment in each heartbeat.
> > > > > >
> > > > > > Luke
> > > > > >
> > > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > > <ja...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > > > Hey David,
> > > > > > >
> > > > > > > Thanks for the response.
> > > > > > >
> > > > > > > > Do you think that we could have a case where two members with
> > the
> > > > > same
> > > > > > > static id are active?
> > > > > > >
> > > > > > > I was just thinking about the handoff when we spin up a new node
> > > > > before the
> > > > > > > old one completes shutdown. It's probably ok to fence the old
> > member
> > > > > > > abruptly. I do recall we had a few issues with this handoff in
> > the
> > > > > past. It
> > > > > > > was due to the fact that the replaced node might have been the
> > > > leader.
> > > > > > > Since we've removed the leader notion, perhaps this kind of issue
> > > > won't
> > > > > > > resurface.
> > > > > > >
> > > > > > > In regard to the RE2 library. I understand the intent to pass
> > this
> > > > > through
> > > > > > > as a string without any validation so that we don't need the
> > > > > dependence on
> > > > > > > the client. I do wonder if that will be restrictive in practice.
> > For
> > > > > > > example, if a regex changes on the client side, would we need to
> > > > > validate
> > > > > > > against the current assignment to see which partitions to
> > continue
> > > > > > > fetching? Or would we instead invalidate the complete current
> > > > > assignment?
> > > > > > > Perhaps this is a rare case and we can do the latter. There
> > could be
> > > > > more
> > > > > > > situations though.
> > > > > > >
> > > > > > > As an alternative, I wonder if we could define a minimalistic
> > regex
> > > > > syntax
> > > > > > > (a subset of RE2 say) that could map more easily to regex support
> > > > that
> > > > > > > already exists in clients? For example, in this scenario, we are
> > > > > matching
> > > > > > > topics, so we only need to be able to represent topic characters
> > and
> > > > > not
> > > > > > > the whole smorgasbord of whitespace and unicode options. Would it
> > > > make
> > > > > > > sense to have a reduced syntax which includes only standard topic
> > > > > > > characters, wildcards, character classes, and repetitions?
> > > > > > >
> > > > > > > Best,
> > > > > > > Jason
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > wrote:
> > > > > > >
> > > > > > > > Hi David, thanks for all the detailed explanations. I think
> > they
> > > > all
> > > > > make
> > > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > > >
> > > > > > > > > I don't really see the benefits here because server side
> > > > assignors
> > > > > > > > don't have metadata at all. They only assign topic-partitions.
> > They
> > > > > > > > are not supposed to generate metadata nor to receive metadata
> > from
> > > > > the
> > > > > > > > members.
> > > > > > > >
> > > > > > > > Just to clarify I was asking about the `version` of the
> > assignor
> > > > > (i.e. up
> > > > > > > > to what version that the client would support), and I do agree
> > we
> > > > > would
> > > > > > > not
> > > > > > > > need metadata. What I have in mind is that, for some specific
> > > > > built-in
> > > > > > > > broker-assignors, e.g. rack-aware assignors, if it's possible
> > that
> > > > > in a
> > > > > > > > newer version we would have a hierarchical rack ID string
> > format,
> > > > > like
> > > > > > > > "tier1-tier2" etc, but if some client has not upgraded their
> > rack
> > > > ID
> > > > > > > > would still be in old format. In this case, the broker then
> > needs
> > > > to
> > > > > > > choose
> > > > > > > > the old versioned assignor. I'm probably making something up
> > here
> > > > for
> > > > > > > rack
> > > > > > > > aware assignors, but I'm wondering if in general such an
> > > > > "auto-downgrade"
> > > > > > > > behavior would be needed still for broker-side assignor, and
> > if yes
> > > > > would
> > > > > > > > "version" still be useful.
> > > > > > > >
> > > > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> > called
> > > > > once
> > > > > > > > when the member transitions to a new epoch. This one contains
> > the
> > > > > full
> > > > > > > > metadata provided by the client side assignor. Then,
> > `onAssignment`
> > > > > > > > can be called max N times where N is the number of partitions
> > > > pending
> > > > > > > > revocation by other members. Let me try to clarify this in the
> > KIP.
> > > > > > > >
> > > > > > > > Okay, my understanding is that the calling ordering of these
> > > > > callbacks
> > > > > > > > would be like the following:
> > > > > > > >
> > > > > > > > ----------------------------------------
> > > > > > > > onPartitionsRevoked();   // just once, since we do not really
> > need
> > > > > > > > to revoke incrementally.
> > > > > > > >
> > > > > > > > onAssignment();    // the first call, with epoch incremented
> > > > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > > > >
> > > > > > > > onAssignment();              // the first onAssignment would
> > bump
> > > > up
> > > > > the
> > > > > > > > epoch, and the metadata reflected.
> > > > > > > > onPartitionsAssigned();   // each time we get an additional
> > > > > assignment,
> > > > > > > we
> > > > > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > > > > ...
> > > > > > > > onAssignment();
> > > > > > > > onPartitionsAssigned();   // on each of the onAssignment
> > calls, the
> > > > > > > encoded
> > > > > > > > metadata would not change, only the incrementally added
> > partitions
> > > > be
> > > > > > > > reflected
> > > > > > > >
> > > > > > > > Is that the case?
> > > > > > > >
> > > > > > > > I'm wondering if we would still call onAssignment just once,
> > that
> > > > > encodes
> > > > > > > > all the assignment for this rebalance, including all the
> > partitions
> > > > > that
> > > > > > > > should be assigned to the member but not yet assigned since
> > they
> > > > > have not
> > > > > > > > been revoked by others. In that case the call ordering would
> > be:
> > > > > > > >
> > > > > > > > ----------------------------------------
> > > > > > > > onPartitionsRevoked();   // just once
> > > > > > > > onAssignment();    // just once, with epoch incremented, and
> > > > metadata
> > > > > > > > encoded changed, the "assignment" field also reflect the final
> > > > target
> > > > > > > > assignment
> > > > > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > > > > incrementally
> > > > > > > > added partitions
> > > > > > > > ...
> > > > > > > > onPartitionsAssigned();
> > > > > > > >
> > > > > > > > The motivation from this is that, most users would only
> > implement
> > > > the
> > > > > > > > rebalance callback listeners and hence we'd definitely need to
> > make
> > > > > sure
> > > > > > > > the semantics of that does not change much, and the time
> > > > > > > > `onPartitionsAssigned` indicate the time when the partitions
> > are
> > > > > actually
> > > > > > > > assigned to it; while for assignors, the `onAssignment` is
> > used to
> > > > > > > indicate
> > > > > > > > what decision is made regarding for this member, i.e. when the
> > > > > partitions
> > > > > > > > are decided to be given to it, but not necessarily meant that
> > it
> > > > has
> > > > > been
> > > > > > > > given, since that time should be determined by the time of
> > > > > > > > `onPartitionsAssigned`. The benefits though, would be that
> > assignor
> > > > > > > > implementers would not need to reason which `onAssignment`
> > would be
> > > > > the
> > > > > > > > last one for this epoch.
> > > > > > > >
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > > <djacot@confluent.io.invalid
> > > > > > > >
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi Guozhang,
> > > > > > > > >
> > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> > Error
> > > > > byte
> > > > > > > > > field,
> > > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> > not.
> > > > > And
> > > > > > > the
> > > > > > > > > > rationale seems to be assuming that we should always be
> > able to
> > > > > do
> > > > > > > the
> > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > Personally I
> > > > > > > > think
> > > > > > > > > > it's still potentially beneficial to add the Error field
> > even
> > > > for
> > > > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > > > subscribed
> > > > > > > > > > topics are not recognized with the current broker's
> > metadata.
> > > > > What do
> > > > > > > > you
> > > > > > > > > > think?
> > > > > > > > >
> > > > > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > > > > better
> > > > > > > > > to use an exception on the server side. This is what we
> > usually
> > > > do
> > > > > for
> > > > > > > > > server side plugins. On the client side, we use a field
> > because
> > > > the
> > > > > > > > > errors are not defined in advance.
> > > > > > > > >
> > > > > > > > > Your comment also makes me think about what we should do
> > when the
> > > > > > > > > server side assignor fails. I suppose that we have to keep
> > the
> > > > > current
> > > > > > > > > assignment until a new event occurs. For instance, in your
> > > > example,
> > > > > > > > > the coordinator would have to trigger a rebalance when
> > > > unrecognized
> > > > > > > > > topics are available. This would be part of the metadata
> > > > > monitoring.
> > > > > > > > >
> > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > fields
> > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > GroupMember. I
> > > > > > > > > agree
> > > > > > > > > > that broker-side assignor would not need reason/metadata
> > since
> > > > > they
> > > > > > > are
> > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > version?
> > > > > E.g.
> > > > > > > is
> > > > > > > > > it
> > > > > > > > > > possible that we evolve our broker-side built-in assignor
> > but
> > > > > the old
> > > > > > > > > > versioned clients would not be able to work with the new
> > > > > version, in
> > > > > > > > > which
> > > > > > > > > > case we need to let the broker being aware of this and
> > upgrade
> > > > > its
> > > > > > > > > behavior
> > > > > > > > > > to cooperate with the clients?
> > > > > > > > >
> > > > > > > > > I don't really see the benefits here because server side
> > > > assignors
> > > > > > > > > don't have metadata at all. They only assign
> > topic-partitions.
> > > > They
> > > > > > > > > are not supposed to generate metadata nor to receive metadata
> > > > from
> > > > > the
> > > > > > > > > members.
> > > > > > > > >
> > > > > > > > > > 3) Also related to 2) above, for the client-side
> > "GroupMember",
> > > > > > > instead
> > > > > > > > > of
> > > > > > > > > > including these three fields, what about just adding the
> > > > > "Metadata"
> > > > > > > > field
> > > > > > > > > > class which has these three fields? Also, there are two
> > > > > "Metadata"
> > > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > > reason/version/metadata, and the second is just the encoded
> > > > > metadata
> > > > > > > > > bytes.
> > > > > > > > > > I'm wondering what about just naming the first as
> > > > memberMetadata,
> > > > > > > which
> > > > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > > > second
> > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > >
> > > > > > > > > That's a good point. Let me try to rationalize this interface
> > > > > based on
> > > > > > > > > your suggestions.
> > > > > > > > >
> > > > > > > > > Best,
> > > > > > > > > David
> > > > > > > > >
> > > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > > wangguoz@gmail.com>
> > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Hello David,
> > > > > > > > > >
> > > > > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > > > > >
> > > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> > Error
> > > > > byte
> > > > > > > > > field,
> > > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> > not.
> > > > > And
> > > > > > > the
> > > > > > > > > > rationale seems to be assuming that we should always be
> > able to
> > > > > do
> > > > > > > the
> > > > > > > > > > assignment at the broker-side assignor without errors.
> > > > > Personally I
> > > > > > > > think
> > > > > > > > > > it's still potentially beneficial to add the Error field
> > even
> > > > for
> > > > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > > > subscribed
> > > > > > > > > > topics are not recognized with the current broker's
> > metadata.
> > > > > What do
> > > > > > > > you
> > > > > > > > > > think?
> > > > > > > > > >
> > > > > > > > > > 2) The client-side "GroupMember" has three additional
> > fields
> > > > > > > > > > reason/version/metadata compared with the broker-side
> > > > > GroupMember. I
> > > > > > > > > agree
> > > > > > > > > > that broker-side assignor would not need reason/metadata
> > since
> > > > > they
> > > > > > > are
> > > > > > > > > > blackbox strings/bytes to the assignor, but what about
> > version?
> > > > > E.g.
> > > > > > > is
> > > > > > > > > it
> > > > > > > > > > possible that we evolve our broker-side built-in assignor
> > but
> > > > > the old
> > > > > > > > > > versioned clients would not be able to work with the new
> > > > > version, in
> > > > > > > > > which
> > > > > > > > > > case we need to let the broker being aware of this and
> > upgrade
> > > > > its
> > > > > > > > > behavior
> > > > > > > > > > to cooperate with the clients?
> > > > > > > > > >
> > > > > > > > > > 3) Also related to 2) above, for the client-side
> > "GroupMember",
> > > > > > > instead
> > > > > > > > > of
> > > > > > > > > > including these three fields, what about just adding the
> > > > > "Metadata"
> > > > > > > > field
> > > > > > > > > > class which has these three fields? Also, there are two
> > > > > "Metadata"
> > > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > > reason/version/metadata, and the second is just the encoded
> > > > > metadata
> > > > > > > > > bytes.
> > > > > > > > > > I'm wondering what about just naming the first as
> > > > memberMetadata,
> > > > > > > which
> > > > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > > > second
> > > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > > >
> > > > > > > > > > Guozhang
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hello David,
> > > > > > > > > > >
> > > > > > > > > > > Thanks for bringing this question up. I think the main
> > > > > benefits as
> > > > > > > > you
> > > > > > > > > > > listed is 2) above if it stays; just to clarify, we would
> > > > only
> > > > > be
> > > > > > > > able
> > > > > > > > > to
> > > > > > > > > > > save one round trip if the rebalance is still triggered
> > by
> > > > the
> > > > > > > > broker;
> > > > > > > > > if
> > > > > > > > > > > the rebalance is triggered by the client then the
> > > > > num.round.trips
> > > > > > > are
> > > > > > > > > the
> > > > > > > > > > > same:
> > > > > > > > > > >
> > > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > > >
> > > > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > > > already
> > > > > > > > sent
> > > > > > > > > a
> > > > > > > > > > > HB and hence has to wait for it to return first since
> > only
> > > > one
> > > > > > > > request
> > > > > > > > > /
> > > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > request.
> > > > > > > > > > >
> > > > > > > > > > > In total, two round trips.
> > > > > > > > > > >
> > > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > > >
> > > > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > > > already
> > > > > > > > sent
> > > > > > > > > a
> > > > > > > > > > > HB and hence has to wait for it to return first since
> > only
> > > > one
> > > > > > > > request
> > > > > > > > > /
> > > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > > T1: client receives the HB response, and then sends the
> > new
> > > > HB
> > > > > > > > request
> > > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > > T2: the HB response with the optional member metadata
> > map is
> > > > > > > > returned.
> > > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > > GroupInstallAssignment
> > > > > > > > > > > request.
> > > > > > > > > > >
> > > > > > > > > > > In total, two round trips as well.
> > > > > > > > > > >
> > > > > > > > > > > -----------------------------
> > > > > > > > > > >
> > > > > > > > > > > So to complete the full picture here, we'd need to modify
> > > > both
> > > > > HB
> > > > > > > > > request
> > > > > > > > > > > and response so that the client can also indicate a new
> > > > > rebalance
> > > > > > > via
> > > > > > > > > the
> > > > > > > > > > > HB request as well, right?
> > > > > > > > > > >
> > > > > > > > > > > Assuming all above is true, I think it's okay to merge
> > the
> > > > > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > > > > additional
> > > > > > > > > fields
> > > > > > > > > > > encoding the full member (subscription) metadata and
> > topic
> > > > > metadata
> > > > > > > > as
> > > > > > > > > > > optional fields.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Guozhang
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > >> Hi all,
> > > > > > > > > > >>
> > > > > > > > > > >> During an offline conversation, someone asked why we
> > need
> > > > the
> > > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we
> > > > could
> > > > > > > > instead
> > > > > > > > > > >> provide the group state in the heartbeat response. This
> > has
> > > > a
> > > > > few
> > > > > > > > > > >> advantages: 1) it does not require using a special error
> > > > code
> > > > > to
> > > > > > > > > > >> signal that a new assignment is required as the signal
> > would
> > > > > be
> > > > > > > the
> > > > > > > > > > >> provided group state; 2) it removes one round trip when
> > a
> > > > > client
> > > > > > > > side
> > > > > > > > > > >> assignor is used. The downside is that it makes the
> > > > heartbeat
> > > > > > > > > > >> response's definition quite large. I recall that I went
> > with
> > > > > the
> > > > > > > > > > >> current approach due to this.
> > > > > > > > > > >>
> > > > > > > > > > >> Providing the group state in the heartbeat response is
> > > > > appealing.
> > > > > > > > What
> > > > > > > > > > >> do you guys think?
> > > > > > > > > > >>
> > > > > > > > > > >> Best,
> > > > > > > > > > >> David
> > > > > > > > > > >>
> > > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > > djacot@confluent.io>
> > > > > > > > > wrote:
> > > > > > > > > > >> >
> > > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > > >> >
> > > > > > > > > > >> > 1. I have added a reference to the relevant chapter
> > > > instead
> > > > > of
> > > > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > > > >> >
> > > > > > > > > > >> > 2. The "Rebalance Triggers" section you are referring
> > to
> > > > is
> > > > > > > about
> > > > > > > > > when
> > > > > > > > > > >> > a rebalance should be triggered for the non-upgraded
> > > > members
> > > > > > > using
> > > > > > > > > the
> > > > > > > > > > >> > old protocol. The section mentions that a rebalance
> > must
> > > > be
> > > > > > > > > triggered
> > > > > > > > > > >> > when a new assignment is installed. This implies that
> > the
> > > > > group
> > > > > > > > > epoch
> > > > > > > > > > >> > was updated either by a native member or a
> > non-upgraded
> > > > > member.
> > > > > > > > For
> > > > > > > > > > >> > the latter, the JoinGroup request would be the
> > trigger. I
> > > > > have
> > > > > > > > > added a
> > > > > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > > > > Handling"
> > > > > > > > > section
> > > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > > >> >
> > > > > > > > > > >> > Thanks,
> > > > > > > > > > >> > David
> > > > > > > > > > >> >
> > > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > > > wangguoz@gmail.com
> > > > > > > > >
> > > > > > > > > > >> wrote:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Hello David,
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > > > > clear in
> > > > > > > > the
> > > > > > > > > doc,
> > > > > > > > > > >> > > could we update:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > 1) the heartbeat request handling section, stating
> > when
> > > > > > > > > coordinator
> > > > > > > > > > >> will
> > > > > > > > > > >> > > trigger rebalance based on the HB's member metadata
> > /
> > > > > reason?
> > > > > > > > > > >> > > 2) the "Rebalance Triggers" section to include what
> > we
> > > > > > > described
> > > > > > > > > in
> > > > > > > > > > >> "Group
> > > > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > > > >> > >
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Guozhang
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > wrote:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > I thought that the assignor will always be
> > consulted
> > > > > when
> > > > > > > the
> > > > > > > > > next
> > > > > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > > > > >> > > > `PartitionAssignor#metadata` will be called for
> > every
> > > > > > > > heartbeat.
> > > > > > > > > > >> This
> > > > > > > > > > >> > > > gives the opportunity for the assignor to enforce
> > a
> > > > > > > rebalance
> > > > > > > > by
> > > > > > > > > > >> > > > setting the reason to a non-zero value or by
> > changing
> > > > > the
> > > > > > > > > bytes. Do
> > > > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > > > concerned by
> > > > > > > > the
> > > > > > > > > > >> delay?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Best,
> > > > > > > > > > >> > > > David
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > > > wangguoz@gmail.com>
> > > > > > > > > > >> wrote:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > Hello David,
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > ```
> > > > > > > > > > >> > > > > In this case, a new assignment is triggered by
> > the
> > > > > client
> > > > > > > > side
> > > > > > > > > > >> > > > > assignor. When constructing the HB, the consumer
> > > > will
> > > > > > > always
> > > > > > > > > > >> consult
> > > > > > > > > > >> > > > > the client side assignor and propagate the
> > > > > information to
> > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > coordinator. In other words, we don't expect
> > users
> > > > to
> > > > > call
> > > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > >> > > > > ```
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > > > > interface,
> > > > > > > we
> > > > > > > > > > >> actually do
> > > > > > > > > > >> > > > > not have a way yet to instruct how to construct
> > the
> > > > > next
> > > > > > > HB
> > > > > > > > > > >> request, e.g.
> > > > > > > > > > >> > > > > when the assignor wants to enforce a new
> > rebalance
> > > > > with a
> > > > > > > > new
> > > > > > > > > > >> assignment,
> > > > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > > > PartitionAssignor
> > > > > > > > > to
> > > > > > > > > > >> indicate
> > > > > > > > > > >> > > > > the next HB telling broker about so. WDYT about
> > > > adding
> > > > > > > such
> > > > > > > > an
> > > > > > > > > > >> API on the
> > > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > Guozhang
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > I have updated the KIP to include your
> > feedback. I
> > > > > have
> > > > > > > > also
> > > > > > > > > > >> tried to
> > > > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Best,
> > > > > > > > > > >> > > > > > David
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > > > > djacot@confluent.io
> > > > > > > > > > >> >
> > > > > > > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > > > > answering
> > > > > > > your
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > benefits
> > > > > of
> > > > > > > the
> > > > > > > > > KIP
> > > > > > > > > > >> seem to
> > > > > > > > > > >> > > > be
> > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > > support of
> > > > > > > > > > >> wildcard. A few
> > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > I would also add that the KIP removes the
> > global
> > > > > sync
> > > > > > > > > barrier
> > > > > > > > > > >> in the
> > > > > > > > > > >> > > > > > > protocol which is essential to improve group
> > > > > stability
> > > > > > > > and
> > > > > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > > > > client by
> > > > > > > > > moving
> > > > > > > > > > >> most of
> > > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> > plan
> > > > > to
> > > > > > > > > support
> > > > > > > > > > >> rolling
> > > > > > > > > > >> > > > > > changing
> > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > consumers?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Definitely. The group coordinator will use
> > the
> > > > > > > assignor
> > > > > > > > > used
> > > > > > > > > > >> by a
> > > > > > > > > > >> > > > > > > majority of the members. This allows the
> > group
> > > > to
> > > > > move
> > > > > > > > > from
> > > > > > > > > > >> one
> > > > > > > > > > >> > > > > > > assignor to another by a roll. This is
> > explained
> > > > > in
> > > > > > > the
> > > > > > > > > > >> Assignor
> > > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > whether
> > > > > it's
> > > > > > > > > > >> required in
> > > > > > > > > > >> > > > every
> > > > > > > > > > >> > > > > > > > request or the scenarios when it needs to
> > be
> > > > > filled?
> > > > > > > > For
> > > > > > > > > > >> example,
> > > > > > > > > > >> > > > it's
> > > > > > > > > > >> > > > > > not
> > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to
> > be
> > > > > filled.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > The client is expected to set those fields
> > in
> > > > > case of
> > > > > > > a
> > > > > > > > > > >> connection
> > > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > > > > changed
> > > > > > > > since
> > > > > > > > > > >> the last
> > > > > > > > > > >> > > > > > > HB. The server populates those fields as
> > long as
> > > > > the
> > > > > > > > > member
> > > > > > > > > > >> is not
> > > > > > > > > > >> > > > > > > fully reconciled - the member should
> > acknowledge
> > > > > that
> > > > > > > it
> > > > > > > > > has
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > expected epoch and assignment. I will
> > clarify
> > > > > this in
> > > > > > > > the
> > > > > > > > > KIP.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> > rack
> > > > > > > > affinity
> > > > > > > > > > >> between the
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > > fetching,
> > > > > > > but
> > > > > > > > > not
> > > > > > > > > > >> during
> > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> > the
> > > > > > > > assignment
> > > > > > > > > is
> > > > > > > > > > >> made,
> > > > > > > > > > >> > > > there
> > > > > > > > > > >> > > > > > is
> > > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> > no
> > > > > replicas
> > > > > > > > of
> > > > > > > > > > >> assigned
> > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> > if we
> > > > > should
> > > > > > > > use
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > GroupMember.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > That's an interesting idea. I don't see any
> > > > issue
> > > > > with
> > > > > > > > > adding
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > rack
> > > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > useful to
> > > > > know
> > > > > > > how
> > > > > > > > > busy
> > > > > > > > > > >> a group
> > > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > > model,
> > > > > it
> > > > > > > > seems
> > > > > > > > > > >> that we
> > > > > > > > > > >> > > > could
> > > > > > > > > > >> > > > > > add
> > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> > time
> > > > > the
> > > > > > > > event
> > > > > > > > > > >> loop is
> > > > > > > > > > >> > > > doing
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > > > failover
> > > > > > > > > > >> handling? For
> > > > > > > > > > >> > > > > > example,
> > > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> > group
> > > > > with
> > > > > > > > the
> > > > > > > > > > >> wildcard
> > > > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Sure. When the new group coordinator takes
> > over,
> > > > > it
> > > > > > > has
> > > > > > > > > to:
> > > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > > > > assignor
> > > > > > > is
> > > > > > > > > used.
> > > > > > > > > > >> We
> > > > > > > > > > >> > > > don't
> > > > > > > > > > >> > > > > > > store the information about the member
> > selected
> > > > > to run
> > > > > > > > the
> > > > > > > > > > >> assignment
> > > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > > > wildcard
> > > > > > > > > > >> subscriptions, and
> > > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > > > document
> > > > > > > > what
> > > > > > > > > > >> the epoch
> > > > > > > > > > >> > > > > > field
> > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > one in
> > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Sure. I will clarify that but it is always
> > the
> > > > > latest
> > > > > > > > > group
> > > > > > > > > > >> epoch.
> > > > > > > > > > >> > > > > > > When the group state is updated, the group
> > epoch
> > > > > is
> > > > > > > > > bumped so
> > > > > > > > > > >> we use
> > > > > > > > > > >> > > > > > > that one for all the change records related
> > to
> > > > the
> > > > > > > > update.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > that
> > > > the
> > > > > > > > > following
> > > > > > > > > > >> > > > invariants
> > > > > > > > > > >> > > > > > are
> > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > possible
> > > > for
> > > > > a
> > > > > > > > member
> > > > > > > > > > >> not to
> > > > > > > > > > >> > > > get any
> > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > That's right. Here I meant that the members
> > > > > provided
> > > > > > > by
> > > > > > > > > the
> > > > > > > > > > >> assignor
> > > > > > > > > > >> > > > > > > in the assignment must exist in the group.
> > The
> > > > > > > assignor
> > > > > > > > > can
> > > > > > > > > > >> not make
> > > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > member
> > > > > epoch
> > > > > > > > > equals to
> > > > > > > > > > >> 0":
> > > > > > > > > > >> > > > When
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > would be
> > > > > used?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > > > partitions and
> > > > > > > > > > >> rejoins when
> > > > > > > > > > >> > > > it
> > > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In
> > this
> > > > > case,
> > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > coordinator will have removed the member
> > from
> > > > the
> > > > > > > group.
> > > > > > > > > The
> > > > > > > > > > >> member
> > > > > > > > > > >> > > > > > > can rejoin the group with the same member
> > id but
> > > > > with
> > > > > > > 0
> > > > > > > > as
> > > > > > > > > > >> epoch. Let
> > > > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > > ability
> > > > > to
> > > > > > > > > trigger a
> > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> > by
> > > > > updating
> > > > > > > > the
> > > > > > > > > > >> assignor
> > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > conflicting
> > > > > with
> > > > > > > the
> > > > > > > > > > >> deprecation
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > In this case, a new assignment is triggered
> > by
> > > > the
> > > > > > > > client
> > > > > > > > > side
> > > > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> > consumer
> > > > > will
> > > > > > > > > always
> > > > > > > > > > >> consult
> > > > > > > > > > >> > > > > > > the client side assignor and propagate the
> > > > > information
> > > > > > > > to
> > > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > > > > users to
> > > > > > > > call
> > > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> > But
> > > > the
> > > > > > > > section
> > > > > > > > > > >> seems to
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> > 2, B
> > > > > > > > > immediately
> > > > > > > > > > >> gets into
> > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > > > incorrect.
> > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> > 3, C
> > > > > seems
> > > > > > > to
> > > > > > > > > get
> > > > > > > > > > >> into
> > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > still
> > > > > has A -
> > > > > > > > > > >> epoch=2,
> > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > > support
> > > > > the
> > > > > > > > > upgrade
> > > > > > > > > > >> from
> > > > > > > > > > >> > > > any
> > > > > > > > > > >> > > > > > old
> > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > We will support upgrading from the consumer
> > > > > protocol
> > > > > > > > > version
> > > > > > > > > > >> 3,
> > > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > > > implemented
> > > > > yet
> > > > > > > so
> > > > > > > > > the
> > > > > > > > > > >> earliest
> > > > > > > > > > >> > > > > > > version is unknown at the moment. This is
> > > > > explained in
> > > > > > > > the
> > > > > > > > > > >> migration
> > > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> > > > > update the
> > > > > > > > KIP
> > > > > > > > > > >> based on
> > > > > > > > > > >> > > > it
> > > > > > > > > > >> > > > > > > next week.
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > Best,
> > > > > > > > > > >> > > > > > > David
> > > > > > > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> > benefits
> > > > > of
> > > > > > > the
> > > > > > > > > KIP
> > > > > > > > > > >> seem to
> > > > > > > > > > >> > > > be
> > > > > > > > > > >> > > > > > fewer
> > > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > > support of
> > > > > > > > > > >> wildcard. A few
> > > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> > plan
> > > > > to
> > > > > > > > > support
> > > > > > > > > > >> rolling
> > > > > > > > > > >> > > > > > changing
> > > > > > > > > > >> > > > > > > > of the partition assignor in the
> > consumers?
> > > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> > whether
> > > > > it's
> > > > > > > > > > >> required in
> > > > > > > > > > >> > > > every
> > > > > > > > > > >> > > > > > > > request or the scenarios when it needs to
> > be
> > > > > filled?
> > > > > > > > For
> > > > > > > > > > >> example,
> > > > > > > > > > >> > > > it's
> > > > > > > > > > >> > > > > > not
> > > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to
> > be
> > > > > filled.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> > rack
> > > > > > > > affinity
> > > > > > > > > > >> between the
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > > fetching,
> > > > > > > but
> > > > > > > > > not
> > > > > > > > > > >> during
> > > > > > > > > > >> > > > > > assigning
> > > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> > the
> > > > > > > > assignment
> > > > > > > > > is
> > > > > > > > > > >> made,
> > > > > > > > > > >> > > > there
> > > > > > > > > > >> > > > > > is
> > > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> > no
> > > > > replicas
> > > > > > > > of
> > > > > > > > > > >> assigned
> > > > > > > > > > >> > > > > > partitions
> > > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> > if we
> > > > > should
> > > > > > > > use
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > > > opportunity
> > > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > > GroupMember.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> > useful to
> > > > > know
> > > > > > > how
> > > > > > > > > busy
> > > > > > > > > > >> a group
> > > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > > model,
> > > > > it
> > > > > > > > seems
> > > > > > > > > > >> that we
> > > > > > > > > > >> > > > could
> > > > > > > > > > >> > > > > > add
> > > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> > time
> > > > > the
> > > > > > > > event
> > > > > > > > > > >> loop is
> > > > > > > > > > >> > > > doing
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > > > failover
> > > > > > > > > > >> handling? For
> > > > > > > > > > >> > > > > > example,
> > > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> > group
> > > > > with
> > > > > > > > the
> > > > > > > > > > >> wildcard
> > > > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > > > document
> > > > > > > > what
> > > > > > > > > > >> the epoch
> > > > > > > > > > >> > > > > > field
> > > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > > >> > > > > > reflect
> > > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> > one in
> > > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> > that
> > > > the
> > > > > > > > > following
> > > > > > > > > > >> > > > invariants
> > > > > > > > > > >> > > > > > are
> > > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> > possible
> > > > for
> > > > > a
> > > > > > > > member
> > > > > > > > > > >> not to
> > > > > > > > > > >> > > > get any
> > > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> > member
> > > > > epoch
> > > > > > > > > equals to
> > > > > > > > > > >> 0":
> > > > > > > > > > >> > > > When
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> > would be
> > > > > used?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > > ability
> > > > > to
> > > > > > > > > trigger a
> > > > > > > > > > >> > > > > > reassignment
> > > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> > by
> > > > > updating
> > > > > > > > the
> > > > > > > > > > >> assignor
> > > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> > conflicting
> > > > > with
> > > > > > > the
> > > > > > > > > > >> deprecation
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> > But
> > > > the
> > > > > > > > section
> > > > > > > > > > >> seems to
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> > 2, B
> > > > > > > > > immediately
> > > > > > > > > > >> gets into
> > > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > > > incorrect.
> > > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> > 3, C
> > > > > seems
> > > > > > > to
> > > > > > > > > get
> > > > > > > > > > >> into
> > > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> > still
> > > > > has A -
> > > > > > > > > > >> epoch=2,
> > > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > > support
> > > > > the
> > > > > > > > > upgrade
> > > > > > > > > > >> from
> > > > > > > > > > >> > > > any
> > > > > > > > > > >> > > > > > old
> > > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David
> > Jacot
> > > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > The KIP states that we will
> > re-implement the
> > > > > > > > > coordinator
> > > > > > > > > > >> in
> > > > > > > > > > >> > > > Java. I
> > > > > > > > > > >> > > > > > > > > discussed this offline with a few folks
> > and
> > > > > folks
> > > > > > > > are
> > > > > > > > > > >> concerned
> > > > > > > > > > >> > > > that
> > > > > > > > > > >> > > > > > > > > we could introduce many regressions in
> > the
> > > > old
> > > > > > > > > protocol
> > > > > > > > > > >> if we do
> > > > > > > > > > >> > > > so.
> > > > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > > > statement
> > > > > > > from
> > > > > > > > > the
> > > > > > > > > > >> KIP. It
> > > > > > > > > > >> > > > is an
> > > > > > > > > > >> > > > > > > > > implementation detail after all so it
> > does
> > > > not
> > > > > > > have
> > > > > > > > > to be
> > > > > > > > > > >> > > > decided at
> > > > > > > > > > >> > > > > > > > > this stage. We will likely start by
> > trying
> > > > to
> > > > > > > > > refactor the
> > > > > > > > > > >> > > > current
> > > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > > >> > > > > > > > > David
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David
> > Jacot
> > > > <
> > > > > > > > > > >> djacot@confluent.io
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > > > "Empty,
> > > > > > > > > assigning,
> > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> > States
> > > > > > > > section,
> > > > > > > > > > >> right?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > This sentence does not refer to those
> > > > group
> > > > > > > states
> > > > > > > > > but
> > > > > > > > > > >> rather
> > > > > > > > > > >> > > > to a
> > > > > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > > > > refers to
> > > > > > > > the
> > > > > > > > > > >> entire
> > > > > > > > > > >> > > > state of
> > > > > > > > > > >> > > > > > > > > > group coordinator which is replicated
> > via
> > > > > the
> > > > > > > log
> > > > > > > > > > >> layer. I will
> > > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > > machine
> > > > > is
> > > > > > > > > modelled
> > > > > > > > > > >> as an
> > > > > > > > > > >> > > > event
> > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > The idea is to follow a model similar
> > to
> > > > > the new
> > > > > > > > > quorum
> > > > > > > > > > >> > > > > > controller. We
> > > > > > > > > > >> > > > > > > > > > will have N threads to process events.
> > > > Each
> > > > > > > > > > >> __consumer_offsets
> > > > > > > > > > >> > > > > > > > > > partition is assigned to a unique
> > thread
> > > > > and all
> > > > > > > > the
> > > > > > > > > > >> events
> > > > > > > > > > >> > > > (e.g.
> > > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> > processed
> > > > by
> > > > > this
> > > > > > > > > > >> thread. This
> > > > > > > > > > >> > > > > > simplify
> > > > > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > > > > simulation
> > > > > > > > > testing
> > > > > > > > > > >> for the
> > > > > > > > > > >> > > > > > group
> > > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> > per
> > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> > group"
> > > > > owned
> > > > > > > > by
> > > > > > > > > a
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> > owns 2
> > > > > > > > consumer
> > > > > > > > > > >> groups, and
> > > > > > > > > > >> > > > > > both
> > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > have 1
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > >> for it,
> > > > > > > > > > >> > > > or
> > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> > there, I
> > > > > > > think.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > "*group.coordinator.threads"
> > > > > > > *is
> > > > > > > > > the
> > > > > > > > > > >> number
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > threads
> > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > > wondering
> > > > > > > if
> > > > > > > > > the
> > > > > > > > > > >> purpose
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > > consumer
> > > > > > > group
> > > > > > > > > (or
> > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > computation,
> > > > > why
> > > > > > > > > should we
> > > > > > > > > > >> need
> > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an
> > ability to
> > > > > shard
> > > > > > > > the
> > > > > > > > > > >> > > > processing as
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> > why
> > > > > does
> > > > > > > the
> > > > > > > > > > >> default
> > > > > > > > > > >> > > > session
> > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > max(60s)? I
> > > > > > > thought
> > > > > > > > > the
> > > > > > > > > > >> min/max
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > bound
> > > > of
> > > > > it,
> > > > > > > > no?
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> > int
> > > > > 30s The
> > > > > > > > > > >> timeout to
> > > > > > > > > > >> > > > detect
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > group
> > > > > > > protocol.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.min.session.timeout.ms
> > > > > int 45s
> > > > > > > > The
> > > > > > > > > > >> minimum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.max.session.timeout.ms
> > > > > int 60s
> > > > > > > > The
> > > > > > > > > > >> maximum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > > > > session
> > > > > > > > > timeout
> > > > > > > > > > >> should
> > > > > > > > > > >> > > > be 45s
> > > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > assignor
> > > > are
> > > > > > > > [range,
> > > > > > > > > > >> uniform],
> > > > > > > > > > >> > > > > > which means
> > > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> > I'd
> > > > > like to
> > > > > > > > > know
> > > > > > > > > > >> why not
> > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > > uniform
> > > > > > > > assignor
> > > > > > > > > > >> (former
> > > > > > > > > > >> > > > sticky
> > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > > > other
> > > > > > > reason
> > > > > > > > > we
> > > > > > > > > > >> choose
> > > > > > > > > > >> > > > range
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > > > uniform
> > > > > > > The
> > > > > > > > > > >> server side
> > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > The order on the server side has no
> > > > > influence
> > > > > > > > > because
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > client
> > > > > > > > > > >> > > > > > must
> > > > > > > > > > >> > > > > > > > > > chose the selector that he wants to
> > use.
> > > > > There
> > > > > > > is
> > > > > > > > no
> > > > > > > > > > >> default
> > > > > > > > > > >> > > > in the
> > > > > > > > > > >> > > > > > > > > > current proposal. If the assignor is
> > not
> > > > > > > specified
> > > > > > > > > by
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > client,
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > request is rejected. The default
> > client
> > > > > value
> > > > > > > for
> > > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> > > > though.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > Thanks for your very good comments,
> > Luke.
> > > > I
> > > > > hope
> > > > > > > > > that my
> > > > > > > > > > >> > > > answers
> > > > > > > > > > >> > > > > > help
> > > > > > > > > > >> > > > > > > > > > to clarify things. I will update the
> > KIP
> > > > as
> > > > > well
> > > > > > > > > based
> > > > > > > > > > >> on your
> > > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke
> > Chen
> > > > <
> > > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > > > > mentioned:
> > > > > > > > > > >> > > > > > > > > > > > The new group coordinator will
> > have a
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > >> per
> > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions,
> > where
> > > > > each
> > > > > > > > state
> > > > > > > > > > >> machine is
> > > > > > > > > > >> > > > > > modelled
> > > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > > >> > > > > > > > > > > event loop. Those state machines
> > will be
> > > > > > > > executed
> > > > > > > > > in
> > > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > > > "Empty,
> > > > > > > > > assigning,
> > > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> > States
> > > > > > > > section,
> > > > > > > > > > >> right?
> > > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > > machine
> > > > > is
> > > > > > > > > modelled
> > > > > > > > > > >> as an
> > > > > > > > > > >> > > > event
> > > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> > per
> > > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> > group"
> > > > > owned
> > > > > > > > by
> > > > > > > > > a
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > > >> > > > > > > > > For
> > > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> > owns 2
> > > > > > > > consumer
> > > > > > > > > > >> groups, and
> > > > > > > > > > >> > > > > > both
> > > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> > have 1
> > > > > state
> > > > > > > > > machine
> > > > > > > > > > >> for it,
> > > > > > > > > > >> > > > or
> > > > > > > > > > >> > > > > > 2?
> > > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > > "*group.coordinator.threads"
> > > > > > > *is
> > > > > > > > > the
> > > > > > > > > > >> number
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > threads
> > > > > > > > > > >> > > > > > > > > used
> > > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > > wondering
> > > > > > > if
> > > > > > > > > the
> > > > > > > > > > >> purpose
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > > consumer
> > > > > > > group
> > > > > > > > > (or
> > > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> > computation,
> > > > > why
> > > > > > > > > should we
> > > > > > > > > > >> need
> > > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > > > > configs:
> > > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> > why
> > > > > does
> > > > > > > the
> > > > > > > > > > >> default
> > > > > > > > > > >> > > > session
> > > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> > max(60s)? I
> > > > > > > thought
> > > > > > > > > the
> > > > > > > > > > >> min/max
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> > bound
> > > > of
> > > > > it,
> > > > > > > > no?
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> > int
> > > > > 30s The
> > > > > > > > > > >> timeout to
> > > > > > > > > > >> > > > detect
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> > group
> > > > > > > protocol.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.min.session.timeout.ms
> > > > > int 45s
> > > > > > > > The
> > > > > > > > > > >> minimum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > > >
> > group.consumer.max.session.timeout.ms
> > > > > int 60s
> > > > > > > > The
> > > > > > > > > > >> maximum
> > > > > > > > > > >> > > > > > session
> > > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> > assignor
> > > > are
> > > > > > > > [range,
> > > > > > > > > > >> uniform],
> > > > > > > > > > >> > > > > > which means
> > > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> > I'd
> > > > > like to
> > > > > > > > > know
> > > > > > > > > > >> why not
> > > > > > > > > > >> > > > > > uniform
> > > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > > uniform
> > > > > > > > assignor
> > > > > > > > > > >> (former
> > > > > > > > > > >> > > > sticky
> > > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > > > other
> > > > > > > reason
> > > > > > > > > we
> > > > > > > > > > >> choose
> > > > > > > > > > >> > > > range
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > > > uniform
> > > > > > > The
> > > > > > > > > > >> server side
> > > > > > > > > > >> > > > > > assignors.
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke
> > > > Chen
> > > > > <
> > > > > > > > > > >> showuon@gmail.com
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> > > > Connect
> > > > > > > > > > >> integrating with
> > > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > > >> > > > > > > > > > > > think we should have a separate
> > > > > discussion
> > > > > > > > > thread
> > > > > > > > > > >> for the
> > > > > > > > > > >> > > > Kafka
> > > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect
> > With
> > > > New
> > > > > > > > Consumer
> > > > > > > > > > >> Rebalance
> > > > > > > > > > >> > > > > > Protocol
> > > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > > >> > > > > > > > > > > > and let this discussion thread
> > focus
> > > > on
> > > > > > > > consumer
> > > > > > > > > > >> rebalance
> > > > > > > > > > >> > > > > > protocol,
> > > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > >
> > > > > > > > > > >>
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM
> > Sagar
> > > > <
> > > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > > > > feedback.
> > > > > > > > > Looks
> > > > > > > > > > >> like
> > > > > > > > > > >> > > > there's
> > > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect.
> > I
> > > > > would
> > > > > > > > > revisit
> > > > > > > > > > >> the doc
> > > > > > > > > > >> > > > and
> > > > > > > > > > >> > > > > > see
> > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM
> > David
> > > > > Jacot
> > > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > > > > document.
> > > > > > > > > That's
> > > > > > > > > > >> really
> > > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that
> > both
> > > > > Connect
> > > > > > > > > and the
> > > > > > > > > > >> > > > Consumer
> > > > > > > > > > >> > > > > > could
> > > > > > > > > > >> > > > > > > > > share
> > > > > > > > > > >> > > > > > > > > > > >> > the same underlying "engine".
> > The
> > > > > main
> > > > > > > > > > >> difference is
> > > > > > > > > > >> > > > that
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> > members
> > > > > > > whereas
> > > > > > > > > > >> Connect
> > > > > > > > > > >> > > > assigns
> > > > > > > > > > >> > > > > > tasks
> > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > > > > forward:
> > > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed
> > APIs
> > > > to
> > > > > > > > support
> > > > > > > > > > >> different
> > > > > > > > > > >> > > > > > resource
> > > > > > > > > > >> > > > > > > > > types
> > > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks,
> > etc.); or
> > > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs
> > for
> > > > > Connect.
> > > > > > > > The
> > > > > > > > > > >> dedicated
> > > > > > > > > > >> > > > APIs
> > > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > > > different on
> > > > > > > > the
> > > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> > engine
> > > > > on the
> > > > > > > > > > >> coordinator
> > > > > > > > > > >> > > > side.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> > > > because
> > > > > I am
> > > > > > > > > not a
> > > > > > > > > > >> fan of
> > > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> > purposes.
> > > > > That
> > > > > > > > being
> > > > > > > > > > >> said, I am
> > > > > > > > > > >> > > > not
> > > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant
> > way to
> > > > > do
> > > > > > > it.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > > > > discuss
> > > > > > > it
> > > > > > > > > here
> > > > > > > > > > >> for now
> > > > > > > > > > >> > > > in
> > > > > > > > > > >> > > > > > order
> > > > > > > > > > >> > > > > > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> > compatible
> > > > > with
> > > > > > > > what
> > > > > > > > > we
> > > > > > > > > > >> will do
> > > > > > > > > > >> > > > for
> > > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM
> > > > David
> > > > > > > Jacot
> > > > > > > > <
> > > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I
> > will
> > > > go
> > > > > > > > through
> > > > > > > > > and
> > > > > > > > > > >> address
> > > > > > > > > > >> > > > > > your
> > > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks
> > for
> > > > your
> > > > > > > > > feedback.
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05
> > PM
> > > > > Gregory
> > > > > > > > > Harris
> > > > > > > > > > >> <
> > > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > > > > wonderful to
> > > > > > > > see
> > > > > > > > > > >> > > > cooperative
> > > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> > Alternatives'
> > > > > > > section
> > > > > > > > > > >> describes how
> > > > > > > > > > >> > > > > > member
> > > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > advance in step with the
> > group
> > > > > epoch
> > > > > > > > and
> > > > > > > > > > >> assignment
> > > > > > > > > > >> > > > > > epoch
> > > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea
> > for
> > > > the
> > > > > > > > reasons
> > > > > > > > > > >> described
> > > > > > > > > > >> > > > in
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > > > > assigning
> > > > > > > > > > >> partitions to a
> > > > > > > > > > >> > > > > > worker,
> > > > > > > > > > >> > > > > > > > > what
> > > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > > > > assignment
> > > > > > > > > use?
> > > > > > > > > > >> Are
> > > > > > > > > > >> > > > member
> > > > > > > > > > >> > > > > > epochs
> > > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > > > > correspond
> > > > > > > to
> > > > > > > > > > >> multiple
> > > > > > > > > > >> > > > > > different
> > > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's
> > 'Reason'
> > > > > field
> > > > > > > > > opaque
> > > > > > > > > > >> to the
> > > > > > > > > > >> > > > group
> > > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> > client-side
> > > > > > > assignor
> > > > > > > > > > >> > > > implementations
> > > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is
> > its
> > > > > common
> > > > > > > > > meaning
> > > > > > > > > > >> agreed
> > > > > > > > > > >> > > > > > upon? If
> > > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct
> > Reason
> > > > > field
> > > > > > > over
> > > > > > > > > > >> including
> > > > > > > > > > >> > > > such
> > > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> > included in
> > > > > the
> > > > > > > > KIP:
> > > > > > > > > > >> "Thanks to
> > > > > > > > > > >> > > > > > this, the
> > > > > > > > > > >> > > > > > > > > input
> > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > > > entirely
> > > > > > > driven
> > > > > > > > > by
> > > > > > > > > > >> the group
> > > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > > > responsible for
> > > > > > > > > > >> maintaining
> > > > > > > > > > >> > > > any
> > > > > > > > > > >> > > > > > state
> > > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does
> > this
> > > > > mean
> > > > > > > > > that the
> > > > > > > > > > >> > > > > > client-side
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > > > non-Metadata
> > > > > > > > state
> > > > > > > > > > >> (such as
> > > > > > > > > > >> > > > > > partition
> > > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > > > > topics,
> > > > > > > > etc),
> > > > > > > > > or
> > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > additional
> > > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > > > separate
> > > > > > > > classes
> > > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > > >> > > > > >
> > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > > >> > > >
> > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > > >> > > > > > that
> > > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is
> > it
> > > > > possible
> > > > > > > > for
> > > > > > > > > > >> these two
> > > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve
> > to
> > > > > promote
> > > > > > > > > feature
> > > > > > > > > > >> parity
> > > > > > > > > > >> > > > of
> > > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and
> > > > would
> > > > > also
> > > > > > > > > > >> facilitate
> > > > > > > > > > >> > > > > > operational
> > > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > > > example,
> > > > > if a
> > > > > > > > > > >> server-side
> > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > > > deploying
> > > > > > > > the
> > > > > > > > > > >> patched
> > > > > > > > > > >> > > > > > assignor to
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer
> > > > group
> > > > > to a
> > > > > > > > > > >> client-side
> > > > > > > > > > >> > > > > > assignor
> > > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> > patching
> > > > all
> > > > > of
> > > > > > > the
> > > > > > > > > > >> brokers.
> > > > > > > > > > >> > > > With
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > > > > non-trivial
> > > > > > > > > > >> > > > reimplementation
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two
> > APIs
> > > > > have
> > > > > > > > > diverged
> > > > > > > > > > >> > > > > > significantly,
> > > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > > reimplementation
> > > > > > > would
> > > > > > > > > not be
> > > > > > > > > > >> > > > possible.
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at
> > 8:39 AM
> > > > > Sagar
> > > > > > > <
> > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence
> > page
> > > > to
> > > > > > > > discuss
> > > > > > > > > how
> > > > > > > > > > >> Connect
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> > rebalance
> > > > > > > protocol.
> > > > > > > > > > >> Here's the
> > > > > > > > > > >> > > > page:
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > >
> > > > > > > > > > >>
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish
> > and
> > > > I
> > > > > have
> > > > > > > > > tried
> > > > > > > > > > >> to keep
> > > > > > > > > > >> > > > a
> > > > > > > > > > >> > > > > > format
> > > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know
> > what you
> > > > > > > think.
> > > > > > > > > Also,
> > > > > > > > > > >> do you
> > > > > > > > > > >> > > > > > think this
> > > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > > > discussion
> > > > > > > thread
> > > > > > > > > or
> > > > > > > > > > >> is this
> > > > > > > > > > >> > > > one
> > > > > > > > > > >> > > > > > fine?
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at
> > 7:37
> > > > AM
> > > > > > > > Sagar <
> > > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for
> > the
> > > > > doc on
> > > > > > > > > Kafka
> > > > > > > > > > >> Streams.
> > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up
> > with
> > > > > such a
> > > > > > > > > > >> document.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at
> > > > 4:47
> > > > > AM
> > > > > > > > > Guozhang
> > > > > > > > > > >> Wang <
> > > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if
> > you
> > > > > could
> > > > > > > > come
> > > > > > > > > > >> back with
> > > > > > > > > > >> > > > some
> > > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect
> > side
> > > > > > > > > integration
> > > > > > > > > > >> with
> > > > > > > > > > >> > > > the new
> > > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on
> > the
> > > > > new
> > > > > > > > > "protocol
> > > > > > > > > > >> type"
> > > > > > > > > > >> > > > we
> > > > > > > > > > >> > > > > > did not
> > > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > > > implementations.
> > > > > > > > > For
> > > > > > > > > > >> Streams
> > > > > > > > > > >> > > > > > here's a
> > > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > >
> > > > > > > > > > >>
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > >
> > > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your
> > analysis
> > > > > on
> > > > > > > > > Connect.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022
> > at
> > > > > 10:48 PM
> > > > > > > > > Sagar <
> > > > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > > > > response.
> > > > > > > The
> > > > > > > > > > >> reason I
> > > > > > > > > > >> > > > thought
> > > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is
> > that
> > > > > even
> > > > > > > > > today the
> > > > > > > > > > >> > > > connect
> > > > > > > > > > >> > > > > > uses a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > > > AbstractCoordinator to
> > > > > > > > > > >> empower
> > > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors.
> > The
> > > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> > metadata()
> > > > > method
> > > > > > > > by
> > > > > > > > > > >> plumbing
> > > > > > > > > > >> > > > into
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > JoinGroupRequestProtocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes
> > to
> > > > > support
> > > > > > > > > > >> connect would
> > > > > > > > > > >> > > > be
> > > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in
> > streams
> > > > > mainly
> > > > > > > > > because
> > > > > > > > > > >> of the
> > > > > > > > > > >> > > > > > Client
> > > > > > > > > > >> > > > > > > > > side
> > > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > > > > > implementation
> > > > > > > > > > >> level, we
> > > > > > > > > > >> > > > > > might
> > > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto
> > this
> > > > > new
> > > > > > > > > > >> assignment
> > > > > > > > > > >> > > > protocol
> > > > > > > > > > >> > > > > > like
> > > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > > request/response and
> > > > > > > > > > >> SyncGroup and
> > > > > > > > > > >> > > > > > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on
> > similar
> > > > > lines
> > > > > > > > to
> > > > > > > > > > >> streams (or
> > > > > > > > > > >> > > > > > there
> > > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> > perform a
> > > > > > > detailed
> > > > > > > > > > >> analysis of
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > same
> > > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> > > > > thread for
> > > > > > > > > that
> > > > > > > > > > >> as that
> > > > > > > > > > >> > > > would
> > > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know
> > if
> > > > > that
> > > > > > > > sounds
> > > > > > > > > > >> good to
> > > > > > > > > > >> > > > you.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15,
> > 2022 at
> > > > > 5:47
> > > > > > > PM
> > > > > > > > > David
> > > > > > > > > > >> Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > > > > comments.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That
> > refers to
> > > > > > > > > > >> `Assignment#error`.
> > > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > > >> > > > > > > > > can
> > > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > > > > > transition C
> > > > > > > > > from
> > > > > > > > > > >> his
> > > > > > > > > > >> > > > current
> > > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment
> > when
> > > > > he
> > > > > > > can
> > > > > > > > > move
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > epoch 3.
> > > > > > > > > > >> > > > > > When
> > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > is
> > > > > updated
> > > > > > > > and
> > > > > > > > > > >> persisted
> > > > > > > > > > >> > > > > > with all
> > > > > > > > > > >> > > > > > > > > its
> > > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if
> > > > they
> > > > > are
> > > > > > > > not
> > > > > > > > > all
> > > > > > > > > > >> revoked
> > > > > > > > > > >> > > > > > yet. In
> > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > > > > becomes
> > > > > > > the
> > > > > > > > > target
> > > > > > > > > > >> > > > > > assignment.
> > > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to
> > avoid
> > > > > having
> > > > > > > > to
> > > > > > > > > > >> write all
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are
> > based on
> > > > > the
> > > > > > > > > persisted
> > > > > > > > > > >> state
> > > > > > > > > > >> > > > so I
> > > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me
> > see
> > > > > if I
> > > > > > > can
> > > > > > > > > > >> improve
> > > > > > > > > > >> > > > this in
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding
> > Connect,
> > > > it
> > > > > > > could
> > > > > > > > > > >> reuse the
> > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it
> > fits in
> > > > > the
> > > > > > > > > > >> protocol. The
> > > > > > > > > > >> > > > > > assignment
> > > > > > > > > > >> > > > > > > > > is
> > > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > topicid-partitions +
> > > > > > > metadata,
> > > > > > > > > could
> > > > > > > > > > >> > > > Connect
> > > > > > > > > > >> > > > > > fit
> > > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15,
> > 2022
> > > > at
> > > > > 1:55
> > > > > > > > PM
> > > > > > > > > > >> Sagar <
> > > > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the
> > KIP.
> > > > I
> > > > > just
> > > > > > > > had
> > > > > > > > > > >> minor
> > > > > > > > > > >> > > > > > observations:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the
> > Assignment
> > > > > Error
> > > > > > > > > > >> section in
> > > > > > > > > > >> > > > Client
> > > > > > > > > > >> > > > > > Side
> > > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned
> > => `In
> > > > > this
> > > > > > > > > case,
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > client
> > > > > > > > > > >> > > > > > side
> > > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the
> > group
> > > > > > > > > coordinator`.
> > > > > > > > > > >> In this
> > > > > > > > > > >> > > > > > case are
> > > > > > > > > > >> > > > > > > > > you
> > > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> > returning
> > > > an
> > > > > > > > > > >> AssignmentError
> > > > > > > > > > >> > > > that's
> > > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you
> > think
> > > > > it
> > > > > > > > would
> > > > > > > > > > >> make sense
> > > > > > > > > > >> > > > to
> > > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case
> > > > Studies
> > > > > > > > > section, I
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > > slight
> > > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the
> > > > same.
> > > > > > > > Consider
> > > > > > > > > > >> this step:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> > heartbeats,
> > > > the
> > > > > > > group
> > > > > > > > > > >> coordinator
> > > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> > partitions
> > > > to
> > > > > > > > > revoke. It
> > > > > > > > > > >> > > > persists
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group
> > (epoch=3)
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > > > Assignment
> > > > > > > > > (epoch=3)
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > > > partitions=[foo-0]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > > > partitions=[foo-2]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > > > partitions=[foo-1]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > > > Assignment
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > epoch=2,
> > > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > epoch=3,
> > > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > epoch=3,
> > > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> > heartbeats,
> > > > it
> > > > > > > > > transitions
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > epoch 3
> > > > > > > > > > >> > > > > > but
> > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's
> > mentioned
> > > > > that
> > > > > > > > > member C
> > > > > > > > > > >> can't
> > > > > > > > > > >> > > > get
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > > > description
> > > > > > > > > above,
> > > > > > > > > > >> it seems
> > > > > > > > > > >> > > > it
> > > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better
> > to
> > > > > remove
> > > > > > > it
> > > > > > > > > and
> > > > > > > > > > >> > > > populate it
> > > > > > > > > > >> > > > > > only
> > > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a
> > lot
> > > > of
> > > > > > > other
> > > > > > > > > > >> places, so
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > I
> > > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding
> > connect ,
> > > > it
> > > > > > > might
> > > > > > > > > be
> > > > > > > > > > >> out of
> > > > > > > > > > >> > > > scope
> > > > > > > > > > >> > > > > > of
> > > > > > > > > > >> > > > > > > > > this
> > > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> > understood it
> > > > > would
> > > > > > > > > > >> probably be
> > > > > > > > > > >> > > > > > running in
> > > > > > > > > > >> > > > > > > > > client
> > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> > > > > rebalance
> > > > > > > > > > >> protocol as it
> > > > > > > > > > >> > > > has
> > > > > > > > > > >> > > > > > its
> > > > > > > > > > >> > > > > > > > > own
> > > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > > IncrementalCooperative).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15,
> > 2022
> > > > > at
> > > > > > > 5:00
> > > > > > > > > PM
> > > > > > > > > > >> David
> > > > > > > > > > >> > > > Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> > Hector! Our
> > > > > goal
> > > > > > > is
> > > > > > > > > to
> > > > > > > > > > >> move
> > > > > > > > > > >> > > > forward
> > > > > > > > > > >> > > > > > with
> > > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of
> > relying
> > > > > on
> > > > > > > one
> > > > > > > > > > >> generic API.
> > > > > > > > > > >> > > > For
> > > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same
> > pattern
> > > > > and
> > > > > > > > > > >> reuse/share the
> > > > > > > > > > >> > > > core
> > > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side.
> > For
> > > > the
> > > > > > > > schema
> > > > > > > > > > >> registry, I
> > > > > > > > > > >> > > > > > think
> > > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a
> > tailored
> > > > > API to
> > > > > > > > do
> > > > > > > > > > >> simple
> > > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul
> > 15,
> > > > > 2022 at
> > > > > > > > > 10:22
> > > > > > > > > > >> AM Ismael
> > > > > > > > > > >> > > > > > Juma <
> > > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> > > > > comments:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > > >> > > > vs
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > com.google.re2j.Pattern,
> > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > > > > > differences
> > > > > > > > > in
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > detail
> > > > > > > > > > >> > > > > > before
> > > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said,
> > if
> > > > > people
> > > > > > > > pass
> > > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics
> > to be
> > > > > > > honored.
> > > > > > > > > If
> > > > > > > > > > >> we are
> > > > > > > > > > >> > > > doing
> > > > > > > > > > >> > > > > > > > > something
> > > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should
> > consider
> > > > > adding
> > > > > > > > an
> > > > > > > > > > >> overload
> > > > > > > > > > >> > > > with
> > > > > > > > > > >> > > > > > our own
> > > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd
> > want
> > > > to
> > > > > > > > expose
> > > > > > > > > > >> re2j's at
> > > > > > > > > > >> > > > this
> > > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding
> > > > topic
> > > > > > > ids,
> > > > > > > > > any
> > > > > > > > > > >> major new
> > > > > > > > > > >> > > > > > protocol
> > > > > > > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and
> > should
> > > > > handle
> > > > > > > the
> > > > > > > > > topic
> > > > > > > > > > >> > > > > > recreation case
> > > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we
> > need to
> > > > > > > handle.
> > > > > > > > I
> > > > > > > > > > >> agree with
> > > > > > > > > > >> > > > > > David
> > > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the
> > relevant
> > > > > > > > protocols
> > > > > > > > > > >> that don't
> > > > > > > > > > >> > > > > > have it
> > > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> > > > > internals
> > > > > > > > > versus
> > > > > > > > > > >> adding
> > > > > > > > > > >> > > > new
> > > > > > > > > > >> > > > > > APIs
> > > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> > > > > adding
> > > > > > > new
> > > > > > > > > APIs
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > required
> > > > > > > > > > >> > > > > > for
> > > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am
> > still
> > > > not
> > > > > sure
> > > > > > > > > about
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> > > > > configs
> > > > > > > to
> > > > > > > > be
> > > > > > > > > > >> > > > centralized
> > > > > > > > > > >> > > > > > in the
> > > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > > > (auditability,
> > > > > > > > > > >> visibility,
> > > > > > > > > > >> > > > > > consistency,
> > > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > > > > automatically
> > > > > > > > > > >> deleting
> > > > > > > > > > >> > > > > > configs in a
> > > > > > > > > > >> > > > > > > > > way
> > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A
> > > > good
> > > > > > > > property
> > > > > > > > > > >> for modern
> > > > > > > > > > >> > > > > > systems
> > > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > unrecoverable
> > > > data
> > > > > > > loss
> > > > > > > > > > >> scenarios.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul
> > 13,
> > > > > 2022
> > > > > > > at
> > > > > > > > > 3:47
> > > > > > > > > > >> PM David
> > > > > > > > > > >> > > > > > Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > > > Guozhang. My
> > > > > > > > > answers
> > > > > > > > > > >> are
> > > > > > > > > > >> > > > below:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > > migration
> > > > > > > > path,
> > > > > > > > > > >> especially
> > > > > > > > > > >> > > > > > the last
> > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > enable
> > > > the
> > > > > new
> > > > > > > > > > >> protocol, in
> > > > > > > > > > >> > > > which
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > protocols /
> > > > > rpcs
> > > > > > > and
> > > > > > > > > old
> > > > > > > > > > >> > > > protocols /
> > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> > How
> > > > the
> > > > > > > > > > >> coordinator could
> > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > protocol
> > > > > is
> > > > > > > > > something
> > > > > > > > > > >> we
> > > > > > > > > > >> > > > need to
> > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I
> > just
> > > > > > > > published
> > > > > > > > > a
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > version
> > > > > > > > > > >> > > > > > of KIP
> > > > > > > > > > >> > > > > > > > > which
> > > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details
> > about
> > > > > this.
> > > > > > > > See
> > > > > > > > > the
> > > > > > > > > > >> > > > "Supporting
> > > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > > > > "Compatibility,
> > > > > > > > > > >> > > > Deprecation,
> > > > > > > > > > >> > > > > > and
> > > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to
> > > > think
> > > > > > > > through
> > > > > > > > > a
> > > > > > > > > > >> few cases
> > > > > > > > > > >> > > > > > now but
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> > > > > should be
> > > > > > > > > > >> understandable.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> > usage
> > > > > of
> > > > > > > > topic
> > > > > > > > > > >> ids. So
> > > > > > > > > > >> > > > far as
> > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> > RPCs
> > > > > and
> > > > > > > > admin
> > > > > > > > > > >> client,
> > > > > > > > > > >> > > > but
> > > > > > > > > > >> > > > > > they
> > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > consumers
> > > > > yet.
> > > > > > > I
> > > > > > > > > think
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> > to be
> > > > > > > > > maintaining
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > names
> > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> > the
> > > > > > > augmented
> > > > > > > > > > >> existing
> > > > > > > > > > >> > > > RPCs
> > > > > > > > > > >> > > > > > and the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > secondly,
> > > > > should
> > > > > > > we
> > > > > > > > > ever
> > > > > > > > > > >> consider
> > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> > APIs
> > > > as
> > > > > > > well
> > > > > > > > > (both
> > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > listener for
> > > > > cases
> > > > > > > > > like
> > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> > Assuming
> > > > > that we
> > > > > > > > > would
> > > > > > > > > > >> include
> > > > > > > > > > >> > > > > > converting
> > > > > > > > > > >> > > > > > > > > all
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to
> > using
> > > > > topic
> > > > > > > > ids
> > > > > > > > > in
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > KIP,
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully
> > operate
> > > > > with
> > > > > > > > topic
> > > > > > > > > > >> ids. That
> > > > > > > > > > >> > > > > > being
> > > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > > > names in
> > > > > > > > > various
> > > > > > > > > > >> APIs so
> > > > > > > > > > >> > > > > > having a
> > > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > > > inevitable to
> > > > > > > > me.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't
> > > > have
> > > > > a
> > > > > > > > strong
> > > > > > > > > > >> opinion on
> > > > > > > > > > >> > > > > > this.
> > > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> > > > > scope of
> > > > > > > > this
> > > > > > > > > > >> KIP. I
> > > > > > > > > > >> > > > would
> > > > > > > > > > >> > > > > > rather
> > > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and
> > we
> > > > can
> > > > > > > > consider
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > > > separately if
> > > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming
> > back to
> > > > > > > > Ismael's
> > > > > > > > > > >> point about
> > > > > > > > > > >> > > > > > using
> > > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > > >> > > > think
> > > > > > > > > > >> > > > > > that
> > > > > > > > > > >> > > > > > > > > there
> > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of
> > it.
> > > > > The
> > > > > > > > > consumer
> > > > > > > > > > >> will
> > > > > > > > > > >> > > > have
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > > > exists
> > > > > > > > before
> > > > > > > > > > >> passing
> > > > > > > > > > >> > > > them
> > > > > > > > > > >> > > > > > into
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> > > > > Obviously,
> > > > > > > > the
> > > > > > > > > > >> > > > coordinator
> > > > > > > > > > >> > > > > > will
> > > > > > > > > > >> > > > > > > > > also
> > > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really
> > > > have
> > > > > a
> > > > > > > way
> > > > > > > > to
> > > > > > > > > > >> reject an
> > > > > > > > > > >> > > > > > invalid
> > > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > agreeing
> > > > > with
> > > > > > > > > David
> > > > > > > > > > >> on all
> > > > > > > > > > >> > > > other
> > > > > > > > > > >> > > > > > minor
> > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > `subscribe(Pattern)`
> > > > > > > > > > >> question:
> > > > > > > > > > >> > > > > > personally
> > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > deprecate
> > > > the
> > > > > > > > > subscribe
> > > > > > > > > > >> API with
> > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> > just
> > > > > > > > documenting
> > > > > > > > > > >> that our
> > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> > the
> > > > > > > > incompatible
> > > > > > > > > > >> case is a
> > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > overloaded
> > > > > > > `String`
> > > > > > > > > based
> > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > > > regexes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> > > > > work. I
> > > > > > > > have
> > > > > > > > > to
> > > > > > > > > > >> look at
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> > > > > better
> > > > > > > > > > >> understand the
> > > > > > > > > > >> > > > > > potential
> > > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would
> > > > work
> > > > > for
> > > > > > > > all
> > > > > > > > > the
> > > > > > > > > > >> basic
> > > > > > > > > > >> > > > > > regular
> > > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between
> > the
> > > > two
> > > > > are
> > > > > > > > > mainly
> > > > > > > > > > >> about
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > various
> > > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder
> > what
> > > > > other
> > > > > > > > people
> > > > > > > > > > >> think
> > > > > > > > > > >> > > > about
> > > > > > > > > > >> > > > > > this.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue,
> > Jul
> > > > 12,
> > > > > 2022
> > > > > > > > at
> > > > > > > > > > >> 11:28 PM
> > > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks
> > > > David!
> > > > > I
> > > > > > > > think
> > > > > > > > > on
> > > > > > > > > > >> the high
> > > > > > > > > > >> > > > > > level
> > > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > > concretize
> > > > > a
> > > > > > > bit
> > > > > > > > > more:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > > migration
> > > > > > > > path,
> > > > > > > > > > >> especially
> > > > > > > > > > >> > > > > > the last
> > > > > > > > > > >> > > > > > > > > step
> > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > enable
> > > > the
> > > > > new
> > > > > > > > > > >> protocol, in
> > > > > > > > > > >> > > > which
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > protocols /
> > > > > rpcs
> > > > > > > and
> > > > > > > > > old
> > > > > > > > > > >> > > > protocols /
> > > > > > > > > > >> > > > > > rpcs
> > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> > How
> > > > the
> > > > > > > > > > >> coordinator could
> > > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> > protocol
> > > > > is
> > > > > > > > > something
> > > > > > > > > > >> we
> > > > > > > > > > >> > > > need to
> > > > > > > > > > >> > > > > > > > > present
> > > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> > usage
> > > > > of
> > > > > > > > topic
> > > > > > > > > > >> ids. So
> > > > > > > > > > >> > > > far as
> > > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> > RPCs
> > > > > and
> > > > > > > > admin
> > > > > > > > > > >> client,
> > > > > > > > > > >> > > > but
> > > > > > > > > > >> > > > > > they
> > > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > consumers
> > > > > yet.
> > > > > > > I
> > > > > > > > > think
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > question is,
> > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> > to be
> > > > > > > > > maintaining
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > names
> > > > > > > > > > >> > > > > > -> ids
> > > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> > the
> > > > > > > augmented
> > > > > > > > > > >> existing
> > > > > > > > > > >> > > > RPCs
> > > > > > > > > > >> > > > > > and the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > secondly,
> > > > > should
> > > > > > > we
> > > > > > > > > ever
> > > > > > > > > > >> consider
> > > > > > > > > > >> > > > > > exposing
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> > APIs
> > > > as
> > > > > > > well
> > > > > > > > > (both
> > > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > listener for
> > > > > cases
> > > > > > > > > like
> > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> > agreeing
> > > > > with
> > > > > > > > > David
> > > > > > > > > > >> on all
> > > > > > > > > > >> > > > other
> > > > > > > > > > >> > > > > > minor
> > > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > `subscribe(Pattern)`
> > > > > > > > > > >> question:
> > > > > > > > > > >> > > > > > personally
> > > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > deprecate
> > > > the
> > > > > > > > > subscribe
> > > > > > > > > > >> API with
> > > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > > >> > > > > > > > > but
> > > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> > just
> > > > > > > > documenting
> > > > > > > > > > >> that our
> > > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> > the
> > > > > > > > incompatible
> > > > > > > > > > >> case is a
> > > > > > > > > > >> > > > > > very rare
> > > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > overloaded
> > > > > > > `String`
> > > > > > > > > based
> > > > > > > > > > >> > > > > > subscription may
> > > > > > > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > > > regexes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue,
> > Jul
> > > > > 12,
> > > > > > > 2022
> > > > > > > > > at
> > > > > > > > > > >> 5:23 AM
> > > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi
> > Ismael,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Thanks for
> > > > > your
> > > > > > > > > > >> feedback. Let
> > > > > > > > > > >> > > > me
> > > > > > > > > > >> > > > > > answer
> > > > > > > > > > >> > > > > > > > > your
> > > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I
> > > > think
> > > > > > > it's
> > > > > > > > > > >> premature to
> > > > > > > > > > >> > > > > > talk about
> > > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > deprecation
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > removal
> > > > > of the
> > > > > > > > > > >> existing group
> > > > > > > > > > >> > > > > > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > client
> > > > > > > protocol
> > > > > > > > > and
> > > > > > > > > > >> hence
> > > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> > > > > quickly
> > > > > > > > > > >> applications
> > > > > > > > > > >> > > > migrate
> > > > > > > > > > >> > > > > > to
> > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That
> > makes
> > > > > > > sense.
> > > > > > > > I
> > > > > > > > > will
> > > > > > > > > > >> > > > remove it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2.
> > The
> > > > KIP
> > > > > > > says
> > > > > > > > we
> > > > > > > > > > >> intend to
> > > > > > > > > > >> > > > > > release
> > > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > why. If
> > > > we
> > > > > > > added
> > > > > > > > > that
> > > > > > > > > > >> as a
> > > > > > > > > > >> > > > way to
> > > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > group
> > > > > > > > > protocol, I
> > > > > > > > > > >> also
> > > > > > > > > > >> > > > > > suggest
> > > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > > explain
> > > > > > > my
> > > > > > > > > > >> reasoning. As
> > > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > coordinator
> > > > > in
> > > > > > > > Java
> > > > > > > > > > >> while we
> > > > > > > > > > >> > > > > > implement
> > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > the
> > > > > > > internals
> > > > > > > > > will
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > slightly
> > > > > > > > > > >> > > > > > > > > different
> > > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > Therefore, I
> > > > > > > > wanted
> > > > > > > > > to
> > > > > > > > > > >> tighten
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > coordinator
> > > > > to
> > > > > > > the
> > > > > > > > > new
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > alternative
> > > > > > > would
> > > > > > > > > be to
> > > > > > > > > > >> use a
> > > > > > > > > > >> > > > flag
> > > > > > > > > > >> > > > > > to do
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> > > > > software
> > > > > > > > > upgrade.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3.
> > We
> > > > > need to
> > > > > > > > > flesh
> > > > > > > > > > >> out the
> > > > > > > > > > >> > > > > > details of
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > we're
> > > > > saying
> > > > > > > we
> > > > > > > > > will
> > > > > > > > > > >> support
> > > > > > > > > > >> > > > > > online
> > > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > explain
> > > > > this
> > > > > > > in
> > > > > > > > > > >> detail. It
> > > > > > > > > > >> > > > could
> > > > > > > > > > >> > > > > > also
> > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > easier.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes,
> > we
> > > > will
> > > > > > > > support
> > > > > > > > > > >> online
> > > > > > > > > > >> > > > > > migrations
> > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > group
> > > > > using
> > > > > > > the
> > > > > > > > > old
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > > > will be
> > > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > > briefly
> > > > > > > > > explain
> > > > > > > > > > >> how that
> > > > > > > > > > >> > > > > > will work
> > > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four
> > step
> > > > > > > process:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> > > > > cluster
> > > > > > > > must
> > > > > > > > > be
> > > > > > > > > > >> > > > upgraded or
> > > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > group
> > > > > > > > > coordinator.
> > > > > > > > > > >> Both the
> > > > > > > > > > >> > > > > > old and
> > > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > support
> > > > the
> > > > > old
> > > > > > > > > > >> protocol and
> > > > > > > > > > >> > > > rely
> > > > > > > > > > >> > > > > > on the
> > > > > > > > > > >> > > > > > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they
> > can
> > > > > work
> > > > > > > > > together.
> > > > > > > > > > >> This
> > > > > > > > > > >> > > > point
> > > > > > > > > > >> > > > > > is an
> > > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > this
> > > > one
> > > > > live
> > > > > > > > > > >> because it
> > > > > > > > > > >> > > > would
> > > > > > > > > > >> > > > > > require
> > > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > coordinator
> > > > > and
> > > > > > > > > > >> starting up the
> > > > > > > > > > >> > > > > > new one
> > > > > > > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > unavailabilities.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> > > > > cluster's
> > > > > > > > > > >> metadata
> > > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > enable
> > > > > the
> > > > > > > new
> > > > > > > > > > >> protocol.
> > > > > > > > > > >> > > > This
> > > > > > > > > > >> > > > > > cannot
> > > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > terminated
> > > > > > > because
> > > > > > > > > the
> > > > > > > > > > >> old
> > > > > > > > > > >> > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > protocol.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> > > > > consumers
> > > > > > > > > must be
> > > > > > > > > > >> > > > upgraded
> > > > > > > > > > >> > > > > > to a
> > > > > > > > > > >> > > > > > > > > version
> > > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > migration
> > > > > (must
> > > > > > > > have
> > > > > > > > > > >> KIP-792).
> > > > > > > > > > >> > > > If
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Nothing
> > > > > must be
> > > > > > > > > done at
> > > > > > > > > > >> this
> > > > > > > > > > >> > > > point.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> > > > > consumers
> > > > > > > > > must
> > > > > > > > > > >> be rolled
> > > > > > > > > > >> > > > > > with the
> > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > consumer
> > > > > group
> > > > > > > is
> > > > > > > > > > >> automatically
> > > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using
> > the
> > > > > new
> > > > > > > > > protocol
> > > > > > > > > > >> joins
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > group.
> > > > > > > > > > >> > > > > > > > > While
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > > > > protocol are
> > > > > > > > > being
> > > > > > > > > > >> > > > upgraded,
> > > > > > > > > > >> > > > > > the old
> > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > one.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > > clarify
> > > > > > > all
> > > > > > > > > of
> > > > > > > > > > >> this in
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > KIP.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4.
> > I am
> > > > > happy
> > > > > > > > > that we
> > > > > > > > > > >> are
> > > > > > > > > > >> > > > > > pushing the
> > > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > server,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but
> > it
> > > > > seems
> > > > > > > > like
> > > > > > > > > > >> there
> > > > > > > > > > >> > > > could be
> > > > > > > > > > >> > > > > > some
> > > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > have a
> > > > > > > mechanism
> > > > > > > > > for
> > > > > > > > > > >> users to
> > > > > > > > > > >> > > > > > detect
> > > > > > > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > before
> > > > > > > switching
> > > > > > > > > to
> > > > > > > > > > >> the new
> > > > > > > > > > >> > > > > > protocol?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > think
> > > > > that I
> > > > > > > am
> > > > > > > > a
> > > > > > > > > bit
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > > optimistic
> > > > > > > > > > >> > > > > > > > > than
> > > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > believe
> > > > > that the
> > > > > > > > > > >> majority of
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > cases
> > > > > > > > > > >> > > > > > > > > are
> > > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work
> > with
> > > > > the
> > > > > > > new
> > > > > > > > > > >> engine. The
> > > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > reject
> > > > > the
> > > > > > > > > consumer
> > > > > > > > > > >> if the
> > > > > > > > > > >> > > > > > regex is
> > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > migration
> > > > > path,
> > > > > > > in
> > > > > > > > > the
> > > > > > > > > > >> worst
> > > > > > > > > > >> > > > case,
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > first
> > > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > group
> > > > > will
> > > > > > > be
> > > > > > > > > > >> rejected.
> > > > > > > > > > >> > > > This
> > > > > > > > > > >> > > > > > should
> > > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > say.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One
> > way
> > > > for
> > > > > > > > > customers to
> > > > > > > > > > >> > > > validate
> > > > > > > > > > >> > > > > > their
> > > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod
> > would
> > > > > be to
> > > > > > > > > test
> > > > > > > > > > >> them with
> > > > > > > > > > >> > > > > > another
> > > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could
> > be
> > > > > done
> > > > > > > in a
> > > > > > > > > > >> pre-prod
> > > > > > > > > > >> > > > > > environment.
> > > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > extend the
> > > > > > > > > > >> consumer-group tool
> > > > > > > > > > >> > > > to
> > > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > mechanism.
> > > > > Would
> > > > > > > > > this be
> > > > > > > > > > >> > > > enough in
> > > > > > > > > > >> > > > > > your
> > > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > > > > Related to
> > > > > > > > the
> > > > > > > > > last
> > > > > > > > > > >> > > > question,
> > > > > > > > > > >> > > > > > will
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > stick
> > > > > with the
> > > > > > > > > > >> current regex
> > > > > > > > > > >> > > > > > engine for
> > > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > example,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it
> > may
> > > > be
> > > > > > > handy
> > > > > > > > to
> > > > > > > > > > >> keep using
> > > > > > > > > > >> > > > > > client
> > > > > > > > > > >> > > > > > > > > based
> > > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > migrations
> > > > > > > > simple
> > > > > > > > > and
> > > > > > > > > > >> then
> > > > > > > > > > >> > > > > > migrate to
> > > > > > > > > > >> > > > > > > > > server
> > > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > > understand
> > > > > > > your
> > > > > > > > > point
> > > > > > > > > > >> but I
> > > > > > > > > > >> > > > am
> > > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > actually
> > > > > stay
> > > > > > > > in
> > > > > > > > > > >> this mode.
> > > > > > > > > > >> > > > That
> > > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > simplifying
> > > > > the
> > > > > > > > > client
> > > > > > > > > > >> because
> > > > > > > > > > >> > > > we
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > metadata on
> > > > > > > > the
> > > > > > > > > > >> client
> > > > > > > > > > >> > > > side. I
> > > > > > > > > > >> > > > > > would
> > > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6.
> > When
> > > > > we say
> > > > > > > > > that
> > > > > > > > > > >> the group
> > > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > > > > configurations
> > > > > > > > > > >> and that
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > group is
> > > > > > > > deleted.
> > > > > > > > > > >> Will a
> > > > > > > > > > >> > > > > > transition to
> > > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > configurations?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> > > > > right.
> > > > > > > The
> > > > > > > > > > >> > > > configurations
> > > > > > > > > > >> > > > > > will be
> > > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > deleted.
> > > > > They go
> > > > > > > > > > >> together.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7.
> > Will
> > > > > the
> > > > > > > > > choice to
> > > > > > > > > > >> store
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > harder
> > > > to
> > > > > list
> > > > > > > > all
> > > > > > > > > > >> cluster
> > > > > > > > > > >> > > > > > configs and
> > > > > > > > > > >> > > > > > > > > their
> > > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > don't
> > > > > think
> > > > > > > so.
> > > > > > > > > The
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > configs.
> > > > If
> > > > > you
> > > > > > > > > want to
> > > > > > > > > > >> know
> > > > > > > > > > >> > > > all
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > ask all
> > > > > the
> > > > > > > > group
> > > > > > > > > > >> > > > coordinators.
> > > > > > > > > > >> > > > > > You
> > > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > instance.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8.
> > How
> > > > > would
> > > > > > > > > someone
> > > > > > > > > > >> > > > configure a
> > > > > > > > > > >> > > > > > group
> > > > > > > > > > >> > > > > > > > > before
> > > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > > > > considered
> > > > > > > > > > >> allowing the
> > > > > > > > > > >> > > > > > explicit
> > > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > Alternatively,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > configs
> > > > > could
> > > > > > > be
> > > > > > > > > > >> decoupled
> > > > > > > > > > >> > > > from
> > > > > > > > > > >> > > > > > the
> > > > > > > > > > >> > > > > > > > > group
> > > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > The
> > > > > group
> > > > > > > > will
> > > > > > > > > be
> > > > > > > > > > >> > > > > > automatically
> > > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > configs
> > > > > will
> > > > > > > > be
> > > > > > > > > > >> lost after
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > passes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9.
> > Will
> > > > > the
> > > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > > >> > > > > > method
> > > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > > >> > > > do
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > That's a
> > > > > very
> > > > > > > > group
> > > > > > > > > > >> question. I
> > > > > > > > > > >> > > > > > forgot
> > > > > > > > > > >> > > > > > > > > about
> > > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > > >> > > > not
> > > > > > > > > > >> > > > > > fully
> > > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > plan to
> > > > > use,
> > > > > > > it
> > > > > > > > > > >> might be
> > > > > > > > > > >> > > > better
> > > > > > > > > > >> > > > > > to
> > > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which
> > > > takes
> > > > > a
> > > > > > > > > string.
> > > > > > > > > > >> We would
> > > > > > > > > > >> > > > > > rely on
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > During the
> > > > > > > > > migration, I
> > > > > > > > > > >> think
> > > > > > > > > > >> > > > that
> > > > > > > > > > >> > > > > > we
> > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > and
> > > > > use
> > > > > > > it.
> > > > > > > > > That
> > > > > > > > > > >> should
> > > > > > > > > > >> > > > > > work, I
> > > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10.
> > I
> > > > > agree
> > > > > > > with
> > > > > > > > > > >> Justine
> > > > > > > > > > >> > > > that we
> > > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > switch
> > > > to
> > > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > > >> > > > > > from the
> > > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> > > > > switch the
> > > > > > > > > > >> default for
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > feature
> > > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > metadata.version
> > > > > > > > > once
> > > > > > > > > > >> we
> > > > > > > > > > >> > > > want to
> > > > > > > > > > >> > > > > > make
> > > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My
> > plan
> > > > was
> > > > > to
> > > > > > > use
> > > > > > > > > that
> > > > > > > > > > >> feature
> > > > > > > > > > >> > > > > > flag
> > > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > phase. I
> > > > > should
> > > > > > > > not
> > > > > > > > > have
> > > > > > > > > > >> > > > mentioned
> > > > > > > > > > >> > > > > > it, I
> > > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > > > internal
> > > > > > > config
> > > > > > > > > for
> > > > > > > > > > >> it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11.
> > Some
> > > > > of
> > > > > > > the
> > > > > > > > > > >> protocol APIs
> > > > > > > > > > >> > > > > > don't
> > > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > good to
> > > > > add
> > > > > > > that
> > > > > > > > > for
> > > > > > > > > > >> > > > consistency.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12.
> > It
> > > > is
> > > > > a
> > > > > > > bit
> > > > > > > > > odd
> > > > > > > > > > >> that
> > > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > though
> > > > it
> > > > > > > seems
> > > > > > > > > to do
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > than
> > > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > agree.
> > > > > This is
> > > > > > > > > how the
> > > > > > > > > > >> > > > current
> > > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > require
> > > > > "Read
> > > > > > > > > Group" to
> > > > > > > > > > >> join a
> > > > > > > > > > >> > > > > > group. We
> > > > > > > > > > >> > > > > > > > > could
> > > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I
> > am
> > > > not
> > > > > > > sure
> > > > > > > > > that
> > > > > > > > > > >> it is
> > > > > > > > > > >> > > > worth
> > > > > > > > > > >> > > > > > it.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13.
> > How
> > > > is
> > > > > > > topic
> > > > > > > > > > >> recreation
> > > > > > > > > > >> > > > > > handled by
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > protocol?
> > > > > It
> > > > > > > > > would be
> > > > > > > > > > >> good to
> > > > > > > > > > >> > > > > > have a
> > > > > > > > > > >> > > > > > > > > section
> > > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > From
> > > > > a
> > > > > > > > > protocol
> > > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic
> > id
> > > > so
> > > > > it
> > > > > > > > will
> > > > > > > > > > >> treat it
> > > > > > > > > > >> > > > like a
> > > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only
> > issue
> > > > > is
> > > > > > > that
> > > > > > > > > the
> > > > > > > > > > >> > > > fetch/commit
> > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs
> > so the
> > > > > > > > consumer
> > > > > > > > > > >> would
> > > > > > > > > > >> > > > reuse the
> > > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > we
> > > > > should
> > > > > > > > > update
> > > > > > > > > > >> those
> > > > > > > > > > >> > > > APIs
> > > > > > > > > > >> > > > > > as well
> > > > > > > > > > >> > > > > > > > > in
> > > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > end.
> > > > That
> > > > > > > would
> > > > > > > > > > >> strengthen
> > > > > > > > > > >> > > > the
> > > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14.
> > The
> > > > > KIP
> > > > > > > > > mentions
> > > > > > > > > > >> we will
> > > > > > > > > > >> > > > > > write the
> > > > > > > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > this is
> > > > an
> > > > > > > > > > >> implementation
> > > > > > > > > > >> > > > > > detail, do
> > > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for
> > it?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15.
> > Do
> > > > we
> > > > > > > have a
> > > > > > > > > > >> scalability
> > > > > > > > > > >> > > > > > goal when
> > > > > > > > > > >> > > > > > > > > it
> > > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > group
> > > > > protocol
> > > > > > > > can
> > > > > > > > > > >> support?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We
> > don't
> > > > > have
> > > > > > > > > numbers
> > > > > > > > > > >> at the
> > > > > > > > > > >> > > > > > moment. The
> > > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of
> > members
> > > > > per
> > > > > > > > > group.
> > > > > > > > > > >> We will
> > > > > > > > > > >> > > > > > measure
> > > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > implementation.
> > > > > > > > Note
> > > > > > > > > > >> that we
> > > > > > > > > > >> > > > might
> > > > > > > > > > >> > > > > > have
> > > > > > > > > > >> > > > > > > > > other
> > > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road
> > (e.g.
> > > > > > > offset
> > > > > > > > > > >> commits).
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16.
> > Did
> > > > we
> > > > > > > > > consider
> > > > > > > > > > >> having
> > > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > SubscribedTopicNames
> > > > > > > > > > >> in
> > > > > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > since we
> > > > > have
> > > > > > > to
> > > > > > > > > > >> resolve the
> > > > > > > > > > >> > > > > > regex on
> > > > > > > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > topic
> > > > > > > name?
> > > > > > > > > The
> > > > > > > > > > >> > > > difference
> > > > > > > > > > >> > > > > > is that
> > > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > efficient
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > whereas
> > > > > > > sending
> > > > > > > > > the
> > > > > > > > > > >> topic
> > > > > > > > > > >> > > > names
> > > > > > > > > > >> > > > > > is less
> > > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > recreation is
> > > > > > > > > easier
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > handle
> > > > > > > > > > >> > > > > > if we
> > > > > > > > > > >> > > > > > > > > have
> > > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The
> > idea
> > > > > was to
> > > > > > > > > > >> consolidate the
> > > > > > > > > > >> > > > > > metadata
> > > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths
> > but
> > > > I
> > > > > do
> > > > > > > > agree
> > > > > > > > > > >> with your
> > > > > > > > > > >> > > > > > point. As
> > > > > > > > > > >> > > > > > > > > a
> > > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids
> > may be
> > > > > > > better
> > > > > > > > > here
> > > > > > > > > > >> for the
> > > > > > > > > > >> > > > > > delete and
> > > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > suppose
> > > > > that we
> > > > > > > > may
> > > > > > > > > > >> allow
> > > > > > > > > > >> > > > users to
> > > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> > > > > because
> > > > > > > > that
> > > > > > > > > is
> > > > > > > > > > >> the only
> > > > > > > > > > >> > > > > > way to be
> > > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > re-creation.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On
> > Tue,
> > > > Jul
> > > > > 12,
> > > > > > > > > 2022 at
> > > > > > > > > > >> 1:38 PM
> > > > > > > > > > >> > > > > > David
> > > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > djacot@confluent.io>
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > > > > Justine,
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > Thanks
> > > > for
> > > > > > > your
> > > > > > > > > > >> comments.
> > > > > > > > > > >> > > > Please
> > > > > > > > > > >> > > > > > find
> > > > > > > > > > >> > > > > > > > > my
> > > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > -
> > Yes,
> > > > > the new
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > > > relies on
> > > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > topic
> > > > > names
> > > > > > > > based
> > > > > > > > > in
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if
> > using
> > > > > topic
> > > > > > > > > names
> > > > > > > > > > >> is the
> > > > > > > > > > >> > > > > > right call
> > > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a
> > little
> > > > > more.
> > > > > > > > > > >> Obviously, the
> > > > > > > > > > >> > > > > > KIP does
> > > > > > > > > > >> > > > > > > > > not
> > > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > > >> > > > > > > > > > > >>
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hello Jun/David,

Just to chime in here for Jun's comment #50, today Streams does not have a
way to map topic names to topic IDs, and in its topology description topics
are encoded with names.

What I originally thought was that, on the consumer side, we'd need to
maintain this mapping from topic names to topic IDs anyways, and Streams
should leverage on the consumer's functionality to get the corresponding
topic names for the given topic IDs. But when I read through the RPC again
I realized it may not be the case, since:

1) in HB request, when clients send the subscription, they are still
sending them in terms of topic names. That is, the consumer would not try
to keep all the mappings from topic names to topic ids for the subscribed
topics.
2) in HB responses, the assignment's topics are encoded as IDs already. And
fetch requests can directly rely on that since they just need topic IDs and
do not care what actual topic names are.
3) in prepare-assignment responses, we return subscriptions as topic IDs as
well.

In a word, the consumer does not try to maintain the mapping of topic names
to IDs at all, and even if they do, they will not expose it in e.g.
assignor APIs. But for client-side assignors like Streams, if they do not
maintain the mapping from topic IDs to topic names, certain logic such as
co-partitioning indeed cannot be executed.

So I think we'd need to consider 1) use topic names in subscriptions in the
prepare-assignment responses, 2) use topic Names in client-side
PartitionAssignor#GroupMember#subscribedTopics and
PartitionAssignor#TopicMetadata#topicName.

On Wed, Sep 28, 2022 at 1:26 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Jun,
>
> Thanks for your comments.
>
> 48. Okay. There is a typo in the text:
> > Note that the persisted target assignment of B already includes foo-2
> here
> but B does not get it yet because foo-2 is not free.
>
> It should be:
> > Note that the persisted member assignment of B already includes foo-2
> here
> but B does not get it yet because foo-2 is not free.
>
> The examples are showing the persisted state in the __consumer_offsets
> topics. When a member transitions to the next epoch, we directly
> persist the whole target assignment in order to reduce the number of
> writes and we manage dependencies in memory. So "B - epoch=2,
> partitions=[foo-2]" is actually correct from a persisted record
> perspective. However, the in-memory state would be different because
> foo-2 is not yet available as you pointed out. The in-memory state
> would like as follow:
>
> Group (epoch=2)
> A
> B
> Target Assignment (epoch=2)
> A - partitions=[foo-0, foo-1]
> B - partitions=[foo-2]
> Member Assignment
> A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending partitions=[]
> B - epoch=2, partitions=[], pending partitions=[foo-2]
>
> I suppose that the confusion comes from this. I will go and convert
> all the examples to their in-memory representation in order to reduce
> the confusion.
>
> 50. Okay. I have to admit that I am not so familiar with the internals
> of KStreams. If I understand correctly, you're saying that KStreams
> may not have topic ids in its topology so topic names are missing here
> to match the topics. I was hoping that KStreams would have topic ids
> in the topology as well. If that is not the case, we can also add
> topic names on our side. I will check this.
>
> 53. It seems that I was not clear. Sorry for that. I did not mean to
> say that MemberAssignment.metadata is not used by the group
> coordinator. Let me re-explain the flow.
>
> PartitionAssignor.assign() returns an Assignment. Assignment contains
> MemberAssignments where each member has MemberAssignment.metadata. The
> Assignment is installed by the client with the
> ConsumerGroupInstallAssignment API and this overrides the current
> target assignment in the coordinator. Then the group coordinator uses
> it to reconcile members. Therefore the target assignment will
> eventually become the current member assignment and it will be
> delivered to the member, including the original
> MemberAssignment.metadata defines by the assignor.
>
> I hope that this helps to clarify how MemberAssignment.metadata is
> used afterwards.
>
> Best,
> David
>
> On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <ju...@confluent.io.invalid> wrote:
> > Hi, David,
> >
> > Thanks for the reply. A few more follow up questions.
> >
> > 48. Hmm, so member assignment is what the client sees. Then, I am not
> sure
> > about the following. The text says B cannot get foo-2 yet. But in Member
> > Assignment, B has foo-2.
> >
> > At this point B can transition to epoch 2 but cannot get foo-2 until A
> > revokes it.
> >
> > Note that the persisted target assignment of B already includes foo-2
> here
> > but B does not get it yet because foo-2 is not free.
> >
> > Group (epoch=2)
> > A
> > B
> > Target Assignment (epoch=2)
> > A - partitions=[foo-0, foo-1]
> > B - partitions=[foo-2]
> > Member Assignment
> > A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> > B - epoch=2, partitions=[foo-2]
> >
> >
> > 50. Hmm, I thought KStreams doesn't use range assignor all the time. It
> > does the assignment based on the topology. For example, if you are
> joining
> > two topics A and B and deriving the join result in a 3rd topic C, which
> > will be aggregated. You want to use range to co-locate the partitions
> from
> > topic A and B. However, for topic C, the partition could potentially be
> > assigned in round robin. Without the topic name, KStreams won't know
> which
> > topics need to be co-located.
> >
> > 53. If MemberAssignment.metadata is not used by the group coordinator,
> > where is it being used?
> >
> > Thanks,
> >
> > Jun
> >
> > On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson
> <ja...@confluent.io.invalid>
> > wrote:
> >
> > > Hi David,
> > >
> > > Thanks for the response. Replied below:
> > >
> > > > I don't have a strong opinion on this one. However, it seems that
> this
> > > could be a pandora box if our subset turns out to not be sufficient
> > > for some existing users. We will have to add more to it and the
> > > compatibility story between versions could be a real nightmare. One
> > > thing that we could do for sure is disabling all unicode groups in the
> > > library. That seems possible and would already reduce the scope.
> > >
> > > Changes to the regex syntax seems not a major problem to me. We would
> bump
> > > the protocol version as we usually do. Actually that begs the same
> question
> > > about RE2. How stable is the syntax and does it differ depending on the
> > > version of the library? In general, I do agree that we'd be better off
> not
> > > having to manage the syntax ourselves if we can convince ourselves that
> > > compatibility won't be a problem and the client won't eventually need
> to
> > > have access to it.
> > >
> > > Thanks,
> > > Jason
> > >
> > > On Tue, Sep 27, 2022 at 6:21 AM David Jacot
> <dj...@confluent.io.invalid>
> > > wrote:
> > >
> > > > Hi Luke,
> > > >
> > > > Thanks for your comments.
> > > >
> > > > Your suggestion makes sense. Jun's made the same one. We are also
> > > > discussing whether we should directly return the group state in the
> > > > response and drop the prepare assignment api. It seems that the
> > > > current consensus is to keep it though. If that stays like this, I
> > > > will apply your suggestion.
> > > >
> > > > Thanks,
> > > > David
> > > >
> > > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com>
> wrote:
> > > > >
> > > > > Hi David,
> > > > >
> > > > > Thanks for your patience for answering all these questions!
> > > > >
> > > > > One more question from me:
> > > > > 1. When the selected consumer is going to be in client-side
> assigning,
> > > we
> > > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm
> wondering if
> > > we
> > > > > could use a flag to indicate the assignment is required. The
> rationale
> > > is
> > > > > when returning error, we lose a chance to catch up with the latest
> > > target
> > > > > assignment or revoke partitions. Since we expected the record
> fetching
> > > is
> > > > > not blocked by the rebalance/assignment computing, we might want to
> > > allow
> > > > > the members to update the member assignment in each heartbeat.
> > > > >
> > > > > Luke
> > > > >
> > > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > > <ja...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > > > Hey David,
> > > > > >
> > > > > > Thanks for the response.
> > > > > >
> > > > > > > Do you think that we could have a case where two members with
> the
> > > > same
> > > > > > static id are active?
> > > > > >
> > > > > > I was just thinking about the handoff when we spin up a new node
> > > > before the
> > > > > > old one completes shutdown. It's probably ok to fence the old
> member
> > > > > > abruptly. I do recall we had a few issues with this handoff in
> the
> > > > past. It
> > > > > > was due to the fact that the replaced node might have been the
> > > leader.
> > > > > > Since we've removed the leader notion, perhaps this kind of issue
> > > won't
> > > > > > resurface.
> > > > > >
> > > > > > In regard to the RE2 library. I understand the intent to pass
> this
> > > > through
> > > > > > as a string without any validation so that we don't need the
> > > > dependence on
> > > > > > the client. I do wonder if that will be restrictive in practice.
> For
> > > > > > example, if a regex changes on the client side, would we need to
> > > > validate
> > > > > > against the current assignment to see which partitions to
> continue
> > > > > > fetching? Or would we instead invalidate the complete current
> > > > assignment?
> > > > > > Perhaps this is a rare case and we can do the latter. There
> could be
> > > > more
> > > > > > situations though.
> > > > > >
> > > > > > As an alternative, I wonder if we could define a minimalistic
> regex
> > > > syntax
> > > > > > (a subset of RE2 say) that could map more easily to regex support
> > > that
> > > > > > already exists in clients? For example, in this scenario, we are
> > > > matching
> > > > > > topics, so we only need to be able to represent topic characters
> and
> > > > not
> > > > > > the whole smorgasbord of whitespace and unicode options. Would it
> > > make
> > > > > > sense to have a reduced syntax which includes only standard topic
> > > > > > characters, wildcards, character classes, and repetitions?
> > > > > >
> > > > > > Best,
> > > > > > Jason
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <
> wangguoz@gmail.com>
> > > > wrote:
> > > > > >
> > > > > > > Hi David, thanks for all the detailed explanations. I think
> they
> > > all
> > > > make
> > > > > > > sense. Just want to have a couple follow-ups here:
> > > > > > >
> > > > > > > > I don't really see the benefits here because server side
> > > assignors
> > > > > > > don't have metadata at all. They only assign topic-partitions.
> They
> > > > > > > are not supposed to generate metadata nor to receive metadata
> from
> > > > the
> > > > > > > members.
> > > > > > >
> > > > > > > Just to clarify I was asking about the `version` of the
> assignor
> > > > (i.e. up
> > > > > > > to what version that the client would support), and I do agree
> we
> > > > would
> > > > > > not
> > > > > > > need metadata. What I have in mind is that, for some specific
> > > > built-in
> > > > > > > broker-assignors, e.g. rack-aware assignors, if it's possible
> that
> > > > in a
> > > > > > > newer version we would have a hierarchical rack ID string
> format,
> > > > like
> > > > > > > "tier1-tier2" etc, but if some client has not upgraded their
> rack
> > > ID
> > > > > > > would still be in old format. In this case, the broker then
> needs
> > > to
> > > > > > choose
> > > > > > > the old versioned assignor. I'm probably making something up
> here
> > > for
> > > > > > rack
> > > > > > > aware assignors, but I'm wondering if in general such an
> > > > "auto-downgrade"
> > > > > > > behavior would be needed still for broker-side assignor, and
> if yes
> > > > would
> > > > > > > "version" still be useful.
> > > > > > >
> > > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> called
> > > > once
> > > > > > > when the member transitions to a new epoch. This one contains
> the
> > > > full
> > > > > > > metadata provided by the client side assignor. Then,
> `onAssignment`
> > > > > > > can be called max N times where N is the number of partitions
> > > pending
> > > > > > > revocation by other members. Let me try to clarify this in the
> KIP.
> > > > > > >
> > > > > > > Okay, my understanding is that the calling ordering of these
> > > > callbacks
> > > > > > > would be like the following:
> > > > > > >
> > > > > > > ----------------------------------------
> > > > > > > onPartitionsRevoked();   // just once, since we do not really
> need
> > > > > > > to revoke incrementally.
> > > > > > >
> > > > > > > onAssignment();    // the first call, with epoch incremented
> > > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > > >
> > > > > > > onAssignment();              // the first onAssignment would
> bump
> > > up
> > > > the
> > > > > > > epoch, and the metadata reflected.
> > > > > > > onPartitionsAssigned();   // each time we get an additional
> > > > assignment,
> > > > > > we
> > > > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > > > ...
> > > > > > > onAssignment();
> > > > > > > onPartitionsAssigned();   // on each of the onAssignment
> calls, the
> > > > > > encoded
> > > > > > > metadata would not change, only the incrementally added
> partitions
> > > be
> > > > > > > reflected
> > > > > > >
> > > > > > > Is that the case?
> > > > > > >
> > > > > > > I'm wondering if we would still call onAssignment just once,
> that
> > > > encodes
> > > > > > > all the assignment for this rebalance, including all the
> partitions
> > > > that
> > > > > > > should be assigned to the member but not yet assigned since
> they
> > > > have not
> > > > > > > been revoked by others. In that case the call ordering would
> be:
> > > > > > >
> > > > > > > ----------------------------------------
> > > > > > > onPartitionsRevoked();   // just once
> > > > > > > onAssignment();    // just once, with epoch incremented, and
> > > metadata
> > > > > > > encoded changed, the "assignment" field also reflect the final
> > > target
> > > > > > > assignment
> > > > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > > > incrementally
> > > > > > > added partitions
> > > > > > > ...
> > > > > > > onPartitionsAssigned();
> > > > > > >
> > > > > > > The motivation from this is that, most users would only
> implement
> > > the
> > > > > > > rebalance callback listeners and hence we'd definitely need to
> make
> > > > sure
> > > > > > > the semantics of that does not change much, and the time
> > > > > > > `onPartitionsAssigned` indicate the time when the partitions
> are
> > > > actually
> > > > > > > assigned to it; while for assignors, the `onAssignment` is
> used to
> > > > > > indicate
> > > > > > > what decision is made regarding for this member, i.e. when the
> > > > partitions
> > > > > > > are decided to be given to it, but not necessarily meant that
> it
> > > has
> > > > been
> > > > > > > given, since that time should be determined by the time of
> > > > > > > `onPartitionsAssigned`. The benefits though, would be that
> assignor
> > > > > > > implementers would not need to reason which `onAssignment`
> would be
> > > > the
> > > > > > > last one for this epoch.
> > > > > > >
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > > <djacot@confluent.io.invalid
> > > > > > >
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hi Guozhang,
> > > > > > > >
> > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> Error
> > > > byte
> > > > > > > > field,
> > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> not.
> > > > And
> > > > > > the
> > > > > > > > > rationale seems to be assuming that we should always be
> able to
> > > > do
> > > > > > the
> > > > > > > > > assignment at the broker-side assignor without errors.
> > > > Personally I
> > > > > > > think
> > > > > > > > > it's still potentially beneficial to add the Error field
> even
> > > for
> > > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > > subscribed
> > > > > > > > > topics are not recognized with the current broker's
> metadata.
> > > > What do
> > > > > > > you
> > > > > > > > > think?
> > > > > > > >
> > > > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > > > better
> > > > > > > > to use an exception on the server side. This is what we
> usually
> > > do
> > > > for
> > > > > > > > server side plugins. On the client side, we use a field
> because
> > > the
> > > > > > > > errors are not defined in advance.
> > > > > > > >
> > > > > > > > Your comment also makes me think about what we should do
> when the
> > > > > > > > server side assignor fails. I suppose that we have to keep
> the
> > > > current
> > > > > > > > assignment until a new event occurs. For instance, in your
> > > example,
> > > > > > > > the coordinator would have to trigger a rebalance when
> > > unrecognized
> > > > > > > > topics are available. This would be part of the metadata
> > > > monitoring.
> > > > > > > >
> > > > > > > > > 2) The client-side "GroupMember" has three additional
> fields
> > > > > > > > > reason/version/metadata compared with the broker-side
> > > > GroupMember. I
> > > > > > > > agree
> > > > > > > > > that broker-side assignor would not need reason/metadata
> since
> > > > they
> > > > > > are
> > > > > > > > > blackbox strings/bytes to the assignor, but what about
> version?
> > > > E.g.
> > > > > > is
> > > > > > > > it
> > > > > > > > > possible that we evolve our broker-side built-in assignor
> but
> > > > the old
> > > > > > > > > versioned clients would not be able to work with the new
> > > > version, in
> > > > > > > > which
> > > > > > > > > case we need to let the broker being aware of this and
> upgrade
> > > > its
> > > > > > > > behavior
> > > > > > > > > to cooperate with the clients?
> > > > > > > >
> > > > > > > > I don't really see the benefits here because server side
> > > assignors
> > > > > > > > don't have metadata at all. They only assign
> topic-partitions.
> > > They
> > > > > > > > are not supposed to generate metadata nor to receive metadata
> > > from
> > > > the
> > > > > > > > members.
> > > > > > > >
> > > > > > > > > 3) Also related to 2) above, for the client-side
> "GroupMember",
> > > > > > instead
> > > > > > > > of
> > > > > > > > > including these three fields, what about just adding the
> > > > "Metadata"
> > > > > > > field
> > > > > > > > > class which has these three fields? Also, there are two
> > > > "Metadata"
> > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > reason/version/metadata, and the second is just the encoded
> > > > metadata
> > > > > > > > bytes.
> > > > > > > > > I'm wondering what about just naming the first as
> > > memberMetadata,
> > > > > > which
> > > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > > second
> > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > >
> > > > > > > > That's a good point. Let me try to rationalize this interface
> > > > based on
> > > > > > > > your suggestions.
> > > > > > > >
> > > > > > > > Best,
> > > > > > > > David
> > > > > > > >
> > > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > Hello David,
> > > > > > > > >
> > > > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > > > >
> > > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> Error
> > > > byte
> > > > > > > > field,
> > > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> not.
> > > > And
> > > > > > the
> > > > > > > > > rationale seems to be assuming that we should always be
> able to
> > > > do
> > > > > > the
> > > > > > > > > assignment at the broker-side assignor without errors.
> > > > Personally I
> > > > > > > think
> > > > > > > > > it's still potentially beneficial to add the Error field
> even
> > > for
> > > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > > subscribed
> > > > > > > > > topics are not recognized with the current broker's
> metadata.
> > > > What do
> > > > > > > you
> > > > > > > > > think?
> > > > > > > > >
> > > > > > > > > 2) The client-side "GroupMember" has three additional
> fields
> > > > > > > > > reason/version/metadata compared with the broker-side
> > > > GroupMember. I
> > > > > > > > agree
> > > > > > > > > that broker-side assignor would not need reason/metadata
> since
> > > > they
> > > > > > are
> > > > > > > > > blackbox strings/bytes to the assignor, but what about
> version?
> > > > E.g.
> > > > > > is
> > > > > > > > it
> > > > > > > > > possible that we evolve our broker-side built-in assignor
> but
> > > > the old
> > > > > > > > > versioned clients would not be able to work with the new
> > > > version, in
> > > > > > > > which
> > > > > > > > > case we need to let the broker being aware of this and
> upgrade
> > > > its
> > > > > > > > behavior
> > > > > > > > > to cooperate with the clients?
> > > > > > > > >
> > > > > > > > > 3) Also related to 2) above, for the client-side
> "GroupMember",
> > > > > > instead
> > > > > > > > of
> > > > > > > > > including these three fields, what about just adding the
> > > > "Metadata"
> > > > > > > field
> > > > > > > > > class which has these three fields? Also, there are two
> > > > "Metadata"
> > > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > > reason/version/metadata, and the second is just the encoded
> > > > metadata
> > > > > > > > bytes.
> > > > > > > > > I'm wondering what about just naming the first as
> > > memberMetadata,
> > > > > > which
> > > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > > second
> > > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > > wangguoz@gmail.com>
> > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hello David,
> > > > > > > > > >
> > > > > > > > > > Thanks for bringing this question up. I think the main
> > > > benefits as
> > > > > > > you
> > > > > > > > > > listed is 2) above if it stays; just to clarify, we would
> > > only
> > > > be
> > > > > > > able
> > > > > > > > to
> > > > > > > > > > save one round trip if the rebalance is still triggered
> by
> > > the
> > > > > > > broker;
> > > > > > > > if
> > > > > > > > > > the rebalance is triggered by the client then the
> > > > num.round.trips
> > > > > > are
> > > > > > > > the
> > > > > > > > > > same:
> > > > > > > > > >
> > > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > > >
> > > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > > already
> > > > > > > sent
> > > > > > > > a
> > > > > > > > > > HB and hence has to wait for it to return first since
> only
> > > one
> > > > > > > request
> > > > > > > > /
> > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > GroupInstallAssignment
> > > > > > > > > > request.
> > > > > > > > > >
> > > > > > > > > > In total, two round trips.
> > > > > > > > > >
> > > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > > >
> > > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > > already
> > > > > > > sent
> > > > > > > > a
> > > > > > > > > > HB and hence has to wait for it to return first since
> only
> > > one
> > > > > > > request
> > > > > > > > /
> > > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > > T1: client receives the HB response, and then sends the
> new
> > > HB
> > > > > > > request
> > > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > > T2: the HB response with the optional member metadata
> map is
> > > > > > > returned.
> > > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > > GroupInstallAssignment
> > > > > > > > > > request.
> > > > > > > > > >
> > > > > > > > > > In total, two round trips as well.
> > > > > > > > > >
> > > > > > > > > > -----------------------------
> > > > > > > > > >
> > > > > > > > > > So to complete the full picture here, we'd need to modify
> > > both
> > > > HB
> > > > > > > > request
> > > > > > > > > > and response so that the client can also indicate a new
> > > > rebalance
> > > > > > via
> > > > > > > > the
> > > > > > > > > > HB request as well, right?
> > > > > > > > > >
> > > > > > > > > > Assuming all above is true, I think it's okay to merge
> the
> > > > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > > > additional
> > > > > > > > fields
> > > > > > > > > > encoding the full member (subscription) metadata and
> topic
> > > > metadata
> > > > > > > as
> > > > > > > > > > optional fields.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Guozhang
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > >> Hi all,
> > > > > > > > > >>
> > > > > > > > > >> During an offline conversation, someone asked why we
> need
> > > the
> > > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we
> > > could
> > > > > > > instead
> > > > > > > > > >> provide the group state in the heartbeat response. This
> has
> > > a
> > > > few
> > > > > > > > > >> advantages: 1) it does not require using a special error
> > > code
> > > > to
> > > > > > > > > >> signal that a new assignment is required as the signal
> would
> > > > be
> > > > > > the
> > > > > > > > > >> provided group state; 2) it removes one round trip when
> a
> > > > client
> > > > > > > side
> > > > > > > > > >> assignor is used. The downside is that it makes the
> > > heartbeat
> > > > > > > > > >> response's definition quite large. I recall that I went
> with
> > > > the
> > > > > > > > > >> current approach due to this.
> > > > > > > > > >>
> > > > > > > > > >> Providing the group state in the heartbeat response is
> > > > appealing.
> > > > > > > What
> > > > > > > > > >> do you guys think?
> > > > > > > > > >>
> > > > > > > > > >> Best,
> > > > > > > > > >> David
> > > > > > > > > >>
> > > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > > djacot@confluent.io>
> > > > > > > > wrote:
> > > > > > > > > >> >
> > > > > > > > > >> > Hi Guozhang,
> > > > > > > > > >> >
> > > > > > > > > >> > 1. I have added a reference to the relevant chapter
> > > instead
> > > > of
> > > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > > >> >
> > > > > > > > > >> > 2. The "Rebalance Triggers" section you are referring
> to
> > > is
> > > > > > about
> > > > > > > > when
> > > > > > > > > >> > a rebalance should be triggered for the non-upgraded
> > > members
> > > > > > using
> > > > > > > > the
> > > > > > > > > >> > old protocol. The section mentions that a rebalance
> must
> > > be
> > > > > > > > triggered
> > > > > > > > > >> > when a new assignment is installed. This implies that
> the
> > > > group
> > > > > > > > epoch
> > > > > > > > > >> > was updated either by a native member or a
> non-upgraded
> > > > member.
> > > > > > > For
> > > > > > > > > >> > the latter, the JoinGroup request would be the
> trigger. I
> > > > have
> > > > > > > > added a
> > > > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > > > Handling"
> > > > > > > > section
> > > > > > > > > >> > as well. Does that make sense?
> > > > > > > > > >> >
> > > > > > > > > >> > Thanks,
> > > > > > > > > >> > David
> > > > > > > > > >> >
> > > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > > wangguoz@gmail.com
> > > > > > > >
> > > > > > > > > >> wrote:
> > > > > > > > > >> > >
> > > > > > > > > >> > > Hello David,
> > > > > > > > > >> > >
> > > > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > > > clear in
> > > > > > > the
> > > > > > > > doc,
> > > > > > > > > >> > > could we update:
> > > > > > > > > >> > >
> > > > > > > > > >> > > 1) the heartbeat request handling section, stating
> when
> > > > > > > > coordinator
> > > > > > > > > >> will
> > > > > > > > > >> > > trigger rebalance based on the HB's member metadata
> /
> > > > reason?
> > > > > > > > > >> > > 2) the "Rebalance Triggers" section to include what
> we
> > > > > > described
> > > > > > > > in
> > > > > > > > > >> "Group
> > > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > > >> > >
> > > > > > > > > >> > >
> > > > > > > > > >> > > Guozhang
> > > > > > > > > >> > >
> > > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > >> > > wrote:
> > > > > > > > > >> > >
> > > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > I thought that the assignor will always be
> consulted
> > > > when
> > > > > > the
> > > > > > > > next
> > > > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > > > >> > > > `PartitionAssignor#metadata` will be called for
> every
> > > > > > > heartbeat.
> > > > > > > > > >> This
> > > > > > > > > >> > > > gives the opportunity for the assignor to enforce
> a
> > > > > > rebalance
> > > > > > > by
> > > > > > > > > >> > > > setting the reason to a non-zero value or by
> changing
> > > > the
> > > > > > > > bytes. Do
> > > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > > concerned by
> > > > > > > the
> > > > > > > > > >> delay?
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > Best,
> > > > > > > > > >> > > > David
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > > wangguoz@gmail.com>
> > > > > > > > > >> wrote:
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > Hello David,
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > ```
> > > > > > > > > >> > > > > In this case, a new assignment is triggered by
> the
> > > > client
> > > > > > > side
> > > > > > > > > >> > > > > assignor. When constructing the HB, the consumer
> > > will
> > > > > > always
> > > > > > > > > >> consult
> > > > > > > > > >> > > > > the client side assignor and propagate the
> > > > information to
> > > > > > > the
> > > > > > > > > >> group
> > > > > > > > > >> > > > > coordinator. In other words, we don't expect
> users
> > > to
> > > > call
> > > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > >> > > > > ```
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > > > interface,
> > > > > > we
> > > > > > > > > >> actually do
> > > > > > > > > >> > > > > not have a way yet to instruct how to construct
> the
> > > > next
> > > > > > HB
> > > > > > > > > >> request, e.g.
> > > > > > > > > >> > > > > when the assignor wants to enforce a new
> rebalance
> > > > with a
> > > > > > > new
> > > > > > > > > >> assignment,
> > > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > > PartitionAssignor
> > > > > > > > to
> > > > > > > > > >> indicate
> > > > > > > > > >> > > > > the next HB telling broker about so. WDYT about
> > > adding
> > > > > > such
> > > > > > > an
> > > > > > > > > >> API on the
> > > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > Guozhang
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > > >> > > > > wrote:
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > I have updated the KIP to include your
> feedback. I
> > > > have
> > > > > > > also
> > > > > > > > > >> tried to
> > > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > Best,
> > > > > > > > > >> > > > > > David
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > > > djacot@confluent.io
> > > > > > > > > >> >
> > > > > > > > > >> > > > wrote:
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > > > answering
> > > > > > your
> > > > > > > > > >> questions
> > > > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> benefits
> > > > of
> > > > > > the
> > > > > > > > KIP
> > > > > > > > > >> seem to
> > > > > > > > > >> > > > be
> > > > > > > > > >> > > > > > fewer
> > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > support of
> > > > > > > > > >> wildcard. A few
> > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > I would also add that the KIP removes the
> global
> > > > sync
> > > > > > > > barrier
> > > > > > > > > >> in the
> > > > > > > > > >> > > > > > > protocol which is essential to improve group
> > > > stability
> > > > > > > and
> > > > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > > > client by
> > > > > > > > moving
> > > > > > > > > >> most of
> > > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> plan
> > > > to
> > > > > > > > support
> > > > > > > > > >> rolling
> > > > > > > > > >> > > > > > changing
> > > > > > > > > >> > > > > > > > of the partition assignor in the
> consumers?
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Definitely. The group coordinator will use
> the
> > > > > > assignor
> > > > > > > > used
> > > > > > > > > >> by a
> > > > > > > > > >> > > > > > > majority of the members. This allows the
> group
> > > to
> > > > move
> > > > > > > > from
> > > > > > > > > >> one
> > > > > > > > > >> > > > > > > assignor to another by a roll. This is
> explained
> > > > in
> > > > > > the
> > > > > > > > > >> Assignor
> > > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> whether
> > > > it's
> > > > > > > > > >> required in
> > > > > > > > > >> > > > every
> > > > > > > > > >> > > > > > > > request or the scenarios when it needs to
> be
> > > > filled?
> > > > > > > For
> > > > > > > > > >> example,
> > > > > > > > > >> > > > it's
> > > > > > > > > >> > > > > > not
> > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to
> be
> > > > filled.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > The client is expected to set those fields
> in
> > > > case of
> > > > > > a
> > > > > > > > > >> connection
> > > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > > > changed
> > > > > > > since
> > > > > > > > > >> the last
> > > > > > > > > >> > > > > > > HB. The server populates those fields as
> long as
> > > > the
> > > > > > > > member
> > > > > > > > > >> is not
> > > > > > > > > >> > > > > > > fully reconciled - the member should
> acknowledge
> > > > that
> > > > > > it
> > > > > > > > has
> > > > > > > > > >> the
> > > > > > > > > >> > > > > > > expected epoch and assignment. I will
> clarify
> > > > this in
> > > > > > > the
> > > > > > > > KIP.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> rack
> > > > > > > affinity
> > > > > > > > > >> between the
> > > > > > > > > >> > > > > > client
> > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > fetching,
> > > > > > but
> > > > > > > > not
> > > > > > > > > >> during
> > > > > > > > > >> > > > > > assigning
> > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> the
> > > > > > > assignment
> > > > > > > > is
> > > > > > > > > >> made,
> > > > > > > > > >> > > > there
> > > > > > > > > >> > > > > > is
> > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> no
> > > > replicas
> > > > > > > of
> > > > > > > > > >> assigned
> > > > > > > > > >> > > > > > partitions
> > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> if we
> > > > should
> > > > > > > use
> > > > > > > > > >> this
> > > > > > > > > >> > > > > > opportunity
> > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > GroupMember.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > That's an interesting idea. I don't see any
> > > issue
> > > > with
> > > > > > > > adding
> > > > > > > > > >> the
> > > > > > > > > >> > > > rack
> > > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> useful to
> > > > know
> > > > > > how
> > > > > > > > busy
> > > > > > > > > >> a group
> > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > model,
> > > > it
> > > > > > > seems
> > > > > > > > > >> that we
> > > > > > > > > >> > > > could
> > > > > > > > > >> > > > > > add
> > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> time
> > > > the
> > > > > > > event
> > > > > > > > > >> loop is
> > > > > > > > > >> > > > doing
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > > failover
> > > > > > > > > >> handling? For
> > > > > > > > > >> > > > > > example,
> > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> group
> > > > with
> > > > > > > the
> > > > > > > > > >> wildcard
> > > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Sure. When the new group coordinator takes
> over,
> > > > it
> > > > > > has
> > > > > > > > to:
> > > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > > > assignor
> > > > > > is
> > > > > > > > used.
> > > > > > > > > >> We
> > > > > > > > > >> > > > don't
> > > > > > > > > >> > > > > > > store the information about the member
> selected
> > > > to run
> > > > > > > the
> > > > > > > > > >> assignment
> > > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > > wildcard
> > > > > > > > > >> subscriptions, and
> > > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > > document
> > > > > > > what
> > > > > > > > > >> the epoch
> > > > > > > > > >> > > > > > field
> > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > >> > > > > > reflect
> > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> one in
> > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Sure. I will clarify that but it is always
> the
> > > > latest
> > > > > > > > group
> > > > > > > > > >> epoch.
> > > > > > > > > >> > > > > > > When the group state is updated, the group
> epoch
> > > > is
> > > > > > > > bumped so
> > > > > > > > > >> we use
> > > > > > > > > >> > > > > > > that one for all the change records related
> to
> > > the
> > > > > > > update.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> that
> > > the
> > > > > > > > following
> > > > > > > > > >> > > > invariants
> > > > > > > > > >> > > > > > are
> > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> possible
> > > for
> > > > a
> > > > > > > member
> > > > > > > > > >> not to
> > > > > > > > > >> > > > get any
> > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > That's right. Here I meant that the members
> > > > provided
> > > > > > by
> > > > > > > > the
> > > > > > > > > >> assignor
> > > > > > > > > >> > > > > > > in the assignment must exist in the group.
> The
> > > > > > assignor
> > > > > > > > can
> > > > > > > > > >> not make
> > > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> member
> > > > epoch
> > > > > > > > equals to
> > > > > > > > > >> 0":
> > > > > > > > > >> > > > When
> > > > > > > > > >> > > > > > would
> > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> would be
> > > > used?
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > > partitions and
> > > > > > > > > >> rejoins when
> > > > > > > > > >> > > > it
> > > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In
> this
> > > > case,
> > > > > > > the
> > > > > > > > > >> group
> > > > > > > > > >> > > > > > > coordinator will have removed the member
> from
> > > the
> > > > > > group.
> > > > > > > > The
> > > > > > > > > >> member
> > > > > > > > > >> > > > > > > can rejoin the group with the same member
> id but
> > > > with
> > > > > > 0
> > > > > > > as
> > > > > > > > > >> epoch. Let
> > > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > ability
> > > > to
> > > > > > > > trigger a
> > > > > > > > > >> > > > > > reassignment
> > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> by
> > > > updating
> > > > > > > the
> > > > > > > > > >> assignor
> > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> conflicting
> > > > with
> > > > > > the
> > > > > > > > > >> deprecation
> > > > > > > > > >> > > > of
> > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > In this case, a new assignment is triggered
> by
> > > the
> > > > > > > client
> > > > > > > > side
> > > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> consumer
> > > > will
> > > > > > > > always
> > > > > > > > > >> consult
> > > > > > > > > >> > > > > > > the client side assignor and propagate the
> > > > information
> > > > > > > to
> > > > > > > > the
> > > > > > > > > >> group
> > > > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > > > users to
> > > > > > > call
> > > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> But
> > > the
> > > > > > > section
> > > > > > > > > >> seems to
> > > > > > > > > >> > > > have
> > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> 2, B
> > > > > > > > immediately
> > > > > > > > > >> gets into
> > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > > incorrect.
> > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> 3, C
> > > > seems
> > > > > > to
> > > > > > > > get
> > > > > > > > > >> into
> > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> still
> > > > has A -
> > > > > > > > > >> epoch=2,
> > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > support
> > > > the
> > > > > > > > upgrade
> > > > > > > > > >> from
> > > > > > > > > >> > > > any
> > > > > > > > > >> > > > > > old
> > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > We will support upgrading from the consumer
> > > > protocol
> > > > > > > > version
> > > > > > > > > >> 3,
> > > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > > implemented
> > > > yet
> > > > > > so
> > > > > > > > the
> > > > > > > > > >> earliest
> > > > > > > > > >> > > > > > > version is unknown at the moment. This is
> > > > explained in
> > > > > > > the
> > > > > > > > > >> migration
> > > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> > > > update the
> > > > > > > KIP
> > > > > > > > > >> based on
> > > > > > > > > >> > > > it
> > > > > > > > > >> > > > > > > next week.
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > Best,
> > > > > > > > > >> > > > > > > David
> > > > > > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > > >> > > > wrote:
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> benefits
> > > > of
> > > > > > the
> > > > > > > > KIP
> > > > > > > > > >> seem to
> > > > > > > > > >> > > > be
> > > > > > > > > >> > > > > > fewer
> > > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > > support of
> > > > > > > > > >> wildcard. A few
> > > > > > > > > >> > > > > > > > comments below.
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> plan
> > > > to
> > > > > > > > support
> > > > > > > > > >> rolling
> > > > > > > > > >> > > > > > changing
> > > > > > > > > >> > > > > > > > of the partition assignor in the
> consumers?
> > > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> whether
> > > > it's
> > > > > > > > > >> required in
> > > > > > > > > >> > > > every
> > > > > > > > > >> > > > > > > > request or the scenarios when it needs to
> be
> > > > filled?
> > > > > > > For
> > > > > > > > > >> example,
> > > > > > > > > >> > > > it's
> > > > > > > > > >> > > > > > not
> > > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to
> be
> > > > filled.
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> rack
> > > > > > > affinity
> > > > > > > > > >> between the
> > > > > > > > > >> > > > > > client
> > > > > > > > > >> > > > > > > > and the broker is only considered during
> > > > fetching,
> > > > > > but
> > > > > > > > not
> > > > > > > > > >> during
> > > > > > > > > >> > > > > > assigning
> > > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once
> the
> > > > > > > assignment
> > > > > > > > is
> > > > > > > > > >> made,
> > > > > > > > > >> > > > there
> > > > > > > > > >> > > > > > is
> > > > > > > > > >> > > > > > > > no opportunity for read affinity because
> no
> > > > replicas
> > > > > > > of
> > > > > > > > > >> assigned
> > > > > > > > > >> > > > > > partitions
> > > > > > > > > >> > > > > > > > are close to the member. I am wondering
> if we
> > > > should
> > > > > > > use
> > > > > > > > > >> this
> > > > > > > > > >> > > > > > opportunity
> > > > > > > > > >> > > > > > > > to address this by including rack in
> > > > GroupMember.
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 32. On the metric side, often, it's
> useful to
> > > > know
> > > > > > how
> > > > > > > > busy
> > > > > > > > > >> a group
> > > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > > model,
> > > > it
> > > > > > > seems
> > > > > > > > > >> that we
> > > > > > > > > >> > > > could
> > > > > > > > > >> > > > > > add
> > > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> time
> > > > the
> > > > > > > event
> > > > > > > > > >> loop is
> > > > > > > > > >> > > > doing
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > actual work.
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > > failover
> > > > > > > > > >> handling? For
> > > > > > > > > >> > > > > > example,
> > > > > > > > > >> > > > > > > > does it need to trigger the check if any
> group
> > > > with
> > > > > > > the
> > > > > > > > > >> wildcard
> > > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > > document
> > > > > > > what
> > > > > > > > > >> the epoch
> > > > > > > > > >> > > > > > field
> > > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > > >> > > > > > reflect
> > > > > > > > > >> > > > > > > > the latest group epoch? What about the
> one in
> > > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure
> that
> > > the
> > > > > > > > following
> > > > > > > > > >> > > > invariants
> > > > > > > > > >> > > > > > are
> > > > > > > > > >> > > > > > > > met: ... All members exists." It's
> possible
> > > for
> > > > a
> > > > > > > member
> > > > > > > > > >> not to
> > > > > > > > > >> > > > get any
> > > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a
> member
> > > > epoch
> > > > > > > > equals to
> > > > > > > > > >> 0":
> > > > > > > > > >> > > > When
> > > > > > > > > >> > > > > > would
> > > > > > > > > >> > > > > > > > a consumer rejoin and what member id
> would be
> > > > used?
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > > ability
> > > > to
> > > > > > > > trigger a
> > > > > > > > > >> > > > > > reassignment
> > > > > > > > > >> > > > > > > > by either providing a non-zero reason or
> by
> > > > updating
> > > > > > > the
> > > > > > > > > >> assignor
> > > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be
> conflicting
> > > > with
> > > > > > the
> > > > > > > > > >> deprecation
> > > > > > > > > >> > > > of
> > > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 38. The reassignment examples are nice.
> But
> > > the
> > > > > > > section
> > > > > > > > > >> seems to
> > > > > > > > > >> > > > have
> > > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch
> 2, B
> > > > > > > > immediately
> > > > > > > > > >> gets into
> > > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > > incorrect.
> > > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch
> 3, C
> > > > seems
> > > > > > to
> > > > > > > > get
> > > > > > > > > >> into
> > > > > > > > > >> > > > > > epoch=3,
> > > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C
> still
> > > > has A -
> > > > > > > > > >> epoch=2,
> > > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > > support
> > > > the
> > > > > > > > upgrade
> > > > > > > > > >> from
> > > > > > > > > >> > > > any
> > > > > > > > > >> > > > > > old
> > > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > Jun
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David
> Jacot
> > > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > >> > > > > > > > wrote:
> > > > > > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > The KIP states that we will
> re-implement the
> > > > > > > > coordinator
> > > > > > > > > >> in
> > > > > > > > > >> > > > Java. I
> > > > > > > > > >> > > > > > > > > discussed this offline with a few folks
> and
> > > > folks
> > > > > > > are
> > > > > > > > > >> concerned
> > > > > > > > > >> > > > that
> > > > > > > > > >> > > > > > > > > we could introduce many regressions in
> the
> > > old
> > > > > > > > protocol
> > > > > > > > > >> if we do
> > > > > > > > > >> > > > so.
> > > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > > statement
> > > > > > from
> > > > > > > > the
> > > > > > > > > >> KIP. It
> > > > > > > > > >> > > > is an
> > > > > > > > > >> > > > > > > > > implementation detail after all so it
> does
> > > not
> > > > > > have
> > > > > > > > to be
> > > > > > > > > >> > > > decided at
> > > > > > > > > >> > > > > > > > > this stage. We will likely start by
> trying
> > > to
> > > > > > > > refactor the
> > > > > > > > > >> > > > current
> > > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > > >> > > > > > > > > David
> > > > > > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David
> Jacot
> > > <
> > > > > > > > > >> djacot@confluent.io
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > wrote:
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > > "Empty,
> > > > > > > > assigning,
> > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> States
> > > > > > > section,
> > > > > > > > > >> right?
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > This sentence does not refer to those
> > > group
> > > > > > states
> > > > > > > > but
> > > > > > > > > >> rather
> > > > > > > > > >> > > > to a
> > > > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > > > refers to
> > > > > > > the
> > > > > > > > > >> entire
> > > > > > > > > >> > > > state of
> > > > > > > > > >> > > > > > > > > > group coordinator which is replicated
> via
> > > > the
> > > > > > log
> > > > > > > > > >> layer. I will
> > > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > machine
> > > > is
> > > > > > > > modelled
> > > > > > > > > >> as an
> > > > > > > > > >> > > > event
> > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > The idea is to follow a model similar
> to
> > > > the new
> > > > > > > > quorum
> > > > > > > > > >> > > > > > controller. We
> > > > > > > > > >> > > > > > > > > > will have N threads to process events.
> > > Each
> > > > > > > > > >> __consumer_offsets
> > > > > > > > > >> > > > > > > > > > partition is assigned to a unique
> thread
> > > > and all
> > > > > > > the
> > > > > > > > > >> events
> > > > > > > > > >> > > > (e.g.
> > > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> processed
> > > by
> > > > this
> > > > > > > > > >> thread. This
> > > > > > > > > >> > > > > > simplify
> > > > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > > > simulation
> > > > > > > > testing
> > > > > > > > > >> for the
> > > > > > > > > >> > > > > > group
> > > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> per
> > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> group"
> > > > owned
> > > > > > > by
> > > > > > > > a
> > > > > > > > > >> group
> > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > >> > > > > > > > > For
> > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> owns 2
> > > > > > > consumer
> > > > > > > > > >> groups, and
> > > > > > > > > >> > > > > > both
> > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> have 1
> > > > state
> > > > > > > > machine
> > > > > > > > > >> for it,
> > > > > > > > > >> > > > or
> > > > > > > > > >> > > > > > 2?
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> there, I
> > > > > > think.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > "*group.coordinator.threads"
> > > > > > *is
> > > > > > > > the
> > > > > > > > > >> number
> > > > > > > > > >> > > > of
> > > > > > > > > >> > > > > > threads
> > > > > > > > > >> > > > > > > > > used
> > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > wondering
> > > > > > if
> > > > > > > > the
> > > > > > > > > >> purpose
> > > > > > > > > >> > > > of
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > consumer
> > > > > > group
> > > > > > > > (or
> > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> computation,
> > > > why
> > > > > > > > should we
> > > > > > > > > >> need
> > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an
> ability to
> > > > shard
> > > > > > > the
> > > > > > > > > >> > > > processing as
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> why
> > > > does
> > > > > > the
> > > > > > > > > >> default
> > > > > > > > > >> > > > session
> > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> max(60s)? I
> > > > > > thought
> > > > > > > > the
> > > > > > > > > >> min/max
> > > > > > > > > >> > > > > > session
> > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> bound
> > > of
> > > > it,
> > > > > > > no?
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> int
> > > > 30s The
> > > > > > > > > >> timeout to
> > > > > > > > > >> > > > detect
> > > > > > > > > >> > > > > > client
> > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> group
> > > > > > protocol.
> > > > > > > > > >> > > > > > > > > > >
> group.consumer.min.session.timeout.ms
> > > > int 45s
> > > > > > > The
> > > > > > > > > >> minimum
> > > > > > > > > >> > > > > > session
> > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > >> > > > > > > > > > >
> group.consumer.max.session.timeout.ms
> > > > int 60s
> > > > > > > The
> > > > > > > > > >> maximum
> > > > > > > > > >> > > > > > session
> > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > > > session
> > > > > > > > timeout
> > > > > > > > > >> should
> > > > > > > > > >> > > > be 45s
> > > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> assignor
> > > are
> > > > > > > [range,
> > > > > > > > > >> uniform],
> > > > > > > > > >> > > > > > which means
> > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> I'd
> > > > like to
> > > > > > > > know
> > > > > > > > > >> why not
> > > > > > > > > >> > > > > > uniform
> > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > uniform
> > > > > > > assignor
> > > > > > > > > >> (former
> > > > > > > > > >> > > > sticky
> > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > > other
> > > > > > reason
> > > > > > > > we
> > > > > > > > > >> choose
> > > > > > > > > >> > > > range
> > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > > uniform
> > > > > > The
> > > > > > > > > >> server side
> > > > > > > > > >> > > > > > assignors.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > The order on the server side has no
> > > > influence
> > > > > > > > because
> > > > > > > > > >> the
> > > > > > > > > >> > > > client
> > > > > > > > > >> > > > > > must
> > > > > > > > > >> > > > > > > > > > chose the selector that he wants to
> use.
> > > > There
> > > > > > is
> > > > > > > no
> > > > > > > > > >> default
> > > > > > > > > >> > > > in the
> > > > > > > > > >> > > > > > > > > > current proposal. If the assignor is
> not
> > > > > > specified
> > > > > > > > by
> > > > > > > > > >> the
> > > > > > > > > >> > > > client,
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > > request is rejected. The default
> client
> > > > value
> > > > > > for
> > > > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> > > though.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > Thanks for your very good comments,
> Luke.
> > > I
> > > > hope
> > > > > > > > that my
> > > > > > > > > >> > > > answers
> > > > > > > > > >> > > > > > help
> > > > > > > > > >> > > > > > > > > > to clarify things. I will update the
> KIP
> > > as
> > > > well
> > > > > > > > based
> > > > > > > > > >> on your
> > > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > > >> > > > > > > > > > David
> > > > > > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke
> Chen
> > > <
> > > > > > > > > >> showuon@gmail.com>
> > > > > > > > > >> > > > > > wrote:
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > > > mentioned:
> > > > > > > > > >> > > > > > > > > > > > The new group coordinator will
> have a
> > > > state
> > > > > > > > machine
> > > > > > > > > >> per
> > > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions,
> where
> > > > each
> > > > > > > state
> > > > > > > > > >> machine is
> > > > > > > > > >> > > > > > modelled
> > > > > > > > > >> > > > > > > > > as an
> > > > > > > > > >> > > > > > > > > > > event loop. Those state machines
> will be
> > > > > > > executed
> > > > > > > > in
> > > > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > > "Empty,
> > > > > > > > assigning,
> > > > > > > > > >> > > > > > reconciling,
> > > > > > > > > >> > > > > > > > > stable,
> > > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> States
> > > > > > > section,
> > > > > > > > > >> right?
> > > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > > machine
> > > > is
> > > > > > > > modelled
> > > > > > > > > >> as an
> > > > > > > > > >> > > > event
> > > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> per
> > > > > > > > > >> *__consumer_offsets*
> > > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> group"
> > > > owned
> > > > > > > by
> > > > > > > > a
> > > > > > > > > >> group
> > > > > > > > > >> > > > > > coordinator?
> > > > > > > > > >> > > > > > > > > For
> > > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> owns 2
> > > > > > > consumer
> > > > > > > > > >> groups, and
> > > > > > > > > >> > > > > > both
> > > > > > > > > >> > > > > > > > > exist in
> > > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we
> have 1
> > > > state
> > > > > > > > machine
> > > > > > > > > >> for it,
> > > > > > > > > >> > > > or
> > > > > > > > > >> > > > > > 2?
> > > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > > "*group.coordinator.threads"
> > > > > > *is
> > > > > > > > the
> > > > > > > > > >> number
> > > > > > > > > >> > > > of
> > > > > > > > > >> > > > > > threads
> > > > > > > > > >> > > > > > > > > used
> > > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > > wondering
> > > > > > if
> > > > > > > > the
> > > > > > > > > >> purpose
> > > > > > > > > >> > > > of
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > threads
> > > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > > consumer
> > > > > > group
> > > > > > > > (or
> > > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> computation,
> > > > why
> > > > > > > > should we
> > > > > > > > > >> need
> > > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > > > configs:
> > > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout,
> why
> > > > does
> > > > > > the
> > > > > > > > > >> default
> > > > > > > > > >> > > > session
> > > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> max(60s)? I
> > > > > > thought
> > > > > > > > the
> > > > > > > > > >> min/max
> > > > > > > > > >> > > > > > session
> > > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> bound
> > > of
> > > > it,
> > > > > > > no?
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms
> int
> > > > 30s The
> > > > > > > > > >> timeout to
> > > > > > > > > >> > > > detect
> > > > > > > > > >> > > > > > client
> > > > > > > > > >> > > > > > > > > > > failures when using the consumer
> group
> > > > > > protocol.
> > > > > > > > > >> > > > > > > > > > >
> group.consumer.min.session.timeout.ms
> > > > int 45s
> > > > > > > The
> > > > > > > > > >> minimum
> > > > > > > > > >> > > > > > session
> > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > >> > > > > > > > > > >
> group.consumer.max.session.timeout.ms
> > > > int 60s
> > > > > > > The
> > > > > > > > > >> maximum
> > > > > > > > > >> > > > > > session
> > > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > 2.2. The default server side
> assignor
> > > are
> > > > > > > [range,
> > > > > > > > > >> uniform],
> > > > > > > > > >> > > > > > which means
> > > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor.
> I'd
> > > > like to
> > > > > > > > know
> > > > > > > > > >> why not
> > > > > > > > > >> > > > > > uniform
> > > > > > > > > >> > > > > > > > > one? I
> > > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > > uniform
> > > > > > > assignor
> > > > > > > > > >> (former
> > > > > > > > > >> > > > sticky
> > > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > > other
> > > > > > reason
> > > > > > > > we
> > > > > > > > > >> choose
> > > > > > > > > >> > > > range
> > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > > uniform
> > > > > > The
> > > > > > > > > >> server side
> > > > > > > > > >> > > > > > assignors.
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke
> > > Chen
> > > > <
> > > > > > > > > >> showuon@gmail.com
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > wrote:
> > > > > > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> > > Connect
> > > > > > > > > >> integrating with
> > > > > > > > > >> > > > > > KIP-848,
> > > > > > > > > >> > > > > > > > > but I
> > > > > > > > > >> > > > > > > > > > > > think we should have a separate
> > > > discussion
> > > > > > > > thread
> > > > > > > > > >> for the
> > > > > > > > > >> > > > Kafka
> > > > > > > > > >> > > > > > > > > Connect
> > > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect
> With
> > > New
> > > > > > > Consumer
> > > > > > > > > >> Rebalance
> > > > > > > > > >> > > > > > Protocol
> > > > > > > > > >> > > > > > > > > [1],
> > > > > > > > > >> > > > > > > > > > > > and let this discussion thread
> focus
> > > on
> > > > > > > consumer
> > > > > > > > > >> rebalance
> > > > > > > > > >> > > > > > protocol,
> > > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > >
> > > > > > > > > >>
> > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM
> Sagar
> > > <
> > > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > > > feedback.
> > > > > > > > Looks
> > > > > > > > > >> like
> > > > > > > > > >> > > > there's
> > > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect.
> I
> > > > would
> > > > > > > > revisit
> > > > > > > > > >> the doc
> > > > > > > > > >> > > > and
> > > > > > > > > >> > > > > > see
> > > > > > > > > >> > > > > > > > > what
> > > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM
> David
> > > > Jacot
> > > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > > > document.
> > > > > > > > That's
> > > > > > > > > >> really
> > > > > > > > > >> > > > > > helpful. I
> > > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that
> both
> > > > Connect
> > > > > > > > and the
> > > > > > > > > >> > > > Consumer
> > > > > > > > > >> > > > > > could
> > > > > > > > > >> > > > > > > > > share
> > > > > > > > > >> > > > > > > > > > > >> > the same underlying "engine".
> The
> > > > main
> > > > > > > > > >> difference is
> > > > > > > > > >> > > > that
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> members
> > > > > > whereas
> > > > > > > > > >> Connect
> > > > > > > > > >> > > > assigns
> > > > > > > > > >> > > > > > tasks
> > > > > > > > > >> > > > > > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > > > forward:
> > > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed
> APIs
> > > to
> > > > > > > support
> > > > > > > > > >> different
> > > > > > > > > >> > > > > > resource
> > > > > > > > > >> > > > > > > > > types
> > > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks,
> etc.); or
> > > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs
> for
> > > > Connect.
> > > > > > > The
> > > > > > > > > >> dedicated
> > > > > > > > > >> > > > APIs
> > > > > > > > > >> > > > > > > > > would be
> > > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > > different on
> > > > > > > the
> > > > > > > > > >> > > > > > content/resources and
> > > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> engine
> > > > on the
> > > > > > > > > >> coordinator
> > > > > > > > > >> > > > side.
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> > > because
> > > > I am
> > > > > > > > not a
> > > > > > > > > >> fan of
> > > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > > >> > > > > > > > > > > >> > APIs to serve different
> purposes.
> > > > That
> > > > > > > being
> > > > > > > > > >> said, I am
> > > > > > > > > >> > > > not
> > > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant
> way to
> > > > do
> > > > > > it.
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > > > discuss
> > > > > > it
> > > > > > > > here
> > > > > > > > > >> for now
> > > > > > > > > >> > > > in
> > > > > > > > > >> > > > > > order
> > > > > > > > > >> > > > > > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> compatible
> > > > with
> > > > > > > what
> > > > > > > > we
> > > > > > > > > >> will do
> > > > > > > > > >> > > > for
> > > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM
> > > David
> > > > > > Jacot
> > > > > > > <
> > > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I
> will
> > > go
> > > > > > > through
> > > > > > > > and
> > > > > > > > > >> address
> > > > > > > > > >> > > > > > your
> > > > > > > > > >> > > > > > > > > comments
> > > > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks
> for
> > > your
> > > > > > > > feedback.
> > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05
> PM
> > > > Gregory
> > > > > > > > Harris
> > > > > > > > > >> <
> > > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > > > wonderful to
> > > > > > > see
> > > > > > > > > >> > > > cooperative
> > > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> Alternatives'
> > > > > > section
> > > > > > > > > >> describes how
> > > > > > > > > >> > > > > > member
> > > > > > > > > >> > > > > > > > > epoch
> > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > >> > > > > > > > > > > >> > > > advance in step with the
> group
> > > > epoch
> > > > > > > and
> > > > > > > > > >> assignment
> > > > > > > > > >> > > > > > epoch
> > > > > > > > > >> > > > > > > > > values. I
> > > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea
> for
> > > the
> > > > > > > reasons
> > > > > > > > > >> described
> > > > > > > > > >> > > > in
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > > > assigning
> > > > > > > > > >> partitions to a
> > > > > > > > > >> > > > > > worker,
> > > > > > > > > >> > > > > > > > > what
> > > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > > > assignment
> > > > > > > > use?
> > > > > > > > > >> Are
> > > > > > > > > >> > > > member
> > > > > > > > > >> > > > > > epochs
> > > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > > > correspond
> > > > > > to
> > > > > > > > > >> multiple
> > > > > > > > > >> > > > > > different
> > > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's
> 'Reason'
> > > > field
> > > > > > > > opaque
> > > > > > > > > >> to the
> > > > > > > > > >> > > > group
> > > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> client-side
> > > > > > assignor
> > > > > > > > > >> > > > implementations
> > > > > > > > > >> > > > > > > > > interact
> > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is
> its
> > > > common
> > > > > > > > meaning
> > > > > > > > > >> agreed
> > > > > > > > > >> > > > > > upon? If
> > > > > > > > > >> > > > > > > > > so, what
> > > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct
> Reason
> > > > field
> > > > > > over
> > > > > > > > > >> including
> > > > > > > > > >> > > > such
> > > > > > > > > >> > > > > > > > > functionality
> > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is
> included in
> > > > the
> > > > > > > KIP:
> > > > > > > > > >> "Thanks to
> > > > > > > > > >> > > > > > this, the
> > > > > > > > > >> > > > > > > > > input
> > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > > entirely
> > > > > > driven
> > > > > > > > by
> > > > > > > > > >> the group
> > > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > > responsible for
> > > > > > > > > >> maintaining
> > > > > > > > > >> > > > any
> > > > > > > > > >> > > > > > state
> > > > > > > > > >> > > > > > > > > besides
> > > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does
> this
> > > > mean
> > > > > > > > that the
> > > > > > > > > >> > > > > > client-side
> > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > > non-Metadata
> > > > > > > state
> > > > > > > > > >> (such as
> > > > > > > > > >> > > > > > partition
> > > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > > > topics,
> > > > > > > etc),
> > > > > > > > or
> > > > > > > > > >> that
> > > > > > > > > >> > > > > > additional
> > > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > > separate
> > > > > > > classes
> > > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > > >> > > > > >
> > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > > >> > > >
> org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > > >> > > > > > that
> > > > > > > > > >> > > > > > > > > seem to
> > > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is
> it
> > > > possible
> > > > > > > for
> > > > > > > > > >> these two
> > > > > > > > > >> > > > > > > > > implementations
> > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve
> to
> > > > promote
> > > > > > > > feature
> > > > > > > > > >> parity
> > > > > > > > > >> > > > of
> > > > > > > > > >> > > > > > > > > server-side
> > > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and
> > > would
> > > > also
> > > > > > > > > >> facilitate
> > > > > > > > > >> > > > > > operational
> > > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > > example,
> > > > if a
> > > > > > > > > >> server-side
> > > > > > > > > >> > > > > > assignor
> > > > > > > > > >> > > > > > > > > has some
> > > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > > deploying
> > > > > > > the
> > > > > > > > > >> patched
> > > > > > > > > >> > > > > > assignor to
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer
> > > group
> > > > to a
> > > > > > > > > >> client-side
> > > > > > > > > >> > > > > > assignor
> > > > > > > > > >> > > > > > > > > may be
> > > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > > >> > > > > > > > > > > >> > > > and less risky than
> patching
> > > all
> > > > of
> > > > > > the
> > > > > > > > > >> brokers.
> > > > > > > > > >> > > > With
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > currently
> > > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > > > non-trivial
> > > > > > > > > >> > > > reimplementation
> > > > > > > > > >> > > > > > would
> > > > > > > > > >> > > > > > > > > have
> > > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two
> APIs
> > > > have
> > > > > > > > diverged
> > > > > > > > > >> > > > > > significantly,
> > > > > > > > > >> > > > > > > > > then it
> > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > > reimplementation
> > > > > > would
> > > > > > > > not be
> > > > > > > > > >> > > > possible.
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at
> 8:39 AM
> > > > Sagar
> > > > > > <
> > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence
> page
> > > to
> > > > > > > discuss
> > > > > > > > how
> > > > > > > > > >> Connect
> > > > > > > > > >> > > > > > would
> > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > > >> > > > > > > > > > > >> > > > > based on the new
> rebalance
> > > > > > protocol.
> > > > > > > > > >> Here's the
> > > > > > > > > >> > > > page:
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > >
> > > > > > > > > >>
> > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish
> and
> > > I
> > > > have
> > > > > > > > tried
> > > > > > > > > >> to keep
> > > > > > > > > >> > > > a
> > > > > > > > > >> > > > > > format
> > > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know
> what you
> > > > > > think.
> > > > > > > > Also,
> > > > > > > > > >> do you
> > > > > > > > > >> > > > > > think this
> > > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > > discussion
> > > > > > thread
> > > > > > > > or
> > > > > > > > > >> is this
> > > > > > > > > >> > > > one
> > > > > > > > > >> > > > > > fine?
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at
> 7:37
> > > AM
> > > > > > > Sagar <
> > > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for
> the
> > > > doc on
> > > > > > > > Kafka
> > > > > > > > > >> Streams.
> > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > >> > > > > > > > > would do
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up
> with
> > > > such a
> > > > > > > > > >> document.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at
> > > 4:47
> > > > AM
> > > > > > > > Guozhang
> > > > > > > > > >> Wang <
> > > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if
> you
> > > > could
> > > > > > > come
> > > > > > > > > >> back with
> > > > > > > > > >> > > > some
> > > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect
> side
> > > > > > > > integration
> > > > > > > > > >> with
> > > > > > > > > >> > > > the new
> > > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on
> the
> > > > new
> > > > > > > > "protocol
> > > > > > > > > >> type"
> > > > > > > > > >> > > > we
> > > > > > > > > >> > > > > > did not
> > > > > > > > > >> > > > > > > > > yet
> > > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > > implementations.
> > > > > > > > For
> > > > > > > > > >> Streams
> > > > > > > > > >> > > > > > here's a
> > > > > > > > > >> > > > > > > > > draft of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > >
> > > > > > > > > >>
> > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your
> analysis
> > > > on
> > > > > > > > Connect.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022
> at
> > > > 10:48 PM
> > > > > > > > Sagar <
> > > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > > > response.
> > > > > > The
> > > > > > > > > >> reason I
> > > > > > > > > >> > > > thought
> > > > > > > > > >> > > > > > > > > connect can
> > > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is
> that
> > > > even
> > > > > > > > today the
> > > > > > > > > >> > > > connect
> > > > > > > > > >> > > > > > uses a
> > > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > > AbstractCoordinator to
> > > > > > > > > >> empower
> > > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors.
> The
> > > > > > > > > >> WorkerCoordinator sets
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> metadata()
> > > > method
> > > > > > > by
> > > > > > > > > >> plumbing
> > > > > > > > > >> > > > into
> > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> JoinGroupRequestProtocol.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes
> to
> > > > support
> > > > > > > > > >> connect would
> > > > > > > > > >> > > > be
> > > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in
> streams
> > > > mainly
> > > > > > > > because
> > > > > > > > > >> of the
> > > > > > > > > >> > > > > > Client
> > > > > > > > > >> > > > > > > > > side
> > > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > > > > implementation
> > > > > > > > > >> level, we
> > > > > > > > > >> > > > > > might
> > > > > > > > > >> > > > > > > > > need to
> > > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto
> this
> > > > new
> > > > > > > > > >> assignment
> > > > > > > > > >> > > > protocol
> > > > > > > > > >> > > > > > like
> > > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > > request/response and
> > > > > > > > > >> SyncGroup and
> > > > > > > > > >> > > > > > using
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on
> similar
> > > > lines
> > > > > > > to
> > > > > > > > > >> streams (or
> > > > > > > > > >> > > > > > there
> > > > > > > > > >> > > > > > > > > might be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to
> perform a
> > > > > > detailed
> > > > > > > > > >> analysis of
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > same
> > > > > > > > > >> > > > > > > > > and we
> > > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> > > > thread for
> > > > > > > > that
> > > > > > > > > >> as that
> > > > > > > > > >> > > > would
> > > > > > > > > >> > > > > > > > > derail this
> > > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know
> if
> > > > that
> > > > > > > sounds
> > > > > > > > > >> good to
> > > > > > > > > >> > > > you.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15,
> 2022 at
> > > > 5:47
> > > > > > PM
> > > > > > > > David
> > > > > > > > > >> Jacot
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > > > comments.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That
> refers to
> > > > > > > > > >> `Assignment#error`.
> > > > > > > > > >> > > > > > Sure, I
> > > > > > > > > >> > > > > > > > > can
> > > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > > > > transition C
> > > > > > > > from
> > > > > > > > > >> his
> > > > > > > > > >> > > > current
> > > > > > > > > >> > > > > > > > > assignment
> > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment
> when
> > > > he
> > > > > > can
> > > > > > > > move
> > > > > > > > > >> to
> > > > > > > > > >> > > > epoch 3.
> > > > > > > > > >> > > > > > When
> > > > > > > > > >> > > > > > > > > that
> > > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> is
> > > > updated
> > > > > > > and
> > > > > > > > > >> persisted
> > > > > > > > > >> > > > > > with all
> > > > > > > > > >> > > > > > > > > its
> > > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if
> > > they
> > > > are
> > > > > > > not
> > > > > > > > all
> > > > > > > > > >> revoked
> > > > > > > > > >> > > > > > yet. In
> > > > > > > > > >> > > > > > > > > other
> > > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > > > becomes
> > > > > > the
> > > > > > > > target
> > > > > > > > > >> > > > > > assignment.
> > > > > > > > > >> > > > > > > > > This is
> > > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to
> avoid
> > > > having
> > > > > > > to
> > > > > > > > > >> write all
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > changes to
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are
> based on
> > > > the
> > > > > > > > persisted
> > > > > > > > > >> state
> > > > > > > > > >> > > > so I
> > > > > > > > > >> > > > > > > > > understand
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me
> see
> > > > if I
> > > > > > can
> > > > > > > > > >> improve
> > > > > > > > > >> > > > this in
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding
> Connect,
> > > it
> > > > > > could
> > > > > > > > > >> reuse the
> > > > > > > > > >> > > > > > protocol
> > > > > > > > > >> > > > > > > > > with a
> > > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it
> fits in
> > > > the
> > > > > > > > > >> protocol. The
> > > > > > > > > >> > > > > > assignment
> > > > > > > > > >> > > > > > > > > is
> > > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> topicid-partitions +
> > > > > > metadata,
> > > > > > > > could
> > > > > > > > > >> > > > Connect
> > > > > > > > > >> > > > > > fit
> > > > > > > > > >> > > > > > > > > into this?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15,
> 2022
> > > at
> > > > 1:55
> > > > > > > PM
> > > > > > > > > >> Sagar <
> > > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the
> KIP.
> > > I
> > > > just
> > > > > > > had
> > > > > > > > > >> minor
> > > > > > > > > >> > > > > > observations:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the
> Assignment
> > > > Error
> > > > > > > > > >> section in
> > > > > > > > > >> > > > Client
> > > > > > > > > >> > > > > > Side
> > > > > > > > > >> > > > > > > > > mode
> > > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned
> => `In
> > > > this
> > > > > > > > case,
> > > > > > > > > >> the
> > > > > > > > > >> > > > client
> > > > > > > > > >> > > > > > side
> > > > > > > > > >> > > > > > > > > assignor
> > > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the
> group
> > > > > > > > coordinator`.
> > > > > > > > > >> In this
> > > > > > > > > >> > > > > > case are
> > > > > > > > > >> > > > > > > > > you
> > > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor
> returning
> > > an
> > > > > > > > > >> AssignmentError
> > > > > > > > > >> > > > that's
> > > > > > > > > >> > > > > > > > > listed down
> > > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you
> think
> > > > it
> > > > > > > would
> > > > > > > > > >> make sense
> > > > > > > > > >> > > > to
> > > > > > > > > >> > > > > > > > > mention this
> > > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case
> > > Studies
> > > > > > > > section, I
> > > > > > > > > >> have a
> > > > > > > > > >> > > > > > slight
> > > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the
> > > same.
> > > > > > > Consider
> > > > > > > > > >> this step:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B
> heartbeats,
> > > the
> > > > > > group
> > > > > > > > > >> coordinator
> > > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> partitions
> > > to
> > > > > > > > revoke. It
> > > > > > > > > >> > > > persists
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > change and
> > > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group
> (epoch=3)
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > > Assignment
> > > > > > > > (epoch=3)
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > > partitions=[foo-0]
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > > partitions=[foo-2]
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > > partitions=[foo-1]
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > > Assignment
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> epoch=2,
> > > > > > > > > >> partitions=[foo-0,
> > > > > > > > > >> > > > > > foo-1]
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> epoch=3,
> > > > > > > > > >> partitions=[foo-2]
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> epoch=3,
> > > > > > > > > >> partitions=[foo-1]
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C
> heartbeats,
> > > it
> > > > > > > > transitions
> > > > > > > > > >> to
> > > > > > > > > >> > > > epoch 3
> > > > > > > > > >> > > > > > but
> > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's
> mentioned
> > > > that
> > > > > > > > member C
> > > > > > > > > >> can't
> > > > > > > > > >> > > > get
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > > description
> > > > > > > > above,
> > > > > > > > > >> it seems
> > > > > > > > > >> > > > it
> > > > > > > > > >> > > > > > > > > already has
> > > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better
> to
> > > > remove
> > > > > > it
> > > > > > > > and
> > > > > > > > > >> > > > populate it
> > > > > > > > > >> > > > > > only
> > > > > > > > > >> > > > > > > > > when it
> > > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a
> lot
> > > of
> > > > > > other
> > > > > > > > > >> places, so
> > > > > > > > > >> > > > have
> > > > > > > > > >> > > > > > I
> > > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding
> connect ,
> > > it
> > > > > > might
> > > > > > > > be
> > > > > > > > > >> out of
> > > > > > > > > >> > > > scope
> > > > > > > > > >> > > > > > of
> > > > > > > > > >> > > > > > > > > this
> > > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I
> understood it
> > > > would
> > > > > > > > > >> probably be
> > > > > > > > > >> > > > > > running in
> > > > > > > > > >> > > > > > > > > client
> > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> > > > rebalance
> > > > > > > > > >> protocol as it
> > > > > > > > > >> > > > has
> > > > > > > > > >> > > > > > its
> > > > > > > > > >> > > > > > > > > own
> > > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > > IncrementalCooperative).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15,
> 2022
> > > > at
> > > > > > 5:00
> > > > > > > > PM
> > > > > > > > > >> David
> > > > > > > > > >> > > > Jacot
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks
> Hector! Our
> > > > goal
> > > > > > is
> > > > > > > > to
> > > > > > > > > >> move
> > > > > > > > > >> > > > forward
> > > > > > > > > >> > > > > > with
> > > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of
> relying
> > > > on
> > > > > > one
> > > > > > > > > >> generic API.
> > > > > > > > > >> > > > For
> > > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same
> pattern
> > > > and
> > > > > > > > > >> reuse/share the
> > > > > > > > > >> > > > core
> > > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side.
> For
> > > the
> > > > > > > schema
> > > > > > > > > >> registry, I
> > > > > > > > > >> > > > > > think
> > > > > > > > > >> > > > > > > > > that we
> > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a
> tailored
> > > > API to
> > > > > > > do
> > > > > > > > > >> simple
> > > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul
> 15,
> > > > 2022 at
> > > > > > > > 10:22
> > > > > > > > > >> AM Ismael
> > > > > > > > > >> > > > > > Juma <
> > > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> > > > comments:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > > > > >> java.util.regex.Pattern
> > > > > > > > > >> > > > vs
> > > > > > > > > >> > > > > > > > > > > >> > > > > >>
> com.google.re2j.Pattern,
> > > we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > > > > differences
> > > > > > > > in
> > > > > > > > > >> more
> > > > > > > > > >> > > > detail
> > > > > > > > > >> > > > > > before
> > > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said,
> if
> > > > people
> > > > > > > pass
> > > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics
> to be
> > > > > > honored.
> > > > > > > > If
> > > > > > > > > >> we are
> > > > > > > > > >> > > > doing
> > > > > > > > > >> > > > > > > > > something
> > > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should
> consider
> > > > adding
> > > > > > > an
> > > > > > > > > >> overload
> > > > > > > > > >> > > > with
> > > > > > > > > >> > > > > > our own
> > > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd
> want
> > > to
> > > > > > > expose
> > > > > > > > > >> re2j's at
> > > > > > > > > >> > > > this
> > > > > > > > > >> > > > > > > > > point).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding
> > > topic
> > > > > > ids,
> > > > > > > > any
> > > > > > > > > >> major new
> > > > > > > > > >> > > > > > protocol
> > > > > > > > > >> > > > > > > > > should
> > > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and
> should
> > > > handle
> > > > > > the
> > > > > > > > topic
> > > > > > > > > >> > > > > > recreation case
> > > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we
> need to
> > > > > > handle.
> > > > > > > I
> > > > > > > > > >> agree with
> > > > > > > > > >> > > > > > David
> > > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the
> relevant
> > > > > > > protocols
> > > > > > > > > >> that don't
> > > > > > > > > >> > > > > > have it
> > > > > > > > > >> > > > > > > > > yet and
> > > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> > > > internals
> > > > > > > > versus
> > > > > > > > > >> adding
> > > > > > > > > >> > > > new
> > > > > > > > > >> > > > > > APIs
> > > > > > > > > >> > > > > > > > > to the
> > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> > > > adding
> > > > > > new
> > > > > > > > APIs
> > > > > > > > > >> is
> > > > > > > > > >> > > > required
> > > > > > > > > >> > > > > > for
> > > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am
> still
> > > not
> > > > sure
> > > > > > > > about
> > > > > > > > > >> the
> > > > > > > > > >> > > > > > coordinator
> > > > > > > > > >> > > > > > > > > storing
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> > > > configs
> > > > > > to
> > > > > > > be
> > > > > > > > > >> > > > centralized
> > > > > > > > > >> > > > > > in the
> > > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > > (auditability,
> > > > > > > > > >> visibility,
> > > > > > > > > >> > > > > > consistency,
> > > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > > > automatically
> > > > > > > > > >> deleting
> > > > > > > > > >> > > > > > configs in a
> > > > > > > > > >> > > > > > > > > way
> > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A
> > > good
> > > > > > > property
> > > > > > > > > >> for modern
> > > > > > > > > >> > > > > > systems
> > > > > > > > > >> > > > > > > > > is to
> > > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> unrecoverable
> > > data
> > > > > > loss
> > > > > > > > > >> scenarios.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul
> 13,
> > > > 2022
> > > > > > at
> > > > > > > > 3:47
> > > > > > > > > >> PM David
> > > > > > > > > >> > > > > > Jacot
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > > Guozhang. My
> > > > > > > > answers
> > > > > > > > > >> are
> > > > > > > > > >> > > > below:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > migration
> > > > > > > path,
> > > > > > > > > >> especially
> > > > > > > > > >> > > > > > the last
> > > > > > > > > >> > > > > > > > > step
> > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> enable
> > > the
> > > > new
> > > > > > > > > >> protocol, in
> > > > > > > > > >> > > > which
> > > > > > > > > >> > > > > > we
> > > > > > > > > >> > > > > > > > > would
> > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> protocols /
> > > > rpcs
> > > > > > and
> > > > > > > > old
> > > > > > > > > >> > > > protocols /
> > > > > > > > > >> > > > > > rpcs
> > > > > > > > > >> > > > > > > > > are
> > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> How
> > > the
> > > > > > > > > >> coordinator could
> > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> protocol
> > > > is
> > > > > > > > something
> > > > > > > > > >> we
> > > > > > > > > >> > > > need to
> > > > > > > > > >> > > > > > > > > present
> > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I
> just
> > > > > > > published
> > > > > > > > a
> > > > > > > > > >> new
> > > > > > > > > >> > > > version
> > > > > > > > > >> > > > > > of KIP
> > > > > > > > > >> > > > > > > > > which
> > > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details
> about
> > > > this.
> > > > > > > See
> > > > > > > > the
> > > > > > > > > >> > > > "Supporting
> > > > > > > > > >> > > > > > > > > Online
> > > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > > > "Compatibility,
> > > > > > > > > >> > > > Deprecation,
> > > > > > > > > >> > > > > > and
> > > > > > > > > >> > > > > > > > > Migration
> > > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to
> > > think
> > > > > > > through
> > > > > > > > a
> > > > > > > > > >> few cases
> > > > > > > > > >> > > > > > now but
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> > > > should be
> > > > > > > > > >> understandable.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> usage
> > > > of
> > > > > > > topic
> > > > > > > > > >> ids. So
> > > > > > > > > >> > > > far as
> > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> RPCs
> > > > and
> > > > > > > admin
> > > > > > > > > >> client,
> > > > > > > > > >> > > > but
> > > > > > > > > >> > > > > > they
> > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> consumers
> > > > yet.
> > > > > > I
> > > > > > > > think
> > > > > > > > > >> the
> > > > > > > > > >> > > > > > question is,
> > > > > > > > > >> > > > > > > > > first
> > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> to be
> > > > > > > > maintaining
> > > > > > > > > >> the
> > > > > > > > > >> > > > names
> > > > > > > > > >> > > > > > -> ids
> > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> the
> > > > > > augmented
> > > > > > > > > >> existing
> > > > > > > > > >> > > > RPCs
> > > > > > > > > >> > > > > > and the
> > > > > > > > > >> > > > > > > > > new
> > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> secondly,
> > > > should
> > > > > > we
> > > > > > > > ever
> > > > > > > > > >> consider
> > > > > > > > > >> > > > > > exposing
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> APIs
> > > as
> > > > > > well
> > > > > > > > (both
> > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> listener for
> > > > cases
> > > > > > > > like
> > > > > > > > > >> topic
> > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a)
> Assuming
> > > > that we
> > > > > > > > would
> > > > > > > > > >> include
> > > > > > > > > >> > > > > > converting
> > > > > > > > > >> > > > > > > > > all
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to
> using
> > > > topic
> > > > > > > ids
> > > > > > > > in
> > > > > > > > > >> this
> > > > > > > > > >> > > > KIP,
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully
> operate
> > > > with
> > > > > > > topic
> > > > > > > > > >> ids. That
> > > > > > > > > >> > > > > > being
> > > > > > > > > >> > > > > > > > > said, it
> > > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > > names in
> > > > > > > > various
> > > > > > > > > >> APIs so
> > > > > > > > > >> > > > > > having a
> > > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > > inevitable to
> > > > > > > me.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't
> > > have
> > > > a
> > > > > > > strong
> > > > > > > > > >> opinion on
> > > > > > > > > >> > > > > > this.
> > > > > > > > > >> > > > > > > > > Here I
> > > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> > > > scope of
> > > > > > > this
> > > > > > > > > >> KIP. I
> > > > > > > > > >> > > > would
> > > > > > > > > >> > > > > > rather
> > > > > > > > > >> > > > > > > > > focus
> > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and
> we
> > > can
> > > > > > > consider
> > > > > > > > > >> this
> > > > > > > > > >> > > > > > separately if
> > > > > > > > > >> > > > > > > > > we see
> > > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming
> back to
> > > > > > > Ismael's
> > > > > > > > > >> point about
> > > > > > > > > >> > > > > > using
> > > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > > >> > > > think
> > > > > > > > > >> > > > > > that
> > > > > > > > > >> > > > > > > > > there
> > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of
> it.
> > > > The
> > > > > > > > consumer
> > > > > > > > > >> will
> > > > > > > > > >> > > > have
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > > exists
> > > > > > > before
> > > > > > > > > >> passing
> > > > > > > > > >> > > > them
> > > > > > > > > >> > > > > > into
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> > > > Obviously,
> > > > > > > the
> > > > > > > > > >> > > > coordinator
> > > > > > > > > >> > > > > > will
> > > > > > > > > >> > > > > > > > > also
> > > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really
> > > have
> > > > a
> > > > > > way
> > > > > > > to
> > > > > > > > > >> reject an
> > > > > > > > > >> > > > > > invalid
> > > > > > > > > >> > > > > > > > > topic in
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> agreeing
> > > > with
> > > > > > > > David
> > > > > > > > > >> on all
> > > > > > > > > >> > > > other
> > > > > > > > > >> > > > > > minor
> > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > `subscribe(Pattern)`
> > > > > > > > > >> question:
> > > > > > > > > >> > > > > > personally
> > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> deprecate
> > > the
> > > > > > > > subscribe
> > > > > > > > > >> API with
> > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > >> > > > > > > > > but
> > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> just
> > > > > > > documenting
> > > > > > > > > >> that our
> > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> the
> > > > > > > incompatible
> > > > > > > > > >> case is a
> > > > > > > > > >> > > > > > very rare
> > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> overloaded
> > > > > > `String`
> > > > > > > > based
> > > > > > > > > >> > > > > > subscription may
> > > > > > > > > >> > > > > > > > > be
> > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > > regexes.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> > > > work. I
> > > > > > > have
> > > > > > > > to
> > > > > > > > > >> look at
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > differences
> > > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> > > > better
> > > > > > > > > >> understand the
> > > > > > > > > >> > > > > > potential
> > > > > > > > > >> > > > > > > > > issues.
> > > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would
> > > work
> > > > for
> > > > > > > all
> > > > > > > > the
> > > > > > > > > >> basic
> > > > > > > > > >> > > > > > regular
> > > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between
> the
> > > two
> > > > are
> > > > > > > > mainly
> > > > > > > > > >> about
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > various
> > > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder
> what
> > > > other
> > > > > > > people
> > > > > > > > > >> think
> > > > > > > > > >> > > > about
> > > > > > > > > >> > > > > > this.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue,
> Jul
> > > 12,
> > > > 2022
> > > > > > > at
> > > > > > > > > >> 11:28 PM
> > > > > > > > > >> > > > > > Guozhang
> > > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks
> > > David!
> > > > I
> > > > > > > think
> > > > > > > > on
> > > > > > > > > >> the high
> > > > > > > > > >> > > > > > level
> > > > > > > > > >> > > > > > > > > there are
> > > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > > concretize
> > > > a
> > > > > > bit
> > > > > > > > more:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > > migration
> > > > > > > path,
> > > > > > > > > >> especially
> > > > > > > > > >> > > > > > the last
> > > > > > > > > >> > > > > > > > > step
> > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> enable
> > > the
> > > > new
> > > > > > > > > >> protocol, in
> > > > > > > > > >> > > > which
> > > > > > > > > >> > > > > > we
> > > > > > > > > >> > > > > > > > > would
> > > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> protocols /
> > > > rpcs
> > > > > > and
> > > > > > > > old
> > > > > > > > > >> > > > protocols /
> > > > > > > > > >> > > > > > rpcs
> > > > > > > > > >> > > > > > > > > are
> > > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group.
> How
> > > the
> > > > > > > > > >> coordinator could
> > > > > > > > > >> > > > > > "mimic"
> > > > > > > > > >> > > > > > > > > the old
> > > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> protocol
> > > > is
> > > > > > > > something
> > > > > > > > > >> we
> > > > > > > > > >> > > > need to
> > > > > > > > > >> > > > > > > > > present
> > > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> usage
> > > > of
> > > > > > > topic
> > > > > > > > > >> ids. So
> > > > > > > > > >> > > > far as
> > > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> RPCs
> > > > and
> > > > > > > admin
> > > > > > > > > >> client,
> > > > > > > > > >> > > > but
> > > > > > > > > >> > > > > > they
> > > > > > > > > >> > > > > > > > > are not
> > > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> consumers
> > > > yet.
> > > > > > I
> > > > > > > > think
> > > > > > > > > >> the
> > > > > > > > > >> > > > > > question is,
> > > > > > > > > >> > > > > > > > > first
> > > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client
> to be
> > > > > > > > maintaining
> > > > > > > > > >> the
> > > > > > > > > >> > > > names
> > > > > > > > > >> > > > > > -> ids
> > > > > > > > > >> > > > > > > > > mapping
> > > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all
> the
> > > > > > augmented
> > > > > > > > > >> existing
> > > > > > > > > >> > > > RPCs
> > > > > > > > > >> > > > > > and the
> > > > > > > > > >> > > > > > > > > new
> > > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> secondly,
> > > > should
> > > > > > we
> > > > > > > > ever
> > > > > > > > > >> consider
> > > > > > > > > >> > > > > > exposing
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> APIs
> > > as
> > > > > > well
> > > > > > > > (both
> > > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> listener for
> > > > cases
> > > > > > > > like
> > > > > > > > > >> topic
> > > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> agreeing
> > > > with
> > > > > > > > David
> > > > > > > > > >> on all
> > > > > > > > > >> > > > other
> > > > > > > > > >> > > > > > minor
> > > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > `subscribe(Pattern)`
> > > > > > > > > >> question:
> > > > > > > > > >> > > > > > personally
> > > > > > > > > >> > > > > > > > > I think
> > > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> deprecate
> > > the
> > > > > > > > subscribe
> > > > > > > > > >> API with
> > > > > > > > > >> > > > > > Pattern,
> > > > > > > > > >> > > > > > > > > but
> > > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while
> just
> > > > > > > documenting
> > > > > > > > > >> that our
> > > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since
> the
> > > > > > > incompatible
> > > > > > > > > >> case is a
> > > > > > > > > >> > > > > > very rare
> > > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> overloaded
> > > > > > `String`
> > > > > > > > based
> > > > > > > > > >> > > > > > subscription may
> > > > > > > > > >> > > > > > > > > be
> > > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > > regexes.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue,
> Jul
> > > > 12,
> > > > > > 2022
> > > > > > > > at
> > > > > > > > > >> 5:23 AM
> > > > > > > > > >> > > > > > David Jacot
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi
> Ismael,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Thanks for
> > > > your
> > > > > > > > > >> feedback. Let
> > > > > > > > > >> > > > me
> > > > > > > > > >> > > > > > answer
> > > > > > > > > >> > > > > > > > > your
> > > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I
> > > think
> > > > > > it's
> > > > > > > > > >> premature to
> > > > > > > > > >> > > > > > talk about
> > > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> deprecation
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> removal
> > > > of the
> > > > > > > > > >> existing group
> > > > > > > > > >> > > > > > protocol.
> > > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> client
> > > > > > protocol
> > > > > > > > and
> > > > > > > > > >> hence
> > > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> > > > quickly
> > > > > > > > > >> applications
> > > > > > > > > >> > > > migrate
> > > > > > > > > >> > > > > > to
> > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That
> makes
> > > > > > sense.
> > > > > > > I
> > > > > > > > will
> > > > > > > > > >> > > > remove it.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2.
> The
> > > KIP
> > > > > > says
> > > > > > > we
> > > > > > > > > >> intend to
> > > > > > > > > >> > > > > > release
> > > > > > > > > >> > > > > > > > > this in
> > > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> why. If
> > > we
> > > > > > added
> > > > > > > > that
> > > > > > > > > >> as a
> > > > > > > > > >> > > > way to
> > > > > > > > > >> > > > > > > > > estimate
> > > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > group
> > > > > > > > protocol, I
> > > > > > > > > >> also
> > > > > > > > > >> > > > > > suggest
> > > > > > > > > >> > > > > > > > > removing
> > > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > explain
> > > > > > my
> > > > > > > > > >> reasoning. As
> > > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > coordinator
> > > > in
> > > > > > > Java
> > > > > > > > > >> while we
> > > > > > > > > >> > > > > > implement
> > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> the
> > > > > > internals
> > > > > > > > will
> > > > > > > > > >> be
> > > > > > > > > >> > > > slightly
> > > > > > > > > >> > > > > > > > > different
> > > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > Therefore, I
> > > > > > > wanted
> > > > > > > > to
> > > > > > > > > >> tighten
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > switch from
> > > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > coordinator
> > > > to
> > > > > > the
> > > > > > > > new
> > > > > > > > > >> group
> > > > > > > > > >> > > > > > coordinator
> > > > > > > > > >> > > > > > > > > to a
> > > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > alternative
> > > > > > would
> > > > > > > > be to
> > > > > > > > > >> use a
> > > > > > > > > >> > > > flag
> > > > > > > > > >> > > > > > to do
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> > > > software
> > > > > > > > upgrade.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3.
> We
> > > > need to
> > > > > > > > flesh
> > > > > > > > > >> out the
> > > > > > > > > >> > > > > > details of
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> we're
> > > > saying
> > > > > > we
> > > > > > > > will
> > > > > > > > > >> support
> > > > > > > > > >> > > > > > online
> > > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> explain
> > > > this
> > > > > > in
> > > > > > > > > >> detail. It
> > > > > > > > > >> > > > could
> > > > > > > > > >> > > > > > also
> > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> easier.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes,
> we
> > > will
> > > > > > > support
> > > > > > > > > >> online
> > > > > > > > > >> > > > > > migrations
> > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> group
> > > > using
> > > > > > the
> > > > > > > > old
> > > > > > > > > >> protocol
> > > > > > > > > >> > > > > > will be
> > > > > > > > > >> > > > > > > > > able to
> > > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> protocol.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > briefly
> > > > > > > > explain
> > > > > > > > > >> how that
> > > > > > > > > >> > > > > > will work
> > > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four
> step
> > > > > > process:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> > > > cluster
> > > > > > > must
> > > > > > > > be
> > > > > > > > > >> > > > upgraded or
> > > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> group
> > > > > > > > coordinator.
> > > > > > > > > >> Both the
> > > > > > > > > >> > > > > > old and
> > > > > > > > > >> > > > > > > > > the new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> support
> > > the
> > > > old
> > > > > > > > > >> protocol and
> > > > > > > > > >> > > > rely
> > > > > > > > > >> > > > > > on the
> > > > > > > > > >> > > > > > > > > same
> > > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they
> can
> > > > work
> > > > > > > > together.
> > > > > > > > > >> This
> > > > > > > > > >> > > > point
> > > > > > > > > >> > > > > > is an
> > > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> this
> > > one
> > > > live
> > > > > > > > > >> because it
> > > > > > > > > >> > > > would
> > > > > > > > > >> > > > > > require
> > > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > coordinator
> > > > and
> > > > > > > > > >> starting up the
> > > > > > > > > >> > > > > > new one
> > > > > > > > > >> > > > > > > > > and
> > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > unavailabilities.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> > > > cluster's
> > > > > > > > > >> metadata
> > > > > > > > > >> > > > > > version/IBP
> > > > > > > > > >> > > > > > > > > must be
> > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> enable
> > > > the
> > > > > > new
> > > > > > > > > >> protocol.
> > > > > > > > > >> > > > This
> > > > > > > > > >> > > > > > cannot
> > > > > > > > > >> > > > > > > > > be done
> > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> terminated
> > > > > > because
> > > > > > > > the
> > > > > > > > > >> old
> > > > > > > > > >> > > > > > coordinator
> > > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> protocol.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> > > > consumers
> > > > > > > > must be
> > > > > > > > > >> > > > upgraded
> > > > > > > > > >> > > > > > to a
> > > > > > > > > >> > > > > > > > > version
> > > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> migration
> > > > (must
> > > > > > > have
> > > > > > > > > >> KIP-792).
> > > > > > > > > >> > > > If
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > consumer
> > > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Nothing
> > > > must be
> > > > > > > > done at
> > > > > > > > > >> this
> > > > > > > > > >> > > > point.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> > > > consumers
> > > > > > > > must
> > > > > > > > > >> be rolled
> > > > > > > > > >> > > > > > with the
> > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> consumer
> > > > group
> > > > > > is
> > > > > > > > > >> automatically
> > > > > > > > > >> > > > > > > > > converted when
> > > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using
> the
> > > > new
> > > > > > > > protocol
> > > > > > > > > >> joins
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > group.
> > > > > > > > > >> > > > > > > > > While
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > > > protocol are
> > > > > > > > being
> > > > > > > > > >> > > > upgraded,
> > > > > > > > > >> > > > > > the old
> > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> one.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > > clarify
> > > > > > all
> > > > > > > > of
> > > > > > > > > >> this in
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > KIP.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4.
> I am
> > > > happy
> > > > > > > > that we
> > > > > > > > > >> are
> > > > > > > > > >> > > > > > pushing the
> > > > > > > > > >> > > > > > > > > pattern
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> server,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but
> it
> > > > seems
> > > > > > > like
> > > > > > > > > >> there
> > > > > > > > > >> > > > could be
> > > > > > > > > >> > > > > > some
> > > > > > > > > >> > > > > > > > > tricky
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> have a
> > > > > > mechanism
> > > > > > > > for
> > > > > > > > > >> users to
> > > > > > > > > >> > > > > > detect
> > > > > > > > > >> > > > > > > > > that
> > > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> before
> > > > > > switching
> > > > > > > > to
> > > > > > > > > >> the new
> > > > > > > > > >> > > > > > protocol?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> think
> > > > that I
> > > > > > am
> > > > > > > a
> > > > > > > > bit
> > > > > > > > > >> more
> > > > > > > > > >> > > > > > optimistic
> > > > > > > > > >> > > > > > > > > than
> > > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> believe
> > > > that the
> > > > > > > > > >> majority of
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > cases
> > > > > > > > > >> > > > > > > > > are
> > > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work
> with
> > > > the
> > > > > > new
> > > > > > > > > >> engine. The
> > > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> reject
> > > > the
> > > > > > > > consumer
> > > > > > > > > >> if the
> > > > > > > > > >> > > > > > regex is
> > > > > > > > > >> > > > > > > > > not
> > > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> migration
> > > > path,
> > > > > > in
> > > > > > > > the
> > > > > > > > > >> worst
> > > > > > > > > >> > > > case,
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > first
> > > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> group
> > > > will
> > > > > > be
> > > > > > > > > >> rejected.
> > > > > > > > > >> > > > This
> > > > > > > > > >> > > > > > should
> > > > > > > > > >> > > > > > > > > be used
> > > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> say.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One
> way
> > > for
> > > > > > > > customers to
> > > > > > > > > >> > > > validate
> > > > > > > > > >> > > > > > their
> > > > > > > > > >> > > > > > > > > regex
> > > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod
> would
> > > > be to
> > > > > > > > test
> > > > > > > > > >> them with
> > > > > > > > > >> > > > > > another
> > > > > > > > > >> > > > > > > > > group.
> > > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could
> be
> > > > done
> > > > > > in a
> > > > > > > > > >> pre-prod
> > > > > > > > > >> > > > > > environment.
> > > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> extend the
> > > > > > > > > >> consumer-group tool
> > > > > > > > > >> > > > to
> > > > > > > > > >> > > > > > > > > provide a
> > > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> mechanism.
> > > > Would
> > > > > > > > this be
> > > > > > > > > >> > > > enough in
> > > > > > > > > >> > > > > > your
> > > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > > > Related to
> > > > > > > the
> > > > > > > > last
> > > > > > > > > >> > > > question,
> > > > > > > > > >> > > > > > will
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> stick
> > > > with the
> > > > > > > > > >> current regex
> > > > > > > > > >> > > > > > engine for
> > > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> example,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it
> may
> > > be
> > > > > > handy
> > > > > > > to
> > > > > > > > > >> keep using
> > > > > > > > > >> > > > > > client
> > > > > > > > > >> > > > > > > > > based
> > > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > migrations
> > > > > > > simple
> > > > > > > > and
> > > > > > > > > >> then
> > > > > > > > > >> > > > > > migrate to
> > > > > > > > > >> > > > > > > > > server
> > > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > > understand
> > > > > > your
> > > > > > > > point
> > > > > > > > > >> but I
> > > > > > > > > >> > > > am
> > > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > actually
> > > > stay
> > > > > > > in
> > > > > > > > > >> this mode.
> > > > > > > > > >> > > > That
> > > > > > > > > >> > > > > > > > > would go
> > > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > simplifying
> > > > the
> > > > > > > > client
> > > > > > > > > >> because
> > > > > > > > > >> > > > we
> > > > > > > > > >> > > > > > would
> > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > metadata on
> > > > > > > the
> > > > > > > > > >> client
> > > > > > > > > >> > > > side. I
> > > > > > > > > >> > > > > > would
> > > > > > > > > >> > > > > > > > > rather
> > > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6.
> When
> > > > we say
> > > > > > > > that
> > > > > > > > > >> the group
> > > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > > > configurations
> > > > > > > > > >> and that
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> group is
> > > > > > > deleted.
> > > > > > > > > >> Will a
> > > > > > > > > >> > > > > > transition to
> > > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > configurations?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> > > > right.
> > > > > > The
> > > > > > > > > >> > > > configurations
> > > > > > > > > >> > > > > > will be
> > > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> deleted.
> > > > They go
> > > > > > > > > >> together.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7.
> Will
> > > > the
> > > > > > > > choice to
> > > > > > > > > >> store
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > configs in
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> harder
> > > to
> > > > list
> > > > > > > all
> > > > > > > > > >> cluster
> > > > > > > > > >> > > > > > configs and
> > > > > > > > > >> > > > > > > > > their
> > > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> don't
> > > > think
> > > > > > so.
> > > > > > > > The
> > > > > > > > > >> group
> > > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> configs.
> > > If
> > > > you
> > > > > > > > want to
> > > > > > > > > >> know
> > > > > > > > > >> > > > all
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > overrides
> > > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> ask all
> > > > the
> > > > > > > group
> > > > > > > > > >> > > > coordinators.
> > > > > > > > > >> > > > > > You
> > > > > > > > > >> > > > > > > > > cannot
> > > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > instance.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8.
> How
> > > > would
> > > > > > > > someone
> > > > > > > > > >> > > > configure a
> > > > > > > > > >> > > > > > group
> > > > > > > > > >> > > > > > > > > before
> > > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > > > considered
> > > > > > > > > >> allowing the
> > > > > > > > > >> > > > > > explicit
> > > > > > > > > >> > > > > > > > > creation
> > > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Alternatively,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> configs
> > > > could
> > > > > > be
> > > > > > > > > >> decoupled
> > > > > > > > > >> > > > from
> > > > > > > > > >> > > > > > the
> > > > > > > > > >> > > > > > > > > group
> > > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> The
> > > > group
> > > > > > > will
> > > > > > > > be
> > > > > > > > > >> > > > > > automatically
> > > > > > > > > >> > > > > > > > > created in
> > > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > configs
> > > > will
> > > > > > > be
> > > > > > > > > >> lost after
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > retention
> > > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> passes.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9.
> Will
> > > > the
> > > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > > >> > > > > > method
> > > > > > > > > >> > > > > > > > > for the
> > > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > > >> > > > do
> > > > > > > > > >> > > > > > we
> > > > > > > > > >> > > > > > > > > have to
> > > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> That's a
> > > > very
> > > > > > > group
> > > > > > > > > >> question. I
> > > > > > > > > >> > > > > > forgot
> > > > > > > > > >> > > > > > > > > about
> > > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > > >> > > > not
> > > > > > > > > >> > > > > > fully
> > > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> plan to
> > > > use,
> > > > > > it
> > > > > > > > > >> might be
> > > > > > > > > >> > > > better
> > > > > > > > > >> > > > > > to
> > > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which
> > > takes
> > > > a
> > > > > > > > string.
> > > > > > > > > >> We would
> > > > > > > > > >> > > > > > rely on
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> During the
> > > > > > > > migration, I
> > > > > > > > > >> think
> > > > > > > > > >> > > > that
> > > > > > > > > >> > > > > > we
> > > > > > > > > >> > > > > > > > > could
> > > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> and
> > > > use
> > > > > > it.
> > > > > > > > That
> > > > > > > > > >> should
> > > > > > > > > >> > > > > > work, I
> > > > > > > > > >> > > > > > > > > think, in
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10.
> I
> > > > agree
> > > > > > with
> > > > > > > > > >> Justine
> > > > > > > > > >> > > > that we
> > > > > > > > > >> > > > > > > > > should be
> > > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> switch
> > > to
> > > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > > >> > > > > > from the
> > > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> > > > switch the
> > > > > > > > > >> default for
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > feature
> > > > > > > > > >> > > > > > > > > flag
> > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > metadata.version
> > > > > > > > once
> > > > > > > > > >> we
> > > > > > > > > >> > > > want to
> > > > > > > > > >> > > > > > make
> > > > > > > > > >> > > > > > > > > it the
> > > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My
> plan
> > > was
> > > > to
> > > > > > use
> > > > > > > > that
> > > > > > > > > >> feature
> > > > > > > > > >> > > > > > flag
> > > > > > > > > >> > > > > > > > > mainly
> > > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> phase. I
> > > > should
> > > > > > > not
> > > > > > > > have
> > > > > > > > > >> > > > mentioned
> > > > > > > > > >> > > > > > it, I
> > > > > > > > > >> > > > > > > > > think,
> > > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > > internal
> > > > > > config
> > > > > > > > for
> > > > > > > > > >> it.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11.
> Some
> > > > of
> > > > > > the
> > > > > > > > > >> protocol APIs
> > > > > > > > > >> > > > > > don't
> > > > > > > > > >> > > > > > > > > mention
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> good to
> > > > add
> > > > > > that
> > > > > > > > for
> > > > > > > > > >> > > > consistency.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12.
> It
> > > is
> > > > a
> > > > > > bit
> > > > > > > > odd
> > > > > > > > > >> that
> > > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> though
> > > it
> > > > > > seems
> > > > > > > > to do
> > > > > > > > > >> more
> > > > > > > > > >> > > > than
> > > > > > > > > >> > > > > > > > > reading.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> agree.
> > > > This is
> > > > > > > > how the
> > > > > > > > > >> > > > current
> > > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> require
> > > > "Read
> > > > > > > > Group" to
> > > > > > > > > >> join a
> > > > > > > > > >> > > > > > group. We
> > > > > > > > > >> > > > > > > > > could
> > > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I
> am
> > > not
> > > > > > sure
> > > > > > > > that
> > > > > > > > > >> it is
> > > > > > > > > >> > > > worth
> > > > > > > > > >> > > > > > it.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13.
> How
> > > is
> > > > > > topic
> > > > > > > > > >> recreation
> > > > > > > > > >> > > > > > handled by
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > protocol?
> > > > It
> > > > > > > > would be
> > > > > > > > > >> good to
> > > > > > > > > >> > > > > > have a
> > > > > > > > > >> > > > > > > > > section
> > > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > From
> > > > a
> > > > > > > > protocol
> > > > > > > > > >> > > > > > perspective, the
> > > > > > > > > >> > > > > > > > > new
> > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic
> id
> > > so
> > > > it
> > > > > > > will
> > > > > > > > > >> treat it
> > > > > > > > > >> > > > like a
> > > > > > > > > >> > > > > > > > > topic with
> > > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only
> issue
> > > > is
> > > > > > that
> > > > > > > > the
> > > > > > > > > >> > > > fetch/commit
> > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs
> so the
> > > > > > > consumer
> > > > > > > > > >> would
> > > > > > > > > >> > > > reuse the
> > > > > > > > > >> > > > > > > > > offsets
> > > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> we
> > > > should
> > > > > > > > update
> > > > > > > > > >> those
> > > > > > > > > >> > > > APIs
> > > > > > > > > >> > > > > > as well
> > > > > > > > > >> > > > > > > > > in
> > > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> end.
> > > That
> > > > > > would
> > > > > > > > > >> strengthen
> > > > > > > > > >> > > > the
> > > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14.
> The
> > > > KIP
> > > > > > > > mentions
> > > > > > > > > >> we will
> > > > > > > > > >> > > > > > write the
> > > > > > > > > >> > > > > > > > > new
> > > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> this is
> > > an
> > > > > > > > > >> implementation
> > > > > > > > > >> > > > > > detail, do
> > > > > > > > > >> > > > > > > > > we plan
> > > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for
> it?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15.
> Do
> > > we
> > > > > > have a
> > > > > > > > > >> scalability
> > > > > > > > > >> > > > > > goal when
> > > > > > > > > >> > > > > > > > > it
> > > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> group
> > > > protocol
> > > > > > > can
> > > > > > > > > >> support?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We
> don't
> > > > have
> > > > > > > > numbers
> > > > > > > > > >> at the
> > > > > > > > > >> > > > > > moment. The
> > > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of
> members
> > > > per
> > > > > > > > group.
> > > > > > > > > >> We will
> > > > > > > > > >> > > > > > measure
> > > > > > > > > >> > > > > > > > > this when
> > > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > implementation.
> > > > > > > Note
> > > > > > > > > >> that we
> > > > > > > > > >> > > > might
> > > > > > > > > >> > > > > > have
> > > > > > > > > >> > > > > > > > > other
> > > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road
> (e.g.
> > > > > > offset
> > > > > > > > > >> commits).
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16.
> Did
> > > we
> > > > > > > > consider
> > > > > > > > > >> having
> > > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > SubscribedTopicNames
> > > > > > > > > >> in
> > > > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> since we
> > > > have
> > > > > > to
> > > > > > > > > >> resolve the
> > > > > > > > > >> > > > > > regex on
> > > > > > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > topic
> > > > > > name?
> > > > > > > > The
> > > > > > > > > >> > > > difference
> > > > > > > > > >> > > > > > is that
> > > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> efficient
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> whereas
> > > > > > sending
> > > > > > > > the
> > > > > > > > > >> topic
> > > > > > > > > >> > > > names
> > > > > > > > > >> > > > > > is less
> > > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > recreation is
> > > > > > > > easier
> > > > > > > > > >> to
> > > > > > > > > >> > > > handle
> > > > > > > > > >> > > > > > if we
> > > > > > > > > >> > > > > > > > > have
> > > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The
> idea
> > > > was to
> > > > > > > > > >> consolidate the
> > > > > > > > > >> > > > > > metadata
> > > > > > > > > >> > > > > > > > > lookup
> > > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths
> but
> > > I
> > > > do
> > > > > > > agree
> > > > > > > > > >> with your
> > > > > > > > > >> > > > > > point. As
> > > > > > > > > >> > > > > > > > > a
> > > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids
> may be
> > > > > > better
> > > > > > > > here
> > > > > > > > > >> for the
> > > > > > > > > >> > > > > > delete and
> > > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> suppose
> > > > that we
> > > > > > > may
> > > > > > > > > >> allow
> > > > > > > > > >> > > > users to
> > > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> > > > because
> > > > > > > that
> > > > > > > > is
> > > > > > > > > >> the only
> > > > > > > > > >> > > > > > way to be
> > > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > re-creation.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On
> Tue,
> > > Jul
> > > > 12,
> > > > > > > > 2022 at
> > > > > > > > > >> 1:38 PM
> > > > > > > > > >> > > > > > David
> > > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> djacot@confluent.io>
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > > > Justine,
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> Thanks
> > > for
> > > > > > your
> > > > > > > > > >> comments.
> > > > > > > > > >> > > > Please
> > > > > > > > > >> > > > > > find
> > > > > > > > > >> > > > > > > > > my
> > > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > -
> Yes,
> > > > the new
> > > > > > > > > >> protocol
> > > > > > > > > >> > > > relies on
> > > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> topic
> > > > names
> > > > > > > based
> > > > > > > > in
> > > > > > > > > >> the
> > > > > > > > > >> > > > > > > > > > > >> > > > >
> > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if
> using
> > > > topic
> > > > > > > > names
> > > > > > > > > >> is the
> > > > > > > > > >> > > > > > right call
> > > > > > > > > >> > > > > > > > > here.
> > > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a
> little
> > > > more.
> > > > > > > > > >> Obviously, the
> > > > > > > > > >> > > > > > KIP does
> > > > > > > > > >> > > > > > > > > not
> > > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > > >> > > > > > > > > > > >>
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Jun,

Thanks for your comments.

48. Okay. There is a typo in the text:
> Note that the persisted target assignment of B already includes foo-2 here
but B does not get it yet because foo-2 is not free.

It should be:
> Note that the persisted member assignment of B already includes foo-2 here
but B does not get it yet because foo-2 is not free.

The examples are showing the persisted state in the __consumer_offsets
topics. When a member transitions to the next epoch, we directly
persist the whole target assignment in order to reduce the number of
writes and we manage dependencies in memory. So "B - epoch=2,
partitions=[foo-2]" is actually correct from a persisted record
perspective. However, the in-memory state would be different because
foo-2 is not yet available as you pointed out. The in-memory state
would like as follow:

Group (epoch=2)
A
B
Target Assignment (epoch=2)
A - partitions=[foo-0, foo-1]
B - partitions=[foo-2]
Member Assignment
A - epoch=1, partitions=[foo-0, foo-1, foo-2], pending partitions=[]
B - epoch=2, partitions=[], pending partitions=[foo-2]

I suppose that the confusion comes from this. I will go and convert
all the examples to their in-memory representation in order to reduce
the confusion.

50. Okay. I have to admit that I am not so familiar with the internals
of KStreams. If I understand correctly, you're saying that KStreams
may not have topic ids in its topology so topic names are missing here
to match the topics. I was hoping that KStreams would have topic ids
in the topology as well. If that is not the case, we can also add
topic names on our side. I will check this.

53. It seems that I was not clear. Sorry for that. I did not mean to
say that MemberAssignment.metadata is not used by the group
coordinator. Let me re-explain the flow.

PartitionAssignor.assign() returns an Assignment. Assignment contains
MemberAssignments where each member has MemberAssignment.metadata. The
Assignment is installed by the client with the
ConsumerGroupInstallAssignment API and this overrides the current
target assignment in the coordinator. Then the group coordinator uses
it to reconcile members. Therefore the target assignment will
eventually become the current member assignment and it will be
delivered to the member, including the original
MemberAssignment.metadata defines by the assignor.

I hope that this helps to clarify how MemberAssignment.metadata is
used afterwards.

Best,
David

On Tue, Sep 27, 2022 at 10:31 PM Jun Rao <ju...@confluent.io.invalid> wrote:
> Hi, David,
>
> Thanks for the reply. A few more follow up questions.
>
> 48. Hmm, so member assignment is what the client sees. Then, I am not sure
> about the following. The text says B cannot get foo-2 yet. But in Member
> Assignment, B has foo-2.
>
> At this point B can transition to epoch 2 but cannot get foo-2 until A
> revokes it.
>
> Note that the persisted target assignment of B already includes foo-2 here
> but B does not get it yet because foo-2 is not free.
>
> Group (epoch=2)
> A
> B
> Target Assignment (epoch=2)
> A - partitions=[foo-0, foo-1]
> B - partitions=[foo-2]
> Member Assignment
> A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> B - epoch=2, partitions=[foo-2]
>
>
> 50. Hmm, I thought KStreams doesn't use range assignor all the time. It
> does the assignment based on the topology. For example, if you are joining
> two topics A and B and deriving the join result in a 3rd topic C, which
> will be aggregated. You want to use range to co-locate the partitions from
> topic A and B. However, for topic C, the partition could potentially be
> assigned in round robin. Without the topic name, KStreams won't know which
> topics need to be co-located.
>
> 53. If MemberAssignment.metadata is not used by the group coordinator,
> where is it being used?
>
> Thanks,
>
> Jun
>
> On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson <ja...@confluent.io.invalid>
> wrote:
>
> > Hi David,
> >
> > Thanks for the response. Replied below:
> >
> > > I don't have a strong opinion on this one. However, it seems that this
> > could be a pandora box if our subset turns out to not be sufficient
> > for some existing users. We will have to add more to it and the
> > compatibility story between versions could be a real nightmare. One
> > thing that we could do for sure is disabling all unicode groups in the
> > library. That seems possible and would already reduce the scope.
> >
> > Changes to the regex syntax seems not a major problem to me. We would bump
> > the protocol version as we usually do. Actually that begs the same question
> > about RE2. How stable is the syntax and does it differ depending on the
> > version of the library? In general, I do agree that we'd be better off not
> > having to manage the syntax ourselves if we can convince ourselves that
> > compatibility won't be a problem and the client won't eventually need to
> > have access to it.
> >
> > Thanks,
> > Jason
> >
> > On Tue, Sep 27, 2022 at 6:21 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi Luke,
> > >
> > > Thanks for your comments.
> > >
> > > Your suggestion makes sense. Jun's made the same one. We are also
> > > discussing whether we should directly return the group state in the
> > > response and drop the prepare assignment api. It seems that the
> > > current consensus is to keep it though. If that stays like this, I
> > > will apply your suggestion.
> > >
> > > Thanks,
> > > David
> > >
> > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com> wrote:
> > > >
> > > > Hi David,
> > > >
> > > > Thanks for your patience for answering all these questions!
> > > >
> > > > One more question from me:
> > > > 1. When the selected consumer is going to be in client-side assigning,
> > we
> > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering if
> > we
> > > > could use a flag to indicate the assignment is required. The rationale
> > is
> > > > when returning error, we lose a chance to catch up with the latest
> > target
> > > > assignment or revoke partitions. Since we expected the record fetching
> > is
> > > > not blocked by the rebalance/assignment computing, we might want to
> > allow
> > > > the members to update the member assignment in each heartbeat.
> > > >
> > > > Luke
> > > >
> > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > <ja...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hey David,
> > > > >
> > > > > Thanks for the response.
> > > > >
> > > > > > Do you think that we could have a case where two members with the
> > > same
> > > > > static id are active?
> > > > >
> > > > > I was just thinking about the handoff when we spin up a new node
> > > before the
> > > > > old one completes shutdown. It's probably ok to fence the old member
> > > > > abruptly. I do recall we had a few issues with this handoff in the
> > > past. It
> > > > > was due to the fact that the replaced node might have been the
> > leader.
> > > > > Since we've removed the leader notion, perhaps this kind of issue
> > won't
> > > > > resurface.
> > > > >
> > > > > In regard to the RE2 library. I understand the intent to pass this
> > > through
> > > > > as a string without any validation so that we don't need the
> > > dependence on
> > > > > the client. I do wonder if that will be restrictive in practice. For
> > > > > example, if a regex changes on the client side, would we need to
> > > validate
> > > > > against the current assignment to see which partitions to continue
> > > > > fetching? Or would we instead invalidate the complete current
> > > assignment?
> > > > > Perhaps this is a rare case and we can do the latter. There could be
> > > more
> > > > > situations though.
> > > > >
> > > > > As an alternative, I wonder if we could define a minimalistic regex
> > > syntax
> > > > > (a subset of RE2 say) that could map more easily to regex support
> > that
> > > > > already exists in clients? For example, in this scenario, we are
> > > matching
> > > > > topics, so we only need to be able to represent topic characters and
> > > not
> > > > > the whole smorgasbord of whitespace and unicode options. Would it
> > make
> > > > > sense to have a reduced syntax which includes only standard topic
> > > > > characters, wildcards, character classes, and repetitions?
> > > > >
> > > > > Best,
> > > > > Jason
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > > >
> > > > > > Hi David, thanks for all the detailed explanations. I think they
> > all
> > > make
> > > > > > sense. Just want to have a couple follow-ups here:
> > > > > >
> > > > > > > I don't really see the benefits here because server side
> > assignors
> > > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > > are not supposed to generate metadata nor to receive metadata from
> > > the
> > > > > > members.
> > > > > >
> > > > > > Just to clarify I was asking about the `version` of the assignor
> > > (i.e. up
> > > > > > to what version that the client would support), and I do agree we
> > > would
> > > > > not
> > > > > > need metadata. What I have in mind is that, for some specific
> > > built-in
> > > > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> > > in a
> > > > > > newer version we would have a hierarchical rack ID string format,
> > > like
> > > > > > "tier1-tier2" etc, but if some client has not upgraded their rack
> > ID
> > > > > > would still be in old format. In this case, the broker then needs
> > to
> > > > > choose
> > > > > > the old versioned assignor. I'm probably making something up here
> > for
> > > > > rack
> > > > > > aware assignors, but I'm wondering if in general such an
> > > "auto-downgrade"
> > > > > > behavior would be needed still for broker-side assignor, and if yes
> > > would
> > > > > > "version" still be useful.
> > > > > >
> > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is called
> > > once
> > > > > > when the member transitions to a new epoch. This one contains the
> > > full
> > > > > > metadata provided by the client side assignor. Then, `onAssignment`
> > > > > > can be called max N times where N is the number of partitions
> > pending
> > > > > > revocation by other members. Let me try to clarify this in the KIP.
> > > > > >
> > > > > > Okay, my understanding is that the calling ordering of these
> > > callbacks
> > > > > > would be like the following:
> > > > > >
> > > > > > ----------------------------------------
> > > > > > onPartitionsRevoked();   // just once, since we do not really need
> > > > > > to revoke incrementally.
> > > > > >
> > > > > > onAssignment();    // the first call, with epoch incremented
> > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > >
> > > > > > onAssignment();              // the first onAssignment would bump
> > up
> > > the
> > > > > > epoch, and the metadata reflected.
> > > > > > onPartitionsAssigned();   // each time we get an additional
> > > assignment,
> > > > > we
> > > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > > ...
> > > > > > onAssignment();
> > > > > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > > > > encoded
> > > > > > metadata would not change, only the incrementally added partitions
> > be
> > > > > > reflected
> > > > > >
> > > > > > Is that the case?
> > > > > >
> > > > > > I'm wondering if we would still call onAssignment just once, that
> > > encodes
> > > > > > all the assignment for this rebalance, including all the partitions
> > > that
> > > > > > should be assigned to the member but not yet assigned since they
> > > have not
> > > > > > been revoked by others. In that case the call ordering would be:
> > > > > >
> > > > > > ----------------------------------------
> > > > > > onPartitionsRevoked();   // just once
> > > > > > onAssignment();    // just once, with epoch incremented, and
> > metadata
> > > > > > encoded changed, the "assignment" field also reflect the final
> > target
> > > > > > assignment
> > > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > > incrementally
> > > > > > added partitions
> > > > > > ...
> > > > > > onPartitionsAssigned();
> > > > > >
> > > > > > The motivation from this is that, most users would only implement
> > the
> > > > > > rebalance callback listeners and hence we'd definitely need to make
> > > sure
> > > > > > the semantics of that does not change much, and the time
> > > > > > `onPartitionsAssigned` indicate the time when the partitions are
> > > actually
> > > > > > assigned to it; while for assignors, the `onAssignment` is used to
> > > > > indicate
> > > > > > what decision is made regarding for this member, i.e. when the
> > > partitions
> > > > > > are decided to be given to it, but not necessarily meant that it
> > has
> > > been
> > > > > > given, since that time should be determined by the time of
> > > > > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > > > > implementers would not need to reason which `onAssignment` would be
> > > the
> > > > > > last one for this epoch.
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > <djacot@confluent.io.invalid
> > > > > >
> > > > > > wrote:
> > > > > >
> > > > > > > Hi Guozhang,
> > > > > > >
> > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > > byte
> > > > > > > field,
> > > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > > And
> > > > > the
> > > > > > > > rationale seems to be assuming that we should always be able to
> > > do
> > > > > the
> > > > > > > > assignment at the broker-side assignor without errors.
> > > Personally I
> > > > > > think
> > > > > > > > it's still potentially beneficial to add the Error field even
> > for
> > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > subscribed
> > > > > > > > topics are not recognized with the current broker's metadata.
> > > What do
> > > > > > you
> > > > > > > > think?
> > > > > > >
> > > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > > better
> > > > > > > to use an exception on the server side. This is what we usually
> > do
> > > for
> > > > > > > server side plugins. On the client side, we use a field because
> > the
> > > > > > > errors are not defined in advance.
> > > > > > >
> > > > > > > Your comment also makes me think about what we should do when the
> > > > > > > server side assignor fails. I suppose that we have to keep the
> > > current
> > > > > > > assignment until a new event occurs. For instance, in your
> > example,
> > > > > > > the coordinator would have to trigger a rebalance when
> > unrecognized
> > > > > > > topics are available. This would be part of the metadata
> > > monitoring.
> > > > > > >
> > > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > > reason/version/metadata compared with the broker-side
> > > GroupMember. I
> > > > > > > agree
> > > > > > > > that broker-side assignor would not need reason/metadata since
> > > they
> > > > > are
> > > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > > E.g.
> > > > > is
> > > > > > > it
> > > > > > > > possible that we evolve our broker-side built-in assignor but
> > > the old
> > > > > > > > versioned clients would not be able to work with the new
> > > version, in
> > > > > > > which
> > > > > > > > case we need to let the broker being aware of this and upgrade
> > > its
> > > > > > > behavior
> > > > > > > > to cooperate with the clients?
> > > > > > >
> > > > > > > I don't really see the benefits here because server side
> > assignors
> > > > > > > don't have metadata at all. They only assign topic-partitions.
> > They
> > > > > > > are not supposed to generate metadata nor to receive metadata
> > from
> > > the
> > > > > > > members.
> > > > > > >
> > > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > > instead
> > > > > > > of
> > > > > > > > including these three fields, what about just adding the
> > > "Metadata"
> > > > > > field
> > > > > > > > class which has these three fields? Also, there are two
> > > "Metadata"
> > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > reason/version/metadata, and the second is just the encoded
> > > metadata
> > > > > > > bytes.
> > > > > > > > I'm wondering what about just naming the first as
> > memberMetadata,
> > > > > which
> > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > second
> > > > > > > > bytebuffer field as metadataBytes?
> > > > > > >
> > > > > > > That's a good point. Let me try to rationalize this interface
> > > based on
> > > > > > > your suggestions.
> > > > > > >
> > > > > > > Best,
> > > > > > > David
> > > > > > >
> > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > > wrote:
> > > > > > > >
> > > > > > > > Hello David,
> > > > > > > >
> > > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > > >
> > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > > byte
> > > > > > > field,
> > > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > > And
> > > > > the
> > > > > > > > rationale seems to be assuming that we should always be able to
> > > do
> > > > > the
> > > > > > > > assignment at the broker-side assignor without errors.
> > > Personally I
> > > > > > think
> > > > > > > > it's still potentially beneficial to add the Error field even
> > for
> > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > subscribed
> > > > > > > > topics are not recognized with the current broker's metadata.
> > > What do
> > > > > > you
> > > > > > > > think?
> > > > > > > >
> > > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > > reason/version/metadata compared with the broker-side
> > > GroupMember. I
> > > > > > > agree
> > > > > > > > that broker-side assignor would not need reason/metadata since
> > > they
> > > > > are
> > > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > > E.g.
> > > > > is
> > > > > > > it
> > > > > > > > possible that we evolve our broker-side built-in assignor but
> > > the old
> > > > > > > > versioned clients would not be able to work with the new
> > > version, in
> > > > > > > which
> > > > > > > > case we need to let the broker being aware of this and upgrade
> > > its
> > > > > > > behavior
> > > > > > > > to cooperate with the clients?
> > > > > > > >
> > > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > > instead
> > > > > > > of
> > > > > > > > including these three fields, what about just adding the
> > > "Metadata"
> > > > > > field
> > > > > > > > class which has these three fields? Also, there are two
> > > "Metadata"
> > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > reason/version/metadata, and the second is just the encoded
> > > metadata
> > > > > > > bytes.
> > > > > > > > I'm wondering what about just naming the first as
> > memberMetadata,
> > > > > which
> > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > second
> > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hello David,
> > > > > > > > >
> > > > > > > > > Thanks for bringing this question up. I think the main
> > > benefits as
> > > > > > you
> > > > > > > > > listed is 2) above if it stays; just to clarify, we would
> > only
> > > be
> > > > > > able
> > > > > > > to
> > > > > > > > > save one round trip if the rebalance is still triggered by
> > the
> > > > > > broker;
> > > > > > > if
> > > > > > > > > the rebalance is triggered by the client then the
> > > num.round.trips
> > > > > are
> > > > > > > the
> > > > > > > > > same:
> > > > > > > > >
> > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > >
> > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > already
> > > > > > sent
> > > > > > > a
> > > > > > > > > HB and hence has to wait for it to return first since only
> > one
> > > > > > request
> > > > > > > /
> > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > GroupInstallAssignment
> > > > > > > > > request.
> > > > > > > > >
> > > > > > > > > In total, two round trips.
> > > > > > > > >
> > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > >
> > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > already
> > > > > > sent
> > > > > > > a
> > > > > > > > > HB and hence has to wait for it to return first since only
> > one
> > > > > > request
> > > > > > > /
> > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > T1: client receives the HB response, and then sends the new
> > HB
> > > > > > request
> > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > T2: the HB response with the optional member metadata map is
> > > > > > returned.
> > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > GroupInstallAssignment
> > > > > > > > > request.
> > > > > > > > >
> > > > > > > > > In total, two round trips as well.
> > > > > > > > >
> > > > > > > > > -----------------------------
> > > > > > > > >
> > > > > > > > > So to complete the full picture here, we'd need to modify
> > both
> > > HB
> > > > > > > request
> > > > > > > > > and response so that the client can also indicate a new
> > > rebalance
> > > > > via
> > > > > > > the
> > > > > > > > > HB request as well, right?
> > > > > > > > >
> > > > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > > additional
> > > > > > > fields
> > > > > > > > > encoding the full member (subscription) metadata and topic
> > > metadata
> > > > > > as
> > > > > > > > > optional fields.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > >> Hi all,
> > > > > > > > >>
> > > > > > > > >> During an offline conversation, someone asked why we need
> > the
> > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we
> > could
> > > > > > instead
> > > > > > > > >> provide the group state in the heartbeat response. This has
> > a
> > > few
> > > > > > > > >> advantages: 1) it does not require using a special error
> > code
> > > to
> > > > > > > > >> signal that a new assignment is required as the signal would
> > > be
> > > > > the
> > > > > > > > >> provided group state; 2) it removes one round trip when a
> > > client
> > > > > > side
> > > > > > > > >> assignor is used. The downside is that it makes the
> > heartbeat
> > > > > > > > >> response's definition quite large. I recall that I went with
> > > the
> > > > > > > > >> current approach due to this.
> > > > > > > > >>
> > > > > > > > >> Providing the group state in the heartbeat response is
> > > appealing.
> > > > > > What
> > > > > > > > >> do you guys think?
> > > > > > > > >>
> > > > > > > > >> Best,
> > > > > > > > >> David
> > > > > > > > >>
> > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > djacot@confluent.io>
> > > > > > > wrote:
> > > > > > > > >> >
> > > > > > > > >> > Hi Guozhang,
> > > > > > > > >> >
> > > > > > > > >> > 1. I have added a reference to the relevant chapter
> > instead
> > > of
> > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > >> >
> > > > > > > > >> > 2. The "Rebalance Triggers" section you are referring to
> > is
> > > > > about
> > > > > > > when
> > > > > > > > >> > a rebalance should be triggered for the non-upgraded
> > members
> > > > > using
> > > > > > > the
> > > > > > > > >> > old protocol. The section mentions that a rebalance must
> > be
> > > > > > > triggered
> > > > > > > > >> > when a new assignment is installed. This implies that the
> > > group
> > > > > > > epoch
> > > > > > > > >> > was updated either by a native member or a non-upgraded
> > > member.
> > > > > > For
> > > > > > > > >> > the latter, the JoinGroup request would be the trigger. I
> > > have
> > > > > > > added a
> > > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > > Handling"
> > > > > > > section
> > > > > > > > >> > as well. Does that make sense?
> > > > > > > > >> >
> > > > > > > > >> > Thanks,
> > > > > > > > >> > David
> > > > > > > > >> >
> > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > wangguoz@gmail.com
> > > > > > >
> > > > > > > > >> wrote:
> > > > > > > > >> > >
> > > > > > > > >> > > Hello David,
> > > > > > > > >> > >
> > > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > > clear in
> > > > > > the
> > > > > > > doc,
> > > > > > > > >> > > could we update:
> > > > > > > > >> > >
> > > > > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > > > > coordinator
> > > > > > > > >> will
> > > > > > > > >> > > trigger rebalance based on the HB's member metadata /
> > > reason?
> > > > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > > > described
> > > > > > > in
> > > > > > > > >> "Group
> > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > >> > >
> > > > > > > > >> > >
> > > > > > > > >> > > Guozhang
> > > > > > > > >> > >
> > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > >> > > wrote:
> > > > > > > > >> > >
> > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > >> > > >
> > > > > > > > >> > > > I thought that the assignor will always be consulted
> > > when
> > > > > the
> > > > > > > next
> > > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > > > > heartbeat.
> > > > > > > > >> This
> > > > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > > > rebalance
> > > > > > by
> > > > > > > > >> > > > setting the reason to a non-zero value or by changing
> > > the
> > > > > > > bytes. Do
> > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > concerned by
> > > > > > the
> > > > > > > > >> delay?
> > > > > > > > >> > > >
> > > > > > > > >> > > > Best,
> > > > > > > > >> > > > David
> > > > > > > > >> > > >
> > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > wangguoz@gmail.com>
> > > > > > > > >> wrote:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > Hello David,
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > ```
> > > > > > > > >> > > > > In this case, a new assignment is triggered by the
> > > client
> > > > > > side
> > > > > > > > >> > > > > assignor. When constructing the HB, the consumer
> > will
> > > > > always
> > > > > > > > >> consult
> > > > > > > > >> > > > > the client side assignor and propagate the
> > > information to
> > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > coordinator. In other words, we don't expect users
> > to
> > > call
> > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > >> > > > > ```
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > > interface,
> > > > > we
> > > > > > > > >> actually do
> > > > > > > > >> > > > > not have a way yet to instruct how to construct the
> > > next
> > > > > HB
> > > > > > > > >> request, e.g.
> > > > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> > > with a
> > > > > > new
> > > > > > > > >> assignment,
> > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > PartitionAssignor
> > > > > > > to
> > > > > > > > >> indicate
> > > > > > > > >> > > > > the next HB telling broker about so. WDYT about
> > adding
> > > > > such
> > > > > > an
> > > > > > > > >> API on the
> > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > Guozhang
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > I have updated the KIP to include your feedback. I
> > > have
> > > > > > also
> > > > > > > > >> tried to
> > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > Best,
> > > > > > > > >> > > > > > David
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > > djacot@confluent.io
> > > > > > > > >> >
> > > > > > > > >> > > > wrote:
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > > answering
> > > > > your
> > > > > > > > >> questions
> > > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > > of
> > > > > the
> > > > > > > KIP
> > > > > > > > >> seem to
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > fewer
> > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > support of
> > > > > > > > >> wildcard. A few
> > > > > > > > >> > > > > > > > comments below.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > I would also add that the KIP removes the global
> > > sync
> > > > > > > barrier
> > > > > > > > >> in the
> > > > > > > > >> > > > > > > protocol which is essential to improve group
> > > stability
> > > > > > and
> > > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > > client by
> > > > > > > moving
> > > > > > > > >> most of
> > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > > to
> > > > > > > support
> > > > > > > > >> rolling
> > > > > > > > >> > > > > > changing
> > > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > > > assignor
> > > > > > > used
> > > > > > > > >> by a
> > > > > > > > >> > > > > > > majority of the members. This allows the group
> > to
> > > move
> > > > > > > from
> > > > > > > > >> one
> > > > > > > > >> > > > > > > assignor to another by a roll. This is explained
> > > in
> > > > > the
> > > > > > > > >> Assignor
> > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > > it's
> > > > > > > > >> required in
> > > > > > > > >> > > > every
> > > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > filled?
> > > > > > For
> > > > > > > > >> example,
> > > > > > > > >> > > > it's
> > > > > > > > >> > > > > > not
> > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > > filled.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > The client is expected to set those fields in
> > > case of
> > > > > a
> > > > > > > > >> connection
> > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > > changed
> > > > > > since
> > > > > > > > >> the last
> > > > > > > > >> > > > > > > HB. The server populates those fields as long as
> > > the
> > > > > > > member
> > > > > > > > >> is not
> > > > > > > > >> > > > > > > fully reconciled - the member should acknowledge
> > > that
> > > > > it
> > > > > > > has
> > > > > > > > >> the
> > > > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> > > this in
> > > > > > the
> > > > > > > KIP.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > > > affinity
> > > > > > > > >> between the
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > and the broker is only considered during
> > > fetching,
> > > > > but
> > > > > > > not
> > > > > > > > >> during
> > > > > > > > >> > > > > > assigning
> > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > > assignment
> > > > > > > is
> > > > > > > > >> made,
> > > > > > > > >> > > > there
> > > > > > > > >> > > > > > is
> > > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > replicas
> > > > > > of
> > > > > > > > >> assigned
> > > > > > > > >> > > > > > partitions
> > > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > > should
> > > > > > use
> > > > > > > > >> this
> > > > > > > > >> > > > > > opportunity
> > > > > > > > >> > > > > > > > to address this by including rack in
> > > GroupMember.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's an interesting idea. I don't see any
> > issue
> > > with
> > > > > > > adding
> > > > > > > > >> the
> > > > > > > > >> > > > rack
> > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > > know
> > > > > how
> > > > > > > busy
> > > > > > > > >> a group
> > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > model,
> > > it
> > > > > > seems
> > > > > > > > >> that we
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > add
> > > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > > the
> > > > > > event
> > > > > > > > >> loop is
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > actual work.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > failover
> > > > > > > > >> handling? For
> > > > > > > > >> > > > > > example,
> > > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > > with
> > > > > > the
> > > > > > > > >> wildcard
> > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sure. When the new group coordinator takes over,
> > > it
> > > > > has
> > > > > > > to:
> > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > > assignor
> > > > > is
> > > > > > > used.
> > > > > > > > >> We
> > > > > > > > >> > > > don't
> > > > > > > > >> > > > > > > store the information about the member selected
> > > to run
> > > > > > the
> > > > > > > > >> assignment
> > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > wildcard
> > > > > > > > >> subscriptions, and
> > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > document
> > > > > > what
> > > > > > > > >> the epoch
> > > > > > > > >> > > > > > field
> > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > >> > > > > > reflect
> > > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> > > latest
> > > > > > > group
> > > > > > > > >> epoch.
> > > > > > > > >> > > > > > > When the group state is updated, the group epoch
> > > is
> > > > > > > bumped so
> > > > > > > > >> we use
> > > > > > > > >> > > > > > > that one for all the change records related to
> > the
> > > > > > update.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> > the
> > > > > > > following
> > > > > > > > >> > > > invariants
> > > > > > > > >> > > > > > are
> > > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> > for
> > > a
> > > > > > member
> > > > > > > > >> not to
> > > > > > > > >> > > > get any
> > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's right. Here I meant that the members
> > > provided
> > > > > by
> > > > > > > the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > > > assignor
> > > > > > > can
> > > > > > > > >> not make
> > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > > epoch
> > > > > > > equals to
> > > > > > > > >> 0":
> > > > > > > > >> > > > When
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > > used?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > partitions and
> > > > > > > > >> rejoins when
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this
> > > case,
> > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > > > coordinator will have removed the member from
> > the
> > > > > group.
> > > > > > > The
> > > > > > > > >> member
> > > > > > > > >> > > > > > > can rejoin the group with the same member id but
> > > with
> > > > > 0
> > > > > > as
> > > > > > > > >> epoch. Let
> > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > ability
> > > to
> > > > > > > trigger a
> > > > > > > > >> > > > > > reassignment
> > > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > updating
> > > > > > the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > > with
> > > > > the
> > > > > > > > >> deprecation
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > In this case, a new assignment is triggered by
> > the
> > > > > > client
> > > > > > > side
> > > > > > > > >> > > > > > > assignor. When constructing the HB, the consumer
> > > will
> > > > > > > always
> > > > > > > > >> consult
> > > > > > > > >> > > > > > > the client side assignor and propagate the
> > > information
> > > > > > to
> > > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > > users to
> > > > > > call
> > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> > the
> > > > > > section
> > > > > > > > >> seems to
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > > immediately
> > > > > > > > >> gets into
> > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > incorrect.
> > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > > seems
> > > > > to
> > > > > > > get
> > > > > > > > >> into
> > > > > > > > >> > > > > > epoch=3,
> > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > > has A -
> > > > > > > > >> epoch=2,
> > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > support
> > > the
> > > > > > > upgrade
> > > > > > > > >> from
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > old
> > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > We will support upgrading from the consumer
> > > protocol
> > > > > > > version
> > > > > > > > >> 3,
> > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > implemented
> > > yet
> > > > > so
> > > > > > > the
> > > > > > > > >> earliest
> > > > > > > > >> > > > > > > version is unknown at the moment. This is
> > > explained in
> > > > > > the
> > > > > > > > >> migration
> > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> > > update the
> > > > > > KIP
> > > > > > > > >> based on
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > next week.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Best,
> > > > > > > > >> > > > > > > David
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > >> > > > wrote:
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > > of
> > > > > the
> > > > > > > KIP
> > > > > > > > >> seem to
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > fewer
> > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > support of
> > > > > > > > >> wildcard. A few
> > > > > > > > >> > > > > > > > comments below.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > > to
> > > > > > > support
> > > > > > > > >> rolling
> > > > > > > > >> > > > > > changing
> > > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > > it's
> > > > > > > > >> required in
> > > > > > > > >> > > > every
> > > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > filled?
> > > > > > For
> > > > > > > > >> example,
> > > > > > > > >> > > > it's
> > > > > > > > >> > > > > > not
> > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > > filled.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > > > affinity
> > > > > > > > >> between the
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > and the broker is only considered during
> > > fetching,
> > > > > but
> > > > > > > not
> > > > > > > > >> during
> > > > > > > > >> > > > > > assigning
> > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > > assignment
> > > > > > > is
> > > > > > > > >> made,
> > > > > > > > >> > > > there
> > > > > > > > >> > > > > > is
> > > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > replicas
> > > > > > of
> > > > > > > > >> assigned
> > > > > > > > >> > > > > > partitions
> > > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > > should
> > > > > > use
> > > > > > > > >> this
> > > > > > > > >> > > > > > opportunity
> > > > > > > > >> > > > > > > > to address this by including rack in
> > > GroupMember.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > > know
> > > > > how
> > > > > > > busy
> > > > > > > > >> a group
> > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > model,
> > > it
> > > > > > seems
> > > > > > > > >> that we
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > add
> > > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > > the
> > > > > > event
> > > > > > > > >> loop is
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > actual work.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > failover
> > > > > > > > >> handling? For
> > > > > > > > >> > > > > > example,
> > > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > > with
> > > > > > the
> > > > > > > > >> wildcard
> > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > document
> > > > > > what
> > > > > > > > >> the epoch
> > > > > > > > >> > > > > > field
> > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > >> > > > > > reflect
> > > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> > the
> > > > > > > following
> > > > > > > > >> > > > invariants
> > > > > > > > >> > > > > > are
> > > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> > for
> > > a
> > > > > > member
> > > > > > > > >> not to
> > > > > > > > >> > > > get any
> > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > > epoch
> > > > > > > equals to
> > > > > > > > >> 0":
> > > > > > > > >> > > > When
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > > used?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > ability
> > > to
> > > > > > > trigger a
> > > > > > > > >> > > > > > reassignment
> > > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > updating
> > > > > > the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > > with
> > > > > the
> > > > > > > > >> deprecation
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> > the
> > > > > > section
> > > > > > > > >> seems to
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > > immediately
> > > > > > > > >> gets into
> > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > incorrect.
> > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > > seems
> > > > > to
> > > > > > > get
> > > > > > > > >> into
> > > > > > > > >> > > > > > epoch=3,
> > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > > has A -
> > > > > > > > >> epoch=2,
> > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > support
> > > the
> > > > > > > upgrade
> > > > > > > > >> from
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > old
> > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Jun
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > > > > wrote:
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > > > > coordinator
> > > > > > > > >> in
> > > > > > > > >> > > > Java. I
> > > > > > > > >> > > > > > > > > discussed this offline with a few folks and
> > > folks
> > > > > > are
> > > > > > > > >> concerned
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > > > > we could introduce many regressions in the
> > old
> > > > > > > protocol
> > > > > > > > >> if we do
> > > > > > > > >> > > > so.
> > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > statement
> > > > > from
> > > > > > > the
> > > > > > > > >> KIP. It
> > > > > > > > >> > > > is an
> > > > > > > > >> > > > > > > > > implementation detail after all so it does
> > not
> > > > > have
> > > > > > > to be
> > > > > > > > >> > > > decided at
> > > > > > > > >> > > > > > > > > this stage. We will likely start by trying
> > to
> > > > > > > refactor the
> > > > > > > > >> > > > current
> > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > >> > > > > > > > > David
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot
> > <
> > > > > > > > >> djacot@confluent.io
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > "Empty,
> > > > > > > assigning,
> > > > > > > > >> > > > > > reconciling,
> > > > > > > > >> > > > > > > > > stable,
> > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > > > section,
> > > > > > > > >> right?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > This sentence does not refer to those
> > group
> > > > > states
> > > > > > > but
> > > > > > > > >> rather
> > > > > > > > >> > > > to a
> > > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > > refers to
> > > > > > the
> > > > > > > > >> entire
> > > > > > > > >> > > > state of
> > > > > > > > >> > > > > > > > > > group coordinator which is replicated via
> > > the
> > > > > log
> > > > > > > > >> layer. I will
> > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > machine
> > > is
> > > > > > > modelled
> > > > > > > > >> as an
> > > > > > > > >> > > > event
> > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> > > the new
> > > > > > > quorum
> > > > > > > > >> > > > > > controller. We
> > > > > > > > >> > > > > > > > > > will have N threads to process events.
> > Each
> > > > > > > > >> __consumer_offsets
> > > > > > > > >> > > > > > > > > > partition is assigned to a unique thread
> > > and all
> > > > > > the
> > > > > > > > >> events
> > > > > > > > >> > > > (e.g.
> > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed
> > by
> > > this
> > > > > > > > >> thread. This
> > > > > > > > >> > > > > > simplify
> > > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > > simulation
> > > > > > > testing
> > > > > > > > >> for the
> > > > > > > > >> > > > > > group
> > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > > >> *__consumer_offsets*
> > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > > owned
> > > > > > by
> > > > > > > a
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator?
> > > > > > > > >> > > > > > > > > For
> > > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > > > consumer
> > > > > > > > >> groups, and
> > > > > > > > >> > > > > > both
> > > > > > > > >> > > > > > > > > exist in
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > > state
> > > > > > > machine
> > > > > > > > >> for it,
> > > > > > > > >> > > > or
> > > > > > > > >> > > > > > 2?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > > > > think.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > "*group.coordinator.threads"
> > > > > *is
> > > > > > > the
> > > > > > > > >> number
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > threads
> > > > > > > > >> > > > > > > > > used
> > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > wondering
> > > > > if
> > > > > > > the
> > > > > > > > >> purpose
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > threads
> > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > consumer
> > > > > group
> > > > > > > (or
> > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > > why
> > > > > > > should we
> > > > > > > > >> need
> > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> > > shard
> > > > > > the
> > > > > > > > >> > > > processing as
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > > does
> > > > > the
> > > > > > > > >> default
> > > > > > > > >> > > > session
> > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > > thought
> > > > > > > the
> > > > > > > > >> min/max
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound
> > of
> > > it,
> > > > > > no?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > > 30s The
> > > > > > > > >> timeout to
> > > > > > > > >> > > > detect
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > > protocol.
> > > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > > int 45s
> > > > > > The
> > > > > > > > >> minimum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > > int 60s
> > > > > > The
> > > > > > > > >> maximum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > > session
> > > > > > > timeout
> > > > > > > > >> should
> > > > > > > > >> > > > be 45s
> > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> > are
> > > > > > [range,
> > > > > > > > >> uniform],
> > > > > > > > >> > > > > > which means
> > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > > like to
> > > > > > > know
> > > > > > > > >> why not
> > > > > > > > >> > > > > > uniform
> > > > > > > > >> > > > > > > > > one? I
> > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > uniform
> > > > > > assignor
> > > > > > > > >> (former
> > > > > > > > >> > > > sticky
> > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > other
> > > > > reason
> > > > > > > we
> > > > > > > > >> choose
> > > > > > > > >> > > > range
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > uniform
> > > > > The
> > > > > > > > >> server side
> > > > > > > > >> > > > > > assignors.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > The order on the server side has no
> > > influence
> > > > > > > because
> > > > > > > > >> the
> > > > > > > > >> > > > client
> > > > > > > > >> > > > > > must
> > > > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> > > There
> > > > > is
> > > > > > no
> > > > > > > > >> default
> > > > > > > > >> > > > in the
> > > > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > > > specified
> > > > > > > by
> > > > > > > > >> the
> > > > > > > > >> > > > client,
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > request is rejected. The default client
> > > value
> > > > > for
> > > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> > though.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Thanks for your very good comments, Luke.
> > I
> > > hope
> > > > > > > that my
> > > > > > > > >> > > > answers
> > > > > > > > >> > > > > > help
> > > > > > > > >> > > > > > > > > > to clarify things. I will update the KIP
> > as
> > > well
> > > > > > > based
> > > > > > > > >> on your
> > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > >> > > > > > > > > > David
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen
> > <
> > > > > > > > >> showuon@gmail.com>
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > > mentioned:
> > > > > > > > >> > > > > > > > > > > > The new group coordinator will have a
> > > state
> > > > > > > machine
> > > > > > > > >> per
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> > > each
> > > > > > state
> > > > > > > > >> machine is
> > > > > > > > >> > > > > > modelled
> > > > > > > > >> > > > > > > > > as an
> > > > > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > > > > executed
> > > > > > > in
> > > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > "Empty,
> > > > > > > assigning,
> > > > > > > > >> > > > > > reconciling,
> > > > > > > > >> > > > > > > > > stable,
> > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > > > section,
> > > > > > > > >> right?
> > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > machine
> > > is
> > > > > > > modelled
> > > > > > > > >> as an
> > > > > > > > >> > > > event
> > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > > >> *__consumer_offsets*
> > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > > owned
> > > > > > by
> > > > > > > a
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator?
> > > > > > > > >> > > > > > > > > For
> > > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > > > consumer
> > > > > > > > >> groups, and
> > > > > > > > >> > > > > > both
> > > > > > > > >> > > > > > > > > exist in
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > > state
> > > > > > > machine
> > > > > > > > >> for it,
> > > > > > > > >> > > > or
> > > > > > > > >> > > > > > 2?
> > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > "*group.coordinator.threads"
> > > > > *is
> > > > > > > the
> > > > > > > > >> number
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > threads
> > > > > > > > >> > > > > > > > > used
> > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > wondering
> > > > > if
> > > > > > > the
> > > > > > > > >> purpose
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > threads
> > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > consumer
> > > > > group
> > > > > > > (or
> > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > > why
> > > > > > > should we
> > > > > > > > >> need
> > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > > configs:
> > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > > does
> > > > > the
> > > > > > > > >> default
> > > > > > > > >> > > > session
> > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > > thought
> > > > > > > the
> > > > > > > > >> min/max
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound
> > of
> > > it,
> > > > > > no?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > > 30s The
> > > > > > > > >> timeout to
> > > > > > > > >> > > > detect
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > > protocol.
> > > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > > int 45s
> > > > > > The
> > > > > > > > >> minimum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > > int 60s
> > > > > > The
> > > > > > > > >> maximum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> > are
> > > > > > [range,
> > > > > > > > >> uniform],
> > > > > > > > >> > > > > > which means
> > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > > like to
> > > > > > > know
> > > > > > > > >> why not
> > > > > > > > >> > > > > > uniform
> > > > > > > > >> > > > > > > > > one? I
> > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > uniform
> > > > > > assignor
> > > > > > > > >> (former
> > > > > > > > >> > > > sticky
> > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > other
> > > > > reason
> > > > > > > we
> > > > > > > > >> choose
> > > > > > > > >> > > > range
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > uniform
> > > > > The
> > > > > > > > >> server side
> > > > > > > > >> > > > > > assignors.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke
> > Chen
> > > <
> > > > > > > > >> showuon@gmail.com
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> > Connect
> > > > > > > > >> integrating with
> > > > > > > > >> > > > > > KIP-848,
> > > > > > > > >> > > > > > > > > but I
> > > > > > > > >> > > > > > > > > > > > think we should have a separate
> > > discussion
> > > > > > > thread
> > > > > > > > >> for the
> > > > > > > > >> > > > Kafka
> > > > > > > > >> > > > > > > > > Connect
> > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With
> > New
> > > > > > Consumer
> > > > > > > > >> Rebalance
> > > > > > > > >> > > > > > Protocol
> > > > > > > > >> > > > > > > > > [1],
> > > > > > > > >> > > > > > > > > > > > and let this discussion thread focus
> > on
> > > > > > consumer
> > > > > > > > >> rebalance
> > > > > > > > >> > > > > > protocol,
> > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > > >
> > > > >
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar
> > <
> > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > > feedback.
> > > > > > > Looks
> > > > > > > > >> like
> > > > > > > > >> > > > there's
> > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> > > would
> > > > > > > revisit
> > > > > > > > >> the doc
> > > > > > > > >> > > > and
> > > > > > > > >> > > > > > see
> > > > > > > > >> > > > > > > > > what
> > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David
> > > Jacot
> > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > > document.
> > > > > > > That's
> > > > > > > > >> really
> > > > > > > > >> > > > > > helpful. I
> > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> > > Connect
> > > > > > > and the
> > > > > > > > >> > > > Consumer
> > > > > > > > >> > > > > > could
> > > > > > > > >> > > > > > > > > share
> > > > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> > > main
> > > > > > > > >> difference is
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > > > > whereas
> > > > > > > > >> Connect
> > > > > > > > >> > > > assigns
> > > > > > > > >> > > > > > tasks
> > > > > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > > forward:
> > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs
> > to
> > > > > > support
> > > > > > > > >> different
> > > > > > > > >> > > > > > resource
> > > > > > > > >> > > > > > > > > types
> > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > > Connect.
> > > > > > The
> > > > > > > > >> dedicated
> > > > > > > > >> > > > APIs
> > > > > > > > >> > > > > > > > > would be
> > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > different on
> > > > > > the
> > > > > > > > >> > > > > > content/resources and
> > > > > > > > >> > > > > > > > > > > >> > they would rely on the same engine
> > > on the
> > > > > > > > >> coordinator
> > > > > > > > >> > > > side.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> > because
> > > I am
> > > > > > > not a
> > > > > > > > >> fan of
> > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> > > That
> > > > > > being
> > > > > > > > >> said, I am
> > > > > > > > >> > > > not
> > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to
> > > do
> > > > > it.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > > discuss
> > > > > it
> > > > > > > here
> > > > > > > > >> for now
> > > > > > > > >> > > > in
> > > > > > > > >> > > > > > order
> > > > > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible
> > > with
> > > > > > what
> > > > > > > we
> > > > > > > > >> will do
> > > > > > > > >> > > > for
> > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM
> > David
> > > > > Jacot
> > > > > > <
> > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will
> > go
> > > > > > through
> > > > > > > and
> > > > > > > > >> address
> > > > > > > > >> > > > > > your
> > > > > > > > >> > > > > > > > > comments
> > > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for
> > your
> > > > > > > feedback.
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> > > Gregory
> > > > > > > Harris
> > > > > > > > >> <
> > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > > wonderful to
> > > > > > see
> > > > > > > > >> > > > cooperative
> > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > > > > section
> > > > > > > > >> describes how
> > > > > > > > >> > > > > > member
> > > > > > > > >> > > > > > > > > epoch
> > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > >> > > > > > > > > > > >> > > > advance in step with the group
> > > epoch
> > > > > > and
> > > > > > > > >> assignment
> > > > > > > > >> > > > > > epoch
> > > > > > > > >> > > > > > > > > values. I
> > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for
> > the
> > > > > > reasons
> > > > > > > > >> described
> > > > > > > > >> > > > in
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > > assigning
> > > > > > > > >> partitions to a
> > > > > > > > >> > > > > > worker,
> > > > > > > > >> > > > > > > > > what
> > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > > assignment
> > > > > > > use?
> > > > > > > > >> Are
> > > > > > > > >> > > > member
> > > > > > > > >> > > > > > epochs
> > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > > correspond
> > > > > to
> > > > > > > > >> multiple
> > > > > > > > >> > > > > > different
> > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> > > field
> > > > > > > opaque
> > > > > > > > >> to the
> > > > > > > > >> > > > group
> > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > > > > assignor
> > > > > > > > >> > > > implementations
> > > > > > > > >> > > > > > > > > interact
> > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> > > common
> > > > > > > meaning
> > > > > > > > >> agreed
> > > > > > > > >> > > > > > upon? If
> > > > > > > > >> > > > > > > > > so, what
> > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> > > field
> > > > > over
> > > > > > > > >> including
> > > > > > > > >> > > > such
> > > > > > > > >> > > > > > > > > functionality
> > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is included in
> > > the
> > > > > > KIP:
> > > > > > > > >> "Thanks to
> > > > > > > > >> > > > > > this, the
> > > > > > > > >> > > > > > > > > input
> > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > entirely
> > > > > driven
> > > > > > > by
> > > > > > > > >> the group
> > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > responsible for
> > > > > > > > >> maintaining
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > state
> > > > > > > > >> > > > > > > > > besides
> > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this
> > > mean
> > > > > > > that the
> > > > > > > > >> > > > > > client-side
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > non-Metadata
> > > > > > state
> > > > > > > > >> (such as
> > > > > > > > >> > > > > > partition
> > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > > topics,
> > > > > > etc),
> > > > > > > or
> > > > > > > > >> that
> > > > > > > > >> > > > > > additional
> > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > separate
> > > > > > classes
> > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > >> > > > > >
> > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > >> > > > > > that
> > > > > > > > >> > > > > > > > > seem to
> > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > > possible
> > > > > > for
> > > > > > > > >> these two
> > > > > > > > >> > > > > > > > > implementations
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> > > promote
> > > > > > > feature
> > > > > > > > >> parity
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > > server-side
> > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and
> > would
> > > also
> > > > > > > > >> facilitate
> > > > > > > > >> > > > > > operational
> > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > example,
> > > if a
> > > > > > > > >> server-side
> > > > > > > > >> > > > > > assignor
> > > > > > > > >> > > > > > > > > has some
> > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > deploying
> > > > > > the
> > > > > > > > >> patched
> > > > > > > > >> > > > > > assignor to
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer
> > group
> > > to a
> > > > > > > > >> client-side
> > > > > > > > >> > > > > > assignor
> > > > > > > > >> > > > > > > > > may be
> > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > >> > > > > > > > > > > >> > > > and less risky than patching
> > all
> > > of
> > > > > the
> > > > > > > > >> brokers.
> > > > > > > > >> > > > With
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > currently
> > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > > non-trivial
> > > > > > > > >> > > > reimplementation
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > have
> > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs
> > > have
> > > > > > > diverged
> > > > > > > > >> > > > > > significantly,
> > > > > > > > >> > > > > > > > > then it
> > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > reimplementation
> > > > > would
> > > > > > > not be
> > > > > > > > >> > > > possible.
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM
> > > Sagar
> > > > > <
> > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page
> > to
> > > > > > discuss
> > > > > > > how
> > > > > > > > >> Connect
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > need to
> > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > > > protocol.
> > > > > > > > >> Here's the
> > > > > > > > >> > > > page:
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > > >
> > > > >
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and
> > I
> > > have
> > > > > > > tried
> > > > > > > > >> to keep
> > > > > > > > >> > > > a
> > > > > > > > >> > > > > > format
> > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > > > > think.
> > > > > > > Also,
> > > > > > > > >> do you
> > > > > > > > >> > > > > > think this
> > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > discussion
> > > > > thread
> > > > > > > or
> > > > > > > > >> is this
> > > > > > > > >> > > > one
> > > > > > > > >> > > > > > fine?
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37
> > AM
> > > > > > Sagar <
> > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> > > doc on
> > > > > > > Kafka
> > > > > > > > >> Streams.
> > > > > > > > >> > > > > > Sure, I
> > > > > > > > >> > > > > > > > > would do
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> > > such a
> > > > > > > > >> document.
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at
> > 4:47
> > > AM
> > > > > > > Guozhang
> > > > > > > > >> Wang <
> > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> > > could
> > > > > > come
> > > > > > > > >> back with
> > > > > > > > >> > > > some
> > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > > > > integration
> > > > > > > > >> with
> > > > > > > > >> > > > the new
> > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the
> > > new
> > > > > > > "protocol
> > > > > > > > >> type"
> > > > > > > > >> > > > we
> > > > > > > > >> > > > > > did not
> > > > > > > > >> > > > > > > > > yet
> > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > implementations.
> > > > > > > For
> > > > > > > > >> Streams
> > > > > > > > >> > > > > > here's a
> > > > > > > > >> > > > > > > > > draft of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > > >
> > > > >
> > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis
> > > on
> > > > > > > Connect.
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at
> > > 10:48 PM
> > > > > > > Sagar <
> > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > > response.
> > > > > The
> > > > > > > > >> reason I
> > > > > > > > >> > > > thought
> > > > > > > > >> > > > > > > > > connect can
> > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that
> > > even
> > > > > > > today the
> > > > > > > > >> > > > connect
> > > > > > > > >> > > > > > uses a
> > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > AbstractCoordinator to
> > > > > > > > >> empower
> > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > > > > > >> WorkerCoordinator sets
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata()
> > > method
> > > > > > by
> > > > > > > > >> plumbing
> > > > > > > > >> > > > into
> > > > > > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to
> > > support
> > > > > > > > >> connect would
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams
> > > mainly
> > > > > > > because
> > > > > > > > >> of the
> > > > > > > > >> > > > > > Client
> > > > > > > > >> > > > > > > > > side
> > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > > > implementation
> > > > > > > > >> level, we
> > > > > > > > >> > > > > > might
> > > > > > > > >> > > > > > > > > need to
> > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this
> > > new
> > > > > > > > >> assignment
> > > > > > > > >> > > > protocol
> > > > > > > > >> > > > > > like
> > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > request/response and
> > > > > > > > >> SyncGroup and
> > > > > > > > >> > > > > > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar
> > > lines
> > > > > > to
> > > > > > > > >> streams (or
> > > > > > > > >> > > > > > there
> > > > > > > > >> > > > > > > > > might be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> > > > > detailed
> > > > > > > > >> analysis of
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > same
> > > > > > > > >> > > > > > > > > and we
> > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> > > thread for
> > > > > > > that
> > > > > > > > >> as that
> > > > > > > > >> > > > would
> > > > > > > > >> > > > > > > > > derail this
> > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if
> > > that
> > > > > > sounds
> > > > > > > > >> good to
> > > > > > > > >> > > > you.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at
> > > 5:47
> > > > > PM
> > > > > > > David
> > > > > > > > >> Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > > comments.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > > > > > >> `Assignment#error`.
> > > > > > > > >> > > > > > Sure, I
> > > > > > > > >> > > > > > > > > can
> > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > > > transition C
> > > > > > > from
> > > > > > > > >> his
> > > > > > > > >> > > > current
> > > > > > > > >> > > > > > > > > assignment
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when
> > > he
> > > > > can
> > > > > > > move
> > > > > > > > >> to
> > > > > > > > >> > > > epoch 3.
> > > > > > > > >> > > > > > When
> > > > > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is
> > > updated
> > > > > > and
> > > > > > > > >> persisted
> > > > > > > > >> > > > > > with all
> > > > > > > > >> > > > > > > > > its
> > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if
> > they
> > > are
> > > > > > not
> > > > > > > all
> > > > > > > > >> revoked
> > > > > > > > >> > > > > > yet. In
> > > > > > > > >> > > > > > > > > other
> > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > > becomes
> > > > > the
> > > > > > > target
> > > > > > > > >> > > > > > assignment.
> > > > > > > > >> > > > > > > > > This is
> > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid
> > > having
> > > > > > to
> > > > > > > > >> write all
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > changes to
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on
> > > the
> > > > > > > persisted
> > > > > > > > >> state
> > > > > > > > >> > > > so I
> > > > > > > > >> > > > > > > > > understand
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see
> > > if I
> > > > > can
> > > > > > > > >> improve
> > > > > > > > >> > > > this in
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect,
> > it
> > > > > could
> > > > > > > > >> reuse the
> > > > > > > > >> > > > > > protocol
> > > > > > > > >> > > > > > > > > with a
> > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in
> > > the
> > > > > > > > >> protocol. The
> > > > > > > > >> > > > > > assignment
> > > > > > > > >> > > > > > > > > is
> > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> > > > > metadata,
> > > > > > > could
> > > > > > > > >> > > > Connect
> > > > > > > > >> > > > > > fit
> > > > > > > > >> > > > > > > > > into this?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022
> > at
> > > 1:55
> > > > > > PM
> > > > > > > > >> Sagar <
> > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP.
> > I
> > > just
> > > > > > had
> > > > > > > > >> minor
> > > > > > > > >> > > > > > observations:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment
> > > Error
> > > > > > > > >> section in
> > > > > > > > >> > > > Client
> > > > > > > > >> > > > > > Side
> > > > > > > > >> > > > > > > > > mode
> > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In
> > > this
> > > > > > > case,
> > > > > > > > >> the
> > > > > > > > >> > > > client
> > > > > > > > >> > > > > > side
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > > > > > coordinator`.
> > > > > > > > >> In this
> > > > > > > > >> > > > > > case are
> > > > > > > > >> > > > > > > > > you
> > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning
> > an
> > > > > > > > >> AssignmentError
> > > > > > > > >> > > > that's
> > > > > > > > >> > > > > > > > > listed down
> > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think
> > > it
> > > > > > would
> > > > > > > > >> make sense
> > > > > > > > >> > > > to
> > > > > > > > >> > > > > > > > > mention this
> > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case
> > Studies
> > > > > > > section, I
> > > > > > > > >> have a
> > > > > > > > >> > > > > > slight
> > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the
> > same.
> > > > > > Consider
> > > > > > > > >> this step:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats,
> > the
> > > > > group
> > > > > > > > >> coordinator
> > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions
> > to
> > > > > > > revoke. It
> > > > > > > > >> > > > persists
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > change and
> > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > Assignment
> > > > > > > (epoch=3)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > partitions=[foo-0]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > partitions=[foo-2]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > partitions=[foo-1]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > Assignment
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > > > > > >> partitions=[foo-0,
> > > > > > > > >> > > > > > foo-1]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > > > > > >> partitions=[foo-2]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > > > > > >> partitions=[foo-1]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats,
> > it
> > > > > > > transitions
> > > > > > > > >> to
> > > > > > > > >> > > > epoch 3
> > > > > > > > >> > > > > > but
> > > > > > > > >> > > > > > > > > cannot
> > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned
> > > that
> > > > > > > member C
> > > > > > > > >> can't
> > > > > > > > >> > > > get
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > description
> > > > > > > above,
> > > > > > > > >> it seems
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > > > already has
> > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to
> > > remove
> > > > > it
> > > > > > > and
> > > > > > > > >> > > > populate it
> > > > > > > > >> > > > > > only
> > > > > > > > >> > > > > > > > > when it
> > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot
> > of
> > > > > other
> > > > > > > > >> places, so
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > I
> > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect ,
> > it
> > > > > might
> > > > > > > be
> > > > > > > > >> out of
> > > > > > > > >> > > > scope
> > > > > > > > >> > > > > > of
> > > > > > > > >> > > > > > > > > this
> > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it
> > > would
> > > > > > > > >> probably be
> > > > > > > > >> > > > > > running in
> > > > > > > > >> > > > > > > > > client
> > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> > > rebalance
> > > > > > > > >> protocol as it
> > > > > > > > >> > > > has
> > > > > > > > >> > > > > > its
> > > > > > > > >> > > > > > > > > own
> > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > IncrementalCooperative).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022
> > > at
> > > > > 5:00
> > > > > > > PM
> > > > > > > > >> David
> > > > > > > > >> > > > Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our
> > > goal
> > > > > is
> > > > > > > to
> > > > > > > > >> move
> > > > > > > > >> > > > forward
> > > > > > > > >> > > > > > with
> > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying
> > > on
> > > > > one
> > > > > > > > >> generic API.
> > > > > > > > >> > > > For
> > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern
> > > and
> > > > > > > > >> reuse/share the
> > > > > > > > >> > > > core
> > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For
> > the
> > > > > > schema
> > > > > > > > >> registry, I
> > > > > > > > >> > > > > > think
> > > > > > > > >> > > > > > > > > that we
> > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored
> > > API to
> > > > > > do
> > > > > > > > >> simple
> > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15,
> > > 2022 at
> > > > > > > 10:22
> > > > > > > > >> AM Ismael
> > > > > > > > >> > > > > > Juma <
> > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> > > comments:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > > > >> java.util.regex.Pattern
> > > > > > > > >> > > > vs
> > > > > > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern,
> > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > > > differences
> > > > > > > in
> > > > > > > > >> more
> > > > > > > > >> > > > detail
> > > > > > > > >> > > > > > before
> > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if
> > > people
> > > > > > pass
> > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be
> > > > > honored.
> > > > > > > If
> > > > > > > > >> we are
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > > > > something
> > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider
> > > adding
> > > > > > an
> > > > > > > > >> overload
> > > > > > > > >> > > > with
> > > > > > > > >> > > > > > our own
> > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want
> > to
> > > > > > expose
> > > > > > > > >> re2j's at
> > > > > > > > >> > > > this
> > > > > > > > >> > > > > > > > > point).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding
> > topic
> > > > > ids,
> > > > > > > any
> > > > > > > > >> major new
> > > > > > > > >> > > > > > protocol
> > > > > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should
> > > handle
> > > > > the
> > > > > > > topic
> > > > > > > > >> > > > > > recreation case
> > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to
> > > > > handle.
> > > > > > I
> > > > > > > > >> agree with
> > > > > > > > >> > > > > > David
> > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > > > > > protocols
> > > > > > > > >> that don't
> > > > > > > > >> > > > > > have it
> > > > > > > > >> > > > > > > > > yet and
> > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> > > internals
> > > > > > > versus
> > > > > > > > >> adding
> > > > > > > > >> > > > new
> > > > > > > > >> > > > > > APIs
> > > > > > > > >> > > > > > > > > to the
> > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> > > adding
> > > > > new
> > > > > > > APIs
> > > > > > > > >> is
> > > > > > > > >> > > > required
> > > > > > > > >> > > > > > for
> > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still
> > not
> > > sure
> > > > > > > about
> > > > > > > > >> the
> > > > > > > > >> > > > > > coordinator
> > > > > > > > >> > > > > > > > > storing
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> > > configs
> > > > > to
> > > > > > be
> > > > > > > > >> > > > centralized
> > > > > > > > >> > > > > > in the
> > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > (auditability,
> > > > > > > > >> visibility,
> > > > > > > > >> > > > > > consistency,
> > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > > automatically
> > > > > > > > >> deleting
> > > > > > > > >> > > > > > configs in a
> > > > > > > > >> > > > > > > > > way
> > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A
> > good
> > > > > > property
> > > > > > > > >> for modern
> > > > > > > > >> > > > > > systems
> > > > > > > > >> > > > > > > > > is to
> > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable
> > data
> > > > > loss
> > > > > > > > >> scenarios.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13,
> > > 2022
> > > > > at
> > > > > > > 3:47
> > > > > > > > >> PM David
> > > > > > > > >> > > > > > Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > Guozhang. My
> > > > > > > answers
> > > > > > > > >> are
> > > > > > > > >> > > > below:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > migration
> > > > > > path,
> > > > > > > > >> especially
> > > > > > > > >> > > > > > the last
> > > > > > > > >> > > > > > > > > step
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable
> > the
> > > new
> > > > > > > > >> protocol, in
> > > > > > > > >> > > > which
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> > > rpcs
> > > > > and
> > > > > > > old
> > > > > > > > >> > > > protocols /
> > > > > > > > >> > > > > > rpcs
> > > > > > > > >> > > > > > > > > are
> > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How
> > the
> > > > > > > > >> coordinator could
> > > > > > > > >> > > > > > "mimic"
> > > > > > > > >> > > > > > > > > the old
> > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> > > is
> > > > > > > something
> > > > > > > > >> we
> > > > > > > > >> > > > need to
> > > > > > > > >> > > > > > > > > present
> > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > > > > > published
> > > > > > > a
> > > > > > > > >> new
> > > > > > > > >> > > > version
> > > > > > > > >> > > > > > of KIP
> > > > > > > > >> > > > > > > > > which
> > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about
> > > this.
> > > > > > See
> > > > > > > the
> > > > > > > > >> > > > "Supporting
> > > > > > > > >> > > > > > > > > Online
> > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > > "Compatibility,
> > > > > > > > >> > > > Deprecation,
> > > > > > > > >> > > > > > and
> > > > > > > > >> > > > > > > > > Migration
> > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to
> > think
> > > > > > through
> > > > > > > a
> > > > > > > > >> few cases
> > > > > > > > >> > > > > > now but
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> > > should be
> > > > > > > > >> understandable.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> > > of
> > > > > > topic
> > > > > > > > >> ids. So
> > > > > > > > >> > > > far as
> > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> > > and
> > > > > > admin
> > > > > > > > >> client,
> > > > > > > > >> > > > but
> > > > > > > > >> > > > > > they
> > > > > > > > >> > > > > > > > > are not
> > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> > > yet.
> > > > > I
> > > > > > > think
> > > > > > > > >> the
> > > > > > > > >> > > > > > question is,
> > > > > > > > >> > > > > > > > > first
> > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > > > maintaining
> > > > > > > > >> the
> > > > > > > > >> > > > names
> > > > > > > > >> > > > > > -> ids
> > > > > > > > >> > > > > > > > > mapping
> > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > > augmented
> > > > > > > > >> existing
> > > > > > > > >> > > > RPCs
> > > > > > > > >> > > > > > and the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > > should
> > > > > we
> > > > > > > ever
> > > > > > > > >> consider
> > > > > > > > >> > > > > > exposing
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs
> > as
> > > > > well
> > > > > > > (both
> > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> > > cases
> > > > > > > like
> > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming
> > > that we
> > > > > > > would
> > > > > > > > >> include
> > > > > > > > >> > > > > > converting
> > > > > > > > >> > > > > > > > > all
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using
> > > topic
> > > > > > ids
> > > > > > > in
> > > > > > > > >> this
> > > > > > > > >> > > > KIP,
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate
> > > with
> > > > > > topic
> > > > > > > > >> ids. That
> > > > > > > > >> > > > > > being
> > > > > > > > >> > > > > > > > > said, it
> > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > names in
> > > > > > > various
> > > > > > > > >> APIs so
> > > > > > > > >> > > > > > having a
> > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > inevitable to
> > > > > > me.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't
> > have
> > > a
> > > > > > strong
> > > > > > > > >> opinion on
> > > > > > > > >> > > > > > this.
> > > > > > > > >> > > > > > > > > Here I
> > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> > > scope of
> > > > > > this
> > > > > > > > >> KIP. I
> > > > > > > > >> > > > would
> > > > > > > > >> > > > > > rather
> > > > > > > > >> > > > > > > > > focus
> > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we
> > can
> > > > > > consider
> > > > > > > > >> this
> > > > > > > > >> > > > > > separately if
> > > > > > > > >> > > > > > > > > we see
> > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > > > > > Ismael's
> > > > > > > > >> point about
> > > > > > > > >> > > > > > using
> > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > >> > > > think
> > > > > > > > >> > > > > > that
> > > > > > > > >> > > > > > > > > there
> > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it.
> > > The
> > > > > > > consumer
> > > > > > > > >> will
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > exists
> > > > > > before
> > > > > > > > >> passing
> > > > > > > > >> > > > them
> > > > > > > > >> > > > > > into
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> > > Obviously,
> > > > > > the
> > > > > > > > >> > > > coordinator
> > > > > > > > >> > > > > > will
> > > > > > > > >> > > > > > > > > also
> > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really
> > have
> > > a
> > > > > way
> > > > > > to
> > > > > > > > >> reject an
> > > > > > > > >> > > > > > invalid
> > > > > > > > >> > > > > > > > > topic in
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> > > with
> > > > > > > David
> > > > > > > > >> on all
> > > > > > > > >> > > > other
> > > > > > > > >> > > > > > minor
> > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > `subscribe(Pattern)`
> > > > > > > > >> question:
> > > > > > > > >> > > > > > personally
> > > > > > > > >> > > > > > > > > I think
> > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate
> > the
> > > > > > > subscribe
> > > > > > > > >> API with
> > > > > > > > >> > > > > > Pattern,
> > > > > > > > >> > > > > > > > > but
> > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > > documenting
> > > > > > > > >> that our
> > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > > incompatible
> > > > > > > > >> case is a
> > > > > > > > >> > > > > > very rare
> > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > > `String`
> > > > > > > based
> > > > > > > > >> > > > > > subscription may
> > > > > > > > >> > > > > > > > > be
> > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > regexes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> > > work. I
> > > > > > have
> > > > > > > to
> > > > > > > > >> look at
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > differences
> > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> > > better
> > > > > > > > >> understand the
> > > > > > > > >> > > > > > potential
> > > > > > > > >> > > > > > > > > issues.
> > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would
> > work
> > > for
> > > > > > all
> > > > > > > the
> > > > > > > > >> basic
> > > > > > > > >> > > > > > regular
> > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the
> > two
> > > are
> > > > > > > mainly
> > > > > > > > >> about
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > various
> > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what
> > > other
> > > > > > people
> > > > > > > > >> think
> > > > > > > > >> > > > about
> > > > > > > > >> > > > > > this.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul
> > 12,
> > > 2022
> > > > > > at
> > > > > > > > >> 11:28 PM
> > > > > > > > >> > > > > > Guozhang
> > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks
> > David!
> > > I
> > > > > > think
> > > > > > > on
> > > > > > > > >> the high
> > > > > > > > >> > > > > > level
> > > > > > > > >> > > > > > > > > there are
> > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > concretize
> > > a
> > > > > bit
> > > > > > > more:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > migration
> > > > > > path,
> > > > > > > > >> especially
> > > > > > > > >> > > > > > the last
> > > > > > > > >> > > > > > > > > step
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable
> > the
> > > new
> > > > > > > > >> protocol, in
> > > > > > > > >> > > > which
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> > > rpcs
> > > > > and
> > > > > > > old
> > > > > > > > >> > > > protocols /
> > > > > > > > >> > > > > > rpcs
> > > > > > > > >> > > > > > > > > are
> > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How
> > the
> > > > > > > > >> coordinator could
> > > > > > > > >> > > > > > "mimic"
> > > > > > > > >> > > > > > > > > the old
> > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> > > is
> > > > > > > something
> > > > > > > > >> we
> > > > > > > > >> > > > need to
> > > > > > > > >> > > > > > > > > present
> > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> > > of
> > > > > > topic
> > > > > > > > >> ids. So
> > > > > > > > >> > > > far as
> > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> > > and
> > > > > > admin
> > > > > > > > >> client,
> > > > > > > > >> > > > but
> > > > > > > > >> > > > > > they
> > > > > > > > >> > > > > > > > > are not
> > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> > > yet.
> > > > > I
> > > > > > > think
> > > > > > > > >> the
> > > > > > > > >> > > > > > question is,
> > > > > > > > >> > > > > > > > > first
> > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > > > maintaining
> > > > > > > > >> the
> > > > > > > > >> > > > names
> > > > > > > > >> > > > > > -> ids
> > > > > > > > >> > > > > > > > > mapping
> > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > > augmented
> > > > > > > > >> existing
> > > > > > > > >> > > > RPCs
> > > > > > > > >> > > > > > and the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > > should
> > > > > we
> > > > > > > ever
> > > > > > > > >> consider
> > > > > > > > >> > > > > > exposing
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs
> > as
> > > > > well
> > > > > > > (both
> > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> > > cases
> > > > > > > like
> > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> > > with
> > > > > > > David
> > > > > > > > >> on all
> > > > > > > > >> > > > other
> > > > > > > > >> > > > > > minor
> > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > `subscribe(Pattern)`
> > > > > > > > >> question:
> > > > > > > > >> > > > > > personally
> > > > > > > > >> > > > > > > > > I think
> > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate
> > the
> > > > > > > subscribe
> > > > > > > > >> API with
> > > > > > > > >> > > > > > Pattern,
> > > > > > > > >> > > > > > > > > but
> > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > > documenting
> > > > > > > > >> that our
> > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > > incompatible
> > > > > > > > >> case is a
> > > > > > > > >> > > > > > very rare
> > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > > `String`
> > > > > > > based
> > > > > > > > >> > > > > > subscription may
> > > > > > > > >> > > > > > > > > be
> > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > regexes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul
> > > 12,
> > > > > 2022
> > > > > > > at
> > > > > > > > >> 5:23 AM
> > > > > > > > >> > > > > > David Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for
> > > your
> > > > > > > > >> feedback. Let
> > > > > > > > >> > > > me
> > > > > > > > >> > > > > > answer
> > > > > > > > >> > > > > > > > > your
> > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I
> > think
> > > > > it's
> > > > > > > > >> premature to
> > > > > > > > >> > > > > > talk about
> > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal
> > > of the
> > > > > > > > >> existing group
> > > > > > > > >> > > > > > protocol.
> > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> > > > > protocol
> > > > > > > and
> > > > > > > > >> hence
> > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> > > quickly
> > > > > > > > >> applications
> > > > > > > > >> > > > migrate
> > > > > > > > >> > > > > > to
> > > > > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes
> > > > > sense.
> > > > > > I
> > > > > > > will
> > > > > > > > >> > > > remove it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The
> > KIP
> > > > > says
> > > > > > we
> > > > > > > > >> intend to
> > > > > > > > >> > > > > > release
> > > > > > > > >> > > > > > > > > this in
> > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If
> > we
> > > > > added
> > > > > > > that
> > > > > > > > >> as a
> > > > > > > > >> > > > way to
> > > > > > > > >> > > > > > > > > estimate
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > group
> > > > > > > protocol, I
> > > > > > > > >> also
> > > > > > > > >> > > > > > suggest
> > > > > > > > >> > > > > > > > > removing
> > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > explain
> > > > > my
> > > > > > > > >> reasoning. As
> > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > coordinator
> > > in
> > > > > > Java
> > > > > > > > >> while we
> > > > > > > > >> > > > > > implement
> > > > > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> > > > > internals
> > > > > > > will
> > > > > > > > >> be
> > > > > > > > >> > > > slightly
> > > > > > > > >> > > > > > > > > different
> > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Therefore, I
> > > > > > wanted
> > > > > > > to
> > > > > > > > >> tighten
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > switch from
> > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > coordinator
> > > to
> > > > > the
> > > > > > > new
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator
> > > > > > > > >> > > > > > > > > to a
> > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > alternative
> > > > > would
> > > > > > > be to
> > > > > > > > >> use a
> > > > > > > > >> > > > flag
> > > > > > > > >> > > > > > to do
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> > > software
> > > > > > > upgrade.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We
> > > need to
> > > > > > > flesh
> > > > > > > > >> out the
> > > > > > > > >> > > > > > details of
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're
> > > saying
> > > > > we
> > > > > > > will
> > > > > > > > >> support
> > > > > > > > >> > > > > > online
> > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain
> > > this
> > > > > in
> > > > > > > > >> detail. It
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > also
> > > > > > > > >> > > > > > > > > be done
> > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we
> > will
> > > > > > support
> > > > > > > > >> online
> > > > > > > > >> > > > > > migrations
> > > > > > > > >> > > > > > > > > for the
> > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group
> > > using
> > > > > the
> > > > > > > old
> > > > > > > > >> protocol
> > > > > > > > >> > > > > > will be
> > > > > > > > >> > > > > > > > > able to
> > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > briefly
> > > > > > > explain
> > > > > > > > >> how that
> > > > > > > > >> > > > > > will work
> > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step
> > > > > process:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> > > cluster
> > > > > > must
> > > > > > > be
> > > > > > > > >> > > > upgraded or
> > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > > > > > coordinator.
> > > > > > > > >> Both the
> > > > > > > > >> > > > > > old and
> > > > > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support
> > the
> > > old
> > > > > > > > >> protocol and
> > > > > > > > >> > > > rely
> > > > > > > > >> > > > > > on the
> > > > > > > > >> > > > > > > > > same
> > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can
> > > work
> > > > > > > together.
> > > > > > > > >> This
> > > > > > > > >> > > > point
> > > > > > > > >> > > > > > is an
> > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this
> > one
> > > live
> > > > > > > > >> because it
> > > > > > > > >> > > > would
> > > > > > > > >> > > > > > require
> > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > coordinator
> > > and
> > > > > > > > >> starting up the
> > > > > > > > >> > > > > > new one
> > > > > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > unavailabilities.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> > > cluster's
> > > > > > > > >> metadata
> > > > > > > > >> > > > > > version/IBP
> > > > > > > > >> > > > > > > > > must be
> > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable
> > > the
> > > > > new
> > > > > > > > >> protocol.
> > > > > > > > >> > > > This
> > > > > > > > >> > > > > > cannot
> > > > > > > > >> > > > > > > > > be done
> > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated
> > > > > because
> > > > > > > the
> > > > > > > > >> old
> > > > > > > > >> > > > > > coordinator
> > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> > > consumers
> > > > > > > must be
> > > > > > > > >> > > > upgraded
> > > > > > > > >> > > > > > to a
> > > > > > > > >> > > > > > > > > version
> > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> > > (must
> > > > > > have
> > > > > > > > >> KIP-792).
> > > > > > > > >> > > > If
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing
> > > must be
> > > > > > > done at
> > > > > > > > >> this
> > > > > > > > >> > > > point.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> > > consumers
> > > > > > > must
> > > > > > > > >> be rolled
> > > > > > > > >> > > > > > with the
> > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer
> > > group
> > > > > is
> > > > > > > > >> automatically
> > > > > > > > >> > > > > > > > > converted when
> > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the
> > > new
> > > > > > > protocol
> > > > > > > > >> joins
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > group.
> > > > > > > > >> > > > > > > > > While
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > > protocol are
> > > > > > > being
> > > > > > > > >> > > > upgraded,
> > > > > > > > >> > > > > > the old
> > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > clarify
> > > > > all
> > > > > > > of
> > > > > > > > >> this in
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > KIP.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am
> > > happy
> > > > > > > that we
> > > > > > > > >> are
> > > > > > > > >> > > > > > pushing the
> > > > > > > > >> > > > > > > > > pattern
> > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it
> > > seems
> > > > > > like
> > > > > > > > >> there
> > > > > > > > >> > > > could be
> > > > > > > > >> > > > > > some
> > > > > > > > >> > > > > > > > > tricky
> > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> > > > > mechanism
> > > > > > > for
> > > > > > > > >> users to
> > > > > > > > >> > > > > > detect
> > > > > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> > > > > switching
> > > > > > > to
> > > > > > > > >> the new
> > > > > > > > >> > > > > > protocol?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think
> > > that I
> > > > > am
> > > > > > a
> > > > > > > bit
> > > > > > > > >> more
> > > > > > > > >> > > > > > optimistic
> > > > > > > > >> > > > > > > > > than
> > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe
> > > that the
> > > > > > > > >> majority of
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > cases
> > > > > > > > >> > > > > > > > > are
> > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with
> > > the
> > > > > new
> > > > > > > > >> engine. The
> > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject
> > > the
> > > > > > > consumer
> > > > > > > > >> if the
> > > > > > > > >> > > > > > regex is
> > > > > > > > >> > > > > > > > > not
> > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> > > path,
> > > > > in
> > > > > > > the
> > > > > > > > >> worst
> > > > > > > > >> > > > case,
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > first
> > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group
> > > will
> > > > > be
> > > > > > > > >> rejected.
> > > > > > > > >> > > > This
> > > > > > > > >> > > > > > should
> > > > > > > > >> > > > > > > > > be used
> > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way
> > for
> > > > > > > customers to
> > > > > > > > >> > > > validate
> > > > > > > > >> > > > > > their
> > > > > > > > >> > > > > > > > > regex
> > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would
> > > be to
> > > > > > > test
> > > > > > > > >> them with
> > > > > > > > >> > > > > > another
> > > > > > > > >> > > > > > > > > group.
> > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be
> > > done
> > > > > in a
> > > > > > > > >> pre-prod
> > > > > > > > >> > > > > > environment.
> > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > > > > > >> consumer-group tool
> > > > > > > > >> > > > to
> > > > > > > > >> > > > > > > > > provide a
> > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism.
> > > Would
> > > > > > > this be
> > > > > > > > >> > > > enough in
> > > > > > > > >> > > > > > your
> > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > > Related to
> > > > > > the
> > > > > > > last
> > > > > > > > >> > > > question,
> > > > > > > > >> > > > > > will
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick
> > > with the
> > > > > > > > >> current regex
> > > > > > > > >> > > > > > engine for
> > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may
> > be
> > > > > handy
> > > > > > to
> > > > > > > > >> keep using
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > based
> > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > migrations
> > > > > > simple
> > > > > > > and
> > > > > > > > >> then
> > > > > > > > >> > > > > > migrate to
> > > > > > > > >> > > > > > > > > server
> > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > understand
> > > > > your
> > > > > > > point
> > > > > > > > >> but I
> > > > > > > > >> > > > am
> > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > actually
> > > stay
> > > > > > in
> > > > > > > > >> this mode.
> > > > > > > > >> > > > That
> > > > > > > > >> > > > > > > > > would go
> > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > simplifying
> > > the
> > > > > > > client
> > > > > > > > >> because
> > > > > > > > >> > > > we
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > have to
> > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > metadata on
> > > > > > the
> > > > > > > > >> client
> > > > > > > > >> > > > side. I
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > rather
> > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When
> > > we say
> > > > > > > that
> > > > > > > > >> the group
> > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > > configurations
> > > > > > > > >> and that
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > > > > > deleted.
> > > > > > > > >> Will a
> > > > > > > > >> > > > > > transition to
> > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > configurations?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> > > right.
> > > > > The
> > > > > > > > >> > > > configurations
> > > > > > > > >> > > > > > will be
> > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted.
> > > They go
> > > > > > > > >> together.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will
> > > the
> > > > > > > choice to
> > > > > > > > >> store
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > configs in
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder
> > to
> > > list
> > > > > > all
> > > > > > > > >> cluster
> > > > > > > > >> > > > > > configs and
> > > > > > > > >> > > > > > > > > their
> > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't
> > > think
> > > > > so.
> > > > > > > The
> > > > > > > > >> group
> > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs.
> > If
> > > you
> > > > > > > want to
> > > > > > > > >> know
> > > > > > > > >> > > > all
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > overrides
> > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all
> > > the
> > > > > > group
> > > > > > > > >> > > > coordinators.
> > > > > > > > >> > > > > > You
> > > > > > > > >> > > > > > > > > cannot
> > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > instance.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How
> > > would
> > > > > > > someone
> > > > > > > > >> > > > configure a
> > > > > > > > >> > > > > > group
> > > > > > > > >> > > > > > > > > before
> > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > > considered
> > > > > > > > >> allowing the
> > > > > > > > >> > > > > > explicit
> > > > > > > > >> > > > > > > > > creation
> > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs
> > > could
> > > > > be
> > > > > > > > >> decoupled
> > > > > > > > >> > > > from
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > group
> > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The
> > > group
> > > > > > will
> > > > > > > be
> > > > > > > > >> > > > > > automatically
> > > > > > > > >> > > > > > > > > created in
> > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > configs
> > > will
> > > > > > be
> > > > > > > > >> lost after
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > retention
> > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will
> > > the
> > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > >> > > > > > method
> > > > > > > > >> > > > > > > > > for the
> > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > >> > > > do
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > have to
> > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a
> > > very
> > > > > > group
> > > > > > > > >> question. I
> > > > > > > > >> > > > > > forgot
> > > > > > > > >> > > > > > > > > about
> > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > >> > > > not
> > > > > > > > >> > > > > > fully
> > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to
> > > use,
> > > > > it
> > > > > > > > >> might be
> > > > > > > > >> > > > better
> > > > > > > > >> > > > > > to
> > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which
> > takes
> > > a
> > > > > > > string.
> > > > > > > > >> We would
> > > > > > > > >> > > > > > rely on
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > > > > > migration, I
> > > > > > > > >> think
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > could
> > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and
> > > use
> > > > > it.
> > > > > > > That
> > > > > > > > >> should
> > > > > > > > >> > > > > > work, I
> > > > > > > > >> > > > > > > > > think, in
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I
> > > agree
> > > > > with
> > > > > > > > >> Justine
> > > > > > > > >> > > > that we
> > > > > > > > >> > > > > > > > > should be
> > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch
> > to
> > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > >> > > > > > from the
> > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> > > switch the
> > > > > > > > >> default for
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > feature
> > > > > > > > >> > > > > > > > > flag
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > metadata.version
> > > > > > > once
> > > > > > > > >> we
> > > > > > > > >> > > > want to
> > > > > > > > >> > > > > > make
> > > > > > > > >> > > > > > > > > it the
> > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan
> > was
> > > to
> > > > > use
> > > > > > > that
> > > > > > > > >> feature
> > > > > > > > >> > > > > > flag
> > > > > > > > >> > > > > > > > > mainly
> > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I
> > > should
> > > > > > not
> > > > > > > have
> > > > > > > > >> > > > mentioned
> > > > > > > > >> > > > > > it, I
> > > > > > > > >> > > > > > > > > think,
> > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > internal
> > > > > config
> > > > > > > for
> > > > > > > > >> it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some
> > > of
> > > > > the
> > > > > > > > >> protocol APIs
> > > > > > > > >> > > > > > don't
> > > > > > > > >> > > > > > > > > mention
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to
> > > add
> > > > > that
> > > > > > > for
> > > > > > > > >> > > > consistency.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It
> > is
> > > a
> > > > > bit
> > > > > > > odd
> > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though
> > it
> > > > > seems
> > > > > > > to do
> > > > > > > > >> more
> > > > > > > > >> > > > than
> > > > > > > > >> > > > > > > > > reading.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree.
> > > This is
> > > > > > > how the
> > > > > > > > >> > > > current
> > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require
> > > "Read
> > > > > > > Group" to
> > > > > > > > >> join a
> > > > > > > > >> > > > > > group. We
> > > > > > > > >> > > > > > > > > could
> > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am
> > not
> > > > > sure
> > > > > > > that
> > > > > > > > >> it is
> > > > > > > > >> > > > worth
> > > > > > > > >> > > > > > it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How
> > is
> > > > > topic
> > > > > > > > >> recreation
> > > > > > > > >> > > > > > handled by
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > protocol?
> > > It
> > > > > > > would be
> > > > > > > > >> good to
> > > > > > > > >> > > > > > have a
> > > > > > > > >> > > > > > > > > section
> > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > From
> > > a
> > > > > > > protocol
> > > > > > > > >> > > > > > perspective, the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id
> > so
> > > it
> > > > > > will
> > > > > > > > >> treat it
> > > > > > > > >> > > > like a
> > > > > > > > >> > > > > > > > > topic with
> > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue
> > > is
> > > > > that
> > > > > > > the
> > > > > > > > >> > > > fetch/commit
> > > > > > > > >> > > > > > > > > offsets
> > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > > > > > consumer
> > > > > > > > >> would
> > > > > > > > >> > > > reuse the
> > > > > > > > >> > > > > > > > > offsets
> > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > should
> > > > > > > update
> > > > > > > > >> those
> > > > > > > > >> > > > APIs
> > > > > > > > >> > > > > > as well
> > > > > > > > >> > > > > > > > > in
> > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end.
> > That
> > > > > would
> > > > > > > > >> strengthen
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The
> > > KIP
> > > > > > > mentions
> > > > > > > > >> we will
> > > > > > > > >> > > > > > write the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is
> > an
> > > > > > > > >> implementation
> > > > > > > > >> > > > > > detail, do
> > > > > > > > >> > > > > > > > > we plan
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do
> > we
> > > > > have a
> > > > > > > > >> scalability
> > > > > > > > >> > > > > > goal when
> > > > > > > > >> > > > > > > > > it
> > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group
> > > protocol
> > > > > > can
> > > > > > > > >> support?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't
> > > have
> > > > > > > numbers
> > > > > > > > >> at the
> > > > > > > > >> > > > > > moment. The
> > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members
> > > per
> > > > > > > group.
> > > > > > > > >> We will
> > > > > > > > >> > > > > > measure
> > > > > > > > >> > > > > > > > > this when
> > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > implementation.
> > > > > > Note
> > > > > > > > >> that we
> > > > > > > > >> > > > might
> > > > > > > > >> > > > > > have
> > > > > > > > >> > > > > > > > > other
> > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g.
> > > > > offset
> > > > > > > > >> commits).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did
> > we
> > > > > > > consider
> > > > > > > > >> having
> > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > SubscribedTopicNames
> > > > > > > > >> in
> > > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we
> > > have
> > > > > to
> > > > > > > > >> resolve the
> > > > > > > > >> > > > > > regex on
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > topic
> > > > > name?
> > > > > > > The
> > > > > > > > >> > > > difference
> > > > > > > > >> > > > > > is that
> > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas
> > > > > sending
> > > > > > > the
> > > > > > > > >> topic
> > > > > > > > >> > > > names
> > > > > > > > >> > > > > > is less
> > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > recreation is
> > > > > > > easier
> > > > > > > > >> to
> > > > > > > > >> > > > handle
> > > > > > > > >> > > > > > if we
> > > > > > > > >> > > > > > > > > have
> > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea
> > > was to
> > > > > > > > >> consolidate the
> > > > > > > > >> > > > > > metadata
> > > > > > > > >> > > > > > > > > lookup
> > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but
> > I
> > > do
> > > > > > agree
> > > > > > > > >> with your
> > > > > > > > >> > > > > > point. As
> > > > > > > > >> > > > > > > > > a
> > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be
> > > > > better
> > > > > > > here
> > > > > > > > >> for the
> > > > > > > > >> > > > > > delete and
> > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose
> > > that we
> > > > > > may
> > > > > > > > >> allow
> > > > > > > > >> > > > users to
> > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> > > because
> > > > > > that
> > > > > > > is
> > > > > > > > >> the only
> > > > > > > > >> > > > > > way to be
> > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > re-creation.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue,
> > Jul
> > > 12,
> > > > > > > 2022 at
> > > > > > > > >> 1:38 PM
> > > > > > > > >> > > > > > David
> > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > > Justine,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks
> > for
> > > > > your
> > > > > > > > >> comments.
> > > > > > > > >> > > > Please
> > > > > > > > >> > > > > > find
> > > > > > > > >> > > > > > > > > my
> > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes,
> > > the new
> > > > > > > > >> protocol
> > > > > > > > >> > > > relies on
> > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic
> > > names
> > > > > > based
> > > > > > > in
> > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > ConsumerGroupHeartbeatRequest.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using
> > > topic
> > > > > > > names
> > > > > > > > >> is the
> > > > > > > > >> > > > > > right call
> > > > > > > > >> > > > > > > > > here.
> > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little
> > > more.
> > > > > > > > >> Obviously, the
> > > > > > > > >> > > > > > KIP does
> > > > > > > > >> > > > > > > > > not
> > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > >> > > > > > > > > > > >>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Luke Chen <sh...@gmail.com>]

Hi Jun and David,

For (48), I believe it is a typo.

   - Group (epoch=2)
      - A
      - B
   - Target Assignment (epoch=2)
      - A - partitions=[foo-0, foo-1]
      - B - partitions=[foo-2]
   - Member Assignment
      - A - epoch=1, partitions=[foo-0, foo-1, foo-2]
      - B - epoch=2, partitions=[]        <-- Here, B should own nothing at
      this moment


On Wed, Sep 28, 2022 at 4:31 AM Jun Rao <ju...@confluent.io.invalid> wrote:

> Hi, David,
>
> Thanks for the reply. A few more follow up questions.
>
> 48. Hmm, so member assignment is what the client sees. Then, I am not sure
> about the following. The text says B cannot get foo-2 yet. But in Member
> Assignment, B has foo-2.
>
> At this point B can transition to epoch 2 but cannot get foo-2 until A
> revokes it.
>
> Note that the persisted target assignment of B already includes foo-2 here
> but B does not get it yet because foo-2 is not free.
>
> Group (epoch=2)
> A
> B
> Target Assignment (epoch=2)
> A - partitions=[foo-0, foo-1]
> B - partitions=[foo-2]
> Member Assignment
> A - epoch=1, partitions=[foo-0, foo-1, foo-2]
> B - epoch=2, partitions=[foo-2]
>
>
> 50. Hmm, I thought KStreams doesn't use range assignor all the time. It
> does the assignment based on the topology. For example, if you are joining
> two topics A and B and deriving the join result in a 3rd topic C, which
> will be aggregated. You want to use range to co-locate the partitions from
> topic A and B. However, for topic C, the partition could potentially be
> assigned in round robin. Without the topic name, KStreams won't know which
> topics need to be co-located.
>
> 53. If MemberAssignment.metadata is not used by the group coordinator,
> where is it being used?
>
> Thanks,
>
> Jun
>
> On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson <jason@confluent.io.invalid
> >
> wrote:
>
> > Hi David,
> >
> > Thanks for the response. Replied below:
> >
> > > I don't have a strong opinion on this one. However, it seems that this
> > could be a pandora box if our subset turns out to not be sufficient
> > for some existing users. We will have to add more to it and the
> > compatibility story between versions could be a real nightmare. One
> > thing that we could do for sure is disabling all unicode groups in the
> > library. That seems possible and would already reduce the scope.
> >
> > Changes to the regex syntax seems not a major problem to me. We would
> bump
> > the protocol version as we usually do. Actually that begs the same
> question
> > about RE2. How stable is the syntax and does it differ depending on the
> > version of the library? In general, I do agree that we'd be better off
> not
> > having to manage the syntax ourselves if we can convince ourselves that
> > compatibility won't be a problem and the client won't eventually need to
> > have access to it.
> >
> > Thanks,
> > Jason
> >
> > On Tue, Sep 27, 2022 at 6:21 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hi Luke,
> > >
> > > Thanks for your comments.
> > >
> > > Your suggestion makes sense. Jun's made the same one. We are also
> > > discussing whether we should directly return the group state in the
> > > response and drop the prepare assignment api. It seems that the
> > > current consensus is to keep it though. If that stays like this, I
> > > will apply your suggestion.
> > >
> > > Thanks,
> > > David
> > >
> > > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com> wrote:
> > > >
> > > > Hi David,
> > > >
> > > > Thanks for your patience for answering all these questions!
> > > >
> > > > One more question from me:
> > > > 1. When the selected consumer is going to be in client-side
> assigning,
> > we
> > > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering
> if
> > we
> > > > could use a flag to indicate the assignment is required. The
> rationale
> > is
> > > > when returning error, we lose a chance to catch up with the latest
> > target
> > > > assignment or revoke partitions. Since we expected the record
> fetching
> > is
> > > > not blocked by the rebalance/assignment computing, we might want to
> > allow
> > > > the members to update the member assignment in each heartbeat.
> > > >
> > > > Luke
> > > >
> > > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > > <ja...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hey David,
> > > > >
> > > > > Thanks for the response.
> > > > >
> > > > > > Do you think that we could have a case where two members with the
> > > same
> > > > > static id are active?
> > > > >
> > > > > I was just thinking about the handoff when we spin up a new node
> > > before the
> > > > > old one completes shutdown. It's probably ok to fence the old
> member
> > > > > abruptly. I do recall we had a few issues with this handoff in the
> > > past. It
> > > > > was due to the fact that the replaced node might have been the
> > leader.
> > > > > Since we've removed the leader notion, perhaps this kind of issue
> > won't
> > > > > resurface.
> > > > >
> > > > > In regard to the RE2 library. I understand the intent to pass this
> > > through
> > > > > as a string without any validation so that we don't need the
> > > dependence on
> > > > > the client. I do wonder if that will be restrictive in practice.
> For
> > > > > example, if a regex changes on the client side, would we need to
> > > validate
> > > > > against the current assignment to see which partitions to continue
> > > > > fetching? Or would we instead invalidate the complete current
> > > assignment?
> > > > > Perhaps this is a rare case and we can do the latter. There could
> be
> > > more
> > > > > situations though.
> > > > >
> > > > > As an alternative, I wonder if we could define a minimalistic regex
> > > syntax
> > > > > (a subset of RE2 say) that could map more easily to regex support
> > that
> > > > > already exists in clients? For example, in this scenario, we are
> > > matching
> > > > > topics, so we only need to be able to represent topic characters
> and
> > > not
> > > > > the whole smorgasbord of whitespace and unicode options. Would it
> > make
> > > > > sense to have a reduced syntax which includes only standard topic
> > > > > characters, wildcards, character classes, and repetitions?
> > > > >
> > > > > Best,
> > > > > Jason
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > > >
> > > > > > Hi David, thanks for all the detailed explanations. I think they
> > all
> > > make
> > > > > > sense. Just want to have a couple follow-ups here:
> > > > > >
> > > > > > > I don't really see the benefits here because server side
> > assignors
> > > > > > don't have metadata at all. They only assign topic-partitions.
> They
> > > > > > are not supposed to generate metadata nor to receive metadata
> from
> > > the
> > > > > > members.
> > > > > >
> > > > > > Just to clarify I was asking about the `version` of the assignor
> > > (i.e. up
> > > > > > to what version that the client would support), and I do agree we
> > > would
> > > > > not
> > > > > > need metadata. What I have in mind is that, for some specific
> > > built-in
> > > > > > broker-assignors, e.g. rack-aware assignors, if it's possible
> that
> > > in a
> > > > > > newer version we would have a hierarchical rack ID string format,
> > > like
> > > > > > "tier1-tier2" etc, but if some client has not upgraded their rack
> > ID
> > > > > > would still be in old format. In this case, the broker then needs
> > to
> > > > > choose
> > > > > > the old versioned assignor. I'm probably making something up here
> > for
> > > > > rack
> > > > > > aware assignors, but I'm wondering if in general such an
> > > "auto-downgrade"
> > > > > > behavior would be needed still for broker-side assignor, and if
> yes
> > > would
> > > > > > "version" still be useful.
> > > > > >
> > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> called
> > > once
> > > > > > when the member transitions to a new epoch. This one contains the
> > > full
> > > > > > metadata provided by the client side assignor. Then,
> `onAssignment`
> > > > > > can be called max N times where N is the number of partitions
> > pending
> > > > > > revocation by other members. Let me try to clarify this in the
> KIP.
> > > > > >
> > > > > > Okay, my understanding is that the calling ordering of these
> > > callbacks
> > > > > > would be like the following:
> > > > > >
> > > > > > ----------------------------------------
> > > > > > onPartitionsRevoked();   // just once, since we do not really
> need
> > > > > > to revoke incrementally.
> > > > > >
> > > > > > onAssignment();    // the first call, with epoch incremented
> > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > >
> > > > > > onAssignment();              // the first onAssignment would bump
> > up
> > > the
> > > > > > epoch, and the metadata reflected.
> > > > > > onPartitionsAssigned();   // each time we get an additional
> > > assignment,
> > > > > we
> > > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > > ...
> > > > > > onAssignment();
> > > > > > onPartitionsAssigned();   // on each of the onAssignment calls,
> the
> > > > > encoded
> > > > > > metadata would not change, only the incrementally added
> partitions
> > be
> > > > > > reflected
> > > > > >
> > > > > > Is that the case?
> > > > > >
> > > > > > I'm wondering if we would still call onAssignment just once, that
> > > encodes
> > > > > > all the assignment for this rebalance, including all the
> partitions
> > > that
> > > > > > should be assigned to the member but not yet assigned since they
> > > have not
> > > > > > been revoked by others. In that case the call ordering would be:
> > > > > >
> > > > > > ----------------------------------------
> > > > > > onPartitionsRevoked();   // just once
> > > > > > onAssignment();    // just once, with epoch incremented, and
> > metadata
> > > > > > encoded changed, the "assignment" field also reflect the final
> > target
> > > > > > assignment
> > > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > > incrementally
> > > > > > added partitions
> > > > > > ...
> > > > > > onPartitionsAssigned();
> > > > > >
> > > > > > The motivation from this is that, most users would only implement
> > the
> > > > > > rebalance callback listeners and hence we'd definitely need to
> make
> > > sure
> > > > > > the semantics of that does not change much, and the time
> > > > > > `onPartitionsAssigned` indicate the time when the partitions are
> > > actually
> > > > > > assigned to it; while for assignors, the `onAssignment` is used
> to
> > > > > indicate
> > > > > > what decision is made regarding for this member, i.e. when the
> > > partitions
> > > > > > are decided to be given to it, but not necessarily meant that it
> > has
> > > been
> > > > > > given, since that time should be determined by the time of
> > > > > > `onPartitionsAssigned`. The benefits though, would be that
> assignor
> > > > > > implementers would not need to reason which `onAssignment` would
> be
> > > the
> > > > > > last one for this epoch.
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > <djacot@confluent.io.invalid
> > > > > >
> > > > > > wrote:
> > > > > >
> > > > > > > Hi Guozhang,
> > > > > > >
> > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> Error
> > > byte
> > > > > > > field,
> > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> not.
> > > And
> > > > > the
> > > > > > > > rationale seems to be assuming that we should always be able
> to
> > > do
> > > > > the
> > > > > > > > assignment at the broker-side assignor without errors.
> > > Personally I
> > > > > > think
> > > > > > > > it's still potentially beneficial to add the Error field even
> > for
> > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > subscribed
> > > > > > > > topics are not recognized with the current broker's metadata.
> > > What do
> > > > > > you
> > > > > > > > think?
> > > > > > >
> > > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > > better
> > > > > > > to use an exception on the server side. This is what we usually
> > do
> > > for
> > > > > > > server side plugins. On the client side, we use a field because
> > the
> > > > > > > errors are not defined in advance.
> > > > > > >
> > > > > > > Your comment also makes me think about what we should do when
> the
> > > > > > > server side assignor fails. I suppose that we have to keep the
> > > current
> > > > > > > assignment until a new event occurs. For instance, in your
> > example,
> > > > > > > the coordinator would have to trigger a rebalance when
> > unrecognized
> > > > > > > topics are available. This would be part of the metadata
> > > monitoring.
> > > > > > >
> > > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > > reason/version/metadata compared with the broker-side
> > > GroupMember. I
> > > > > > > agree
> > > > > > > > that broker-side assignor would not need reason/metadata
> since
> > > they
> > > > > are
> > > > > > > > blackbox strings/bytes to the assignor, but what about
> version?
> > > E.g.
> > > > > is
> > > > > > > it
> > > > > > > > possible that we evolve our broker-side built-in assignor but
> > > the old
> > > > > > > > versioned clients would not be able to work with the new
> > > version, in
> > > > > > > which
> > > > > > > > case we need to let the broker being aware of this and
> upgrade
> > > its
> > > > > > > behavior
> > > > > > > > to cooperate with the clients?
> > > > > > >
> > > > > > > I don't really see the benefits here because server side
> > assignors
> > > > > > > don't have metadata at all. They only assign topic-partitions.
> > They
> > > > > > > are not supposed to generate metadata nor to receive metadata
> > from
> > > the
> > > > > > > members.
> > > > > > >
> > > > > > > > 3) Also related to 2) above, for the client-side
> "GroupMember",
> > > > > instead
> > > > > > > of
> > > > > > > > including these three fields, what about just adding the
> > > "Metadata"
> > > > > > field
> > > > > > > > class which has these three fields? Also, there are two
> > > "Metadata"
> > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > reason/version/metadata, and the second is just the encoded
> > > metadata
> > > > > > > bytes.
> > > > > > > > I'm wondering what about just naming the first as
> > memberMetadata,
> > > > > which
> > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > second
> > > > > > > > bytebuffer field as metadataBytes?
> > > > > > >
> > > > > > > That's a good point. Let me try to rationalize this interface
> > > based on
> > > > > > > your suggestions.
> > > > > > >
> > > > > > > Best,
> > > > > > > David
> > > > > > >
> > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > > wrote:
> > > > > > > >
> > > > > > > > Hello David,
> > > > > > > >
> > > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > > >
> > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> Error
> > > byte
> > > > > > > field,
> > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> not.
> > > And
> > > > > the
> > > > > > > > rationale seems to be assuming that we should always be able
> to
> > > do
> > > > > the
> > > > > > > > assignment at the broker-side assignor without errors.
> > > Personally I
> > > > > > think
> > > > > > > > it's still potentially beneficial to add the Error field even
> > for
> > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > subscribed
> > > > > > > > topics are not recognized with the current broker's metadata.
> > > What do
> > > > > > you
> > > > > > > > think?
> > > > > > > >
> > > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > > reason/version/metadata compared with the broker-side
> > > GroupMember. I
> > > > > > > agree
> > > > > > > > that broker-side assignor would not need reason/metadata
> since
> > > they
> > > > > are
> > > > > > > > blackbox strings/bytes to the assignor, but what about
> version?
> > > E.g.
> > > > > is
> > > > > > > it
> > > > > > > > possible that we evolve our broker-side built-in assignor but
> > > the old
> > > > > > > > versioned clients would not be able to work with the new
> > > version, in
> > > > > > > which
> > > > > > > > case we need to let the broker being aware of this and
> upgrade
> > > its
> > > > > > > behavior
> > > > > > > > to cooperate with the clients?
> > > > > > > >
> > > > > > > > 3) Also related to 2) above, for the client-side
> "GroupMember",
> > > > > instead
> > > > > > > of
> > > > > > > > including these three fields, what about just adding the
> > > "Metadata"
> > > > > > field
> > > > > > > > class which has these three fields? Also, there are two
> > > "Metadata"
> > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > reason/version/metadata, and the second is just the encoded
> > > metadata
> > > > > > > bytes.
> > > > > > > > I'm wondering what about just naming the first as
> > memberMetadata,
> > > > > which
> > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > second
> > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hello David,
> > > > > > > > >
> > > > > > > > > Thanks for bringing this question up. I think the main
> > > benefits as
> > > > > > you
> > > > > > > > > listed is 2) above if it stays; just to clarify, we would
> > only
> > > be
> > > > > > able
> > > > > > > to
> > > > > > > > > save one round trip if the rebalance is still triggered by
> > the
> > > > > > broker;
> > > > > > > if
> > > > > > > > > the rebalance is triggered by the client then the
> > > num.round.trips
> > > > > are
> > > > > > > the
> > > > > > > > > same:
> > > > > > > > >
> > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > >
> > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > already
> > > > > > sent
> > > > > > > a
> > > > > > > > > HB and hence has to wait for it to return first since only
> > one
> > > > > > request
> > > > > > > /
> > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > GroupInstallAssignment
> > > > > > > > > request.
> > > > > > > > >
> > > > > > > > > In total, two round trips.
> > > > > > > > >
> > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > >
> > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > already
> > > > > > sent
> > > > > > > a
> > > > > > > > > HB and hence has to wait for it to return first since only
> > one
> > > > > > request
> > > > > > > /
> > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > T1: client receives the HB response, and then sends the new
> > HB
> > > > > > request
> > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > T2: the HB response with the optional member metadata map
> is
> > > > > > returned.
> > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > GroupInstallAssignment
> > > > > > > > > request.
> > > > > > > > >
> > > > > > > > > In total, two round trips as well.
> > > > > > > > >
> > > > > > > > > -----------------------------
> > > > > > > > >
> > > > > > > > > So to complete the full picture here, we'd need to modify
> > both
> > > HB
> > > > > > > request
> > > > > > > > > and response so that the client can also indicate a new
> > > rebalance
> > > > > via
> > > > > > > the
> > > > > > > > > HB request as well, right?
> > > > > > > > >
> > > > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > > additional
> > > > > > > fields
> > > > > > > > > encoding the full member (subscription) metadata and topic
> > > metadata
> > > > > > as
> > > > > > > > > optional fields.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > >> Hi all,
> > > > > > > > >>
> > > > > > > > >> During an offline conversation, someone asked why we need
> > the
> > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we
> > could
> > > > > > instead
> > > > > > > > >> provide the group state in the heartbeat response. This
> has
> > a
> > > few
> > > > > > > > >> advantages: 1) it does not require using a special error
> > code
> > > to
> > > > > > > > >> signal that a new assignment is required as the signal
> would
> > > be
> > > > > the
> > > > > > > > >> provided group state; 2) it removes one round trip when a
> > > client
> > > > > > side
> > > > > > > > >> assignor is used. The downside is that it makes the
> > heartbeat
> > > > > > > > >> response's definition quite large. I recall that I went
> with
> > > the
> > > > > > > > >> current approach due to this.
> > > > > > > > >>
> > > > > > > > >> Providing the group state in the heartbeat response is
> > > appealing.
> > > > > > What
> > > > > > > > >> do you guys think?
> > > > > > > > >>
> > > > > > > > >> Best,
> > > > > > > > >> David
> > > > > > > > >>
> > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > djacot@confluent.io>
> > > > > > > wrote:
> > > > > > > > >> >
> > > > > > > > >> > Hi Guozhang,
> > > > > > > > >> >
> > > > > > > > >> > 1. I have added a reference to the relevant chapter
> > instead
> > > of
> > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > >> >
> > > > > > > > >> > 2. The "Rebalance Triggers" section you are referring to
> > is
> > > > > about
> > > > > > > when
> > > > > > > > >> > a rebalance should be triggered for the non-upgraded
> > members
> > > > > using
> > > > > > > the
> > > > > > > > >> > old protocol. The section mentions that a rebalance must
> > be
> > > > > > > triggered
> > > > > > > > >> > when a new assignment is installed. This implies that
> the
> > > group
> > > > > > > epoch
> > > > > > > > >> > was updated either by a native member or a non-upgraded
> > > member.
> > > > > > For
> > > > > > > > >> > the latter, the JoinGroup request would be the trigger.
> I
> > > have
> > > > > > > added a
> > > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > > Handling"
> > > > > > > section
> > > > > > > > >> > as well. Does that make sense?
> > > > > > > > >> >
> > > > > > > > >> > Thanks,
> > > > > > > > >> > David
> > > > > > > > >> >
> > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > wangguoz@gmail.com
> > > > > > >
> > > > > > > > >> wrote:
> > > > > > > > >> > >
> > > > > > > > >> > > Hello David,
> > > > > > > > >> > >
> > > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > > clear in
> > > > > > the
> > > > > > > doc,
> > > > > > > > >> > > could we update:
> > > > > > > > >> > >
> > > > > > > > >> > > 1) the heartbeat request handling section, stating
> when
> > > > > > > coordinator
> > > > > > > > >> will
> > > > > > > > >> > > trigger rebalance based on the HB's member metadata /
> > > reason?
> > > > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > > > described
> > > > > > > in
> > > > > > > > >> "Group
> > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > >> > >
> > > > > > > > >> > >
> > > > > > > > >> > > Guozhang
> > > > > > > > >> > >
> > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > >> > > wrote:
> > > > > > > > >> > >
> > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > >> > > >
> > > > > > > > >> > > > I thought that the assignor will always be consulted
> > > when
> > > > > the
> > > > > > > next
> > > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > > >> > > > `PartitionAssignor#metadata` will be called for
> every
> > > > > > heartbeat.
> > > > > > > > >> This
> > > > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > > > rebalance
> > > > > > by
> > > > > > > > >> > > > setting the reason to a non-zero value or by
> changing
> > > the
> > > > > > > bytes. Do
> > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > concerned by
> > > > > > the
> > > > > > > > >> delay?
> > > > > > > > >> > > >
> > > > > > > > >> > > > Best,
> > > > > > > > >> > > > David
> > > > > > > > >> > > >
> > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > wangguoz@gmail.com>
> > > > > > > > >> wrote:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > Hello David,
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > ```
> > > > > > > > >> > > > > In this case, a new assignment is triggered by the
> > > client
> > > > > > side
> > > > > > > > >> > > > > assignor. When constructing the HB, the consumer
> > will
> > > > > always
> > > > > > > > >> consult
> > > > > > > > >> > > > > the client side assignor and propagate the
> > > information to
> > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > coordinator. In other words, we don't expect users
> > to
> > > call
> > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > >> > > > > ```
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > > interface,
> > > > > we
> > > > > > > > >> actually do
> > > > > > > > >> > > > > not have a way yet to instruct how to construct
> the
> > > next
> > > > > HB
> > > > > > > > >> request, e.g.
> > > > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> > > with a
> > > > > > new
> > > > > > > > >> assignment,
> > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > > PartitionAssignor
> > > > > > > to
> > > > > > > > >> indicate
> > > > > > > > >> > > > > the next HB telling broker about so. WDYT about
> > adding
> > > > > such
> > > > > > an
> > > > > > > > >> API on the
> > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > Guozhang
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > I have updated the KIP to include your
> feedback. I
> > > have
> > > > > > also
> > > > > > > > >> tried to
> > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > Best,
> > > > > > > > >> > > > > > David
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > > djacot@confluent.io
> > > > > > > > >> >
> > > > > > > > >> > > > wrote:
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > > answering
> > > > > your
> > > > > > > > >> questions
> > > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> benefits
> > > of
> > > > > the
> > > > > > > KIP
> > > > > > > > >> seem to
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > fewer
> > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > support of
> > > > > > > > >> wildcard. A few
> > > > > > > > >> > > > > > > > comments below.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > I would also add that the KIP removes the
> global
> > > sync
> > > > > > > barrier
> > > > > > > > >> in the
> > > > > > > > >> > > > > > > protocol which is essential to improve group
> > > stability
> > > > > > and
> > > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > > client by
> > > > > > > moving
> > > > > > > > >> most of
> > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> plan
> > > to
> > > > > > > support
> > > > > > > > >> rolling
> > > > > > > > >> > > > > > changing
> > > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > > > assignor
> > > > > > > used
> > > > > > > > >> by a
> > > > > > > > >> > > > > > > majority of the members. This allows the group
> > to
> > > move
> > > > > > > from
> > > > > > > > >> one
> > > > > > > > >> > > > > > > assignor to another by a roll. This is
> explained
> > > in
> > > > > the
> > > > > > > > >> Assignor
> > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> whether
> > > it's
> > > > > > > > >> required in
> > > > > > > > >> > > > every
> > > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > filled?
> > > > > > For
> > > > > > > > >> example,
> > > > > > > > >> > > > it's
> > > > > > > > >> > > > > > not
> > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > > filled.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > The client is expected to set those fields in
> > > case of
> > > > > a
> > > > > > > > >> connection
> > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > > changed
> > > > > > since
> > > > > > > > >> the last
> > > > > > > > >> > > > > > > HB. The server populates those fields as long
> as
> > > the
> > > > > > > member
> > > > > > > > >> is not
> > > > > > > > >> > > > > > > fully reconciled - the member should
> acknowledge
> > > that
> > > > > it
> > > > > > > has
> > > > > > > > >> the
> > > > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> > > this in
> > > > > > the
> > > > > > > KIP.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> rack
> > > > > > affinity
> > > > > > > > >> between the
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > and the broker is only considered during
> > > fetching,
> > > > > but
> > > > > > > not
> > > > > > > > >> during
> > > > > > > > >> > > > > > assigning
> > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > > assignment
> > > > > > > is
> > > > > > > > >> made,
> > > > > > > > >> > > > there
> > > > > > > > >> > > > > > is
> > > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > replicas
> > > > > > of
> > > > > > > > >> assigned
> > > > > > > > >> > > > > > partitions
> > > > > > > > >> > > > > > > > are close to the member. I am wondering if
> we
> > > should
> > > > > > use
> > > > > > > > >> this
> > > > > > > > >> > > > > > opportunity
> > > > > > > > >> > > > > > > > to address this by including rack in
> > > GroupMember.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's an interesting idea. I don't see any
> > issue
> > > with
> > > > > > > adding
> > > > > > > > >> the
> > > > > > > > >> > > > rack
> > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful
> to
> > > know
> > > > > how
> > > > > > > busy
> > > > > > > > >> a group
> > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > model,
> > > it
> > > > > > seems
> > > > > > > > >> that we
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > add
> > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> time
> > > the
> > > > > > event
> > > > > > > > >> loop is
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > actual work.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > failover
> > > > > > > > >> handling? For
> > > > > > > > >> > > > > > example,
> > > > > > > > >> > > > > > > > does it need to trigger the check if any
> group
> > > with
> > > > > > the
> > > > > > > > >> wildcard
> > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sure. When the new group coordinator takes
> over,
> > > it
> > > > > has
> > > > > > > to:
> > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > > assignor
> > > > > is
> > > > > > > used.
> > > > > > > > >> We
> > > > > > > > >> > > > don't
> > > > > > > > >> > > > > > > store the information about the member
> selected
> > > to run
> > > > > > the
> > > > > > > > >> assignment
> > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> > wildcard
> > > > > > > > >> subscriptions, and
> > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > document
> > > > > > what
> > > > > > > > >> the epoch
> > > > > > > > >> > > > > > field
> > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > >> > > > > > reflect
> > > > > > > > >> > > > > > > > the latest group epoch? What about the one
> in
> > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> > > latest
> > > > > > > group
> > > > > > > > >> epoch.
> > > > > > > > >> > > > > > > When the group state is updated, the group
> epoch
> > > is
> > > > > > > bumped so
> > > > > > > > >> we use
> > > > > > > > >> > > > > > > that one for all the change records related to
> > the
> > > > > > update.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> > the
> > > > > > > following
> > > > > > > > >> > > > invariants
> > > > > > > > >> > > > > > are
> > > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> > for
> > > a
> > > > > > member
> > > > > > > > >> not to
> > > > > > > > >> > > > get any
> > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's right. Here I meant that the members
> > > provided
> > > > > by
> > > > > > > the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > > > assignor
> > > > > > > can
> > > > > > > > >> not make
> > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > > epoch
> > > > > > > equals to
> > > > > > > > >> 0":
> > > > > > > > >> > > > When
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > a consumer rejoin and what member id would
> be
> > > used?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > partitions and
> > > > > > > > >> rejoins when
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In
> this
> > > case,
> > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > > > coordinator will have removed the member from
> > the
> > > > > group.
> > > > > > > The
> > > > > > > > >> member
> > > > > > > > >> > > > > > > can rejoin the group with the same member id
> but
> > > with
> > > > > 0
> > > > > > as
> > > > > > > > >> epoch. Let
> > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > ability
> > > to
> > > > > > > trigger a
> > > > > > > > >> > > > > > reassignment
> > > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > updating
> > > > > > the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > > with
> > > > > the
> > > > > > > > >> deprecation
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > In this case, a new assignment is triggered by
> > the
> > > > > > client
> > > > > > > side
> > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> consumer
> > > will
> > > > > > > always
> > > > > > > > >> consult
> > > > > > > > >> > > > > > > the client side assignor and propagate the
> > > information
> > > > > > to
> > > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > > users to
> > > > > > call
> > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> > the
> > > > > > section
> > > > > > > > >> seems to
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2,
> B
> > > > > > > immediately
> > > > > > > > >> gets into
> > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > incorrect.
> > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3,
> C
> > > seems
> > > > > to
> > > > > > > get
> > > > > > > > >> into
> > > > > > > > >> > > > > > epoch=3,
> > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > > has A -
> > > > > > > > >> epoch=2,
> > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > support
> > > the
> > > > > > > upgrade
> > > > > > > > >> from
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > old
> > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > We will support upgrading from the consumer
> > > protocol
> > > > > > > version
> > > > > > > > >> 3,
> > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> > implemented
> > > yet
> > > > > so
> > > > > > > the
> > > > > > > > >> earliest
> > > > > > > > >> > > > > > > version is unknown at the moment. This is
> > > explained in
> > > > > > the
> > > > > > > > >> migration
> > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> > > update the
> > > > > > KIP
> > > > > > > > >> based on
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > next week.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Best,
> > > > > > > > >> > > > > > > David
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > >> > > > wrote:
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> benefits
> > > of
> > > > > the
> > > > > > > KIP
> > > > > > > > >> seem to
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > fewer
> > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > support of
> > > > > > > > >> wildcard. A few
> > > > > > > > >> > > > > > > > comments below.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> plan
> > > to
> > > > > > > support
> > > > > > > > >> rolling
> > > > > > > > >> > > > > > changing
> > > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> whether
> > > it's
> > > > > > > > >> required in
> > > > > > > > >> > > > every
> > > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > filled?
> > > > > > For
> > > > > > > > >> example,
> > > > > > > > >> > > > it's
> > > > > > > > >> > > > > > not
> > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > > filled.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> rack
> > > > > > affinity
> > > > > > > > >> between the
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > and the broker is only considered during
> > > fetching,
> > > > > but
> > > > > > > not
> > > > > > > > >> during
> > > > > > > > >> > > > > > assigning
> > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > > assignment
> > > > > > > is
> > > > > > > > >> made,
> > > > > > > > >> > > > there
> > > > > > > > >> > > > > > is
> > > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > replicas
> > > > > > of
> > > > > > > > >> assigned
> > > > > > > > >> > > > > > partitions
> > > > > > > > >> > > > > > > > are close to the member. I am wondering if
> we
> > > should
> > > > > > use
> > > > > > > > >> this
> > > > > > > > >> > > > > > opportunity
> > > > > > > > >> > > > > > > > to address this by including rack in
> > > GroupMember.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful
> to
> > > know
> > > > > how
> > > > > > > busy
> > > > > > > > >> a group
> > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> > model,
> > > it
> > > > > > seems
> > > > > > > > >> that we
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > add
> > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> time
> > > the
> > > > > > event
> > > > > > > > >> loop is
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > actual work.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > failover
> > > > > > > > >> handling? For
> > > > > > > > >> > > > > > example,
> > > > > > > > >> > > > > > > > does it need to trigger the check if any
> group
> > > with
> > > > > > the
> > > > > > > > >> wildcard
> > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > document
> > > > > > what
> > > > > > > > >> the epoch
> > > > > > > > >> > > > > > field
> > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > >> > > > > > reflect
> > > > > > > > >> > > > > > > > the latest group epoch? What about the one
> in
> > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> > the
> > > > > > > following
> > > > > > > > >> > > > invariants
> > > > > > > > >> > > > > > are
> > > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> > for
> > > a
> > > > > > member
> > > > > > > > >> not to
> > > > > > > > >> > > > get any
> > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > > epoch
> > > > > > > equals to
> > > > > > > > >> 0":
> > > > > > > > >> > > > When
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > a consumer rejoin and what member id would
> be
> > > used?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> > ability
> > > to
> > > > > > > trigger a
> > > > > > > > >> > > > > > reassignment
> > > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > updating
> > > > > > the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > > with
> > > > > the
> > > > > > > > >> deprecation
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> > the
> > > > > > section
> > > > > > > > >> seems to
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2,
> B
> > > > > > > immediately
> > > > > > > > >> gets into
> > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > incorrect.
> > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3,
> C
> > > seems
> > > > > to
> > > > > > > get
> > > > > > > > >> into
> > > > > > > > >> > > > > > epoch=3,
> > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > > has A -
> > > > > > > > >> epoch=2,
> > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> > support
> > > the
> > > > > > > upgrade
> > > > > > > > >> from
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > old
> > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Jun
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > > > > wrote:
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > The KIP states that we will re-implement
> the
> > > > > > > coordinator
> > > > > > > > >> in
> > > > > > > > >> > > > Java. I
> > > > > > > > >> > > > > > > > > discussed this offline with a few folks
> and
> > > folks
> > > > > > are
> > > > > > > > >> concerned
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > > > > we could introduce many regressions in the
> > old
> > > > > > > protocol
> > > > > > > > >> if we do
> > > > > > > > >> > > > so.
> > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> > statement
> > > > > from
> > > > > > > the
> > > > > > > > >> KIP. It
> > > > > > > > >> > > > is an
> > > > > > > > >> > > > > > > > > implementation detail after all so it does
> > not
> > > > > have
> > > > > > > to be
> > > > > > > > >> > > > decided at
> > > > > > > > >> > > > > > > > > this stage. We will likely start by trying
> > to
> > > > > > > refactor the
> > > > > > > > >> > > > current
> > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > >> > > > > > > > > David
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David
> Jacot
> > <
> > > > > > > > >> djacot@confluent.io
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > "Empty,
> > > > > > > assigning,
> > > > > > > > >> > > > > > reconciling,
> > > > > > > > >> > > > > > > > > stable,
> > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> States
> > > > > > section,
> > > > > > > > >> right?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > This sentence does not refer to those
> > group
> > > > > states
> > > > > > > but
> > > > > > > > >> rather
> > > > > > > > >> > > > to a
> > > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > > refers to
> > > > > > the
> > > > > > > > >> entire
> > > > > > > > >> > > > state of
> > > > > > > > >> > > > > > > > > > group coordinator which is replicated
> via
> > > the
> > > > > log
> > > > > > > > >> layer. I will
> > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > machine
> > > is
> > > > > > > modelled
> > > > > > > > >> as an
> > > > > > > > >> > > > event
> > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> > > the new
> > > > > > > quorum
> > > > > > > > >> > > > > > controller. We
> > > > > > > > >> > > > > > > > > > will have N threads to process events.
> > Each
> > > > > > > > >> __consumer_offsets
> > > > > > > > >> > > > > > > > > > partition is assigned to a unique thread
> > > and all
> > > > > > the
> > > > > > > > >> events
> > > > > > > > >> > > > (e.g.
> > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed
> > by
> > > this
> > > > > > > > >> thread. This
> > > > > > > > >> > > > > > simplify
> > > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > > simulation
> > > > > > > testing
> > > > > > > > >> for the
> > > > > > > > >> > > > > > group
> > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> per
> > > > > > > > >> *__consumer_offsets*
> > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> group"
> > > owned
> > > > > > by
> > > > > > > a
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator?
> > > > > > > > >> > > > > > > > > For
> > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> owns 2
> > > > > > consumer
> > > > > > > > >> groups, and
> > > > > > > > >> > > > > > both
> > > > > > > > >> > > > > > > > > exist in
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > > state
> > > > > > > machine
> > > > > > > > >> for it,
> > > > > > > > >> > > > or
> > > > > > > > >> > > > > > 2?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> there, I
> > > > > think.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > "*group.coordinator.threads"
> > > > > *is
> > > > > > > the
> > > > > > > > >> number
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > threads
> > > > > > > > >> > > > > > > > > used
> > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > wondering
> > > > > if
> > > > > > > the
> > > > > > > > >> purpose
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > threads
> > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > consumer
> > > > > group
> > > > > > > (or
> > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> computation,
> > > why
> > > > > > > should we
> > > > > > > > >> need
> > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability
> to
> > > shard
> > > > > > the
> > > > > > > > >> > > > processing as
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > > does
> > > > > the
> > > > > > > > >> default
> > > > > > > > >> > > > session
> > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> max(60s)? I
> > > > > thought
> > > > > > > the
> > > > > > > > >> min/max
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound
> > of
> > > it,
> > > > > > no?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > > 30s The
> > > > > > > > >> timeout to
> > > > > > > > >> > > > detect
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > > protocol.
> > > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > > int 45s
> > > > > > The
> > > > > > > > >> minimum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > > int 60s
> > > > > > The
> > > > > > > > >> maximum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > > session
> > > > > > > timeout
> > > > > > > > >> should
> > > > > > > > >> > > > be 45s
> > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> > are
> > > > > > [range,
> > > > > > > > >> uniform],
> > > > > > > > >> > > > > > which means
> > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > > like to
> > > > > > > know
> > > > > > > > >> why not
> > > > > > > > >> > > > > > uniform
> > > > > > > > >> > > > > > > > > one? I
> > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > uniform
> > > > > > assignor
> > > > > > > > >> (former
> > > > > > > > >> > > > sticky
> > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > other
> > > > > reason
> > > > > > > we
> > > > > > > > >> choose
> > > > > > > > >> > > > range
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > uniform
> > > > > The
> > > > > > > > >> server side
> > > > > > > > >> > > > > > assignors.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > The order on the server side has no
> > > influence
> > > > > > > because
> > > > > > > > >> the
> > > > > > > > >> > > > client
> > > > > > > > >> > > > > > must
> > > > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> > > There
> > > > > is
> > > > > > no
> > > > > > > > >> default
> > > > > > > > >> > > > in the
> > > > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > > > specified
> > > > > > > by
> > > > > > > > >> the
> > > > > > > > >> > > > client,
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > request is rejected. The default client
> > > value
> > > > > for
> > > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> > though.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Thanks for your very good comments,
> Luke.
> > I
> > > hope
> > > > > > > that my
> > > > > > > > >> > > > answers
> > > > > > > > >> > > > > > help
> > > > > > > > >> > > > > > > > > > to clarify things. I will update the KIP
> > as
> > > well
> > > > > > > based
> > > > > > > > >> on your
> > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > >> > > > > > > > > > David
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke
> Chen
> > <
> > > > > > > > >> showuon@gmail.com>
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > > mentioned:
> > > > > > > > >> > > > > > > > > > > > The new group coordinator will have
> a
> > > state
> > > > > > > machine
> > > > > > > > >> per
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> > > each
> > > > > > state
> > > > > > > > >> machine is
> > > > > > > > >> > > > > > modelled
> > > > > > > > >> > > > > > > > > as an
> > > > > > > > >> > > > > > > > > > > event loop. Those state machines will
> be
> > > > > > executed
> > > > > > > in
> > > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > "Empty,
> > > > > > > assigning,
> > > > > > > > >> > > > > > reconciling,
> > > > > > > > >> > > > > > > > > stable,
> > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> States
> > > > > > section,
> > > > > > > > >> right?
> > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> > machine
> > > is
> > > > > > > modelled
> > > > > > > > >> as an
> > > > > > > > >> > > > event
> > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> per
> > > > > > > > >> *__consumer_offsets*
> > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> group"
> > > owned
> > > > > > by
> > > > > > > a
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator?
> > > > > > > > >> > > > > > > > > For
> > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> owns 2
> > > > > > consumer
> > > > > > > > >> groups, and
> > > > > > > > >> > > > > > both
> > > > > > > > >> > > > > > > > > exist in
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > > state
> > > > > > > machine
> > > > > > > > >> for it,
> > > > > > > > >> > > > or
> > > > > > > > >> > > > > > 2?
> > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > "*group.coordinator.threads"
> > > > > *is
> > > > > > > the
> > > > > > > > >> number
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > threads
> > > > > > > > >> > > > > > > > > used
> > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > wondering
> > > > > if
> > > > > > > the
> > > > > > > > >> purpose
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > threads
> > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> > consumer
> > > > > group
> > > > > > > (or
> > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> computation,
> > > why
> > > > > > > should we
> > > > > > > > >> need
> > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > > configs:
> > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > > does
> > > > > the
> > > > > > > > >> default
> > > > > > > > >> > > > session
> > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> max(60s)? I
> > > > > thought
> > > > > > > the
> > > > > > > > >> min/max
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound
> > of
> > > it,
> > > > > > no?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > > 30s The
> > > > > > > > >> timeout to
> > > > > > > > >> > > > detect
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > > protocol.
> > > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > > int 45s
> > > > > > The
> > > > > > > > >> minimum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > > int 60s
> > > > > > The
> > > > > > > > >> maximum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> > are
> > > > > > [range,
> > > > > > > > >> uniform],
> > > > > > > > >> > > > > > which means
> > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > > like to
> > > > > > > know
> > > > > > > > >> why not
> > > > > > > > >> > > > > > uniform
> > > > > > > > >> > > > > > > > > one? I
> > > > > > > > >> > > > > > > > > > > thought usually users will choose
> > uniform
> > > > > > assignor
> > > > > > > > >> (former
> > > > > > > > >> > > > sticky
> > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> > other
> > > > > reason
> > > > > > > we
> > > > > > > > >> choose
> > > > > > > > >> > > > range
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > uniform
> > > > > The
> > > > > > > > >> server side
> > > > > > > > >> > > > > > assignors.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke
> > Chen
> > > <
> > > > > > > > >> showuon@gmail.com
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> > Connect
> > > > > > > > >> integrating with
> > > > > > > > >> > > > > > KIP-848,
> > > > > > > > >> > > > > > > > > but I
> > > > > > > > >> > > > > > > > > > > > think we should have a separate
> > > discussion
> > > > > > > thread
> > > > > > > > >> for the
> > > > > > > > >> > > > Kafka
> > > > > > > > >> > > > > > > > > Connect
> > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With
> > New
> > > > > > Consumer
> > > > > > > > >> Rebalance
> > > > > > > > >> > > > > > Protocol
> > > > > > > > >> > > > > > > > > [1],
> > > > > > > > >> > > > > > > > > > > > and let this discussion thread focus
> > on
> > > > > > consumer
> > > > > > > > >> rebalance
> > > > > > > > >> > > > > > protocol,
> > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > > >
> > > > >
> > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM
> Sagar
> > <
> > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > > feedback.
> > > > > > > Looks
> > > > > > > > >> like
> > > > > > > > >> > > > there's
> > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> > > would
> > > > > > > revisit
> > > > > > > > >> the doc
> > > > > > > > >> > > > and
> > > > > > > > >> > > > > > see
> > > > > > > > >> > > > > > > > > what
> > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM
> David
> > > Jacot
> > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > > document.
> > > > > > > That's
> > > > > > > > >> really
> > > > > > > > >> > > > > > helpful. I
> > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> > > Connect
> > > > > > > and the
> > > > > > > > >> > > > Consumer
> > > > > > > > >> > > > > > could
> > > > > > > > >> > > > > > > > > share
> > > > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> > > main
> > > > > > > > >> difference is
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> members
> > > > > whereas
> > > > > > > > >> Connect
> > > > > > > > >> > > > assigns
> > > > > > > > >> > > > > > tasks
> > > > > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > > forward:
> > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed
> APIs
> > to
> > > > > > support
> > > > > > > > >> different
> > > > > > > > >> > > > > > resource
> > > > > > > > >> > > > > > > > > types
> > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.);
> or
> > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > > Connect.
> > > > > > The
> > > > > > > > >> dedicated
> > > > > > > > >> > > > APIs
> > > > > > > > >> > > > > > > > > would be
> > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > different on
> > > > > > the
> > > > > > > > >> > > > > > content/resources and
> > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> engine
> > > on the
> > > > > > > > >> coordinator
> > > > > > > > >> > > > side.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> > because
> > > I am
> > > > > > > not a
> > > > > > > > >> fan of
> > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> > > That
> > > > > > being
> > > > > > > > >> said, I am
> > > > > > > > >> > > > not
> > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way
> to
> > > do
> > > > > it.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > > discuss
> > > > > it
> > > > > > > here
> > > > > > > > >> for now
> > > > > > > > >> > > > in
> > > > > > > > >> > > > > > order
> > > > > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> compatible
> > > with
> > > > > > what
> > > > > > > we
> > > > > > > > >> will do
> > > > > > > > >> > > > for
> > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM
> > David
> > > > > Jacot
> > > > > > <
> > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will
> > go
> > > > > > through
> > > > > > > and
> > > > > > > > >> address
> > > > > > > > >> > > > > > your
> > > > > > > > >> > > > > > > > > comments
> > > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for
> > your
> > > > > > > feedback.
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> > > Gregory
> > > > > > > Harris
> > > > > > > > >> <
> > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > > wonderful to
> > > > > > see
> > > > > > > > >> > > > cooperative
> > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> Alternatives'
> > > > > section
> > > > > > > > >> describes how
> > > > > > > > >> > > > > > member
> > > > > > > > >> > > > > > > > > epoch
> > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > >> > > > > > > > > > > >> > > > advance in step with the
> group
> > > epoch
> > > > > > and
> > > > > > > > >> assignment
> > > > > > > > >> > > > > > epoch
> > > > > > > > >> > > > > > > > > values. I
> > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for
> > the
> > > > > > reasons
> > > > > > > > >> described
> > > > > > > > >> > > > in
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > > assigning
> > > > > > > > >> partitions to a
> > > > > > > > >> > > > > > worker,
> > > > > > > > >> > > > > > > > > what
> > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > > assignment
> > > > > > > use?
> > > > > > > > >> Are
> > > > > > > > >> > > > member
> > > > > > > > >> > > > > > epochs
> > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > > correspond
> > > > > to
> > > > > > > > >> multiple
> > > > > > > > >> > > > > > different
> > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> > > field
> > > > > > > opaque
> > > > > > > > >> to the
> > > > > > > > >> > > > group
> > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> client-side
> > > > > assignor
> > > > > > > > >> > > > implementations
> > > > > > > > >> > > > > > > > > interact
> > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> > > common
> > > > > > > meaning
> > > > > > > > >> agreed
> > > > > > > > >> > > > > > upon? If
> > > > > > > > >> > > > > > > > > so, what
> > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> > > field
> > > > > over
> > > > > > > > >> including
> > > > > > > > >> > > > such
> > > > > > > > >> > > > > > > > > functionality
> > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is included
> in
> > > the
> > > > > > KIP:
> > > > > > > > >> "Thanks to
> > > > > > > > >> > > > > > this, the
> > > > > > > > >> > > > > > > > > input
> > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> > entirely
> > > > > driven
> > > > > > > by
> > > > > > > > >> the group
> > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > responsible for
> > > > > > > > >> maintaining
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > state
> > > > > > > > >> > > > > > > > > besides
> > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does
> this
> > > mean
> > > > > > > that the
> > > > > > > > >> > > > > > client-side
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > non-Metadata
> > > > > > state
> > > > > > > > >> (such as
> > > > > > > > >> > > > > > partition
> > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > > topics,
> > > > > > etc),
> > > > > > > or
> > > > > > > > >> that
> > > > > > > > >> > > > > > additional
> > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> > separate
> > > > > > classes
> > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > >> > > > > >
> > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > >> > > > > > that
> > > > > > > > >> > > > > > > > > seem to
> > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > > possible
> > > > > > for
> > > > > > > > >> these two
> > > > > > > > >> > > > > > > > > implementations
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> > > promote
> > > > > > > feature
> > > > > > > > >> parity
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > > server-side
> > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and
> > would
> > > also
> > > > > > > > >> facilitate
> > > > > > > > >> > > > > > operational
> > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> > example,
> > > if a
> > > > > > > > >> server-side
> > > > > > > > >> > > > > > assignor
> > > > > > > > >> > > > > > > > > has some
> > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > deploying
> > > > > > the
> > > > > > > > >> patched
> > > > > > > > >> > > > > > assignor to
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer
> > group
> > > to a
> > > > > > > > >> client-side
> > > > > > > > >> > > > > > assignor
> > > > > > > > >> > > > > > > > > may be
> > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > >> > > > > > > > > > > >> > > > and less risky than patching
> > all
> > > of
> > > > > the
> > > > > > > > >> brokers.
> > > > > > > > >> > > > With
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > currently
> > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > > non-trivial
> > > > > > > > >> > > > reimplementation
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > have
> > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two
> APIs
> > > have
> > > > > > > diverged
> > > > > > > > >> > > > > > significantly,
> > > > > > > > >> > > > > > > > > then it
> > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> > reimplementation
> > > > > would
> > > > > > > not be
> > > > > > > > >> > > > possible.
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39
> AM
> > > Sagar
> > > > > <
> > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page
> > to
> > > > > > discuss
> > > > > > > how
> > > > > > > > >> Connect
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > need to
> > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > > > protocol.
> > > > > > > > >> Here's the
> > > > > > > > >> > > > page:
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > > >
> > > > >
> > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish
> and
> > I
> > > have
> > > > > > > tried
> > > > > > > > >> to keep
> > > > > > > > >> > > > a
> > > > > > > > >> > > > > > format
> > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what
> you
> > > > > think.
> > > > > > > Also,
> > > > > > > > >> do you
> > > > > > > > >> > > > > > think this
> > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> > discussion
> > > > > thread
> > > > > > > or
> > > > > > > > >> is this
> > > > > > > > >> > > > one
> > > > > > > > >> > > > > > fine?
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at
> 7:37
> > AM
> > > > > > Sagar <
> > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> > > doc on
> > > > > > > Kafka
> > > > > > > > >> Streams.
> > > > > > > > >> > > > > > Sure, I
> > > > > > > > >> > > > > > > > > would do
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> > > such a
> > > > > > > > >> document.
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at
> > 4:47
> > > AM
> > > > > > > Guozhang
> > > > > > > > >> Wang <
> > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> > > could
> > > > > > come
> > > > > > > > >> back with
> > > > > > > > >> > > > some
> > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect
> side
> > > > > > > integration
> > > > > > > > >> with
> > > > > > > > >> > > > the new
> > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on
> the
> > > new
> > > > > > > "protocol
> > > > > > > > >> type"
> > > > > > > > >> > > > we
> > > > > > > > >> > > > > > did not
> > > > > > > > >> > > > > > > > > yet
> > > > > > > > >> > > > > > > > > > > >> think
> > > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > > implementations.
> > > > > > > For
> > > > > > > > >> Streams
> > > > > > > > >> > > > > > here's a
> > > > > > > > >> > > > > > > > > draft of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > > >
> > > > >
> > >
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your
> analysis
> > > on
> > > > > > > Connect.
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at
> > > 10:48 PM
> > > > > > > Sagar <
> > > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > > response.
> > > > > The
> > > > > > > > >> reason I
> > > > > > > > >> > > > thought
> > > > > > > > >> > > > > > > > > connect can
> > > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is
> that
> > > even
> > > > > > > today the
> > > > > > > > >> > > > connect
> > > > > > > > >> > > > > > uses a
> > > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > > AbstractCoordinator to
> > > > > > > > >> empower
> > > > > > > > >> > > > > > > > > rebalances of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > > > > > >> WorkerCoordinator sets
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the
> metadata()
> > > method
> > > > > > by
> > > > > > > > >> plumbing
> > > > > > > > >> > > > into
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> JoinGroupRequestProtocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to
> > > support
> > > > > > > > >> connect would
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > > > > similar at a
> > > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams
> > > mainly
> > > > > > > because
> > > > > > > > >> of the
> > > > > > > > >> > > > > > Client
> > > > > > > > >> > > > > > > > > side
> > > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > > > implementation
> > > > > > > > >> level, we
> > > > > > > > >> > > > > > might
> > > > > > > > >> > > > > > > > > need to
> > > > > > > > >> > > > > > > > > > > >> make
> > > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto
> this
> > > new
> > > > > > > > >> assignment
> > > > > > > > >> > > > protocol
> > > > > > > > >> > > > > > like
> > > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > > request/response and
> > > > > > > > >> SyncGroup and
> > > > > > > > >> > > > > > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on
> similar
> > > lines
> > > > > > to
> > > > > > > > >> streams (or
> > > > > > > > >> > > > > > there
> > > > > > > > >> > > > > > > > > might be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> > > > > detailed
> > > > > > > > >> analysis of
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > same
> > > > > > > > >> > > > > > > > > and we
> > > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> > > thread for
> > > > > > > that
> > > > > > > > >> as that
> > > > > > > > >> > > > would
> > > > > > > > >> > > > > > > > > derail this
> > > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if
> > > that
> > > > > > sounds
> > > > > > > > >> good to
> > > > > > > > >> > > > you.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022
> at
> > > 5:47
> > > > > PM
> > > > > > > David
> > > > > > > > >> Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > >
> <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > > comments.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers
> to
> > > > > > > > >> `Assignment#error`.
> > > > > > > > >> > > > > > Sure, I
> > > > > > > > >> > > > > > > > > can
> > > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > > > transition C
> > > > > > > from
> > > > > > > > >> his
> > > > > > > > >> > > > current
> > > > > > > > >> > > > > > > > > assignment
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment
> when
> > > he
> > > > > can
> > > > > > > move
> > > > > > > > >> to
> > > > > > > > >> > > > epoch 3.
> > > > > > > > >> > > > > > When
> > > > > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is
> > > updated
> > > > > > and
> > > > > > > > >> persisted
> > > > > > > > >> > > > > > with all
> > > > > > > > >> > > > > > > > > its
> > > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if
> > they
> > > are
> > > > > > not
> > > > > > > all
> > > > > > > > >> revoked
> > > > > > > > >> > > > > > yet. In
> > > > > > > > >> > > > > > > > > other
> > > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > > becomes
> > > > > the
> > > > > > > target
> > > > > > > > >> > > > > > assignment.
> > > > > > > > >> > > > > > > > > This is
> > > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to
> avoid
> > > having
> > > > > > to
> > > > > > > > >> write all
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > changes to
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are based
> on
> > > the
> > > > > > > persisted
> > > > > > > > >> state
> > > > > > > > >> > > > so I
> > > > > > > > >> > > > > > > > > understand
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me
> see
> > > if I
> > > > > can
> > > > > > > > >> improve
> > > > > > > > >> > > > this in
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding
> Connect,
> > it
> > > > > could
> > > > > > > > >> reuse the
> > > > > > > > >> > > > > > protocol
> > > > > > > > >> > > > > > > > > with a
> > > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits
> in
> > > the
> > > > > > > > >> protocol. The
> > > > > > > > >> > > > > > assignment
> > > > > > > > >> > > > > > > > > is
> > > > > > > > >> > > > > > > > > > > >> about
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> > > > > metadata,
> > > > > > > could
> > > > > > > > >> > > > Connect
> > > > > > > > >> > > > > > fit
> > > > > > > > >> > > > > > > > > into this?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022
> > at
> > > 1:55
> > > > > > PM
> > > > > > > > >> Sagar <
> > > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the
> KIP.
> > I
> > > just
> > > > > > had
> > > > > > > > >> minor
> > > > > > > > >> > > > > > observations:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the
> Assignment
> > > Error
> > > > > > > > >> section in
> > > > > > > > >> > > > Client
> > > > > > > > >> > > > > > Side
> > > > > > > > >> > > > > > > > > mode
> > > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned =>
> `In
> > > this
> > > > > > > case,
> > > > > > > > >> the
> > > > > > > > >> > > > client
> > > > > > > > >> > > > > > side
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > >> can
> > > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > > > > > coordinator`.
> > > > > > > > >> In this
> > > > > > > > >> > > > > > case are
> > > > > > > > >> > > > > > > > > you
> > > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning
> > an
> > > > > > > > >> AssignmentError
> > > > > > > > >> > > > that's
> > > > > > > > >> > > > > > > > > listed down
> > > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you
> think
> > > it
> > > > > > would
> > > > > > > > >> make sense
> > > > > > > > >> > > > to
> > > > > > > > >> > > > > > > > > mention this
> > > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case
> > Studies
> > > > > > > section, I
> > > > > > > > >> have a
> > > > > > > > >> > > > > > slight
> > > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the
> > same.
> > > > > > Consider
> > > > > > > > >> this step:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats,
> > the
> > > > > group
> > > > > > > > >> coordinator
> > > > > > > > >> > > > > > > > > transitions him
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no
> partitions
> > to
> > > > > > > revoke. It
> > > > > > > > >> > > > persists
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > change and
> > > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group
> (epoch=3)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> > Assignment
> > > > > > > (epoch=3)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > > partitions=[foo-0]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > > partitions=[foo-2]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > > partitions=[foo-1]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> > Assignment
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> epoch=2,
> > > > > > > > >> partitions=[foo-0,
> > > > > > > > >> > > > > > foo-1]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> epoch=3,
> > > > > > > > >> partitions=[foo-2]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> epoch=3,
> > > > > > > > >> partitions=[foo-1]
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats,
> > it
> > > > > > > transitions
> > > > > > > > >> to
> > > > > > > > >> > > > epoch 3
> > > > > > > > >> > > > > > but
> > > > > > > > >> > > > > > > > > cannot
> > > > > > > > >> > > > > > > > > > > >> get
> > > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's
> mentioned
> > > that
> > > > > > > member C
> > > > > > > > >> can't
> > > > > > > > >> > > > get
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > foo-1
> > > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > > description
> > > > > > > above,
> > > > > > > > >> it seems
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > > > already has
> > > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to
> > > remove
> > > > > it
> > > > > > > and
> > > > > > > > >> > > > populate it
> > > > > > > > >> > > > > > only
> > > > > > > > >> > > > > > > > > when it
> > > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a
> lot
> > of
> > > > > other
> > > > > > > > >> places, so
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > I
> > > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect
> ,
> > it
> > > > > might
> > > > > > > be
> > > > > > > > >> out of
> > > > > > > > >> > > > scope
> > > > > > > > >> > > > > > of
> > > > > > > > >> > > > > > > > > this
> > > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood
> it
> > > would
> > > > > > > > >> probably be
> > > > > > > > >> > > > > > running in
> > > > > > > > >> > > > > > > > > client
> > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> > > rebalance
> > > > > > > > >> protocol as it
> > > > > > > > >> > > > has
> > > > > > > > >> > > > > > its
> > > > > > > > >> > > > > > > > > own
> > > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > > IncrementalCooperative).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15,
> 2022
> > > at
> > > > > 5:00
> > > > > > > PM
> > > > > > > > >> David
> > > > > > > > >> > > > Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector!
> Our
> > > goal
> > > > > is
> > > > > > > to
> > > > > > > > >> move
> > > > > > > > >> > > > forward
> > > > > > > > >> > > > > > with
> > > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of
> relying
> > > on
> > > > > one
> > > > > > > > >> generic API.
> > > > > > > > >> > > > For
> > > > > > > > >> > > > > > > > > Connect, we
> > > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same
> pattern
> > > and
> > > > > > > > >> reuse/share the
> > > > > > > > >> > > > core
> > > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For
> > the
> > > > > > schema
> > > > > > > > >> registry, I
> > > > > > > > >> > > > > > think
> > > > > > > > >> > > > > > > > > that we
> > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a
> tailored
> > > API to
> > > > > > do
> > > > > > > > >> simple
> > > > > > > > >> > > > > > > > > membership/leader
> > > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15,
> > > 2022 at
> > > > > > > 10:22
> > > > > > > > >> AM Ismael
> > > > > > > > >> > > > > > Juma <
> > > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> > > comments:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > > > >> java.util.regex.Pattern
> > > > > > > > >> > > > vs
> > > > > > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern,
> > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > > > differences
> > > > > > > in
> > > > > > > > >> more
> > > > > > > > >> > > > detail
> > > > > > > > >> > > > > > before
> > > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if
> > > people
> > > > > > pass
> > > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to
> be
> > > > > honored.
> > > > > > > If
> > > > > > > > >> we are
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > > > > something
> > > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should
> consider
> > > adding
> > > > > > an
> > > > > > > > >> overload
> > > > > > > > >> > > > with
> > > > > > > > >> > > > > > our own
> > > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd
> want
> > to
> > > > > > expose
> > > > > > > > >> re2j's at
> > > > > > > > >> > > > this
> > > > > > > > >> > > > > > > > > point).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding
> > topic
> > > > > ids,
> > > > > > > any
> > > > > > > > >> major new
> > > > > > > > >> > > > > > protocol
> > > > > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should
> > > handle
> > > > > the
> > > > > > > topic
> > > > > > > > >> > > > > > recreation case
> > > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need
> to
> > > > > handle.
> > > > > > I
> > > > > > > > >> agree with
> > > > > > > > >> > > > > > David
> > > > > > > > >> > > > > > > > > that we'd
> > > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the
> relevant
> > > > > > protocols
> > > > > > > > >> that don't
> > > > > > > > >> > > > > > have it
> > > > > > > > >> > > > > > > > > yet and
> > > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> > > internals
> > > > > > > versus
> > > > > > > > >> adding
> > > > > > > > >> > > > new
> > > > > > > > >> > > > > > APIs
> > > > > > > > >> > > > > > > > > to the
> > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> > > adding
> > > > > new
> > > > > > > APIs
> > > > > > > > >> is
> > > > > > > > >> > > > required
> > > > > > > > >> > > > > > for
> > > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still
> > not
> > > sure
> > > > > > > about
> > > > > > > > >> the
> > > > > > > > >> > > > > > coordinator
> > > > > > > > >> > > > > > > > > storing
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> > > configs
> > > > > to
> > > > > > be
> > > > > > > > >> > > > centralized
> > > > > > > > >> > > > > > in the
> > > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > > (auditability,
> > > > > > > > >> visibility,
> > > > > > > > >> > > > > > consistency,
> > > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > > automatically
> > > > > > > > >> deleting
> > > > > > > > >> > > > > > configs in a
> > > > > > > > >> > > > > > > > > way
> > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A
> > good
> > > > > > property
> > > > > > > > >> for modern
> > > > > > > > >> > > > > > systems
> > > > > > > > >> > > > > > > > > is to
> > > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable
> > data
> > > > > loss
> > > > > > > > >> scenarios.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul
> 13,
> > > 2022
> > > > > at
> > > > > > > 3:47
> > > > > > > > >> PM David
> > > > > > > > >> > > > > > Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > > Guozhang. My
> > > > > > > answers
> > > > > > > > >> are
> > > > > > > > >> > > > below:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > migration
> > > > > > path,
> > > > > > > > >> especially
> > > > > > > > >> > > > > > the last
> > > > > > > > >> > > > > > > > > step
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable
> > the
> > > new
> > > > > > > > >> protocol, in
> > > > > > > > >> > > > which
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols
> /
> > > rpcs
> > > > > and
> > > > > > > old
> > > > > > > > >> > > > protocols /
> > > > > > > > >> > > > > > rpcs
> > > > > > > > >> > > > > > > > > are
> > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How
> > the
> > > > > > > > >> coordinator could
> > > > > > > > >> > > > > > "mimic"
> > > > > > > > >> > > > > > > > > the old
> > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> protocol
> > > is
> > > > > > > something
> > > > > > > > >> we
> > > > > > > > >> > > > need to
> > > > > > > > >> > > > > > > > > present
> > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I
> just
> > > > > > published
> > > > > > > a
> > > > > > > > >> new
> > > > > > > > >> > > > version
> > > > > > > > >> > > > > > of KIP
> > > > > > > > >> > > > > > > > > which
> > > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details
> about
> > > this.
> > > > > > See
> > > > > > > the
> > > > > > > > >> > > > "Supporting
> > > > > > > > >> > > > > > > > > Online
> > > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > > "Compatibility,
> > > > > > > > >> > > > Deprecation,
> > > > > > > > >> > > > > > and
> > > > > > > > >> > > > > > > > > Migration
> > > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to
> > think
> > > > > > through
> > > > > > > a
> > > > > > > > >> few cases
> > > > > > > > >> > > > > > now but
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> > > should be
> > > > > > > > >> understandable.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> usage
> > > of
> > > > > > topic
> > > > > > > > >> ids. So
> > > > > > > > >> > > > far as
> > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> RPCs
> > > and
> > > > > > admin
> > > > > > > > >> client,
> > > > > > > > >> > > > but
> > > > > > > > >> > > > > > they
> > > > > > > > >> > > > > > > > > are not
> > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> consumers
> > > yet.
> > > > > I
> > > > > > > think
> > > > > > > > >> the
> > > > > > > > >> > > > > > question is,
> > > > > > > > >> > > > > > > > > first
> > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to
> be
> > > > > > > maintaining
> > > > > > > > >> the
> > > > > > > > >> > > > names
> > > > > > > > >> > > > > > -> ids
> > > > > > > > >> > > > > > > > > mapping
> > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > > augmented
> > > > > > > > >> existing
> > > > > > > > >> > > > RPCs
> > > > > > > > >> > > > > > and the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > > should
> > > > > we
> > > > > > > ever
> > > > > > > > >> consider
> > > > > > > > >> > > > > > exposing
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> APIs
> > as
> > > > > well
> > > > > > > (both
> > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener
> for
> > > cases
> > > > > > > like
> > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming
> > > that we
> > > > > > > would
> > > > > > > > >> include
> > > > > > > > >> > > > > > converting
> > > > > > > > >> > > > > > > > > all
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to
> using
> > > topic
> > > > > > ids
> > > > > > > in
> > > > > > > > >> this
> > > > > > > > >> > > > KIP,
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully
> operate
> > > with
> > > > > > topic
> > > > > > > > >> ids. That
> > > > > > > > >> > > > > > being
> > > > > > > > >> > > > > > > > > said, it
> > > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > names in
> > > > > > > various
> > > > > > > > >> APIs so
> > > > > > > > >> > > > > > having a
> > > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > > inevitable to
> > > > > > me.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't
> > have
> > > a
> > > > > > strong
> > > > > > > > >> opinion on
> > > > > > > > >> > > > > > this.
> > > > > > > > >> > > > > > > > > Here I
> > > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> > > scope of
> > > > > > this
> > > > > > > > >> KIP. I
> > > > > > > > >> > > > would
> > > > > > > > >> > > > > > rather
> > > > > > > > >> > > > > > > > > focus
> > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we
> > can
> > > > > > consider
> > > > > > > > >> this
> > > > > > > > >> > > > > > separately if
> > > > > > > > >> > > > > > > > > we see
> > > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back
> to
> > > > > > Ismael's
> > > > > > > > >> point about
> > > > > > > > >> > > > > > using
> > > > > > > > >> > > > > > > > > topic ids
> > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > > >> > > > think
> > > > > > > > >> > > > > > that
> > > > > > > > >> > > > > > > > > there
> > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of
> it.
> > > The
> > > > > > > consumer
> > > > > > > > >> will
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > > exists
> > > > > > before
> > > > > > > > >> passing
> > > > > > > > >> > > > them
> > > > > > > > >> > > > > > into
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> group
> > > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> > > Obviously,
> > > > > > the
> > > > > > > > >> > > > coordinator
> > > > > > > > >> > > > > > will
> > > > > > > > >> > > > > > > > > also
> > > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really
> > have
> > > a
> > > > > way
> > > > > > to
> > > > > > > > >> reject an
> > > > > > > > >> > > > > > invalid
> > > > > > > > >> > > > > > > > > topic in
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> agreeing
> > > with
> > > > > > > David
> > > > > > > > >> on all
> > > > > > > > >> > > > other
> > > > > > > > >> > > > > > minor
> > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > `subscribe(Pattern)`
> > > > > > > > >> question:
> > > > > > > > >> > > > > > personally
> > > > > > > > >> > > > > > > > > I think
> > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate
> > the
> > > > > > > subscribe
> > > > > > > > >> API with
> > > > > > > > >> > > > > > Pattern,
> > > > > > > > >> > > > > > > > > but
> > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > > documenting
> > > > > > > > >> that our
> > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > > incompatible
> > > > > > > > >> case is a
> > > > > > > > >> > > > > > very rare
> > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > > `String`
> > > > > > > based
> > > > > > > > >> > > > > > subscription may
> > > > > > > > >> > > > > > > > > be
> > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > regexes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> > > work. I
> > > > > > have
> > > > > > > to
> > > > > > > > >> look at
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > differences
> > > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> > > better
> > > > > > > > >> understand the
> > > > > > > > >> > > > > > potential
> > > > > > > > >> > > > > > > > > issues.
> > > > > > > > >> > > > > > > > > > > >> My
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would
> > work
> > > for
> > > > > > all
> > > > > > > the
> > > > > > > > >> basic
> > > > > > > > >> > > > > > regular
> > > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the
> > two
> > > are
> > > > > > > mainly
> > > > > > > > >> about
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > various
> > > > > > > > >> > > > > > > > > > > >> character
> > > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what
> > > other
> > > > > > people
> > > > > > > > >> think
> > > > > > > > >> > > > about
> > > > > > > > >> > > > > > this.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul
> > 12,
> > > 2022
> > > > > > at
> > > > > > > > >> 11:28 PM
> > > > > > > > >> > > > > > Guozhang
> > > > > > > > >> > > > > > > > > Wang <
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks
> > David!
> > > I
> > > > > > think
> > > > > > > on
> > > > > > > > >> the high
> > > > > > > > >> > > > > > level
> > > > > > > > >> > > > > > > > > there are
> > > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> > concretize
> > > a
> > > > > bit
> > > > > > > more:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > > migration
> > > > > > path,
> > > > > > > > >> especially
> > > > > > > > >> > > > > > the last
> > > > > > > > >> > > > > > > > > step
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable
> > the
> > > new
> > > > > > > > >> protocol, in
> > > > > > > > >> > > > which
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols
> /
> > > rpcs
> > > > > and
> > > > > > > old
> > > > > > > > >> > > > protocols /
> > > > > > > > >> > > > > > rpcs
> > > > > > > > >> > > > > > > > > are
> > > > > > > > >> > > > > > > > > > > >> used
> > > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How
> > the
> > > > > > > > >> coordinator could
> > > > > > > > >> > > > > > "mimic"
> > > > > > > > >> > > > > > > > > the old
> > > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new
> protocol
> > > is
> > > > > > > something
> > > > > > > > >> we
> > > > > > > > >> > > > need to
> > > > > > > > >> > > > > > > > > present
> > > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the
> usage
> > > of
> > > > > > topic
> > > > > > > > >> ids. So
> > > > > > > > >> > > > far as
> > > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of
> RPCs
> > > and
> > > > > > admin
> > > > > > > > >> client,
> > > > > > > > >> > > > but
> > > > > > > > >> > > > > > they
> > > > > > > > >> > > > > > > > > are not
> > > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> consumers
> > > yet.
> > > > > I
> > > > > > > think
> > > > > > > > >> the
> > > > > > > > >> > > > > > question is,
> > > > > > > > >> > > > > > > > > first
> > > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to
> be
> > > > > > > maintaining
> > > > > > > > >> the
> > > > > > > > >> > > > names
> > > > > > > > >> > > > > > -> ids
> > > > > > > > >> > > > > > > > > mapping
> > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > > augmented
> > > > > > > > >> existing
> > > > > > > > >> > > > RPCs
> > > > > > > > >> > > > > > and the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > > should
> > > > > we
> > > > > > > ever
> > > > > > > > >> consider
> > > > > > > > >> > > > > > exposing
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public
> APIs
> > as
> > > > > well
> > > > > > > (both
> > > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > > >> > > > > > > > > > > >> well
> > > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener
> for
> > > cases
> > > > > > > like
> > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm
> agreeing
> > > with
> > > > > > > David
> > > > > > > > >> on all
> > > > > > > > >> > > > other
> > > > > > > > >> > > > > > minor
> > > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > `subscribe(Pattern)`
> > > > > > > > >> question:
> > > > > > > > >> > > > > > personally
> > > > > > > > >> > > > > > > > > I think
> > > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate
> > the
> > > > > > > subscribe
> > > > > > > > >> API with
> > > > > > > > >> > > > > > Pattern,
> > > > > > > > >> > > > > > > > > but
> > > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > > documenting
> > > > > > > > >> that our
> > > > > > > > >> > > > > > > > > subscription may
> > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > > incompatible
> > > > > > > > >> case is a
> > > > > > > > >> > > > > > very rare
> > > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > > `String`
> > > > > > > based
> > > > > > > > >> > > > > > subscription may
> > > > > > > > >> > > > > > > > > be
> > > > > > > > >> > > > > > > > > > > >> more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > > regexes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue,
> Jul
> > > 12,
> > > > > 2022
> > > > > > > at
> > > > > > > > >> 5:23 AM
> > > > > > > > >> > > > > > David Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi
> Ismael,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks
> for
> > > your
> > > > > > > > >> feedback. Let
> > > > > > > > >> > > > me
> > > > > > > > >> > > > > > answer
> > > > > > > > >> > > > > > > > > your
> > > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I
> > think
> > > > > it's
> > > > > > > > >> premature to
> > > > > > > > >> > > > > > talk about
> > > > > > > > >> > > > > > > > > > > >> target
> > > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> removal
> > > of the
> > > > > > > > >> existing group
> > > > > > > > >> > > > > > protocol.
> > > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> > > > > protocol
> > > > > > > and
> > > > > > > > >> hence
> > > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> > > quickly
> > > > > > > > >> applications
> > > > > > > > >> > > > migrate
> > > > > > > > >> > > > > > to
> > > > > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That
> makes
> > > > > sense.
> > > > > > I
> > > > > > > will
> > > > > > > > >> > > > remove it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The
> > KIP
> > > > > says
> > > > > > we
> > > > > > > > >> intend to
> > > > > > > > >> > > > > > release
> > > > > > > > >> > > > > > > > > this in
> > > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why.
> If
> > we
> > > > > added
> > > > > > > that
> > > > > > > > >> as a
> > > > > > > > >> > > > way to
> > > > > > > > >> > > > > > > > > estimate
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > group
> > > > > > > protocol, I
> > > > > > > > >> also
> > > > > > > > >> > > > > > suggest
> > > > > > > > >> > > > > > > > > removing
> > > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > explain
> > > > > my
> > > > > > > > >> reasoning. As
> > > > > > > > >> > > > > > > > > explained, I
> > > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > coordinator
> > > in
> > > > > > Java
> > > > > > > > >> while we
> > > > > > > > >> > > > > > implement
> > > > > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> > > > > internals
> > > > > > > will
> > > > > > > > >> be
> > > > > > > > >> > > > slightly
> > > > > > > > >> > > > > > > > > different
> > > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Therefore, I
> > > > > > wanted
> > > > > > > to
> > > > > > > > >> tighten
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > switch from
> > > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > coordinator
> > > to
> > > > > the
> > > > > > > new
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator
> > > > > > > > >> > > > > > > > > to a
> > > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > alternative
> > > > > would
> > > > > > > be to
> > > > > > > > >> use a
> > > > > > > > >> > > > flag
> > > > > > > > >> > > > > > to do
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> > > software
> > > > > > > upgrade.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We
> > > need to
> > > > > > > flesh
> > > > > > > > >> out the
> > > > > > > > >> > > > > > details of
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're
> > > saying
> > > > > we
> > > > > > > will
> > > > > > > > >> support
> > > > > > > > >> > > > > > online
> > > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> explain
> > > this
> > > > > in
> > > > > > > > >> detail. It
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > also
> > > > > > > > >> > > > > > > > > be done
> > > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> easier.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we
> > will
> > > > > > support
> > > > > > > > >> online
> > > > > > > > >> > > > > > migrations
> > > > > > > > >> > > > > > > > > for the
> > > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group
> > > using
> > > > > the
> > > > > > > old
> > > > > > > > >> protocol
> > > > > > > > >> > > > > > will be
> > > > > > > > >> > > > > > > > > able to
> > > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > briefly
> > > > > > > explain
> > > > > > > > >> how that
> > > > > > > > >> > > > > > will work
> > > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four
> step
> > > > > process:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> > > cluster
> > > > > > must
> > > > > > > be
> > > > > > > > >> > > > upgraded or
> > > > > > > > >> > > > > > > > > rolled to a
> > > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> group
> > > > > > > coordinator.
> > > > > > > > >> Both the
> > > > > > > > >> > > > > > old and
> > > > > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support
> > the
> > > old
> > > > > > > > >> protocol and
> > > > > > > > >> > > > rely
> > > > > > > > >> > > > > > on the
> > > > > > > > >> > > > > > > > > same
> > > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can
> > > work
> > > > > > > together.
> > > > > > > > >> This
> > > > > > > > >> > > > point
> > > > > > > > >> > > > > > is an
> > > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this
> > one
> > > live
> > > > > > > > >> because it
> > > > > > > > >> > > > would
> > > > > > > > >> > > > > > require
> > > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > coordinator
> > > and
> > > > > > > > >> starting up the
> > > > > > > > >> > > > > > new one
> > > > > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > unavailabilities.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> > > cluster's
> > > > > > > > >> metadata
> > > > > > > > >> > > > > > version/IBP
> > > > > > > > >> > > > > > > > > must be
> > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> enable
> > > the
> > > > > new
> > > > > > > > >> protocol.
> > > > > > > > >> > > > This
> > > > > > > > >> > > > > > cannot
> > > > > > > > >> > > > > > > > > be done
> > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> terminated
> > > > > because
> > > > > > > the
> > > > > > > > >> old
> > > > > > > > >> > > > > > coordinator
> > > > > > > > >> > > > > > > > > doesn't
> > > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> > > consumers
> > > > > > > must be
> > > > > > > > >> > > > upgraded
> > > > > > > > >> > > > > > to a
> > > > > > > > >> > > > > > > > > version
> > > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> migration
> > > (must
> > > > > > have
> > > > > > > > >> KIP-792).
> > > > > > > > >> > > > If
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing
> > > must be
> > > > > > > done at
> > > > > > > > >> this
> > > > > > > > >> > > > point.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> > > consumers
> > > > > > > must
> > > > > > > > >> be rolled
> > > > > > > > >> > > > > > with the
> > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer
> > > group
> > > > > is
> > > > > > > > >> automatically
> > > > > > > > >> > > > > > > > > converted when
> > > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using
> the
> > > new
> > > > > > > protocol
> > > > > > > > >> joins
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > group.
> > > > > > > > >> > > > > > > > > While
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > > protocol are
> > > > > > > being
> > > > > > > > >> > > > upgraded,
> > > > > > > > >> > > > > > the old
> > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > > clarify
> > > > > all
> > > > > > > of
> > > > > > > > >> this in
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > KIP.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I
> am
> > > happy
> > > > > > > that we
> > > > > > > > >> are
> > > > > > > > >> > > > > > pushing the
> > > > > > > > >> > > > > > > > > pattern
> > > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it
> > > seems
> > > > > > like
> > > > > > > > >> there
> > > > > > > > >> > > > could be
> > > > > > > > >> > > > > > some
> > > > > > > > >> > > > > > > > > tricky
> > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> > > > > mechanism
> > > > > > > for
> > > > > > > > >> users to
> > > > > > > > >> > > > > > detect
> > > > > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> they
> > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> > > > > switching
> > > > > > > to
> > > > > > > > >> the new
> > > > > > > > >> > > > > > protocol?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think
> > > that I
> > > > > am
> > > > > > a
> > > > > > > bit
> > > > > > > > >> more
> > > > > > > > >> > > > > > optimistic
> > > > > > > > >> > > > > > > > > than
> > > > > > > > >> > > > > > > > > > > >> you
> > > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe
> > > that the
> > > > > > > > >> majority of
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > cases
> > > > > > > > >> > > > > > > > > are
> > > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work
> with
> > > the
> > > > > new
> > > > > > > > >> engine. The
> > > > > > > > >> > > > > > > > > coordinator will
> > > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> reject
> > > the
> > > > > > > consumer
> > > > > > > > >> if the
> > > > > > > > >> > > > > > regex is
> > > > > > > > >> > > > > > > > > not
> > > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> migration
> > > path,
> > > > > in
> > > > > > > the
> > > > > > > > >> worst
> > > > > > > > >> > > > case,
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > first
> > > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> group
> > > will
> > > > > be
> > > > > > > > >> rejected.
> > > > > > > > >> > > > This
> > > > > > > > >> > > > > > should
> > > > > > > > >> > > > > > > > > be used
> > > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> say.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way
> > for
> > > > > > > customers to
> > > > > > > > >> > > > validate
> > > > > > > > >> > > > > > their
> > > > > > > > >> > > > > > > > > regex
> > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod
> would
> > > be to
> > > > > > > test
> > > > > > > > >> them with
> > > > > > > > >> > > > > > another
> > > > > > > > >> > > > > > > > > group.
> > > > > > > > >> > > > > > > > > > > >> For
> > > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be
> > > done
> > > > > in a
> > > > > > > > >> pre-prod
> > > > > > > > >> > > > > > environment.
> > > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend
> the
> > > > > > > > >> consumer-group tool
> > > > > > > > >> > > > to
> > > > > > > > >> > > > > > > > > provide a
> > > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> mechanism.
> > > Would
> > > > > > > this be
> > > > > > > > >> > > > enough in
> > > > > > > > >> > > > > > your
> > > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > > Related to
> > > > > > the
> > > > > > > last
> > > > > > > > >> > > > question,
> > > > > > > > >> > > > > > will
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick
> > > with the
> > > > > > > > >> current regex
> > > > > > > > >> > > > > > engine for
> > > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may
> > be
> > > > > handy
> > > > > > to
> > > > > > > > >> keep using
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > based
> > > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > migrations
> > > > > > simple
> > > > > > > and
> > > > > > > > >> then
> > > > > > > > >> > > > > > migrate to
> > > > > > > > >> > > > > > > > > server
> > > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> > understand
> > > > > your
> > > > > > > point
> > > > > > > > >> but I
> > > > > > > > >> > > > am
> > > > > > > > >> > > > > > > > > concerned that
> > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > actually
> > > stay
> > > > > > in
> > > > > > > > >> this mode.
> > > > > > > > >> > > > That
> > > > > > > > >> > > > > > > > > would go
> > > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > simplifying
> > > the
> > > > > > > client
> > > > > > > > >> because
> > > > > > > > >> > > > we
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > have to
> > > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > metadata on
> > > > > > the
> > > > > > > > >> client
> > > > > > > > >> > > > side. I
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > rather
> > > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6.
> When
> > > we say
> > > > > > > that
> > > > > > > > >> the group
> > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > >> > > > > > > > > > > >> will
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > > configurations
> > > > > > > > >> and that
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group
> is
> > > > > > deleted.
> > > > > > > > >> Will a
> > > > > > > > >> > > > > > transition to
> > > > > > > > >> > > > > > > > > DEAD
> > > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > configurations?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> > > right.
> > > > > The
> > > > > > > > >> > > > configurations
> > > > > > > > >> > > > > > will be
> > > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted.
> > > They go
> > > > > > > > >> together.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7.
> Will
> > > the
> > > > > > > choice to
> > > > > > > > >> store
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > configs in
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder
> > to
> > > list
> > > > > > all
> > > > > > > > >> cluster
> > > > > > > > >> > > > > > configs and
> > > > > > > > >> > > > > > > > > their
> > > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't
> > > think
> > > > > so.
> > > > > > > The
> > > > > > > > >> group
> > > > > > > > >> > > > > > > > > configurations are
> > > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs.
> > If
> > > you
> > > > > > > want to
> > > > > > > > >> know
> > > > > > > > >> > > > all
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > overrides
> > > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask
> all
> > > the
> > > > > > group
> > > > > > > > >> > > > coordinators.
> > > > > > > > >> > > > > > You
> > > > > > > > >> > > > > > > > > cannot
> > > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > instance.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How
> > > would
> > > > > > > someone
> > > > > > > > >> > > > configure a
> > > > > > > > >> > > > > > group
> > > > > > > > >> > > > > > > > > before
> > > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > > considered
> > > > > > > > >> allowing the
> > > > > > > > >> > > > > > explicit
> > > > > > > > >> > > > > > > > > creation
> > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> Alternatively,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> configs
> > > could
> > > > > be
> > > > > > > > >> decoupled
> > > > > > > > >> > > > from
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > group
> > > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The
> > > group
> > > > > > will
> > > > > > > be
> > > > > > > > >> > > > > > automatically
> > > > > > > > >> > > > > > > > > created in
> > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > configs
> > > will
> > > > > > be
> > > > > > > > >> lost after
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > retention
> > > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9.
> Will
> > > the
> > > > > > > > >> > > > Consumer.subscribe
> > > > > > > > >> > > > > > method
> > > > > > > > >> > > > > > > > > for the
> > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > > >> > > > do
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > have to
> > > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a
> > > very
> > > > > > group
> > > > > > > > >> question. I
> > > > > > > > >> > > > > > forgot
> > > > > > > > >> > > > > > > > > about
> > > > > > > > >> > > > > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > > >> > > > not
> > > > > > > > >> > > > > > fully
> > > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan
> to
> > > use,
> > > > > it
> > > > > > > > >> might be
> > > > > > > > >> > > > better
> > > > > > > > >> > > > > > to
> > > > > > > > >> > > > > > > > > deprecate
> > > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which
> > takes
> > > a
> > > > > > > string.
> > > > > > > > >> We would
> > > > > > > > >> > > > > > rely on
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> server
> > > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During
> the
> > > > > > > migration, I
> > > > > > > > >> think
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > could
> > > > > > > > >> > > > > > > > > > > >> still
> > > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> and
> > > use
> > > > > it.
> > > > > > > That
> > > > > > > > >> should
> > > > > > > > >> > > > > > work, I
> > > > > > > > >> > > > > > > > > think, in
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I
> > > agree
> > > > > with
> > > > > > > > >> Justine
> > > > > > > > >> > > > that we
> > > > > > > > >> > > > > > > > > should be
> > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch
> > to
> > > > > > > > >> > > > IBP/metadata.version
> > > > > > > > >> > > > > > from the
> > > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> > > switch the
> > > > > > > > >> default for
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > feature
> > > > > > > > >> > > > > > > > > flag
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > metadata.version
> > > > > > > once
> > > > > > > > >> we
> > > > > > > > >> > > > want to
> > > > > > > > >> > > > > > make
> > > > > > > > >> > > > > > > > > it the
> > > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan
> > was
> > > to
> > > > > use
> > > > > > > that
> > > > > > > > >> feature
> > > > > > > > >> > > > > > flag
> > > > > > > > >> > > > > > > > > mainly
> > > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I
> > > should
> > > > > > not
> > > > > > > have
> > > > > > > > >> > > > mentioned
> > > > > > > > >> > > > > > it, I
> > > > > > > > >> > > > > > > > > think,
> > > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> > internal
> > > > > config
> > > > > > > for
> > > > > > > > >> it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11.
> Some
> > > of
> > > > > the
> > > > > > > > >> protocol APIs
> > > > > > > > >> > > > > > don't
> > > > > > > > >> > > > > > > > > mention
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good
> to
> > > add
> > > > > that
> > > > > > > for
> > > > > > > > >> > > > consistency.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It
> > is
> > > a
> > > > > bit
> > > > > > > odd
> > > > > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though
> > it
> > > > > seems
> > > > > > > to do
> > > > > > > > >> more
> > > > > > > > >> > > > than
> > > > > > > > >> > > > > > > > > reading.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree.
> > > This is
> > > > > > > how the
> > > > > > > > >> > > > current
> > > > > > > > >> > > > > > > > > protocol works
> > > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require
> > > "Read
> > > > > > > Group" to
> > > > > > > > >> join a
> > > > > > > > >> > > > > > group. We
> > > > > > > > >> > > > > > > > > could
> > > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am
> > not
> > > > > sure
> > > > > > > that
> > > > > > > > >> it is
> > > > > > > > >> > > > worth
> > > > > > > > >> > > > > > it.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13.
> How
> > is
> > > > > topic
> > > > > > > > >> recreation
> > > > > > > > >> > > > > > handled by
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > protocol?
> > > It
> > > > > > > would be
> > > > > > > > >> good to
> > > > > > > > >> > > > > > have a
> > > > > > > > >> > > > > > > > > section
> > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > From
> > > a
> > > > > > > protocol
> > > > > > > > >> > > > > > perspective, the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id
> > so
> > > it
> > > > > > will
> > > > > > > > >> treat it
> > > > > > > > >> > > > like a
> > > > > > > > >> > > > > > > > > topic with
> > > > > > > > >> > > > > > > > > > > >> a
> > > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only
> issue
> > > is
> > > > > that
> > > > > > > the
> > > > > > > > >> > > > fetch/commit
> > > > > > > > >> > > > > > > > > offsets
> > > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so
> the
> > > > > > consumer
> > > > > > > > >> would
> > > > > > > > >> > > > reuse the
> > > > > > > > >> > > > > > > > > offsets
> > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > should
> > > > > > > update
> > > > > > > > >> those
> > > > > > > > >> > > > APIs
> > > > > > > > >> > > > > > as well
> > > > > > > > >> > > > > > > > > in
> > > > > > > > >> > > > > > > > > > > >> order
> > > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end.
> > That
> > > > > would
> > > > > > > > >> strengthen
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > semantics of
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14.
> The
> > > KIP
> > > > > > > mentions
> > > > > > > > >> we will
> > > > > > > > >> > > > > > write the
> > > > > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this
> is
> > an
> > > > > > > > >> implementation
> > > > > > > > >> > > > > > detail, do
> > > > > > > > >> > > > > > > > > we plan
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for
> it?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do
> > we
> > > > > have a
> > > > > > > > >> scalability
> > > > > > > > >> > > > > > goal when
> > > > > > > > >> > > > > > > > > it
> > > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group
> > > protocol
> > > > > > can
> > > > > > > > >> support?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't
> > > have
> > > > > > > numbers
> > > > > > > > >> at the
> > > > > > > > >> > > > > > moment. The
> > > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of
> members
> > > per
> > > > > > > group.
> > > > > > > > >> We will
> > > > > > > > >> > > > > > measure
> > > > > > > > >> > > > > > > > > this when
> > > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > implementation.
> > > > > > Note
> > > > > > > > >> that we
> > > > > > > > >> > > > might
> > > > > > > > >> > > > > > have
> > > > > > > > >> > > > > > > > > other
> > > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road
> (e.g.
> > > > > offset
> > > > > > > > >> commits).
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16.
> Did
> > we
> > > > > > > consider
> > > > > > > > >> having
> > > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > SubscribedTopicNames
> > > > > > > > >> in
> > > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since
> we
> > > have
> > > > > to
> > > > > > > > >> resolve the
> > > > > > > > >> > > > > > regex on
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > topic
> > > > > name?
> > > > > > > The
> > > > > > > > >> > > > difference
> > > > > > > > >> > > > > > is that
> > > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> efficient
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> whereas
> > > > > sending
> > > > > > > the
> > > > > > > > >> topic
> > > > > > > > >> > > > names
> > > > > > > > >> > > > > > is less
> > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > recreation is
> > > > > > > easier
> > > > > > > > >> to
> > > > > > > > >> > > > handle
> > > > > > > > >> > > > > > if we
> > > > > > > > >> > > > > > > > > have
> > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea
> > > was to
> > > > > > > > >> consolidate the
> > > > > > > > >> > > > > > metadata
> > > > > > > > >> > > > > > > > > lookup
> > > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths
> but
> > I
> > > do
> > > > > > agree
> > > > > > > > >> with your
> > > > > > > > >> > > > > > point. As
> > > > > > > > >> > > > > > > > > a
> > > > > > > > >> > > > > > > > > > > >> second
> > > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may
> be
> > > > > better
> > > > > > > here
> > > > > > > > >> for the
> > > > > > > > >> > > > > > delete and
> > > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose
> > > that we
> > > > > > may
> > > > > > > > >> allow
> > > > > > > > >> > > > users to
> > > > > > > > >> > > > > > > > > subscribe
> > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> > > because
> > > > > > that
> > > > > > > is
> > > > > > > > >> the only
> > > > > > > > >> > > > > > way to be
> > > > > > > > >> > > > > > > > > > > >> really
> > > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > re-creation.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue,
> > Jul
> > > 12,
> > > > > > > 2022 at
> > > > > > > > >> 1:38 PM
> > > > > > > > >> > > > > > David
> > > > > > > > >> > > > > > > > > Jacot <
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io
> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > > Justine,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks
> > for
> > > > > your
> > > > > > > > >> comments.
> > > > > > > > >> > > > Please
> > > > > > > > >> > > > > > find
> > > > > > > > >> > > > > > > > > my
> > > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes,
> > > the new
> > > > > > > > >> protocol
> > > > > > > > >> > > > relies on
> > > > > > > > >> > > > > > > > > topic IDs
> > > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic
> > > names
> > > > > > based
> > > > > > > in
> > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > ConsumerGroupHeartbeatRequest.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if
> using
> > > topic
> > > > > > > names
> > > > > > > > >> is the
> > > > > > > > >> > > > > > right call
> > > > > > > > >> > > > > > > > > here.
> > > > > > > > >> > > > > > > > > > > >> I
> > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a
> little
> > > more.
> > > > > > > > >> Obviously, the
> > > > > > > > >> > > > > > KIP does
> > > > > > > > >> > > > > > > > > not
> > > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> offsets
> > > RPCs
> > > > > to
> > > > > > > use
> > > > > > > > >> topic
> > > > > > > > >> > > > IDs.
> > > > > > > > >> > > > > > This
> > > > > > > > >> > > > > > > > > may be
> > > > > > > > >> > > > > > > > > > > >> > something
> > > > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> include
> > > though
> > > > > > as
> > > > > > > it
> > > > > > > > >> would
> > > > > > > > >> > > > give
> > > > > > > > >> > > > > > better
> > > > > > > > >> > > > > > > > > > > >> overall
> > > > > > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > producer.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > -
> You're
> > > > > right.
> > > > > > I
> > > > > > > > >> think that
> > > > > > > > >> > > > I
> > > > > > > > >> > > > > > should
> > > > > > > > >> > > > > > > > > not
> > > > > > > > >> > > > > > > > > > > >> have
> > > > > > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I
> > > will
> > > > > > remove
> > > > > > > > >> it. We can
> > > > > > > > >> > > > > > use an
> > > > > > > > >> > > > > > > > > internal
> > > > > > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > developing the
> > > > > > > > >> feature.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both
> > > cluster
> > > > > > > types
> > > > > > > > >> will be
> > > > > > > > >> > > > > > > > > supported. The
> > > > > > > > >> > > > > > > > > > > >> > change is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only
> > > > > requirement
> > > > > > > is
> > > > > > > > >> that the
> > > > > > > > >> > > > > > cluster
> > > > > > > > >> > > > > > > > > uses
> > > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > > >> > > > > > > > > > > >> > IDs.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On
> Mon,
> > > Jul
> > > > > 11,
> > > > > > > 2022
> > > > > > > > >> at 9:53
> > > > > > > > >> > > > PM
> > > > > > > > >> > > > > > > > > Guozhang
> > > > > > > > >> > > > > > > > > > > >> Wang <
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > wangguoz@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi
> > > Ismael,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> Thanks
> > > for
> > > > > the
> > > > > > > > >> feedback.
> > > > > > > > >> > > > Here
> > > > > > > > >> > > > > > are
> > > > > > > > >> > > > > > > > > some
> > > > > > > > >> > > > > > > > > > > >> replies
> > > > > > > > >> > > > > > > > > > > >> > > > > inlined
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On
> > Sat,
> > > Jul
> > > > > 9,
> > > > > > > 2022
> > > > > > > > >> at
> > > > > > > > >> > > > 2:53 AM
> > > > > > > > >> > > > > > > > > Ismael Juma
> > > > > > > > >> > > > > > > > > > > >> <
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > Thanks for
> > > > > > the
> > > > > > > > >> KIP. This
> > > > > > > > >> > > > has
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > > >> potential
> > > > > > > > >> > > > > > > > > > > >> > to be a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > improvement. A
> > > > > few
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > initial
> > > > > > > > >> > > > questions/comments:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 1. I
> > > think
> > > > > > > it's
> > > > > > > > >> > > > premature to
> > > > > > > > >> > > > > > talk
> > > > > > > > >> > > > > > > > > about
> > > > > > > > >> > > > > > > > > > > >> > target
> > > > > > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > deprecation
> > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > removal of
> > > > > > the
> > > > > > > > >> existing
> > > > > > > > >> > > > group
> > > > > > > > >> > > > > > > > > protocol.
> > > > > > > > >> > > > > > > > > > > >> > Unlike
> > > > > > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > client
> > > > > > > protocol
> > > > > > > > >> and hence
> > > > > > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> dependent
> > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> how
> > > > > quickly
> > > > > > > > >> applications
> > > > > > > > >> > > > > > migrate
> > > > > > > > >> > > > > > > > > to the
> > > > > > > > >> > > > > > > > > > > >> new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> Yeah I
> > > agree
> > > > > > > with
> > > > > > > > >> you. I
> > > > > > > > >> > > > think
> > > > > > > > >> > > > > > we can
> > > > > > > > >> > > > > > > > > > > >> remove
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > `Compatibility,
> > > > > > > > >> > > > Deprecation,
> > > > > > > > >> > > > > > and
> > > > > > > > >> > > > > > > > > Migration
> > > > > > > > >> > > > > > > > > > > >> > Plan` and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that
> > we
> > > will
> > > > > > > decide
> > > > > > > > >> in the
> > > > > > > > >> > > > > > future
> > > > > > > > >> > > > > > > > > about
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > we
> > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > protocol and
> > > > > > > > >> behaviors.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2.
> > > The KIP
> > > > > > > says we
> > > > > > > > >> > > > intend to
> > > > > > > > >> > > > > > > > > release this
> > > > > > > > >> > > > > > > > > > > >> > in 4.x,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> why.
> > > If we
> > > > > > > added
> > > > > > > > >> that as
> > > > > > > > >> > > > a
> > > > > > > > >> > > > > > way to
> > > > > > > > >> > > > > > > > > > > >> estimate
> > > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > > >> > > > > > > > > > > >> > > > > we'd
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> the
> > > group
> > > > > > > > >> protocol, I
> > > > > > > > >> > > > also
> > > > > > > > >> > > > > > suggest
> > > > > > > > >> > > > > > > > > > > >> removing
> > > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I
> > think
> > > > > that's
> > > > > > > not
> > > > > > > > >> > > > specifically
> > > > > > > > >> > > > > > > > > related to
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > deprecation/removal
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> plan,
> > > but
> > > > > it's
> > > > > > > more
> > > > > > > > >> for
> > > > > > > > >> > > > client
> > > > > > > > >> > > > > > > > > upgrades.
> > > > > > > > >> > > > > > > > > > > >> I.e.
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > implementation
> > > > > > > may
> > > > > > > > >> be done
> > > > > > > > >> > > > > > first,
> > > > > > > > >> > > > > > > > > and then
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only
> > > mark it
> > > > > > as
> > > > > > > > >> "released"
> > > > > > > > >> > > > by
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > time
> > > > > > > > >> > > > > > > > > > > >> clients
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that
> > > time,
> > > > > to
> > > > > > > > >> enable the
> > > > > > > > >> > > > > > feature the
> > > > > > > > >> > > > > > > > > > > >> clients
> > > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > bytecode
> > > > > with
> > > > > > a
> > > > > > > > >> rolling
> > > > > > > > >> > > > bounce
> > > > > > > > >> > > > > > and
> > > > > > > > >> > > > > > > > > then set
> > > > > > > > >> > > > > > > > > > > >> > the flag
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > bounce,
> > > and
> > > > > > > hence
> > > > > > > > >> we feel
> > > > > > > > >> > > > it's
> > > > > > > > >> > > > > > > > > better to be
> > > > > > > > >> > > > > > > > > > > >> > released
> > > > > > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3.
> > We
> > > need
> > > > > > to
> > > > > > > > >> flesh out
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > > > > details of
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > migration
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds
> > like
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > we're
> > > > > saying
> > > > > > > we
> > > > > > > > >> will
> > > > > > > > >> > > > support
> > > > > > > > >> > > > > > online
> > > > > > > > >> > > > > > > > > > > >> > migrations. Is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > explain
> > > > > this
> > > > > > > in
> > > > > > > > >> detail.
> > > > > > > > >> > > > It
> > > > > > > > >> > > > > > could
> > > > > > > > >> > > > > > > > > also be
> > > > > > > > >> > > > > > > > > > > >> > done as a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > easier.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes
> I
> > > think
> > > > > > > that's
> > > > > > > > >> the
> > > > > > > > >> > > > part we
> > > > > > > > >> > > > > > can
> > > > > > > > >> > > > > > > > > be more
> > > > > > > > >> > > > > > > > > > > >> > concrete
> > > > > > > > >> > > > > > > > > > > >> > > > > >> about
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this
> > is
> > > > > > related
> > > > > > > to
> > > > > > > > >> your
> > > > > > > > >> > > > > > question 2)
> > > > > > > > >> > > > > > > > > above).
> > > > > > > > >> > > > > > > > > > > >> > We will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> work
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > explicit in
> > > > > > > > >> parallel as we
> > > > > > > > >> > > > > > solicit
> > > > > > > > >> > > > > > > > > more
> > > > > > > > >> > > > > > > > > > > >> > feedback.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 4. I
> > > am
> > > > > > happy
> > > > > > > > >> that we are
> > > > > > > > >> > > > > > pushing
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > pattern
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> but
> > it
> > > > > seems
> > > > > > > like
> > > > > > > > >> there
> > > > > > > > >> > > > > > could be
> > > > > > > > >> > > > > > > > > some
> > > > > > > > >> > > > > > > > > > > >> tricky
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > have a
> > > > > > > mechanism
> > > > > > > > >> for
> > > > > > > > >> > > > users to
> > > > > > > > >> > > > > > > > > detect that
> > > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > > >> > > > > > > > > > > >> > > > > need
> > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > before
> > > > > > > switching
> > > > > > > > >> to the
> > > > > > > > >> > > > new
> > > > > > > > >> > > > > > > > > protocol?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes
> I
> > > think
> > > > > we
> > > > > > > need
> > > > > > > > >> some
> > > > > > > > >> > > > > > tooling for
> > > > > > > > >> > > > > > > > > > > >> non-java
> > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > "dry-run"
> > > > > the
> > > > > > > > >> client before
> > > > > > > > >> > > > > > > > > switching to
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > specific
> > > > > idea
> > > > > > on
> > > > > > > > >> top of my
> > > > > > > > >> > > > head
> > > > > > > > >> > > > > > > > > though,
> > > > > > > > >> > > > > > > > > > > >> maybe
> > > > > > > > >> > > > > > > > > > > >> > others
> > > > > > > > >> > > > > > > > > > > >> > > > > >> like
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett
> > can
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > chime-in
> > > > > here?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5.
> > > Related
> > > > > > to
> > > > > > > the
> > > > > > > > >> last
> > > > > > > > >> > > > > > question,
> > > > > > > > >> > > > > > > > > will the
> > > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > stick
> > > with
> > > > > > the
> > > > > > > > >> current
> > > > > > > > >> > > > regex
> > > > > > > > >> > > > > > > > > engine for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it
> > > may be
> > > > > > > handy
> > > > > > > > >> to keep
> > > > > > > > >> > > > using
> > > > > > > > >> > > > > > > > > client
> > > > > > > > >> > > > > > > > > > > >> based
> > > > > > > > >> > > > > > > > > > > >> > regex
> > > > > > > > >> > > > > > > > > > > >> > > > > at
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > migrations
> > > > > > > simple
> > > > > > > > >> and
> > > > > > > > >> > > > then
> > > > > > > > >> > > > > > migrate
> > > > > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > server based
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > step.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > Honestly I
> > > > > > have
> > > > > > > not
> > > > > > > > >> thought
> > > > > > > > >> > > > > > about
> > > > > > > > >> > > > > > > > > that for
> > > > > > > > >> > > > > > > > > > > >> > java
> > > > > > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> that.
> > > What
> > > > > > kind
> > > > > > > of
> > > > > > > > >> > > > > > compatibility
> > > > > > > > >> > > > > > > > > issues do
> > > > > > > > >> > > > > > > > > > > >> > you have
> > > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6.
> > > When we
> > > > > > say
> > > > > > > > >> that the
> > > > > > > > >> > > > group
> > > > > > > > >> > > > > > > > > coordinator
> > > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> the
> > > > > > > > >> configurations and
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > configurations
> > > > > > > > >> > > > > > > > > > > >> > > > > will
> > > > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > group
> > > is
> > > > > > > deleted.
> > > > > > > > >> Will a
> > > > > > > > >> > > > > > > > > transition to
> > > > > > > > >> > > > > > > > > > > >> DEAD
> > > > > > > > >> > > > > > > > > > > >> > > > > trigger
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > configurations?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes,
> > > since
> > > > > the
> > > > > > > DEAD
> > > > > > > > >> state
> > > > > > > > >> > > > is an
> > > > > > > > >> > > > > > > > > ending
> > > > > > > > >> > > > > > > > > > > >> state
> > > > > > > > >> > > > > > > > > > > >> > (we
> > > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> state
> > > when
> > > > > the
> > > > > > > > >> group is
> > > > > > > > >> > > > EMPTY
> > > > > > > > >> > > > > > and
> > > > > > > > >> > > > > > > > > also all
> > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > its
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's
> > > > > transited
> > > > > > > to
> > > > > > > > >> DEAD this
> > > > > > > > >> > > > > > group
> > > > > > > > >> > > > > > > > > would
> > > > > > > > >> > > > > > > > > > > >> never
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7.
> > > Will
> > > > > the
> > > > > > > > >> choice to
> > > > > > > > >> > > > store
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > configs
> > > > > > > > >> > > > > > > > > > > >> in
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > harder to
> > > > > > > list all
> > > > > > > > >> > > > cluster
> > > > > > > > >> > > > > > configs
> > > > > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > > > > > > > >> their
> > > > > > > > >> > > > > > > > > > > >> > > > > values?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > That's a
> > > > > good
> > > > > > > > >> question,
> > > > > > > > >> > > > and our
> > > > > > > > >> > > > > > > > > thoughts
> > > > > > > > >> > > > > > > > > > > >> are
> > > > > > > > >> > > > > > > > > > > >> > that
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > configurations"
> > > > > > > are
> > > > > > > > >> > > > overrides
> > > > > > > > >> > > > > > of the
> > > > > > > > >> > > > > > > > > > > >> > cluster-level
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > customized
> > > > > per
> > > > > > > > >> group so
> > > > > > > > >> > > > when an
> > > > > > > > >> > > > > > > > > admin list
> > > > > > > > >> > > > > > > > > > > >> > cluster
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list
> > > just
> > > > > the
> > > > > > > > >> cluster-level
> > > > > > > > >> > > > > > > > > "defaults", not
> > > > > > > > >> > > > > > > > > > > >> > showing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > customizations.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8.
> > How
> > > > > would
> > > > > > > > >> someone
> > > > > > > > >> > > > > > configure a
> > > > > > > > >> > > > > > > > > group
> > > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> consumers?
> > > Have
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > > > > > considered
> > > > > > > > >> allowing
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > explicit
> > > > > > > > >> > > > > > > > > > > >> > creation of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > Alternatively,
> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > configs
> > > > > > could
> > > > > > > be
> > > > > > > > >> > > > decoupled
> > > > > > > > >> > > > > > from
> > > > > > > > >> > > > > > > > > the group
> > > > > > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The
> > > configs
> > > > > > can
> > > > > > > be
> > > > > > > > >> created
> > > > > > > > >> > > > > > before
> > > > > > > > >> > > > > > > > > the group
> > > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > > >> > > > > > > > > > > >> > > > > as
> > > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > ---
> of
> > > > > course,
> > > > > > > this
> > > > > > > > >> > > > requires
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > > >> corresponding
> > > > > > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> right
> > > > > > > coordinator
> > > > > > > > >> based on
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > group
> > > > > > > > >> > > > > > > > > id ---
> > > > > > > > >> > > > > > > > > > > >> > the only
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs
> > is,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when
> > the
> > > > > group
> > > > > > > > >> itself is
> > > > > > > > >> > > > gone
> > > > > > > > >> > > > > > we
> > > > > > > > >> > > > > > > > > also check
> > > > > > > > >> > > > > > > > > > > >> > if there
> > > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > configuration
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > entities
> > > > > > > related to
> > > > > > > > >> that
> > > > > > > > >> > > > group
> > > > > > > > >> > > > > > and
> > > > > > > > >> > > > > > > > > delete
> > > > > > > > >> > > > > > > > > > > >> as
> > > > > > > > >> > > > > > > > > > > >> > well.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > Admittedly
> > > > > > this
> > > > > > > > >> indeed
> > > > > > > > >> > > > > > introduces an
> > > > > > > > >> > > > > > > > > > > >> > asymmetry on
> > > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > lifecycles
> > > > > of
> > > > > > > the
> > > > > > > > >> config
> > > > > > > > >> > > > > > entities,
> > > > > > > > >> > > > > > > > > and we
> > > > > > > > >> > > > > > > > > > > >> > would like
> > > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > feelings
> > > > > > > whether we
> > > > > > > > >> should
> > > > > > > > >> > > > aim
> > > > > > > > >> > > > > > for
> > > > > > > > >> > > > > > > > > symmetry
> > > > > > > > >> > > > > > > > > > > >> > i.e.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > configs
> > > and
> > > > > > > hence
> > > > > > > > >> not
> > > > > > > > >> > > > delete
> > > > > > > > >> > > > > > them at
> > > > > > > > >> > > > > > > > > all
> > > > > > > > >> > > > > > > > > > > >> when
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > require
> > > > > > explicit
> > > > > > > > >> deletion
> > > > > > > > >> > > > > > operations
> > > > > > > > >> > > > > > > > > by
> > > > > > > > >> > > > > > > > > > > >> > themselves.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9.
> > > Will
> > > > > the
> > > > > > > > >> > > > > > Consumer.subscribe
> > > > > > > > >> > > > > > > > > method for
> > > > > > > > >> > > > > > > > > > > >> > the Java
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > `java.util.regex.Pattern` of
> > > > > > > > >> > > > > > do we
> > > > > > > > >> > > > > > > > > have
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I
> > think
> > > we
> > > > > do
> > > > > > > not
> > > > > > > > >> need to
> > > > > > > > >> > > > > > introduce
> > > > > > > > >> > > > > > > > > an
> > > > > > > > >> > > > > > > > > > > >> > overload, but
> > > > > > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may
> be
> > > some
> > > > > > > > >> compatibility
> > > > > > > > >> > > > > > issues
> > > > > > > > >> > > > > > > > > that we
> > > > > > > > >> > > > > > > > > > > >> may
> > > > > > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 10.
> > I
> > > > > agree
> > > > > > > with
> > > > > > > > >> Justine
> > > > > > > > >> > > > > > that we
> > > > > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > > > > > > > >> be
> > > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > switch to
> > > > > > > > >> > > > > > IBP/metadata.version
> > > > > > > > >> > > > > > > > > from the
> > > > > > > > >> > > > > > > > > > > >> > feature
> > > > > > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> can
> > > switch
> > > > > > the
> > > > > > > > >> default
> > > > > > > > >> > > > for
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > feature
> > > > > > > > >> > > > > > > > > > > >> flag
> > > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > > >> > > > > > > > > > > >> > > > > true
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > metadata.version
> > > > > > > > >> once we
> > > > > > > > >> > > > > > want to
> > > > > > > > >> > > > > > > > > make it
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11.
> > > Some of
> > > > > > the
> > > > > > > > >> protocol
> > > > > > > > >> > > > APIs
> > > > > > > > >> > > > > > don't
> > > > > > > > >> > > > > > > > > mention
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> required
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> good
> > > to
> > > > > add
> > > > > > > that
> > > > > > > > >> for
> > > > > > > > >> > > > > > consistency.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack,
> > we
> > > can
> > > > > > > > >> certainly do
> > > > > > > > >> > > > that.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 12.
> > > It is
> > > > > a
> > > > > > > bit
> > > > > > > > >> odd that
> > > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group"
> > even
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > though it
> > > > > > > seems
> > > > > > > > >> to do
> > > > > > > > >> > > > more
> > > > > > > > >> > > > > > than
> > > > > > > > >> > > > > > > > > reading.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I
> had
> > > that
> > > > > > > thought
> > > > > > > > >> myself
> > > > > > > > >> > > > as
> > > > > > > > >> > > > > > well,
> > > > > > > > >> > > > > > > > > but in
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > end we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> better
> > > > > > > alternative:
> > > > > > > > >> adding
> > > > > > > > >> > > > > > Write
> > > > > > > > >> > > > > > > > > Group
> > > > > > > > >> > > > > > > > > > > >> seems
> > > > > > > > >> > > > > > > > > > > >> > an
> > > > > > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not
> > > have it
> > > > > > > > >> elsewhere (we
> > > > > > > > >> > > > only
> > > > > > > > >> > > > > > have
> > > > > > > > >> > > > > > > > > Read /
> > > > > > > > >> > > > > > > > > > > >> > Delete
> > > > > > > > >> > > > > > > > > > > >> > > > > and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups
> so
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> far).
> > > Would
> > > > > > > like to
> > > > > > > > >> hear
> > > > > > > > >> > > > others
> > > > > > > > >> > > > > > > > > thoughts.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 13.
> > > How is
> > > > > > > topic
> > > > > > > > >> > > > recreation
> > > > > > > > >> > > > > > > > > handled by
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > consumer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > protocol?
> > > > > It
> > > > > > > > >> would be
> > > > > > > > >> > > > good
> > > > > > > > >> > > > > > to have
> > > > > > > > >> > > > > > > > > a
> > > > > > > > >> > > > > > > > > > > >> > section on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You
> > mean
> > > > > with
> > > > > > > regex
> > > > > > > > >> > > > > > subscription
> > > > > > > > >> > > > > > > > > right? Yes
> > > > > > > > >> > > > > > > > > > > >> > we can
> > > > > > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> that,
> > > but
> > > > > > > basically
> > > > > > > > >> the
> > > > > > > > >> > > > idea
> > > > > > > > >> > > > > > is that
> > > > > > > > >> > > > > > > > > > > >> consumer
> > > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > > >> > > > > > > > > > > >> > > > > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic
> > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the
> > new
> > > > > > > protocol as
> > > > > > > > >> it's
> > > > > > > > >> > > > > > handled all
> > > > > > > > >> > > > > > > > > by the
> > > > > > > > >> > > > > > > > > > > >> > brokers.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 14.
> > > The
> > > > > KIP
> > > > > > > > >> mentions we
> > > > > > > > >> > > > will
> > > > > > > > >> > > > > > write
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> this
> > > is an
> > > > > > > > >> implementation
> > > > > > > > >> > > > > > detail,
> > > > > > > > >> > > > > > > > > do we
> > > > > > > > >> > > > > > > > > > > >> > plan to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> for
> > > it?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We
> > have
> > > not
> > > > > > > thought
> > > > > > > > >> about
> > > > > > > > >> > > > > > that. But
> > > > > > > > >> > > > > > > > > I think
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > answer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 15.
> > > Do we
> > > > > > > have a
> > > > > > > > >> > > > scalability
> > > > > > > > >> > > > > > goal
> > > > > > > > >> > > > > > > > > when it
> > > > > > > > >> > > > > > > > > > > >> > comes to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> how
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > group
> > > > > > > protocol can
> > > > > > > > >> > > > support?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > Within a
> > > > > > group,
> > > > > > > I
> > > > > > > > >> think we
> > > > > > > > >> > > > > > should
> > > > > > > > >> > > > > > > > > shoot for
> > > > > > > > >> > > > > > > > > > > >> > 1000s of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > scalability
> > > > > > > goals
> > > > > > > > >> also
> > > > > > > > >> > > > depend
> > > > > > > > >> > > > > > on the
> > > > > > > > >> > > > > > > > > offset
> > > > > > > > >> > > > > > > > > > > >> > > > > management
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > capabilities
> > > > > > of
> > > > > > > the
> > > > > > > > >> > > > coordinator
> > > > > > > > >> > > > > > > > > which we
> > > > > > > > >> > > > > > > > > > > >> did
> > > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard
> > to
> > > > > give a
> > > > > > > > >> number that
> > > > > > > > >> > > > > > applies
> > > > > > > > >> > > > > > > > > > > >> > universally.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> 16.
> > > Did we
> > > > > > > > >> consider
> > > > > > > > >> > > > having
> > > > > > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> SubscribedTopicNames in
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > ConsumerGroupHeartbeatRequest?
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > since
> > > we
> > > > > > have
> > > > > > > to
> > > > > > > > >> resolve
> > > > > > > > >> > > > the
> > > > > > > > >> > > > > > regex
> > > > > > > > >> > > > > > > > > on the
> > > > > > > > >> > > > > > > > > > > >> > server,
> > > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> the
> > > topic
> > > > > > > name?
> > > > > > > > >> The
> > > > > > > > >> > > > > > difference is
> > > > > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > sending the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> efficient
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > whereas
> > > > > > > sending
> > > > > > > > >> the topic
> > > > > > > > >> > > > > > names is
> > > > > > > > >> > > > > > > > > less
> > > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete
> and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > recreation
> > > > > > is
> > > > > > > > >> easier to
> > > > > > > > >> > > > > > handle if
> > > > > > > > >> > > > > > > > > we have
> > > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > > >> > > > > > > > > > > >> > > > > ids.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The
> > main
> > > > > > reason
> > > > > > > to
> > > > > > > > >> still
> > > > > > > > >> > > > let
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > clients
> > > > > > > > >> > > > > > > > > > > >> send
> > > > > > > > >> > > > > > > > > > > >> > names
> > > > > > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > reasoning of
> > > > > > > names
> > > > > > > > >> -> ids
> > > > > > > > >> > > > on
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > broker /
> > > > > > > > >> > > > > > > > > > > >> > admin
> > > > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > although we
> > > > > > > added
> > > > > > > > >> topic id
> > > > > > > > >> > > > in
> > > > > > > > >> > > > > > > > > KIP-516, we
> > > > > > > > >> > > > > > > > > > > >> > never
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > consumer/producers
> > > > > > > > >> > > > leveraging
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > related
> > > > > > > > >> > > > > > > > > > > >> > newer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead
> we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just
> > > set the
> > > > > > > topic
> > > > > > > > >> id as
> > > > > > > > >> > > > empty
> > > > > > > > >> > > > > > UUID.
> > > > > > > > >> > > > > > > > > We
> > > > > > > > >> > > > > > > > > > > >> want
> > > > > > > > >> > > > > > > > > > > >> > to keep
> > > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > consumer/producer
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to
> be
> > > thin
> > > > > and
> > > > > > > only
> > > > > > > > >> > > > delegate
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > reasoning
> > > > > > > > >> > > > > > > > > > > >> on
> > > > > > > > >> > > > > > > > > > > >> > broker
> > > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > clients.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > Thanks,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > Ismael
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On
> > > Wed,
> > > > > Jul
> > > > > > 6,
> > > > > > > > >> 2022 at
> > > > > > > > >> > > > 10:45
> > > > > > > > >> > > > > > AM
> > > > > > > > >> > > > > > > > > David
> > > > > > > > >> > > > > > > > > > > >> Jacot
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> Hi
> > > all,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> I
> > > would
> > > > > > > like to
> > > > > > > > >> start a
> > > > > > > > >> > > > > > > > > discussion
> > > > > > > > >> > > > > > > > > > > >> thread
> > > > > > > > >> > > > > > > > > > > >> > on
> > > > > > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > Generation
> > > > > > > of
> > > > > > > > >> the
> > > > > > > > >> > > > Consumer
> > > > > > > > >> > > > > > > > > Rebalance
> > > > > > > > >> > > > > > > > > > > >> > Protocol.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> With
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> to
> > > make
> > > > > > the
> > > > > > > > >> rebalance
> > > > > > > > >> > > > > > protocol
> > > > > > > > >> > > > > > > > > (for
> > > > > > > > >> > > > > > > > > > > >> > consumers)
> > > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > scalable,
> > > > > > > > >> easier to
> > > > > > > > >> > > > > > implement for
> > > > > > > > >> > > > > > > > > > > >> > clients, and
> > > > > > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > operators.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > The
> > > KIP
> > > > > is
> > > > > > > here:
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > Please
> > > > > > take
> > > > > > > a
> > > > > > > > >> look and
> > > > > > > > >> > > > let
> > > > > > > > >> > > > > > me
> > > > > > > > >> > > > > > > > > know what
> > > > > > > > >> > > > > > > > > > > >> > you
> > > > > > > > >> > > > > > > > > > > >> > > > > think.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > Best,
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > David
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > PS:
> > > I
> > > > > will
> > > > > > > be
> > > > > > > > >> away from
> > > > > > > > >> > > > > > July
> > > > > > > > >> > > > > > > > > 18th to
> > > > > > > > >> > > > > > > > > > > >> > August 8th.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> That
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> of
> > > time
> > > > > to
> > > > > > > read
> > > > > > > > >> and
> > > > > > > > >> > > > digest
> > > > > > > > >> > > > > > this
> > > > > > > > >> > > > > > > > > long
> > > > > > > > >> > > > > > > > > > > >> KIP.
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > Guozhang
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> Guozhang
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> --
> > > > > > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > --
> > > > > > > > >> > > > > -- Guozhang
> > > > > > > > >> > > >
> > > > > > > > >> > >
> > > > > > > > >> > >
> > > > > > > > >> > > --
> > > > > > > > >> > > -- Guozhang
> > > > > > > > >>
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > --
> > > > > > > > > -- Guozhang
> > > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > --
> > > > > > > > -- Guozhang
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > -- Guozhang
> > > > > >
> > > > >
> > >
> >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Jun Rao <ju...@confluent.io.INVALID>]

Hi, David,

Thanks for the reply. A few more follow up questions.

48. Hmm, so member assignment is what the client sees. Then, I am not sure
about the following. The text says B cannot get foo-2 yet. But in Member
Assignment, B has foo-2.

At this point B can transition to epoch 2 but cannot get foo-2 until A
revokes it.

Note that the persisted target assignment of B already includes foo-2 here
but B does not get it yet because foo-2 is not free.

Group (epoch=2)
A
B
Target Assignment (epoch=2)
A - partitions=[foo-0, foo-1]
B - partitions=[foo-2]
Member Assignment
A - epoch=1, partitions=[foo-0, foo-1, foo-2]
B - epoch=2, partitions=[foo-2]


50. Hmm, I thought KStreams doesn't use range assignor all the time. It
does the assignment based on the topology. For example, if you are joining
two topics A and B and deriving the join result in a 3rd topic C, which
will be aggregated. You want to use range to co-locate the partitions from
topic A and B. However, for topic C, the partition could potentially be
assigned in round robin. Without the topic name, KStreams won't know which
topics need to be co-located.

53. If MemberAssignment.metadata is not used by the group coordinator,
where is it being used?

Thanks,

Jun

On Tue, Sep 27, 2022 at 1:20 PM Jason Gustafson <ja...@confluent.io.invalid>
wrote:

> Hi David,
>
> Thanks for the response. Replied below:
>
> > I don't have a strong opinion on this one. However, it seems that this
> could be a pandora box if our subset turns out to not be sufficient
> for some existing users. We will have to add more to it and the
> compatibility story between versions could be a real nightmare. One
> thing that we could do for sure is disabling all unicode groups in the
> library. That seems possible and would already reduce the scope.
>
> Changes to the regex syntax seems not a major problem to me. We would bump
> the protocol version as we usually do. Actually that begs the same question
> about RE2. How stable is the syntax and does it differ depending on the
> version of the library? In general, I do agree that we'd be better off not
> having to manage the syntax ourselves if we can convince ourselves that
> compatibility won't be a problem and the client won't eventually need to
> have access to it.
>
> Thanks,
> Jason
>
> On Tue, Sep 27, 2022 at 6:21 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Luke,
> >
> > Thanks for your comments.
> >
> > Your suggestion makes sense. Jun's made the same one. We are also
> > discussing whether we should directly return the group state in the
> > response and drop the prepare assignment api. It seems that the
> > current consensus is to keep it though. If that stays like this, I
> > will apply your suggestion.
> >
> > Thanks,
> > David
> >
> > On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com> wrote:
> > >
> > > Hi David,
> > >
> > > Thanks for your patience for answering all these questions!
> > >
> > > One more question from me:
> > > 1. When the selected consumer is going to be in client-side assigning,
> we
> > > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering if
> we
> > > could use a flag to indicate the assignment is required. The rationale
> is
> > > when returning error, we lose a chance to catch up with the latest
> target
> > > assignment or revoke partitions. Since we expected the record fetching
> is
> > > not blocked by the rebalance/assignment computing, we might want to
> allow
> > > the members to update the member assignment in each heartbeat.
> > >
> > > Luke
> > >
> > > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> > <ja...@confluent.io.invalid>
> > > wrote:
> > >
> > > > Hey David,
> > > >
> > > > Thanks for the response.
> > > >
> > > > > Do you think that we could have a case where two members with the
> > same
> > > > static id are active?
> > > >
> > > > I was just thinking about the handoff when we spin up a new node
> > before the
> > > > old one completes shutdown. It's probably ok to fence the old member
> > > > abruptly. I do recall we had a few issues with this handoff in the
> > past. It
> > > > was due to the fact that the replaced node might have been the
> leader.
> > > > Since we've removed the leader notion, perhaps this kind of issue
> won't
> > > > resurface.
> > > >
> > > > In regard to the RE2 library. I understand the intent to pass this
> > through
> > > > as a string without any validation so that we don't need the
> > dependence on
> > > > the client. I do wonder if that will be restrictive in practice. For
> > > > example, if a regex changes on the client side, would we need to
> > validate
> > > > against the current assignment to see which partitions to continue
> > > > fetching? Or would we instead invalidate the complete current
> > assignment?
> > > > Perhaps this is a rare case and we can do the latter. There could be
> > more
> > > > situations though.
> > > >
> > > > As an alternative, I wonder if we could define a minimalistic regex
> > syntax
> > > > (a subset of RE2 say) that could map more easily to regex support
> that
> > > > already exists in clients? For example, in this scenario, we are
> > matching
> > > > topics, so we only need to be able to represent topic characters and
> > not
> > > > the whole smorgasbord of whitespace and unicode options. Would it
> make
> > > > sense to have a reduced syntax which includes only standard topic
> > > > characters, wildcards, character classes, and repetitions?
> > > >
> > > > Best,
> > > > Jason
> > > >
> > > >
> > > >
> > > >
> > > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > >
> > > > > Hi David, thanks for all the detailed explanations. I think they
> all
> > make
> > > > > sense. Just want to have a couple follow-ups here:
> > > > >
> > > > > > I don't really see the benefits here because server side
> assignors
> > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > are not supposed to generate metadata nor to receive metadata from
> > the
> > > > > members.
> > > > >
> > > > > Just to clarify I was asking about the `version` of the assignor
> > (i.e. up
> > > > > to what version that the client would support), and I do agree we
> > would
> > > > not
> > > > > need metadata. What I have in mind is that, for some specific
> > built-in
> > > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> > in a
> > > > > newer version we would have a hierarchical rack ID string format,
> > like
> > > > > "tier1-tier2" etc, but if some client has not upgraded their rack
> ID
> > > > > would still be in old format. In this case, the broker then needs
> to
> > > > choose
> > > > > the old versioned assignor. I'm probably making something up here
> for
> > > > rack
> > > > > aware assignors, but I'm wondering if in general such an
> > "auto-downgrade"
> > > > > behavior would be needed still for broker-side assignor, and if yes
> > would
> > > > > "version" still be useful.
> > > > >
> > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is called
> > once
> > > > > when the member transitions to a new epoch. This one contains the
> > full
> > > > > metadata provided by the client side assignor. Then, `onAssignment`
> > > > > can be called max N times where N is the number of partitions
> pending
> > > > > revocation by other members. Let me try to clarify this in the KIP.
> > > > >
> > > > > Okay, my understanding is that the calling ordering of these
> > callbacks
> > > > > would be like the following:
> > > > >
> > > > > ----------------------------------------
> > > > > onPartitionsRevoked();   // just once, since we do not really need
> > > > > to revoke incrementally.
> > > > >
> > > > > onAssignment();    // the first call, with epoch incremented
> > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > >
> > > > > onAssignment();              // the first onAssignment would bump
> up
> > the
> > > > > epoch, and the metadata reflected.
> > > > > onPartitionsAssigned();   // each time we get an additional
> > assignment,
> > > > we
> > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > ...
> > > > > onAssignment();
> > > > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > > > encoded
> > > > > metadata would not change, only the incrementally added partitions
> be
> > > > > reflected
> > > > >
> > > > > Is that the case?
> > > > >
> > > > > I'm wondering if we would still call onAssignment just once, that
> > encodes
> > > > > all the assignment for this rebalance, including all the partitions
> > that
> > > > > should be assigned to the member but not yet assigned since they
> > have not
> > > > > been revoked by others. In that case the call ordering would be:
> > > > >
> > > > > ----------------------------------------
> > > > > onPartitionsRevoked();   // just once
> > > > > onAssignment();    // just once, with epoch incremented, and
> metadata
> > > > > encoded changed, the "assignment" field also reflect the final
> target
> > > > > assignment
> > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > incrementally
> > > > > added partitions
> > > > > ...
> > > > > onPartitionsAssigned();
> > > > >
> > > > > The motivation from this is that, most users would only implement
> the
> > > > > rebalance callback listeners and hence we'd definitely need to make
> > sure
> > > > > the semantics of that does not change much, and the time
> > > > > `onPartitionsAssigned` indicate the time when the partitions are
> > actually
> > > > > assigned to it; while for assignors, the `onAssignment` is used to
> > > > indicate
> > > > > what decision is made regarding for this member, i.e. when the
> > partitions
> > > > > are decided to be given to it, but not necessarily meant that it
> has
> > been
> > > > > given, since that time should be determined by the time of
> > > > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > > > implementers would not need to reason which `onAssignment` would be
> > the
> > > > > last one for this epoch.
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > <djacot@confluent.io.invalid
> > > > >
> > > > > wrote:
> > > > >
> > > > > > Hi Guozhang,
> > > > > >
> > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > byte
> > > > > > field,
> > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > And
> > > > the
> > > > > > > rationale seems to be assuming that we should always be able to
> > do
> > > > the
> > > > > > > assignment at the broker-side assignor without errors.
> > Personally I
> > > > > think
> > > > > > > it's still potentially beneficial to add the Error field even
> for
> > > > > > > broker-side assignors, e.g. for some edge cases where some
> > subscribed
> > > > > > > topics are not recognized with the current broker's metadata.
> > What do
> > > > > you
> > > > > > > think?
> > > > > >
> > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > better
> > > > > > to use an exception on the server side. This is what we usually
> do
> > for
> > > > > > server side plugins. On the client side, we use a field because
> the
> > > > > > errors are not defined in advance.
> > > > > >
> > > > > > Your comment also makes me think about what we should do when the
> > > > > > server side assignor fails. I suppose that we have to keep the
> > current
> > > > > > assignment until a new event occurs. For instance, in your
> example,
> > > > > > the coordinator would have to trigger a rebalance when
> unrecognized
> > > > > > topics are available. This would be part of the metadata
> > monitoring.
> > > > > >
> > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > reason/version/metadata compared with the broker-side
> > GroupMember. I
> > > > > > agree
> > > > > > > that broker-side assignor would not need reason/metadata since
> > they
> > > > are
> > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > E.g.
> > > > is
> > > > > > it
> > > > > > > possible that we evolve our broker-side built-in assignor but
> > the old
> > > > > > > versioned clients would not be able to work with the new
> > version, in
> > > > > > which
> > > > > > > case we need to let the broker being aware of this and upgrade
> > its
> > > > > > behavior
> > > > > > > to cooperate with the clients?
> > > > > >
> > > > > > I don't really see the benefits here because server side
> assignors
> > > > > > don't have metadata at all. They only assign topic-partitions.
> They
> > > > > > are not supposed to generate metadata nor to receive metadata
> from
> > the
> > > > > > members.
> > > > > >
> > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > instead
> > > > > > of
> > > > > > > including these three fields, what about just adding the
> > "Metadata"
> > > > > field
> > > > > > > class which has these three fields? Also, there are two
> > "Metadata"
> > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > reason/version/metadata, and the second is just the encoded
> > metadata
> > > > > > bytes.
> > > > > > > I'm wondering what about just naming the first as
> memberMetadata,
> > > > which
> > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > second
> > > > > > > bytebuffer field as metadataBytes?
> > > > > >
> > > > > > That's a good point. Let me try to rationalize this interface
> > based on
> > > > > > your suggestions.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> wangguoz@gmail.com>
> > > > > wrote:
> > > > > > >
> > > > > > > Hello David,
> > > > > > >
> > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > >
> > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > byte
> > > > > > field,
> > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > And
> > > > the
> > > > > > > rationale seems to be assuming that we should always be able to
> > do
> > > > the
> > > > > > > assignment at the broker-side assignor without errors.
> > Personally I
> > > > > think
> > > > > > > it's still potentially beneficial to add the Error field even
> for
> > > > > > > broker-side assignors, e.g. for some edge cases where some
> > subscribed
> > > > > > > topics are not recognized with the current broker's metadata.
> > What do
> > > > > you
> > > > > > > think?
> > > > > > >
> > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > reason/version/metadata compared with the broker-side
> > GroupMember. I
> > > > > > agree
> > > > > > > that broker-side assignor would not need reason/metadata since
> > they
> > > > are
> > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > E.g.
> > > > is
> > > > > > it
> > > > > > > possible that we evolve our broker-side built-in assignor but
> > the old
> > > > > > > versioned clients would not be able to work with the new
> > version, in
> > > > > > which
> > > > > > > case we need to let the broker being aware of this and upgrade
> > its
> > > > > > behavior
> > > > > > > to cooperate with the clients?
> > > > > > >
> > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > instead
> > > > > > of
> > > > > > > including these three fields, what about just adding the
> > "Metadata"
> > > > > field
> > > > > > > class which has these three fields? Also, there are two
> > "Metadata"
> > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > reason/version/metadata, and the second is just the encoded
> > metadata
> > > > > > bytes.
> > > > > > > I'm wondering what about just naming the first as
> memberMetadata,
> > > > which
> > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > second
> > > > > > > bytebuffer field as metadataBytes?
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > > wrote:
> > > > > > >
> > > > > > > > Hello David,
> > > > > > > >
> > > > > > > > Thanks for bringing this question up. I think the main
> > benefits as
> > > > > you
> > > > > > > > listed is 2) above if it stays; just to clarify, we would
> only
> > be
> > > > > able
> > > > > > to
> > > > > > > > save one round trip if the rebalance is still triggered by
> the
> > > > > broker;
> > > > > > if
> > > > > > > > the rebalance is triggered by the client then the
> > num.round.trips
> > > > are
> > > > > > the
> > > > > > > > same:
> > > > > > > >
> > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > >
> > > > > > > > T0: client decides to do a new assignment, suppose it has
> > already
> > > > > sent
> > > > > > a
> > > > > > > > HB and hence has to wait for it to return first since only
> one
> > > > > request
> > > > > > /
> > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > GroupPrepareAssignment request.
> > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > GroupInstallAssignment
> > > > > > > > request.
> > > > > > > >
> > > > > > > > In total, two round trips.
> > > > > > > >
> > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > >
> > > > > > > > T0: client decides to do a new assignment, suppose it has
> > already
> > > > > sent
> > > > > > a
> > > > > > > > HB and hence has to wait for it to return first since only
> one
> > > > > request
> > > > > > /
> > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > T1: client receives the HB response, and then sends the new
> HB
> > > > > request
> > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > T2: the HB response with the optional member metadata map is
> > > > > returned.
> > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > GroupInstallAssignment
> > > > > > > > request.
> > > > > > > >
> > > > > > > > In total, two round trips as well.
> > > > > > > >
> > > > > > > > -----------------------------
> > > > > > > >
> > > > > > > > So to complete the full picture here, we'd need to modify
> both
> > HB
> > > > > > request
> > > > > > > > and response so that the client can also indicate a new
> > rebalance
> > > > via
> > > > > > the
> > > > > > > > HB request as well, right?
> > > > > > > >
> > > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > additional
> > > > > > fields
> > > > > > > > encoding the full member (subscription) metadata and topic
> > metadata
> > > > > as
> > > > > > > > optional fields.
> > > > > > > >
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > >> Hi all,
> > > > > > > >>
> > > > > > > >> During an offline conversation, someone asked why we need
> the
> > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we
> could
> > > > > instead
> > > > > > > >> provide the group state in the heartbeat response. This has
> a
> > few
> > > > > > > >> advantages: 1) it does not require using a special error
> code
> > to
> > > > > > > >> signal that a new assignment is required as the signal would
> > be
> > > > the
> > > > > > > >> provided group state; 2) it removes one round trip when a
> > client
> > > > > side
> > > > > > > >> assignor is used. The downside is that it makes the
> heartbeat
> > > > > > > >> response's definition quite large. I recall that I went with
> > the
> > > > > > > >> current approach due to this.
> > > > > > > >>
> > > > > > > >> Providing the group state in the heartbeat response is
> > appealing.
> > > > > What
> > > > > > > >> do you guys think?
> > > > > > > >>
> > > > > > > >> Best,
> > > > > > > >> David
> > > > > > > >>
> > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > djacot@confluent.io>
> > > > > > wrote:
> > > > > > > >> >
> > > > > > > >> > Hi Guozhang,
> > > > > > > >> >
> > > > > > > >> > 1. I have added a reference to the relevant chapter
> instead
> > of
> > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > >> >
> > > > > > > >> > 2. The "Rebalance Triggers" section you are referring to
> is
> > > > about
> > > > > > when
> > > > > > > >> > a rebalance should be triggered for the non-upgraded
> members
> > > > using
> > > > > > the
> > > > > > > >> > old protocol. The section mentions that a rebalance must
> be
> > > > > > triggered
> > > > > > > >> > when a new assignment is installed. This implies that the
> > group
> > > > > > epoch
> > > > > > > >> > was updated either by a native member or a non-upgraded
> > member.
> > > > > For
> > > > > > > >> > the latter, the JoinGroup request would be the trigger. I
> > have
> > > > > > added a
> > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > Handling"
> > > > > > section
> > > > > > > >> > as well. Does that make sense?
> > > > > > > >> >
> > > > > > > >> > Thanks,
> > > > > > > >> > David
> > > > > > > >> >
> > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > wangguoz@gmail.com
> > > > > >
> > > > > > > >> wrote:
> > > > > > > >> > >
> > > > > > > >> > > Hello David,
> > > > > > > >> > >
> > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > clear in
> > > > > the
> > > > > > doc,
> > > > > > > >> > > could we update:
> > > > > > > >> > >
> > > > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > > > coordinator
> > > > > > > >> will
> > > > > > > >> > > trigger rebalance based on the HB's member metadata /
> > reason?
> > > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > > described
> > > > > > in
> > > > > > > >> "Group
> > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > >> > >
> > > > > > > >> > >
> > > > > > > >> > > Guozhang
> > > > > > > >> > >
> > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > wrote:
> > > > > > > >> > >
> > > > > > > >> > > > Hi Guozhang,
> > > > > > > >> > > >
> > > > > > > >> > > > I thought that the assignor will always be consulted
> > when
> > > > the
> > > > > > next
> > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > > > heartbeat.
> > > > > > > >> This
> > > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > > rebalance
> > > > > by
> > > > > > > >> > > > setting the reason to a non-zero value or by changing
> > the
> > > > > > bytes. Do
> > > > > > > >> > > > you think that this is not sufficient? Are you
> > concerned by
> > > > > the
> > > > > > > >> delay?
> > > > > > > >> > > >
> > > > > > > >> > > > Best,
> > > > > > > >> > > > David
> > > > > > > >> > > >
> > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > wangguoz@gmail.com>
> > > > > > > >> wrote:
> > > > > > > >> > > > >
> > > > > > > >> > > > > Hello David,
> > > > > > > >> > > > >
> > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > >> > > > >
> > > > > > > >> > > > > ```
> > > > > > > >> > > > > In this case, a new assignment is triggered by the
> > client
> > > > > side
> > > > > > > >> > > > > assignor. When constructing the HB, the consumer
> will
> > > > always
> > > > > > > >> consult
> > > > > > > >> > > > > the client side assignor and propagate the
> > information to
> > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > coordinator. In other words, we don't expect users
> to
> > call
> > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > >> > > > > ```
> > > > > > > >> > > > >
> > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > interface,
> > > > we
> > > > > > > >> actually do
> > > > > > > >> > > > > not have a way yet to instruct how to construct the
> > next
> > > > HB
> > > > > > > >> request, e.g.
> > > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> > with a
> > > > > new
> > > > > > > >> assignment,
> > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > PartitionAssignor
> > > > > > to
> > > > > > > >> indicate
> > > > > > > >> > > > > the next HB telling broker about so. WDYT about
> adding
> > > > such
> > > > > an
> > > > > > > >> API on the
> > > > > > > >> > > > > PartitionAssignor?
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > Guozhang
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > > > wrote:
> > > > > > > >> > > > >
> > > > > > > >> > > > > > Hi Jun,
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > I have updated the KIP to include your feedback. I
> > have
> > > > > also
> > > > > > > >> tried to
> > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > Best,
> > > > > > > >> > > > > > David
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > djacot@confluent.io
> > > > > > > >> >
> > > > > > > >> > > > wrote:
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Hi Jun,
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > answering
> > > > your
> > > > > > > >> questions
> > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > of
> > > > the
> > > > > > KIP
> > > > > > > >> seem to
> > > > > > > >> > > > be
> > > > > > > >> > > > > > fewer
> > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > support of
> > > > > > > >> wildcard. A few
> > > > > > > >> > > > > > > > comments below.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > I would also add that the KIP removes the global
> > sync
> > > > > > barrier
> > > > > > > >> in the
> > > > > > > >> > > > > > > protocol which is essential to improve group
> > stability
> > > > > and
> > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > client by
> > > > > > moving
> > > > > > > >> most of
> > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > to
> > > > > > support
> > > > > > > >> rolling
> > > > > > > >> > > > > > changing
> > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > > assignor
> > > > > > used
> > > > > > > >> by a
> > > > > > > >> > > > > > > majority of the members. This allows the group
> to
> > move
> > > > > > from
> > > > > > > >> one
> > > > > > > >> > > > > > > assignor to another by a roll. This is explained
> > in
> > > > the
> > > > > > > >> Assignor
> > > > > > > >> > > > > > > Selection chapter.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > it's
> > > > > > > >> required in
> > > > > > > >> > > > every
> > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > filled?
> > > > > For
> > > > > > > >> example,
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > not
> > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > filled.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > The client is expected to set those fields in
> > case of
> > > > a
> > > > > > > >> connection
> > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > changed
> > > > > since
> > > > > > > >> the last
> > > > > > > >> > > > > > > HB. The server populates those fields as long as
> > the
> > > > > > member
> > > > > > > >> is not
> > > > > > > >> > > > > > > fully reconciled - the member should acknowledge
> > that
> > > > it
> > > > > > has
> > > > > > > >> the
> > > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> > this in
> > > > > the
> > > > > > KIP.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > > affinity
> > > > > > > >> between the
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > and the broker is only considered during
> > fetching,
> > > > but
> > > > > > not
> > > > > > > >> during
> > > > > > > >> > > > > > assigning
> > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > assignment
> > > > > > is
> > > > > > > >> made,
> > > > > > > >> > > > there
> > > > > > > >> > > > > > is
> > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > replicas
> > > > > of
> > > > > > > >> assigned
> > > > > > > >> > > > > > partitions
> > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > should
> > > > > use
> > > > > > > >> this
> > > > > > > >> > > > > > opportunity
> > > > > > > >> > > > > > > > to address this by including rack in
> > GroupMember.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's an interesting idea. I don't see any
> issue
> > with
> > > > > > adding
> > > > > > > >> the
> > > > > > > >> > > > rack
> > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > know
> > > > how
> > > > > > busy
> > > > > > > >> a group
> > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> model,
> > it
> > > > > seems
> > > > > > > >> that we
> > > > > > > >> > > > could
> > > > > > > >> > > > > > add
> > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > the
> > > > > event
> > > > > > > >> loop is
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > actual work.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > failover
> > > > > > > >> handling? For
> > > > > > > >> > > > > > example,
> > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > with
> > > > > the
> > > > > > > >> wildcard
> > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sure. When the new group coordinator takes over,
> > it
> > > > has
> > > > > > to:
> > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > assignor
> > > > is
> > > > > > used.
> > > > > > > >> We
> > > > > > > >> > > > don't
> > > > > > > >> > > > > > > store the information about the member selected
> > to run
> > > > > the
> > > > > > > >> assignment
> > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > >> > > > > > > * Update the topics metadata, verify the
> wildcard
> > > > > > > >> subscriptions, and
> > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > document
> > > > > what
> > > > > > > >> the epoch
> > > > > > > >> > > > > > field
> > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > >> > > > > > reflect
> > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> > latest
> > > > > > group
> > > > > > > >> epoch.
> > > > > > > >> > > > > > > When the group state is updated, the group epoch
> > is
> > > > > > bumped so
> > > > > > > >> we use
> > > > > > > >> > > > > > > that one for all the change records related to
> the
> > > > > update.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> the
> > > > > > following
> > > > > > > >> > > > invariants
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> for
> > a
> > > > > member
> > > > > > > >> not to
> > > > > > > >> > > > get any
> > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's right. Here I meant that the members
> > provided
> > > > by
> > > > > > the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > > assignor
> > > > > > can
> > > > > > > >> not make
> > > > > > > >> > > > > > > up new member ids.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > epoch
> > > > > > equals to
> > > > > > > >> 0":
> > > > > > > >> > > > When
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > used?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > A member is expected to abandon all its
> > partitions and
> > > > > > > >> rejoins when
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this
> > case,
> > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > > > coordinator will have removed the member from
> the
> > > > group.
> > > > > > The
> > > > > > > >> member
> > > > > > > >> > > > > > > can rejoin the group with the same member id but
> > with
> > > > 0
> > > > > as
> > > > > > > >> epoch. Let
> > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> ability
> > to
> > > > > > trigger a
> > > > > > > >> > > > > > reassignment
> > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > updating
> > > > > the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > with
> > > > the
> > > > > > > >> deprecation
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > In this case, a new assignment is triggered by
> the
> > > > > client
> > > > > > side
> > > > > > > >> > > > > > > assignor. When constructing the HB, the consumer
> > will
> > > > > > always
> > > > > > > >> consult
> > > > > > > >> > > > > > > the client side assignor and propagate the
> > information
> > > > > to
> > > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > users to
> > > > > call
> > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> the
> > > > > section
> > > > > > > >> seems to
> > > > > > > >> > > > have
> > > > > > > >> > > > > > > > multiple typos.
> > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > immediately
> > > > > > > >> gets into
> > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > incorrect.
> > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > seems
> > > > to
> > > > > > get
> > > > > > > >> into
> > > > > > > >> > > > > > epoch=3,
> > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > has A -
> > > > > > > >> epoch=2,
> > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> support
> > the
> > > > > > upgrade
> > > > > > > >> from
> > > > > > > >> > > > any
> > > > > > > >> > > > > > old
> > > > > > > >> > > > > > > > version to new one?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > We will support upgrading from the consumer
> > protocol
> > > > > > version
> > > > > > > >> 3,
> > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> implemented
> > yet
> > > > so
> > > > > > the
> > > > > > > >> earliest
> > > > > > > >> > > > > > > version is unknown at the moment. This is
> > explained in
> > > > > the
> > > > > > > >> migration
> > > > > > > >> > > > > > > plan chapter.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> > update the
> > > > > KIP
> > > > > > > >> based on
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > next week.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Best,
> > > > > > > >> > > > > > > David
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > >> > > > wrote:
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Hi, David,
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > of
> > > > the
> > > > > > KIP
> > > > > > > >> seem to
> > > > > > > >> > > > be
> > > > > > > >> > > > > > fewer
> > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > support of
> > > > > > > >> wildcard. A few
> > > > > > > >> > > > > > > > comments below.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > to
> > > > > > support
> > > > > > > >> rolling
> > > > > > > >> > > > > > changing
> > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > it's
> > > > > > > >> required in
> > > > > > > >> > > > every
> > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > filled?
> > > > > For
> > > > > > > >> example,
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > not
> > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > filled.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > > affinity
> > > > > > > >> between the
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > and the broker is only considered during
> > fetching,
> > > > but
> > > > > > not
> > > > > > > >> during
> > > > > > > >> > > > > > assigning
> > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > assignment
> > > > > > is
> > > > > > > >> made,
> > > > > > > >> > > > there
> > > > > > > >> > > > > > is
> > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > replicas
> > > > > of
> > > > > > > >> assigned
> > > > > > > >> > > > > > partitions
> > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > should
> > > > > use
> > > > > > > >> this
> > > > > > > >> > > > > > opportunity
> > > > > > > >> > > > > > > > to address this by including rack in
> > GroupMember.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > know
> > > > how
> > > > > > busy
> > > > > > > >> a group
> > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> model,
> > it
> > > > > seems
> > > > > > > >> that we
> > > > > > > >> > > > could
> > > > > > > >> > > > > > add
> > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > the
> > > > > event
> > > > > > > >> loop is
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > actual work.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > failover
> > > > > > > >> handling? For
> > > > > > > >> > > > > > example,
> > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > with
> > > > > the
> > > > > > > >> wildcard
> > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > document
> > > > > what
> > > > > > > >> the epoch
> > > > > > > >> > > > > > field
> > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > >> > > > > > reflect
> > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> the
> > > > > > following
> > > > > > > >> > > > invariants
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> for
> > a
> > > > > member
> > > > > > > >> not to
> > > > > > > >> > > > get any
> > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > epoch
> > > > > > equals to
> > > > > > > >> 0":
> > > > > > > >> > > > When
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > used?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> ability
> > to
> > > > > > trigger a
> > > > > > > >> > > > > > reassignment
> > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > updating
> > > > > the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > with
> > > > the
> > > > > > > >> deprecation
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> the
> > > > > section
> > > > > > > >> seems to
> > > > > > > >> > > > have
> > > > > > > >> > > > > > > > multiple typos.
> > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > immediately
> > > > > > > >> gets into
> > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > incorrect.
> > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > seems
> > > > to
> > > > > > get
> > > > > > > >> into
> > > > > > > >> > > > > > epoch=3,
> > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > has A -
> > > > > > > >> epoch=2,
> > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> support
> > the
> > > > > > upgrade
> > > > > > > >> from
> > > > > > > >> > > > any
> > > > > > > >> > > > > > old
> > > > > > > >> > > > > > > > version to new one?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Thanks,
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Jun
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > wrote:
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > Hi all,
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > > > coordinator
> > > > > > > >> in
> > > > > > > >> > > > Java. I
> > > > > > > >> > > > > > > > > discussed this offline with a few folks and
> > folks
> > > > > are
> > > > > > > >> concerned
> > > > > > > >> > > > that
> > > > > > > >> > > > > > > > > we could introduce many regressions in the
> old
> > > > > > protocol
> > > > > > > >> if we do
> > > > > > > >> > > > so.
> > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> statement
> > > > from
> > > > > > the
> > > > > > > >> KIP. It
> > > > > > > >> > > > is an
> > > > > > > >> > > > > > > > > implementation detail after all so it does
> not
> > > > have
> > > > > > to be
> > > > > > > >> > > > decided at
> > > > > > > >> > > > > > > > > this stage. We will likely start by trying
> to
> > > > > > refactor the
> > > > > > > >> > > > current
> > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > Cheers,
> > > > > > > >> > > > > > > > > David
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot
> <
> > > > > > > >> djacot@confluent.io
> > > > > > > >> > > > >
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > "Empty,
> > > > > > assigning,
> > > > > > > >> > > > > > reconciling,
> > > > > > > >> > > > > > > > > stable,
> > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > > section,
> > > > > > > >> right?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > This sentence does not refer to those
> group
> > > > states
> > > > > > but
> > > > > > > >> rather
> > > > > > > >> > > > to a
> > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > refers to
> > > > > the
> > > > > > > >> entire
> > > > > > > >> > > > state of
> > > > > > > >> > > > > > > > > > group coordinator which is replicated via
> > the
> > > > log
> > > > > > > >> layer. I will
> > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> machine
> > is
> > > > > > modelled
> > > > > > > >> as an
> > > > > > > >> > > > event
> > > > > > > >> > > > > > > > > loop"?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> > the new
> > > > > > quorum
> > > > > > > >> > > > > > controller. We
> > > > > > > >> > > > > > > > > > will have N threads to process events.
> Each
> > > > > > > >> __consumer_offsets
> > > > > > > >> > > > > > > > > > partition is assigned to a unique thread
> > and all
> > > > > the
> > > > > > > >> events
> > > > > > > >> > > > (e.g.
> > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed
> by
> > this
> > > > > > > >> thread. This
> > > > > > > >> > > > > > simplify
> > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > simulation
> > > > > > testing
> > > > > > > >> for the
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > >> *__consumer_offsets*
> > > > > > > >> > > > > > > > > partitions?
> > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > owned
> > > > > by
> > > > > > a
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator?
> > > > > > > >> > > > > > > > > For
> > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > > consumer
> > > > > > > >> groups, and
> > > > > > > >> > > > > > both
> > > > > > > >> > > > > > > > > exist in
> > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > state
> > > > > > machine
> > > > > > > >> for it,
> > > > > > > >> > > > or
> > > > > > > >> > > > > > 2?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > > > think.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.4. I know the
> > "*group.coordinator.threads"
> > > > *is
> > > > > > the
> > > > > > > >> number
> > > > > > > >> > > > of
> > > > > > > >> > > > > > threads
> > > > > > > >> > > > > > > > > used
> > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > wondering
> > > > if
> > > > > > the
> > > > > > > >> purpose
> > > > > > > >> > > > of
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > threads
> > > > > > > >> > > > > > > > > > > is only to keep the state of each
> consumer
> > > > group
> > > > > > (or
> > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > why
> > > > > > should we
> > > > > > > >> need
> > > > > > > >> > > > > > > > > multi-threads
> > > > > > > >> > > > > > > > > > > here?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> > shard
> > > > > the
> > > > > > > >> > > > processing as
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > does
> > > > the
> > > > > > > >> default
> > > > > > > >> > > > session
> > > > > > > >> > > > > > > > > timeout not
> > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > thought
> > > > > > the
> > > > > > > >> min/max
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound
> of
> > it,
> > > > > no?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > 30s The
> > > > > > > >> timeout to
> > > > > > > >> > > > detect
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > protocol.
> > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > int 45s
> > > > > The
> > > > > > > >> minimum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > int 60s
> > > > > The
> > > > > > > >> maximum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > session
> > > > > > timeout
> > > > > > > >> should
> > > > > > > >> > > > be 45s
> > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> are
> > > > > [range,
> > > > > > > >> uniform],
> > > > > > > >> > > > > > which means
> > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > like to
> > > > > > know
> > > > > > > >> why not
> > > > > > > >> > > > > > uniform
> > > > > > > >> > > > > > > > > one? I
> > > > > > > >> > > > > > > > > > > thought usually users will choose
> uniform
> > > > > assignor
> > > > > > > >> (former
> > > > > > > >> > > > sticky
> > > > > > > >> > > > > > > > > assinor)
> > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> other
> > > > reason
> > > > > > we
> > > > > > > >> choose
> > > > > > > >> > > > range
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > as default?
> > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > uniform
> > > > The
> > > > > > > >> server side
> > > > > > > >> > > > > > assignors.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > The order on the server side has no
> > influence
> > > > > > because
> > > > > > > >> the
> > > > > > > >> > > > client
> > > > > > > >> > > > > > must
> > > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> > There
> > > > is
> > > > > no
> > > > > > > >> default
> > > > > > > >> > > > in the
> > > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > > specified
> > > > > > by
> > > > > > > >> the
> > > > > > > >> > > > client,
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > request is rejected. The default client
> > value
> > > > for
> > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> though.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Thanks for your very good comments, Luke.
> I
> > hope
> > > > > > that my
> > > > > > > >> > > > answers
> > > > > > > >> > > > > > help
> > > > > > > >> > > > > > > > > > to clarify things. I will update the KIP
> as
> > well
> > > > > > based
> > > > > > > >> on your
> > > > > > > >> > > > > > > > > > feedback.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > >> > > > > > > > > > David
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen
> <
> > > > > > > >> showuon@gmail.com>
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > mentioned:
> > > > > > > >> > > > > > > > > > > > The new group coordinator will have a
> > state
> > > > > > machine
> > > > > > > >> per
> > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> > each
> > > > > state
> > > > > > > >> machine is
> > > > > > > >> > > > > > modelled
> > > > > > > >> > > > > > > > > as an
> > > > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > > > executed
> > > > > > in
> > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > "Empty,
> > > > > > assigning,
> > > > > > > >> > > > > > reconciling,
> > > > > > > >> > > > > > > > > stable,
> > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > > section,
> > > > > > > >> right?
> > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> machine
> > is
> > > > > > modelled
> > > > > > > >> as an
> > > > > > > >> > > > event
> > > > > > > >> > > > > > > > > loop"?
> > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > >> *__consumer_offsets*
> > > > > > > >> > > > > > > > > partitions?
> > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > owned
> > > > > by
> > > > > > a
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator?
> > > > > > > >> > > > > > > > > For
> > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > > consumer
> > > > > > > >> groups, and
> > > > > > > >> > > > > > both
> > > > > > > >> > > > > > > > > exist in
> > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > state
> > > > > > machine
> > > > > > > >> for it,
> > > > > > > >> > > > or
> > > > > > > >> > > > > > 2?
> > > > > > > >> > > > > > > > > > > 1.4. I know the
> > "*group.coordinator.threads"
> > > > *is
> > > > > > the
> > > > > > > >> number
> > > > > > > >> > > > of
> > > > > > > >> > > > > > threads
> > > > > > > >> > > > > > > > > used
> > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > wondering
> > > > if
> > > > > > the
> > > > > > > >> purpose
> > > > > > > >> > > > of
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > threads
> > > > > > > >> > > > > > > > > > > is only to keep the state of each
> consumer
> > > > group
> > > > > > (or
> > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > why
> > > > > > should we
> > > > > > > >> need
> > > > > > > >> > > > > > > > > multi-threads
> > > > > > > >> > > > > > > > > > > here?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > configs:
> > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > does
> > > > the
> > > > > > > >> default
> > > > > > > >> > > > session
> > > > > > > >> > > > > > > > > timeout not
> > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > thought
> > > > > > the
> > > > > > > >> min/max
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound
> of
> > it,
> > > > > no?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > 30s The
> > > > > > > >> timeout to
> > > > > > > >> > > > detect
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > protocol.
> > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> > int 45s
> > > > > The
> > > > > > > >> minimum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> > int 60s
> > > > > The
> > > > > > > >> maximum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> are
> > > > > [range,
> > > > > > > >> uniform],
> > > > > > > >> > > > > > which means
> > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > like to
> > > > > > know
> > > > > > > >> why not
> > > > > > > >> > > > > > uniform
> > > > > > > >> > > > > > > > > one? I
> > > > > > > >> > > > > > > > > > > thought usually users will choose
> uniform
> > > > > assignor
> > > > > > > >> (former
> > > > > > > >> > > > sticky
> > > > > > > >> > > > > > > > > assinor)
> > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> other
> > > > reason
> > > > > > we
> > > > > > > >> choose
> > > > > > > >> > > > range
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > as default?
> > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > uniform
> > > > The
> > > > > > > >> server side
> > > > > > > >> > > > > > assignors.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > >> > > > > > > > > > > Luke
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke
> Chen
> > <
> > > > > > > >> showuon@gmail.com
> > > > > > > >> > > > >
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> Connect
> > > > > > > >> integrating with
> > > > > > > >> > > > > > KIP-848,
> > > > > > > >> > > > > > > > > but I
> > > > > > > >> > > > > > > > > > > > think we should have a separate
> > discussion
> > > > > > thread
> > > > > > > >> for the
> > > > > > > >> > > > Kafka
> > > > > > > >> > > > > > > > > Connect
> > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With
> New
> > > > > Consumer
> > > > > > > >> Rebalance
> > > > > > > >> > > > > > Protocol
> > > > > > > >> > > > > > > > > [1],
> > > > > > > >> > > > > > > > > > > > and let this discussion thread focus
> on
> > > > > consumer
> > > > > > > >> rebalance
> > > > > > > >> > > > > > protocol,
> > > > > > > >> > > > > > > > > WDYT?
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > > >
> > > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar
> <
> > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > feedback.
> > > > > > Looks
> > > > > > > >> like
> > > > > > > >> > > > there's
> > > > > > > >> > > > > > > > > agreement on
> > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> > would
> > > > > > revisit
> > > > > > > >> the doc
> > > > > > > >> > > > and
> > > > > > > >> > > > > > see
> > > > > > > >> > > > > > > > > what
> > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David
> > Jacot
> > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > document.
> > > > > > That's
> > > > > > > >> really
> > > > > > > >> > > > > > helpful. I
> > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> > Connect
> > > > > > and the
> > > > > > > >> > > > Consumer
> > > > > > > >> > > > > > could
> > > > > > > >> > > > > > > > > share
> > > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> > main
> > > > > > > >> difference is
> > > > > > > >> > > > that
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > Consumer
> > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > > > whereas
> > > > > > > >> Connect
> > > > > > > >> > > > assigns
> > > > > > > >> > > > > > tasks
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > forward:
> > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs
> to
> > > > > support
> > > > > > > >> different
> > > > > > > >> > > > > > resource
> > > > > > > >> > > > > > > > > types
> > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > Connect.
> > > > > The
> > > > > > > >> dedicated
> > > > > > > >> > > > APIs
> > > > > > > >> > > > > > > > > would be
> > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > different on
> > > > > the
> > > > > > > >> > > > > > content/resources and
> > > > > > > >> > > > > > > > > > > >> > they would rely on the same engine
> > on the
> > > > > > > >> coordinator
> > > > > > > >> > > > side.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> because
> > I am
> > > > > > not a
> > > > > > > >> fan of
> > > > > > > >> > > > > > > > > overcharging
> > > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> > That
> > > > > being
> > > > > > > >> said, I am
> > > > > > > >> > > > not
> > > > > > > >> > > > > > > > > opposed to
> > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to
> > do
> > > > it.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > discuss
> > > > it
> > > > > > here
> > > > > > > >> for now
> > > > > > > >> > > > in
> > > > > > > >> > > > > > order
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible
> > with
> > > > > what
> > > > > > we
> > > > > > > >> will do
> > > > > > > >> > > > for
> > > > > > > >> > > > > > > > > Connect in
> > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM
> David
> > > > Jacot
> > > > > <
> > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will
> go
> > > > > through
> > > > > > and
> > > > > > > >> address
> > > > > > > >> > > > > > your
> > > > > > > >> > > > > > > > > comments
> > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for
> your
> > > > > > feedback.
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> > Gregory
> > > > > > Harris
> > > > > > > >> <
> > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > wonderful to
> > > > > see
> > > > > > > >> > > > cooperative
> > > > > > > >> > > > > > > > > rebalancing
> > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > > > section
> > > > > > > >> describes how
> > > > > > > >> > > > > > member
> > > > > > > >> > > > > > > > > epoch
> > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > >> > > > > > > > > > > >> > > > advance in step with the group
> > epoch
> > > > > and
> > > > > > > >> assignment
> > > > > > > >> > > > > > epoch
> > > > > > > >> > > > > > > > > values. I
> > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for
> the
> > > > > reasons
> > > > > > > >> described
> > > > > > > >> > > > in
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > KIP. When
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > assigning
> > > > > > > >> partitions to a
> > > > > > > >> > > > > > worker,
> > > > > > > >> > > > > > > > > what
> > > > > > > >> > > > > > > > > > > >> member
> > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > assignment
> > > > > > use?
> > > > > > > >> Are
> > > > > > > >> > > > member
> > > > > > > >> > > > > > epochs
> > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > correspond
> > > > to
> > > > > > > >> multiple
> > > > > > > >> > > > > > different
> > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> > field
> > > > > > opaque
> > > > > > > >> to the
> > > > > > > >> > > > group
> > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > > > assignor
> > > > > > > >> > > > implementations
> > > > > > > >> > > > > > > > > interact
> > > > > > > >> > > > > > > > > > > >> with
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> > common
> > > > > > meaning
> > > > > > > >> agreed
> > > > > > > >> > > > > > upon? If
> > > > > > > >> > > > > > > > > so, what
> > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> > field
> > > > over
> > > > > > > >> including
> > > > > > > >> > > > such
> > > > > > > >> > > > > > > > > functionality
> > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 3. The following is included in
> > the
> > > > > KIP:
> > > > > > > >> "Thanks to
> > > > > > > >> > > > > > this, the
> > > > > > > >> > > > > > > > > input
> > > > > > > >> > > > > > > > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> entirely
> > > > driven
> > > > > > by
> > > > > > > >> the group
> > > > > > > >> > > > > > > > > coordinator.
> > > > > > > >> > > > > > > > > > > >> The
> > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > responsible for
> > > > > > > >> maintaining
> > > > > > > >> > > > any
> > > > > > > >> > > > > > state
> > > > > > > >> > > > > > > > > besides
> > > > > > > >> > > > > > > > > > > >> its
> > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this
> > mean
> > > > > > that the
> > > > > > > >> > > > > > client-side
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > non-Metadata
> > > > > state
> > > > > > > >> (such as
> > > > > > > >> > > > > > partition
> > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > topics,
> > > > > etc),
> > > > > > or
> > > > > > > >> that
> > > > > > > >> > > > > > additional
> > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> separate
> > > > > classes
> > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > >> > > > > >
> > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > >> > > > > > that
> > > > > > > >> > > > > > > > > seem to
> > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > possible
> > > > > for
> > > > > > > >> these two
> > > > > > > >> > > > > > > > > implementations
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> > promote
> > > > > > feature
> > > > > > > >> parity
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > > server-side
> > > > > > > >> > > > > > > > > > > >> and
> > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and
> would
> > also
> > > > > > > >> facilitate
> > > > > > > >> > > > > > operational
> > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> example,
> > if a
> > > > > > > >> server-side
> > > > > > > >> > > > > > assignor
> > > > > > > >> > > > > > > > > has some
> > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > deploying
> > > > > the
> > > > > > > >> patched
> > > > > > > >> > > > > > assignor to
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer
> group
> > to a
> > > > > > > >> client-side
> > > > > > > >> > > > > > assignor
> > > > > > > >> > > > > > > > > may be
> > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > >> > > > > > > > > > > >> > > > and less risky than patching
> all
> > of
> > > > the
> > > > > > > >> brokers.
> > > > > > > >> > > > With
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > currently
> > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > non-trivial
> > > > > > > >> > > > reimplementation
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > have
> > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs
> > have
> > > > > > diverged
> > > > > > > >> > > > > > significantly,
> > > > > > > >> > > > > > > > > then it
> > > > > > > >> > > > > > > > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > > > possible that a
> reimplementation
> > > > would
> > > > > > not be
> > > > > > > >> > > > possible.
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM
> > Sagar
> > > > <
> > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page
> to
> > > > > discuss
> > > > > > how
> > > > > > > >> Connect
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > need to
> > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > > protocol.
> > > > > > > >> Here's the
> > > > > > > >> > > > page:
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > > >
> > > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and
> I
> > have
> > > > > > tried
> > > > > > > >> to keep
> > > > > > > >> > > > a
> > > > > > > >> > > > > > format
> > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > > > think.
> > > > > > Also,
> > > > > > > >> do you
> > > > > > > >> > > > > > think this
> > > > > > > >> > > > > > > > > > > >> should
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> discussion
> > > > thread
> > > > > > or
> > > > > > > >> is this
> > > > > > > >> > > > one
> > > > > > > >> > > > > > fine?
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37
> AM
> > > > > Sagar <
> > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> > doc on
> > > > > > Kafka
> > > > > > > >> Streams.
> > > > > > > >> > > > > > Sure, I
> > > > > > > >> > > > > > > > > would do
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> > such a
> > > > > > > >> document.
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at
> 4:47
> > AM
> > > > > > Guozhang
> > > > > > > >> Wang <
> > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> > could
> > > > > come
> > > > > > > >> back with
> > > > > > > >> > > > some
> > > > > > > >> > > > > > > > > analysis on
> > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > > > integration
> > > > > > > >> with
> > > > > > > >> > > > the new
> > > > > > > >> > > > > > > > > protocol;
> > > > > > > >> > > > > > > > > > > >> so
> > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the
> > new
> > > > > > "protocol
> > > > > > > >> type"
> > > > > > > >> > > > we
> > > > > > > >> > > > > > did not
> > > > > > > >> > > > > > > > > yet
> > > > > > > >> > > > > > > > > > > >> think
> > > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > > implementations.
> > > > > > For
> > > > > > > >> Streams
> > > > > > > >> > > > > > here's a
> > > > > > > >> > > > > > > > > draft of
> > > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > > >
> > > >
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis
> > on
> > > > > > Connect.
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at
> > 10:48 PM
> > > > > > Sagar <
> > > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> > response.
> > > > The
> > > > > > > >> reason I
> > > > > > > >> > > > thought
> > > > > > > >> > > > > > > > > connect can
> > > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that
> > even
> > > > > > today the
> > > > > > > >> > > > connect
> > > > > > > >> > > > > > uses a
> > > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > > AbstractCoordinator to
> > > > > > > >> empower
> > > > > > > >> > > > > > > > > rebalances of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > > > > >> WorkerCoordinator sets
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata()
> > method
> > > > > by
> > > > > > > >> plumbing
> > > > > > > >> > > > into
> > > > > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to
> > support
> > > > > > > >> connect would
> > > > > > > >> > > > be
> > > > > > > >> > > > > > > > > similar at a
> > > > > > > >> > > > > > > > > > > >> > high
> > > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams
> > mainly
> > > > > > because
> > > > > > > >> of the
> > > > > > > >> > > > > > Client
> > > > > > > >> > > > > > > > > side
> > > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > > implementation
> > > > > > > >> level, we
> > > > > > > >> > > > > > might
> > > > > > > >> > > > > > > > > need to
> > > > > > > >> > > > > > > > > > > >> make
> > > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this
> > new
> > > > > > > >> assignment
> > > > > > > >> > > > protocol
> > > > > > > >> > > > > > like
> > > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> > request/response and
> > > > > > > >> SyncGroup and
> > > > > > > >> > > > > > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar
> > lines
> > > > > to
> > > > > > > >> streams (or
> > > > > > > >> > > > > > there
> > > > > > > >> > > > > > > > > might be
> > > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> > > > detailed
> > > > > > > >> analysis of
> > > > > > > >> > > > the
> > > > > > > >> > > > > > same
> > > > > > > >> > > > > > > > > and we
> > > > > > > >> > > > > > > > > > > >> > can have
> > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> > thread for
> > > > > > that
> > > > > > > >> as that
> > > > > > > >> > > > would
> > > > > > > >> > > > > > > > > derail this
> > > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if
> > that
> > > > > sounds
> > > > > > > >> good to
> > > > > > > >> > > > you.
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at
> > 5:47
> > > > PM
> > > > > > David
> > > > > > > >> Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> > comments.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > > > > >> `Assignment#error`.
> > > > > > > >> > > > > > Sure, I
> > > > > > > >> > > > > > > > > can
> > > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > > transition C
> > > > > > from
> > > > > > > >> his
> > > > > > > >> > > > current
> > > > > > > >> > > > > > > > > assignment
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > his
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when
> > he
> > > > can
> > > > > > move
> > > > > > > >> to
> > > > > > > >> > > > epoch 3.
> > > > > > > >> > > > > > When
> > > > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is
> > updated
> > > > > and
> > > > > > > >> persisted
> > > > > > > >> > > > > > with all
> > > > > > > >> > > > > > > > > its
> > > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if
> they
> > are
> > > > > not
> > > > > > all
> > > > > > > >> revoked
> > > > > > > >> > > > > > yet. In
> > > > > > > >> > > > > > > > > other
> > > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> > becomes
> > > > the
> > > > > > target
> > > > > > > >> > > > > > assignment.
> > > > > > > >> > > > > > > > > This is
> > > > > > > >> > > > > > > > > > > >> > basically
> > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid
> > having
> > > > > to
> > > > > > > >> write all
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > changes to
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > log.
> > > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on
> > the
> > > > > > persisted
> > > > > > > >> state
> > > > > > > >> > > > so I
> > > > > > > >> > > > > > > > > understand
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see
> > if I
> > > > can
> > > > > > > >> improve
> > > > > > > >> > > > this in
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> > description.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect,
> it
> > > > could
> > > > > > > >> reuse the
> > > > > > > >> > > > > > protocol
> > > > > > > >> > > > > > > > > with a
> > > > > > > >> > > > > > > > > > > >> > client side
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in
> > the
> > > > > > > >> protocol. The
> > > > > > > >> > > > > > assignment
> > > > > > > >> > > > > > > > > is
> > > > > > > >> > > > > > > > > > > >> about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> > > > metadata,
> > > > > > could
> > > > > > > >> > > > Connect
> > > > > > > >> > > > > > fit
> > > > > > > >> > > > > > > > > into this?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022
> at
> > 1:55
> > > > > PM
> > > > > > > >> Sagar <
> > > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP.
> I
> > just
> > > > > had
> > > > > > > >> minor
> > > > > > > >> > > > > > observations:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment
> > Error
> > > > > > > >> section in
> > > > > > > >> > > > Client
> > > > > > > >> > > > > > Side
> > > > > > > >> > > > > > > > > mode
> > > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In
> > this
> > > > > > case,
> > > > > > > >> the
> > > > > > > >> > > > client
> > > > > > > >> > > > > > side
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > >> can
> > > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > > > > coordinator`.
> > > > > > > >> In this
> > > > > > > >> > > > > > case are
> > > > > > > >> > > > > > > > > you
> > > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning
> an
> > > > > > > >> AssignmentError
> > > > > > > >> > > > that's
> > > > > > > >> > > > > > > > > listed down
> > > > > > > >> > > > > > > > > > > >> > towards
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think
> > it
> > > > > would
> > > > > > > >> make sense
> > > > > > > >> > > > to
> > > > > > > >> > > > > > > > > mention this
> > > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case
> Studies
> > > > > > section, I
> > > > > > > >> have a
> > > > > > > >> > > > > > slight
> > > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the
> same.
> > > > > Consider
> > > > > > > >> this step:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats,
> the
> > > > group
> > > > > > > >> coordinator
> > > > > > > >> > > > > > > > > transitions him
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions
> to
> > > > > > revoke. It
> > > > > > > >> > > > persists
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > change and
> > > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target
> Assignment
> > > > > > (epoch=3)
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > > partitions=[foo-0]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > > partitions=[foo-2]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > > partitions=[foo-1]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member
> Assignment
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > > > > >> partitions=[foo-0,
> > > > > > > >> > > > > > foo-1]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > > > > >> partitions=[foo-2]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > > > > >> partitions=[foo-1]
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats,
> it
> > > > > > transitions
> > > > > > > >> to
> > > > > > > >> > > > epoch 3
> > > > > > > >> > > > > > but
> > > > > > > >> > > > > > > > > cannot
> > > > > > > >> > > > > > > > > > > >> get
> > > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned
> > that
> > > > > > member C
> > > > > > > >> can't
> > > > > > > >> > > > get
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > foo-1
> > > > > > > >> > > > > > > > > > > >> > partition
> > > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> > description
> > > > > > above,
> > > > > > > >> it seems
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > > > already has
> > > > > > > >> > > > > > > > > > > >> it.
> > > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to
> > remove
> > > > it
> > > > > > and
> > > > > > > >> > > > populate it
> > > > > > > >> > > > > > only
> > > > > > > >> > > > > > > > > when it
> > > > > > > >> > > > > > > > > > > >> > actually
> > > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot
> of
> > > > other
> > > > > > > >> places, so
> > > > > > > >> > > > have
> > > > > > > >> > > > > > I
> > > > > > > >> > > > > > > > > > > >> understood it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect ,
> it
> > > > might
> > > > > > be
> > > > > > > >> out of
> > > > > > > >> > > > scope
> > > > > > > >> > > > > > of
> > > > > > > >> > > > > > > > > this
> > > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it
> > would
> > > > > > > >> probably be
> > > > > > > >> > > > > > running in
> > > > > > > >> > > > > > > > > client
> > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> > rebalance
> > > > > > > >> protocol as it
> > > > > > > >> > > > has
> > > > > > > >> > > > > > its
> > > > > > > >> > > > > > > > > own
> > > > > > > >> > > > > > > > > > > >> Custom
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > > IncrementalCooperative).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022
> > at
> > > > 5:00
> > > > > > PM
> > > > > > > >> David
> > > > > > > >> > > > Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our
> > goal
> > > > is
> > > > > > to
> > > > > > > >> move
> > > > > > > >> > > > forward
> > > > > > > >> > > > > > with
> > > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying
> > on
> > > > one
> > > > > > > >> generic API.
> > > > > > > >> > > > For
> > > > > > > >> > > > > > > > > Connect, we
> > > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern
> > and
> > > > > > > >> reuse/share the
> > > > > > > >> > > > core
> > > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For
> the
> > > > > schema
> > > > > > > >> registry, I
> > > > > > > >> > > > > > think
> > > > > > > >> > > > > > > > > that we
> > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored
> > API to
> > > > > do
> > > > > > > >> simple
> > > > > > > >> > > > > > > > > membership/leader
> > > > > > > >> > > > > > > > > > > >> > election.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15,
> > 2022 at
> > > > > > 10:22
> > > > > > > >> AM Ismael
> > > > > > > >> > > > > > Juma <
> > > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> > comments:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > > >> java.util.regex.Pattern
> > > > > > > >> > > > vs
> > > > > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern,
> we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > > differences
> > > > > > in
> > > > > > > >> more
> > > > > > > >> > > > detail
> > > > > > > >> > > > > > before
> > > > > > > >> > > > > > > > > > > >> deciding
> > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if
> > people
> > > > > pass
> > > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > > >> > > > > > > > > > > >> they
> > > > > > > >> > > > > > > > > > > >> > expect
> > > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be
> > > > honored.
> > > > > > If
> > > > > > > >> we are
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > > > > something
> > > > > > > >> > > > > > > > > > > >> > different,
> > > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider
> > adding
> > > > > an
> > > > > > > >> overload
> > > > > > > >> > > > with
> > > > > > > >> > > > > > our own
> > > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > > >> > > > > > > > > > > >> > class
> > > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want
> to
> > > > > expose
> > > > > > > >> re2j's at
> > > > > > > >> > > > this
> > > > > > > >> > > > > > > > > point).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding
> topic
> > > > ids,
> > > > > > any
> > > > > > > >> major new
> > > > > > > >> > > > > > protocol
> > > > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should
> > handle
> > > > the
> > > > > > topic
> > > > > > > >> > > > > > recreation case
> > > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to
> > > > handle.
> > > > > I
> > > > > > > >> agree with
> > > > > > > >> > > > > > David
> > > > > > > >> > > > > > > > > that we'd
> > > > > > > >> > > > > > > > > > > >> > want to
> > > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > > > > protocols
> > > > > > > >> that don't
> > > > > > > >> > > > > > have it
> > > > > > > >> > > > > > > > > yet and
> > > > > > > >> > > > > > > > > > > >> > that we
> > > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> > internals
> > > > > > versus
> > > > > > > >> adding
> > > > > > > >> > > > new
> > > > > > > >> > > > > > APIs
> > > > > > > >> > > > > > > > > to the
> > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> > adding
> > > > new
> > > > > > APIs
> > > > > > > >> is
> > > > > > > >> > > > required
> > > > > > > >> > > > > > for
> > > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still
> not
> > sure
> > > > > > about
> > > > > > > >> the
> > > > > > > >> > > > > > coordinator
> > > > > > > >> > > > > > > > > storing
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> > configs
> > > > to
> > > > > be
> > > > > > > >> > > > centralized
> > > > > > > >> > > > > > in the
> > > > > > > >> > > > > > > > > > > >> metadata
> > > > > > > >> > > > > > > > > > > >> > log for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> > (auditability,
> > > > > > > >> visibility,
> > > > > > > >> > > > > > consistency,
> > > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > > automatically
> > > > > > > >> deleting
> > > > > > > >> > > > > > configs in a
> > > > > > > >> > > > > > > > > way
> > > > > > > >> > > > > > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A
> good
> > > > > property
> > > > > > > >> for modern
> > > > > > > >> > > > > > systems
> > > > > > > >> > > > > > > > > is to
> > > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable
> data
> > > > loss
> > > > > > > >> scenarios.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13,
> > 2022
> > > > at
> > > > > > 3:47
> > > > > > > >> PM David
> > > > > > > >> > > > > > Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> > Guozhang. My
> > > > > > answers
> > > > > > > >> are
> > > > > > > >> > > > below:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > migration
> > > > > path,
> > > > > > > >> especially
> > > > > > > >> > > > > > the last
> > > > > > > >> > > > > > > > > step
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable
> the
> > new
> > > > > > > >> protocol, in
> > > > > > > >> > > > which
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> > rpcs
> > > > and
> > > > > > old
> > > > > > > >> > > > protocols /
> > > > > > > >> > > > > > rpcs
> > > > > > > >> > > > > > > > > are
> > > > > > > >> > > > > > > > > > > >> used
> > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How
> the
> > > > > > > >> coordinator could
> > > > > > > >> > > > > > "mimic"
> > > > > > > >> > > > > > > > > the old
> > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> > is
> > > > > > something
> > > > > > > >> we
> > > > > > > >> > > > need to
> > > > > > > >> > > > > > > > > present
> > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > > > > published
> > > > > > a
> > > > > > > >> new
> > > > > > > >> > > > version
> > > > > > > >> > > > > > of KIP
> > > > > > > >> > > > > > > > > which
> > > > > > > >> > > > > > > > > > > >> > includes
> > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about
> > this.
> > > > > See
> > > > > > the
> > > > > > > >> > > > "Supporting
> > > > > > > >> > > > > > > > > Online
> > > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > > "Compatibility,
> > > > > > > >> > > > Deprecation,
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > Migration
> > > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to
> think
> > > > > through
> > > > > > a
> > > > > > > >> few cases
> > > > > > > >> > > > > > now but
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> > should be
> > > > > > > >> understandable.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> > of
> > > > > topic
> > > > > > > >> ids. So
> > > > > > > >> > > > far as
> > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> > and
> > > > > admin
> > > > > > > >> client,
> > > > > > > >> > > > but
> > > > > > > >> > > > > > they
> > > > > > > >> > > > > > > > > are not
> > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> > yet.
> > > > I
> > > > > > think
> > > > > > > >> the
> > > > > > > >> > > > > > question is,
> > > > > > > >> > > > > > > > > first
> > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > > maintaining
> > > > > > > >> the
> > > > > > > >> > > > names
> > > > > > > >> > > > > > -> ids
> > > > > > > >> > > > > > > > > mapping
> > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > augmented
> > > > > > > >> existing
> > > > > > > >> > > > RPCs
> > > > > > > >> > > > > > and the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > should
> > > > we
> > > > > > ever
> > > > > > > >> consider
> > > > > > > >> > > > > > exposing
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs
> as
> > > > well
> > > > > > (both
> > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > >> > > > > > > > > > > >> well
> > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> > cases
> > > > > > like
> > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming
> > that we
> > > > > > would
> > > > > > > >> include
> > > > > > > >> > > > > > converting
> > > > > > > >> > > > > > > > > all
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using
> > topic
> > > > > ids
> > > > > > in
> > > > > > > >> this
> > > > > > > >> > > > KIP,
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > would be
> > > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate
> > with
> > > > > topic
> > > > > > > >> ids. That
> > > > > > > >> > > > > > being
> > > > > > > >> > > > > > > > > said, it
> > > > > > > >> > > > > > > > > > > >> > still has
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > names in
> > > > > > various
> > > > > > > >> APIs so
> > > > > > > >> > > > > > having a
> > > > > > > >> > > > > > > > > > > >> mapping
> > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> > inevitable to
> > > > > me.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't
> have
> > a
> > > > > strong
> > > > > > > >> opinion on
> > > > > > > >> > > > > > this.
> > > > > > > >> > > > > > > > > Here I
> > > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> > scope of
> > > > > this
> > > > > > > >> KIP. I
> > > > > > > >> > > > would
> > > > > > > >> > > > > > rather
> > > > > > > >> > > > > > > > > focus
> > > > > > > >> > > > > > > > > > > >> on
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we
> can
> > > > > consider
> > > > > > > >> this
> > > > > > > >> > > > > > separately if
> > > > > > > >> > > > > > > > > we see
> > > > > > > >> > > > > > > > > > > >> > value in
> > > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > > > > Ismael's
> > > > > > > >> point about
> > > > > > > >> > > > > > using
> > > > > > > >> > > > > > > > > topic ids
> > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > > >> > > > think
> > > > > > > >> > > > > > that
> > > > > > > >> > > > > > > > > there
> > > > > > > >> > > > > > > > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > one
> > > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it.
> > The
> > > > > > consumer
> > > > > > > >> will
> > > > > > > >> > > > have
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> > exists
> > > > > before
> > > > > > > >> passing
> > > > > > > >> > > > them
> > > > > > > >> > > > > > into
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> group
> > > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> > Obviously,
> > > > > the
> > > > > > > >> > > > coordinator
> > > > > > > >> > > > > > will
> > > > > > > >> > > > > > > > > also
> > > > > > > >> > > > > > > > > > > >> notice
> > > > > > > >> > > > > > > > > > > >> > it but
> > > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really
> have
> > a
> > > > way
> > > > > to
> > > > > > > >> reject an
> > > > > > > >> > > > > > invalid
> > > > > > > >> > > > > > > > > topic in
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> > with
> > > > > > David
> > > > > > > >> on all
> > > > > > > >> > > > other
> > > > > > > >> > > > > > minor
> > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > `subscribe(Pattern)`
> > > > > > > >> question:
> > > > > > > >> > > > > > personally
> > > > > > > >> > > > > > > > > I think
> > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate
> the
> > > > > > subscribe
> > > > > > > >> API with
> > > > > > > >> > > > > > Pattern,
> > > > > > > >> > > > > > > > > but
> > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > documenting
> > > > > > > >> that our
> > > > > > > >> > > > > > > > > subscription may
> > > > > > > >> > > > > > > > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > incompatible
> > > > > > > >> case is a
> > > > > > > >> > > > > > very rare
> > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > `String`
> > > > > > based
> > > > > > > >> > > > > > subscription may
> > > > > > > >> > > > > > > > > be
> > > > > > > >> > > > > > > > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > regexes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> > work. I
> > > > > have
> > > > > > to
> > > > > > > >> look at
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > differences
> > > > > > > >> > > > > > > > > > > >> > between
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> > better
> > > > > > > >> understand the
> > > > > > > >> > > > > > potential
> > > > > > > >> > > > > > > > > issues.
> > > > > > > >> > > > > > > > > > > >> My
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would
> work
> > for
> > > > > all
> > > > > > the
> > > > > > > >> basic
> > > > > > > >> > > > > > regular
> > > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the
> two
> > are
> > > > > > mainly
> > > > > > > >> about
> > > > > > > >> > > > the
> > > > > > > >> > > > > > various
> > > > > > > >> > > > > > > > > > > >> character
> > > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what
> > other
> > > > > people
> > > > > > > >> think
> > > > > > > >> > > > about
> > > > > > > >> > > > > > this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul
> 12,
> > 2022
> > > > > at
> > > > > > > >> 11:28 PM
> > > > > > > >> > > > > > Guozhang
> > > > > > > >> > > > > > > > > Wang <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks
> David!
> > I
> > > > > think
> > > > > > on
> > > > > > > >> the high
> > > > > > > >> > > > > > level
> > > > > > > >> > > > > > > > > there are
> > > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to
> concretize
> > a
> > > > bit
> > > > > > more:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> > migration
> > > > > path,
> > > > > > > >> especially
> > > > > > > >> > > > > > the last
> > > > > > > >> > > > > > > > > step
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable
> the
> > new
> > > > > > > >> protocol, in
> > > > > > > >> > > > which
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> > rpcs
> > > > and
> > > > > > old
> > > > > > > >> > > > protocols /
> > > > > > > >> > > > > > rpcs
> > > > > > > >> > > > > > > > > are
> > > > > > > >> > > > > > > > > > > >> used
> > > > > > > >> > > > > > > > > > > >> > by
> > > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How
> the
> > > > > > > >> coordinator could
> > > > > > > >> > > > > > "mimic"
> > > > > > > >> > > > > > > > > the old
> > > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> > is
> > > > > > something
> > > > > > > >> we
> > > > > > > >> > > > need to
> > > > > > > >> > > > > > > > > present
> > > > > > > >> > > > > > > > > > > >> about.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> > of
> > > > > topic
> > > > > > > >> ids. So
> > > > > > > >> > > > far as
> > > > > > > >> > > > > > > > > KIP-516 the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> > and
> > > > > admin
> > > > > > > >> client,
> > > > > > > >> > > > but
> > > > > > > >> > > > > > they
> > > > > > > >> > > > > > > > > are not
> > > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> > yet.
> > > > I
> > > > > > think
> > > > > > > >> the
> > > > > > > >> > > > > > question is,
> > > > > > > >> > > > > > > > > first
> > > > > > > >> > > > > > > > > > > >> > should we
> > > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > > maintaining
> > > > > > > >> the
> > > > > > > >> > > > names
> > > > > > > >> > > > > > -> ids
> > > > > > > >> > > > > > > > > mapping
> > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > > augmented
> > > > > > > >> existing
> > > > > > > >> > > > RPCs
> > > > > > > >> > > > > > and the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> > should
> > > > we
> > > > > > ever
> > > > > > > >> consider
> > > > > > > >> > > > > > exposing
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs
> as
> > > > well
> > > > > > (both
> > > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > > >> > > > > > > > > > > >> well
> > > > > > > >> > > > > > > > > > > >> > as in
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> > cases
> > > > > > like
> > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> > with
> > > > > > David
> > > > > > > >> on all
> > > > > > > >> > > > other
> > > > > > > >> > > > > > minor
> > > > > > > >> > > > > > > > > > > >> questions
> > > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > `subscribe(Pattern)`
> > > > > > > >> question:
> > > > > > > >> > > > > > personally
> > > > > > > >> > > > > > > > > I think
> > > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate
> the
> > > > > > subscribe
> > > > > > > >> API with
> > > > > > > >> > > > > > Pattern,
> > > > > > > >> > > > > > > > > but
> > > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > > documenting
> > > > > > > >> that our
> > > > > > > >> > > > > > > > > subscription may
> > > > > > > >> > > > > > > > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > > incompatible
> > > > > > > >> case is a
> > > > > > > >> > > > > > very rare
> > > > > > > >> > > > > > > > > > > >> scenario
> > > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > > `String`
> > > > > > based
> > > > > > > >> > > > > > subscription may
> > > > > > > >> > > > > > > > > be
> > > > > > > >> > > > > > > > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> > regexes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul
> > 12,
> > > > 2022
> > > > > > at
> > > > > > > >> 5:23 AM
> > > > > > > >> > > > > > David Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > <djacot@confluent.io.invalid
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for
> > your
> > > > > > > >> feedback. Let
> > > > > > > >> > > > me
> > > > > > > >> > > > > > answer
> > > > > > > >> > > > > > > > > your
> > > > > > > >> > > > > > > > > > > >> > questions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I
> think
> > > > it's
> > > > > > > >> premature to
> > > > > > > >> > > > > > talk about
> > > > > > > >> > > > > > > > > > > >> target
> > > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal
> > of the
> > > > > > > >> existing group
> > > > > > > >> > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> > > > protocol
> > > > > > and
> > > > > > > >> hence
> > > > > > > >> > > > > > > > > deprecation/removal
> > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> > quickly
> > > > > > > >> applications
> > > > > > > >> > > > migrate
> > > > > > > >> > > > > > to
> > > > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes
> > > > sense.
> > > > > I
> > > > > > will
> > > > > > > >> > > > remove it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The
> KIP
> > > > says
> > > > > we
> > > > > > > >> intend to
> > > > > > > >> > > > > > release
> > > > > > > >> > > > > > > > > this in
> > > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If
> we
> > > > added
> > > > > > that
> > > > > > > >> as a
> > > > > > > >> > > > way to
> > > > > > > >> > > > > > > > > estimate
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> group
> > > > > > protocol, I
> > > > > > > >> also
> > > > > > > >> > > > > > suggest
> > > > > > > >> > > > > > > > > removing
> > > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > explain
> > > > my
> > > > > > > >> reasoning. As
> > > > > > > >> > > > > > > > > explained, I
> > > > > > > >> > > > > > > > > > > >> plan
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> coordinator
> > in
> > > > > Java
> > > > > > > >> while we
> > > > > > > >> > > > > > implement
> > > > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> > > > internals
> > > > > > will
> > > > > > > >> be
> > > > > > > >> > > > slightly
> > > > > > > >> > > > > > > > > different
> > > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Therefore, I
> > > > > wanted
> > > > > > to
> > > > > > > >> tighten
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > switch from
> > > > > > > >> > > > > > > > > > > >> > the old
> > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> coordinator
> > to
> > > > the
> > > > > > new
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator
> > > > > > > >> > > > > > > > > to a
> > > > > > > >> > > > > > > > > > > >> > major
> > > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> alternative
> > > > would
> > > > > > be to
> > > > > > > >> use a
> > > > > > > >> > > > flag
> > > > > > > >> > > > > > to do
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > switch
> > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> > software
> > > > > > upgrade.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We
> > need to
> > > > > > flesh
> > > > > > > >> out the
> > > > > > > >> > > > > > details of
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > migration
> > > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're
> > saying
> > > > we
> > > > > > will
> > > > > > > >> support
> > > > > > > >> > > > > > online
> > > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > > >> > > > > > > > > > > >> > Is
> > > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain
> > this
> > > > in
> > > > > > > >> detail. It
> > > > > > > >> > > > could
> > > > > > > >> > > > > > also
> > > > > > > >> > > > > > > > > be done
> > > > > > > >> > > > > > > > > > > >> > as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we
> will
> > > > > support
> > > > > > > >> online
> > > > > > > >> > > > > > migrations
> > > > > > > >> > > > > > > > > for the
> > > > > > > >> > > > > > > > > > > >> > group.
> > > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group
> > using
> > > > the
> > > > > > old
> > > > > > > >> protocol
> > > > > > > >> > > > > > will be
> > > > > > > >> > > > > > > > > able to
> > > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > briefly
> > > > > > explain
> > > > > > > >> how that
> > > > > > > >> > > > > > will work
> > > > > > > >> > > > > > > > > > > >> though.
> > > > > > > >> > > > > > > > > > > >> > It is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step
> > > > process:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> > cluster
> > > > > must
> > > > > > be
> > > > > > > >> > > > upgraded or
> > > > > > > >> > > > > > > > > rolled to a
> > > > > > > >> > > > > > > > > > > >> > software
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > > > > coordinator.
> > > > > > > >> Both the
> > > > > > > >> > > > > > old and
> > > > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support
> the
> > old
> > > > > > > >> protocol and
> > > > > > > >> > > > rely
> > > > > > > >> > > > > > on the
> > > > > > > >> > > > > > > > > same
> > > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can
> > work
> > > > > > together.
> > > > > > > >> This
> > > > > > > >> > > > point
> > > > > > > >> > > > > > is an
> > > > > > > >> > > > > > > > > > > >> offline
> > > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this
> one
> > live
> > > > > > > >> because it
> > > > > > > >> > > > would
> > > > > > > >> > > > > > require
> > > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> coordinator
> > and
> > > > > > > >> starting up the
> > > > > > > >> > > > > > new one
> > > > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > unavailabilities.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> > cluster's
> > > > > > > >> metadata
> > > > > > > >> > > > > > version/IBP
> > > > > > > >> > > > > > > > > must be
> > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable
> > the
> > > > new
> > > > > > > >> protocol.
> > > > > > > >> > > > This
> > > > > > > >> > > > > > cannot
> > > > > > > >> > > > > > > > > be done
> > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated
> > > > because
> > > > > > the
> > > > > > > >> old
> > > > > > > >> > > > > > coordinator
> > > > > > > >> > > > > > > > > doesn't
> > > > > > > >> > > > > > > > > > > >> > support
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> > consumers
> > > > > > must be
> > > > > > > >> > > > upgraded
> > > > > > > >> > > > > > to a
> > > > > > > >> > > > > > > > > version
> > > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> > (must
> > > > > have
> > > > > > > >> KIP-792).
> > > > > > > >> > > > If
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > consumer
> > > > > > > >> > > > > > > > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing
> > must be
> > > > > > done at
> > > > > > > >> this
> > > > > > > >> > > > point.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> > consumers
> > > > > > must
> > > > > > > >> be rolled
> > > > > > > >> > > > > > with the
> > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > >> > > > > > > > > > > >> > flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer
> > group
> > > > is
> > > > > > > >> automatically
> > > > > > > >> > > > > > > > > converted when
> > > > > > > >> > > > > > > > > > > >> > the first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the
> > new
> > > > > > protocol
> > > > > > > >> joins
> > > > > > > >> > > > the
> > > > > > > >> > > > > > group.
> > > > > > > >> > > > > > > > > While
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> > protocol are
> > > > > > being
> > > > > > > >> > > > upgraded,
> > > > > > > >> > > > > > the old
> > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > >> > > > > > > > > > > >> > is
> > > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> > clarify
> > > > all
> > > > > > of
> > > > > > > >> this in
> > > > > > > >> > > > the
> > > > > > > >> > > > > > KIP.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am
> > happy
> > > > > > that we
> > > > > > > >> are
> > > > > > > >> > > > > > pushing the
> > > > > > > >> > > > > > > > > pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it
> > seems
> > > > > like
> > > > > > > >> there
> > > > > > > >> > > > could be
> > > > > > > >> > > > > > some
> > > > > > > >> > > > > > > > > tricky
> > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> > > > mechanism
> > > > > > for
> > > > > > > >> users to
> > > > > > > >> > > > > > detect
> > > > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> they
> > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> > > > switching
> > > > > > to
> > > > > > > >> the new
> > > > > > > >> > > > > > protocol?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think
> > that I
> > > > am
> > > > > a
> > > > > > bit
> > > > > > > >> more
> > > > > > > >> > > > > > optimistic
> > > > > > > >> > > > > > > > > than
> > > > > > > >> > > > > > > > > > > >> you
> > > > > > > >> > > > > > > > > > > >> > on this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe
> > that the
> > > > > > > >> majority of
> > > > > > > >> > > > the
> > > > > > > >> > > > > > cases
> > > > > > > >> > > > > > > > > are
> > > > > > > >> > > > > > > > > > > >> simple
> > > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with
> > the
> > > > new
> > > > > > > >> engine. The
> > > > > > > >> > > > > > > > > coordinator will
> > > > > > > >> > > > > > > > > > > >> > verify
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject
> > the
> > > > > > consumer
> > > > > > > >> if the
> > > > > > > >> > > > > > regex is
> > > > > > > >> > > > > > > > > not
> > > > > > > >> > > > > > > > > > > >> valid.
> > > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> > path,
> > > > in
> > > > > > the
> > > > > > > >> worst
> > > > > > > >> > > > case,
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > first
> > > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group
> > will
> > > > be
> > > > > > > >> rejected.
> > > > > > > >> > > > This
> > > > > > > >> > > > > > should
> > > > > > > >> > > > > > > > > be used
> > > > > > > >> > > > > > > > > > > >> > as the
> > > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way
> for
> > > > > > customers to
> > > > > > > >> > > > validate
> > > > > > > >> > > > > > their
> > > > > > > >> > > > > > > > > regex
> > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would
> > be to
> > > > > > test
> > > > > > > >> them with
> > > > > > > >> > > > > > another
> > > > > > > >> > > > > > > > > group.
> > > > > > > >> > > > > > > > > > > >> For
> > > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be
> > done
> > > > in a
> > > > > > > >> pre-prod
> > > > > > > >> > > > > > environment.
> > > > > > > >> > > > > > > > > > > >> Another
> > > > > > > >> > > > > > > > > > > >> > way
> > > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > > > > >> consumer-group tool
> > > > > > > >> > > > to
> > > > > > > >> > > > > > > > > provide a
> > > > > > > >> > > > > > > > > > > >> regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism.
> > Would
> > > > > > this be
> > > > > > > >> > > > enough in
> > > > > > > >> > > > > > your
> > > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> > Related to
> > > > > the
> > > > > > last
> > > > > > > >> > > > question,
> > > > > > > >> > > > > > will
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> Java
> > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick
> > with the
> > > > > > > >> current regex
> > > > > > > >> > > > > > engine for
> > > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may
> be
> > > > handy
> > > > > to
> > > > > > > >> keep using
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > based
> > > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> migrations
> > > > > simple
> > > > > > and
> > > > > > > >> then
> > > > > > > >> > > > > > migrate to
> > > > > > > >> > > > > > > > > server
> > > > > > > >> > > > > > > > > > > >> > based
> > > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I
> understand
> > > > your
> > > > > > point
> > > > > > > >> but I
> > > > > > > >> > > > am
> > > > > > > >> > > > > > > > > concerned that
> > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> actually
> > stay
> > > > > in
> > > > > > > >> this mode.
> > > > > > > >> > > > That
> > > > > > > >> > > > > > > > > would go
> > > > > > > >> > > > > > > > > > > >> > against our
> > > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> simplifying
> > the
> > > > > > client
> > > > > > > >> because
> > > > > > > >> > > > we
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > have to
> > > > > > > >> > > > > > > > > > > >> > continue
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > metadata on
> > > > > the
> > > > > > > >> client
> > > > > > > >> > > > side. I
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > rather
> > > > > > > >> > > > > > > > > > > >> > not do
> > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When
> > we say
> > > > > > that
> > > > > > > >> the group
> > > > > > > >> > > > > > > > > coordinator
> > > > > > > >> > > > > > > > > > > >> will
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > > configurations
> > > > > > > >> and that
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > > > >> configurations
> > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > > > > deleted.
> > > > > > > >> Will a
> > > > > > > >> > > > > > transition to
> > > > > > > >> > > > > > > > > DEAD
> > > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > configurations?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> > right.
> > > > The
> > > > > > > >> > > > configurations
> > > > > > > >> > > > > > will be
> > > > > > > >> > > > > > > > > > > >> deleted
> > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted.
> > They go
> > > > > > > >> together.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will
> > the
> > > > > > choice to
> > > > > > > >> store
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > configs in
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder
> to
> > list
> > > > > all
> > > > > > > >> cluster
> > > > > > > >> > > > > > configs and
> > > > > > > >> > > > > > > > > their
> > > > > > > >> > > > > > > > > > > >> > values?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't
> > think
> > > > so.
> > > > > > The
> > > > > > > >> group
> > > > > > > >> > > > > > > > > configurations are
> > > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs.
> If
> > you
> > > > > > want to
> > > > > > > >> know
> > > > > > > >> > > > all
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > overrides
> > > > > > > >> > > > > > > > > > > >> > though,
> > > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all
> > the
> > > > > group
> > > > > > > >> > > > coordinators.
> > > > > > > >> > > > > > You
> > > > > > > >> > > > > > > > > cannot
> > > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > instance.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How
> > would
> > > > > > someone
> > > > > > > >> > > > configure a
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > before
> > > > > > > >> > > > > > > > > > > >> > starting
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> > considered
> > > > > > > >> allowing the
> > > > > > > >> > > > > > explicit
> > > > > > > >> > > > > > > > > creation
> > > > > > > >> > > > > > > > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs
> > could
> > > > be
> > > > > > > >> decoupled
> > > > > > > >> > > > from
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The
> > group
> > > > > will
> > > > > > be
> > > > > > > >> > > > > > automatically
> > > > > > > >> > > > > > > > > created in
> > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> configs
> > will
> > > > > be
> > > > > > > >> lost after
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > retention
> > > > > > > >> > > > > > > > > > > >> > period of
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will
> > the
> > > > > > > >> > > > Consumer.subscribe
> > > > > > > >> > > > > > method
> > > > > > > >> > > > > > > > > for the
> > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> `java.util.regex.Pattern` of
> > > > > > > >> > > > do
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > have to
> > > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a
> > very
> > > > > group
> > > > > > > >> question. I
> > > > > > > >> > > > > > forgot
> > > > > > > >> > > > > > > > > about
> > > > > > > >> > > > > > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > one.
> > > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> `java.util.regex.Pattern` is
> > > > > > > >> > > > not
> > > > > > > >> > > > > > fully
> > > > > > > >> > > > > > > > > > > >> compatible
> > > > > > > >> > > > > > > > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to
> > use,
> > > > it
> > > > > > > >> might be
> > > > > > > >> > > > better
> > > > > > > >> > > > > > to
> > > > > > > >> > > > > > > > > deprecate
> > > > > > > >> > > > > > > > > > > >> > it and
> > > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which
> takes
> > a
> > > > > > string.
> > > > > > > >> We would
> > > > > > > >> > > > > > rely on
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> server
> > > > > > > >> > > > > > > > > > > >> > side
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > > > > migration, I
> > > > > > > >> think
> > > > > > > >> > > > that
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > could
> > > > > > > >> > > > > > > > > > > >> still
> > > > > > > >> > > > > > > > > > > >> > try to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and
> > use
> > > > it.
> > > > > > That
> > > > > > > >> should
> > > > > > > >> > > > > > work, I
> > > > > > > >> > > > > > > > > think, in
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I
> > agree
> > > > with
> > > > > > > >> Justine
> > > > > > > >> > > > that we
> > > > > > > >> > > > > > > > > should be
> > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch
> to
> > > > > > > >> > > > IBP/metadata.version
> > > > > > > >> > > > > > from the
> > > > > > > >> > > > > > > > > > > >> feature
> > > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> > switch the
> > > > > > > >> default for
> > > > > > > >> > > > the
> > > > > > > >> > > > > > feature
> > > > > > > >> > > > > > > > > flag
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > true
> > > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > metadata.version
> > > > > > once
> > > > > > > >> we
> > > > > > > >> > > > want to
> > > > > > > >> > > > > > make
> > > > > > > >> > > > > > > > > it the
> > > > > > > >> > > > > > > > > > > >> > default.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan
> was
> > to
> > > > use
> > > > > > that
> > > > > > > >> feature
> > > > > > > >> > > > > > flag
> > > > > > > >> > > > > > > > > mainly
> > > > > > > >> > > > > > > > > > > >> > during the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I
> > should
> > > > > not
> > > > > > have
> > > > > > > >> > > > mentioned
> > > > > > > >> > > > > > it, I
> > > > > > > >> > > > > > > > > think,
> > > > > > > >> > > > > > > > > > > >> > because
> > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an
> internal
> > > > config
> > > > > > for
> > > > > > > >> it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some
> > of
> > > > the
> > > > > > > >> protocol APIs
> > > > > > > >> > > > > > don't
> > > > > > > >> > > > > > > > > mention
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to
> > add
> > > > that
> > > > > > for
> > > > > > > >> > > > consistency.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It
> is
> > a
> > > > bit
> > > > > > odd
> > > > > > > >> that
> > > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though
> it
> > > > seems
> > > > > > to do
> > > > > > > >> more
> > > > > > > >> > > > than
> > > > > > > >> > > > > > > > > reading.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree.
> > This is
> > > > > > how the
> > > > > > > >> > > > current
> > > > > > > >> > > > > > > > > protocol works
> > > > > > > >> > > > > > > > > > > >> > though.
> > > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require
> > "Read
> > > > > > Group" to
> > > > > > > >> join a
> > > > > > > >> > > > > > group. We
> > > > > > > >> > > > > > > > > could
> > > > > > > >> > > > > > > > > > > >> > consider
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am
> not
> > > > sure
> > > > > > that
> > > > > > > >> it is
> > > > > > > >> > > > worth
> > > > > > > >> > > > > > it.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How
> is
> > > > topic
> > > > > > > >> recreation
> > > > > > > >> > > > > > handled by
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> protocol?
> > It
> > > > > > would be
> > > > > > > >> good to
> > > > > > > >> > > > > > have a
> > > > > > > >> > > > > > > > > section
> > > > > > > >> > > > > > > > > > > >> on
> > > > > > > >> > > > > > > > > > > >> > this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> From
> > a
> > > > > > protocol
> > > > > > > >> > > > > > perspective, the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > will
> > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id
> so
> > it
> > > > > will
> > > > > > > >> treat it
> > > > > > > >> > > > like a
> > > > > > > >> > > > > > > > > topic with
> > > > > > > >> > > > > > > > > > > >> a
> > > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue
> > is
> > > > that
> > > > > > the
> > > > > > > >> > > > fetch/commit
> > > > > > > >> > > > > > > > > offsets
> > > > > > > >> > > > > > > > > > > >> APIs
> > > > > > > >> > > > > > > > > > > >> > do not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > > > > consumer
> > > > > > > >> would
> > > > > > > >> > > > reuse the
> > > > > > > >> > > > > > > > > offsets
> > > > > > > >> > > > > > > > > > > >> based
> > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > should
> > > > > > update
> > > > > > > >> those
> > > > > > > >> > > > APIs
> > > > > > > >> > > > > > as well
> > > > > > > >> > > > > > > > > in
> > > > > > > >> > > > > > > > > > > >> order
> > > > > > > >> > > > > > > > > > > >> > to be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end.
> That
> > > > would
> > > > > > > >> strengthen
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > semantics of
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The
> > KIP
> > > > > > mentions
> > > > > > > >> we will
> > > > > > > >> > > > > > write the
> > > > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is
> an
> > > > > > > >> implementation
> > > > > > > >> > > > > > detail, do
> > > > > > > >> > > > > > > > > we plan
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do
> we
> > > > have a
> > > > > > > >> scalability
> > > > > > > >> > > > > > goal when
> > > > > > > >> > > > > > > > > it
> > > > > > > >> > > > > > > > > > > >> comes
> > > > > > > >> > > > > > > > > > > >> > to how
> > > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group
> > protocol
> > > > > can
> > > > > > > >> support?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't
> > have
> > > > > > numbers
> > > > > > > >> at the
> > > > > > > >> > > > > > moment. The
> > > > > > > >> > > > > > > > > > > >> protocol
> > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members
> > per
> > > > > > group.
> > > > > > > >> We will
> > > > > > > >> > > > > > measure
> > > > > > > >> > > > > > > > > this when
> > > > > > > >> > > > > > > > > > > >> > we have
> > > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > implementation.
> > > > > Note
> > > > > > > >> that we
> > > > > > > >> > > > might
> > > > > > > >> > > > > > have
> > > > > > > >> > > > > > > > > other
> > > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g.
> > > > offset
> > > > > > > >> commits).
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did
> we
> > > > > > consider
> > > > > > > >> having
> > > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > > >> > > > > > > > > > > >> > instead
> > > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > SubscribedTopicNames
> > > > > > > >> in
> > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we
> > have
> > > > to
> > > > > > > >> resolve the
> > > > > > > >> > > > > > regex on
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> topic
> > > > name?
> > > > > > The
> > > > > > > >> > > > difference
> > > > > > > >> > > > > > is that
> > > > > > > >> > > > > > > > > > > >> sending
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas
> > > > sending
> > > > > > the
> > > > > > > >> topic
> > > > > > > >> > > > names
> > > > > > > >> > > > > > is less
> > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > recreation is
> > > > > > easier
> > > > > > > >> to
> > > > > > > >> > > > handle
> > > > > > > >> > > > > > if we
> > > > > > > >> > > > > > > > > have
> > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea
> > was to
> > > > > > > >> consolidate the
> > > > > > > >> > > > > > metadata
> > > > > > > >> > > > > > > > > lookup
> > > > > > > >> > > > > > > > > > > >> > on the
> > > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but
> I
> > do
> > > > > agree
> > > > > > > >> with your
> > > > > > > >> > > > > > point. As
> > > > > > > >> > > > > > > > > a
> > > > > > > >> > > > > > > > > > > >> second
> > > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be
> > > > better
> > > > > > here
> > > > > > > >> for the
> > > > > > > >> > > > > > delete and
> > > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose
> > that we
> > > > > may
> > > > > > > >> allow
> > > > > > > >> > > > users to
> > > > > > > >> > > > > > > > > subscribe
> > > > > > > >> > > > > > > > > > > >> with
> > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> > because
> > > > > that
> > > > > > is
> > > > > > > >> the only
> > > > > > > >> > > > > > way to be
> > > > > > > >> > > > > > > > > > > >> really
> > > > > > > >> > > > > > > > > > > >> > robust
> > > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> re-creation.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue,
> Jul
> > 12,
> > > > > > 2022 at
> > > > > > > >> 1:38 PM
> > > > > > > >> > > > > > David
> > > > > > > >> > > > > > > > > Jacot <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> > Justine,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks
> for
> > > > your
> > > > > > > >> comments.
> > > > > > > >> > > > Please
> > > > > > > >> > > > > > find
> > > > > > > >> > > > > > > > > my
> > > > > > > >> > > > > > > > > > > >> answers
> > > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes,
> > the new
> > > > > > > >> protocol
> > > > > > > >> > > > relies on
> > > > > > > >> > > > > > > > > topic IDs
> > > > > > > >> > > > > > > > > > > >> > with the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic
> > names
> > > > > based
> > > > > > in
> > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > >
> ConsumerGroupHeartbeatRequest.
> > > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using
> > topic
> > > > > > names
> > > > > > > >> is the
> > > > > > > >> > > > > > right call
> > > > > > > >> > > > > > > > > here.
> > > > > > > >> > > > > > > > > > > >> I
> > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little
> > more.
> > > > > > > >> Obviously, the
> > > > > > > >> > > > > > KIP does
> > > > > > > >> > > > > > > > > not
> > > > > > > >> > > > > > > > > > > >> > change the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets
> > RPCs
> > > > to
> > > > > > use
> > > > > > > >> topic
> > > > > > > >> > > > IDs.
> > > > > > > >> > > > > > This
> > > > > > > >> > > > > > > > > may be
> > > > > > > >> > > > > > > > > > > >> > something
> > > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include
> > though
> > > > > as
> > > > > > it
> > > > > > > >> would
> > > > > > > >> > > > give
> > > > > > > >> > > > > > better
> > > > > > > >> > > > > > > > > > > >> overall
> > > > > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> producer.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're
> > > > right.
> > > > > I
> > > > > > > >> think that
> > > > > > > >> > > > I
> > > > > > > >> > > > > > should
> > > > > > > >> > > > > > > > > not
> > > > > > > >> > > > > > > > > > > >> have
> > > > > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I
> > will
> > > > > remove
> > > > > > > >> it. We can
> > > > > > > >> > > > > > use an
> > > > > > > >> > > > > > > > > internal
> > > > > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > developing the
> > > > > > > >> feature.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both
> > cluster
> > > > > > types
> > > > > > > >> will be
> > > > > > > >> > > > > > > > > supported. The
> > > > > > > >> > > > > > > > > > > >> > change is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only
> > > > requirement
> > > > > > is
> > > > > > > >> that the
> > > > > > > >> > > > > > cluster
> > > > > > > >> > > > > > > > > uses
> > > > > > > >> > > > > > > > > > > >> topic
> > > > > > > >> > > > > > > > > > > >> > IDs.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon,
> > Jul
> > > > 11,
> > > > > > 2022
> > > > > > > >> at 9:53
> > > > > > > >> > > > PM
> > > > > > > >> > > > > > > > > Guozhang
> > > > > > > >> > > > > > > > > > > >> Wang <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> wangguoz@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi
> > Ismael,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks
> > for
> > > > the
> > > > > > > >> feedback.
> > > > > > > >> > > > Here
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > > some
> > > > > > > >> > > > > > > > > > > >> replies
> > > > > > > >> > > > > > > > > > > >> > > > > inlined
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On
> Sat,
> > Jul
> > > > 9,
> > > > > > 2022
> > > > > > > >> at
> > > > > > > >> > > > 2:53 AM
> > > > > > > >> > > > > > > > > Ismael Juma
> > > > > > > >> > > > > > > > > > > >> <
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > Thanks for
> > > > > the
> > > > > > > >> KIP. This
> > > > > > > >> > > > has
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> potential
> > > > > > > >> > > > > > > > > > > >> > to be a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > improvement. A
> > > > few
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > initial
> > > > > > > >> > > > questions/comments:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I
> > think
> > > > > > it's
> > > > > > > >> > > > premature to
> > > > > > > >> > > > > > talk
> > > > > > > >> > > > > > > > > about
> > > > > > > >> > > > > > > > > > > >> > target
> > > > > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> deprecation
> > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > removal of
> > > > > the
> > > > > > > >> existing
> > > > > > > >> > > > group
> > > > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > > > > > > > >> > Unlike
> > > > > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> client
> > > > > > protocol
> > > > > > > >> and hence
> > > > > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent
> on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how
> > > > quickly
> > > > > > > >> applications
> > > > > > > >> > > > > > migrate
> > > > > > > >> > > > > > > > > to the
> > > > > > > >> > > > > > > > > > > >> new
> > > > > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I
> > agree
> > > > > > with
> > > > > > > >> you. I
> > > > > > > >> > > > think
> > > > > > > >> > > > > > we can
> > > > > > > >> > > > > > > > > > > >> remove
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > `Compatibility,
> > > > > > > >> > > > Deprecation,
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > Migration
> > > > > > > >> > > > > > > > > > > >> > Plan` and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that
> we
> > will
> > > > > > decide
> > > > > > > >> in the
> > > > > > > >> > > > > > future
> > > > > > > >> > > > > > > > > about
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > we
> > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > protocol and
> > > > > > > >> behaviors.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2.
> > The KIP
> > > > > > says we
> > > > > > > >> > > > intend to
> > > > > > > >> > > > > > > > > release this
> > > > > > > >> > > > > > > > > > > >> > in 4.x,
> > > > > > > >> > > > > > > > > > > >> > > > > >> but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why.
> > If we
> > > > > > added
> > > > > > > >> that as
> > > > > > > >> > > > a
> > > > > > > >> > > > > > way to
> > > > > > > >> > > > > > > > > > > >> estimate
> > > > > > > >> > > > > > > > > > > >> > when
> > > > > > > >> > > > > > > > > > > >> > > > > we'd
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > group
> > > > > > > >> protocol, I
> > > > > > > >> > > > also
> > > > > > > >> > > > > > suggest
> > > > > > > >> > > > > > > > > > > >> removing
> > > > > > > >> > > > > > > > > > > >> > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I
> think
> > > > that's
> > > > > > not
> > > > > > > >> > > > specifically
> > > > > > > >> > > > > > > > > related to
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> deprecation/removal
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan,
> > but
> > > > it's
> > > > > > more
> > > > > > > >> for
> > > > > > > >> > > > client
> > > > > > > >> > > > > > > > > upgrades.
> > > > > > > >> > > > > > > > > > > >> I.e.
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > implementation
> > > > > > may
> > > > > > > >> be done
> > > > > > > >> > > > > > first,
> > > > > > > >> > > > > > > > > and then
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only
> > mark it
> > > > > as
> > > > > > > >> "released"
> > > > > > > >> > > > by
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > time
> > > > > > > >> > > > > > > > > > > >> clients
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that
> > time,
> > > > to
> > > > > > > >> enable the
> > > > > > > >> > > > > > feature the
> > > > > > > >> > > > > > > > > > > >> clients
> > > > > > > >> > > > > > > > > > > >> > need to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> bytecode
> > > > with
> > > > > a
> > > > > > > >> rolling
> > > > > > > >> > > > bounce
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > then set
> > > > > > > >> > > > > > > > > > > >> > the flag
> > > > > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> bounce,
> > and
> > > > > > hence
> > > > > > > >> we feel
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > > > > better to be
> > > > > > > >> > > > > > > > > > > >> > released
> > > > > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3.
> We
> > need
> > > > > to
> > > > > > > >> flesh out
> > > > > > > >> > > > the
> > > > > > > >> > > > > > > > > details of
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > migration
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds
> like
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> we're
> > > > saying
> > > > > > we
> > > > > > > >> will
> > > > > > > >> > > > support
> > > > > > > >> > > > > > online
> > > > > > > >> > > > > > > > > > > >> > migrations. Is
> > > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > explain
> > > > this
> > > > > > in
> > > > > > > >> detail.
> > > > > > > >> > > > It
> > > > > > > >> > > > > > could
> > > > > > > >> > > > > > > > > also be
> > > > > > > >> > > > > > > > > > > >> > done as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > easier.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> > think
> > > > > > that's
> > > > > > > >> the
> > > > > > > >> > > > part we
> > > > > > > >> > > > > > can
> > > > > > > >> > > > > > > > > be more
> > > > > > > >> > > > > > > > > > > >> > concrete
> > > > > > > >> > > > > > > > > > > >> > > > > >> about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this
> is
> > > > > related
> > > > > > to
> > > > > > > >> your
> > > > > > > >> > > > > > question 2)
> > > > > > > >> > > > > > > > > above).
> > > > > > > >> > > > > > > > > > > >> > We will
> > > > > > > >> > > > > > > > > > > >> > > > > >> work
> > > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > explicit in
> > > > > > > >> parallel as we
> > > > > > > >> > > > > > solicit
> > > > > > > >> > > > > > > > > more
> > > > > > > >> > > > > > > > > > > >> > feedback.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I
> > am
> > > > > happy
> > > > > > > >> that we are
> > > > > > > >> > > > > > pushing
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > pattern
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but
> it
> > > > seems
> > > > > > like
> > > > > > > >> there
> > > > > > > >> > > > > > could be
> > > > > > > >> > > > > > > > > some
> > > > > > > >> > > > > > > > > > > >> tricky
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> have a
> > > > > > mechanism
> > > > > > > >> for
> > > > > > > >> > > > users to
> > > > > > > >> > > > > > > > > detect that
> > > > > > > >> > > > > > > > > > > >> > they
> > > > > > > >> > > > > > > > > > > >> > > > > need
> > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> before
> > > > > > switching
> > > > > > > >> to the
> > > > > > > >> > > > new
> > > > > > > >> > > > > > > > > protocol?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> > think
> > > > we
> > > > > > need
> > > > > > > >> some
> > > > > > > >> > > > > > tooling for
> > > > > > > >> > > > > > > > > > > >> non-java
> > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > "dry-run"
> > > > the
> > > > > > > >> client before
> > > > > > > >> > > > > > > > > switching to
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> specific
> > > > idea
> > > > > on
> > > > > > > >> top of my
> > > > > > > >> > > > head
> > > > > > > >> > > > > > > > > though,
> > > > > > > >> > > > > > > > > > > >> maybe
> > > > > > > >> > > > > > > > > > > >> > others
> > > > > > > >> > > > > > > > > > > >> > > > > >> like
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett
> can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> chime-in
> > > > here?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5.
> > Related
> > > > > to
> > > > > > the
> > > > > > > >> last
> > > > > > > >> > > > > > question,
> > > > > > > >> > > > > > > > > will the
> > > > > > > >> > > > > > > > > > > >> > Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> stick
> > with
> > > > > the
> > > > > > > >> current
> > > > > > > >> > > > regex
> > > > > > > >> > > > > > > > > engine for
> > > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it
> > may be
> > > > > > handy
> > > > > > > >> to keep
> > > > > > > >> > > > using
> > > > > > > >> > > > > > > > > client
> > > > > > > >> > > > > > > > > > > >> based
> > > > > > > >> > > > > > > > > > > >> > regex
> > > > > > > >> > > > > > > > > > > >> > > > > at
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > migrations
> > > > > > simple
> > > > > > > >> and
> > > > > > > >> > > > then
> > > > > > > >> > > > > > migrate
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > server based
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> step.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > Honestly I
> > > > > have
> > > > > > not
> > > > > > > >> thought
> > > > > > > >> > > > > > about
> > > > > > > >> > > > > > > > > that for
> > > > > > > >> > > > > > > > > > > >> > java
> > > > > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that.
> > What
> > > > > kind
> > > > > > of
> > > > > > > >> > > > > > compatibility
> > > > > > > >> > > > > > > > > issues do
> > > > > > > >> > > > > > > > > > > >> > you have
> > > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6.
> > When we
> > > > > say
> > > > > > > >> that the
> > > > > > > >> > > > group
> > > > > > > >> > > > > > > > > coordinator
> > > > > > > >> > > > > > > > > > > >> > will be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > > > > > >> configurations and
> > > > > > > >> > > > that
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> > configurations
> > > > > > > >> > > > > > > > > > > >> > > > > will
> > > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> group
> > is
> > > > > > deleted.
> > > > > > > >> Will a
> > > > > > > >> > > > > > > > > transition to
> > > > > > > >> > > > > > > > > > > >> DEAD
> > > > > > > >> > > > > > > > > > > >> > > > > trigger
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > configurations?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes,
> > since
> > > > the
> > > > > > DEAD
> > > > > > > >> state
> > > > > > > >> > > > is an
> > > > > > > >> > > > > > > > > ending
> > > > > > > >> > > > > > > > > > > >> state
> > > > > > > >> > > > > > > > > > > >> > (we
> > > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state
> > when
> > > > the
> > > > > > > >> group is
> > > > > > > >> > > > EMPTY
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > also all
> > > > > > > >> > > > > > > > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > its
> > > > > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's
> > > > transited
> > > > > > to
> > > > > > > >> DEAD this
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > would
> > > > > > > >> > > > > > > > > > > >> never
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7.
> > Will
> > > > the
> > > > > > > >> choice to
> > > > > > > >> > > > store
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > configs
> > > > > > > >> > > > > > > > > > > >> in
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > harder to
> > > > > > list all
> > > > > > > >> > > > cluster
> > > > > > > >> > > > > > configs
> > > > > > > >> > > > > > > > > and
> > > > > > > >> > > > > > > > > > > >> their
> > > > > > > >> > > > > > > > > > > >> > > > > values?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> That's a
> > > > good
> > > > > > > >> question,
> > > > > > > >> > > > and our
> > > > > > > >> > > > > > > > > thoughts
> > > > > > > >> > > > > > > > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > that
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > configurations"
> > > > > > are
> > > > > > > >> > > > overrides
> > > > > > > >> > > > > > of the
> > > > > > > >> > > > > > > > > > > >> > cluster-level
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > customized
> > > > per
> > > > > > > >> group so
> > > > > > > >> > > > when an
> > > > > > > >> > > > > > > > > admin list
> > > > > > > >> > > > > > > > > > > >> > cluster
> > > > > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list
> > just
> > > > the
> > > > > > > >> cluster-level
> > > > > > > >> > > > > > > > > "defaults", not
> > > > > > > >> > > > > > > > > > > >> > showing
> > > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > customizations.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8.
> How
> > > > would
> > > > > > > >> someone
> > > > > > > >> > > > > > configure a
> > > > > > > >> > > > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > before
> > > > > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers?
> > Have
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > > > > considered
> > > > > > > >> allowing
> > > > > > > >> > > > the
> > > > > > > >> > > > > > explicit
> > > > > > > >> > > > > > > > > > > >> > creation of
> > > > > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > Alternatively,
> > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > configs
> > > > > could
> > > > > > be
> > > > > > > >> > > > decoupled
> > > > > > > >> > > > > > from
> > > > > > > >> > > > > > > > > the group
> > > > > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The
> > configs
> > > > > can
> > > > > > be
> > > > > > > >> created
> > > > > > > >> > > > > > before
> > > > > > > >> > > > > > > > > the group
> > > > > > > >> > > > > > > > > > > >> > itself
> > > > > > > >> > > > > > > > > > > >> > > > > as
> > > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of
> > > > course,
> > > > > > this
> > > > > > > >> > > > requires
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > > >> corresponding
> > > > > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > > > > > coordinator
> > > > > > > >> based on
> > > > > > > >> > > > the
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > id ---
> > > > > > > >> > > > > > > > > > > >> > the only
> > > > > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs
> is,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when
> the
> > > > group
> > > > > > > >> itself is
> > > > > > > >> > > > gone
> > > > > > > >> > > > > > we
> > > > > > > >> > > > > > > > > also check
> > > > > > > >> > > > > > > > > > > >> > if there
> > > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > configuration
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> entities
> > > > > > related to
> > > > > > > >> that
> > > > > > > >> > > > group
> > > > > > > >> > > > > > and
> > > > > > > >> > > > > > > > > delete
> > > > > > > >> > > > > > > > > > > >> as
> > > > > > > >> > > > > > > > > > > >> > well.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > Admittedly
> > > > > this
> > > > > > > >> indeed
> > > > > > > >> > > > > > introduces an
> > > > > > > >> > > > > > > > > > > >> > asymmetry on
> > > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > lifecycles
> > > > of
> > > > > > the
> > > > > > > >> config
> > > > > > > >> > > > > > entities,
> > > > > > > >> > > > > > > > > and we
> > > > > > > >> > > > > > > > > > > >> > would like
> > > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> feelings
> > > > > > whether we
> > > > > > > >> should
> > > > > > > >> > > > aim
> > > > > > > >> > > > > > for
> > > > > > > >> > > > > > > > > symmetry
> > > > > > > >> > > > > > > > > > > >> > i.e.
> > > > > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> configs
> > and
> > > > > > hence
> > > > > > > >> not
> > > > > > > >> > > > delete
> > > > > > > >> > > > > > them at
> > > > > > > >> > > > > > > > > all
> > > > > > > >> > > > > > > > > > > >> when
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> require
> > > > > explicit
> > > > > > > >> deletion
> > > > > > > >> > > > > > operations
> > > > > > > >> > > > > > > > > by
> > > > > > > >> > > > > > > > > > > >> > themselves.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9.
> > Will
> > > > the
> > > > > > > >> > > > > > Consumer.subscribe
> > > > > > > >> > > > > > > > > method for
> > > > > > > >> > > > > > > > > > > >> > the Java
> > > > > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > `java.util.regex.Pattern` of
> > > > > > > >> > > > > > do we
> > > > > > > >> > > > > > > > > have
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I
> think
> > we
> > > > do
> > > > > > not
> > > > > > > >> need to
> > > > > > > >> > > > > > introduce
> > > > > > > >> > > > > > > > > an
> > > > > > > >> > > > > > > > > > > >> > overload, but
> > > > > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be
> > some
> > > > > > > >> compatibility
> > > > > > > >> > > > > > issues
> > > > > > > >> > > > > > > > > that we
> > > > > > > >> > > > > > > > > > > >> may
> > > > > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10.
> I
> > > > agree
> > > > > > with
> > > > > > > >> Justine
> > > > > > > >> > > > > > that we
> > > > > > > >> > > > > > > > > should
> > > > > > > >> > > > > > > > > > > >> be
> > > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > switch to
> > > > > > > >> > > > > > IBP/metadata.version
> > > > > > > >> > > > > > > > > from the
> > > > > > > >> > > > > > > > > > > >> > feature
> > > > > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can
> > switch
> > > > > the
> > > > > > > >> default
> > > > > > > >> > > > for
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > feature
> > > > > > > >> > > > > > > > > > > >> flag
> > > > > > > >> > > > > > > > > > > >> > to
> > > > > > > >> > > > > > > > > > > >> > > > > true
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > metadata.version
> > > > > > > >> once we
> > > > > > > >> > > > > > want to
> > > > > > > >> > > > > > > > > make it
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11.
> > Some of
> > > > > the
> > > > > > > >> protocol
> > > > > > > >> > > > APIs
> > > > > > > >> > > > > > don't
> > > > > > > >> > > > > > > > > mention
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> required
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good
> > to
> > > > add
> > > > > > that
> > > > > > > >> for
> > > > > > > >> > > > > > consistency.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack,
> we
> > can
> > > > > > > >> certainly do
> > > > > > > >> > > > that.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12.
> > It is
> > > > a
> > > > > > bit
> > > > > > > >> odd that
> > > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group"
> even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > though it
> > > > > > seems
> > > > > > > >> to do
> > > > > > > >> > > > more
> > > > > > > >> > > > > > than
> > > > > > > >> > > > > > > > > reading.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had
> > that
> > > > > > thought
> > > > > > > >> myself
> > > > > > > >> > > > as
> > > > > > > >> > > > > > well,
> > > > > > > >> > > > > > > > > but in
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > end we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > > > > alternative:
> > > > > > > >> adding
> > > > > > > >> > > > > > Write
> > > > > > > >> > > > > > > > > Group
> > > > > > > >> > > > > > > > > > > >> seems
> > > > > > > >> > > > > > > > > > > >> > an
> > > > > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not
> > have it
> > > > > > > >> elsewhere (we
> > > > > > > >> > > > only
> > > > > > > >> > > > > > have
> > > > > > > >> > > > > > > > > Read /
> > > > > > > >> > > > > > > > > > > >> > Delete
> > > > > > > >> > > > > > > > > > > >> > > > > and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far).
> > Would
> > > > > > like to
> > > > > > > >> hear
> > > > > > > >> > > > others
> > > > > > > >> > > > > > > > > thoughts.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13.
> > How is
> > > > > > topic
> > > > > > > >> > > > recreation
> > > > > > > >> > > > > > > > > handled by
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > consumer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > protocol?
> > > > It
> > > > > > > >> would be
> > > > > > > >> > > > good
> > > > > > > >> > > > > > to have
> > > > > > > >> > > > > > > > > a
> > > > > > > >> > > > > > > > > > > >> > section on
> > > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You
> mean
> > > > with
> > > > > > regex
> > > > > > > >> > > > > > subscription
> > > > > > > >> > > > > > > > > right? Yes
> > > > > > > >> > > > > > > > > > > >> > we can
> > > > > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that,
> > but
> > > > > > basically
> > > > > > > >> the
> > > > > > > >> > > > idea
> > > > > > > >> > > > > > is that
> > > > > > > >> > > > > > > > > > > >> consumer
> > > > > > > >> > > > > > > > > > > >> > would
> > > > > > > >> > > > > > > > > > > >> > > > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic
> in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the
> new
> > > > > > protocol as
> > > > > > > >> it's
> > > > > > > >> > > > > > handled all
> > > > > > > >> > > > > > > > > by the
> > > > > > > >> > > > > > > > > > > >> > brokers.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14.
> > The
> > > > KIP
> > > > > > > >> mentions we
> > > > > > > >> > > > will
> > > > > > > >> > > > > > write
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> new
> > > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this
> > is an
> > > > > > > >> implementation
> > > > > > > >> > > > > > detail,
> > > > > > > >> > > > > > > > > do we
> > > > > > > >> > > > > > > > > > > >> > plan to
> > > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for
> > it?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We
> have
> > not
> > > > > > thought
> > > > > > > >> about
> > > > > > > >> > > > > > that. But
> > > > > > > >> > > > > > > > > I think
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > answer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15.
> > Do we
> > > > > > have a
> > > > > > > >> > > > scalability
> > > > > > > >> > > > > > goal
> > > > > > > >> > > > > > > > > when it
> > > > > > > >> > > > > > > > > > > >> > comes to
> > > > > > > >> > > > > > > > > > > >> > > > > >> how
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> group
> > > > > > protocol can
> > > > > > > >> > > > support?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> Within a
> > > > > group,
> > > > > > I
> > > > > > > >> think we
> > > > > > > >> > > > > > should
> > > > > > > >> > > > > > > > > shoot for
> > > > > > > >> > > > > > > > > > > >> > 1000s of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > scalability
> > > > > > goals
> > > > > > > >> also
> > > > > > > >> > > > depend
> > > > > > > >> > > > > > on the
> > > > > > > >> > > > > > > > > offset
> > > > > > > >> > > > > > > > > > > >> > > > > management
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > capabilities
> > > > > of
> > > > > > the
> > > > > > > >> > > > coordinator
> > > > > > > >> > > > > > > > > which we
> > > > > > > >> > > > > > > > > > > >> did
> > > > > > > >> > > > > > > > > > > >> > not
> > > > > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard
> to
> > > > give a
> > > > > > > >> number that
> > > > > > > >> > > > > > applies
> > > > > > > >> > > > > > > > > > > >> > universally.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16.
> > Did we
> > > > > > > >> consider
> > > > > > > >> > > > having
> > > > > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> SubscribedTopicNames in
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > ConsumerGroupHeartbeatRequest?
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> since
> > we
> > > > > have
> > > > > > to
> > > > > > > >> resolve
> > > > > > > >> > > > the
> > > > > > > >> > > > > > regex
> > > > > > > >> > > > > > > > > on the
> > > > > > > >> > > > > > > > > > > >> > server,
> > > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > topic
> > > > > > name?
> > > > > > > >> The
> > > > > > > >> > > > > > difference is
> > > > > > > >> > > > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > sending the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > whereas
> > > > > > sending
> > > > > > > >> the topic
> > > > > > > >> > > > > > names is
> > > > > > > >> > > > > > > > > less
> > > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > recreation
> > > > > is
> > > > > > > >> easier to
> > > > > > > >> > > > > > handle if
> > > > > > > >> > > > > > > > > we have
> > > > > > > >> > > > > > > > > > > >> > topic
> > > > > > > >> > > > > > > > > > > >> > > > > ids.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The
> main
> > > > > reason
> > > > > > to
> > > > > > > >> still
> > > > > > > >> > > > let
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > clients
> > > > > > > >> > > > > > > > > > > >> send
> > > > > > > >> > > > > > > > > > > >> > names
> > > > > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > reasoning of
> > > > > > names
> > > > > > > >> -> ids
> > > > > > > >> > > > on
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > broker /
> > > > > > > >> > > > > > > > > > > >> > admin
> > > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > although we
> > > > > > added
> > > > > > > >> topic id
> > > > > > > >> > > > in
> > > > > > > >> > > > > > > > > KIP-516, we
> > > > > > > >> > > > > > > > > > > >> > never
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > consumer/producers
> > > > > > > >> > > > leveraging
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > related
> > > > > > > >> > > > > > > > > > > >> > newer
> > > > > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just
> > set the
> > > > > > topic
> > > > > > > >> id as
> > > > > > > >> > > > empty
> > > > > > > >> > > > > > UUID.
> > > > > > > >> > > > > > > > > We
> > > > > > > >> > > > > > > > > > > >> want
> > > > > > > >> > > > > > > > > > > >> > to keep
> > > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > consumer/producer
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be
> > thin
> > > > and
> > > > > > only
> > > > > > > >> > > > delegate
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > reasoning
> > > > > > > >> > > > > > > > > > > >> on
> > > > > > > >> > > > > > > > > > > >> > broker
> > > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> clients.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > Thanks,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> Ismael
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On
> > Wed,
> > > > Jul
> > > > > 6,
> > > > > > > >> 2022 at
> > > > > > > >> > > > 10:45
> > > > > > > >> > > > > > AM
> > > > > > > >> > > > > > > > > David
> > > > > > > >> > > > > > > > > > > >> Jacot
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi
> > all,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I
> > would
> > > > > > like to
> > > > > > > >> start a
> > > > > > > >> > > > > > > > > discussion
> > > > > > > >> > > > > > > > > > > >> thread
> > > > > > > >> > > > > > > > > > > >> > on
> > > > > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > Generation
> > > > > > of
> > > > > > > >> the
> > > > > > > >> > > > Consumer
> > > > > > > >> > > > > > > > > Rebalance
> > > > > > > >> > > > > > > > > > > >> > Protocol.
> > > > > > > >> > > > > > > > > > > >> > > > > >> With
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to
> > make
> > > > > the
> > > > > > > >> rebalance
> > > > > > > >> > > > > > protocol
> > > > > > > >> > > > > > > > > (for
> > > > > > > >> > > > > > > > > > > >> > consumers)
> > > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > scalable,
> > > > > > > >> easier to
> > > > > > > >> > > > > > implement for
> > > > > > > >> > > > > > > > > > > >> > clients, and
> > > > > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > operators.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> The
> > KIP
> > > > is
> > > > > > here:
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > Please
> > > > > take
> > > > > > a
> > > > > > > >> look and
> > > > > > > >> > > > let
> > > > > > > >> > > > > > me
> > > > > > > >> > > > > > > > > know what
> > > > > > > >> > > > > > > > > > > >> > you
> > > > > > > >> > > > > > > > > > > >> > > > > think.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > Best,
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > David
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> PS:
> > I
> > > > will
> > > > > > be
> > > > > > > >> away from
> > > > > > > >> > > > > > July
> > > > > > > >> > > > > > > > > 18th to
> > > > > > > >> > > > > > > > > > > >> > August 8th.
> > > > > > > >> > > > > > > > > > > >> > > > > >> That
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of
> > time
> > > > to
> > > > > > read
> > > > > > > >> and
> > > > > > > >> > > > digest
> > > > > > > >> > > > > > this
> > > > > > > >> > > > > > > > > long
> > > > > > > >> > > > > > > > > > > >> KIP.
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> --
> > > > > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > --
> > > > > > > >> > > > > -- Guozhang
> > > > > > > >> > > >
> > > > > > > >> > >
> > > > > > > >> > >
> > > > > > > >> > > --
> > > > > > > >> > > -- Guozhang
> > > > > > > >>
> > > > > > > >
> > > > > > > >
> > > > > > > > --
> > > > > > > > -- Guozhang
> > > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > --
> > > > > > > -- Guozhang
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > > -- Guozhang
> > > > >
> > > >
> >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Jason Gustafson <ja...@confluent.io.INVALID>]

Hi David,

Thanks for the response. Replied below:

> I don't have a strong opinion on this one. However, it seems that this
could be a pandora box if our subset turns out to not be sufficient
for some existing users. We will have to add more to it and the
compatibility story between versions could be a real nightmare. One
thing that we could do for sure is disabling all unicode groups in the
library. That seems possible and would already reduce the scope.

Changes to the regex syntax seems not a major problem to me. We would bump
the protocol version as we usually do. Actually that begs the same question
about RE2. How stable is the syntax and does it differ depending on the
version of the library? In general, I do agree that we'd be better off not
having to manage the syntax ourselves if we can convince ourselves that
compatibility won't be a problem and the client won't eventually need to
have access to it.

Thanks,
Jason

On Tue, Sep 27, 2022 at 6:21 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Luke,
>
> Thanks for your comments.
>
> Your suggestion makes sense. Jun's made the same one. We are also
> discussing whether we should directly return the group state in the
> response and drop the prepare assignment api. It seems that the
> current consensus is to keep it though. If that stays like this, I
> will apply your suggestion.
>
> Thanks,
> David
>
> On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com> wrote:
> >
> > Hi David,
> >
> > Thanks for your patience for answering all these questions!
> >
> > One more question from me:
> > 1. When the selected consumer is going to be in client-side assigning, we
> > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering if we
> > could use a flag to indicate the assignment is required. The rationale is
> > when returning error, we lose a chance to catch up with the latest target
> > assignment or revoke partitions. Since we expected the record fetching is
> > not blocked by the rebalance/assignment computing, we might want to allow
> > the members to update the member assignment in each heartbeat.
> >
> > Luke
> >
> > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> <ja...@confluent.io.invalid>
> > wrote:
> >
> > > Hey David,
> > >
> > > Thanks for the response.
> > >
> > > > Do you think that we could have a case where two members with the
> same
> > > static id are active?
> > >
> > > I was just thinking about the handoff when we spin up a new node
> before the
> > > old one completes shutdown. It's probably ok to fence the old member
> > > abruptly. I do recall we had a few issues with this handoff in the
> past. It
> > > was due to the fact that the replaced node might have been the leader.
> > > Since we've removed the leader notion, perhaps this kind of issue won't
> > > resurface.
> > >
> > > In regard to the RE2 library. I understand the intent to pass this
> through
> > > as a string without any validation so that we don't need the
> dependence on
> > > the client. I do wonder if that will be restrictive in practice. For
> > > example, if a regex changes on the client side, would we need to
> validate
> > > against the current assignment to see which partitions to continue
> > > fetching? Or would we instead invalidate the complete current
> assignment?
> > > Perhaps this is a rare case and we can do the latter. There could be
> more
> > > situations though.
> > >
> > > As an alternative, I wonder if we could define a minimalistic regex
> syntax
> > > (a subset of RE2 say) that could map more easily to regex support that
> > > already exists in clients? For example, in this scenario, we are
> matching
> > > topics, so we only need to be able to represent topic characters and
> not
> > > the whole smorgasbord of whitespace and unicode options. Would it make
> > > sense to have a reduced syntax which includes only standard topic
> > > characters, wildcards, character classes, and repetitions?
> > >
> > > Best,
> > > Jason
> > >
> > >
> > >
> > >
> > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com>
> wrote:
> > >
> > > > Hi David, thanks for all the detailed explanations. I think they all
> make
> > > > sense. Just want to have a couple follow-ups here:
> > > >
> > > > > I don't really see the benefits here because server side assignors
> > > > don't have metadata at all. They only assign topic-partitions. They
> > > > are not supposed to generate metadata nor to receive metadata from
> the
> > > > members.
> > > >
> > > > Just to clarify I was asking about the `version` of the assignor
> (i.e. up
> > > > to what version that the client would support), and I do agree we
> would
> > > not
> > > > need metadata. What I have in mind is that, for some specific
> built-in
> > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> in a
> > > > newer version we would have a hierarchical rack ID string format,
> like
> > > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > > would still be in old format. In this case, the broker then needs to
> > > choose
> > > > the old versioned assignor. I'm probably making something up here for
> > > rack
> > > > aware assignors, but I'm wondering if in general such an
> "auto-downgrade"
> > > > behavior would be needed still for broker-side assignor, and if yes
> would
> > > > "version" still be useful.
> > > >
> > > > > Yeah, that's right. Within a rebalance, `onAssignment` is called
> once
> > > > when the member transitions to a new epoch. This one contains the
> full
> > > > metadata provided by the client side assignor. Then, `onAssignment`
> > > > can be called max N times where N is the number of partitions pending
> > > > revocation by other members. Let me try to clarify this in the KIP.
> > > >
> > > > Okay, my understanding is that the calling ordering of these
> callbacks
> > > > would be like the following:
> > > >
> > > > ----------------------------------------
> > > > onPartitionsRevoked();   // just once, since we do not really need
> > > > to revoke incrementally.
> > > >
> > > > onAssignment();    // the first call, with epoch incremented
> > > > onPartitionsAssigned();   // paired with the onAssignment
> > > >
> > > > onAssignment();              // the first onAssignment would bump up
> the
> > > > epoch, and the metadata reflected.
> > > > onPartitionsAssigned();   // each time we get an additional
> assignment,
> > > we
> > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > ...
> > > > onAssignment();
> > > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > > encoded
> > > > metadata would not change, only the incrementally added partitions be
> > > > reflected
> > > >
> > > > Is that the case?
> > > >
> > > > I'm wondering if we would still call onAssignment just once, that
> encodes
> > > > all the assignment for this rebalance, including all the partitions
> that
> > > > should be assigned to the member but not yet assigned since they
> have not
> > > > been revoked by others. In that case the call ordering would be:
> > > >
> > > > ----------------------------------------
> > > > onPartitionsRevoked();   // just once
> > > > onAssignment();    // just once, with epoch incremented, and metadata
> > > > encoded changed, the "assignment" field also reflect the final target
> > > > assignment
> > > > onPartitionsAssigned();   // multiple times, which represent
> > > incrementally
> > > > added partitions
> > > > ...
> > > > onPartitionsAssigned();
> > > >
> > > > The motivation from this is that, most users would only implement the
> > > > rebalance callback listeners and hence we'd definitely need to make
> sure
> > > > the semantics of that does not change much, and the time
> > > > `onPartitionsAssigned` indicate the time when the partitions are
> actually
> > > > assigned to it; while for assignors, the `onAssignment` is used to
> > > indicate
> > > > what decision is made regarding for this member, i.e. when the
> partitions
> > > > are decided to be given to it, but not necessarily meant that it has
> been
> > > > given, since that time should be determined by the time of
> > > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > > implementers would not need to reason which `onAssignment` would be
> the
> > > > last one for this epoch.
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > > Guozhang
> > > >
> > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> <djacot@confluent.io.invalid
> > > >
> > > > wrote:
> > > >
> > > > > Hi Guozhang,
> > > > >
> > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> byte
> > > > > field,
> > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> And
> > > the
> > > > > > rationale seems to be assuming that we should always be able to
> do
> > > the
> > > > > > assignment at the broker-side assignor without errors.
> Personally I
> > > > think
> > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > broker-side assignors, e.g. for some edge cases where some
> subscribed
> > > > > > topics are not recognized with the current broker's metadata.
> What do
> > > > you
> > > > > > think?
> > > > >
> > > > > Yeah, that seems reasonable. However, I wonder if it would be
> better
> > > > > to use an exception on the server side. This is what we usually do
> for
> > > > > server side plugins. On the client side, we use a field because the
> > > > > errors are not defined in advance.
> > > > >
> > > > > Your comment also makes me think about what we should do when the
> > > > > server side assignor fails. I suppose that we have to keep the
> current
> > > > > assignment until a new event occurs. For instance, in your example,
> > > > > the coordinator would have to trigger a rebalance when unrecognized
> > > > > topics are available. This would be part of the metadata
> monitoring.
> > > > >
> > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > reason/version/metadata compared with the broker-side
> GroupMember. I
> > > > > agree
> > > > > > that broker-side assignor would not need reason/metadata since
> they
> > > are
> > > > > > blackbox strings/bytes to the assignor, but what about version?
> E.g.
> > > is
> > > > > it
> > > > > > possible that we evolve our broker-side built-in assignor but
> the old
> > > > > > versioned clients would not be able to work with the new
> version, in
> > > > > which
> > > > > > case we need to let the broker being aware of this and upgrade
> its
> > > > > behavior
> > > > > > to cooperate with the clients?
> > > > >
> > > > > I don't really see the benefits here because server side assignors
> > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > are not supposed to generate metadata nor to receive metadata from
> the
> > > > > members.
> > > > >
> > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > instead
> > > > > of
> > > > > > including these three fields, what about just adding the
> "Metadata"
> > > > field
> > > > > > class which has these three fields? Also, there are two
> "Metadata"
> > > > > > currently in the APIs, the first is a class that encodes
> > > > > > reason/version/metadata, and the second is just the encoded
> metadata
> > > > > bytes.
> > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > which
> > > > > > then has a bytebuffer field Metadata, or instead naming the
> second
> > > > > > bytebuffer field as metadataBytes?
> > > > >
> > > > > That's a good point. Let me try to rationalize this interface
> based on
> > > > > your suggestions.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > > > wrote:
> > > > > >
> > > > > > Hello David,
> > > > > >
> > > > > > Just had a few more nit questions about the public APIs:
> > > > > >
> > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> byte
> > > > > field,
> > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> And
> > > the
> > > > > > rationale seems to be assuming that we should always be able to
> do
> > > the
> > > > > > assignment at the broker-side assignor without errors.
> Personally I
> > > > think
> > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > broker-side assignors, e.g. for some edge cases where some
> subscribed
> > > > > > topics are not recognized with the current broker's metadata.
> What do
> > > > you
> > > > > > think?
> > > > > >
> > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > reason/version/metadata compared with the broker-side
> GroupMember. I
> > > > > agree
> > > > > > that broker-side assignor would not need reason/metadata since
> they
> > > are
> > > > > > blackbox strings/bytes to the assignor, but what about version?
> E.g.
> > > is
> > > > > it
> > > > > > possible that we evolve our broker-side built-in assignor but
> the old
> > > > > > versioned clients would not be able to work with the new
> version, in
> > > > > which
> > > > > > case we need to let the broker being aware of this and upgrade
> its
> > > > > behavior
> > > > > > to cooperate with the clients?
> > > > > >
> > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > instead
> > > > > of
> > > > > > including these three fields, what about just adding the
> "Metadata"
> > > > field
> > > > > > class which has these three fields? Also, there are two
> "Metadata"
> > > > > > currently in the APIs, the first is a class that encodes
> > > > > > reason/version/metadata, and the second is just the encoded
> metadata
> > > > > bytes.
> > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > which
> > > > > > then has a bytebuffer field Metadata, or instead naming the
> second
> > > > > > bytebuffer field as metadataBytes?
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> wangguoz@gmail.com>
> > > > > wrote:
> > > > > >
> > > > > > > Hello David,
> > > > > > >
> > > > > > > Thanks for bringing this question up. I think the main
> benefits as
> > > > you
> > > > > > > listed is 2) above if it stays; just to clarify, we would only
> be
> > > > able
> > > > > to
> > > > > > > save one round trip if the rebalance is still triggered by the
> > > > broker;
> > > > > if
> > > > > > > the rebalance is triggered by the client then the
> num.round.trips
> > > are
> > > > > the
> > > > > > > same:
> > > > > > >
> > > > > > > 1) With GroupPrepareAssignment:
> > > > > > >
> > > > > > > T0: client decides to do a new assignment, suppose it has
> already
> > > > sent
> > > > > a
> > > > > > > HB and hence has to wait for it to return first since only one
> > > > request
> > > > > /
> > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > GroupPrepareAssignment request.
> > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > T3: it calculates the new assignment, and sends a
> > > > > GroupInstallAssignment
> > > > > > > request.
> > > > > > >
> > > > > > > In total, two round trips.
> > > > > > >
> > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > >
> > > > > > > T0: client decides to do a new assignment, suppose it has
> already
> > > > sent
> > > > > a
> > > > > > > HB and hence has to wait for it to return first since only one
> > > > request
> > > > > /
> > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > T1: client receives the HB response, and then sends the new HB
> > > > request
> > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > T2: the HB response with the optional member metadata map is
> > > > returned.
> > > > > > > T3: it calculates the new assignment, and sends a
> > > > > GroupInstallAssignment
> > > > > > > request.
> > > > > > >
> > > > > > > In total, two round trips as well.
> > > > > > >
> > > > > > > -----------------------------
> > > > > > >
> > > > > > > So to complete the full picture here, we'd need to modify both
> HB
> > > > > request
> > > > > > > and response so that the client can also indicate a new
> rebalance
> > > via
> > > > > the
> > > > > > > HB request as well, right?
> > > > > > >
> > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > additional
> > > > > fields
> > > > > > > encoding the full member (subscription) metadata and topic
> metadata
> > > > as
> > > > > > > optional fields.
> > > > > > >
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > >> Hi all,
> > > > > > >>
> > > > > > >> During an offline conversation, someone asked why we need the
> > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > > > instead
> > > > > > >> provide the group state in the heartbeat response. This has a
> few
> > > > > > >> advantages: 1) it does not require using a special error code
> to
> > > > > > >> signal that a new assignment is required as the signal would
> be
> > > the
> > > > > > >> provided group state; 2) it removes one round trip when a
> client
> > > > side
> > > > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > > > >> response's definition quite large. I recall that I went with
> the
> > > > > > >> current approach due to this.
> > > > > > >>
> > > > > > >> Providing the group state in the heartbeat response is
> appealing.
> > > > What
> > > > > > >> do you guys think?
> > > > > > >>
> > > > > > >> Best,
> > > > > > >> David
> > > > > > >>
> > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> djacot@confluent.io>
> > > > > wrote:
> > > > > > >> >
> > > > > > >> > Hi Guozhang,
> > > > > > >> >
> > > > > > >> > 1. I have added a reference to the relevant chapter instead
> of
> > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > >> >
> > > > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> > > about
> > > > > when
> > > > > > >> > a rebalance should be triggered for the non-upgraded members
> > > using
> > > > > the
> > > > > > >> > old protocol. The section mentions that a rebalance must be
> > > > > triggered
> > > > > > >> > when a new assignment is installed. This implies that the
> group
> > > > > epoch
> > > > > > >> > was updated either by a native member or a non-upgraded
> member.
> > > > For
> > > > > > >> > the latter, the JoinGroup request would be the trigger. I
> have
> > > > > added a
> > > > > > >> > reference to the relevant chapter in the "JoinGroup
> Handling"
> > > > > section
> > > > > > >> > as well. Does that make sense?
> > > > > > >> >
> > > > > > >> > Thanks,
> > > > > > >> > David
> > > > > > >> >
> > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > wangguoz@gmail.com
> > > > >
> > > > > > >> wrote:
> > > > > > >> > >
> > > > > > >> > > Hello David,
> > > > > > >> > >
> > > > > > >> > > Alright I think that's sufficient. Just to make that
> clear in
> > > > the
> > > > > doc,
> > > > > > >> > > could we update:
> > > > > > >> > >
> > > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > > coordinator
> > > > > > >> will
> > > > > > >> > > trigger rebalance based on the HB's member metadata /
> reason?
> > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > described
> > > > > in
> > > > > > >> "Group
> > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> > > Guozhang
> > > > > > >> > >
> > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > wrote:
> > > > > > >> > >
> > > > > > >> > > > Hi Guozhang,
> > > > > > >> > > >
> > > > > > >> > > > I thought that the assignor will always be consulted
> when
> > > the
> > > > > next
> > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > > heartbeat.
> > > > > > >> This
> > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > rebalance
> > > > by
> > > > > > >> > > > setting the reason to a non-zero value or by changing
> the
> > > > > bytes. Do
> > > > > > >> > > > you think that this is not sufficient? Are you
> concerned by
> > > > the
> > > > > > >> delay?
> > > > > > >> > > >
> > > > > > >> > > > Best,
> > > > > > >> > > > David
> > > > > > >> > > >
> > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > >> wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > Hello David,
> > > > > > >> > > > >
> > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > >> > > > >
> > > > > > >> > > > > ```
> > > > > > >> > > > > In this case, a new assignment is triggered by the
> client
> > > > side
> > > > > > >> > > > > assignor. When constructing the HB, the consumer will
> > > always
> > > > > > >> consult
> > > > > > >> > > > > the client side assignor and propagate the
> information to
> > > > the
> > > > > > >> group
> > > > > > >> > > > > coordinator. In other words, we don't expect users to
> call
> > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > >> > > > > ```
> > > > > > >> > > > >
> > > > > > >> > > > > As I looked at the current PartitionAssignor's
> interface,
> > > we
> > > > > > >> actually do
> > > > > > >> > > > > not have a way yet to instruct how to construct the
> next
> > > HB
> > > > > > >> request, e.g.
> > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> with a
> > > > new
> > > > > > >> assignment,
> > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > PartitionAssignor
> > > > > to
> > > > > > >> indicate
> > > > > > >> > > > > the next HB telling broker about so. WDYT about adding
> > > such
> > > > an
> > > > > > >> API on the
> > > > > > >> > > > > PartitionAssignor?
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > Guozhang
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > > > wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > > Hi Jun,
> > > > > > >> > > > > >
> > > > > > >> > > > > > I have updated the KIP to include your feedback. I
> have
> > > > also
> > > > > > >> tried to
> > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > >> > > > > >
> > > > > > >> > > > > > Best,
> > > > > > >> > > > > > David
> > > > > > >> > > > > >
> > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > djacot@confluent.io
> > > > > > >> >
> > > > > > >> > > > wrote:
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Hi Jun,
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> answering
> > > your
> > > > > > >> questions
> > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> of
> > > the
> > > > > KIP
> > > > > > >> seem to
> > > > > > >> > > > be
> > > > > > >> > > > > > fewer
> > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> support of
> > > > > > >> wildcard. A few
> > > > > > >> > > > > > > > comments below.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > I would also add that the KIP removes the global
> sync
> > > > > barrier
> > > > > > >> in the
> > > > > > >> > > > > > > protocol which is essential to improve group
> stability
> > > > and
> > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> client by
> > > > > moving
> > > > > > >> most of
> > > > > > >> > > > > > > the logic to the server side.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> to
> > > > > support
> > > > > > >> rolling
> > > > > > >> > > > > > changing
> > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > assignor
> > > > > used
> > > > > > >> by a
> > > > > > >> > > > > > > majority of the members. This allows the group to
> move
> > > > > from
> > > > > > >> one
> > > > > > >> > > > > > > assignor to another by a roll. This is explained
> in
> > > the
> > > > > > >> Assignor
> > > > > > >> > > > > > > Selection chapter.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> it's
> > > > > > >> required in
> > > > > > >> > > > every
> > > > > > >> > > > > > > > request or the scenarios when it needs to be
> filled?
> > > > For
> > > > > > >> example,
> > > > > > >> > > > it's
> > > > > > >> > > > > > not
> > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> filled.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > The client is expected to set those fields in
> case of
> > > a
> > > > > > >> connection
> > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> changed
> > > > since
> > > > > > >> the last
> > > > > > >> > > > > > > HB. The server populates those fields as long as
> the
> > > > > member
> > > > > > >> is not
> > > > > > >> > > > > > > fully reconciled - the member should acknowledge
> that
> > > it
> > > > > has
> > > > > > >> the
> > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> this in
> > > > the
> > > > > KIP.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > affinity
> > > > > > >> between the
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > and the broker is only considered during
> fetching,
> > > but
> > > > > not
> > > > > > >> during
> > > > > > >> > > > > > assigning
> > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > assignment
> > > > > is
> > > > > > >> made,
> > > > > > >> > > > there
> > > > > > >> > > > > > is
> > > > > > >> > > > > > > > no opportunity for read affinity because no
> replicas
> > > > of
> > > > > > >> assigned
> > > > > > >> > > > > > partitions
> > > > > > >> > > > > > > > are close to the member. I am wondering if we
> should
> > > > use
> > > > > > >> this
> > > > > > >> > > > > > opportunity
> > > > > > >> > > > > > > > to address this by including rack in
> GroupMember.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's an interesting idea. I don't see any issue
> with
> > > > > adding
> > > > > > >> the
> > > > > > >> > > > rack
> > > > > > >> > > > > > > to the members. I will do so.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> know
> > > how
> > > > > busy
> > > > > > >> a group
> > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> it
> > > > seems
> > > > > > >> that we
> > > > > > >> > > > could
> > > > > > >> > > > > > add
> > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> the
> > > > event
> > > > > > >> loop is
> > > > > > >> > > > doing
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > actual work.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> failover
> > > > > > >> handling? For
> > > > > > >> > > > > > example,
> > > > > > >> > > > > > > > does it need to trigger the check if any group
> with
> > > > the
> > > > > > >> wildcard
> > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sure. When the new group coordinator takes over,
> it
> > > has
> > > > > to:
> > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > >> > > > > > > * Trigger a new assignment if a client side
> assignor
> > > is
> > > > > used.
> > > > > > >> We
> > > > > > >> > > > don't
> > > > > > >> > > > > > > store the information about the member selected
> to run
> > > > the
> > > > > > >> assignment
> > > > > > >> > > > > > > so we have to start a new one.
> > > > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > > > >> subscriptions, and
> > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> document
> > > > what
> > > > > > >> the epoch
> > > > > > >> > > > > > field
> > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > >> ConsumerGroupMetadataValue
> > > > > > >> > > > > > reflect
> > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> latest
> > > > > group
> > > > > > >> epoch.
> > > > > > >> > > > > > > When the group state is updated, the group epoch
> is
> > > > > bumped so
> > > > > > >> we use
> > > > > > >> > > > > > > that one for all the change records related to the
> > > > update.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > following
> > > > > > >> > > > invariants
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> a
> > > > member
> > > > > > >> not to
> > > > > > >> > > > get any
> > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's right. Here I meant that the members
> provided
> > > by
> > > > > the
> > > > > > >> assignor
> > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > assignor
> > > > > can
> > > > > > >> not make
> > > > > > >> > > > > > > up new member ids.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> epoch
> > > > > equals to
> > > > > > >> 0":
> > > > > > >> > > > When
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> used?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > A member is expected to abandon all its
> partitions and
> > > > > > >> rejoins when
> > > > > > >> > > > it
> > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this
> case,
> > > > the
> > > > > > >> group
> > > > > > >> > > > > > > coordinator will have removed the member from the
> > > group.
> > > > > The
> > > > > > >> member
> > > > > > >> > > > > > > can rejoin the group with the same member id but
> with
> > > 0
> > > > as
> > > > > > >> epoch. Let
> > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> to
> > > > > trigger a
> > > > > > >> > > > > > reassignment
> > > > > > >> > > > > > > > by either providing a non-zero reason or by
> updating
> > > > the
> > > > > > >> assignor
> > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> with
> > > the
> > > > > > >> deprecation
> > > > > > >> > > > of
> > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > In this case, a new assignment is triggered by the
> > > > client
> > > > > side
> > > > > > >> > > > > > > assignor. When constructing the HB, the consumer
> will
> > > > > always
> > > > > > >> consult
> > > > > > >> > > > > > > the client side assignor and propagate the
> information
> > > > to
> > > > > the
> > > > > > >> group
> > > > > > >> > > > > > > coordinator. In other words, we don't expect
> users to
> > > > call
> > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > section
> > > > > > >> seems to
> > > > > > >> > > > have
> > > > > > >> > > > > > > > multiple typos.
> > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > immediately
> > > > > > >> gets into
> > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > incorrect.
> > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> seems
> > > to
> > > > > get
> > > > > > >> into
> > > > > > >> > > > > > epoch=3,
> > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> has A -
> > > > > > >> epoch=2,
> > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> the
> > > > > upgrade
> > > > > > >> from
> > > > > > >> > > > any
> > > > > > >> > > > > > old
> > > > > > >> > > > > > > > version to new one?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > We will support upgrading from the consumer
> protocol
> > > > > version
> > > > > > >> 3,
> > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented
> yet
> > > so
> > > > > the
> > > > > > >> earliest
> > > > > > >> > > > > > > version is unknown at the moment. This is
> explained in
> > > > the
> > > > > > >> migration
> > > > > > >> > > > > > > plan chapter.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> update the
> > > > KIP
> > > > > > >> based on
> > > > > > >> > > > it
> > > > > > >> > > > > > > next week.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Best,
> > > > > > >> > > > > > > David
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > >> <ju...@confluent.io.invalid>
> > > > > > >> > > > wrote:
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Hi, David,
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> of
> > > the
> > > > > KIP
> > > > > > >> seem to
> > > > > > >> > > > be
> > > > > > >> > > > > > fewer
> > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> support of
> > > > > > >> wildcard. A few
> > > > > > >> > > > > > > > comments below.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> to
> > > > > support
> > > > > > >> rolling
> > > > > > >> > > > > > changing
> > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> it's
> > > > > > >> required in
> > > > > > >> > > > every
> > > > > > >> > > > > > > > request or the scenarios when it needs to be
> filled?
> > > > For
> > > > > > >> example,
> > > > > > >> > > > it's
> > > > > > >> > > > > > not
> > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> filled.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > affinity
> > > > > > >> between the
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > and the broker is only considered during
> fetching,
> > > but
> > > > > not
> > > > > > >> during
> > > > > > >> > > > > > assigning
> > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > assignment
> > > > > is
> > > > > > >> made,
> > > > > > >> > > > there
> > > > > > >> > > > > > is
> > > > > > >> > > > > > > > no opportunity for read affinity because no
> replicas
> > > > of
> > > > > > >> assigned
> > > > > > >> > > > > > partitions
> > > > > > >> > > > > > > > are close to the member. I am wondering if we
> should
> > > > use
> > > > > > >> this
> > > > > > >> > > > > > opportunity
> > > > > > >> > > > > > > > to address this by including rack in
> GroupMember.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> know
> > > how
> > > > > busy
> > > > > > >> a group
> > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> it
> > > > seems
> > > > > > >> that we
> > > > > > >> > > > could
> > > > > > >> > > > > > add
> > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> the
> > > > event
> > > > > > >> loop is
> > > > > > >> > > > doing
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > actual work.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> failover
> > > > > > >> handling? For
> > > > > > >> > > > > > example,
> > > > > > >> > > > > > > > does it need to trigger the check if any group
> with
> > > > the
> > > > > > >> wildcard
> > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> document
> > > > what
> > > > > > >> the epoch
> > > > > > >> > > > > > field
> > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > >> ConsumerGroupMetadataValue
> > > > > > >> > > > > > reflect
> > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > following
> > > > > > >> > > > invariants
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> a
> > > > member
> > > > > > >> not to
> > > > > > >> > > > get any
> > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> epoch
> > > > > equals to
> > > > > > >> 0":
> > > > > > >> > > > When
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> used?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> to
> > > > > trigger a
> > > > > > >> > > > > > reassignment
> > > > > > >> > > > > > > > by either providing a non-zero reason or by
> updating
> > > > the
> > > > > > >> assignor
> > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> with
> > > the
> > > > > > >> deprecation
> > > > > > >> > > > of
> > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > section
> > > > > > >> seems to
> > > > > > >> > > > have
> > > > > > >> > > > > > > > multiple typos.
> > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > immediately
> > > > > > >> gets into
> > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > incorrect.
> > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> seems
> > > to
> > > > > get
> > > > > > >> into
> > > > > > >> > > > > > epoch=3,
> > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> has A -
> > > > > > >> epoch=2,
> > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> the
> > > > > upgrade
> > > > > > >> from
> > > > > > >> > > > any
> > > > > > >> > > > > > old
> > > > > > >> > > > > > > > version to new one?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Thanks,
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Jun
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > wrote:
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > Hi all,
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > > coordinator
> > > > > > >> in
> > > > > > >> > > > Java. I
> > > > > > >> > > > > > > > > discussed this offline with a few folks and
> folks
> > > > are
> > > > > > >> concerned
> > > > > > >> > > > that
> > > > > > >> > > > > > > > > we could introduce many regressions in the old
> > > > > protocol
> > > > > > >> if we do
> > > > > > >> > > > so.
> > > > > > >> > > > > > > > > Therefore, I am going to remove this statement
> > > from
> > > > > the
> > > > > > >> KIP. It
> > > > > > >> > > > is an
> > > > > > >> > > > > > > > > implementation detail after all so it does not
> > > have
> > > > > to be
> > > > > > >> > > > decided at
> > > > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > > > refactor the
> > > > > > >> > > > current
> > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > Cheers,
> > > > > > >> > > > > > > > > David
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > > > >> djacot@confluent.io
> > > > > > >> > > > >
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> "Empty,
> > > > > assigning,
> > > > > > >> > > > > > reconciling,
> > > > > > >> > > > > > > > > stable,
> > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > section,
> > > > > > >> right?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > This sentence does not refer to those group
> > > states
> > > > > but
> > > > > > >> rather
> > > > > > >> > > > to a
> > > > > > >> > > > > > > > > > state machine replication (SMR). This
> refers to
> > > > the
> > > > > > >> entire
> > > > > > >> > > > state of
> > > > > > >> > > > > > > > > > group coordinator which is replicated via
> the
> > > log
> > > > > > >> layer. I will
> > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> is
> > > > > modelled
> > > > > > >> as an
> > > > > > >> > > > event
> > > > > > >> > > > > > > > > loop"?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> the new
> > > > > quorum
> > > > > > >> > > > > > controller. We
> > > > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > > > >> __consumer_offsets
> > > > > > >> > > > > > > > > > partition is assigned to a unique thread
> and all
> > > > the
> > > > > > >> events
> > > > > > >> > > > (e.g.
> > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by
> this
> > > > > > >> thread. This
> > > > > > >> > > > > > simplify
> > > > > > >> > > > > > > > > > concurrency and will enable us to do
> simulation
> > > > > testing
> > > > > > >> for the
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > > coordinator.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > >> *__consumer_offsets*
> > > > > > >> > > > > > > > > partitions?
> > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> owned
> > > > by
> > > > > a
> > > > > > >> group
> > > > > > >> > > > > > coordinator?
> > > > > > >> > > > > > > > > For
> > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > consumer
> > > > > > >> groups, and
> > > > > > >> > > > > > both
> > > > > > >> > > > > > > > > exist in
> > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> state
> > > > > machine
> > > > > > >> for it,
> > > > > > >> > > > or
> > > > > > >> > > > > > 2?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > > think.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.4. I know the
> "*group.coordinator.threads"
> > > *is
> > > > > the
> > > > > > >> number
> > > > > > >> > > > of
> > > > > > >> > > > > > threads
> > > > > > >> > > > > > > > > used
> > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> wondering
> > > if
> > > > > the
> > > > > > >> purpose
> > > > > > >> > > > of
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > threads
> > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > group
> > > > > (or
> > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> why
> > > > > should we
> > > > > > >> need
> > > > > > >> > > > > > > > > multi-threads
> > > > > > >> > > > > > > > > > > here?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> shard
> > > > the
> > > > > > >> > > > processing as
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> does
> > > the
> > > > > > >> default
> > > > > > >> > > > session
> > > > > > >> > > > > > > > > timeout not
> > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > thought
> > > > > the
> > > > > > >> min/max
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> it,
> > > > no?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> 30s The
> > > > > > >> timeout to
> > > > > > >> > > > detect
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > protocol.
> > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> int 45s
> > > > The
> > > > > > >> minimum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> int 60s
> > > > The
> > > > > > >> maximum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> session
> > > > > timeout
> > > > > > >> should
> > > > > > >> > > > be 45s
> > > > > > >> > > > > > > > > > (the current default).
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > [range,
> > > > > > >> uniform],
> > > > > > >> > > > > > which means
> > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> like to
> > > > > know
> > > > > > >> why not
> > > > > > >> > > > > > uniform
> > > > > > >> > > > > > > > > one? I
> > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > assignor
> > > > > > >> (former
> > > > > > >> > > > sticky
> > > > > > >> > > > > > > > > assinor)
> > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > reason
> > > > > we
> > > > > > >> choose
> > > > > > >> > > > range
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > as default?
> > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> uniform
> > > The
> > > > > > >> server side
> > > > > > >> > > > > > assignors.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > The order on the server side has no
> influence
> > > > > because
> > > > > > >> the
> > > > > > >> > > > client
> > > > > > >> > > > > > must
> > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> There
> > > is
> > > > no
> > > > > > >> default
> > > > > > >> > > > in the
> > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > specified
> > > > > by
> > > > > > >> the
> > > > > > >> > > > client,
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > request is rejected. The default client
> value
> > > for
> > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I
> hope
> > > > > that my
> > > > > > >> > > > answers
> > > > > > >> > > > > > help
> > > > > > >> > > > > > > > > > to clarify things. I will update the KIP as
> well
> > > > > based
> > > > > > >> on your
> > > > > > >> > > > > > > > > > feedback.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Cheers,
> > > > > > >> > > > > > > > > > David
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > > > >> showuon@gmail.com>
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Hi David,
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > mentioned:
> > > > > > >> > > > > > > > > > > > The new group coordinator will have a
> state
> > > > > machine
> > > > > > >> per
> > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> each
> > > > state
> > > > > > >> machine is
> > > > > > >> > > > > > modelled
> > > > > > >> > > > > > > > > as an
> > > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > > executed
> > > > > in
> > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> "Empty,
> > > > > assigning,
> > > > > > >> > > > > > reconciling,
> > > > > > >> > > > > > > > > stable,
> > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > section,
> > > > > > >> right?
> > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> is
> > > > > modelled
> > > > > > >> as an
> > > > > > >> > > > event
> > > > > > >> > > > > > > > > loop"?
> > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > >> *__consumer_offsets*
> > > > > > >> > > > > > > > > partitions?
> > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> owned
> > > > by
> > > > > a
> > > > > > >> group
> > > > > > >> > > > > > coordinator?
> > > > > > >> > > > > > > > > For
> > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > consumer
> > > > > > >> groups, and
> > > > > > >> > > > > > both
> > > > > > >> > > > > > > > > exist in
> > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> state
> > > > > machine
> > > > > > >> for it,
> > > > > > >> > > > or
> > > > > > >> > > > > > 2?
> > > > > > >> > > > > > > > > > > 1.4. I know the
> "*group.coordinator.threads"
> > > *is
> > > > > the
> > > > > > >> number
> > > > > > >> > > > of
> > > > > > >> > > > > > threads
> > > > > > >> > > > > > > > > used
> > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> wondering
> > > if
> > > > > the
> > > > > > >> purpose
> > > > > > >> > > > of
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > threads
> > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > group
> > > > > (or
> > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> why
> > > > > should we
> > > > > > >> need
> > > > > > >> > > > > > > > > multi-threads
> > > > > > >> > > > > > > > > > > here?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2. For the default value in the new
> configs:
> > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> does
> > > the
> > > > > > >> default
> > > > > > >> > > > session
> > > > > > >> > > > > > > > > timeout not
> > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > thought
> > > > > the
> > > > > > >> min/max
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> it,
> > > > no?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> 30s The
> > > > > > >> timeout to
> > > > > > >> > > > detect
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > protocol.
> > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> int 45s
> > > > The
> > > > > > >> minimum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> int 60s
> > > > The
> > > > > > >> maximum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > [range,
> > > > > > >> uniform],
> > > > > > >> > > > > > which means
> > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> like to
> > > > > know
> > > > > > >> why not
> > > > > > >> > > > > > uniform
> > > > > > >> > > > > > > > > one? I
> > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > assignor
> > > > > > >> (former
> > > > > > >> > > > sticky
> > > > > > >> > > > > > > > > assinor)
> > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > reason
> > > > > we
> > > > > > >> choose
> > > > > > >> > > > range
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > as default?
> > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> uniform
> > > The
> > > > > > >> server side
> > > > > > >> > > > > > assignors.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Thank you.
> > > > > > >> > > > > > > > > > > Luke
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen
> <
> > > > > > >> showuon@gmail.com
> > > > > > >> > > > >
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > > > >> integrating with
> > > > > > >> > > > > > KIP-848,
> > > > > > >> > > > > > > > > but I
> > > > > > >> > > > > > > > > > > > think we should have a separate
> discussion
> > > > > thread
> > > > > > >> for the
> > > > > > >> > > > Kafka
> > > > > > >> > > > > > > > > Connect
> > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > > > Consumer
> > > > > > >> Rebalance
> > > > > > >> > > > > > Protocol
> > > > > > >> > > > > > > > > [1],
> > > > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > > > consumer
> > > > > > >> rebalance
> > > > > > >> > > > > > protocol,
> > > > > > >> > > > > > > > > WDYT?
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > [1]
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > >> > > > > > > > > > > > Luke
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> feedback.
> > > > > Looks
> > > > > > >> like
> > > > > > >> > > > there's
> > > > > > >> > > > > > > > > agreement on
> > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> would
> > > > > revisit
> > > > > > >> the doc
> > > > > > >> > > > and
> > > > > > >> > > > > > see
> > > > > > >> > > > > > > > > what
> > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David
> Jacot
> > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> document.
> > > > > That's
> > > > > > >> really
> > > > > > >> > > > > > helpful. I
> > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> Connect
> > > > > and the
> > > > > > >> > > > Consumer
> > > > > > >> > > > > > could
> > > > > > >> > > > > > > > > share
> > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> main
> > > > > > >> difference is
> > > > > > >> > > > that
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > Consumer
> > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > > whereas
> > > > > > >> Connect
> > > > > > >> > > > assigns
> > > > > > >> > > > > > tasks
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> forward:
> > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > > > support
> > > > > > >> different
> > > > > > >> > > > > > resource
> > > > > > >> > > > > > > > > types
> > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> Connect.
> > > > The
> > > > > > >> dedicated
> > > > > > >> > > > APIs
> > > > > > >> > > > > > > > > would be
> > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> different on
> > > > the
> > > > > > >> > > > > > content/resources and
> > > > > > >> > > > > > > > > > > >> > they would rely on the same engine
> on the
> > > > > > >> coordinator
> > > > > > >> > > > side.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > I personally lean towards 2) because
> I am
> > > > > not a
> > > > > > >> fan of
> > > > > > >> > > > > > > > > overcharging
> > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> That
> > > > being
> > > > > > >> said, I am
> > > > > > >> > > > not
> > > > > > >> > > > > > > > > opposed to
> > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to
> do
> > > it.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> discuss
> > > it
> > > > > here
> > > > > > >> for now
> > > > > > >> > > > in
> > > > > > >> > > > > > order
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible
> with
> > > > what
> > > > > we
> > > > > > >> will do
> > > > > > >> > > > for
> > > > > > >> > > > > > > > > Connect in
> > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > >> > > > > > > > > > > >> > David
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> > > Jacot
> > > > <
> > > > > > >> > > > > > djacot@confluent.io>
> > > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > > > through
> > > > > and
> > > > > > >> address
> > > > > > >> > > > > > your
> > > > > > >> > > > > > > > > comments
> > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > > > feedback.
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > >> > > > > > > > > > > >> > > David
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> Gregory
> > > > > Harris
> > > > > > >> <
> > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> wonderful to
> > > > see
> > > > > > >> > > > cooperative
> > > > > > >> > > > > > > > > rebalancing
> > > > > > >> > > > > > > > > > > >> > making it
> > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > > section
> > > > > > >> describes how
> > > > > > >> > > > > > member
> > > > > > >> > > > > > > > > epoch
> > > > > > >> > > > > > > > > > > >> > should
> > > > > > >> > > > > > > > > > > >> > > > advance in step with the group
> epoch
> > > > and
> > > > > > >> assignment
> > > > > > >> > > > > > epoch
> > > > > > >> > > > > > > > > values. I
> > > > > > >> > > > > > > > > > > >> > think
> > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > > > reasons
> > > > > > >> described
> > > > > > >> > > > in
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > KIP. When
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> assigning
> > > > > > >> partitions to a
> > > > > > >> > > > > > worker,
> > > > > > >> > > > > > > > > what
> > > > > > >> > > > > > > > > > > >> member
> > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > assignment
> > > > > use?
> > > > > > >> Are
> > > > > > >> > > > member
> > > > > > >> > > > > > epochs
> > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > >> > > > > > > > > > > >> > and
> > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> correspond
> > > to
> > > > > > >> multiple
> > > > > > >> > > > > > different
> > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> field
> > > > > opaque
> > > > > > >> to the
> > > > > > >> > > > group
> > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > > assignor
> > > > > > >> > > > implementations
> > > > > > >> > > > > > > > > interact
> > > > > > >> > > > > > > > > > > >> with
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> common
> > > > > meaning
> > > > > > >> agreed
> > > > > > >> > > > > > upon? If
> > > > > > >> > > > > > > > > so, what
> > > > > > >> > > > > > > > > > > >> > is the
> > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> field
> > > over
> > > > > > >> including
> > > > > > >> > > > such
> > > > > > >> > > > > > > > > functionality
> > > > > > >> > > > > > > > > > > >> > in the
> > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 3. The following is included in
> the
> > > > KIP:
> > > > > > >> "Thanks to
> > > > > > >> > > > > > this, the
> > > > > > >> > > > > > > > > input
> > > > > > >> > > > > > > > > > > >> of
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> > > driven
> > > > > by
> > > > > > >> the group
> > > > > > >> > > > > > > > > coordinator.
> > > > > > >> > > > > > > > > > > >> The
> > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> responsible for
> > > > > > >> maintaining
> > > > > > >> > > > any
> > > > > > >> > > > > > state
> > > > > > >> > > > > > > > > besides
> > > > > > >> > > > > > > > > > > >> its
> > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this
> mean
> > > > > that the
> > > > > > >> > > > > > client-side
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > >> MAY
> > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> non-Metadata
> > > > state
> > > > > > >> (such as
> > > > > > >> > > > > > partition
> > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> topics,
> > > > etc),
> > > > > or
> > > > > > >> that
> > > > > > >> > > > > > additional
> > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > > > classes
> > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > >> > > > > >
> org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > >> > > > > > that
> > > > > > >> > > > > > > > > seem to
> > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> possible
> > > > for
> > > > > > >> these two
> > > > > > >> > > > > > > > > implementations
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> promote
> > > > > feature
> > > > > > >> parity
> > > > > > >> > > > of
> > > > > > >> > > > > > > > > server-side
> > > > > > >> > > > > > > > > > > >> and
> > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and would
> also
> > > > > > >> facilitate
> > > > > > >> > > > > > operational
> > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > >> > > > > > > > > > > >> > > > certain situations. For example,
> if a
> > > > > > >> server-side
> > > > > > >> > > > > > assignor
> > > > > > >> > > > > > > > > has some
> > > > > > >> > > > > > > > > > > >> > poor
> > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> deploying
> > > > the
> > > > > > >> patched
> > > > > > >> > > > > > assignor to
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > and switching one consumer group
> to a
> > > > > > >> client-side
> > > > > > >> > > > > > assignor
> > > > > > >> > > > > > > > > may be
> > > > > > >> > > > > > > > > > > >> > faster
> > > > > > >> > > > > > > > > > > >> > > > and less risky than patching all
> of
> > > the
> > > > > > >> brokers.
> > > > > > >> > > > With
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > currently
> > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> non-trivial
> > > > > > >> > > > reimplementation
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > have
> > > > > > >> > > > > > > > > > > >> to be
> > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs
> have
> > > > > diverged
> > > > > > >> > > > > > significantly,
> > > > > > >> > > > > > > > > then it
> > > > > > >> > > > > > > > > > > >> is
> > > > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> > > would
> > > > > not be
> > > > > > >> > > > possible.
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM
> Sagar
> > > <
> > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > > > discuss
> > > > > how
> > > > > > >> Connect
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > need to
> > > > > > >> > > > > > > > > > > >> > change
> > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > protocol.
> > > > > > >> Here's the
> > > > > > >> > > > page:
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I
> have
> > > > > tried
> > > > > > >> to keep
> > > > > > >> > > > a
> > > > > > >> > > > > > format
> > > > > > >> > > > > > > > > > > >> similar to
> > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > > think.
> > > > > Also,
> > > > > > >> do you
> > > > > > >> > > > > > think this
> > > > > > >> > > > > > > > > > > >> should
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> > > thread
> > > > > or
> > > > > > >> is this
> > > > > > >> > > > one
> > > > > > >> > > > > > fine?
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > > > Sagar <
> > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> doc on
> > > > > Kafka
> > > > > > >> Streams.
> > > > > > >> > > > > > Sure, I
> > > > > > >> > > > > > > > > would do
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> such a
> > > > > > >> document.
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47
> AM
> > > > > Guozhang
> > > > > > >> Wang <
> > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> could
> > > > come
> > > > > > >> back with
> > > > > > >> > > > some
> > > > > > >> > > > > > > > > analysis on
> > > > > > >> > > > > > > > > > > >> > how to
> > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > > integration
> > > > > > >> with
> > > > > > >> > > > the new
> > > > > > >> > > > > > > > > protocol;
> > > > > > >> > > > > > > > > > > >> so
> > > > > > >> > > > > > > > > > > >> > far
> > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the
> new
> > > > > "protocol
> > > > > > >> type"
> > > > > > >> > > > we
> > > > > > >> > > > > > did not
> > > > > > >> > > > > > > > > yet
> > > > > > >> > > > > > > > > > > >> think
> > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > implementations.
> > > > > For
> > > > > > >> Streams
> > > > > > >> > > > > > here's a
> > > > > > >> > > > > > > > > draft of
> > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis
> on
> > > > > Connect.
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at
> 10:48 PM
> > > > > Sagar <
> > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> response.
> > > The
> > > > > > >> reason I
> > > > > > >> > > > thought
> > > > > > >> > > > > > > > > connect can
> > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that
> even
> > > > > today the
> > > > > > >> > > > connect
> > > > > > >> > > > > > uses a
> > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > AbstractCoordinator to
> > > > > > >> empower
> > > > > > >> > > > > > > > > rebalances of
> > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > > > >> WorkerCoordinator sets
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata()
> method
> > > > by
> > > > > > >> plumbing
> > > > > > >> > > > into
> > > > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to
> support
> > > > > > >> connect would
> > > > > > >> > > > be
> > > > > > >> > > > > > > > > similar at a
> > > > > > >> > > > > > > > > > > >> > high
> > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams
> mainly
> > > > > because
> > > > > > >> of the
> > > > > > >> > > > > > Client
> > > > > > >> > > > > > > > > side
> > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > implementation
> > > > > > >> level, we
> > > > > > >> > > > > > might
> > > > > > >> > > > > > > > > need to
> > > > > > >> > > > > > > > > > > >> make
> > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this
> new
> > > > > > >> assignment
> > > > > > >> > > > protocol
> > > > > > >> > > > > > like
> > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> request/response and
> > > > > > >> SyncGroup and
> > > > > > >> > > > > > using
> > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar
> lines
> > > > to
> > > > > > >> streams (or
> > > > > > >> > > > > > there
> > > > > > >> > > > > > > > > might be
> > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> > > detailed
> > > > > > >> analysis of
> > > > > > >> > > > the
> > > > > > >> > > > > > same
> > > > > > >> > > > > > > > > and we
> > > > > > >> > > > > > > > > > > >> > can have
> > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> thread for
> > > > > that
> > > > > > >> as that
> > > > > > >> > > > would
> > > > > > >> > > > > > > > > derail this
> > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if
> that
> > > > sounds
> > > > > > >> good to
> > > > > > >> > > > you.
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at
> 5:47
> > > PM
> > > > > David
> > > > > > >> Jacot
> > > > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> comments.
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > > > >> `Assignment#error`.
> > > > > > >> > > > > > Sure, I
> > > > > > >> > > > > > > > > can
> > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > transition C
> > > > > from
> > > > > > >> his
> > > > > > >> > > > current
> > > > > > >> > > > > > > > > assignment
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > his
> > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when
> he
> > > can
> > > > > move
> > > > > > >> to
> > > > > > >> > > > epoch 3.
> > > > > > >> > > > > > When
> > > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is
> updated
> > > > and
> > > > > > >> persisted
> > > > > > >> > > > > > with all
> > > > > > >> > > > > > > > > its
> > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if they
> are
> > > > not
> > > > > all
> > > > > > >> revoked
> > > > > > >> > > > > > yet. In
> > > > > > >> > > > > > > > > other
> > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> becomes
> > > the
> > > > > target
> > > > > > >> > > > > > assignment.
> > > > > > >> > > > > > > > > This is
> > > > > > >> > > > > > > > > > > >> > basically
> > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid
> having
> > > > to
> > > > > > >> write all
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > changes to
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > log.
> > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on
> the
> > > > > persisted
> > > > > > >> state
> > > > > > >> > > > so I
> > > > > > >> > > > > > > > > understand
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see
> if I
> > > can
> > > > > > >> improve
> > > > > > >> > > > this in
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> > description.
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it
> > > could
> > > > > > >> reuse the
> > > > > > >> > > > > > protocol
> > > > > > >> > > > > > > > > with a
> > > > > > >> > > > > > > > > > > >> > client side
> > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in
> the
> > > > > > >> protocol. The
> > > > > > >> > > > > > assignment
> > > > > > >> > > > > > > > > is
> > > > > > >> > > > > > > > > > > >> about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> > > metadata,
> > > > > could
> > > > > > >> > > > Connect
> > > > > > >> > > > > > fit
> > > > > > >> > > > > > > > > into this?
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at
> 1:55
> > > > PM
> > > > > > >> Sagar <
> > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I
> just
> > > > had
> > > > > > >> minor
> > > > > > >> > > > > > observations:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment
> Error
> > > > > > >> section in
> > > > > > >> > > > Client
> > > > > > >> > > > > > Side
> > > > > > >> > > > > > > > > mode
> > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In
> this
> > > > > case,
> > > > > > >> the
> > > > > > >> > > > client
> > > > > > >> > > > > > side
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > >> can
> > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > > > coordinator`.
> > > > > > >> In this
> > > > > > >> > > > > > case are
> > > > > > >> > > > > > > > > you
> > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > > > > >> AssignmentError
> > > > > > >> > > > that's
> > > > > > >> > > > > > > > > listed down
> > > > > > >> > > > > > > > > > > >> > towards
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think
> it
> > > > would
> > > > > > >> make sense
> > > > > > >> > > > to
> > > > > > >> > > > > > > > > mention this
> > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > > > > section, I
> > > > > > >> have a
> > > > > > >> > > > > > slight
> > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > >> > > > > > > > > > > >> > not
> > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the same.
> > > > Consider
> > > > > > >> this step:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the
> > > group
> > > > > > >> coordinator
> > > > > > >> > > > > > > > > transitions him
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > > > > revoke. It
> > > > > > >> > > > persists
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > change and
> > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > > > > (epoch=3)
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > partitions=[foo-0]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > partitions=[foo-2]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > partitions=[foo-1]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > > > >> partitions=[foo-0,
> > > > > > >> > > > > > foo-1]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > > > >> partitions=[foo-2]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > > > >> partitions=[foo-1]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > > > > transitions
> > > > > > >> to
> > > > > > >> > > > epoch 3
> > > > > > >> > > > > > but
> > > > > > >> > > > > > > > > cannot
> > > > > > >> > > > > > > > > > > >> get
> > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned
> that
> > > > > member C
> > > > > > >> can't
> > > > > > >> > > > get
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > foo-1
> > > > > > >> > > > > > > > > > > >> > partition
> > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> description
> > > > > above,
> > > > > > >> it seems
> > > > > > >> > > > it
> > > > > > >> > > > > > > > > already has
> > > > > > >> > > > > > > > > > > >> it.
> > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to
> remove
> > > it
> > > > > and
> > > > > > >> > > > populate it
> > > > > > >> > > > > > only
> > > > > > >> > > > > > > > > when it
> > > > > > >> > > > > > > > > > > >> > actually
> > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of
> > > other
> > > > > > >> places, so
> > > > > > >> > > > have
> > > > > > >> > > > > > I
> > > > > > >> > > > > > > > > > > >> understood it
> > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it
> > > might
> > > > > be
> > > > > > >> out of
> > > > > > >> > > > scope
> > > > > > >> > > > > > of
> > > > > > >> > > > > > > > > this
> > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it
> would
> > > > > > >> probably be
> > > > > > >> > > > > > running in
> > > > > > >> > > > > > > > > client
> > > > > > >> > > > > > > > > > > >> > side
> > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> rebalance
> > > > > > >> protocol as it
> > > > > > >> > > > has
> > > > > > >> > > > > > its
> > > > > > >> > > > > > > > > own
> > > > > > >> > > > > > > > > > > >> Custom
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > IncrementalCooperative).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022
> at
> > > 5:00
> > > > > PM
> > > > > > >> David
> > > > > > >> > > > Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our
> goal
> > > is
> > > > > to
> > > > > > >> move
> > > > > > >> > > > forward
> > > > > > >> > > > > > with
> > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying
> on
> > > one
> > > > > > >> generic API.
> > > > > > >> > > > For
> > > > > > >> > > > > > > > > Connect, we
> > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern
> and
> > > > > > >> reuse/share the
> > > > > > >> > > > core
> > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the
> > > > schema
> > > > > > >> registry, I
> > > > > > >> > > > > > think
> > > > > > >> > > > > > > > > that we
> > > > > > >> > > > > > > > > > > >> > should
> > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored
> API to
> > > > do
> > > > > > >> simple
> > > > > > >> > > > > > > > > membership/leader
> > > > > > >> > > > > > > > > > > >> > election.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15,
> 2022 at
> > > > > 10:22
> > > > > > >> AM Ismael
> > > > > > >> > > > > > Juma <
> > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> comments:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > >> java.util.regex.Pattern
> > > > > > >> > > > vs
> > > > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > differences
> > > > > in
> > > > > > >> more
> > > > > > >> > > > detail
> > > > > > >> > > > > > before
> > > > > > >> > > > > > > > > > > >> deciding
> > > > > > >> > > > > > > > > > > >> > one
> > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if
> people
> > > > pass
> > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > >> > > > > > > > > > > >> they
> > > > > > >> > > > > > > > > > > >> > expect
> > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be
> > > honored.
> > > > > If
> > > > > > >> we are
> > > > > > >> > > > doing
> > > > > > >> > > > > > > > > something
> > > > > > >> > > > > > > > > > > >> > different,
> > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider
> adding
> > > > an
> > > > > > >> overload
> > > > > > >> > > > with
> > > > > > >> > > > > > our own
> > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > >> > > > > > > > > > > >> > class
> > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to
> > > > expose
> > > > > > >> re2j's at
> > > > > > >> > > > this
> > > > > > >> > > > > > > > > point).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic
> > > ids,
> > > > > any
> > > > > > >> major new
> > > > > > >> > > > > > protocol
> > > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should
> handle
> > > the
> > > > > topic
> > > > > > >> > > > > > recreation case
> > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to
> > > handle.
> > > > I
> > > > > > >> agree with
> > > > > > >> > > > > > David
> > > > > > >> > > > > > > > > that we'd
> > > > > > >> > > > > > > > > > > >> > want to
> > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > > > protocols
> > > > > > >> that don't
> > > > > > >> > > > > > have it
> > > > > > >> > > > > > > > > yet and
> > > > > > >> > > > > > > > > > > >> > that we
> > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> internals
> > > > > versus
> > > > > > >> adding
> > > > > > >> > > > new
> > > > > > >> > > > > > APIs
> > > > > > >> > > > > > > > > to the
> > > > > > >> > > > > > > > > > > >> > Java
> > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> adding
> > > new
> > > > > APIs
> > > > > > >> is
> > > > > > >> > > > required
> > > > > > >> > > > > > for
> > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not
> sure
> > > > > about
> > > > > > >> the
> > > > > > >> > > > > > coordinator
> > > > > > >> > > > > > > > > storing
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> configs
> > > to
> > > > be
> > > > > > >> > > > centralized
> > > > > > >> > > > > > in the
> > > > > > >> > > > > > > > > > > >> metadata
> > > > > > >> > > > > > > > > > > >> > log for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> (auditability,
> > > > > > >> visibility,
> > > > > > >> > > > > > consistency,
> > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > automatically
> > > > > > >> deleting
> > > > > > >> > > > > > configs in a
> > > > > > >> > > > > > > > > way
> > > > > > >> > > > > > > > > > > >> that
> > > > > > >> > > > > > > > > > > >> > they
> > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> > > > property
> > > > > > >> for modern
> > > > > > >> > > > > > systems
> > > > > > >> > > > > > > > > is to
> > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data
> > > loss
> > > > > > >> scenarios.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13,
> 2022
> > > at
> > > > > 3:47
> > > > > > >> PM David
> > > > > > >> > > > > > Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> Guozhang. My
> > > > > answers
> > > > > > >> are
> > > > > > >> > > > below:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> migration
> > > > path,
> > > > > > >> especially
> > > > > > >> > > > > > the last
> > > > > > >> > > > > > > > > step
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the
> new
> > > > > > >> protocol, in
> > > > > > >> > > > which
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> rpcs
> > > and
> > > > > old
> > > > > > >> > > > protocols /
> > > > > > >> > > > > > rpcs
> > > > > > >> > > > > > > > > are
> > > > > > >> > > > > > > > > > > >> used
> > > > > > >> > > > > > > > > > > >> > by
> > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > > >> coordinator could
> > > > > > >> > > > > > "mimic"
> > > > > > >> > > > > > > > > the old
> > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> is
> > > > > something
> > > > > > >> we
> > > > > > >> > > > need to
> > > > > > >> > > > > > > > > present
> > > > > > >> > > > > > > > > > > >> about.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > > > published
> > > > > a
> > > > > > >> new
> > > > > > >> > > > version
> > > > > > >> > > > > > of KIP
> > > > > > >> > > > > > > > > which
> > > > > > >> > > > > > > > > > > >> > includes
> > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about
> this.
> > > > See
> > > > > the
> > > > > > >> > > > "Supporting
> > > > > > >> > > > > > > > > Online
> > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > "Compatibility,
> > > > > > >> > > > Deprecation,
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > Migration
> > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> > > > through
> > > > > a
> > > > > > >> few cases
> > > > > > >> > > > > > now but
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> should be
> > > > > > >> understandable.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> of
> > > > topic
> > > > > > >> ids. So
> > > > > > >> > > > far as
> > > > > > >> > > > > > > > > KIP-516 the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> and
> > > > admin
> > > > > > >> client,
> > > > > > >> > > > but
> > > > > > >> > > > > > they
> > > > > > >> > > > > > > > > are not
> > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> yet.
> > > I
> > > > > think
> > > > > > >> the
> > > > > > >> > > > > > question is,
> > > > > > >> > > > > > > > > first
> > > > > > >> > > > > > > > > > > >> > should we
> > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > maintaining
> > > > > > >> the
> > > > > > >> > > > names
> > > > > > >> > > > > > -> ids
> > > > > > >> > > > > > > > > mapping
> > > > > > >> > > > > > > > > > > >> > itself
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > augmented
> > > > > > >> existing
> > > > > > >> > > > RPCs
> > > > > > >> > > > > > and the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > >> > > > > > > > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> should
> > > we
> > > > > ever
> > > > > > >> consider
> > > > > > >> > > > > > exposing
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > > well
> > > > > (both
> > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > >> > > > > > > > > > > >> well
> > > > > > >> > > > > > > > > > > >> > as in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> cases
> > > > > like
> > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming
> that we
> > > > > would
> > > > > > >> include
> > > > > > >> > > > > > converting
> > > > > > >> > > > > > > > > all
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using
> topic
> > > > ids
> > > > > in
> > > > > > >> this
> > > > > > >> > > > KIP,
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> > would be
> > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate
> with
> > > > topic
> > > > > > >> ids. That
> > > > > > >> > > > > > being
> > > > > > >> > > > > > > > > said, it
> > > > > > >> > > > > > > > > > > >> > still has
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> names in
> > > > > various
> > > > > > >> APIs so
> > > > > > >> > > > > > having a
> > > > > > >> > > > > > > > > > > >> mapping
> > > > > > >> > > > > > > > > > > >> > in the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> inevitable to
> > > > me.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have
> a
> > > > strong
> > > > > > >> opinion on
> > > > > > >> > > > > > this.
> > > > > > >> > > > > > > > > Here I
> > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> scope of
> > > > this
> > > > > > >> KIP. I
> > > > > > >> > > > would
> > > > > > >> > > > > > rather
> > > > > > >> > > > > > > > > focus
> > > > > > >> > > > > > > > > > > >> on
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> > > > consider
> > > > > > >> this
> > > > > > >> > > > > > separately if
> > > > > > >> > > > > > > > > we see
> > > > > > >> > > > > > > > > > > >> > value in
> > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > > > Ismael's
> > > > > > >> point about
> > > > > > >> > > > > > using
> > > > > > >> > > > > > > > > topic ids
> > > > > > >> > > > > > > > > > > >> > in the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > >> > > > think
> > > > > > >> > > > > > that
> > > > > > >> > > > > > > > > there
> > > > > > >> > > > > > > > > > > >> is
> > > > > > >> > > > > > > > > > > >> > one
> > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it.
> The
> > > > > consumer
> > > > > > >> will
> > > > > > >> > > > have
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> exists
> > > > before
> > > > > > >> passing
> > > > > > >> > > > them
> > > > > > >> > > > > > into
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> group
> > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> Obviously,
> > > > the
> > > > > > >> > > > coordinator
> > > > > > >> > > > > > will
> > > > > > >> > > > > > > > > also
> > > > > > >> > > > > > > > > > > >> notice
> > > > > > >> > > > > > > > > > > >> > it but
> > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have
> a
> > > way
> > > > to
> > > > > > >> reject an
> > > > > > >> > > > > > invalid
> > > > > > >> > > > > > > > > topic in
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> with
> > > > > David
> > > > > > >> on all
> > > > > > >> > > > other
> > > > > > >> > > > > > minor
> > > > > > >> > > > > > > > > > > >> questions
> > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > `subscribe(Pattern)`
> > > > > > >> question:
> > > > > > >> > > > > > personally
> > > > > > >> > > > > > > > > I think
> > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > > subscribe
> > > > > > >> API with
> > > > > > >> > > > > > Pattern,
> > > > > > >> > > > > > > > > but
> > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > documenting
> > > > > > >> that our
> > > > > > >> > > > > > > > > subscription may
> > > > > > >> > > > > > > > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > incompatible
> > > > > > >> case is a
> > > > > > >> > > > > > very rare
> > > > > > >> > > > > > > > > > > >> scenario
> > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > `String`
> > > > > based
> > > > > > >> > > > > > subscription may
> > > > > > >> > > > > > > > > be
> > > > > > >> > > > > > > > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> regexes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> work. I
> > > > have
> > > > > to
> > > > > > >> look at
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > differences
> > > > > > >> > > > > > > > > > > >> > between
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> better
> > > > > > >> understand the
> > > > > > >> > > > > > potential
> > > > > > >> > > > > > > > > issues.
> > > > > > >> > > > > > > > > > > >> My
> > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work
> for
> > > > all
> > > > > the
> > > > > > >> basic
> > > > > > >> > > > > > regular
> > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two
> are
> > > > > mainly
> > > > > > >> about
> > > > > > >> > > > the
> > > > > > >> > > > > > various
> > > > > > >> > > > > > > > > > > >> character
> > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what
> other
> > > > people
> > > > > > >> think
> > > > > > >> > > > about
> > > > > > >> > > > > > this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12,
> 2022
> > > > at
> > > > > > >> 11:28 PM
> > > > > > >> > > > > > Guozhang
> > > > > > >> > > > > > > > > Wang <
> > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David!
> I
> > > > think
> > > > > on
> > > > > > >> the high
> > > > > > >> > > > > > level
> > > > > > >> > > > > > > > > there are
> > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize
> a
> > > bit
> > > > > more:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> migration
> > > > path,
> > > > > > >> especially
> > > > > > >> > > > > > the last
> > > > > > >> > > > > > > > > step
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the
> new
> > > > > > >> protocol, in
> > > > > > >> > > > which
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> rpcs
> > > and
> > > > > old
> > > > > > >> > > > protocols /
> > > > > > >> > > > > > rpcs
> > > > > > >> > > > > > > > > are
> > > > > > >> > > > > > > > > > > >> used
> > > > > > >> > > > > > > > > > > >> > by
> > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > > >> coordinator could
> > > > > > >> > > > > > "mimic"
> > > > > > >> > > > > > > > > the old
> > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> is
> > > > > something
> > > > > > >> we
> > > > > > >> > > > need to
> > > > > > >> > > > > > > > > present
> > > > > > >> > > > > > > > > > > >> about.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> of
> > > > topic
> > > > > > >> ids. So
> > > > > > >> > > > far as
> > > > > > >> > > > > > > > > KIP-516 the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> and
> > > > admin
> > > > > > >> client,
> > > > > > >> > > > but
> > > > > > >> > > > > > they
> > > > > > >> > > > > > > > > are not
> > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> yet.
> > > I
> > > > > think
> > > > > > >> the
> > > > > > >> > > > > > question is,
> > > > > > >> > > > > > > > > first
> > > > > > >> > > > > > > > > > > >> > should we
> > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > maintaining
> > > > > > >> the
> > > > > > >> > > > names
> > > > > > >> > > > > > -> ids
> > > > > > >> > > > > > > > > mapping
> > > > > > >> > > > > > > > > > > >> > itself
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > augmented
> > > > > > >> existing
> > > > > > >> > > > RPCs
> > > > > > >> > > > > > and the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > >> > > > > > > > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> should
> > > we
> > > > > ever
> > > > > > >> consider
> > > > > > >> > > > > > exposing
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > > well
> > > > > (both
> > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > >> > > > > > > > > > > >> well
> > > > > > >> > > > > > > > > > > >> > as in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> cases
> > > > > like
> > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> with
> > > > > David
> > > > > > >> on all
> > > > > > >> > > > other
> > > > > > >> > > > > > minor
> > > > > > >> > > > > > > > > > > >> questions
> > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > `subscribe(Pattern)`
> > > > > > >> question:
> > > > > > >> > > > > > personally
> > > > > > >> > > > > > > > > I think
> > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > > subscribe
> > > > > > >> API with
> > > > > > >> > > > > > Pattern,
> > > > > > >> > > > > > > > > but
> > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > documenting
> > > > > > >> that our
> > > > > > >> > > > > > > > > subscription may
> > > > > > >> > > > > > > > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > incompatible
> > > > > > >> case is a
> > > > > > >> > > > > > very rare
> > > > > > >> > > > > > > > > > > >> scenario
> > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > `String`
> > > > > based
> > > > > > >> > > > > > subscription may
> > > > > > >> > > > > > > > > be
> > > > > > >> > > > > > > > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> regexes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul
> 12,
> > > 2022
> > > > > at
> > > > > > >> 5:23 AM
> > > > > > >> > > > > > David Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for
> your
> > > > > > >> feedback. Let
> > > > > > >> > > > me
> > > > > > >> > > > > > answer
> > > > > > >> > > > > > > > > your
> > > > > > >> > > > > > > > > > > >> > questions
> > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think
> > > it's
> > > > > > >> premature to
> > > > > > >> > > > > > talk about
> > > > > > >> > > > > > > > > > > >> target
> > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal
> of the
> > > > > > >> existing group
> > > > > > >> > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> > > protocol
> > > > > and
> > > > > > >> hence
> > > > > > >> > > > > > > > > deprecation/removal
> > > > > > >> > > > > > > > > > > >> > will be
> > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> quickly
> > > > > > >> applications
> > > > > > >> > > > migrate
> > > > > > >> > > > > > to
> > > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes
> > > sense.
> > > > I
> > > > > will
> > > > > > >> > > > remove it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP
> > > says
> > > > we
> > > > > > >> intend to
> > > > > > >> > > > > > release
> > > > > > >> > > > > > > > > this in
> > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we
> > > added
> > > > > that
> > > > > > >> as a
> > > > > > >> > > > way to
> > > > > > >> > > > > > > > > estimate
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > > > > protocol, I
> > > > > > >> also
> > > > > > >> > > > > > suggest
> > > > > > >> > > > > > > > > removing
> > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> explain
> > > my
> > > > > > >> reasoning. As
> > > > > > >> > > > > > > > > explained, I
> > > > > > >> > > > > > > > > > > >> plan
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> in
> > > > Java
> > > > > > >> while we
> > > > > > >> > > > > > implement
> > > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> > > internals
> > > > > will
> > > > > > >> be
> > > > > > >> > > > slightly
> > > > > > >> > > > > > > > > different
> > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I
> > > > wanted
> > > > > to
> > > > > > >> tighten
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > switch from
> > > > > > >> > > > > > > > > > > >> > the old
> > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> to
> > > the
> > > > > new
> > > > > > >> group
> > > > > > >> > > > > > coordinator
> > > > > > >> > > > > > > > > to a
> > > > > > >> > > > > > > > > > > >> > major
> > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative
> > > would
> > > > > be to
> > > > > > >> use a
> > > > > > >> > > > flag
> > > > > > >> > > > > > to do
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > switch
> > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> software
> > > > > upgrade.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We
> need to
> > > > > flesh
> > > > > > >> out the
> > > > > > >> > > > > > details of
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > migration
> > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're
> saying
> > > we
> > > > > will
> > > > > > >> support
> > > > > > >> > > > > > online
> > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > >> > > > > > > > > > > >> > Is
> > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain
> this
> > > in
> > > > > > >> detail. It
> > > > > > >> > > > could
> > > > > > >> > > > > > also
> > > > > > >> > > > > > > > > be done
> > > > > > >> > > > > > > > > > > >> > as a
> > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> > > > support
> > > > > > >> online
> > > > > > >> > > > > > migrations
> > > > > > >> > > > > > > > > for the
> > > > > > >> > > > > > > > > > > >> > group.
> > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group
> using
> > > the
> > > > > old
> > > > > > >> protocol
> > > > > > >> > > > > > will be
> > > > > > >> > > > > > > > > able to
> > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> briefly
> > > > > explain
> > > > > > >> how that
> > > > > > >> > > > > > will work
> > > > > > >> > > > > > > > > > > >> though.
> > > > > > >> > > > > > > > > > > >> > It is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step
> > > process:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> cluster
> > > > must
> > > > > be
> > > > > > >> > > > upgraded or
> > > > > > >> > > > > > > > > rolled to a
> > > > > > >> > > > > > > > > > > >> > software
> > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > > > coordinator.
> > > > > > >> Both the
> > > > > > >> > > > > > old and
> > > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the
> old
> > > > > > >> protocol and
> > > > > > >> > > > rely
> > > > > > >> > > > > > on the
> > > > > > >> > > > > > > > > same
> > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can
> work
> > > > > together.
> > > > > > >> This
> > > > > > >> > > > point
> > > > > > >> > > > > > is an
> > > > > > >> > > > > > > > > > > >> offline
> > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one
> live
> > > > > > >> because it
> > > > > > >> > > > would
> > > > > > >> > > > > > require
> > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> and
> > > > > > >> starting up the
> > > > > > >> > > > > > new one
> > > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> that
> > > > > > >> > > > > > > > > > > >> > would
> > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > unavailabilities.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> cluster's
> > > > > > >> metadata
> > > > > > >> > > > > > version/IBP
> > > > > > >> > > > > > > > > must be
> > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable
> the
> > > new
> > > > > > >> protocol.
> > > > > > >> > > > This
> > > > > > >> > > > > > cannot
> > > > > > >> > > > > > > > > be done
> > > > > > >> > > > > > > > > > > >> > before
> > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated
> > > because
> > > > > the
> > > > > > >> old
> > > > > > >> > > > > > coordinator
> > > > > > >> > > > > > > > > doesn't
> > > > > > >> > > > > > > > > > > >> > support
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> consumers
> > > > > must be
> > > > > > >> > > > upgraded
> > > > > > >> > > > > > to a
> > > > > > >> > > > > > > > > version
> > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> (must
> > > > have
> > > > > > >> KIP-792).
> > > > > > >> > > > If
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> is
> > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing
> must be
> > > > > done at
> > > > > > >> this
> > > > > > >> > > > point.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> consumers
> > > > > must
> > > > > > >> be rolled
> > > > > > >> > > > > > with the
> > > > > > >> > > > > > > > > > > >> feature
> > > > > > >> > > > > > > > > > > >> > flag
> > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer
> group
> > > is
> > > > > > >> automatically
> > > > > > >> > > > > > > > > converted when
> > > > > > >> > > > > > > > > > > >> > the first
> > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the
> new
> > > > > protocol
> > > > > > >> joins
> > > > > > >> > > > the
> > > > > > >> > > > > > group.
> > > > > > >> > > > > > > > > While
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> protocol are
> > > > > being
> > > > > > >> > > > upgraded,
> > > > > > >> > > > > > the old
> > > > > > >> > > > > > > > > > > >> protocol
> > > > > > >> > > > > > > > > > > >> > is
> > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> clarify
> > > all
> > > > > of
> > > > > > >> this in
> > > > > > >> > > > the
> > > > > > >> > > > > > KIP.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am
> happy
> > > > > that we
> > > > > > >> are
> > > > > > >> > > > > > pushing the
> > > > > > >> > > > > > > > > pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it
> seems
> > > > like
> > > > > > >> there
> > > > > > >> > > > could be
> > > > > > >> > > > > > some
> > > > > > >> > > > > > > > > tricky
> > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> > > mechanism
> > > > > for
> > > > > > >> users to
> > > > > > >> > > > > > detect
> > > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> they
> > > > > > >> > > > > > > > > > > >> > need to
> > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> > > switching
> > > > > to
> > > > > > >> the new
> > > > > > >> > > > > > protocol?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think
> that I
> > > am
> > > > a
> > > > > bit
> > > > > > >> more
> > > > > > >> > > > > > optimistic
> > > > > > >> > > > > > > > > than
> > > > > > >> > > > > > > > > > > >> you
> > > > > > >> > > > > > > > > > > >> > on this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe
> that the
> > > > > > >> majority of
> > > > > > >> > > > the
> > > > > > >> > > > > > cases
> > > > > > >> > > > > > > > > are
> > > > > > >> > > > > > > > > > > >> simple
> > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with
> the
> > > new
> > > > > > >> engine. The
> > > > > > >> > > > > > > > > coordinator will
> > > > > > >> > > > > > > > > > > >> > verify
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject
> the
> > > > > consumer
> > > > > > >> if the
> > > > > > >> > > > > > regex is
> > > > > > >> > > > > > > > > not
> > > > > > >> > > > > > > > > > > >> valid.
> > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> path,
> > > in
> > > > > the
> > > > > > >> worst
> > > > > > >> > > > case,
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > first
> > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group
> will
> > > be
> > > > > > >> rejected.
> > > > > > >> > > > This
> > > > > > >> > > > > > should
> > > > > > >> > > > > > > > > be used
> > > > > > >> > > > > > > > > > > >> > as the
> > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > > > > customers to
> > > > > > >> > > > validate
> > > > > > >> > > > > > their
> > > > > > >> > > > > > > > > regex
> > > > > > >> > > > > > > > > > > >> > before
> > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would
> be to
> > > > > test
> > > > > > >> them with
> > > > > > >> > > > > > another
> > > > > > >> > > > > > > > > group.
> > > > > > >> > > > > > > > > > > >> For
> > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be
> done
> > > in a
> > > > > > >> pre-prod
> > > > > > >> > > > > > environment.
> > > > > > >> > > > > > > > > > > >> Another
> > > > > > >> > > > > > > > > > > >> > way
> > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > > > >> consumer-group tool
> > > > > > >> > > > to
> > > > > > >> > > > > > > > > provide a
> > > > > > >> > > > > > > > > > > >> regex
> > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism.
> Would
> > > > > this be
> > > > > > >> > > > enough in
> > > > > > >> > > > > > your
> > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> Related to
> > > > the
> > > > > last
> > > > > > >> > > > question,
> > > > > > >> > > > > > will
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> Java
> > > > > > >> > > > > > > > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick
> with the
> > > > > > >> current regex
> > > > > > >> > > > > > engine for
> > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be
> > > handy
> > > > to
> > > > > > >> keep using
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > based
> > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations
> > > > simple
> > > > > and
> > > > > > >> then
> > > > > > >> > > > > > migrate to
> > > > > > >> > > > > > > > > server
> > > > > > >> > > > > > > > > > > >> > based
> > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand
> > > your
> > > > > point
> > > > > > >> but I
> > > > > > >> > > > am
> > > > > > >> > > > > > > > > concerned that
> > > > > > >> > > > > > > > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually
> stay
> > > > in
> > > > > > >> this mode.
> > > > > > >> > > > That
> > > > > > >> > > > > > > > > would go
> > > > > > >> > > > > > > > > > > >> > against our
> > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying
> the
> > > > > client
> > > > > > >> because
> > > > > > >> > > > we
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > have to
> > > > > > >> > > > > > > > > > > >> > continue
> > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> metadata on
> > > > the
> > > > > > >> client
> > > > > > >> > > > side. I
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > rather
> > > > > > >> > > > > > > > > > > >> > not do
> > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When
> we say
> > > > > that
> > > > > > >> the group
> > > > > > >> > > > > > > > > coordinator
> > > > > > >> > > > > > > > > > > >> will
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > configurations
> > > > > > >> and that
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > > > >> configurations
> > > > > > >> > > > > > > > > > > >> > will be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > > > deleted.
> > > > > > >> Will a
> > > > > > >> > > > > > transition to
> > > > > > >> > > > > > > > > DEAD
> > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > configurations?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> right.
> > > The
> > > > > > >> > > > configurations
> > > > > > >> > > > > > will be
> > > > > > >> > > > > > > > > > > >> deleted
> > > > > > >> > > > > > > > > > > >> > when
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted.
> They go
> > > > > > >> together.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will
> the
> > > > > choice to
> > > > > > >> store
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > configs in
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to
> list
> > > > all
> > > > > > >> cluster
> > > > > > >> > > > > > configs and
> > > > > > >> > > > > > > > > their
> > > > > > >> > > > > > > > > > > >> > values?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't
> think
> > > so.
> > > > > The
> > > > > > >> group
> > > > > > >> > > > > > > > > configurations are
> > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If
> you
> > > > > want to
> > > > > > >> know
> > > > > > >> > > > all
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > overrides
> > > > > > >> > > > > > > > > > > >> > though,
> > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all
> the
> > > > group
> > > > > > >> > > > coordinators.
> > > > > > >> > > > > > You
> > > > > > >> > > > > > > > > cannot
> > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> instance.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How
> would
> > > > > someone
> > > > > > >> > > > configure a
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > before
> > > > > > >> > > > > > > > > > > >> > starting
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> considered
> > > > > > >> allowing the
> > > > > > >> > > > > > explicit
> > > > > > >> > > > > > > > > creation
> > > > > > >> > > > > > > > > > > >> of
> > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs
> could
> > > be
> > > > > > >> decoupled
> > > > > > >> > > > from
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > group
> > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The
> group
> > > > will
> > > > > be
> > > > > > >> > > > > > automatically
> > > > > > >> > > > > > > > > created in
> > > > > > >> > > > > > > > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs
> will
> > > > be
> > > > > > >> lost after
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > retention
> > > > > > >> > > > > > > > > > > >> > period of
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will
> the
> > > > > > >> > > > Consumer.subscribe
> > > > > > >> > > > > > method
> > > > > > >> > > > > > > > > for the
> > > > > > >> > > > > > > > > > > >> > Java
> > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> `java.util.regex.Pattern` of
> > > > > > >> > > > do
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > have to
> > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a
> very
> > > > group
> > > > > > >> question. I
> > > > > > >> > > > > > forgot
> > > > > > >> > > > > > > > > about
> > > > > > >> > > > > > > > > > > >> that
> > > > > > >> > > > > > > > > > > >> > one.
> > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> `java.util.regex.Pattern` is
> > > > > > >> > > > not
> > > > > > >> > > > > > fully
> > > > > > >> > > > > > > > > > > >> compatible
> > > > > > >> > > > > > > > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to
> use,
> > > it
> > > > > > >> might be
> > > > > > >> > > > better
> > > > > > >> > > > > > to
> > > > > > >> > > > > > > > > deprecate
> > > > > > >> > > > > > > > > > > >> > it and
> > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes
> a
> > > > > string.
> > > > > > >> We would
> > > > > > >> > > > > > rely on
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> server
> > > > > > >> > > > > > > > > > > >> > side
> > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > > > migration, I
> > > > > > >> think
> > > > > > >> > > > that
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > could
> > > > > > >> > > > > > > > > > > >> still
> > > > > > >> > > > > > > > > > > >> > try to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and
> use
> > > it.
> > > > > That
> > > > > > >> should
> > > > > > >> > > > > > work, I
> > > > > > >> > > > > > > > > think, in
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I
> agree
> > > with
> > > > > > >> Justine
> > > > > > >> > > > that we
> > > > > > >> > > > > > > > > should be
> > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > > > >> > > > IBP/metadata.version
> > > > > > >> > > > > > from the
> > > > > > >> > > > > > > > > > > >> feature
> > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> switch the
> > > > > > >> default for
> > > > > > >> > > > the
> > > > > > >> > > > > > feature
> > > > > > >> > > > > > > > > flag
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > true
> > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > metadata.version
> > > > > once
> > > > > > >> we
> > > > > > >> > > > want to
> > > > > > >> > > > > > make
> > > > > > >> > > > > > > > > it the
> > > > > > >> > > > > > > > > > > >> > default.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was
> to
> > > use
> > > > > that
> > > > > > >> feature
> > > > > > >> > > > > > flag
> > > > > > >> > > > > > > > > mainly
> > > > > > >> > > > > > > > > > > >> > during the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I
> should
> > > > not
> > > > > have
> > > > > > >> > > > mentioned
> > > > > > >> > > > > > it, I
> > > > > > >> > > > > > > > > think,
> > > > > > >> > > > > > > > > > > >> > because
> > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal
> > > config
> > > > > for
> > > > > > >> it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some
> of
> > > the
> > > > > > >> protocol APIs
> > > > > > >> > > > > > don't
> > > > > > >> > > > > > > > > mention
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to
> add
> > > that
> > > > > for
> > > > > > >> > > > consistency.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is
> a
> > > bit
> > > > > odd
> > > > > > >> that
> > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it
> > > seems
> > > > > to do
> > > > > > >> more
> > > > > > >> > > > than
> > > > > > >> > > > > > > > > reading.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree.
> This is
> > > > > how the
> > > > > > >> > > > current
> > > > > > >> > > > > > > > > protocol works
> > > > > > >> > > > > > > > > > > >> > though.
> > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require
> "Read
> > > > > Group" to
> > > > > > >> join a
> > > > > > >> > > > > > group. We
> > > > > > >> > > > > > > > > could
> > > > > > >> > > > > > > > > > > >> > consider
> > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not
> > > sure
> > > > > that
> > > > > > >> it is
> > > > > > >> > > > worth
> > > > > > >> > > > > > it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is
> > > topic
> > > > > > >> recreation
> > > > > > >> > > > > > handled by
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol?
> It
> > > > > would be
> > > > > > >> good to
> > > > > > >> > > > > > have a
> > > > > > >> > > > > > > > > section
> > > > > > >> > > > > > > > > > > >> on
> > > > > > >> > > > > > > > > > > >> > this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From
> a
> > > > > protocol
> > > > > > >> > > > > > perspective, the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > will
> > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so
> it
> > > > will
> > > > > > >> treat it
> > > > > > >> > > > like a
> > > > > > >> > > > > > > > > topic with
> > > > > > >> > > > > > > > > > > >> a
> > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue
> is
> > > that
> > > > > the
> > > > > > >> > > > fetch/commit
> > > > > > >> > > > > > > > > offsets
> > > > > > >> > > > > > > > > > > >> APIs
> > > > > > >> > > > > > > > > > > >> > do not
> > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > > > consumer
> > > > > > >> would
> > > > > > >> > > > reuse the
> > > > > > >> > > > > > > > > offsets
> > > > > > >> > > > > > > > > > > >> based
> > > > > > >> > > > > > > > > > > >> > on the
> > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> should
> > > > > update
> > > > > > >> those
> > > > > > >> > > > APIs
> > > > > > >> > > > > > as well
> > > > > > >> > > > > > > > > in
> > > > > > >> > > > > > > > > > > >> order
> > > > > > >> > > > > > > > > > > >> > to be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That
> > > would
> > > > > > >> strengthen
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > semantics of
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The
> KIP
> > > > > mentions
> > > > > > >> we will
> > > > > > >> > > > > > write the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > > > > >> implementation
> > > > > > >> > > > > > detail, do
> > > > > > >> > > > > > > > > we plan
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > have a
> > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we
> > > have a
> > > > > > >> scalability
> > > > > > >> > > > > > goal when
> > > > > > >> > > > > > > > > it
> > > > > > >> > > > > > > > > > > >> comes
> > > > > > >> > > > > > > > > > > >> > to how
> > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group
> protocol
> > > > can
> > > > > > >> support?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't
> have
> > > > > numbers
> > > > > > >> at the
> > > > > > >> > > > > > moment. The
> > > > > > >> > > > > > > > > > > >> protocol
> > > > > > >> > > > > > > > > > > >> > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members
> per
> > > > > group.
> > > > > > >> We will
> > > > > > >> > > > > > measure
> > > > > > >> > > > > > > > > this when
> > > > > > >> > > > > > > > > > > >> > we have
> > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> implementation.
> > > > Note
> > > > > > >> that we
> > > > > > >> > > > might
> > > > > > >> > > > > > have
> > > > > > >> > > > > > > > > other
> > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g.
> > > offset
> > > > > > >> commits).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > > > > consider
> > > > > > >> having
> > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > >> > > > > > > > > > > >> > instead
> > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > SubscribedTopicNames
> > > > > > >> in
> > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we
> have
> > > to
> > > > > > >> resolve the
> > > > > > >> > > > > > regex on
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic
> > > name?
> > > > > The
> > > > > > >> > > > difference
> > > > > > >> > > > > > is that
> > > > > > >> > > > > > > > > > > >> sending
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas
> > > sending
> > > > > the
> > > > > > >> topic
> > > > > > >> > > > names
> > > > > > >> > > > > > is less
> > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> recreation is
> > > > > easier
> > > > > > >> to
> > > > > > >> > > > handle
> > > > > > >> > > > > > if we
> > > > > > >> > > > > > > > > have
> > > > > > >> > > > > > > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea
> was to
> > > > > > >> consolidate the
> > > > > > >> > > > > > metadata
> > > > > > >> > > > > > > > > lookup
> > > > > > >> > > > > > > > > > > >> > on the
> > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I
> do
> > > > agree
> > > > > > >> with your
> > > > > > >> > > > > > point. As
> > > > > > >> > > > > > > > > a
> > > > > > >> > > > > > > > > > > >> second
> > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be
> > > better
> > > > > here
> > > > > > >> for the
> > > > > > >> > > > > > delete and
> > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose
> that we
> > > > may
> > > > > > >> allow
> > > > > > >> > > > users to
> > > > > > >> > > > > > > > > subscribe
> > > > > > >> > > > > > > > > > > >> with
> > > > > > >> > > > > > > > > > > >> > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> because
> > > > that
> > > > > is
> > > > > > >> the only
> > > > > > >> > > > > > way to be
> > > > > > >> > > > > > > > > > > >> really
> > > > > > >> > > > > > > > > > > >> > robust
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul
> 12,
> > > > > 2022 at
> > > > > > >> 1:38 PM
> > > > > > >> > > > > > David
> > > > > > >> > > > > > > > > Jacot <
> > > > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> Justine,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for
> > > your
> > > > > > >> comments.
> > > > > > >> > > > Please
> > > > > > >> > > > > > find
> > > > > > >> > > > > > > > > my
> > > > > > >> > > > > > > > > > > >> answers
> > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes,
> the new
> > > > > > >> protocol
> > > > > > >> > > > relies on
> > > > > > >> > > > > > > > > topic IDs
> > > > > > >> > > > > > > > > > > >> > with the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic
> names
> > > > based
> > > > > in
> > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using
> topic
> > > > > names
> > > > > > >> is the
> > > > > > >> > > > > > right call
> > > > > > >> > > > > > > > > here.
> > > > > > >> > > > > > > > > > > >> I
> > > > > > >> > > > > > > > > > > >> > need to
> > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little
> more.
> > > > > > >> Obviously, the
> > > > > > >> > > > > > KIP does
> > > > > > >> > > > > > > > > not
> > > > > > >> > > > > > > > > > > >> > change the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets
> RPCs
> > > to
> > > > > use
> > > > > > >> topic
> > > > > > >> > > > IDs.
> > > > > > >> > > > > > This
> > > > > > >> > > > > > > > > may be
> > > > > > >> > > > > > > > > > > >> > something
> > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include
> though
> > > > as
> > > > > it
> > > > > > >> would
> > > > > > >> > > > give
> > > > > > >> > > > > > better
> > > > > > >> > > > > > > > > > > >> overall
> > > > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're
> > > right.
> > > > I
> > > > > > >> think that
> > > > > > >> > > > I
> > > > > > >> > > > > > should
> > > > > > >> > > > > > > > > not
> > > > > > >> > > > > > > > > > > >> have
> > > > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I
> will
> > > > remove
> > > > > > >> it. We can
> > > > > > >> > > > > > use an
> > > > > > >> > > > > > > > > internal
> > > > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> developing the
> > > > > > >> feature.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both
> cluster
> > > > > types
> > > > > > >> will be
> > > > > > >> > > > > > > > > supported. The
> > > > > > >> > > > > > > > > > > >> > change is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only
> > > requirement
> > > > > is
> > > > > > >> that the
> > > > > > >> > > > > > cluster
> > > > > > >> > > > > > > > > uses
> > > > > > >> > > > > > > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > IDs.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon,
> Jul
> > > 11,
> > > > > 2022
> > > > > > >> at 9:53
> > > > > > >> > > > PM
> > > > > > >> > > > > > > > > Guozhang
> > > > > > >> > > > > > > > > > > >> Wang <
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi
> Ismael,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks
> for
> > > the
> > > > > > >> feedback.
> > > > > > >> > > > Here
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > > some
> > > > > > >> > > > > > > > > > > >> replies
> > > > > > >> > > > > > > > > > > >> > > > > inlined
> > > > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat,
> Jul
> > > 9,
> > > > > 2022
> > > > > > >> at
> > > > > > >> > > > 2:53 AM
> > > > > > >> > > > > > > > > Ismael Juma
> > > > > > >> > > > > > > > > > > >> <
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> Thanks for
> > > > the
> > > > > > >> KIP. This
> > > > > > >> > > > has
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> potential
> > > > > > >> > > > > > > > > > > >> > to be a
> > > > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> improvement. A
> > > few
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> initial
> > > > > > >> > > > questions/comments:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I
> think
> > > > > it's
> > > > > > >> > > > premature to
> > > > > > >> > > > > > talk
> > > > > > >> > > > > > > > > about
> > > > > > >> > > > > > > > > > > >> > target
> > > > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation
> and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> removal of
> > > > the
> > > > > > >> existing
> > > > > > >> > > > group
> > > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> > Unlike
> > > > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > > > > protocol
> > > > > > >> and hence
> > > > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how
> > > quickly
> > > > > > >> applications
> > > > > > >> > > > > > migrate
> > > > > > >> > > > > > > > > to the
> > > > > > >> > > > > > > > > > > >> new
> > > > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I
> agree
> > > > > with
> > > > > > >> you. I
> > > > > > >> > > > think
> > > > > > >> > > > > > we can
> > > > > > >> > > > > > > > > > > >> remove
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > `Compatibility,
> > > > > > >> > > > Deprecation,
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > Migration
> > > > > > >> > > > > > > > > > > >> > Plan` and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we
> will
> > > > > decide
> > > > > > >> in the
> > > > > > >> > > > > > future
> > > > > > >> > > > > > > > > about
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > we
> > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> protocol and
> > > > > > >> behaviors.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2.
> The KIP
> > > > > says we
> > > > > > >> > > > intend to
> > > > > > >> > > > > > > > > release this
> > > > > > >> > > > > > > > > > > >> > in 4.x,
> > > > > > >> > > > > > > > > > > >> > > > > >> but
> > > > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why.
> If we
> > > > > added
> > > > > > >> that as
> > > > > > >> > > > a
> > > > > > >> > > > > > way to
> > > > > > >> > > > > > > > > > > >> estimate
> > > > > > >> > > > > > > > > > > >> > when
> > > > > > >> > > > > > > > > > > >> > > > > we'd
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> group
> > > > > > >> protocol, I
> > > > > > >> > > > also
> > > > > > >> > > > > > suggest
> > > > > > >> > > > > > > > > > > >> removing
> > > > > > >> > > > > > > > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> > > that's
> > > > > not
> > > > > > >> > > > specifically
> > > > > > >> > > > > > > > > related to
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan,
> but
> > > it's
> > > > > more
> > > > > > >> for
> > > > > > >> > > > client
> > > > > > >> > > > > > > > > upgrades.
> > > > > > >> > > > > > > > > > > >> I.e.
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > implementation
> > > > > may
> > > > > > >> be done
> > > > > > >> > > > > > first,
> > > > > > >> > > > > > > > > and then
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only
> mark it
> > > > as
> > > > > > >> "released"
> > > > > > >> > > > by
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > time
> > > > > > >> > > > > > > > > > > >> clients
> > > > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that
> time,
> > > to
> > > > > > >> enable the
> > > > > > >> > > > > > feature the
> > > > > > >> > > > > > > > > > > >> clients
> > > > > > >> > > > > > > > > > > >> > need to
> > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode
> > > with
> > > > a
> > > > > > >> rolling
> > > > > > >> > > > bounce
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > then set
> > > > > > >> > > > > > > > > > > >> > the flag
> > > > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce,
> and
> > > > > hence
> > > > > > >> we feel
> > > > > > >> > > > it's
> > > > > > >> > > > > > > > > better to be
> > > > > > >> > > > > > > > > > > >> > released
> > > > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We
> need
> > > > to
> > > > > > >> flesh out
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > details of
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > migration
> > > > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're
> > > saying
> > > > > we
> > > > > > >> will
> > > > > > >> > > > support
> > > > > > >> > > > > > online
> > > > > > >> > > > > > > > > > > >> > migrations. Is
> > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> explain
> > > this
> > > > > in
> > > > > > >> detail.
> > > > > > >> > > > It
> > > > > > >> > > > > > could
> > > > > > >> > > > > > > > > also be
> > > > > > >> > > > > > > > > > > >> > done as a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> easier.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> think
> > > > > that's
> > > > > > >> the
> > > > > > >> > > > part we
> > > > > > >> > > > > > can
> > > > > > >> > > > > > > > > be more
> > > > > > >> > > > > > > > > > > >> > concrete
> > > > > > >> > > > > > > > > > > >> > > > > >> about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is
> > > > related
> > > > > to
> > > > > > >> your
> > > > > > >> > > > > > question 2)
> > > > > > >> > > > > > > > > above).
> > > > > > >> > > > > > > > > > > >> > We will
> > > > > > >> > > > > > > > > > > >> > > > > >> work
> > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> explicit in
> > > > > > >> parallel as we
> > > > > > >> > > > > > solicit
> > > > > > >> > > > > > > > > more
> > > > > > >> > > > > > > > > > > >> > feedback.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I
> am
> > > > happy
> > > > > > >> that we are
> > > > > > >> > > > > > pushing
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it
> > > seems
> > > > > like
> > > > > > >> there
> > > > > > >> > > > > > could be
> > > > > > >> > > > > > > > > some
> > > > > > >> > > > > > > > > > > >> tricky
> > > > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > > > > mechanism
> > > > > > >> for
> > > > > > >> > > > users to
> > > > > > >> > > > > > > > > detect that
> > > > > > >> > > > > > > > > > > >> > they
> > > > > > >> > > > > > > > > > > >> > > > > need
> > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > > > > switching
> > > > > > >> to the
> > > > > > >> > > > new
> > > > > > >> > > > > > > > > protocol?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> think
> > > we
> > > > > need
> > > > > > >> some
> > > > > > >> > > > > > tooling for
> > > > > > >> > > > > > > > > > > >> non-java
> > > > > > >> > > > > > > > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> "dry-run"
> > > the
> > > > > > >> client before
> > > > > > >> > > > > > > > > switching to
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific
> > > idea
> > > > on
> > > > > > >> top of my
> > > > > > >> > > > head
> > > > > > >> > > > > > > > > though,
> > > > > > >> > > > > > > > > > > >> maybe
> > > > > > >> > > > > > > > > > > >> > others
> > > > > > >> > > > > > > > > > > >> > > > > >> like
> > > > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in
> > > here?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5.
> Related
> > > > to
> > > > > the
> > > > > > >> last
> > > > > > >> > > > > > question,
> > > > > > >> > > > > > > > > will the
> > > > > > >> > > > > > > > > > > >> > Java
> > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick
> with
> > > > the
> > > > > > >> current
> > > > > > >> > > > regex
> > > > > > >> > > > > > > > > engine for
> > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it
> may be
> > > > > handy
> > > > > > >> to keep
> > > > > > >> > > > using
> > > > > > >> > > > > > > > > client
> > > > > > >> > > > > > > > > > > >> based
> > > > > > >> > > > > > > > > > > >> > regex
> > > > > > >> > > > > > > > > > > >> > > > > at
> > > > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> migrations
> > > > > simple
> > > > > > >> and
> > > > > > >> > > > then
> > > > > > >> > > > > > migrate
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > server based
> > > > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> Honestly I
> > > > have
> > > > > not
> > > > > > >> thought
> > > > > > >> > > > > > about
> > > > > > >> > > > > > > > > that for
> > > > > > >> > > > > > > > > > > >> > java
> > > > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that.
> What
> > > > kind
> > > > > of
> > > > > > >> > > > > > compatibility
> > > > > > >> > > > > > > > > issues do
> > > > > > >> > > > > > > > > > > >> > you have
> > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6.
> When we
> > > > say
> > > > > > >> that the
> > > > > > >> > > > group
> > > > > > >> > > > > > > > > coordinator
> > > > > > >> > > > > > > > > > > >> > will be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > > > > >> configurations and
> > > > > > >> > > > that
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> > configurations
> > > > > > >> > > > > > > > > > > >> > > > > will
> > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> is
> > > > > deleted.
> > > > > > >> Will a
> > > > > > >> > > > > > > > > transition to
> > > > > > >> > > > > > > > > > > >> DEAD
> > > > > > >> > > > > > > > > > > >> > > > > trigger
> > > > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > configurations?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes,
> since
> > > the
> > > > > DEAD
> > > > > > >> state
> > > > > > >> > > > is an
> > > > > > >> > > > > > > > > ending
> > > > > > >> > > > > > > > > > > >> state
> > > > > > >> > > > > > > > > > > >> > (we
> > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state
> when
> > > the
> > > > > > >> group is
> > > > > > >> > > > EMPTY
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > also all
> > > > > > >> > > > > > > > > > > >> of
> > > > > > >> > > > > > > > > > > >> > its
> > > > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's
> > > transited
> > > > > to
> > > > > > >> DEAD this
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> never
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7.
> Will
> > > the
> > > > > > >> choice to
> > > > > > >> > > > store
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > configs
> > > > > > >> > > > > > > > > > > >> in
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> harder to
> > > > > list all
> > > > > > >> > > > cluster
> > > > > > >> > > > > > configs
> > > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> their
> > > > > > >> > > > > > > > > > > >> > > > > values?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a
> > > good
> > > > > > >> question,
> > > > > > >> > > > and our
> > > > > > >> > > > > > > > > thoughts
> > > > > > >> > > > > > > > > > > >> are
> > > > > > >> > > > > > > > > > > >> > that
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > configurations"
> > > > > are
> > > > > > >> > > > overrides
> > > > > > >> > > > > > of the
> > > > > > >> > > > > > > > > > > >> > cluster-level
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> customized
> > > per
> > > > > > >> group so
> > > > > > >> > > > when an
> > > > > > >> > > > > > > > > admin list
> > > > > > >> > > > > > > > > > > >> > cluster
> > > > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list
> just
> > > the
> > > > > > >> cluster-level
> > > > > > >> > > > > > > > > "defaults", not
> > > > > > >> > > > > > > > > > > >> > showing
> > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > customizations.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How
> > > would
> > > > > > >> someone
> > > > > > >> > > > > > configure a
> > > > > > >> > > > > > > > > group
> > > > > > >> > > > > > > > > > > >> > before
> > > > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers?
> Have
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > > > considered
> > > > > > >> allowing
> > > > > > >> > > > the
> > > > > > >> > > > > > explicit
> > > > > > >> > > > > > > > > > > >> > creation of
> > > > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Alternatively,
> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> configs
> > > > could
> > > > > be
> > > > > > >> > > > decoupled
> > > > > > >> > > > > > from
> > > > > > >> > > > > > > > > the group
> > > > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The
> configs
> > > > can
> > > > > be
> > > > > > >> created
> > > > > > >> > > > > > before
> > > > > > >> > > > > > > > > the group
> > > > > > >> > > > > > > > > > > >> > itself
> > > > > > >> > > > > > > > > > > >> > > > > as
> > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of
> > > course,
> > > > > this
> > > > > > >> > > > requires
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> corresponding
> > > > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > > > > coordinator
> > > > > > >> based on
> > > > > > >> > > > the
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > id ---
> > > > > > >> > > > > > > > > > > >> > the only
> > > > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the
> > > group
> > > > > > >> itself is
> > > > > > >> > > > gone
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > also check
> > > > > > >> > > > > > > > > > > >> > if there
> > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> configuration
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > > > > related to
> > > > > > >> that
> > > > > > >> > > > group
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > delete
> > > > > > >> > > > > > > > > > > >> as
> > > > > > >> > > > > > > > > > > >> > well.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> Admittedly
> > > > this
> > > > > > >> indeed
> > > > > > >> > > > > > introduces an
> > > > > > >> > > > > > > > > > > >> > asymmetry on
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> lifecycles
> > > of
> > > > > the
> > > > > > >> config
> > > > > > >> > > > > > entities,
> > > > > > >> > > > > > > > > and we
> > > > > > >> > > > > > > > > > > >> > would like
> > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > > > > whether we
> > > > > > >> should
> > > > > > >> > > > aim
> > > > > > >> > > > > > for
> > > > > > >> > > > > > > > > symmetry
> > > > > > >> > > > > > > > > > > >> > i.e.
> > > > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs
> and
> > > > > hence
> > > > > > >> not
> > > > > > >> > > > delete
> > > > > > >> > > > > > them at
> > > > > > >> > > > > > > > > all
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> > > > explicit
> > > > > > >> deletion
> > > > > > >> > > > > > operations
> > > > > > >> > > > > > > > > by
> > > > > > >> > > > > > > > > > > >> > themselves.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9.
> Will
> > > the
> > > > > > >> > > > > > Consumer.subscribe
> > > > > > >> > > > > > > > > method for
> > > > > > >> > > > > > > > > > > >> > the Java
> > > > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > `java.util.regex.Pattern` of
> > > > > > >> > > > > > do we
> > > > > > >> > > > > > > > > have
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> we
> > > do
> > > > > not
> > > > > > >> need to
> > > > > > >> > > > > > introduce
> > > > > > >> > > > > > > > > an
> > > > > > >> > > > > > > > > > > >> > overload, but
> > > > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be
> some
> > > > > > >> compatibility
> > > > > > >> > > > > > issues
> > > > > > >> > > > > > > > > that we
> > > > > > >> > > > > > > > > > > >> may
> > > > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I
> > > agree
> > > > > with
> > > > > > >> Justine
> > > > > > >> > > > > > that we
> > > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> be
> > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> switch to
> > > > > > >> > > > > > IBP/metadata.version
> > > > > > >> > > > > > > > > from the
> > > > > > >> > > > > > > > > > > >> > feature
> > > > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can
> switch
> > > > the
> > > > > > >> default
> > > > > > >> > > > for
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > feature
> > > > > > >> > > > > > > > > > > >> flag
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > true
> > > > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > metadata.version
> > > > > > >> once we
> > > > > > >> > > > > > want to
> > > > > > >> > > > > > > > > make it
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11.
> Some of
> > > > the
> > > > > > >> protocol
> > > > > > >> > > > APIs
> > > > > > >> > > > > > don't
> > > > > > >> > > > > > > > > mention
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> required
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good
> to
> > > add
> > > > > that
> > > > > > >> for
> > > > > > >> > > > > > consistency.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we
> can
> > > > > > >> certainly do
> > > > > > >> > > > that.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12.
> It is
> > > a
> > > > > bit
> > > > > > >> odd that
> > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> though it
> > > > > seems
> > > > > > >> to do
> > > > > > >> > > > more
> > > > > > >> > > > > > than
> > > > > > >> > > > > > > > > reading.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had
> that
> > > > > thought
> > > > > > >> myself
> > > > > > >> > > > as
> > > > > > >> > > > > > well,
> > > > > > >> > > > > > > > > but in
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > end we
> > > > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > > > alternative:
> > > > > > >> adding
> > > > > > >> > > > > > Write
> > > > > > >> > > > > > > > > Group
> > > > > > >> > > > > > > > > > > >> seems
> > > > > > >> > > > > > > > > > > >> > an
> > > > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not
> have it
> > > > > > >> elsewhere (we
> > > > > > >> > > > only
> > > > > > >> > > > > > have
> > > > > > >> > > > > > > > > Read /
> > > > > > >> > > > > > > > > > > >> > Delete
> > > > > > >> > > > > > > > > > > >> > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far).
> Would
> > > > > like to
> > > > > > >> hear
> > > > > > >> > > > others
> > > > > > >> > > > > > > > > thoughts.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13.
> How is
> > > > > topic
> > > > > > >> > > > recreation
> > > > > > >> > > > > > > > > handled by
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> protocol?
> > > It
> > > > > > >> would be
> > > > > > >> > > > good
> > > > > > >> > > > > > to have
> > > > > > >> > > > > > > > > a
> > > > > > >> > > > > > > > > > > >> > section on
> > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean
> > > with
> > > > > regex
> > > > > > >> > > > > > subscription
> > > > > > >> > > > > > > > > right? Yes
> > > > > > >> > > > > > > > > > > >> > we can
> > > > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that,
> but
> > > > > basically
> > > > > > >> the
> > > > > > >> > > > idea
> > > > > > >> > > > > > is that
> > > > > > >> > > > > > > > > > > >> consumer
> > > > > > >> > > > > > > > > > > >> > would
> > > > > > >> > > > > > > > > > > >> > > > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > > > > protocol as
> > > > > > >> it's
> > > > > > >> > > > > > handled all
> > > > > > >> > > > > > > > > by the
> > > > > > >> > > > > > > > > > > >> > brokers.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14.
> The
> > > KIP
> > > > > > >> mentions we
> > > > > > >> > > > will
> > > > > > >> > > > > > write
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> new
> > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this
> is an
> > > > > > >> implementation
> > > > > > >> > > > > > detail,
> > > > > > >> > > > > > > > > do we
> > > > > > >> > > > > > > > > > > >> > plan to
> > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for
> it?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have
> not
> > > > > thought
> > > > > > >> about
> > > > > > >> > > > > > that. But
> > > > > > >> > > > > > > > > I think
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > answer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15.
> Do we
> > > > > have a
> > > > > > >> > > > scalability
> > > > > > >> > > > > > goal
> > > > > > >> > > > > > > > > when it
> > > > > > >> > > > > > > > > > > >> > comes to
> > > > > > >> > > > > > > > > > > >> > > > > >> how
> > > > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > > > > protocol can
> > > > > > >> > > > support?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> > > > group,
> > > > > I
> > > > > > >> think we
> > > > > > >> > > > > > should
> > > > > > >> > > > > > > > > shoot for
> > > > > > >> > > > > > > > > > > >> > 1000s of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> scalability
> > > > > goals
> > > > > > >> also
> > > > > > >> > > > depend
> > > > > > >> > > > > > on the
> > > > > > >> > > > > > > > > offset
> > > > > > >> > > > > > > > > > > >> > > > > management
> > > > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> capabilities
> > > > of
> > > > > the
> > > > > > >> > > > coordinator
> > > > > > >> > > > > > > > > which we
> > > > > > >> > > > > > > > > > > >> did
> > > > > > >> > > > > > > > > > > >> > not
> > > > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to
> > > give a
> > > > > > >> number that
> > > > > > >> > > > > > applies
> > > > > > >> > > > > > > > > > > >> > universally.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16.
> Did we
> > > > > > >> consider
> > > > > > >> > > > having
> > > > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> SubscribedTopicNames in
> > > > > > >> > > > > > > > > > > >> > > > > >>
> ConsumerGroupHeartbeatRequest?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since
> we
> > > > have
> > > > > to
> > > > > > >> resolve
> > > > > > >> > > > the
> > > > > > >> > > > > > regex
> > > > > > >> > > > > > > > > on the
> > > > > > >> > > > > > > > > > > >> > server,
> > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> topic
> > > > > name?
> > > > > > >> The
> > > > > > >> > > > > > difference is
> > > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > sending the
> > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> whereas
> > > > > sending
> > > > > > >> the topic
> > > > > > >> > > > > > names is
> > > > > > >> > > > > > > > > less
> > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> recreation
> > > > is
> > > > > > >> easier to
> > > > > > >> > > > > > handle if
> > > > > > >> > > > > > > > > we have
> > > > > > >> > > > > > > > > > > >> > topic
> > > > > > >> > > > > > > > > > > >> > > > > ids.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main
> > > > reason
> > > > > to
> > > > > > >> still
> > > > > > >> > > > let
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > clients
> > > > > > >> > > > > > > > > > > >> send
> > > > > > >> > > > > > > > > > > >> > names
> > > > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> reasoning of
> > > > > names
> > > > > > >> -> ids
> > > > > > >> > > > on
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > broker /
> > > > > > >> > > > > > > > > > > >> > admin
> > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> although we
> > > > > added
> > > > > > >> topic id
> > > > > > >> > > > in
> > > > > > >> > > > > > > > > KIP-516, we
> > > > > > >> > > > > > > > > > > >> > never
> > > > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > consumer/producers
> > > > > > >> > > > leveraging
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > related
> > > > > > >> > > > > > > > > > > >> > newer
> > > > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just
> set the
> > > > > topic
> > > > > > >> id as
> > > > > > >> > > > empty
> > > > > > >> > > > > > UUID.
> > > > > > >> > > > > > > > > We
> > > > > > >> > > > > > > > > > > >> want
> > > > > > >> > > > > > > > > > > >> > to keep
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > consumer/producer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be
> thin
> > > and
> > > > > only
> > > > > > >> > > > delegate
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > reasoning
> > > > > > >> > > > > > > > > > > >> on
> > > > > > >> > > > > > > > > > > >> > broker
> > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> Thanks,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On
> Wed,
> > > Jul
> > > > 6,
> > > > > > >> 2022 at
> > > > > > >> > > > 10:45
> > > > > > >> > > > > > AM
> > > > > > >> > > > > > > > > David
> > > > > > >> > > > > > > > > > > >> Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi
> all,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I
> would
> > > > > like to
> > > > > > >> start a
> > > > > > >> > > > > > > > > discussion
> > > > > > >> > > > > > > > > > > >> thread
> > > > > > >> > > > > > > > > > > >> > on
> > > > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > Generation
> > > > > of
> > > > > > >> the
> > > > > > >> > > > Consumer
> > > > > > >> > > > > > > > > Rebalance
> > > > > > >> > > > > > > > > > > >> > Protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> With
> > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to
> make
> > > > the
> > > > > > >> rebalance
> > > > > > >> > > > > > protocol
> > > > > > >> > > > > > > > > (for
> > > > > > >> > > > > > > > > > > >> > consumers)
> > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > scalable,
> > > > > > >> easier to
> > > > > > >> > > > > > implement for
> > > > > > >> > > > > > > > > > > >> > clients, and
> > > > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > operators.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The
> KIP
> > > is
> > > > > here:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> Please
> > > > take
> > > > > a
> > > > > > >> look and
> > > > > > >> > > > let
> > > > > > >> > > > > > me
> > > > > > >> > > > > > > > > know what
> > > > > > >> > > > > > > > > > > >> > you
> > > > > > >> > > > > > > > > > > >> > > > > think.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS:
> I
> > > will
> > > > > be
> > > > > > >> away from
> > > > > > >> > > > > > July
> > > > > > >> > > > > > > > > 18th to
> > > > > > >> > > > > > > > > > > >> > August 8th.
> > > > > > >> > > > > > > > > > > >> > > > > >> That
> > > > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of
> time
> > > to
> > > > > read
> > > > > > >> and
> > > > > > >> > > > digest
> > > > > > >> > > > > > this
> > > > > > >> > > > > > > > > long
> > > > > > >> > > > > > > > > > > >> KIP.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> --
> > > > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > --
> > > > > > >> > > > > -- Guozhang
> > > > > > >> > > >
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> > > --
> > > > > > >> > > -- Guozhang
> > > > > > >>
> > > > > > >
> > > > > > >
> > > > > > > --
> > > > > > > -- Guozhang
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > -- Guozhang
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > > >
> > >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

+1. I like this idea too --- even I'm a nit leaning towards piggy-backing
the prepare assignment with the HB, if the group agrees on keeping them
separate I'm also fine, and I think we should consider doing this to have
an extra round of catching up assignment and hence keeping the fetch
throughput during rebalance.

On Tue, Sep 27, 2022 at 6:21 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Luke,
>
> Thanks for your comments.
>
> Your suggestion makes sense. Jun's made the same one. We are also
> discussing whether we should directly return the group state in the
> response and drop the prepare assignment api. It seems that the
> current consensus is to keep it though. If that stays like this, I
> will apply your suggestion.
>
> Thanks,
> David
>
> On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com> wrote:
> >
> > Hi David,
> >
> > Thanks for your patience for answering all these questions!
> >
> > One more question from me:
> > 1. When the selected consumer is going to be in client-side assigning, we
> > will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering if we
> > could use a flag to indicate the assignment is required. The rationale is
> > when returning error, we lose a chance to catch up with the latest target
> > assignment or revoke partitions. Since we expected the record fetching is
> > not blocked by the rebalance/assignment computing, we might want to allow
> > the members to update the member assignment in each heartbeat.
> >
> > Luke
> >
> > On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson
> <ja...@confluent.io.invalid>
> > wrote:
> >
> > > Hey David,
> > >
> > > Thanks for the response.
> > >
> > > > Do you think that we could have a case where two members with the
> same
> > > static id are active?
> > >
> > > I was just thinking about the handoff when we spin up a new node
> before the
> > > old one completes shutdown. It's probably ok to fence the old member
> > > abruptly. I do recall we had a few issues with this handoff in the
> past. It
> > > was due to the fact that the replaced node might have been the leader.
> > > Since we've removed the leader notion, perhaps this kind of issue won't
> > > resurface.
> > >
> > > In regard to the RE2 library. I understand the intent to pass this
> through
> > > as a string without any validation so that we don't need the
> dependence on
> > > the client. I do wonder if that will be restrictive in practice. For
> > > example, if a regex changes on the client side, would we need to
> validate
> > > against the current assignment to see which partitions to continue
> > > fetching? Or would we instead invalidate the complete current
> assignment?
> > > Perhaps this is a rare case and we can do the latter. There could be
> more
> > > situations though.
> > >
> > > As an alternative, I wonder if we could define a minimalistic regex
> syntax
> > > (a subset of RE2 say) that could map more easily to regex support that
> > > already exists in clients? For example, in this scenario, we are
> matching
> > > topics, so we only need to be able to represent topic characters and
> not
> > > the whole smorgasbord of whitespace and unicode options. Would it make
> > > sense to have a reduced syntax which includes only standard topic
> > > characters, wildcards, character classes, and repetitions?
> > >
> > > Best,
> > > Jason
> > >
> > >
> > >
> > >
> > > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com>
> wrote:
> > >
> > > > Hi David, thanks for all the detailed explanations. I think they all
> make
> > > > sense. Just want to have a couple follow-ups here:
> > > >
> > > > > I don't really see the benefits here because server side assignors
> > > > don't have metadata at all. They only assign topic-partitions. They
> > > > are not supposed to generate metadata nor to receive metadata from
> the
> > > > members.
> > > >
> > > > Just to clarify I was asking about the `version` of the assignor
> (i.e. up
> > > > to what version that the client would support), and I do agree we
> would
> > > not
> > > > need metadata. What I have in mind is that, for some specific
> built-in
> > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> in a
> > > > newer version we would have a hierarchical rack ID string format,
> like
> > > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > > would still be in old format. In this case, the broker then needs to
> > > choose
> > > > the old versioned assignor. I'm probably making something up here for
> > > rack
> > > > aware assignors, but I'm wondering if in general such an
> "auto-downgrade"
> > > > behavior would be needed still for broker-side assignor, and if yes
> would
> > > > "version" still be useful.
> > > >
> > > > > Yeah, that's right. Within a rebalance, `onAssignment` is called
> once
> > > > when the member transitions to a new epoch. This one contains the
> full
> > > > metadata provided by the client side assignor. Then, `onAssignment`
> > > > can be called max N times where N is the number of partitions pending
> > > > revocation by other members. Let me try to clarify this in the KIP.
> > > >
> > > > Okay, my understanding is that the calling ordering of these
> callbacks
> > > > would be like the following:
> > > >
> > > > ----------------------------------------
> > > > onPartitionsRevoked();   // just once, since we do not really need
> > > > to revoke incrementally.
> > > >
> > > > onAssignment();    // the first call, with epoch incremented
> > > > onPartitionsAssigned();   // paired with the onAssignment
> > > >
> > > > onAssignment();              // the first onAssignment would bump up
> the
> > > > epoch, and the metadata reflected.
> > > > onPartitionsAssigned();   // each time we get an additional
> assignment,
> > > we
> > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > ...
> > > > onAssignment();
> > > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > > encoded
> > > > metadata would not change, only the incrementally added partitions be
> > > > reflected
> > > >
> > > > Is that the case?
> > > >
> > > > I'm wondering if we would still call onAssignment just once, that
> encodes
> > > > all the assignment for this rebalance, including all the partitions
> that
> > > > should be assigned to the member but not yet assigned since they
> have not
> > > > been revoked by others. In that case the call ordering would be:
> > > >
> > > > ----------------------------------------
> > > > onPartitionsRevoked();   // just once
> > > > onAssignment();    // just once, with epoch incremented, and metadata
> > > > encoded changed, the "assignment" field also reflect the final target
> > > > assignment
> > > > onPartitionsAssigned();   // multiple times, which represent
> > > incrementally
> > > > added partitions
> > > > ...
> > > > onPartitionsAssigned();
> > > >
> > > > The motivation from this is that, most users would only implement the
> > > > rebalance callback listeners and hence we'd definitely need to make
> sure
> > > > the semantics of that does not change much, and the time
> > > > `onPartitionsAssigned` indicate the time when the partitions are
> actually
> > > > assigned to it; while for assignors, the `onAssignment` is used to
> > > indicate
> > > > what decision is made regarding for this member, i.e. when the
> partitions
> > > > are decided to be given to it, but not necessarily meant that it has
> been
> > > > given, since that time should be determined by the time of
> > > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > > implementers would not need to reason which `onAssignment` would be
> the
> > > > last one for this epoch.
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > > Guozhang
> > > >
> > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> <djacot@confluent.io.invalid
> > > >
> > > > wrote:
> > > >
> > > > > Hi Guozhang,
> > > > >
> > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> byte
> > > > > field,
> > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> And
> > > the
> > > > > > rationale seems to be assuming that we should always be able to
> do
> > > the
> > > > > > assignment at the broker-side assignor without errors.
> Personally I
> > > > think
> > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > broker-side assignors, e.g. for some edge cases where some
> subscribed
> > > > > > topics are not recognized with the current broker's metadata.
> What do
> > > > you
> > > > > > think?
> > > > >
> > > > > Yeah, that seems reasonable. However, I wonder if it would be
> better
> > > > > to use an exception on the server side. This is what we usually do
> for
> > > > > server side plugins. On the client side, we use a field because the
> > > > > errors are not defined in advance.
> > > > >
> > > > > Your comment also makes me think about what we should do when the
> > > > > server side assignor fails. I suppose that we have to keep the
> current
> > > > > assignment until a new event occurs. For instance, in your example,
> > > > > the coordinator would have to trigger a rebalance when unrecognized
> > > > > topics are available. This would be part of the metadata
> monitoring.
> > > > >
> > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > reason/version/metadata compared with the broker-side
> GroupMember. I
> > > > > agree
> > > > > > that broker-side assignor would not need reason/metadata since
> they
> > > are
> > > > > > blackbox strings/bytes to the assignor, but what about version?
> E.g.
> > > is
> > > > > it
> > > > > > possible that we evolve our broker-side built-in assignor but
> the old
> > > > > > versioned clients would not be able to work with the new
> version, in
> > > > > which
> > > > > > case we need to let the broker being aware of this and upgrade
> its
> > > > > behavior
> > > > > > to cooperate with the clients?
> > > > >
> > > > > I don't really see the benefits here because server side assignors
> > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > are not supposed to generate metadata nor to receive metadata from
> the
> > > > > members.
> > > > >
> > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > instead
> > > > > of
> > > > > > including these three fields, what about just adding the
> "Metadata"
> > > > field
> > > > > > class which has these three fields? Also, there are two
> "Metadata"
> > > > > > currently in the APIs, the first is a class that encodes
> > > > > > reason/version/metadata, and the second is just the encoded
> metadata
> > > > > bytes.
> > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > which
> > > > > > then has a bytebuffer field Metadata, or instead naming the
> second
> > > > > > bytebuffer field as metadataBytes?
> > > > >
> > > > > That's a good point. Let me try to rationalize this interface
> based on
> > > > > your suggestions.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > > > wrote:
> > > > > >
> > > > > > Hello David,
> > > > > >
> > > > > > Just had a few more nit questions about the public APIs:
> > > > > >
> > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> byte
> > > > > field,
> > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> And
> > > the
> > > > > > rationale seems to be assuming that we should always be able to
> do
> > > the
> > > > > > assignment at the broker-side assignor without errors.
> Personally I
> > > > think
> > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > broker-side assignors, e.g. for some edge cases where some
> subscribed
> > > > > > topics are not recognized with the current broker's metadata.
> What do
> > > > you
> > > > > > think?
> > > > > >
> > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > reason/version/metadata compared with the broker-side
> GroupMember. I
> > > > > agree
> > > > > > that broker-side assignor would not need reason/metadata since
> they
> > > are
> > > > > > blackbox strings/bytes to the assignor, but what about version?
> E.g.
> > > is
> > > > > it
> > > > > > possible that we evolve our broker-side built-in assignor but
> the old
> > > > > > versioned clients would not be able to work with the new
> version, in
> > > > > which
> > > > > > case we need to let the broker being aware of this and upgrade
> its
> > > > > behavior
> > > > > > to cooperate with the clients?
> > > > > >
> > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > instead
> > > > > of
> > > > > > including these three fields, what about just adding the
> "Metadata"
> > > > field
> > > > > > class which has these three fields? Also, there are two
> "Metadata"
> > > > > > currently in the APIs, the first is a class that encodes
> > > > > > reason/version/metadata, and the second is just the encoded
> metadata
> > > > > bytes.
> > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > which
> > > > > > then has a bytebuffer field Metadata, or instead naming the
> second
> > > > > > bytebuffer field as metadataBytes?
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> wangguoz@gmail.com>
> > > > > wrote:
> > > > > >
> > > > > > > Hello David,
> > > > > > >
> > > > > > > Thanks for bringing this question up. I think the main
> benefits as
> > > > you
> > > > > > > listed is 2) above if it stays; just to clarify, we would only
> be
> > > > able
> > > > > to
> > > > > > > save one round trip if the rebalance is still triggered by the
> > > > broker;
> > > > > if
> > > > > > > the rebalance is triggered by the client then the
> num.round.trips
> > > are
> > > > > the
> > > > > > > same:
> > > > > > >
> > > > > > > 1) With GroupPrepareAssignment:
> > > > > > >
> > > > > > > T0: client decides to do a new assignment, suppose it has
> already
> > > > sent
> > > > > a
> > > > > > > HB and hence has to wait for it to return first since only one
> > > > request
> > > > > /
> > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > GroupPrepareAssignment request.
> > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > T3: it calculates the new assignment, and sends a
> > > > > GroupInstallAssignment
> > > > > > > request.
> > > > > > >
> > > > > > > In total, two round trips.
> > > > > > >
> > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > >
> > > > > > > T0: client decides to do a new assignment, suppose it has
> already
> > > > sent
> > > > > a
> > > > > > > HB and hence has to wait for it to return first since only one
> > > > request
> > > > > /
> > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > T1: client receives the HB response, and then sends the new HB
> > > > request
> > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > T2: the HB response with the optional member metadata map is
> > > > returned.
> > > > > > > T3: it calculates the new assignment, and sends a
> > > > > GroupInstallAssignment
> > > > > > > request.
> > > > > > >
> > > > > > > In total, two round trips as well.
> > > > > > >
> > > > > > > -----------------------------
> > > > > > >
> > > > > > > So to complete the full picture here, we'd need to modify both
> HB
> > > > > request
> > > > > > > and response so that the client can also indicate a new
> rebalance
> > > via
> > > > > the
> > > > > > > HB request as well, right?
> > > > > > >
> > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > additional
> > > > > fields
> > > > > > > encoding the full member (subscription) metadata and topic
> metadata
> > > > as
> > > > > > > optional fields.
> > > > > > >
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > >> Hi all,
> > > > > > >>
> > > > > > >> During an offline conversation, someone asked why we need the
> > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > > > instead
> > > > > > >> provide the group state in the heartbeat response. This has a
> few
> > > > > > >> advantages: 1) it does not require using a special error code
> to
> > > > > > >> signal that a new assignment is required as the signal would
> be
> > > the
> > > > > > >> provided group state; 2) it removes one round trip when a
> client
> > > > side
> > > > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > > > >> response's definition quite large. I recall that I went with
> the
> > > > > > >> current approach due to this.
> > > > > > >>
> > > > > > >> Providing the group state in the heartbeat response is
> appealing.
> > > > What
> > > > > > >> do you guys think?
> > > > > > >>
> > > > > > >> Best,
> > > > > > >> David
> > > > > > >>
> > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> djacot@confluent.io>
> > > > > wrote:
> > > > > > >> >
> > > > > > >> > Hi Guozhang,
> > > > > > >> >
> > > > > > >> > 1. I have added a reference to the relevant chapter instead
> of
> > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > >> >
> > > > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> > > about
> > > > > when
> > > > > > >> > a rebalance should be triggered for the non-upgraded members
> > > using
> > > > > the
> > > > > > >> > old protocol. The section mentions that a rebalance must be
> > > > > triggered
> > > > > > >> > when a new assignment is installed. This implies that the
> group
> > > > > epoch
> > > > > > >> > was updated either by a native member or a non-upgraded
> member.
> > > > For
> > > > > > >> > the latter, the JoinGroup request would be the trigger. I
> have
> > > > > added a
> > > > > > >> > reference to the relevant chapter in the "JoinGroup
> Handling"
> > > > > section
> > > > > > >> > as well. Does that make sense?
> > > > > > >> >
> > > > > > >> > Thanks,
> > > > > > >> > David
> > > > > > >> >
> > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > wangguoz@gmail.com
> > > > >
> > > > > > >> wrote:
> > > > > > >> > >
> > > > > > >> > > Hello David,
> > > > > > >> > >
> > > > > > >> > > Alright I think that's sufficient. Just to make that
> clear in
> > > > the
> > > > > doc,
> > > > > > >> > > could we update:
> > > > > > >> > >
> > > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > > coordinator
> > > > > > >> will
> > > > > > >> > > trigger rebalance based on the HB's member metadata /
> reason?
> > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > described
> > > > > in
> > > > > > >> "Group
> > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> > > Guozhang
> > > > > > >> > >
> > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > wrote:
> > > > > > >> > >
> > > > > > >> > > > Hi Guozhang,
> > > > > > >> > > >
> > > > > > >> > > > I thought that the assignor will always be consulted
> when
> > > the
> > > > > next
> > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > > heartbeat.
> > > > > > >> This
> > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > rebalance
> > > > by
> > > > > > >> > > > setting the reason to a non-zero value or by changing
> the
> > > > > bytes. Do
> > > > > > >> > > > you think that this is not sufficient? Are you
> concerned by
> > > > the
> > > > > > >> delay?
> > > > > > >> > > >
> > > > > > >> > > > Best,
> > > > > > >> > > > David
> > > > > > >> > > >
> > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > >> wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > Hello David,
> > > > > > >> > > > >
> > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > >> > > > >
> > > > > > >> > > > > ```
> > > > > > >> > > > > In this case, a new assignment is triggered by the
> client
> > > > side
> > > > > > >> > > > > assignor. When constructing the HB, the consumer will
> > > always
> > > > > > >> consult
> > > > > > >> > > > > the client side assignor and propagate the
> information to
> > > > the
> > > > > > >> group
> > > > > > >> > > > > coordinator. In other words, we don't expect users to
> call
> > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > >> > > > > ```
> > > > > > >> > > > >
> > > > > > >> > > > > As I looked at the current PartitionAssignor's
> interface,
> > > we
> > > > > > >> actually do
> > > > > > >> > > > > not have a way yet to instruct how to construct the
> next
> > > HB
> > > > > > >> request, e.g.
> > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> with a
> > > > new
> > > > > > >> assignment,
> > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > PartitionAssignor
> > > > > to
> > > > > > >> indicate
> > > > > > >> > > > > the next HB telling broker about so. WDYT about adding
> > > such
> > > > an
> > > > > > >> API on the
> > > > > > >> > > > > PartitionAssignor?
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > Guozhang
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > > > wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > > Hi Jun,
> > > > > > >> > > > > >
> > > > > > >> > > > > > I have updated the KIP to include your feedback. I
> have
> > > > also
> > > > > > >> tried to
> > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > >> > > > > >
> > > > > > >> > > > > > Best,
> > > > > > >> > > > > > David
> > > > > > >> > > > > >
> > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > djacot@confluent.io
> > > > > > >> >
> > > > > > >> > > > wrote:
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Hi Jun,
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> answering
> > > your
> > > > > > >> questions
> > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> of
> > > the
> > > > > KIP
> > > > > > >> seem to
> > > > > > >> > > > be
> > > > > > >> > > > > > fewer
> > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> support of
> > > > > > >> wildcard. A few
> > > > > > >> > > > > > > > comments below.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > I would also add that the KIP removes the global
> sync
> > > > > barrier
> > > > > > >> in the
> > > > > > >> > > > > > > protocol which is essential to improve group
> stability
> > > > and
> > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> client by
> > > > > moving
> > > > > > >> most of
> > > > > > >> > > > > > > the logic to the server side.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> to
> > > > > support
> > > > > > >> rolling
> > > > > > >> > > > > > changing
> > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > assignor
> > > > > used
> > > > > > >> by a
> > > > > > >> > > > > > > majority of the members. This allows the group to
> move
> > > > > from
> > > > > > >> one
> > > > > > >> > > > > > > assignor to another by a roll. This is explained
> in
> > > the
> > > > > > >> Assignor
> > > > > > >> > > > > > > Selection chapter.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> it's
> > > > > > >> required in
> > > > > > >> > > > every
> > > > > > >> > > > > > > > request or the scenarios when it needs to be
> filled?
> > > > For
> > > > > > >> example,
> > > > > > >> > > > it's
> > > > > > >> > > > > > not
> > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> filled.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > The client is expected to set those fields in
> case of
> > > a
> > > > > > >> connection
> > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> changed
> > > > since
> > > > > > >> the last
> > > > > > >> > > > > > > HB. The server populates those fields as long as
> the
> > > > > member
> > > > > > >> is not
> > > > > > >> > > > > > > fully reconciled - the member should acknowledge
> that
> > > it
> > > > > has
> > > > > > >> the
> > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> this in
> > > > the
> > > > > KIP.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > affinity
> > > > > > >> between the
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > and the broker is only considered during
> fetching,
> > > but
> > > > > not
> > > > > > >> during
> > > > > > >> > > > > > assigning
> > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > assignment
> > > > > is
> > > > > > >> made,
> > > > > > >> > > > there
> > > > > > >> > > > > > is
> > > > > > >> > > > > > > > no opportunity for read affinity because no
> replicas
> > > > of
> > > > > > >> assigned
> > > > > > >> > > > > > partitions
> > > > > > >> > > > > > > > are close to the member. I am wondering if we
> should
> > > > use
> > > > > > >> this
> > > > > > >> > > > > > opportunity
> > > > > > >> > > > > > > > to address this by including rack in
> GroupMember.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's an interesting idea. I don't see any issue
> with
> > > > > adding
> > > > > > >> the
> > > > > > >> > > > rack
> > > > > > >> > > > > > > to the members. I will do so.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> know
> > > how
> > > > > busy
> > > > > > >> a group
> > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> it
> > > > seems
> > > > > > >> that we
> > > > > > >> > > > could
> > > > > > >> > > > > > add
> > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> the
> > > > event
> > > > > > >> loop is
> > > > > > >> > > > doing
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > actual work.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> failover
> > > > > > >> handling? For
> > > > > > >> > > > > > example,
> > > > > > >> > > > > > > > does it need to trigger the check if any group
> with
> > > > the
> > > > > > >> wildcard
> > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sure. When the new group coordinator takes over,
> it
> > > has
> > > > > to:
> > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > >> > > > > > > * Trigger a new assignment if a client side
> assignor
> > > is
> > > > > used.
> > > > > > >> We
> > > > > > >> > > > don't
> > > > > > >> > > > > > > store the information about the member selected
> to run
> > > > the
> > > > > > >> assignment
> > > > > > >> > > > > > > so we have to start a new one.
> > > > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > > > >> subscriptions, and
> > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> document
> > > > what
> > > > > > >> the epoch
> > > > > > >> > > > > > field
> > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > >> ConsumerGroupMetadataValue
> > > > > > >> > > > > > reflect
> > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> latest
> > > > > group
> > > > > > >> epoch.
> > > > > > >> > > > > > > When the group state is updated, the group epoch
> is
> > > > > bumped so
> > > > > > >> we use
> > > > > > >> > > > > > > that one for all the change records related to the
> > > > update.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > following
> > > > > > >> > > > invariants
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> a
> > > > member
> > > > > > >> not to
> > > > > > >> > > > get any
> > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's right. Here I meant that the members
> provided
> > > by
> > > > > the
> > > > > > >> assignor
> > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > assignor
> > > > > can
> > > > > > >> not make
> > > > > > >> > > > > > > up new member ids.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> epoch
> > > > > equals to
> > > > > > >> 0":
> > > > > > >> > > > When
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> used?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > A member is expected to abandon all its
> partitions and
> > > > > > >> rejoins when
> > > > > > >> > > > it
> > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this
> case,
> > > > the
> > > > > > >> group
> > > > > > >> > > > > > > coordinator will have removed the member from the
> > > group.
> > > > > The
> > > > > > >> member
> > > > > > >> > > > > > > can rejoin the group with the same member id but
> with
> > > 0
> > > > as
> > > > > > >> epoch. Let
> > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> to
> > > > > trigger a
> > > > > > >> > > > > > reassignment
> > > > > > >> > > > > > > > by either providing a non-zero reason or by
> updating
> > > > the
> > > > > > >> assignor
> > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> with
> > > the
> > > > > > >> deprecation
> > > > > > >> > > > of
> > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > In this case, a new assignment is triggered by the
> > > > client
> > > > > side
> > > > > > >> > > > > > > assignor. When constructing the HB, the consumer
> will
> > > > > always
> > > > > > >> consult
> > > > > > >> > > > > > > the client side assignor and propagate the
> information
> > > > to
> > > > > the
> > > > > > >> group
> > > > > > >> > > > > > > coordinator. In other words, we don't expect
> users to
> > > > call
> > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > section
> > > > > > >> seems to
> > > > > > >> > > > have
> > > > > > >> > > > > > > > multiple typos.
> > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > immediately
> > > > > > >> gets into
> > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > incorrect.
> > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> seems
> > > to
> > > > > get
> > > > > > >> into
> > > > > > >> > > > > > epoch=3,
> > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> has A -
> > > > > > >> epoch=2,
> > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> the
> > > > > upgrade
> > > > > > >> from
> > > > > > >> > > > any
> > > > > > >> > > > > > old
> > > > > > >> > > > > > > > version to new one?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > We will support upgrading from the consumer
> protocol
> > > > > version
> > > > > > >> 3,
> > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented
> yet
> > > so
> > > > > the
> > > > > > >> earliest
> > > > > > >> > > > > > > version is unknown at the moment. This is
> explained in
> > > > the
> > > > > > >> migration
> > > > > > >> > > > > > > plan chapter.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> update the
> > > > KIP
> > > > > > >> based on
> > > > > > >> > > > it
> > > > > > >> > > > > > > next week.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Best,
> > > > > > >> > > > > > > David
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > >> <ju...@confluent.io.invalid>
> > > > > > >> > > > wrote:
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Hi, David,
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> of
> > > the
> > > > > KIP
> > > > > > >> seem to
> > > > > > >> > > > be
> > > > > > >> > > > > > fewer
> > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> support of
> > > > > > >> wildcard. A few
> > > > > > >> > > > > > > > comments below.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> to
> > > > > support
> > > > > > >> rolling
> > > > > > >> > > > > > changing
> > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> it's
> > > > > > >> required in
> > > > > > >> > > > every
> > > > > > >> > > > > > > > request or the scenarios when it needs to be
> filled?
> > > > For
> > > > > > >> example,
> > > > > > >> > > > it's
> > > > > > >> > > > > > not
> > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> filled.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > affinity
> > > > > > >> between the
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > and the broker is only considered during
> fetching,
> > > but
> > > > > not
> > > > > > >> during
> > > > > > >> > > > > > assigning
> > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > assignment
> > > > > is
> > > > > > >> made,
> > > > > > >> > > > there
> > > > > > >> > > > > > is
> > > > > > >> > > > > > > > no opportunity for read affinity because no
> replicas
> > > > of
> > > > > > >> assigned
> > > > > > >> > > > > > partitions
> > > > > > >> > > > > > > > are close to the member. I am wondering if we
> should
> > > > use
> > > > > > >> this
> > > > > > >> > > > > > opportunity
> > > > > > >> > > > > > > > to address this by including rack in
> GroupMember.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> know
> > > how
> > > > > busy
> > > > > > >> a group
> > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> it
> > > > seems
> > > > > > >> that we
> > > > > > >> > > > could
> > > > > > >> > > > > > add
> > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> the
> > > > event
> > > > > > >> loop is
> > > > > > >> > > > doing
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > actual work.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> failover
> > > > > > >> handling? For
> > > > > > >> > > > > > example,
> > > > > > >> > > > > > > > does it need to trigger the check if any group
> with
> > > > the
> > > > > > >> wildcard
> > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> document
> > > > what
> > > > > > >> the epoch
> > > > > > >> > > > > > field
> > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > >> ConsumerGroupMetadataValue
> > > > > > >> > > > > > reflect
> > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > following
> > > > > > >> > > > invariants
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> a
> > > > member
> > > > > > >> not to
> > > > > > >> > > > get any
> > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> epoch
> > > > > equals to
> > > > > > >> 0":
> > > > > > >> > > > When
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> used?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> to
> > > > > trigger a
> > > > > > >> > > > > > reassignment
> > > > > > >> > > > > > > > by either providing a non-zero reason or by
> updating
> > > > the
> > > > > > >> assignor
> > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> with
> > > the
> > > > > > >> deprecation
> > > > > > >> > > > of
> > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > section
> > > > > > >> seems to
> > > > > > >> > > > have
> > > > > > >> > > > > > > > multiple typos.
> > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > immediately
> > > > > > >> gets into
> > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > incorrect.
> > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> seems
> > > to
> > > > > get
> > > > > > >> into
> > > > > > >> > > > > > epoch=3,
> > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> has A -
> > > > > > >> epoch=2,
> > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> the
> > > > > upgrade
> > > > > > >> from
> > > > > > >> > > > any
> > > > > > >> > > > > > old
> > > > > > >> > > > > > > > version to new one?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Thanks,
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Jun
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > wrote:
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > Hi all,
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > > coordinator
> > > > > > >> in
> > > > > > >> > > > Java. I
> > > > > > >> > > > > > > > > discussed this offline with a few folks and
> folks
> > > > are
> > > > > > >> concerned
> > > > > > >> > > > that
> > > > > > >> > > > > > > > > we could introduce many regressions in the old
> > > > > protocol
> > > > > > >> if we do
> > > > > > >> > > > so.
> > > > > > >> > > > > > > > > Therefore, I am going to remove this statement
> > > from
> > > > > the
> > > > > > >> KIP. It
> > > > > > >> > > > is an
> > > > > > >> > > > > > > > > implementation detail after all so it does not
> > > have
> > > > > to be
> > > > > > >> > > > decided at
> > > > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > > > refactor the
> > > > > > >> > > > current
> > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > Cheers,
> > > > > > >> > > > > > > > > David
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > > > >> djacot@confluent.io
> > > > > > >> > > > >
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> "Empty,
> > > > > assigning,
> > > > > > >> > > > > > reconciling,
> > > > > > >> > > > > > > > > stable,
> > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > section,
> > > > > > >> right?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > This sentence does not refer to those group
> > > states
> > > > > but
> > > > > > >> rather
> > > > > > >> > > > to a
> > > > > > >> > > > > > > > > > state machine replication (SMR). This
> refers to
> > > > the
> > > > > > >> entire
> > > > > > >> > > > state of
> > > > > > >> > > > > > > > > > group coordinator which is replicated via
> the
> > > log
> > > > > > >> layer. I will
> > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> is
> > > > > modelled
> > > > > > >> as an
> > > > > > >> > > > event
> > > > > > >> > > > > > > > > loop"?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> the new
> > > > > quorum
> > > > > > >> > > > > > controller. We
> > > > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > > > >> __consumer_offsets
> > > > > > >> > > > > > > > > > partition is assigned to a unique thread
> and all
> > > > the
> > > > > > >> events
> > > > > > >> > > > (e.g.
> > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by
> this
> > > > > > >> thread. This
> > > > > > >> > > > > > simplify
> > > > > > >> > > > > > > > > > concurrency and will enable us to do
> simulation
> > > > > testing
> > > > > > >> for the
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > > coordinator.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > >> *__consumer_offsets*
> > > > > > >> > > > > > > > > partitions?
> > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> owned
> > > > by
> > > > > a
> > > > > > >> group
> > > > > > >> > > > > > coordinator?
> > > > > > >> > > > > > > > > For
> > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > consumer
> > > > > > >> groups, and
> > > > > > >> > > > > > both
> > > > > > >> > > > > > > > > exist in
> > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> state
> > > > > machine
> > > > > > >> for it,
> > > > > > >> > > > or
> > > > > > >> > > > > > 2?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > > think.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.4. I know the
> "*group.coordinator.threads"
> > > *is
> > > > > the
> > > > > > >> number
> > > > > > >> > > > of
> > > > > > >> > > > > > threads
> > > > > > >> > > > > > > > > used
> > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> wondering
> > > if
> > > > > the
> > > > > > >> purpose
> > > > > > >> > > > of
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > threads
> > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > group
> > > > > (or
> > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> why
> > > > > should we
> > > > > > >> need
> > > > > > >> > > > > > > > > multi-threads
> > > > > > >> > > > > > > > > > > here?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> shard
> > > > the
> > > > > > >> > > > processing as
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> does
> > > the
> > > > > > >> default
> > > > > > >> > > > session
> > > > > > >> > > > > > > > > timeout not
> > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > thought
> > > > > the
> > > > > > >> min/max
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> it,
> > > > no?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> 30s The
> > > > > > >> timeout to
> > > > > > >> > > > detect
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > protocol.
> > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> int 45s
> > > > The
> > > > > > >> minimum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> int 60s
> > > > The
> > > > > > >> maximum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> session
> > > > > timeout
> > > > > > >> should
> > > > > > >> > > > be 45s
> > > > > > >> > > > > > > > > > (the current default).
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > [range,
> > > > > > >> uniform],
> > > > > > >> > > > > > which means
> > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> like to
> > > > > know
> > > > > > >> why not
> > > > > > >> > > > > > uniform
> > > > > > >> > > > > > > > > one? I
> > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > assignor
> > > > > > >> (former
> > > > > > >> > > > sticky
> > > > > > >> > > > > > > > > assinor)
> > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > reason
> > > > > we
> > > > > > >> choose
> > > > > > >> > > > range
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > as default?
> > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> uniform
> > > The
> > > > > > >> server side
> > > > > > >> > > > > > assignors.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > The order on the server side has no
> influence
> > > > > because
> > > > > > >> the
> > > > > > >> > > > client
> > > > > > >> > > > > > must
> > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> There
> > > is
> > > > no
> > > > > > >> default
> > > > > > >> > > > in the
> > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > specified
> > > > > by
> > > > > > >> the
> > > > > > >> > > > client,
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > request is rejected. The default client
> value
> > > for
> > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I
> hope
> > > > > that my
> > > > > > >> > > > answers
> > > > > > >> > > > > > help
> > > > > > >> > > > > > > > > > to clarify things. I will update the KIP as
> well
> > > > > based
> > > > > > >> on your
> > > > > > >> > > > > > > > > > feedback.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Cheers,
> > > > > > >> > > > > > > > > > David
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > > > >> showuon@gmail.com>
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Hi David,
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > mentioned:
> > > > > > >> > > > > > > > > > > > The new group coordinator will have a
> state
> > > > > machine
> > > > > > >> per
> > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> each
> > > > state
> > > > > > >> machine is
> > > > > > >> > > > > > modelled
> > > > > > >> > > > > > > > > as an
> > > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > > executed
> > > > > in
> > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> "Empty,
> > > > > assigning,
> > > > > > >> > > > > > reconciling,
> > > > > > >> > > > > > > > > stable,
> > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > section,
> > > > > > >> right?
> > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> is
> > > > > modelled
> > > > > > >> as an
> > > > > > >> > > > event
> > > > > > >> > > > > > > > > loop"?
> > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > >> *__consumer_offsets*
> > > > > > >> > > > > > > > > partitions?
> > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> owned
> > > > by
> > > > > a
> > > > > > >> group
> > > > > > >> > > > > > coordinator?
> > > > > > >> > > > > > > > > For
> > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > consumer
> > > > > > >> groups, and
> > > > > > >> > > > > > both
> > > > > > >> > > > > > > > > exist in
> > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> state
> > > > > machine
> > > > > > >> for it,
> > > > > > >> > > > or
> > > > > > >> > > > > > 2?
> > > > > > >> > > > > > > > > > > 1.4. I know the
> "*group.coordinator.threads"
> > > *is
> > > > > the
> > > > > > >> number
> > > > > > >> > > > of
> > > > > > >> > > > > > threads
> > > > > > >> > > > > > > > > used
> > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> wondering
> > > if
> > > > > the
> > > > > > >> purpose
> > > > > > >> > > > of
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > threads
> > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > group
> > > > > (or
> > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> why
> > > > > should we
> > > > > > >> need
> > > > > > >> > > > > > > > > multi-threads
> > > > > > >> > > > > > > > > > > here?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2. For the default value in the new
> configs:
> > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> does
> > > the
> > > > > > >> default
> > > > > > >> > > > session
> > > > > > >> > > > > > > > > timeout not
> > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > thought
> > > > > the
> > > > > > >> min/max
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> it,
> > > > no?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> 30s The
> > > > > > >> timeout to
> > > > > > >> > > > detect
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > protocol.
> > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> int 45s
> > > > The
> > > > > > >> minimum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> int 60s
> > > > The
> > > > > > >> maximum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > [range,
> > > > > > >> uniform],
> > > > > > >> > > > > > which means
> > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> like to
> > > > > know
> > > > > > >> why not
> > > > > > >> > > > > > uniform
> > > > > > >> > > > > > > > > one? I
> > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > assignor
> > > > > > >> (former
> > > > > > >> > > > sticky
> > > > > > >> > > > > > > > > assinor)
> > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > reason
> > > > > we
> > > > > > >> choose
> > > > > > >> > > > range
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > as default?
> > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> uniform
> > > The
> > > > > > >> server side
> > > > > > >> > > > > > assignors.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Thank you.
> > > > > > >> > > > > > > > > > > Luke
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen
> <
> > > > > > >> showuon@gmail.com
> > > > > > >> > > > >
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > > > >> integrating with
> > > > > > >> > > > > > KIP-848,
> > > > > > >> > > > > > > > > but I
> > > > > > >> > > > > > > > > > > > think we should have a separate
> discussion
> > > > > thread
> > > > > > >> for the
> > > > > > >> > > > Kafka
> > > > > > >> > > > > > > > > Connect
> > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > > > Consumer
> > > > > > >> Rebalance
> > > > > > >> > > > > > Protocol
> > > > > > >> > > > > > > > > [1],
> > > > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > > > consumer
> > > > > > >> rebalance
> > > > > > >> > > > > > protocol,
> > > > > > >> > > > > > > > > WDYT?
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > [1]
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > >> > > > > > > > > > > > Luke
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> feedback.
> > > > > Looks
> > > > > > >> like
> > > > > > >> > > > there's
> > > > > > >> > > > > > > > > agreement on
> > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> would
> > > > > revisit
> > > > > > >> the doc
> > > > > > >> > > > and
> > > > > > >> > > > > > see
> > > > > > >> > > > > > > > > what
> > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David
> Jacot
> > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> document.
> > > > > That's
> > > > > > >> really
> > > > > > >> > > > > > helpful. I
> > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> Connect
> > > > > and the
> > > > > > >> > > > Consumer
> > > > > > >> > > > > > could
> > > > > > >> > > > > > > > > share
> > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> main
> > > > > > >> difference is
> > > > > > >> > > > that
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > Consumer
> > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > > whereas
> > > > > > >> Connect
> > > > > > >> > > > assigns
> > > > > > >> > > > > > tasks
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> forward:
> > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > > > support
> > > > > > >> different
> > > > > > >> > > > > > resource
> > > > > > >> > > > > > > > > types
> > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> Connect.
> > > > The
> > > > > > >> dedicated
> > > > > > >> > > > APIs
> > > > > > >> > > > > > > > > would be
> > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> different on
> > > > the
> > > > > > >> > > > > > content/resources and
> > > > > > >> > > > > > > > > > > >> > they would rely on the same engine
> on the
> > > > > > >> coordinator
> > > > > > >> > > > side.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > I personally lean towards 2) because
> I am
> > > > > not a
> > > > > > >> fan of
> > > > > > >> > > > > > > > > overcharging
> > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> That
> > > > being
> > > > > > >> said, I am
> > > > > > >> > > > not
> > > > > > >> > > > > > > > > opposed to
> > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to
> do
> > > it.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> discuss
> > > it
> > > > > here
> > > > > > >> for now
> > > > > > >> > > > in
> > > > > > >> > > > > > order
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible
> with
> > > > what
> > > > > we
> > > > > > >> will do
> > > > > > >> > > > for
> > > > > > >> > > > > > > > > Connect in
> > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > >> > > > > > > > > > > >> > David
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> > > Jacot
> > > > <
> > > > > > >> > > > > > djacot@confluent.io>
> > > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > > > through
> > > > > and
> > > > > > >> address
> > > > > > >> > > > > > your
> > > > > > >> > > > > > > > > comments
> > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > > > feedback.
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > >> > > > > > > > > > > >> > > David
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> Gregory
> > > > > Harris
> > > > > > >> <
> > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> wonderful to
> > > > see
> > > > > > >> > > > cooperative
> > > > > > >> > > > > > > > > rebalancing
> > > > > > >> > > > > > > > > > > >> > making it
> > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > > section
> > > > > > >> describes how
> > > > > > >> > > > > > member
> > > > > > >> > > > > > > > > epoch
> > > > > > >> > > > > > > > > > > >> > should
> > > > > > >> > > > > > > > > > > >> > > > advance in step with the group
> epoch
> > > > and
> > > > > > >> assignment
> > > > > > >> > > > > > epoch
> > > > > > >> > > > > > > > > values. I
> > > > > > >> > > > > > > > > > > >> > think
> > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > > > reasons
> > > > > > >> described
> > > > > > >> > > > in
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > KIP. When
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> assigning
> > > > > > >> partitions to a
> > > > > > >> > > > > > worker,
> > > > > > >> > > > > > > > > what
> > > > > > >> > > > > > > > > > > >> member
> > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > assignment
> > > > > use?
> > > > > > >> Are
> > > > > > >> > > > member
> > > > > > >> > > > > > epochs
> > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > >> > > > > > > > > > > >> > and
> > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> correspond
> > > to
> > > > > > >> multiple
> > > > > > >> > > > > > different
> > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> field
> > > > > opaque
> > > > > > >> to the
> > > > > > >> > > > group
> > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > > assignor
> > > > > > >> > > > implementations
> > > > > > >> > > > > > > > > interact
> > > > > > >> > > > > > > > > > > >> with
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> common
> > > > > meaning
> > > > > > >> agreed
> > > > > > >> > > > > > upon? If
> > > > > > >> > > > > > > > > so, what
> > > > > > >> > > > > > > > > > > >> > is the
> > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> field
> > > over
> > > > > > >> including
> > > > > > >> > > > such
> > > > > > >> > > > > > > > > functionality
> > > > > > >> > > > > > > > > > > >> > in the
> > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 3. The following is included in
> the
> > > > KIP:
> > > > > > >> "Thanks to
> > > > > > >> > > > > > this, the
> > > > > > >> > > > > > > > > input
> > > > > > >> > > > > > > > > > > >> of
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> > > driven
> > > > > by
> > > > > > >> the group
> > > > > > >> > > > > > > > > coordinator.
> > > > > > >> > > > > > > > > > > >> The
> > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> responsible for
> > > > > > >> maintaining
> > > > > > >> > > > any
> > > > > > >> > > > > > state
> > > > > > >> > > > > > > > > besides
> > > > > > >> > > > > > > > > > > >> its
> > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this
> mean
> > > > > that the
> > > > > > >> > > > > > client-side
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > >> MAY
> > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> non-Metadata
> > > > state
> > > > > > >> (such as
> > > > > > >> > > > > > partition
> > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> topics,
> > > > etc),
> > > > > or
> > > > > > >> that
> > > > > > >> > > > > > additional
> > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > > > classes
> > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > >> > > > > >
> org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > >> > > > > > that
> > > > > > >> > > > > > > > > seem to
> > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> possible
> > > > for
> > > > > > >> these two
> > > > > > >> > > > > > > > > implementations
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> promote
> > > > > feature
> > > > > > >> parity
> > > > > > >> > > > of
> > > > > > >> > > > > > > > > server-side
> > > > > > >> > > > > > > > > > > >> and
> > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and would
> also
> > > > > > >> facilitate
> > > > > > >> > > > > > operational
> > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > >> > > > > > > > > > > >> > > > certain situations. For example,
> if a
> > > > > > >> server-side
> > > > > > >> > > > > > assignor
> > > > > > >> > > > > > > > > has some
> > > > > > >> > > > > > > > > > > >> > poor
> > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> deploying
> > > > the
> > > > > > >> patched
> > > > > > >> > > > > > assignor to
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > and switching one consumer group
> to a
> > > > > > >> client-side
> > > > > > >> > > > > > assignor
> > > > > > >> > > > > > > > > may be
> > > > > > >> > > > > > > > > > > >> > faster
> > > > > > >> > > > > > > > > > > >> > > > and less risky than patching all
> of
> > > the
> > > > > > >> brokers.
> > > > > > >> > > > With
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > currently
> > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> non-trivial
> > > > > > >> > > > reimplementation
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > have
> > > > > > >> > > > > > > > > > > >> to be
> > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs
> have
> > > > > diverged
> > > > > > >> > > > > > significantly,
> > > > > > >> > > > > > > > > then it
> > > > > > >> > > > > > > > > > > >> is
> > > > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> > > would
> > > > > not be
> > > > > > >> > > > possible.
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM
> Sagar
> > > <
> > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > > > discuss
> > > > > how
> > > > > > >> Connect
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > need to
> > > > > > >> > > > > > > > > > > >> > change
> > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > protocol.
> > > > > > >> Here's the
> > > > > > >> > > > page:
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I
> have
> > > > > tried
> > > > > > >> to keep
> > > > > > >> > > > a
> > > > > > >> > > > > > format
> > > > > > >> > > > > > > > > > > >> similar to
> > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > > think.
> > > > > Also,
> > > > > > >> do you
> > > > > > >> > > > > > think this
> > > > > > >> > > > > > > > > > > >> should
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> > > thread
> > > > > or
> > > > > > >> is this
> > > > > > >> > > > one
> > > > > > >> > > > > > fine?
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > > > Sagar <
> > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> doc on
> > > > > Kafka
> > > > > > >> Streams.
> > > > > > >> > > > > > Sure, I
> > > > > > >> > > > > > > > > would do
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> such a
> > > > > > >> document.
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47
> AM
> > > > > Guozhang
> > > > > > >> Wang <
> > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> could
> > > > come
> > > > > > >> back with
> > > > > > >> > > > some
> > > > > > >> > > > > > > > > analysis on
> > > > > > >> > > > > > > > > > > >> > how to
> > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > > integration
> > > > > > >> with
> > > > > > >> > > > the new
> > > > > > >> > > > > > > > > protocol;
> > > > > > >> > > > > > > > > > > >> so
> > > > > > >> > > > > > > > > > > >> > far
> > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the
> new
> > > > > "protocol
> > > > > > >> type"
> > > > > > >> > > > we
> > > > > > >> > > > > > did not
> > > > > > >> > > > > > > > > yet
> > > > > > >> > > > > > > > > > > >> think
> > > > > > >> > > > > > > > > > > >> > > > > through
> > > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > > implementations.
> > > > > For
> > > > > > >> Streams
> > > > > > >> > > > > > here's a
> > > > > > >> > > > > > > > > draft of
> > > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis
> on
> > > > > Connect.
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at
> 10:48 PM
> > > > > Sagar <
> > > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your
> response.
> > > The
> > > > > > >> reason I
> > > > > > >> > > > thought
> > > > > > >> > > > > > > > > connect can
> > > > > > >> > > > > > > > > > > >> > also fit
> > > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that
> even
> > > > > today the
> > > > > > >> > > > connect
> > > > > > >> > > > > > uses a
> > > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > > AbstractCoordinator to
> > > > > > >> empower
> > > > > > >> > > > > > > > > rebalances of
> > > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > > > >> WorkerCoordinator sets
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > > > >> protocolType()
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata()
> method
> > > > by
> > > > > > >> plumbing
> > > > > > >> > > > into
> > > > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to
> support
> > > > > > >> connect would
> > > > > > >> > > > be
> > > > > > >> > > > > > > > > similar at a
> > > > > > >> > > > > > > > > > > >> > high
> > > > > > >> > > > > > > > > > > >> > > > > level
> > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams
> mainly
> > > > > because
> > > > > > >> of the
> > > > > > >> > > > > > Client
> > > > > > >> > > > > > > > > side
> > > > > > >> > > > > > > > > > > >> > assignors
> > > > > > >> > > > > > > > > > > >> > > > > being
> > > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > > implementation
> > > > > > >> level, we
> > > > > > >> > > > > > might
> > > > > > >> > > > > > > > > need to
> > > > > > >> > > > > > > > > > > >> make
> > > > > > >> > > > > > > > > > > >> > a lot
> > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this
> new
> > > > > > >> assignment
> > > > > > >> > > > protocol
> > > > > > >> > > > > > like
> > > > > > >> > > > > > > > > > > >> enhancing
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup
> request/response and
> > > > > > >> SyncGroup and
> > > > > > >> > > > > > using
> > > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar
> lines
> > > > to
> > > > > > >> streams (or
> > > > > > >> > > > > > there
> > > > > > >> > > > > > > > > might be
> > > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> > > detailed
> > > > > > >> analysis of
> > > > > > >> > > > the
> > > > > > >> > > > > > same
> > > > > > >> > > > > > > > > and we
> > > > > > >> > > > > > > > > > > >> > can have
> > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion
> thread for
> > > > > that
> > > > > > >> as that
> > > > > > >> > > > would
> > > > > > >> > > > > > > > > derail this
> > > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if
> that
> > > > sounds
> > > > > > >> good to
> > > > > > >> > > > you.
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at
> 5:47
> > > PM
> > > > > David
> > > > > > >> Jacot
> > > > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your
> comments.
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > > > >> `Assignment#error`.
> > > > > > >> > > > > > Sure, I
> > > > > > >> > > > > > > > > can
> > > > > > >> > > > > > > > > > > >> > mention it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > > transition C
> > > > > from
> > > > > > >> his
> > > > > > >> > > > current
> > > > > > >> > > > > > > > > assignment
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > his
> > > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when
> he
> > > can
> > > > > move
> > > > > > >> to
> > > > > > >> > > > epoch 3.
> > > > > > >> > > > > > When
> > > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > happens,
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is
> updated
> > > > and
> > > > > > >> persisted
> > > > > > >> > > > > > with all
> > > > > > >> > > > > > > > > its
> > > > > > >> > > > > > > > > > > >> > assigned
> > > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if they
> are
> > > > not
> > > > > all
> > > > > > >> revoked
> > > > > > >> > > > > > yet. In
> > > > > > >> > > > > > > > > other
> > > > > > >> > > > > > > > > > > >> > words, the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment
> becomes
> > > the
> > > > > target
> > > > > > >> > > > > > assignment.
> > > > > > >> > > > > > > > > This is
> > > > > > >> > > > > > > > > > > >> > basically
> > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid
> having
> > > > to
> > > > > > >> write all
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > changes to
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > log.
> > > > > > >> > > > > > > > > > > >> > > > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on
> the
> > > > > persisted
> > > > > > >> state
> > > > > > >> > > > so I
> > > > > > >> > > > > > > > > understand
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see
> if I
> > > can
> > > > > > >> improve
> > > > > > >> > > > this in
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> > description.
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it
> > > could
> > > > > > >> reuse the
> > > > > > >> > > > > > protocol
> > > > > > >> > > > > > > > > with a
> > > > > > >> > > > > > > > > > > >> > client side
> > > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in
> the
> > > > > > >> protocol. The
> > > > > > >> > > > > > assignment
> > > > > > >> > > > > > > > > is
> > > > > > >> > > > > > > > > > > >> about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> > > metadata,
> > > > > could
> > > > > > >> > > > Connect
> > > > > > >> > > > > > fit
> > > > > > >> > > > > > > > > into this?
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at
> 1:55
> > > > PM
> > > > > > >> Sagar <
> > > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I
> just
> > > > had
> > > > > > >> minor
> > > > > > >> > > > > > observations:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment
> Error
> > > > > > >> section in
> > > > > > >> > > > Client
> > > > > > >> > > > > > Side
> > > > > > >> > > > > > > > > mode
> > > > > > >> > > > > > > > > > > >> > Assignment
> > > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In
> this
> > > > > case,
> > > > > > >> the
> > > > > > >> > > > client
> > > > > > >> > > > > > side
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > >> can
> > > > > > >> > > > > > > > > > > >> > > > > return
> > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > > > coordinator`.
> > > > > > >> In this
> > > > > > >> > > > > > case are
> > > > > > >> > > > > > > > > you
> > > > > > >> > > > > > > > > > > >> > referring to
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > > > > >> AssignmentError
> > > > > > >> > > > that's
> > > > > > >> > > > > > > > > listed down
> > > > > > >> > > > > > > > > > > >> > towards
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think
> it
> > > > would
> > > > > > >> make sense
> > > > > > >> > > > to
> > > > > > >> > > > > > > > > mention this
> > > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > > > > section, I
> > > > > > >> have a
> > > > > > >> > > > > > slight
> > > > > > >> > > > > > > > > > > >> confusion,
> > > > > > >> > > > > > > > > > > >> > not
> > > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the same.
> > > > Consider
> > > > > > >> this step:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the
> > > group
> > > > > > >> coordinator
> > > > > > >> > > > > > > > > transitions him
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > epoch
> > > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > > > > revoke. It
> > > > > > >> > > > persists
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > change and
> > > > > > >> > > > > > > > > > > >> > reply.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > > > > (epoch=3)
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > > partitions=[foo-0]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > > partitions=[foo-2]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > > partitions=[foo-1]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > > > >> partitions=[foo-0,
> > > > > > >> > > > > > foo-1]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > > > >> partitions=[foo-2]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > > > >> partitions=[foo-1]
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > > > > transitions
> > > > > > >> to
> > > > > > >> > > > epoch 3
> > > > > > >> > > > > > but
> > > > > > >> > > > > > > > > cannot
> > > > > > >> > > > > > > > > > > >> get
> > > > > > >> > > > > > > > > > > >> > foo-1
> > > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned
> that
> > > > > member C
> > > > > > >> can't
> > > > > > >> > > > get
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > foo-1
> > > > > > >> > > > > > > > > > > >> > partition
> > > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the
> description
> > > > > above,
> > > > > > >> it seems
> > > > > > >> > > > it
> > > > > > >> > > > > > > > > already has
> > > > > > >> > > > > > > > > > > >> it.
> > > > > > >> > > > > > > > > > > >> > Do you
> > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to
> remove
> > > it
> > > > > and
> > > > > > >> > > > populate it
> > > > > > >> > > > > > only
> > > > > > >> > > > > > > > > when it
> > > > > > >> > > > > > > > > > > >> > actually
> > > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of
> > > other
> > > > > > >> places, so
> > > > > > >> > > > have
> > > > > > >> > > > > > I
> > > > > > >> > > > > > > > > > > >> understood it
> > > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it
> > > might
> > > > > be
> > > > > > >> out of
> > > > > > >> > > > scope
> > > > > > >> > > > > > of
> > > > > > >> > > > > > > > > this
> > > > > > >> > > > > > > > > > > >> > discussion,
> > > > > > >> > > > > > > > > > > >> > > > > but
> > > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it
> would
> > > > > > >> probably be
> > > > > > >> > > > > > running in
> > > > > > >> > > > > > > > > client
> > > > > > >> > > > > > > > > > > >> > side
> > > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new
> rebalance
> > > > > > >> protocol as it
> > > > > > >> > > > has
> > > > > > >> > > > > > its
> > > > > > >> > > > > > > > > own
> > > > > > >> > > > > > > > > > > >> Custom
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > > IncrementalCooperative).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022
> at
> > > 5:00
> > > > > PM
> > > > > > >> David
> > > > > > >> > > > Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our
> goal
> > > is
> > > > > to
> > > > > > >> move
> > > > > > >> > > > forward
> > > > > > >> > > > > > with
> > > > > > >> > > > > > > > > > > >> > specialized API
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying
> on
> > > one
> > > > > > >> generic API.
> > > > > > >> > > > For
> > > > > > >> > > > > > > > > Connect, we
> > > > > > >> > > > > > > > > > > >> > can apply
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern
> and
> > > > > > >> reuse/share the
> > > > > > >> > > > core
> > > > > > >> > > > > > > > > > > >> > implementation on
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the
> > > > schema
> > > > > > >> registry, I
> > > > > > >> > > > > > think
> > > > > > >> > > > > > > > > that we
> > > > > > >> > > > > > > > > > > >> > should
> > > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored
> API to
> > > > do
> > > > > > >> simple
> > > > > > >> > > > > > > > > membership/leader
> > > > > > >> > > > > > > > > > > >> > election.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15,
> 2022 at
> > > > > 10:22
> > > > > > >> AM Ismael
> > > > > > >> > > > > > Juma <
> > > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick
> comments:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > > >> java.util.regex.Pattern
> > > > > > >> > > > vs
> > > > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > > differences
> > > > > in
> > > > > > >> more
> > > > > > >> > > > detail
> > > > > > >> > > > > > before
> > > > > > >> > > > > > > > > > > >> deciding
> > > > > > >> > > > > > > > > > > >> > one
> > > > > > >> > > > > > > > > > > >> > > > > way
> > > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if
> people
> > > > pass
> > > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > > >> > > > > > > > > > > >> they
> > > > > > >> > > > > > > > > > > >> > expect
> > > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be
> > > honored.
> > > > > If
> > > > > > >> we are
> > > > > > >> > > > doing
> > > > > > >> > > > > > > > > something
> > > > > > >> > > > > > > > > > > >> > different,
> > > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider
> adding
> > > > an
> > > > > > >> overload
> > > > > > >> > > > with
> > > > > > >> > > > > > our own
> > > > > > >> > > > > > > > > > > >> Pattern
> > > > > > >> > > > > > > > > > > >> > class
> > > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to
> > > > expose
> > > > > > >> re2j's at
> > > > > > >> > > > this
> > > > > > >> > > > > > > > > point).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic
> > > ids,
> > > > > any
> > > > > > >> major new
> > > > > > >> > > > > > protocol
> > > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should
> handle
> > > the
> > > > > topic
> > > > > > >> > > > > > recreation case
> > > > > > >> > > > > > > > > > > >> > correctly.
> > > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to
> > > handle.
> > > > I
> > > > > > >> agree with
> > > > > > >> > > > > > David
> > > > > > >> > > > > > > > > that we'd
> > > > > > >> > > > > > > > > > > >> > want to
> > > > > > >> > > > > > > > > > > >> > > > > add
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > > > protocols
> > > > > > >> that don't
> > > > > > >> > > > > > have it
> > > > > > >> > > > > > > > > yet and
> > > > > > >> > > > > > > > > > > >> > that we
> > > > > > >> > > > > > > > > > > >> > > > > can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the
> internals
> > > > > versus
> > > > > > >> adding
> > > > > > >> > > > new
> > > > > > >> > > > > > APIs
> > > > > > >> > > > > > > > > to the
> > > > > > >> > > > > > > > > > > >> > Java
> > > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that
> adding
> > > new
> > > > > APIs
> > > > > > >> is
> > > > > > >> > > > required
> > > > > > >> > > > > > for
> > > > > > >> > > > > > > > > > > >> reasonable
> > > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not
> sure
> > > > > about
> > > > > > >> the
> > > > > > >> > > > > > coordinator
> > > > > > >> > > > > > > > > storing
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for
> configs
> > > to
> > > > be
> > > > > > >> > > > centralized
> > > > > > >> > > > > > in the
> > > > > > >> > > > > > > > > > > >> metadata
> > > > > > >> > > > > > > > > > > >> > log for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons
> (auditability,
> > > > > > >> visibility,
> > > > > > >> > > > > > consistency,
> > > > > > >> > > > > > > > > > > >> etc.).
> > > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > > automatically
> > > > > > >> deleting
> > > > > > >> > > > > > configs in a
> > > > > > >> > > > > > > > > way
> > > > > > >> > > > > > > > > > > >> that
> > > > > > >> > > > > > > > > > > >> > they
> > > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> > > > property
> > > > > > >> for modern
> > > > > > >> > > > > > systems
> > > > > > >> > > > > > > > > is to
> > > > > > >> > > > > > > > > > > >> > minimize
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data
> > > loss
> > > > > > >> scenarios.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13,
> 2022
> > > at
> > > > > 3:47
> > > > > > >> PM David
> > > > > > >> > > > > > Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks
> Guozhang. My
> > > > > answers
> > > > > > >> are
> > > > > > >> > > > below:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> migration
> > > > path,
> > > > > > >> especially
> > > > > > >> > > > > > the last
> > > > > > >> > > > > > > > > step
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the
> new
> > > > > > >> protocol, in
> > > > > > >> > > > which
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> rpcs
> > > and
> > > > > old
> > > > > > >> > > > protocols /
> > > > > > >> > > > > > rpcs
> > > > > > >> > > > > > > > > are
> > > > > > >> > > > > > > > > > > >> used
> > > > > > >> > > > > > > > > > > >> > by
> > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > > >> coordinator could
> > > > > > >> > > > > > "mimic"
> > > > > > >> > > > > > > > > the old
> > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> is
> > > > > something
> > > > > > >> we
> > > > > > >> > > > need to
> > > > > > >> > > > > > > > > present
> > > > > > >> > > > > > > > > > > >> about.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > > > published
> > > > > a
> > > > > > >> new
> > > > > > >> > > > version
> > > > > > >> > > > > > of KIP
> > > > > > >> > > > > > > > > which
> > > > > > >> > > > > > > > > > > >> > includes
> > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about
> this.
> > > > See
> > > > > the
> > > > > > >> > > > "Supporting
> > > > > > >> > > > > > > > > Online
> > > > > > >> > > > > > > > > > > >> > Consumer
> > > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > > "Compatibility,
> > > > > > >> > > > Deprecation,
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > Migration
> > > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> > > > through
> > > > > a
> > > > > > >> few cases
> > > > > > >> > > > > > now but
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > overall idea
> > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism
> should be
> > > > > > >> understandable.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> of
> > > > topic
> > > > > > >> ids. So
> > > > > > >> > > > far as
> > > > > > >> > > > > > > > > KIP-516 the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> and
> > > > admin
> > > > > > >> client,
> > > > > > >> > > > but
> > > > > > >> > > > > > they
> > > > > > >> > > > > > > > > are not
> > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> yet.
> > > I
> > > > > think
> > > > > > >> the
> > > > > > >> > > > > > question is,
> > > > > > >> > > > > > > > > first
> > > > > > >> > > > > > > > > > > >> > should we
> > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > maintaining
> > > > > > >> the
> > > > > > >> > > > names
> > > > > > >> > > > > > -> ids
> > > > > > >> > > > > > > > > mapping
> > > > > > >> > > > > > > > > > > >> > itself
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > augmented
> > > > > > >> existing
> > > > > > >> > > > RPCs
> > > > > > >> > > > > > and the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > >> > > > > > > > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> should
> > > we
> > > > > ever
> > > > > > >> consider
> > > > > > >> > > > > > exposing
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > > well
> > > > > (both
> > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > >> > > > > > > > > > > >> well
> > > > > > >> > > > > > > > > > > >> > as in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> cases
> > > > > like
> > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming
> that we
> > > > > would
> > > > > > >> include
> > > > > > >> > > > > > converting
> > > > > > >> > > > > > > > > all
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > offsets
> > > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using
> topic
> > > > ids
> > > > > in
> > > > > > >> this
> > > > > > >> > > > KIP,
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> > would be
> > > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate
> with
> > > > topic
> > > > > > >> ids. That
> > > > > > >> > > > > > being
> > > > > > >> > > > > > > > > said, it
> > > > > > >> > > > > > > > > > > >> > still has
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> names in
> > > > > various
> > > > > > >> APIs so
> > > > > > >> > > > > > having a
> > > > > > >> > > > > > > > > > > >> mapping
> > > > > > >> > > > > > > > > > > >> > in the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems
> inevitable to
> > > > me.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have
> a
> > > > strong
> > > > > > >> opinion on
> > > > > > >> > > > > > this.
> > > > > > >> > > > > > > > > Here I
> > > > > > >> > > > > > > > > > > >> > wonder if
> > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the
> scope of
> > > > this
> > > > > > >> KIP. I
> > > > > > >> > > > would
> > > > > > >> > > > > > rather
> > > > > > >> > > > > > > > > focus
> > > > > > >> > > > > > > > > > > >> on
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> > > > consider
> > > > > > >> this
> > > > > > >> > > > > > separately if
> > > > > > >> > > > > > > > > we see
> > > > > > >> > > > > > > > > > > >> > value in
> > > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > > > Ismael's
> > > > > > >> point about
> > > > > > >> > > > > > using
> > > > > > >> > > > > > > > > topic ids
> > > > > > >> > > > > > > > > > > >> > in the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > > >> > > > think
> > > > > > >> > > > > > that
> > > > > > >> > > > > > > > > there
> > > > > > >> > > > > > > > > > > >> is
> > > > > > >> > > > > > > > > > > >> > one
> > > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it.
> The
> > > > > consumer
> > > > > > >> will
> > > > > > >> > > > have
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> opportunity
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics
> exists
> > > > before
> > > > > > >> passing
> > > > > > >> > > > them
> > > > > > >> > > > > > into
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> group
> > > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol.
> Obviously,
> > > > the
> > > > > > >> > > > coordinator
> > > > > > >> > > > > > will
> > > > > > >> > > > > > > > > also
> > > > > > >> > > > > > > > > > > >> notice
> > > > > > >> > > > > > > > > > > >> > it but
> > > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have
> a
> > > way
> > > > to
> > > > > > >> reject an
> > > > > > >> > > > > > invalid
> > > > > > >> > > > > > > > > topic in
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> with
> > > > > David
> > > > > > >> on all
> > > > > > >> > > > other
> > > > > > >> > > > > > minor
> > > > > > >> > > > > > > > > > > >> questions
> > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > `subscribe(Pattern)`
> > > > > > >> question:
> > > > > > >> > > > > > personally
> > > > > > >> > > > > > > > > I think
> > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > > subscribe
> > > > > > >> API with
> > > > > > >> > > > > > Pattern,
> > > > > > >> > > > > > > > > but
> > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > documenting
> > > > > > >> that our
> > > > > > >> > > > > > > > > subscription may
> > > > > > >> > > > > > > > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > incompatible
> > > > > > >> case is a
> > > > > > >> > > > > > very rare
> > > > > > >> > > > > > > > > > > >> scenario
> > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > `String`
> > > > > based
> > > > > > >> > > > > > subscription may
> > > > > > >> > > > > > > > > be
> > > > > > >> > > > > > > > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> regexes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could
> work. I
> > > > have
> > > > > to
> > > > > > >> look at
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > differences
> > > > > > >> > > > > > > > > > > >> > between
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to
> better
> > > > > > >> understand the
> > > > > > >> > > > > > potential
> > > > > > >> > > > > > > > > issues.
> > > > > > >> > > > > > > > > > > >> My
> > > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work
> for
> > > > all
> > > > > the
> > > > > > >> basic
> > > > > > >> > > > > > regular
> > > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two
> are
> > > > > mainly
> > > > > > >> about
> > > > > > >> > > > the
> > > > > > >> > > > > > various
> > > > > > >> > > > > > > > > > > >> character
> > > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what
> other
> > > > people
> > > > > > >> think
> > > > > > >> > > > about
> > > > > > >> > > > > > this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12,
> 2022
> > > > at
> > > > > > >> 11:28 PM
> > > > > > >> > > > > > Guozhang
> > > > > > >> > > > > > > > > Wang <
> > > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David!
> I
> > > > think
> > > > > on
> > > > > > >> the high
> > > > > > >> > > > > > level
> > > > > > >> > > > > > > > > there are
> > > > > > >> > > > > > > > > > > >> > two meta
> > > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize
> a
> > > bit
> > > > > more:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the
> migration
> > > > path,
> > > > > > >> especially
> > > > > > >> > > > > > the last
> > > > > > >> > > > > > > > > step
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the
> new
> > > > > > >> protocol, in
> > > > > > >> > > > which
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > window
> > > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols /
> rpcs
> > > and
> > > > > old
> > > > > > >> > > > protocols /
> > > > > > >> > > > > > rpcs
> > > > > > >> > > > > > > > > are
> > > > > > >> > > > > > > > > > > >> used
> > > > > > >> > > > > > > > > > > >> > by
> > > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > > >> coordinator could
> > > > > > >> > > > > > "mimic"
> > > > > > >> > > > > > > > > the old
> > > > > > >> > > > > > > > > > > >> > behavior
> > > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol
> is
> > > > > something
> > > > > > >> we
> > > > > > >> > > > need to
> > > > > > >> > > > > > > > > present
> > > > > > >> > > > > > > > > > > >> about.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage
> of
> > > > topic
> > > > > > >> ids. So
> > > > > > >> > > > far as
> > > > > > >> > > > > > > > > KIP-516 the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs
> and
> > > > admin
> > > > > > >> client,
> > > > > > >> > > > but
> > > > > > >> > > > > > they
> > > > > > >> > > > > > > > > are not
> > > > > > >> > > > > > > > > > > >> > exposed
> > > > > > >> > > > > > > > > > > >> > > > > via
> > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers
> yet.
> > > I
> > > > > think
> > > > > > >> the
> > > > > > >> > > > > > question is,
> > > > > > >> > > > > > > > > first
> > > > > > >> > > > > > > > > > > >> > should we
> > > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > > maintaining
> > > > > > >> the
> > > > > > >> > > > names
> > > > > > >> > > > > > -> ids
> > > > > > >> > > > > > > > > mapping
> > > > > > >> > > > > > > > > > > >> > itself
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > > augmented
> > > > > > >> existing
> > > > > > >> > > > RPCs
> > > > > > >> > > > > > and the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> RPCs
> > > > > > >> > > > > > > > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly,
> should
> > > we
> > > > > ever
> > > > > > >> consider
> > > > > > >> > > > > > exposing
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > topic ids
> > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > > well
> > > > > (both
> > > > > > >> > > > > > > > > subscribe/assign, as
> > > > > > >> > > > > > > > > > > >> well
> > > > > > >> > > > > > > > > > > >> > as in
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for
> cases
> > > > > like
> > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing
> with
> > > > > David
> > > > > > >> on all
> > > > > > >> > > > other
> > > > > > >> > > > > > minor
> > > > > > >> > > > > > > > > > > >> questions
> > > > > > >> > > > > > > > > > > >> > > > > except
> > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > `subscribe(Pattern)`
> > > > > > >> question:
> > > > > > >> > > > > > personally
> > > > > > >> > > > > > > > > I think
> > > > > > >> > > > > > > > > > > >> > it's not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > > subscribe
> > > > > > >> API with
> > > > > > >> > > > > > Pattern,
> > > > > > >> > > > > > > > > but
> > > > > > >> > > > > > > > > > > >> > instead we
> > > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > > documenting
> > > > > > >> that our
> > > > > > >> > > > > > > > > subscription may
> > > > > > >> > > > > > > > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > > incompatible
> > > > > > >> case is a
> > > > > > >> > > > > > very rare
> > > > > > >> > > > > > > > > > > >> scenario
> > > > > > >> > > > > > > > > > > >> > I felt
> > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > > `String`
> > > > > based
> > > > > > >> > > > > > subscription may
> > > > > > >> > > > > > > > > be
> > > > > > >> > > > > > > > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid
> regexes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul
> 12,
> > > 2022
> > > > > at
> > > > > > >> 5:23 AM
> > > > > > >> > > > > > David Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > <djacot@confluent.io.invalid
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for
> your
> > > > > > >> feedback. Let
> > > > > > >> > > > me
> > > > > > >> > > > > > answer
> > > > > > >> > > > > > > > > your
> > > > > > >> > > > > > > > > > > >> > questions
> > > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think
> > > it's
> > > > > > >> premature to
> > > > > > >> > > > > > talk about
> > > > > > >> > > > > > > > > > > >> target
> > > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal
> of the
> > > > > > >> existing group
> > > > > > >> > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> Unlike
> > > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> > > protocol
> > > > > and
> > > > > > >> hence
> > > > > > >> > > > > > > > > deprecation/removal
> > > > > > >> > > > > > > > > > > >> > will be
> > > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how
> quickly
> > > > > > >> applications
> > > > > > >> > > > migrate
> > > > > > >> > > > > > to
> > > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes
> > > sense.
> > > > I
> > > > > will
> > > > > > >> > > > remove it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP
> > > says
> > > > we
> > > > > > >> intend to
> > > > > > >> > > > > > release
> > > > > > >> > > > > > > > > this in
> > > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > > >> > > > > > > > > > > >> > > > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we
> > > added
> > > > > that
> > > > > > >> as a
> > > > > > >> > > > way to
> > > > > > >> > > > > > > > > estimate
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > we'd
> > > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > > > > protocol, I
> > > > > > >> also
> > > > > > >> > > > > > suggest
> > > > > > >> > > > > > > > > removing
> > > > > > >> > > > > > > > > > > >> > this part.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> explain
> > > my
> > > > > > >> reasoning. As
> > > > > > >> > > > > > > > > explained, I
> > > > > > >> > > > > > > > > > > >> plan
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> in
> > > > Java
> > > > > > >> while we
> > > > > > >> > > > > > implement
> > > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> > > internals
> > > > > will
> > > > > > >> be
> > > > > > >> > > > slightly
> > > > > > >> > > > > > > > > different
> > > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I
> > > > wanted
> > > > > to
> > > > > > >> tighten
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > switch from
> > > > > > >> > > > > > > > > > > >> > the old
> > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> to
> > > the
> > > > > new
> > > > > > >> group
> > > > > > >> > > > > > coordinator
> > > > > > >> > > > > > > > > to a
> > > > > > >> > > > > > > > > > > >> > major
> > > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative
> > > would
> > > > > be to
> > > > > > >> use a
> > > > > > >> > > > flag
> > > > > > >> > > > > > to do
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > switch
> > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the
> software
> > > > > upgrade.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We
> need to
> > > > > flesh
> > > > > > >> out the
> > > > > > >> > > > > > details of
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > migration
> > > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're
> saying
> > > we
> > > > > will
> > > > > > >> support
> > > > > > >> > > > > > online
> > > > > > >> > > > > > > > > > > >> migrations.
> > > > > > >> > > > > > > > > > > >> > Is
> > > > > > >> > > > > > > > > > > >> > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain
> this
> > > in
> > > > > > >> detail. It
> > > > > > >> > > > could
> > > > > > >> > > > > > also
> > > > > > >> > > > > > > > > be done
> > > > > > >> > > > > > > > > > > >> > as a
> > > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> > > > support
> > > > > > >> online
> > > > > > >> > > > > > migrations
> > > > > > >> > > > > > > > > for the
> > > > > > >> > > > > > > > > > > >> > group.
> > > > > > >> > > > > > > > > > > >> > > > > That
> > > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group
> using
> > > the
> > > > > old
> > > > > > >> protocol
> > > > > > >> > > > > > will be
> > > > > > >> > > > > > > > > able to
> > > > > > >> > > > > > > > > > > >> > switch to
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> briefly
> > > > > explain
> > > > > > >> how that
> > > > > > >> > > > > > will work
> > > > > > >> > > > > > > > > > > >> though.
> > > > > > >> > > > > > > > > > > >> > It is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step
> > > process:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The
> cluster
> > > > must
> > > > > be
> > > > > > >> > > > upgraded or
> > > > > > >> > > > > > > > > rolled to a
> > > > > > >> > > > > > > > > > > >> > software
> > > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > > > coordinator.
> > > > > > >> Both the
> > > > > > >> > > > > > old and
> > > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the
> old
> > > > > > >> protocol and
> > > > > > >> > > > rely
> > > > > > >> > > > > > on the
> > > > > > >> > > > > > > > > same
> > > > > > >> > > > > > > > > > > >> > persisted
> > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can
> work
> > > > > together.
> > > > > > >> This
> > > > > > >> > > > point
> > > > > > >> > > > > > is an
> > > > > > >> > > > > > > > > > > >> offline
> > > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one
> live
> > > > > > >> because it
> > > > > > >> > > > would
> > > > > > >> > > > > > require
> > > > > > >> > > > > > > > > > > >> > shutting down
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator
> and
> > > > > > >> starting up the
> > > > > > >> > > > > > new one
> > > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> that
> > > > > > >> > > > > > > > > > > >> > would
> > > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > unavailabilities.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The
> cluster's
> > > > > > >> metadata
> > > > > > >> > > > > > version/IBP
> > > > > > >> > > > > > > > > must be
> > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable
> the
> > > new
> > > > > > >> protocol.
> > > > > > >> > > > This
> > > > > > >> > > > > > cannot
> > > > > > >> > > > > > > > > be done
> > > > > > >> > > > > > > > > > > >> > before
> > > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated
> > > because
> > > > > the
> > > > > > >> old
> > > > > > >> > > > > > coordinator
> > > > > > >> > > > > > > > > doesn't
> > > > > > >> > > > > > > > > > > >> > support
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The
> consumers
> > > > > must be
> > > > > > >> > > > upgraded
> > > > > > >> > > > > > to a
> > > > > > >> > > > > > > > > version
> > > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> (must
> > > > have
> > > > > > >> KIP-792).
> > > > > > >> > > > If
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > consumer
> > > > > > >> > > > > > > > > > > >> is
> > > > > > >> > > > > > > > > > > >> > > > > already
> > > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing
> must be
> > > > > done at
> > > > > > >> this
> > > > > > >> > > > point.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The
> consumers
> > > > > must
> > > > > > >> be rolled
> > > > > > >> > > > > > with the
> > > > > > >> > > > > > > > > > > >> feature
> > > > > > >> > > > > > > > > > > >> > flag
> > > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer
> group
> > > is
> > > > > > >> automatically
> > > > > > >> > > > > > > > > converted when
> > > > > > >> > > > > > > > > > > >> > the first
> > > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the
> new
> > > > > protocol
> > > > > > >> joins
> > > > > > >> > > > the
> > > > > > >> > > > > > group.
> > > > > > >> > > > > > > > > While
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > members
> > > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old
> protocol are
> > > > > being
> > > > > > >> > > > upgraded,
> > > > > > >> > > > > > the old
> > > > > > >> > > > > > > > > > > >> protocol
> > > > > > >> > > > > > > > > > > >> > is
> > > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me
> clarify
> > > all
> > > > > of
> > > > > > >> this in
> > > > > > >> > > > the
> > > > > > >> > > > > > KIP.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am
> happy
> > > > > that we
> > > > > > >> are
> > > > > > >> > > > > > pushing the
> > > > > > >> > > > > > > > > pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it
> seems
> > > > like
> > > > > > >> there
> > > > > > >> > > > could be
> > > > > > >> > > > > > some
> > > > > > >> > > > > > > > > tricky
> > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> > > mechanism
> > > > > for
> > > > > > >> users to
> > > > > > >> > > > > > detect
> > > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> they
> > > > > > >> > > > > > > > > > > >> > need to
> > > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> > > switching
> > > > > to
> > > > > > >> the new
> > > > > > >> > > > > > protocol?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think
> that I
> > > am
> > > > a
> > > > > bit
> > > > > > >> more
> > > > > > >> > > > > > optimistic
> > > > > > >> > > > > > > > > than
> > > > > > >> > > > > > > > > > > >> you
> > > > > > >> > > > > > > > > > > >> > on this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe
> that the
> > > > > > >> majority of
> > > > > > >> > > > the
> > > > > > >> > > > > > cases
> > > > > > >> > > > > > > > > are
> > > > > > >> > > > > > > > > > > >> simple
> > > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with
> the
> > > new
> > > > > > >> engine. The
> > > > > > >> > > > > > > > > coordinator will
> > > > > > >> > > > > > > > > > > >> > verify
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject
> the
> > > > > consumer
> > > > > > >> if the
> > > > > > >> > > > > > regex is
> > > > > > >> > > > > > > > > not
> > > > > > >> > > > > > > > > > > >> valid.
> > > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration
> path,
> > > in
> > > > > the
> > > > > > >> worst
> > > > > > >> > > > case,
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > first
> > > > > > >> > > > > > > > > > > >> > upgraded
> > > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group
> will
> > > be
> > > > > > >> rejected.
> > > > > > >> > > > This
> > > > > > >> > > > > > should
> > > > > > >> > > > > > > > > be used
> > > > > > >> > > > > > > > > > > >> > as the
> > > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > > > > customers to
> > > > > > >> > > > validate
> > > > > > >> > > > > > their
> > > > > > >> > > > > > > > > regex
> > > > > > >> > > > > > > > > > > >> > before
> > > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would
> be to
> > > > > test
> > > > > > >> them with
> > > > > > >> > > > > > another
> > > > > > >> > > > > > > > > group.
> > > > > > >> > > > > > > > > > > >> For
> > > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be
> done
> > > in a
> > > > > > >> pre-prod
> > > > > > >> > > > > > environment.
> > > > > > >> > > > > > > > > > > >> Another
> > > > > > >> > > > > > > > > > > >> > way
> > > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > > > >> consumer-group tool
> > > > > > >> > > > to
> > > > > > >> > > > > > > > > provide a
> > > > > > >> > > > > > > > > > > >> regex
> > > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism.
> Would
> > > > > this be
> > > > > > >> > > > enough in
> > > > > > >> > > > > > your
> > > > > > >> > > > > > > > > > > >> opinion?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5.
> Related to
> > > > the
> > > > > last
> > > > > > >> > > > question,
> > > > > > >> > > > > > will
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> Java
> > > > > > >> > > > > > > > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick
> with the
> > > > > > >> current regex
> > > > > > >> > > > > > engine for
> > > > > > >> > > > > > > > > > > >> > compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be
> > > handy
> > > > to
> > > > > > >> keep using
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > based
> > > > > > >> > > > > > > > > > > >> > regex at
> > > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations
> > > > simple
> > > > > and
> > > > > > >> then
> > > > > > >> > > > > > migrate to
> > > > > > >> > > > > > > > > server
> > > > > > >> > > > > > > > > > > >> > based
> > > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand
> > > your
> > > > > point
> > > > > > >> but I
> > > > > > >> > > > am
> > > > > > >> > > > > > > > > concerned that
> > > > > > >> > > > > > > > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually
> stay
> > > > in
> > > > > > >> this mode.
> > > > > > >> > > > That
> > > > > > >> > > > > > > > > would go
> > > > > > >> > > > > > > > > > > >> > against our
> > > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying
> the
> > > > > client
> > > > > > >> because
> > > > > > >> > > > we
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > have to
> > > > > > >> > > > > > > > > > > >> > continue
> > > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> metadata on
> > > > the
> > > > > > >> client
> > > > > > >> > > > side. I
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > rather
> > > > > > >> > > > > > > > > > > >> > not do
> > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When
> we say
> > > > > that
> > > > > > >> the group
> > > > > > >> > > > > > > > > coordinator
> > > > > > >> > > > > > > > > > > >> will
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > > configurations
> > > > > > >> and that
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > > > >> configurations
> > > > > > >> > > > > > > > > > > >> > will be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > > > deleted.
> > > > > > >> Will a
> > > > > > >> > > > > > transition to
> > > > > > >> > > > > > > > > DEAD
> > > > > > >> > > > > > > > > > > >> > trigger
> > > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > configurations?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's
> right.
> > > The
> > > > > > >> > > > configurations
> > > > > > >> > > > > > will be
> > > > > > >> > > > > > > > > > > >> deleted
> > > > > > >> > > > > > > > > > > >> > when
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted.
> They go
> > > > > > >> together.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will
> the
> > > > > choice to
> > > > > > >> store
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > configs in
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to
> list
> > > > all
> > > > > > >> cluster
> > > > > > >> > > > > > configs and
> > > > > > >> > > > > > > > > their
> > > > > > >> > > > > > > > > > > >> > values?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't
> think
> > > so.
> > > > > The
> > > > > > >> group
> > > > > > >> > > > > > > > > configurations are
> > > > > > >> > > > > > > > > > > >> > overrides
> > > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If
> you
> > > > > want to
> > > > > > >> know
> > > > > > >> > > > all
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > overrides
> > > > > > >> > > > > > > > > > > >> > though,
> > > > > > >> > > > > > > > > > > >> > > > > you
> > > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all
> the
> > > > group
> > > > > > >> > > > coordinators.
> > > > > > >> > > > > > You
> > > > > > >> > > > > > > > > cannot
> > > > > > >> > > > > > > > > > > >> > rely on
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> instance.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How
> would
> > > > > someone
> > > > > > >> > > > configure a
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > before
> > > > > > >> > > > > > > > > > > >> > starting
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we
> considered
> > > > > > >> allowing the
> > > > > > >> > > > > > explicit
> > > > > > >> > > > > > > > > creation
> > > > > > >> > > > > > > > > > > >> of
> > > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs
> could
> > > be
> > > > > > >> decoupled
> > > > > > >> > > > from
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > group
> > > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The
> group
> > > > will
> > > > > be
> > > > > > >> > > > > > automatically
> > > > > > >> > > > > > > > > created in
> > > > > > >> > > > > > > > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs
> will
> > > > be
> > > > > > >> lost after
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > retention
> > > > > > >> > > > > > > > > > > >> > period of
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will
> the
> > > > > > >> > > > Consumer.subscribe
> > > > > > >> > > > > > method
> > > > > > >> > > > > > > > > for the
> > > > > > >> > > > > > > > > > > >> > Java
> > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> `java.util.regex.Pattern` of
> > > > > > >> > > > do
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > have to
> > > > > > >> > > > > > > > > > > >> > introduce
> > > > > > >> > > > > > > > > > > >> > > > > an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a
> very
> > > > group
> > > > > > >> question. I
> > > > > > >> > > > > > forgot
> > > > > > >> > > > > > > > > about
> > > > > > >> > > > > > > > > > > >> that
> > > > > > >> > > > > > > > > > > >> > one.
> > > > > > >> > > > > > > > > > > >> > > > > As
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> `java.util.regex.Pattern` is
> > > > > > >> > > > not
> > > > > > >> > > > > > fully
> > > > > > >> > > > > > > > > > > >> compatible
> > > > > > >> > > > > > > > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to
> use,
> > > it
> > > > > > >> might be
> > > > > > >> > > > better
> > > > > > >> > > > > > to
> > > > > > >> > > > > > > > > deprecate
> > > > > > >> > > > > > > > > > > >> > it and
> > > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes
> a
> > > > > string.
> > > > > > >> We would
> > > > > > >> > > > > > rely on
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> server
> > > > > > >> > > > > > > > > > > >> > side
> > > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > > > migration, I
> > > > > > >> think
> > > > > > >> > > > that
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > could
> > > > > > >> > > > > > > > > > > >> still
> > > > > > >> > > > > > > > > > > >> > try to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and
> use
> > > it.
> > > > > That
> > > > > > >> should
> > > > > > >> > > > > > work, I
> > > > > > >> > > > > > > > > think, in
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I
> agree
> > > with
> > > > > > >> Justine
> > > > > > >> > > > that we
> > > > > > >> > > > > > > > > should be
> > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > >> > > > > > > > > > > >> > > > > about
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > > > >> > > > IBP/metadata.version
> > > > > > >> > > > > > from the
> > > > > > >> > > > > > > > > > > >> feature
> > > > > > >> > > > > > > > > > > >> > flag.
> > > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can
> switch the
> > > > > > >> default for
> > > > > > >> > > > the
> > > > > > >> > > > > > feature
> > > > > > >> > > > > > > > > flag
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > true
> > > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > metadata.version
> > > > > once
> > > > > > >> we
> > > > > > >> > > > want to
> > > > > > >> > > > > > make
> > > > > > >> > > > > > > > > it the
> > > > > > >> > > > > > > > > > > >> > default.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was
> to
> > > use
> > > > > that
> > > > > > >> feature
> > > > > > >> > > > > > flag
> > > > > > >> > > > > > > > > mainly
> > > > > > >> > > > > > > > > > > >> > during the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I
> should
> > > > not
> > > > > have
> > > > > > >> > > > mentioned
> > > > > > >> > > > > > it, I
> > > > > > >> > > > > > > > > think,
> > > > > > >> > > > > > > > > > > >> > because
> > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal
> > > config
> > > > > for
> > > > > > >> it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some
> of
> > > the
> > > > > > >> protocol APIs
> > > > > > >> > > > > > don't
> > > > > > >> > > > > > > > > mention
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > required
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to
> add
> > > that
> > > > > for
> > > > > > >> > > > consistency.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is
> a
> > > bit
> > > > > odd
> > > > > > >> that
> > > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it
> > > seems
> > > > > to do
> > > > > > >> more
> > > > > > >> > > > than
> > > > > > >> > > > > > > > > reading.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree.
> This is
> > > > > how the
> > > > > > >> > > > current
> > > > > > >> > > > > > > > > protocol works
> > > > > > >> > > > > > > > > > > >> > though.
> > > > > > >> > > > > > > > > > > >> > > > > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require
> "Read
> > > > > Group" to
> > > > > > >> join a
> > > > > > >> > > > > > group. We
> > > > > > >> > > > > > > > > could
> > > > > > >> > > > > > > > > > > >> > consider
> > > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not
> > > sure
> > > > > that
> > > > > > >> it is
> > > > > > >> > > > worth
> > > > > > >> > > > > > it.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is
> > > topic
> > > > > > >> recreation
> > > > > > >> > > > > > handled by
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > consumer
> > > > > > >> > > > > > > > > > > >> > > > > with
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol?
> It
> > > > > would be
> > > > > > >> good to
> > > > > > >> > > > > > have a
> > > > > > >> > > > > > > > > section
> > > > > > >> > > > > > > > > > > >> on
> > > > > > >> > > > > > > > > > > >> > this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From
> a
> > > > > protocol
> > > > > > >> > > > > > perspective, the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > will
> > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so
> it
> > > > will
> > > > > > >> treat it
> > > > > > >> > > > like a
> > > > > > >> > > > > > > > > topic with
> > > > > > >> > > > > > > > > > > >> a
> > > > > > >> > > > > > > > > > > >> > > > > different
> > > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue
> is
> > > that
> > > > > the
> > > > > > >> > > > fetch/commit
> > > > > > >> > > > > > > > > offsets
> > > > > > >> > > > > > > > > > > >> APIs
> > > > > > >> > > > > > > > > > > >> > do not
> > > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > > > consumer
> > > > > > >> would
> > > > > > >> > > > reuse the
> > > > > > >> > > > > > > > > offsets
> > > > > > >> > > > > > > > > > > >> based
> > > > > > >> > > > > > > > > > > >> > on the
> > > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> should
> > > > > update
> > > > > > >> those
> > > > > > >> > > > APIs
> > > > > > >> > > > > > as well
> > > > > > >> > > > > > > > > in
> > > > > > >> > > > > > > > > > > >> order
> > > > > > >> > > > > > > > > > > >> > to be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That
> > > would
> > > > > > >> strengthen
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > semantics of
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The
> KIP
> > > > > mentions
> > > > > > >> we will
> > > > > > >> > > > > > write the
> > > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> > coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > > > > >> implementation
> > > > > > >> > > > > > detail, do
> > > > > > >> > > > > > > > > we plan
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > have a
> > > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we
> > > have a
> > > > > > >> scalability
> > > > > > >> > > > > > goal when
> > > > > > >> > > > > > > > > it
> > > > > > >> > > > > > > > > > > >> comes
> > > > > > >> > > > > > > > > > > >> > to how
> > > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group
> protocol
> > > > can
> > > > > > >> support?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't
> have
> > > > > numbers
> > > > > > >> at the
> > > > > > >> > > > > > moment. The
> > > > > > >> > > > > > > > > > > >> protocol
> > > > > > >> > > > > > > > > > > >> > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members
> per
> > > > > group.
> > > > > > >> We will
> > > > > > >> > > > > > measure
> > > > > > >> > > > > > > > > this when
> > > > > > >> > > > > > > > > > > >> > we have
> > > > > > >> > > > > > > > > > > >> > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> implementation.
> > > > Note
> > > > > > >> that we
> > > > > > >> > > > might
> > > > > > >> > > > > > have
> > > > > > >> > > > > > > > > other
> > > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g.
> > > offset
> > > > > > >> commits).
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > > > > consider
> > > > > > >> having
> > > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > > >> > > > > > > > > > > >> > instead
> > > > > > >> > > > > > > > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > SubscribedTopicNames
> > > > > > >> in
> > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we
> have
> > > to
> > > > > > >> resolve the
> > > > > > >> > > > > > regex on
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > server, we
> > > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic
> > > name?
> > > > > The
> > > > > > >> > > > difference
> > > > > > >> > > > > > is that
> > > > > > >> > > > > > > > > > > >> sending
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas
> > > sending
> > > > > the
> > > > > > >> topic
> > > > > > >> > > > names
> > > > > > >> > > > > > is less
> > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> recreation is
> > > > > easier
> > > > > > >> to
> > > > > > >> > > > handle
> > > > > > >> > > > > > if we
> > > > > > >> > > > > > > > > have
> > > > > > >> > > > > > > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > ids.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea
> was to
> > > > > > >> consolidate the
> > > > > > >> > > > > > metadata
> > > > > > >> > > > > > > > > lookup
> > > > > > >> > > > > > > > > > > >> > on the
> > > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I
> do
> > > > agree
> > > > > > >> with your
> > > > > > >> > > > > > point. As
> > > > > > >> > > > > > > > > a
> > > > > > >> > > > > > > > > > > >> second
> > > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be
> > > better
> > > > > here
> > > > > > >> for the
> > > > > > >> > > > > > delete and
> > > > > > >> > > > > > > > > > > >> > recreation
> > > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose
> that we
> > > > may
> > > > > > >> allow
> > > > > > >> > > > users to
> > > > > > >> > > > > > > > > subscribe
> > > > > > >> > > > > > > > > > > >> with
> > > > > > >> > > > > > > > > > > >> > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future
> because
> > > > that
> > > > > is
> > > > > > >> the only
> > > > > > >> > > > > > way to be
> > > > > > >> > > > > > > > > > > >> really
> > > > > > >> > > > > > > > > > > >> > robust
> > > > > > >> > > > > > > > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul
> 12,
> > > > > 2022 at
> > > > > > >> 1:38 PM
> > > > > > >> > > > > > David
> > > > > > >> > > > > > > > > Jacot <
> > > > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi
> Justine,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for
> > > your
> > > > > > >> comments.
> > > > > > >> > > > Please
> > > > > > >> > > > > > find
> > > > > > >> > > > > > > > > my
> > > > > > >> > > > > > > > > > > >> answers
> > > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes,
> the new
> > > > > > >> protocol
> > > > > > >> > > > relies on
> > > > > > >> > > > > > > > > topic IDs
> > > > > > >> > > > > > > > > > > >> > with the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic
> names
> > > > based
> > > > > in
> > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using
> topic
> > > > > names
> > > > > > >> is the
> > > > > > >> > > > > > right call
> > > > > > >> > > > > > > > > here.
> > > > > > >> > > > > > > > > > > >> I
> > > > > > >> > > > > > > > > > > >> > need to
> > > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little
> more.
> > > > > > >> Obviously, the
> > > > > > >> > > > > > KIP does
> > > > > > >> > > > > > > > > not
> > > > > > >> > > > > > > > > > > >> > change the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets
> RPCs
> > > to
> > > > > use
> > > > > > >> topic
> > > > > > >> > > > IDs.
> > > > > > >> > > > > > This
> > > > > > >> > > > > > > > > may be
> > > > > > >> > > > > > > > > > > >> > something
> > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include
> though
> > > > as
> > > > > it
> > > > > > >> would
> > > > > > >> > > > give
> > > > > > >> > > > > > better
> > > > > > >> > > > > > > > > > > >> overall
> > > > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're
> > > right.
> > > > I
> > > > > > >> think that
> > > > > > >> > > > I
> > > > > > >> > > > > > should
> > > > > > >> > > > > > > > > not
> > > > > > >> > > > > > > > > > > >> have
> > > > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I
> will
> > > > remove
> > > > > > >> it. We can
> > > > > > >> > > > > > use an
> > > > > > >> > > > > > > > > internal
> > > > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> developing the
> > > > > > >> feature.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both
> cluster
> > > > > types
> > > > > > >> will be
> > > > > > >> > > > > > > > > supported. The
> > > > > > >> > > > > > > > > > > >> > change is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only
> > > requirement
> > > > > is
> > > > > > >> that the
> > > > > > >> > > > > > cluster
> > > > > > >> > > > > > > > > uses
> > > > > > >> > > > > > > > > > > >> topic
> > > > > > >> > > > > > > > > > > >> > IDs.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon,
> Jul
> > > 11,
> > > > > 2022
> > > > > > >> at 9:53
> > > > > > >> > > > PM
> > > > > > >> > > > > > > > > Guozhang
> > > > > > >> > > > > > > > > > > >> Wang <
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi
> Ismael,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks
> for
> > > the
> > > > > > >> feedback.
> > > > > > >> > > > Here
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > > some
> > > > > > >> > > > > > > > > > > >> replies
> > > > > > >> > > > > > > > > > > >> > > > > inlined
> > > > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat,
> Jul
> > > 9,
> > > > > 2022
> > > > > > >> at
> > > > > > >> > > > 2:53 AM
> > > > > > >> > > > > > > > > Ismael Juma
> > > > > > >> > > > > > > > > > > >> <
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> Thanks for
> > > > the
> > > > > > >> KIP. This
> > > > > > >> > > > has
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> potential
> > > > > > >> > > > > > > > > > > >> > to be a
> > > > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> improvement. A
> > > few
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> initial
> > > > > > >> > > > questions/comments:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I
> think
> > > > > it's
> > > > > > >> > > > premature to
> > > > > > >> > > > > > talk
> > > > > > >> > > > > > > > > about
> > > > > > >> > > > > > > > > > > >> > target
> > > > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation
> and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> removal of
> > > > the
> > > > > > >> existing
> > > > > > >> > > > group
> > > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > > > > > > > >> > Unlike
> > > > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > > > > protocol
> > > > > > >> and hence
> > > > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how
> > > quickly
> > > > > > >> applications
> > > > > > >> > > > > > migrate
> > > > > > >> > > > > > > > > to the
> > > > > > >> > > > > > > > > > > >> new
> > > > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I
> agree
> > > > > with
> > > > > > >> you. I
> > > > > > >> > > > think
> > > > > > >> > > > > > we can
> > > > > > >> > > > > > > > > > > >> remove
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > `Compatibility,
> > > > > > >> > > > Deprecation,
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > Migration
> > > > > > >> > > > > > > > > > > >> > Plan` and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we
> will
> > > > > decide
> > > > > > >> in the
> > > > > > >> > > > > > future
> > > > > > >> > > > > > > > > about
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > we
> > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> protocol and
> > > > > > >> behaviors.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2.
> The KIP
> > > > > says we
> > > > > > >> > > > intend to
> > > > > > >> > > > > > > > > release this
> > > > > > >> > > > > > > > > > > >> > in 4.x,
> > > > > > >> > > > > > > > > > > >> > > > > >> but
> > > > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why.
> If we
> > > > > added
> > > > > > >> that as
> > > > > > >> > > > a
> > > > > > >> > > > > > way to
> > > > > > >> > > > > > > > > > > >> estimate
> > > > > > >> > > > > > > > > > > >> > when
> > > > > > >> > > > > > > > > > > >> > > > > we'd
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> group
> > > > > > >> protocol, I
> > > > > > >> > > > also
> > > > > > >> > > > > > suggest
> > > > > > >> > > > > > > > > > > >> removing
> > > > > > >> > > > > > > > > > > >> > this
> > > > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> > > that's
> > > > > not
> > > > > > >> > > > specifically
> > > > > > >> > > > > > > > > related to
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan,
> but
> > > it's
> > > > > more
> > > > > > >> for
> > > > > > >> > > > client
> > > > > > >> > > > > > > > > upgrades.
> > > > > > >> > > > > > > > > > > >> I.e.
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > implementation
> > > > > may
> > > > > > >> be done
> > > > > > >> > > > > > first,
> > > > > > >> > > > > > > > > and then
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only
> mark it
> > > > as
> > > > > > >> "released"
> > > > > > >> > > > by
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > time
> > > > > > >> > > > > > > > > > > >> clients
> > > > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that
> time,
> > > to
> > > > > > >> enable the
> > > > > > >> > > > > > feature the
> > > > > > >> > > > > > > > > > > >> clients
> > > > > > >> > > > > > > > > > > >> > need to
> > > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode
> > > with
> > > > a
> > > > > > >> rolling
> > > > > > >> > > > bounce
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > then set
> > > > > > >> > > > > > > > > > > >> > the flag
> > > > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce,
> and
> > > > > hence
> > > > > > >> we feel
> > > > > > >> > > > it's
> > > > > > >> > > > > > > > > better to be
> > > > > > >> > > > > > > > > > > >> > released
> > > > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We
> need
> > > > to
> > > > > > >> flesh out
> > > > > > >> > > > the
> > > > > > >> > > > > > > > > details of
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > migration
> > > > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're
> > > saying
> > > > > we
> > > > > > >> will
> > > > > > >> > > > support
> > > > > > >> > > > > > online
> > > > > > >> > > > > > > > > > > >> > migrations. Is
> > > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> explain
> > > this
> > > > > in
> > > > > > >> detail.
> > > > > > >> > > > It
> > > > > > >> > > > > > could
> > > > > > >> > > > > > > > > also be
> > > > > > >> > > > > > > > > > > >> > done as a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> easier.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> think
> > > > > that's
> > > > > > >> the
> > > > > > >> > > > part we
> > > > > > >> > > > > > can
> > > > > > >> > > > > > > > > be more
> > > > > > >> > > > > > > > > > > >> > concrete
> > > > > > >> > > > > > > > > > > >> > > > > >> about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is
> > > > related
> > > > > to
> > > > > > >> your
> > > > > > >> > > > > > question 2)
> > > > > > >> > > > > > > > > above).
> > > > > > >> > > > > > > > > > > >> > We will
> > > > > > >> > > > > > > > > > > >> > > > > >> work
> > > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> explicit in
> > > > > > >> parallel as we
> > > > > > >> > > > > > solicit
> > > > > > >> > > > > > > > > more
> > > > > > >> > > > > > > > > > > >> > feedback.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I
> am
> > > > happy
> > > > > > >> that we are
> > > > > > >> > > > > > pushing
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > pattern
> > > > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it
> > > seems
> > > > > like
> > > > > > >> there
> > > > > > >> > > > > > could be
> > > > > > >> > > > > > > > > some
> > > > > > >> > > > > > > > > > > >> tricky
> > > > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > > > > mechanism
> > > > > > >> for
> > > > > > >> > > > users to
> > > > > > >> > > > > > > > > detect that
> > > > > > >> > > > > > > > > > > >> > they
> > > > > > >> > > > > > > > > > > >> > > > > need
> > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > > > > switching
> > > > > > >> to the
> > > > > > >> > > > new
> > > > > > >> > > > > > > > > protocol?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I
> think
> > > we
> > > > > need
> > > > > > >> some
> > > > > > >> > > > > > tooling for
> > > > > > >> > > > > > > > > > > >> non-java
> > > > > > >> > > > > > > > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> "dry-run"
> > > the
> > > > > > >> client before
> > > > > > >> > > > > > > > > switching to
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific
> > > idea
> > > > on
> > > > > > >> top of my
> > > > > > >> > > > head
> > > > > > >> > > > > > > > > though,
> > > > > > >> > > > > > > > > > > >> maybe
> > > > > > >> > > > > > > > > > > >> > others
> > > > > > >> > > > > > > > > > > >> > > > > >> like
> > > > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in
> > > here?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5.
> Related
> > > > to
> > > > > the
> > > > > > >> last
> > > > > > >> > > > > > question,
> > > > > > >> > > > > > > > > will the
> > > > > > >> > > > > > > > > > > >> > Java
> > > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick
> with
> > > > the
> > > > > > >> current
> > > > > > >> > > > regex
> > > > > > >> > > > > > > > > engine for
> > > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it
> may be
> > > > > handy
> > > > > > >> to keep
> > > > > > >> > > > using
> > > > > > >> > > > > > > > > client
> > > > > > >> > > > > > > > > > > >> based
> > > > > > >> > > > > > > > > > > >> > regex
> > > > > > >> > > > > > > > > > > >> > > > > at
> > > > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> migrations
> > > > > simple
> > > > > > >> and
> > > > > > >> > > > then
> > > > > > >> > > > > > migrate
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > server based
> > > > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> Honestly I
> > > > have
> > > > > not
> > > > > > >> thought
> > > > > > >> > > > > > about
> > > > > > >> > > > > > > > > that for
> > > > > > >> > > > > > > > > > > >> > java
> > > > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that.
> What
> > > > kind
> > > > > of
> > > > > > >> > > > > > compatibility
> > > > > > >> > > > > > > > > issues do
> > > > > > >> > > > > > > > > > > >> > you have
> > > > > > >> > > > > > > > > > > >> > > > > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6.
> When we
> > > > say
> > > > > > >> that the
> > > > > > >> > > > group
> > > > > > >> > > > > > > > > coordinator
> > > > > > >> > > > > > > > > > > >> > will be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > > > > >> configurations and
> > > > > > >> > > > that
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> > configurations
> > > > > > >> > > > > > > > > > > >> > > > > will
> > > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> is
> > > > > deleted.
> > > > > > >> Will a
> > > > > > >> > > > > > > > > transition to
> > > > > > >> > > > > > > > > > > >> DEAD
> > > > > > >> > > > > > > > > > > >> > > > > trigger
> > > > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > configurations?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes,
> since
> > > the
> > > > > DEAD
> > > > > > >> state
> > > > > > >> > > > is an
> > > > > > >> > > > > > > > > ending
> > > > > > >> > > > > > > > > > > >> state
> > > > > > >> > > > > > > > > > > >> > (we
> > > > > > >> > > > > > > > > > > >> > > > > would
> > > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state
> when
> > > the
> > > > > > >> group is
> > > > > > >> > > > EMPTY
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > also all
> > > > > > >> > > > > > > > > > > >> of
> > > > > > >> > > > > > > > > > > >> > its
> > > > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's
> > > transited
> > > > > to
> > > > > > >> DEAD this
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > would
> > > > > > >> > > > > > > > > > > >> never
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7.
> Will
> > > the
> > > > > > >> choice to
> > > > > > >> > > > store
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > configs
> > > > > > >> > > > > > > > > > > >> in
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> harder to
> > > > > list all
> > > > > > >> > > > cluster
> > > > > > >> > > > > > configs
> > > > > > >> > > > > > > > > and
> > > > > > >> > > > > > > > > > > >> their
> > > > > > >> > > > > > > > > > > >> > > > > values?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a
> > > good
> > > > > > >> question,
> > > > > > >> > > > and our
> > > > > > >> > > > > > > > > thoughts
> > > > > > >> > > > > > > > > > > >> are
> > > > > > >> > > > > > > > > > > >> > that
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > configurations"
> > > > > are
> > > > > > >> > > > overrides
> > > > > > >> > > > > > of the
> > > > > > >> > > > > > > > > > > >> > cluster-level
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> customized
> > > per
> > > > > > >> group so
> > > > > > >> > > > when an
> > > > > > >> > > > > > > > > admin list
> > > > > > >> > > > > > > > > > > >> > cluster
> > > > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list
> just
> > > the
> > > > > > >> cluster-level
> > > > > > >> > > > > > > > > "defaults", not
> > > > > > >> > > > > > > > > > > >> > showing
> > > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > customizations.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How
> > > would
> > > > > > >> someone
> > > > > > >> > > > > > configure a
> > > > > > >> > > > > > > > > group
> > > > > > >> > > > > > > > > > > >> > before
> > > > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers?
> Have
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > > > considered
> > > > > > >> allowing
> > > > > > >> > > > the
> > > > > > >> > > > > > explicit
> > > > > > >> > > > > > > > > > > >> > creation of
> > > > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> Alternatively,
> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> configs
> > > > could
> > > > > be
> > > > > > >> > > > decoupled
> > > > > > >> > > > > > from
> > > > > > >> > > > > > > > > the group
> > > > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The
> configs
> > > > can
> > > > > be
> > > > > > >> created
> > > > > > >> > > > > > before
> > > > > > >> > > > > > > > > the group
> > > > > > >> > > > > > > > > > > >> > itself
> > > > > > >> > > > > > > > > > > >> > > > > as
> > > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of
> > > course,
> > > > > this
> > > > > > >> > > > requires
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > > >> corresponding
> > > > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > > > > coordinator
> > > > > > >> based on
> > > > > > >> > > > the
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > id ---
> > > > > > >> > > > > > > > > > > >> > the only
> > > > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the
> > > group
> > > > > > >> itself is
> > > > > > >> > > > gone
> > > > > > >> > > > > > we
> > > > > > >> > > > > > > > > also check
> > > > > > >> > > > > > > > > > > >> > if there
> > > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> configuration
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > > > > related to
> > > > > > >> that
> > > > > > >> > > > group
> > > > > > >> > > > > > and
> > > > > > >> > > > > > > > > delete
> > > > > > >> > > > > > > > > > > >> as
> > > > > > >> > > > > > > > > > > >> > well.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> Admittedly
> > > > this
> > > > > > >> indeed
> > > > > > >> > > > > > introduces an
> > > > > > >> > > > > > > > > > > >> > asymmetry on
> > > > > > >> > > > > > > > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> lifecycles
> > > of
> > > > > the
> > > > > > >> config
> > > > > > >> > > > > > entities,
> > > > > > >> > > > > > > > > and we
> > > > > > >> > > > > > > > > > > >> > would like
> > > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > > > > whether we
> > > > > > >> should
> > > > > > >> > > > aim
> > > > > > >> > > > > > for
> > > > > > >> > > > > > > > > symmetry
> > > > > > >> > > > > > > > > > > >> > i.e.
> > > > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs
> and
> > > > > hence
> > > > > > >> not
> > > > > > >> > > > delete
> > > > > > >> > > > > > them at
> > > > > > >> > > > > > > > > all
> > > > > > >> > > > > > > > > > > >> when
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> > > > explicit
> > > > > > >> deletion
> > > > > > >> > > > > > operations
> > > > > > >> > > > > > > > > by
> > > > > > >> > > > > > > > > > > >> > themselves.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9.
> Will
> > > the
> > > > > > >> > > > > > Consumer.subscribe
> > > > > > >> > > > > > > > > method for
> > > > > > >> > > > > > > > > > > >> > the Java
> > > > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > `java.util.regex.Pattern` of
> > > > > > >> > > > > > do we
> > > > > > >> > > > > > > > > have
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> we
> > > do
> > > > > not
> > > > > > >> need to
> > > > > > >> > > > > > introduce
> > > > > > >> > > > > > > > > an
> > > > > > >> > > > > > > > > > > >> > overload, but
> > > > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be
> some
> > > > > > >> compatibility
> > > > > > >> > > > > > issues
> > > > > > >> > > > > > > > > that we
> > > > > > >> > > > > > > > > > > >> may
> > > > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I
> > > agree
> > > > > with
> > > > > > >> Justine
> > > > > > >> > > > > > that we
> > > > > > >> > > > > > > > > should
> > > > > > >> > > > > > > > > > > >> be
> > > > > > >> > > > > > > > > > > >> > clearer
> > > > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> switch to
> > > > > > >> > > > > > IBP/metadata.version
> > > > > > >> > > > > > > > > from the
> > > > > > >> > > > > > > > > > > >> > feature
> > > > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can
> switch
> > > > the
> > > > > > >> default
> > > > > > >> > > > for
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > feature
> > > > > > >> > > > > > > > > > > >> flag
> > > > > > >> > > > > > > > > > > >> > to
> > > > > > >> > > > > > > > > > > >> > > > > true
> > > > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > metadata.version
> > > > > > >> once we
> > > > > > >> > > > > > want to
> > > > > > >> > > > > > > > > make it
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11.
> Some of
> > > > the
> > > > > > >> protocol
> > > > > > >> > > > APIs
> > > > > > >> > > > > > don't
> > > > > > >> > > > > > > > > mention
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > >> required
> > > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good
> to
> > > add
> > > > > that
> > > > > > >> for
> > > > > > >> > > > > > consistency.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we
> can
> > > > > > >> certainly do
> > > > > > >> > > > that.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12.
> It is
> > > a
> > > > > bit
> > > > > > >> odd that
> > > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> though it
> > > > > seems
> > > > > > >> to do
> > > > > > >> > > > more
> > > > > > >> > > > > > than
> > > > > > >> > > > > > > > > reading.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had
> that
> > > > > thought
> > > > > > >> myself
> > > > > > >> > > > as
> > > > > > >> > > > > > well,
> > > > > > >> > > > > > > > > but in
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > end we
> > > > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > > > alternative:
> > > > > > >> adding
> > > > > > >> > > > > > Write
> > > > > > >> > > > > > > > > Group
> > > > > > >> > > > > > > > > > > >> seems
> > > > > > >> > > > > > > > > > > >> > an
> > > > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not
> have it
> > > > > > >> elsewhere (we
> > > > > > >> > > > only
> > > > > > >> > > > > > have
> > > > > > >> > > > > > > > > Read /
> > > > > > >> > > > > > > > > > > >> > Delete
> > > > > > >> > > > > > > > > > > >> > > > > and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far).
> Would
> > > > > like to
> > > > > > >> hear
> > > > > > >> > > > others
> > > > > > >> > > > > > > > > thoughts.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13.
> How is
> > > > > topic
> > > > > > >> > > > recreation
> > > > > > >> > > > > > > > > handled by
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > consumer
> > > > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> protocol?
> > > It
> > > > > > >> would be
> > > > > > >> > > > good
> > > > > > >> > > > > > to have
> > > > > > >> > > > > > > > > a
> > > > > > >> > > > > > > > > > > >> > section on
> > > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean
> > > with
> > > > > regex
> > > > > > >> > > > > > subscription
> > > > > > >> > > > > > > > > right? Yes
> > > > > > >> > > > > > > > > > > >> > we can
> > > > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that,
> but
> > > > > basically
> > > > > > >> the
> > > > > > >> > > > idea
> > > > > > >> > > > > > is that
> > > > > > >> > > > > > > > > > > >> consumer
> > > > > > >> > > > > > > > > > > >> > would
> > > > > > >> > > > > > > > > > > >> > > > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > > > > protocol as
> > > > > > >> it's
> > > > > > >> > > > > > handled all
> > > > > > >> > > > > > > > > by the
> > > > > > >> > > > > > > > > > > >> > brokers.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14.
> The
> > > KIP
> > > > > > >> mentions we
> > > > > > >> > > > will
> > > > > > >> > > > > > write
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> new
> > > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this
> is an
> > > > > > >> implementation
> > > > > > >> > > > > > detail,
> > > > > > >> > > > > > > > > do we
> > > > > > >> > > > > > > > > > > >> > plan to
> > > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for
> it?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have
> not
> > > > > thought
> > > > > > >> about
> > > > > > >> > > > > > that. But
> > > > > > >> > > > > > > > > I think
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > answer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15.
> Do we
> > > > > have a
> > > > > > >> > > > scalability
> > > > > > >> > > > > > goal
> > > > > > >> > > > > > > > > when it
> > > > > > >> > > > > > > > > > > >> > comes to
> > > > > > >> > > > > > > > > > > >> > > > > >> how
> > > > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > > > > protocol can
> > > > > > >> > > > support?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> > > > group,
> > > > > I
> > > > > > >> think we
> > > > > > >> > > > > > should
> > > > > > >> > > > > > > > > shoot for
> > > > > > >> > > > > > > > > > > >> > 1000s of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> scalability
> > > > > goals
> > > > > > >> also
> > > > > > >> > > > depend
> > > > > > >> > > > > > on the
> > > > > > >> > > > > > > > > offset
> > > > > > >> > > > > > > > > > > >> > > > > management
> > > > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> capabilities
> > > > of
> > > > > the
> > > > > > >> > > > coordinator
> > > > > > >> > > > > > > > > which we
> > > > > > >> > > > > > > > > > > >> did
> > > > > > >> > > > > > > > > > > >> > not
> > > > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to
> > > give a
> > > > > > >> number that
> > > > > > >> > > > > > applies
> > > > > > >> > > > > > > > > > > >> > universally.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16.
> Did we
> > > > > > >> consider
> > > > > > >> > > > having
> > > > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> SubscribedTopicNames in
> > > > > > >> > > > > > > > > > > >> > > > > >>
> ConsumerGroupHeartbeatRequest?
> > > > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since
> we
> > > > have
> > > > > to
> > > > > > >> resolve
> > > > > > >> > > > the
> > > > > > >> > > > > > regex
> > > > > > >> > > > > > > > > on the
> > > > > > >> > > > > > > > > > > >> > server,
> > > > > > >> > > > > > > > > > > >> > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> topic
> > > > > name?
> > > > > > >> The
> > > > > > >> > > > > > difference is
> > > > > > >> > > > > > > > > that
> > > > > > >> > > > > > > > > > > >> > sending the
> > > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> whereas
> > > > > sending
> > > > > > >> the topic
> > > > > > >> > > > > > names is
> > > > > > >> > > > > > > > > less
> > > > > > >> > > > > > > > > > > >> > efficient.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> recreation
> > > > is
> > > > > > >> easier to
> > > > > > >> > > > > > handle if
> > > > > > >> > > > > > > > > we have
> > > > > > >> > > > > > > > > > > >> > topic
> > > > > > >> > > > > > > > > > > >> > > > > ids.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main
> > > > reason
> > > > > to
> > > > > > >> still
> > > > > > >> > > > let
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > clients
> > > > > > >> > > > > > > > > > > >> send
> > > > > > >> > > > > > > > > > > >> > names
> > > > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> reasoning of
> > > > > names
> > > > > > >> -> ids
> > > > > > >> > > > on
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > broker /
> > > > > > >> > > > > > > > > > > >> > admin
> > > > > > >> > > > > > > > > > > >> > > > > client
> > > > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> although we
> > > > > added
> > > > > > >> topic id
> > > > > > >> > > > in
> > > > > > >> > > > > > > > > KIP-516, we
> > > > > > >> > > > > > > > > > > >> > never
> > > > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > consumer/producers
> > > > > > >> > > > leveraging
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > related
> > > > > > >> > > > > > > > > > > >> > newer
> > > > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just
> set the
> > > > > topic
> > > > > > >> id as
> > > > > > >> > > > empty
> > > > > > >> > > > > > UUID.
> > > > > > >> > > > > > > > > We
> > > > > > >> > > > > > > > > > > >> want
> > > > > > >> > > > > > > > > > > >> > to keep
> > > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > consumer/producer
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be
> thin
> > > and
> > > > > only
> > > > > > >> > > > delegate
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > reasoning
> > > > > > >> > > > > > > > > > > >> on
> > > > > > >> > > > > > > > > > > >> > broker
> > > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> Thanks,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On
> Wed,
> > > Jul
> > > > 6,
> > > > > > >> 2022 at
> > > > > > >> > > > 10:45
> > > > > > >> > > > > > AM
> > > > > > >> > > > > > > > > David
> > > > > > >> > > > > > > > > > > >> Jacot
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi
> all,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I
> would
> > > > > like to
> > > > > > >> start a
> > > > > > >> > > > > > > > > discussion
> > > > > > >> > > > > > > > > > > >> thread
> > > > > > >> > > > > > > > > > > >> > on
> > > > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > Generation
> > > > > of
> > > > > > >> the
> > > > > > >> > > > Consumer
> > > > > > >> > > > > > > > > Rebalance
> > > > > > >> > > > > > > > > > > >> > Protocol.
> > > > > > >> > > > > > > > > > > >> > > > > >> With
> > > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to
> make
> > > > the
> > > > > > >> rebalance
> > > > > > >> > > > > > protocol
> > > > > > >> > > > > > > > > (for
> > > > > > >> > > > > > > > > > > >> > consumers)
> > > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > scalable,
> > > > > > >> easier to
> > > > > > >> > > > > > implement for
> > > > > > >> > > > > > > > > > > >> > clients, and
> > > > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > operators.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The
> KIP
> > > is
> > > > > here:
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> Please
> > > > take
> > > > > a
> > > > > > >> look and
> > > > > > >> > > > let
> > > > > > >> > > > > > me
> > > > > > >> > > > > > > > > know what
> > > > > > >> > > > > > > > > > > >> > you
> > > > > > >> > > > > > > > > > > >> > > > > think.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> Best,
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> David
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS:
> I
> > > will
> > > > > be
> > > > > > >> away from
> > > > > > >> > > > > > July
> > > > > > >> > > > > > > > > 18th to
> > > > > > >> > > > > > > > > > > >> > August 8th.
> > > > > > >> > > > > > > > > > > >> > > > > >> That
> > > > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of
> time
> > > to
> > > > > read
> > > > > > >> and
> > > > > > >> > > > digest
> > > > > > >> > > > > > this
> > > > > > >> > > > > > > > > long
> > > > > > >> > > > > > > > > > > >> KIP.
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> --
> > > > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > --
> > > > > > >> > > > > -- Guozhang
> > > > > > >> > > >
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> > > --
> > > > > > >> > > -- Guozhang
> > > > > > >>
> > > > > > >
> > > > > > >
> > > > > > > --
> > > > > > > -- Guozhang
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > -- Guozhang
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > > >
> > >
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Luke,

Thanks for your comments.

Your suggestion makes sense. Jun's made the same one. We are also
discussing whether we should directly return the group state in the
response and drop the prepare assignment api. It seems that the
current consensus is to keep it though. If that stays like this, I
will apply your suggestion.

Thanks,
David

On Tue, Sep 27, 2022 at 10:40 AM Luke Chen <sh...@gmail.com> wrote:
>
> Hi David,
>
> Thanks for your patience for answering all these questions!
>
> One more question from me:
> 1. When the selected consumer is going to be in client-side assigning, we
> will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering if we
> could use a flag to indicate the assignment is required. The rationale is
> when returning error, we lose a chance to catch up with the latest target
> assignment or revoke partitions. Since we expected the record fetching is
> not blocked by the rebalance/assignment computing, we might want to allow
> the members to update the member assignment in each heartbeat.
>
> Luke
>
> On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson <ja...@confluent.io.invalid>
> wrote:
>
> > Hey David,
> >
> > Thanks for the response.
> >
> > > Do you think that we could have a case where two members with the same
> > static id are active?
> >
> > I was just thinking about the handoff when we spin up a new node before the
> > old one completes shutdown. It's probably ok to fence the old member
> > abruptly. I do recall we had a few issues with this handoff in the past. It
> > was due to the fact that the replaced node might have been the leader.
> > Since we've removed the leader notion, perhaps this kind of issue won't
> > resurface.
> >
> > In regard to the RE2 library. I understand the intent to pass this through
> > as a string without any validation so that we don't need the dependence on
> > the client. I do wonder if that will be restrictive in practice. For
> > example, if a regex changes on the client side, would we need to validate
> > against the current assignment to see which partitions to continue
> > fetching? Or would we instead invalidate the complete current assignment?
> > Perhaps this is a rare case and we can do the latter. There could be more
> > situations though.
> >
> > As an alternative, I wonder if we could define a minimalistic regex syntax
> > (a subset of RE2 say) that could map more easily to regex support that
> > already exists in clients? For example, in this scenario, we are matching
> > topics, so we only need to be able to represent topic characters and not
> > the whole smorgasbord of whitespace and unicode options. Would it make
> > sense to have a reduced syntax which includes only standard topic
> > characters, wildcards, character classes, and repetitions?
> >
> > Best,
> > Jason
> >
> >
> >
> >
> > On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > > Hi David, thanks for all the detailed explanations. I think they all make
> > > sense. Just want to have a couple follow-ups here:
> > >
> > > > I don't really see the benefits here because server side assignors
> > > don't have metadata at all. They only assign topic-partitions. They
> > > are not supposed to generate metadata nor to receive metadata from the
> > > members.
> > >
> > > Just to clarify I was asking about the `version` of the assignor (i.e. up
> > > to what version that the client would support), and I do agree we would
> > not
> > > need metadata. What I have in mind is that, for some specific built-in
> > > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > > newer version we would have a hierarchical rack ID string format, like
> > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > would still be in old format. In this case, the broker then needs to
> > choose
> > > the old versioned assignor. I'm probably making something up here for
> > rack
> > > aware assignors, but I'm wondering if in general such an "auto-downgrade"
> > > behavior would be needed still for broker-side assignor, and if yes would
> > > "version" still be useful.
> > >
> > > > Yeah, that's right. Within a rebalance, `onAssignment` is called once
> > > when the member transitions to a new epoch. This one contains the full
> > > metadata provided by the client side assignor. Then, `onAssignment`
> > > can be called max N times where N is the number of partitions pending
> > > revocation by other members. Let me try to clarify this in the KIP.
> > >
> > > Okay, my understanding is that the calling ordering of these callbacks
> > > would be like the following:
> > >
> > > ----------------------------------------
> > > onPartitionsRevoked();   // just once, since we do not really need
> > > to revoke incrementally.
> > >
> > > onAssignment();    // the first call, with epoch incremented
> > > onPartitionsAssigned();   // paired with the onAssignment
> > >
> > > onAssignment();              // the first onAssignment would bump up the
> > > epoch, and the metadata reflected.
> > > onPartitionsAssigned();   // each time we get an additional assignment,
> > we
> > > call onAssignment and then paired with an onPartitionsAssigned
> > > ...
> > > onAssignment();
> > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > encoded
> > > metadata would not change, only the incrementally added partitions be
> > > reflected
> > >
> > > Is that the case?
> > >
> > > I'm wondering if we would still call onAssignment just once, that encodes
> > > all the assignment for this rebalance, including all the partitions that
> > > should be assigned to the member but not yet assigned since they have not
> > > been revoked by others. In that case the call ordering would be:
> > >
> > > ----------------------------------------
> > > onPartitionsRevoked();   // just once
> > > onAssignment();    // just once, with epoch incremented, and metadata
> > > encoded changed, the "assignment" field also reflect the final target
> > > assignment
> > > onPartitionsAssigned();   // multiple times, which represent
> > incrementally
> > > added partitions
> > > ...
> > > onPartitionsAssigned();
> > >
> > > The motivation from this is that, most users would only implement the
> > > rebalance callback listeners and hence we'd definitely need to make sure
> > > the semantics of that does not change much, and the time
> > > `onPartitionsAssigned` indicate the time when the partitions are actually
> > > assigned to it; while for assignors, the `onAssignment` is used to
> > indicate
> > > what decision is made regarding for this member, i.e. when the partitions
> > > are decided to be given to it, but not necessarily meant that it has been
> > > given, since that time should be determined by the time of
> > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > implementers would not need to reason which `onAssignment` would be the
> > > last one for this epoch.
> > >
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > > Guozhang
> > >
> > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot <djacot@confluent.io.invalid
> > >
> > > wrote:
> > >
> > > > Hi Guozhang,
> > > >
> > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > > field,
> > > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> > the
> > > > > rationale seems to be assuming that we should always be able to do
> > the
> > > > > assignment at the broker-side assignor without errors. Personally I
> > > think
> > > > > it's still potentially beneficial to add the Error field even for
> > > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > > topics are not recognized with the current broker's metadata. What do
> > > you
> > > > > think?
> > > >
> > > > Yeah, that seems reasonable. However, I wonder if it would be better
> > > > to use an exception on the server side. This is what we usually do for
> > > > server side plugins. On the client side, we use a field because the
> > > > errors are not defined in advance.
> > > >
> > > > Your comment also makes me think about what we should do when the
> > > > server side assignor fails. I suppose that we have to keep the current
> > > > assignment until a new event occurs. For instance, in your example,
> > > > the coordinator would have to trigger a rebalance when unrecognized
> > > > topics are available. This would be part of the metadata monitoring.
> > > >
> > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > > agree
> > > > > that broker-side assignor would not need reason/metadata since they
> > are
> > > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> > is
> > > > it
> > > > > possible that we evolve our broker-side built-in assignor but the old
> > > > > versioned clients would not be able to work with the new version, in
> > > > which
> > > > > case we need to let the broker being aware of this and upgrade its
> > > > behavior
> > > > > to cooperate with the clients?
> > > >
> > > > I don't really see the benefits here because server side assignors
> > > > don't have metadata at all. They only assign topic-partitions. They
> > > > are not supposed to generate metadata nor to receive metadata from the
> > > > members.
> > > >
> > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > instead
> > > > of
> > > > > including these three fields, what about just adding the "Metadata"
> > > field
> > > > > class which has these three fields? Also, there are two "Metadata"
> > > > > currently in the APIs, the first is a class that encodes
> > > > > reason/version/metadata, and the second is just the encoded metadata
> > > > bytes.
> > > > > I'm wondering what about just naming the first as memberMetadata,
> > which
> > > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > > bytebuffer field as metadataBytes?
> > > >
> > > > That's a good point. Let me try to rationalize this interface based on
> > > > your suggestions.
> > > >
> > > > Best,
> > > > David
> > > >
> > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > > >
> > > > > Hello David,
> > > > >
> > > > > Just had a few more nit questions about the public APIs:
> > > > >
> > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > > field,
> > > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> > the
> > > > > rationale seems to be assuming that we should always be able to do
> > the
> > > > > assignment at the broker-side assignor without errors. Personally I
> > > think
> > > > > it's still potentially beneficial to add the Error field even for
> > > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > > topics are not recognized with the current broker's metadata. What do
> > > you
> > > > > think?
> > > > >
> > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > > agree
> > > > > that broker-side assignor would not need reason/metadata since they
> > are
> > > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> > is
> > > > it
> > > > > possible that we evolve our broker-side built-in assignor but the old
> > > > > versioned clients would not be able to work with the new version, in
> > > > which
> > > > > case we need to let the broker being aware of this and upgrade its
> > > > behavior
> > > > > to cooperate with the clients?
> > > > >
> > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > instead
> > > > of
> > > > > including these three fields, what about just adding the "Metadata"
> > > field
> > > > > class which has these three fields? Also, there are two "Metadata"
> > > > > currently in the APIs, the first is a class that encodes
> > > > > reason/version/metadata, and the second is just the encoded metadata
> > > > bytes.
> > > > > I'm wondering what about just naming the first as memberMetadata,
> > which
> > > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > > bytebuffer field as metadataBytes?
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > > > wrote:
> > > > >
> > > > > > Hello David,
> > > > > >
> > > > > > Thanks for bringing this question up. I think the main benefits as
> > > you
> > > > > > listed is 2) above if it stays; just to clarify, we would only be
> > > able
> > > > to
> > > > > > save one round trip if the rebalance is still triggered by the
> > > broker;
> > > > if
> > > > > > the rebalance is triggered by the client then the num.round.trips
> > are
> > > > the
> > > > > > same:
> > > > > >
> > > > > > 1) With GroupPrepareAssignment:
> > > > > >
> > > > > > T0: client decides to do a new assignment, suppose it has already
> > > sent
> > > > a
> > > > > > HB and hence has to wait for it to return first since only one
> > > request
> > > > /
> > > > > > response can be inflight with the coordinator's socket.
> > > > > > T1: client receives the HB response, and then sends the
> > > > > > GroupPrepareAssignment request.
> > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > T3: it calculates the new assignment, and sends a
> > > > GroupInstallAssignment
> > > > > > request.
> > > > > >
> > > > > > In total, two round trips.
> > > > > >
> > > > > > 2) Without GroupPrepareAssignment:
> > > > > >
> > > > > > T0: client decides to do a new assignment, suppose it has already
> > > sent
> > > > a
> > > > > > HB and hence has to wait for it to return first since only one
> > > request
> > > > /
> > > > > > response can be inflight with the coordinator's socket.
> > > > > > T1: client receives the HB response, and then sends the new HB
> > > request
> > > > > > with the flag indicating a new rebalance needed..
> > > > > > T2: the HB response with the optional member metadata map is
> > > returned.
> > > > > > T3: it calculates the new assignment, and sends a
> > > > GroupInstallAssignment
> > > > > > request.
> > > > > >
> > > > > > In total, two round trips as well.
> > > > > >
> > > > > > -----------------------------
> > > > > >
> > > > > > So to complete the full picture here, we'd need to modify both HB
> > > > request
> > > > > > and response so that the client can also indicate a new rebalance
> > via
> > > > the
> > > > > > HB request as well, right?
> > > > > >
> > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > GroupPrepareAssignment into HB given that we can make the
> > additional
> > > > fields
> > > > > > encoding the full member (subscription) metadata and topic metadata
> > > as
> > > > > > optional fields.
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > <dj...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > >> Hi all,
> > > > > >>
> > > > > >> During an offline conversation, someone asked why we need the
> > > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > > instead
> > > > > >> provide the group state in the heartbeat response. This has a few
> > > > > >> advantages: 1) it does not require using a special error code to
> > > > > >> signal that a new assignment is required as the signal would be
> > the
> > > > > >> provided group state; 2) it removes one round trip when a client
> > > side
> > > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > > >> response's definition quite large. I recall that I went with the
> > > > > >> current approach due to this.
> > > > > >>
> > > > > >> Providing the group state in the heartbeat response is appealing.
> > > What
> > > > > >> do you guys think?
> > > > > >>
> > > > > >> Best,
> > > > > >> David
> > > > > >>
> > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > > > wrote:
> > > > > >> >
> > > > > >> > Hi Guozhang,
> > > > > >> >
> > > > > >> > 1. I have added a reference to the relevant chapter instead of
> > > > > >> > repeating the whole thing. Does that work for you?
> > > > > >> >
> > > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> > about
> > > > when
> > > > > >> > a rebalance should be triggered for the non-upgraded members
> > using
> > > > the
> > > > > >> > old protocol. The section mentions that a rebalance must be
> > > > triggered
> > > > > >> > when a new assignment is installed. This implies that the group
> > > > epoch
> > > > > >> > was updated either by a native member or a non-upgraded member.
> > > For
> > > > > >> > the latter, the JoinGroup request would be the trigger. I have
> > > > added a
> > > > > >> > reference to the relevant chapter in the "JoinGroup Handling"
> > > > section
> > > > > >> > as well. Does that make sense?
> > > > > >> >
> > > > > >> > Thanks,
> > > > > >> > David
> > > > > >> >
> > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > wangguoz@gmail.com
> > > >
> > > > > >> wrote:
> > > > > >> > >
> > > > > >> > > Hello David,
> > > > > >> > >
> > > > > >> > > Alright I think that's sufficient. Just to make that clear in
> > > the
> > > > doc,
> > > > > >> > > could we update:
> > > > > >> > >
> > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > coordinator
> > > > > >> will
> > > > > >> > > trigger rebalance based on the HB's member metadata / reason?
> > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > described
> > > > in
> > > > > >> "Group
> > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > >> > >
> > > > > >> > >
> > > > > >> > > Guozhang
> > > > > >> > >
> > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > >> <dj...@confluent.io.invalid>
> > > > > >> > > wrote:
> > > > > >> > >
> > > > > >> > > > Hi Guozhang,
> > > > > >> > > >
> > > > > >> > > > I thought that the assignor will always be consulted when
> > the
> > > > next
> > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > heartbeat.
> > > > > >> This
> > > > > >> > > > gives the opportunity for the assignor to enforce a
> > rebalance
> > > by
> > > > > >> > > > setting the reason to a non-zero value or by changing the
> > > > bytes. Do
> > > > > >> > > > you think that this is not sufficient? Are you concerned by
> > > the
> > > > > >> delay?
> > > > > >> > > >
> > > > > >> > > > Best,
> > > > > >> > > > David
> > > > > >> > > >
> > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > wangguoz@gmail.com>
> > > > > >> wrote:
> > > > > >> > > > >
> > > > > >> > > > > Hello David,
> > > > > >> > > > >
> > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > >> > > > >
> > > > > >> > > > > ```
> > > > > >> > > > > In this case, a new assignment is triggered by the client
> > > side
> > > > > >> > > > > assignor. When constructing the HB, the consumer will
> > always
> > > > > >> consult
> > > > > >> > > > > the client side assignor and propagate the information to
> > > the
> > > > > >> group
> > > > > >> > > > > coordinator. In other words, we don't expect users to call
> > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > >> > > > > ```
> > > > > >> > > > >
> > > > > >> > > > > As I looked at the current PartitionAssignor's interface,
> > we
> > > > > >> actually do
> > > > > >> > > > > not have a way yet to instruct how to construct the next
> > HB
> > > > > >> request, e.g.
> > > > > >> > > > > when the assignor wants to enforce a new rebalance with a
> > > new
> > > > > >> assignment,
> > > > > >> > > > > we'd need some customizable APIs inside the
> > > PartitionAssignor
> > > > to
> > > > > >> indicate
> > > > > >> > > > > the next HB telling broker about so. WDYT about adding
> > such
> > > an
> > > > > >> API on the
> > > > > >> > > > > PartitionAssignor?
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> > > > > Guozhang
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > >> <dj...@confluent.io.invalid>
> > > > > >> > > > > wrote:
> > > > > >> > > > >
> > > > > >> > > > > > Hi Jun,
> > > > > >> > > > > >
> > > > > >> > > > > > I have updated the KIP to include your feedback. I have
> > > also
> > > > > >> tried to
> > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > >> > > > > >
> > > > > >> > > > > > Best,
> > > > > >> > > > > > David
> > > > > >> > > > > >
> > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > djacot@confluent.io
> > > > > >> >
> > > > > >> > > > wrote:
> > > > > >> > > > > > >
> > > > > >> > > > > > > Hi Jun,
> > > > > >> > > > > > >
> > > > > >> > > > > > > Thanks for your feedback. Let me start by answering
> > your
> > > > > >> questions
> > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> > the
> > > > KIP
> > > > > >> seem to
> > > > > >> > > > be
> > > > > >> > > > > > fewer
> > > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > > >> wildcard. A few
> > > > > >> > > > > > > > comments below.
> > > > > >> > > > > > >
> > > > > >> > > > > > > I would also add that the KIP removes the global sync
> > > > barrier
> > > > > >> in the
> > > > > >> > > > > > > protocol which is essential to improve group stability
> > > and
> > > > > >> > > > > > > scalability, and the KIP also simplifies the client by
> > > > moving
> > > > > >> most of
> > > > > >> > > > > > > the logic to the server side.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > > support
> > > > > >> rolling
> > > > > >> > > > > > changing
> > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > >> > > > > > >
> > > > > >> > > > > > > Definitely. The group coordinator will use the
> > assignor
> > > > used
> > > > > >> by a
> > > > > >> > > > > > > majority of the members. This allows the group to move
> > > > from
> > > > > >> one
> > > > > >> > > > > > > assignor to another by a roll. This is explained in
> > the
> > > > > >> Assignor
> > > > > >> > > > > > > Selection chapter.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > > >> required in
> > > > > >> > > > every
> > > > > >> > > > > > > > request or the scenarios when it needs to be filled?
> > > For
> > > > > >> example,
> > > > > >> > > > it's
> > > > > >> > > > > > not
> > > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >> > > > > > >
> > > > > >> > > > > > > The client is expected to set those fields in case of
> > a
> > > > > >> connection
> > > > > >> > > > > > > issue (e.g. timeout) or when the fields have changed
> > > since
> > > > > >> the last
> > > > > >> > > > > > > HB. The server populates those fields as long as the
> > > > member
> > > > > >> is not
> > > > > >> > > > > > > fully reconciled - the member should acknowledge that
> > it
> > > > has
> > > > > >> the
> > > > > >> > > > > > > expected epoch and assignment. I will clarify this in
> > > the
> > > > KIP.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > affinity
> > > > > >> between the
> > > > > >> > > > > > client
> > > > > >> > > > > > > > and the broker is only considered during fetching,
> > but
> > > > not
> > > > > >> during
> > > > > >> > > > > > assigning
> > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > assignment
> > > > is
> > > > > >> made,
> > > > > >> > > > there
> > > > > >> > > > > > is
> > > > > >> > > > > > > > no opportunity for read affinity because no replicas
> > > of
> > > > > >> assigned
> > > > > >> > > > > > partitions
> > > > > >> > > > > > > > are close to the member. I am wondering if we should
> > > use
> > > > > >> this
> > > > > >> > > > > > opportunity
> > > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > > >> > > > > > >
> > > > > >> > > > > > > That's an interesting idea. I don't see any issue with
> > > > adding
> > > > > >> the
> > > > > >> > > > rack
> > > > > >> > > > > > > to the members. I will do so.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> > how
> > > > busy
> > > > > >> a group
> > > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > > seems
> > > > > >> that we
> > > > > >> > > > could
> > > > > >> > > > > > add
> > > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > > event
> > > > > >> loop is
> > > > > >> > > > doing
> > > > > >> > > > > > the
> > > > > >> > > > > > > > actual work.
> > > > > >> > > > > > >
> > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > > >> handling? For
> > > > > >> > > > > > example,
> > > > > >> > > > > > > > does it need to trigger the check if any group with
> > > the
> > > > > >> wildcard
> > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > >> > > > > > >
> > > > > >> > > > > > > Sure. When the new group coordinator takes over, it
> > has
> > > > to:
> > > > > >> > > > > > > * Setup the session timeouts.
> > > > > >> > > > > > > * Trigger a new assignment if a client side assignor
> > is
> > > > used.
> > > > > >> We
> > > > > >> > > > don't
> > > > > >> > > > > > > store the information about the member selected to run
> > > the
> > > > > >> assignment
> > > > > >> > > > > > > so we have to start a new one.
> > > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > > >> subscriptions, and
> > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > > what
> > > > > >> the epoch
> > > > > >> > > > > > field
> > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > >> ConsumerGroupMetadataValue
> > > > > >> > > > > > reflect
> > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > >> > > > > > >
> > > > > >> > > > > > > Sure. I will clarify that but it is always the latest
> > > > group
> > > > > >> epoch.
> > > > > >> > > > > > > When the group state is updated, the group epoch is
> > > > bumped so
> > > > > >> we use
> > > > > >> > > > > > > that one for all the change records related to the
> > > update.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > following
> > > > > >> > > > invariants
> > > > > >> > > > > > are
> > > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > > member
> > > > > >> not to
> > > > > >> > > > get any
> > > > > >> > > > > > > > assigned partitions, right?
> > > > > >> > > > > > >
> > > > > >> > > > > > > That's right. Here I meant that the members provided
> > by
> > > > the
> > > > > >> assignor
> > > > > >> > > > > > > in the assignment must exist in the group. The
> > assignor
> > > > can
> > > > > >> not make
> > > > > >> > > > > > > up new member ids.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > > equals to
> > > > > >> 0":
> > > > > >> > > > When
> > > > > >> > > > > > would
> > > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > > >> > > > > > >
> > > > > >> > > > > > > A member is expected to abandon all its partitions and
> > > > > >> rejoins when
> > > > > >> > > > it
> > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case,
> > > the
> > > > > >> group
> > > > > >> > > > > > > coordinator will have removed the member from the
> > group.
> > > > The
> > > > > >> member
> > > > > >> > > > > > > can rejoin the group with the same member id but with
> > 0
> > > as
> > > > > >> epoch. Let
> > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > > trigger a
> > > > > >> > > > > > reassignment
> > > > > >> > > > > > > > by either providing a non-zero reason or by updating
> > > the
> > > > > >> assignor
> > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> > the
> > > > > >> deprecation
> > > > > >> > > > of
> > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > >> > > > > > >
> > > > > >> > > > > > > In this case, a new assignment is triggered by the
> > > client
> > > > side
> > > > > >> > > > > > > assignor. When constructing the HB, the consumer will
> > > > always
> > > > > >> consult
> > > > > >> > > > > > > the client side assignor and propagate the information
> > > to
> > > > the
> > > > > >> group
> > > > > >> > > > > > > coordinator. In other words, we don't expect users to
> > > call
> > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > section
> > > > > >> seems to
> > > > > >> > > > have
> > > > > >> > > > > > > > multiple typos.
> > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > immediately
> > > > > >> gets into
> > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > incorrect.
> > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> > to
> > > > get
> > > > > >> into
> > > > > >> > > > > > epoch=3,
> > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > > >> epoch=2,
> > > > > >> > > > > > > > partitions=[foo-0].
> > > > > >> > > > > > >
> > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > > upgrade
> > > > > >> from
> > > > > >> > > > any
> > > > > >> > > > > > old
> > > > > >> > > > > > > > version to new one?
> > > > > >> > > > > > >
> > > > > >> > > > > > > We will support upgrading from the consumer protocol
> > > > version
> > > > > >> 3,
> > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet
> > so
> > > > the
> > > > > >> earliest
> > > > > >> > > > > > > version is unknown at the moment. This is explained in
> > > the
> > > > > >> migration
> > > > > >> > > > > > > plan chapter.
> > > > > >> > > > > > >
> > > > > >> > > > > > > Thanks again for your feedback, Jun. I will update the
> > > KIP
> > > > > >> based on
> > > > > >> > > > it
> > > > > >> > > > > > > next week.
> > > > > >> > > > > > >
> > > > > >> > > > > > > Best,
> > > > > >> > > > > > > David
> > > > > >> > > > > > >
> > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > >> <ju...@confluent.io.invalid>
> > > > > >> > > > wrote:
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Hi, David,
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> > the
> > > > KIP
> > > > > >> seem to
> > > > > >> > > > be
> > > > > >> > > > > > fewer
> > > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > > >> wildcard. A few
> > > > > >> > > > > > > > comments below.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > > support
> > > > > >> rolling
> > > > > >> > > > > > changing
> > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > > >> required in
> > > > > >> > > > every
> > > > > >> > > > > > > > request or the scenarios when it needs to be filled?
> > > For
> > > > > >> example,
> > > > > >> > > > it's
> > > > > >> > > > > > not
> > > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > affinity
> > > > > >> between the
> > > > > >> > > > > > client
> > > > > >> > > > > > > > and the broker is only considered during fetching,
> > but
> > > > not
> > > > > >> during
> > > > > >> > > > > > assigning
> > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > assignment
> > > > is
> > > > > >> made,
> > > > > >> > > > there
> > > > > >> > > > > > is
> > > > > >> > > > > > > > no opportunity for read affinity because no replicas
> > > of
> > > > > >> assigned
> > > > > >> > > > > > partitions
> > > > > >> > > > > > > > are close to the member. I am wondering if we should
> > > use
> > > > > >> this
> > > > > >> > > > > > opportunity
> > > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> > how
> > > > busy
> > > > > >> a group
> > > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > > seems
> > > > > >> that we
> > > > > >> > > > could
> > > > > >> > > > > > add
> > > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > > event
> > > > > >> loop is
> > > > > >> > > > doing
> > > > > >> > > > > > the
> > > > > >> > > > > > > > actual work.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > > >> handling? For
> > > > > >> > > > > > example,
> > > > > >> > > > > > > > does it need to trigger the check if any group with
> > > the
> > > > > >> wildcard
> > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > > what
> > > > > >> the epoch
> > > > > >> > > > > > field
> > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > >> ConsumerGroupMetadataValue
> > > > > >> > > > > > reflect
> > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > following
> > > > > >> > > > invariants
> > > > > >> > > > > > are
> > > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > > member
> > > > > >> not to
> > > > > >> > > > get any
> > > > > >> > > > > > > > assigned partitions, right?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > > equals to
> > > > > >> 0":
> > > > > >> > > > When
> > > > > >> > > > > > would
> > > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > > trigger a
> > > > > >> > > > > > reassignment
> > > > > >> > > > > > > > by either providing a non-zero reason or by updating
> > > the
> > > > > >> assignor
> > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> > the
> > > > > >> deprecation
> > > > > >> > > > of
> > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > section
> > > > > >> seems to
> > > > > >> > > > have
> > > > > >> > > > > > > > multiple typos.
> > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > immediately
> > > > > >> gets into
> > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > incorrect.
> > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> > to
> > > > get
> > > > > >> into
> > > > > >> > > > > > epoch=3,
> > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > > >> epoch=2,
> > > > > >> > > > > > > > partitions=[foo-0].
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > > upgrade
> > > > > >> from
> > > > > >> > > > any
> > > > > >> > > > > > old
> > > > > >> > > > > > > > version to new one?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Thanks,
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Jun
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > >> > > > > > > > wrote:
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > Hi all,
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > coordinator
> > > > > >> in
> > > > > >> > > > Java. I
> > > > > >> > > > > > > > > discussed this offline with a few folks and folks
> > > are
> > > > > >> concerned
> > > > > >> > > > that
> > > > > >> > > > > > > > > we could introduce many regressions in the old
> > > > protocol
> > > > > >> if we do
> > > > > >> > > > so.
> > > > > >> > > > > > > > > Therefore, I am going to remove this statement
> > from
> > > > the
> > > > > >> KIP. It
> > > > > >> > > > is an
> > > > > >> > > > > > > > > implementation detail after all so it does not
> > have
> > > > to be
> > > > > >> > > > decided at
> > > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > > refactor the
> > > > > >> > > > current
> > > > > >> > > > > > > > > implementation as a first step.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Cheers,
> > > > > >> > > > > > > > > David
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > > >> djacot@confluent.io
> > > > > >> > > > >
> > > > > >> > > > > > wrote:
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Hi Luke,
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > > assigning,
> > > > > >> > > > > > reconciling,
> > > > > >> > > > > > > > > stable,
> > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > section,
> > > > > >> right?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > This sentence does not refer to those group
> > states
> > > > but
> > > > > >> rather
> > > > > >> > > > to a
> > > > > >> > > > > > > > > > state machine replication (SMR). This refers to
> > > the
> > > > > >> entire
> > > > > >> > > > state of
> > > > > >> > > > > > > > > > group coordinator which is replicated via the
> > log
> > > > > >> layer. I will
> > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > > modelled
> > > > > >> as an
> > > > > >> > > > event
> > > > > >> > > > > > > > > loop"?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > The idea is to follow a model similar to the new
> > > > quorum
> > > > > >> > > > > > controller. We
> > > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > > >> __consumer_offsets
> > > > > >> > > > > > > > > > partition is assigned to a unique thread and all
> > > the
> > > > > >> events
> > > > > >> > > > (e.g.
> > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > > > >> thread. This
> > > > > >> > > > > > simplify
> > > > > >> > > > > > > > > > concurrency and will enable us to do simulation
> > > > testing
> > > > > >> for the
> > > > > >> > > > > > group
> > > > > >> > > > > > > > > > coordinator.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > >> *__consumer_offsets*
> > > > > >> > > > > > > > > partitions?
> > > > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> > > by
> > > > a
> > > > > >> group
> > > > > >> > > > > > coordinator?
> > > > > >> > > > > > > > > For
> > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > consumer
> > > > > >> groups, and
> > > > > >> > > > > > both
> > > > > >> > > > > > > > > exist in
> > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > > machine
> > > > > >> for it,
> > > > > >> > > > or
> > > > > >> > > > > > 2?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > think.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> > *is
> > > > the
> > > > > >> number
> > > > > >> > > > of
> > > > > >> > > > > > threads
> > > > > >> > > > > > > > > used
> > > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> > if
> > > > the
> > > > > >> purpose
> > > > > >> > > > of
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > threads
> > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > group
> > > > (or
> > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > > should we
> > > > > >> need
> > > > > >> > > > > > > > > multi-threads
> > > > > >> > > > > > > > > > > here?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to shard
> > > the
> > > > > >> > > > processing as
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > computation could be heavy.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> > the
> > > > > >> default
> > > > > >> > > > session
> > > > > >> > > > > > > > > timeout not
> > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > thought
> > > > the
> > > > > >> min/max
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > > no?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > > >> timeout to
> > > > > >> > > > detect
> > > > > >> > > > > > client
> > > > > >> > > > > > > > > > > failures when using the consumer group
> > protocol.
> > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > > The
> > > > > >> minimum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > > The
> > > > > >> maximum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > This is indeed a mistake. The default session
> > > > timeout
> > > > > >> should
> > > > > >> > > > be 45s
> > > > > >> > > > > > > > > > (the current default).
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > [range,
> > > > > >> uniform],
> > > > > >> > > > > > which means
> > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > > know
> > > > > >> why not
> > > > > >> > > > > > uniform
> > > > > >> > > > > > > > > one? I
> > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > assignor
> > > > > >> (former
> > > > > >> > > > sticky
> > > > > >> > > > > > > > > assinor)
> > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > reason
> > > > we
> > > > > >> choose
> > > > > >> > > > range
> > > > > >> > > > > > > > > assignor
> > > > > >> > > > > > > > > > > as default?
> > > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> > The
> > > > > >> server side
> > > > > >> > > > > > assignors.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > The order on the server side has no influence
> > > > because
> > > > > >> the
> > > > > >> > > > client
> > > > > >> > > > > > must
> > > > > >> > > > > > > > > > chose the selector that he wants to use. There
> > is
> > > no
> > > > > >> default
> > > > > >> > > > in the
> > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > specified
> > > > by
> > > > > >> the
> > > > > >> > > > client,
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > request is rejected. The default client value
> > for
> > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > > > that my
> > > > > >> > > > answers
> > > > > >> > > > > > help
> > > > > >> > > > > > > > > > to clarify things. I will update the KIP as well
> > > > based
> > > > > >> on your
> > > > > >> > > > > > > > > > feedback.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Cheers,
> > > > > >> > > > > > > > > > David
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > > >> showuon@gmail.com>
> > > > > >> > > > > > wrote:
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Hi David,
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Some more questions:
> > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > mentioned:
> > > > > >> > > > > > > > > > > > The new group coordinator will have a state
> > > > machine
> > > > > >> per
> > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where each
> > > state
> > > > > >> machine is
> > > > > >> > > > > > modelled
> > > > > >> > > > > > > > > as an
> > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > executed
> > > > in
> > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > > assigning,
> > > > > >> > > > > > reconciling,
> > > > > >> > > > > > > > > stable,
> > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > section,
> > > > > >> right?
> > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > > modelled
> > > > > >> as an
> > > > > >> > > > event
> > > > > >> > > > > > > > > loop"?
> > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > >> *__consumer_offsets*
> > > > > >> > > > > > > > > partitions?
> > > > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> > > by
> > > > a
> > > > > >> group
> > > > > >> > > > > > coordinator?
> > > > > >> > > > > > > > > For
> > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > consumer
> > > > > >> groups, and
> > > > > >> > > > > > both
> > > > > >> > > > > > > > > exist in
> > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > > machine
> > > > > >> for it,
> > > > > >> > > > or
> > > > > >> > > > > > 2?
> > > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> > *is
> > > > the
> > > > > >> number
> > > > > >> > > > of
> > > > > >> > > > > > threads
> > > > > >> > > > > > > > > used
> > > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> > if
> > > > the
> > > > > >> purpose
> > > > > >> > > > of
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > threads
> > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > group
> > > > (or
> > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > > should we
> > > > > >> need
> > > > > >> > > > > > > > > multi-threads
> > > > > >> > > > > > > > > > > here?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > 2. For the default value in the new configs:
> > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> > the
> > > > > >> default
> > > > > >> > > > session
> > > > > >> > > > > > > > > timeout not
> > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > thought
> > > > the
> > > > > >> min/max
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > > no?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > > >> timeout to
> > > > > >> > > > detect
> > > > > >> > > > > > client
> > > > > >> > > > > > > > > > > failures when using the consumer group
> > protocol.
> > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > > The
> > > > > >> minimum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > > The
> > > > > >> maximum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > [range,
> > > > > >> uniform],
> > > > > >> > > > > > which means
> > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > > know
> > > > > >> why not
> > > > > >> > > > > > uniform
> > > > > >> > > > > > > > > one? I
> > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > assignor
> > > > > >> (former
> > > > > >> > > > sticky
> > > > > >> > > > > > > > > assinor)
> > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > reason
> > > > we
> > > > > >> choose
> > > > > >> > > > range
> > > > > >> > > > > > > > > assignor
> > > > > >> > > > > > > > > > > as default?
> > > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> > The
> > > > > >> server side
> > > > > >> > > > > > assignors.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Thank you.
> > > > > >> > > > > > > > > > > Luke
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > > > >> showuon@gmail.com
> > > > > >> > > > >
> > > > > >> > > > > > wrote:
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > > >> integrating with
> > > > > >> > > > > > KIP-848,
> > > > > >> > > > > > > > > but I
> > > > > >> > > > > > > > > > > > think we should have a separate discussion
> > > > thread
> > > > > >> for the
> > > > > >> > > > Kafka
> > > > > >> > > > > > > > > Connect
> > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > > Consumer
> > > > > >> Rebalance
> > > > > >> > > > > > Protocol
> > > > > >> > > > > > > > > [1],
> > > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > > consumer
> > > > > >> rebalance
> > > > > >> > > > > > protocol,
> > > > > >> > > > > > > > > WDYT?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > [1]
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > >
> > > > > >> > > > > >
> > > > > >> > > >
> > > > > >>
> > > >
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > Thank you.
> > > > > >> > > > > > > > > > > > Luke
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > >> > > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > > > Looks
> > > > > >> like
> > > > > >> > > > there's
> > > > > >> > > > > > > > > agreement on
> > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> > > > revisit
> > > > > >> the doc
> > > > > >> > > > and
> > > > > >> > > > > > see
> > > > > >> > > > > > > > > what
> > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > > > > >> Thanks!
> > > > > >> > > > > > > > > > > >> Sagar.
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > >> > > > > > > > > > > >> wrote:
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > > > That's
> > > > > >> really
> > > > > >> > > > > > helpful. I
> > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > > > and the
> > > > > >> > > > Consumer
> > > > > >> > > > > > could
> > > > > >> > > > > > > > > share
> > > > > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > > > > >> difference is
> > > > > >> > > > that
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > Consumer
> > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > whereas
> > > > > >> Connect
> > > > > >> > > > assigns
> > > > > >> > > > > > tasks
> > > > > >> > > > > > > > > to
> > > > > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > > support
> > > > > >> different
> > > > > >> > > > > > resource
> > > > > >> > > > > > > > > types
> > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect.
> > > The
> > > > > >> dedicated
> > > > > >> > > > APIs
> > > > > >> > > > > > > > > would be
> > > > > >> > > > > > > > > > > >> > similar to the new ones but different on
> > > the
> > > > > >> > > > > > content/resources and
> > > > > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > > > > >> coordinator
> > > > > >> > > > side.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > > > not a
> > > > > >> fan of
> > > > > >> > > > > > > > > overcharging
> > > > > >> > > > > > > > > > > >> > APIs to serve different purposes. That
> > > being
> > > > > >> said, I am
> > > > > >> > > > not
> > > > > >> > > > > > > > > opposed to
> > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do
> > it.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > I think that we can continue to discuss
> > it
> > > > here
> > > > > >> for now
> > > > > >> > > > in
> > > > > >> > > > > > order
> > > > > >> > > > > > > > > to
> > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible with
> > > what
> > > > we
> > > > > >> will do
> > > > > >> > > > for
> > > > > >> > > > > > > > > Connect in
> > > > > >> > > > > > > > > > > >> > the future.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > Best,
> > > > > >> > > > > > > > > > > >> > David
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> > Jacot
> > > <
> > > > > >> > > > > > djacot@confluent.io>
> > > > > >> > > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > > through
> > > > and
> > > > > >> address
> > > > > >> > > > > > your
> > > > > >> > > > > > > > > comments
> > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > > feedback.
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > >> > > > > > > > > > > >> > > David
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > > > Harris
> > > > > >> <
> > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > wrote:
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to
> > > see
> > > > > >> > > > cooperative
> > > > > >> > > > > > > > > rebalancing
> > > > > >> > > > > > > > > > > >> > making it
> > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > section
> > > > > >> describes how
> > > > > >> > > > > > member
> > > > > >> > > > > > > > > epoch
> > > > > >> > > > > > > > > > > >> > should
> > > > > >> > > > > > > > > > > >> > > > advance in step with the group epoch
> > > and
> > > > > >> assignment
> > > > > >> > > > > > epoch
> > > > > >> > > > > > > > > values. I
> > > > > >> > > > > > > > > > > >> > think
> > > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > > reasons
> > > > > >> described
> > > > > >> > > > in
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > KIP. When
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > > > >> partitions to a
> > > > > >> > > > > > worker,
> > > > > >> > > > > > > > > what
> > > > > >> > > > > > > > > > > >> member
> > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > assignment
> > > > use?
> > > > > >> Are
> > > > > >> > > > member
> > > > > >> > > > > > epochs
> > > > > >> > > > > > > > > > > >> re-used,
> > > > > >> > > > > > > > > > > >> > and
> > > > > >> > > > > > > > > > > >> > > > a single member epoch can correspond
> > to
> > > > > >> multiple
> > > > > >> > > > > > different
> > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > > > opaque
> > > > > >> to the
> > > > > >> > > > group
> > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > assignor
> > > > > >> > > > implementations
> > > > > >> > > > > > > > > interact
> > > > > >> > > > > > > > > > > >> with
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> > > > meaning
> > > > > >> agreed
> > > > > >> > > > > > upon? If
> > > > > >> > > > > > > > > so, what
> > > > > >> > > > > > > > > > > >> > is the
> > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field
> > over
> > > > > >> including
> > > > > >> > > > such
> > > > > >> > > > > > > > > functionality
> > > > > >> > > > > > > > > > > >> > in the
> > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 3. The following is included in the
> > > KIP:
> > > > > >> "Thanks to
> > > > > >> > > > > > this, the
> > > > > >> > > > > > > > > input
> > > > > >> > > > > > > > > > > >> of
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> > driven
> > > > by
> > > > > >> the group
> > > > > >> > > > > > > > > coordinator.
> > > > > >> > > > > > > > > > > >> The
> > > > > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > > > >> maintaining
> > > > > >> > > > any
> > > > > >> > > > > > state
> > > > > >> > > > > > > > > besides
> > > > > >> > > > > > > > > > > >> its
> > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > > > that the
> > > > > >> > > > > > client-side
> > > > > >> > > > > > > > > assignor
> > > > > >> > > > > > > > > > > >> MAY
> > > > > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> > > state
> > > > > >> (such as
> > > > > >> > > > > > partition
> > > > > >> > > > > > > > > > > >> > throughput,
> > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics,
> > > etc),
> > > > or
> > > > > >> that
> > > > > >> > > > > > additional
> > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > > classes
> > > > > >> > > > > > > > > > > >> > > > for
> > > > > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > >> > > > > > > > > > > >> > > > and
> > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > >> > > > > > that
> > > > > >> > > > > > > > > seem to
> > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it possible
> > > for
> > > > > >> these two
> > > > > >> > > > > > > > > implementations
> > > > > >> > > > > > > > > > > >> to
> > > > > >> > > > > > > > > > > >> > be
> > > > > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> > > > feature
> > > > > >> parity
> > > > > >> > > > of
> > > > > >> > > > > > > > > server-side
> > > > > >> > > > > > > > > > > >> and
> > > > > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > > > >> facilitate
> > > > > >> > > > > > operational
> > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > > > >> server-side
> > > > > >> > > > > > assignor
> > > > > >> > > > > > > > > has some
> > > > > >> > > > > > > > > > > >> > poor
> > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch, deploying
> > > the
> > > > > >> patched
> > > > > >> > > > > > assignor to
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > client
> > > > > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > > > >> client-side
> > > > > >> > > > > > assignor
> > > > > >> > > > > > > > > may be
> > > > > >> > > > > > > > > > > >> > faster
> > > > > >> > > > > > > > > > > >> > > > and less risky than patching all of
> > the
> > > > > >> brokers.
> > > > > >> > > > With
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > currently
> > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > > > >> > > > reimplementation
> > > > > >> > > > > > would
> > > > > >> > > > > > > > > have
> > > > > >> > > > > > > > > > > >> to be
> > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > > > diverged
> > > > > >> > > > > > significantly,
> > > > > >> > > > > > > > > then it
> > > > > >> > > > > > > > > > > >> is
> > > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> > would
> > > > not be
> > > > > >> > > > possible.
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > --
> > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar
> > <
> > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > >> > > > > > > > > > > >> > wrote:
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > > discuss
> > > > how
> > > > > >> Connect
> > > > > >> > > > > > would
> > > > > >> > > > > > > > > need to
> > > > > >> > > > > > > > > > > >> > change
> > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > protocol.
> > > > > >> Here's the
> > > > > >> > > > page:
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > >
> > > > > >> > > > > >
> > > > > >> > > >
> > > > > >>
> > > >
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > > > tried
> > > > > >> to keep
> > > > > >> > > > a
> > > > > >> > > > > > format
> > > > > >> > > > > > > > > > > >> similar to
> > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > think.
> > > > Also,
> > > > > >> do you
> > > > > >> > > > > > think this
> > > > > >> > > > > > > > > > > >> should
> > > > > >> > > > > > > > > > > >> > be
> > > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> > thread
> > > > or
> > > > > >> is this
> > > > > >> > > > one
> > > > > >> > > > > > fine?
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > > Sagar <
> > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > >> > > > > > > > > > > >> > wrote:
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > > > Kafka
> > > > > >> Streams.
> > > > > >> > > > > > Sure, I
> > > > > >> > > > > > > > > would do
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > > > >> document.
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > > > Guozhang
> > > > > >> Wang <
> > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you could
> > > come
> > > > > >> back with
> > > > > >> > > > some
> > > > > >> > > > > > > > > analysis on
> > > > > >> > > > > > > > > > > >> > how to
> > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > integration
> > > > > >> with
> > > > > >> > > > the new
> > > > > >> > > > > > > > > protocol;
> > > > > >> > > > > > > > > > > >> so
> > > > > >> > > > > > > > > > > >> > far
> > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > > > "protocol
> > > > > >> type"
> > > > > >> > > > we
> > > > > >> > > > > > did not
> > > > > >> > > > > > > > > yet
> > > > > >> > > > > > > > > > > >> think
> > > > > >> > > > > > > > > > > >> > > > > through
> > > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> > implementations.
> > > > For
> > > > > >> Streams
> > > > > >> > > > > > here's a
> > > > > >> > > > > > > > > draft of
> > > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > >
> > > > > >> > > > > >
> > > > > >> > > >
> > > > > >>
> > > >
> > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> > > > Connect.
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> > > > Sagar <
> > > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your response.
> > The
> > > > > >> reason I
> > > > > >> > > > thought
> > > > > >> > > > > > > > > connect can
> > > > > >> > > > > > > > > > > >> > also fit
> > > > > >> > > > > > > > > > > >> > > > > >> into
> > > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that even
> > > > today the
> > > > > >> > > > connect
> > > > > >> > > > > > uses a
> > > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > > AbstractCoordinator to
> > > > > >> empower
> > > > > >> > > > > > > > > rebalances of
> > > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > > >> WorkerCoordinator sets
> > > > > >> > > > the
> > > > > >> > > > > > > > > > > >> protocolType()
> > > > > >> > > > > > > > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata() method
> > > by
> > > > > >> plumbing
> > > > > >> > > > into
> > > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > > > > >> connect would
> > > > > >> > > > be
> > > > > >> > > > > > > > > similar at a
> > > > > >> > > > > > > > > > > >> > high
> > > > > >> > > > > > > > > > > >> > > > > level
> > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> > > > because
> > > > > >> of the
> > > > > >> > > > > > Client
> > > > > >> > > > > > > > > side
> > > > > >> > > > > > > > > > > >> > assignors
> > > > > >> > > > > > > > > > > >> > > > > being
> > > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > > implementation
> > > > > >> level, we
> > > > > >> > > > > > might
> > > > > >> > > > > > > > > need to
> > > > > >> > > > > > > > > > > >> make
> > > > > >> > > > > > > > > > > >> > a lot
> > > > > >> > > > > > > > > > > >> > > > > of
> > > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > > > > >> assignment
> > > > > >> > > > protocol
> > > > > >> > > > > > like
> > > > > >> > > > > > > > > > > >> enhancing
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > > > > >> SyncGroup and
> > > > > >> > > > > > using
> > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar lines
> > > to
> > > > > >> streams (or
> > > > > >> > > > > > there
> > > > > >> > > > > > > > > might be
> > > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> > detailed
> > > > > >> analysis of
> > > > > >> > > > the
> > > > > >> > > > > > same
> > > > > >> > > > > > > > > and we
> > > > > >> > > > > > > > > > > >> > can have
> > > > > >> > > > > > > > > > > >> > > > > a
> > > > > >> > > > > > > > > > > >> > > > > >> > separate discussion thread for
> > > > that
> > > > > >> as that
> > > > > >> > > > would
> > > > > >> > > > > > > > > derail this
> > > > > >> > > > > > > > > > > >> > > > > discussion
> > > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if that
> > > sounds
> > > > > >> good to
> > > > > >> > > > you.
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47
> > PM
> > > > David
> > > > > >> Jacot
> > > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > > >> `Assignment#error`.
> > > > > >> > > > > > Sure, I
> > > > > >> > > > > > > > > can
> > > > > >> > > > > > > > > > > >> > mention it.
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> > transition C
> > > > from
> > > > > >> his
> > > > > >> > > > current
> > > > > >> > > > > > > > > assignment
> > > > > >> > > > > > > > > > > >> to
> > > > > >> > > > > > > > > > > >> > his
> > > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when he
> > can
> > > > move
> > > > > >> to
> > > > > >> > > > epoch 3.
> > > > > >> > > > > > When
> > > > > >> > > > > > > > > that
> > > > > >> > > > > > > > > > > >> > happens,
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is updated
> > > and
> > > > > >> persisted
> > > > > >> > > > > > with all
> > > > > >> > > > > > > > > its
> > > > > >> > > > > > > > > > > >> > assigned
> > > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if they are
> > > not
> > > > all
> > > > > >> revoked
> > > > > >> > > > > > yet. In
> > > > > >> > > > > > > > > other
> > > > > >> > > > > > > > > > > >> > words, the
> > > > > >> > > > > > > > > > > >> > > > > >> > > member assignment becomes
> > the
> > > > target
> > > > > >> > > > > > assignment.
> > > > > >> > > > > > > > > This is
> > > > > >> > > > > > > > > > > >> > basically
> > > > > >> > > > > > > > > > > >> > > > > an
> > > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid having
> > > to
> > > > > >> write all
> > > > > >> > > > the
> > > > > >> > > > > > > > > changes to
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > log.
> > > > > >> > > > > > > > > > > >> > > > > The
> > > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on the
> > > > persisted
> > > > > >> state
> > > > > >> > > > so I
> > > > > >> > > > > > > > > understand
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I
> > can
> > > > > >> improve
> > > > > >> > > > this in
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > > >> > description.
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it
> > could
> > > > > >> reuse the
> > > > > >> > > > > > protocol
> > > > > >> > > > > > > > > with a
> > > > > >> > > > > > > > > > > >> > client side
> > > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > > > > >> protocol. The
> > > > > >> > > > > > assignment
> > > > > >> > > > > > > > > is
> > > > > >> > > > > > > > > > > >> about
> > > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> > metadata,
> > > > could
> > > > > >> > > > Connect
> > > > > >> > > > > > fit
> > > > > >> > > > > > > > > into this?
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55
> > > PM
> > > > > >> Sagar <
> > > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just
> > > had
> > > > > >> minor
> > > > > >> > > > > > observations:
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > > > > >> section in
> > > > > >> > > > Client
> > > > > >> > > > > > Side
> > > > > >> > > > > > > > > mode
> > > > > >> > > > > > > > > > > >> > Assignment
> > > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this
> > > > case,
> > > > > >> the
> > > > > >> > > > client
> > > > > >> > > > > > side
> > > > > >> > > > > > > > > assignor
> > > > > >> > > > > > > > > > > >> can
> > > > > >> > > > > > > > > > > >> > > > > return
> > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > > coordinator`.
> > > > > >> In this
> > > > > >> > > > > > case are
> > > > > >> > > > > > > > > you
> > > > > >> > > > > > > > > > > >> > referring to
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > > > >> AssignmentError
> > > > > >> > > > that's
> > > > > >> > > > > > > > > listed down
> > > > > >> > > > > > > > > > > >> > towards
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it
> > > would
> > > > > >> make sense
> > > > > >> > > > to
> > > > > >> > > > > > > > > mention this
> > > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > > > section, I
> > > > > >> have a
> > > > > >> > > > > > slight
> > > > > >> > > > > > > > > > > >> confusion,
> > > > > >> > > > > > > > > > > >> > not
> > > > > >> > > > > > > > > > > >> > > > > sure
> > > > > >> > > > > > > > > > > >> > > > > >> if
> > > > > >> > > > > > > > > > > >> > > > > >> > > > others have the same.
> > > Consider
> > > > > >> this step:
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the
> > group
> > > > > >> coordinator
> > > > > >> > > > > > > > > transitions him
> > > > > >> > > > > > > > > > > >> to
> > > > > >> > > > > > > > > > > >> > epoch
> > > > > >> > > > > > > > > > > >> > > > > 3
> > > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > > > revoke. It
> > > > > >> > > > persists
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > change and
> > > > > >> > > > > > > > > > > >> > reply.
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > > > (epoch=3)
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > > partitions=[foo-0]
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > > partitions=[foo-2]
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > > partitions=[foo-1]
> > > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > > >> partitions=[foo-0,
> > > > > >> > > > > > foo-1]
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > > >> partitions=[foo-2]
> > > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > > >> partitions=[foo-1]
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > > > transitions
> > > > > >> to
> > > > > >> > > > epoch 3
> > > > > >> > > > > > but
> > > > > >> > > > > > > > > cannot
> > > > > >> > > > > > > > > > > >> get
> > > > > >> > > > > > > > > > > >> > foo-1
> > > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that
> > > > member C
> > > > > >> can't
> > > > > >> > > > get
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > foo-1
> > > > > >> > > > > > > > > > > >> > partition
> > > > > >> > > > > > > > > > > >> > > > > yet,
> > > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > > >> > > > > > > > > > > >> > > > > >> > > > based on the description
> > > > above,
> > > > > >> it seems
> > > > > >> > > > it
> > > > > >> > > > > > > > > already has
> > > > > >> > > > > > > > > > > >> it.
> > > > > >> > > > > > > > > > > >> > Do you
> > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to remove
> > it
> > > > and
> > > > > >> > > > populate it
> > > > > >> > > > > > only
> > > > > >> > > > > > > > > when it
> > > > > >> > > > > > > > > > > >> > actually
> > > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of
> > other
> > > > > >> places, so
> > > > > >> > > > have
> > > > > >> > > > > > I
> > > > > >> > > > > > > > > > > >> understood it
> > > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it
> > might
> > > > be
> > > > > >> out of
> > > > > >> > > > scope
> > > > > >> > > > > > of
> > > > > >> > > > > > > > > this
> > > > > >> > > > > > > > > > > >> > discussion,
> > > > > >> > > > > > > > > > > >> > > > > but
> > > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > > > > >> probably be
> > > > > >> > > > > > running in
> > > > > >> > > > > > > > > client
> > > > > >> > > > > > > > > > > >> > side
> > > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > > > > >> protocol as it
> > > > > >> > > > has
> > > > > >> > > > > > its
> > > > > >> > > > > > > > > own
> > > > > >> > > > > > > > > > > >> Custom
> > > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > >> > > > > > > > > > > >> > > > > >> > > > and
> > IncrementalCooperative).
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at
> > 5:00
> > > > PM
> > > > > >> David
> > > > > >> > > > Jacot
> > > > > >> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal
> > is
> > > > to
> > > > > >> move
> > > > > >> > > > forward
> > > > > >> > > > > > with
> > > > > >> > > > > > > > > > > >> > specialized API
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying on
> > one
> > > > > >> generic API.
> > > > > >> > > > For
> > > > > >> > > > > > > > > Connect, we
> > > > > >> > > > > > > > > > > >> > can apply
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > > > > >> reuse/share the
> > > > > >> > > > core
> > > > > >> > > > > > > > > > > >> > implementation on
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the
> > > schema
> > > > > >> registry, I
> > > > > >> > > > > > think
> > > > > >> > > > > > > > > that we
> > > > > >> > > > > > > > > > > >> > should
> > > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to
> > > do
> > > > > >> simple
> > > > > >> > > > > > > > > membership/leader
> > > > > >> > > > > > > > > > > >> > election.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at
> > > > 10:22
> > > > > >> AM Ismael
> > > > > >> > > > > > Juma <
> > > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > > >> java.util.regex.Pattern
> > > > > >> > > > vs
> > > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> > differences
> > > > in
> > > > > >> more
> > > > > >> > > > detail
> > > > > >> > > > > > before
> > > > > >> > > > > > > > > > > >> deciding
> > > > > >> > > > > > > > > > > >> > one
> > > > > >> > > > > > > > > > > >> > > > > way
> > > > > >> > > > > > > > > > > >> > > > > >> or
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if people
> > > pass
> > > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > > >> > > > > > > > > > > >> they
> > > > > >> > > > > > > > > > > >> > expect
> > > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be
> > honored.
> > > > If
> > > > > >> we are
> > > > > >> > > > doing
> > > > > >> > > > > > > > > something
> > > > > >> > > > > > > > > > > >> > different,
> > > > > >> > > > > > > > > > > >> > > > > >> then
> > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider adding
> > > an
> > > > > >> overload
> > > > > >> > > > with
> > > > > >> > > > > > our own
> > > > > >> > > > > > > > > > > >> Pattern
> > > > > >> > > > > > > > > > > >> > class
> > > > > >> > > > > > > > > > > >> > > > > (I
> > > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to
> > > expose
> > > > > >> re2j's at
> > > > > >> > > > this
> > > > > >> > > > > > > > > point).
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic
> > ids,
> > > > any
> > > > > >> major new
> > > > > >> > > > > > protocol
> > > > > >> > > > > > > > > should
> > > > > >> > > > > > > > > > > >> > > > > integrate
> > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should handle
> > the
> > > > topic
> > > > > >> > > > > > recreation case
> > > > > >> > > > > > > > > > > >> > correctly.
> > > > > >> > > > > > > > > > > >> > > > > That's
> > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to
> > handle.
> > > I
> > > > > >> agree with
> > > > > >> > > > > > David
> > > > > >> > > > > > > > > that we'd
> > > > > >> > > > > > > > > > > >> > want to
> > > > > >> > > > > > > > > > > >> > > > > add
> > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > > protocols
> > > > > >> that don't
> > > > > >> > > > > > have it
> > > > > >> > > > > > > > > yet and
> > > > > >> > > > > > > > > > > >> > that we
> > > > > >> > > > > > > > > > > >> > > > > can
> > > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> > > > versus
> > > > > >> adding
> > > > > >> > > > new
> > > > > >> > > > > > APIs
> > > > > >> > > > > > > > > to the
> > > > > >> > > > > > > > > > > >> > Java
> > > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that adding
> > new
> > > > APIs
> > > > > >> is
> > > > > >> > > > required
> > > > > >> > > > > > for
> > > > > >> > > > > > > > > > > >> reasonable
> > > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> > > > about
> > > > > >> the
> > > > > >> > > > > > coordinator
> > > > > >> > > > > > > > > storing
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs
> > to
> > > be
> > > > > >> > > > centralized
> > > > > >> > > > > > in the
> > > > > >> > > > > > > > > > > >> metadata
> > > > > >> > > > > > > > > > > >> > log for
> > > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > > > > >> visibility,
> > > > > >> > > > > > consistency,
> > > > > >> > > > > > > > > > > >> etc.).
> > > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> > automatically
> > > > > >> deleting
> > > > > >> > > > > > configs in a
> > > > > >> > > > > > > > > way
> > > > > >> > > > > > > > > > > >> that
> > > > > >> > > > > > > > > > > >> > they
> > > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> > > property
> > > > > >> for modern
> > > > > >> > > > > > systems
> > > > > >> > > > > > > > > is to
> > > > > >> > > > > > > > > > > >> > minimize
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data
> > loss
> > > > > >> scenarios.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022
> > at
> > > > 3:47
> > > > > >> PM David
> > > > > >> > > > > > Jacot
> > > > > >> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> > > > answers
> > > > > >> are
> > > > > >> > > > below:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > > path,
> > > > > >> especially
> > > > > >> > > > > > the last
> > > > > >> > > > > > > > > step
> > > > > >> > > > > > > > > > > >> when
> > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > > > >> protocol, in
> > > > > >> > > > which
> > > > > >> > > > > > we
> > > > > >> > > > > > > > > would
> > > > > >> > > > > > > > > > > >> have a
> > > > > >> > > > > > > > > > > >> > > > > window
> > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs
> > and
> > > > old
> > > > > >> > > > protocols /
> > > > > >> > > > > > rpcs
> > > > > >> > > > > > > > > are
> > > > > >> > > > > > > > > > > >> used
> > > > > >> > > > > > > > > > > >> > by
> > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > >> coordinator could
> > > > > >> > > > > > "mimic"
> > > > > >> > > > > > > > > the old
> > > > > >> > > > > > > > > > > >> > behavior
> > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > > something
> > > > > >> we
> > > > > >> > > > need to
> > > > > >> > > > > > > > > present
> > > > > >> > > > > > > > > > > >> about.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > > published
> > > > a
> > > > > >> new
> > > > > >> > > > version
> > > > > >> > > > > > of KIP
> > > > > >> > > > > > > > > which
> > > > > >> > > > > > > > > > > >> > includes
> > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about this.
> > > See
> > > > the
> > > > > >> > > > "Supporting
> > > > > >> > > > > > > > > Online
> > > > > >> > > > > > > > > > > >> > Consumer
> > > > > >> > > > > > > > > > > >> > > > > Group
> > > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > > "Compatibility,
> > > > > >> > > > Deprecation,
> > > > > >> > > > > > and
> > > > > >> > > > > > > > > Migration
> > > > > >> > > > > > > > > > > >> > Plan". I
> > > > > >> > > > > > > > > > > >> > > > > >> think
> > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> > > through
> > > > a
> > > > > >> few cases
> > > > > >> > > > > > now but
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > overall idea
> > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > > > > >> understandable.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> > > topic
> > > > > >> ids. So
> > > > > >> > > > far as
> > > > > >> > > > > > > > > KIP-516 the
> > > > > >> > > > > > > > > > > >> > topic ids
> > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > > admin
> > > > > >> client,
> > > > > >> > > > but
> > > > > >> > > > > > they
> > > > > >> > > > > > > > > are not
> > > > > >> > > > > > > > > > > >> > exposed
> > > > > >> > > > > > > > > > > >> > > > > via
> > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet.
> > I
> > > > think
> > > > > >> the
> > > > > >> > > > > > question is,
> > > > > >> > > > > > > > > first
> > > > > >> > > > > > > > > > > >> > should we
> > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > maintaining
> > > > > >> the
> > > > > >> > > > names
> > > > > >> > > > > > -> ids
> > > > > >> > > > > > > > > mapping
> > > > > >> > > > > > > > > > > >> > itself
> > > > > >> > > > > > > > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > augmented
> > > > > >> existing
> > > > > >> > > > RPCs
> > > > > >> > > > > > and the
> > > > > >> > > > > > > > > new
> > > > > >> > > > > > > > > > > >> RPCs
> > > > > >> > > > > > > > > > > >> > with
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should
> > we
> > > > ever
> > > > > >> consider
> > > > > >> > > > > > exposing
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > topic ids
> > > > > >> > > > > > > > > > > >> > > > > in
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > well
> > > > (both
> > > > > >> > > > > > > > > subscribe/assign, as
> > > > > >> > > > > > > > > > > >> well
> > > > > >> > > > > > > > > > > >> > as in
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > > > like
> > > > > >> topic
> > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we
> > > > would
> > > > > >> include
> > > > > >> > > > > > converting
> > > > > >> > > > > > > > > all
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > offsets
> > > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic
> > > ids
> > > > in
> > > > > >> this
> > > > > >> > > > KIP,
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > consumer
> > > > > >> > > > > > > > > > > >> > would be
> > > > > >> > > > > > > > > > > >> > > > > >> able
> > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with
> > > topic
> > > > > >> ids. That
> > > > > >> > > > > > being
> > > > > >> > > > > > > > > said, it
> > > > > >> > > > > > > > > > > >> > still has
> > > > > >> > > > > > > > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> > > > various
> > > > > >> APIs so
> > > > > >> > > > > > having a
> > > > > >> > > > > > > > > > > >> mapping
> > > > > >> > > > > > > > > > > >> > in the
> > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to
> > > me.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a
> > > strong
> > > > > >> opinion on
> > > > > >> > > > > > this.
> > > > > >> > > > > > > > > Here I
> > > > > >> > > > > > > > > > > >> > wonder if
> > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of
> > > this
> > > > > >> KIP. I
> > > > > >> > > > would
> > > > > >> > > > > > rather
> > > > > >> > > > > > > > > focus
> > > > > >> > > > > > > > > > > >> on
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> > > consider
> > > > > >> this
> > > > > >> > > > > > separately if
> > > > > >> > > > > > > > > we see
> > > > > >> > > > > > > > > > > >> > value in
> > > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > > Ismael's
> > > > > >> point about
> > > > > >> > > > > > using
> > > > > >> > > > > > > > > topic ids
> > > > > >> > > > > > > > > > > >> > in the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> ConsumerGroupHeartbeatRequest, I
> > > > > >> > > > think
> > > > > >> > > > > > that
> > > > > >> > > > > > > > > there
> > > > > >> > > > > > > > > > > >> is
> > > > > >> > > > > > > > > > > >> > one
> > > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> > > > consumer
> > > > > >> will
> > > > > >> > > > have
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > > >> opportunity
> > > > > >> > > > > > > > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists
> > > before
> > > > > >> passing
> > > > > >> > > > them
> > > > > >> > > > > > into
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> group
> > > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously,
> > > the
> > > > > >> > > > coordinator
> > > > > >> > > > > > will
> > > > > >> > > > > > > > > also
> > > > > >> > > > > > > > > > > >> notice
> > > > > >> > > > > > > > > > > >> > it but
> > > > > >> > > > > > > > > > > >> > > > > >> it
> > > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have a
> > way
> > > to
> > > > > >> reject an
> > > > > >> > > > > > invalid
> > > > > >> > > > > > > > > topic in
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > > > David
> > > > > >> on all
> > > > > >> > > > other
> > > > > >> > > > > > minor
> > > > > >> > > > > > > > > > > >> questions
> > > > > >> > > > > > > > > > > >> > > > > except
> > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > `subscribe(Pattern)`
> > > > > >> question:
> > > > > >> > > > > > personally
> > > > > >> > > > > > > > > I think
> > > > > >> > > > > > > > > > > >> > it's not
> > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > subscribe
> > > > > >> API with
> > > > > >> > > > > > Pattern,
> > > > > >> > > > > > > > > but
> > > > > >> > > > > > > > > > > >> > instead we
> > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > documenting
> > > > > >> that our
> > > > > >> > > > > > > > > subscription may
> > > > > >> > > > > > > > > > > >> be
> > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > incompatible
> > > > > >> case is a
> > > > > >> > > > > > very rare
> > > > > >> > > > > > > > > > > >> scenario
> > > > > >> > > > > > > > > > > >> > I felt
> > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > `String`
> > > > based
> > > > > >> > > > > > subscription may
> > > > > >> > > > > > > > > be
> > > > > >> > > > > > > > > > > >> more
> > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I
> > > have
> > > > to
> > > > > >> look at
> > > > > >> > > > the
> > > > > >> > > > > > > > > differences
> > > > > >> > > > > > > > > > > >> > between
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > > > > >> understand the
> > > > > >> > > > > > potential
> > > > > >> > > > > > > > > issues.
> > > > > >> > > > > > > > > > > >> My
> > > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work for
> > > all
> > > > the
> > > > > >> basic
> > > > > >> > > > > > regular
> > > > > >> > > > > > > > > > > >> > expressions. The
> > > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> > > > mainly
> > > > > >> about
> > > > > >> > > > the
> > > > > >> > > > > > various
> > > > > >> > > > > > > > > > > >> character
> > > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other
> > > people
> > > > > >> think
> > > > > >> > > > about
> > > > > >> > > > > > this.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022
> > > at
> > > > > >> 11:28 PM
> > > > > >> > > > > > Guozhang
> > > > > >> > > > > > > > > Wang <
> > > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I
> > > think
> > > > on
> > > > > >> the high
> > > > > >> > > > > > level
> > > > > >> > > > > > > > > there are
> > > > > >> > > > > > > > > > > >> > two meta
> > > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a
> > bit
> > > > more:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > > path,
> > > > > >> especially
> > > > > >> > > > > > the last
> > > > > >> > > > > > > > > step
> > > > > >> > > > > > > > > > > >> when
> > > > > >> > > > > > > > > > > >> > > > > clients
> > > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > > > >> protocol, in
> > > > > >> > > > which
> > > > > >> > > > > > we
> > > > > >> > > > > > > > > would
> > > > > >> > > > > > > > > > > >> have a
> > > > > >> > > > > > > > > > > >> > > > > window
> > > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs
> > and
> > > > old
> > > > > >> > > > protocols /
> > > > > >> > > > > > rpcs
> > > > > >> > > > > > > > > are
> > > > > >> > > > > > > > > > > >> used
> > > > > >> > > > > > > > > > > >> > by
> > > > > >> > > > > > > > > > > >> > > > > >> members
> > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > > >> coordinator could
> > > > > >> > > > > > "mimic"
> > > > > >> > > > > > > > > the old
> > > > > >> > > > > > > > > > > >> > behavior
> > > > > >> > > > > > > > > > > >> > > > > >> while
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > > something
> > > > > >> we
> > > > > >> > > > need to
> > > > > >> > > > > > > > > present
> > > > > >> > > > > > > > > > > >> about.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> > > topic
> > > > > >> ids. So
> > > > > >> > > > far as
> > > > > >> > > > > > > > > KIP-516 the
> > > > > >> > > > > > > > > > > >> > topic ids
> > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > > admin
> > > > > >> client,
> > > > > >> > > > but
> > > > > >> > > > > > they
> > > > > >> > > > > > > > > are not
> > > > > >> > > > > > > > > > > >> > exposed
> > > > > >> > > > > > > > > > > >> > > > > via
> > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet.
> > I
> > > > think
> > > > > >> the
> > > > > >> > > > > > question is,
> > > > > >> > > > > > > > > first
> > > > > >> > > > > > > > > > > >> > should we
> > > > > >> > > > > > > > > > > >> > > > > >> let
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > > maintaining
> > > > > >> the
> > > > > >> > > > names
> > > > > >> > > > > > -> ids
> > > > > >> > > > > > > > > mapping
> > > > > >> > > > > > > > > > > >> > itself
> > > > > >> > > > > > > > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> > augmented
> > > > > >> existing
> > > > > >> > > > RPCs
> > > > > >> > > > > > and the
> > > > > >> > > > > > > > > new
> > > > > >> > > > > > > > > > > >> RPCs
> > > > > >> > > > > > > > > > > >> > with
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should
> > we
> > > > ever
> > > > > >> consider
> > > > > >> > > > > > exposing
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > topic ids
> > > > > >> > > > > > > > > > > >> > > > > in
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> > well
> > > > (both
> > > > > >> > > > > > > > > subscribe/assign, as
> > > > > >> > > > > > > > > > > >> well
> > > > > >> > > > > > > > > > > >> > as in
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > > > like
> > > > > >> topic
> > > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > > > David
> > > > > >> on all
> > > > > >> > > > other
> > > > > >> > > > > > minor
> > > > > >> > > > > > > > > > > >> questions
> > > > > >> > > > > > > > > > > >> > > > > except
> > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > `subscribe(Pattern)`
> > > > > >> question:
> > > > > >> > > > > > personally
> > > > > >> > > > > > > > > I think
> > > > > >> > > > > > > > > > > >> > it's not
> > > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > > subscribe
> > > > > >> API with
> > > > > >> > > > > > Pattern,
> > > > > >> > > > > > > > > but
> > > > > >> > > > > > > > > > > >> > instead we
> > > > > >> > > > > > > > > > > >> > > > > >> still
> > > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > > documenting
> > > > > >> that our
> > > > > >> > > > > > > > > subscription may
> > > > > >> > > > > > > > > > > >> be
> > > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > > incompatible
> > > > > >> case is a
> > > > > >> > > > > > very rare
> > > > > >> > > > > > > > > > > >> scenario
> > > > > >> > > > > > > > > > > >> > I felt
> > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> > `String`
> > > > based
> > > > > >> > > > > > subscription may
> > > > > >> > > > > > > > > be
> > > > > >> > > > > > > > > > > >> more
> > > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12,
> > 2022
> > > > at
> > > > > >> 5:23 AM
> > > > > >> > > > > > David Jacot
> > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > <djacot@confluent.io.invalid
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > > > > >> feedback. Let
> > > > > >> > > > me
> > > > > >> > > > > > answer
> > > > > >> > > > > > > > > your
> > > > > >> > > > > > > > > > > >> > questions
> > > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think
> > it's
> > > > > >> premature to
> > > > > >> > > > > > talk about
> > > > > >> > > > > > > > > > > >> target
> > > > > >> > > > > > > > > > > >> > > > > versions
> > > > > >> > > > > > > > > > > >> > > > > >> for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > > > > >> existing group
> > > > > >> > > > > > protocol.
> > > > > >> > > > > > > > > > > >> Unlike
> > > > > >> > > > > > > > > > > >> > KRaft,
> > > > > >> > > > > > > > > > > >> > > > > >> this
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> > protocol
> > > > and
> > > > > >> hence
> > > > > >> > > > > > > > > deprecation/removal
> > > > > >> > > > > > > > > > > >> > will be
> > > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > > > > >> applications
> > > > > >> > > > migrate
> > > > > >> > > > > > to
> > > > > >> > > > > > > > > the new
> > > > > >> > > > > > > > > > > >> > protocol.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes
> > sense.
> > > I
> > > > will
> > > > > >> > > > remove it.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP
> > says
> > > we
> > > > > >> intend to
> > > > > >> > > > > > release
> > > > > >> > > > > > > > > this in
> > > > > >> > > > > > > > > > > >> > 4.x, but
> > > > > >> > > > > > > > > > > >> > > > > it
> > > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we
> > added
> > > > that
> > > > > >> as a
> > > > > >> > > > way to
> > > > > >> > > > > > > > > estimate
> > > > > >> > > > > > > > > > > >> when
> > > > > >> > > > > > > > > > > >> > we'd
> > > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > > > protocol, I
> > > > > >> also
> > > > > >> > > > > > suggest
> > > > > >> > > > > > > > > removing
> > > > > >> > > > > > > > > > > >> > this part.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain
> > my
> > > > > >> reasoning. As
> > > > > >> > > > > > > > > explained, I
> > > > > >> > > > > > > > > > > >> plan
> > > > > >> > > > > > > > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in
> > > Java
> > > > > >> while we
> > > > > >> > > > > > implement
> > > > > >> > > > > > > > > the new
> > > > > >> > > > > > > > > > > >> > protocol.
> > > > > >> > > > > > > > > > > >> > > > > >> This
> > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> > internals
> > > > will
> > > > > >> be
> > > > > >> > > > slightly
> > > > > >> > > > > > > > > different
> > > > > >> > > > > > > > > > > >> > (e.g.
> > > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I
> > > wanted
> > > > to
> > > > > >> tighten
> > > > > >> > > > the
> > > > > >> > > > > > > > > switch from
> > > > > >> > > > > > > > > > > >> > the old
> > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to
> > the
> > > > new
> > > > > >> group
> > > > > >> > > > > > coordinator
> > > > > >> > > > > > > > > to a
> > > > > >> > > > > > > > > > > >> > major
> > > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative
> > would
> > > > be to
> > > > > >> use a
> > > > > >> > > > flag
> > > > > >> > > > > > to do
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > switch
> > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> > > > upgrade.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to
> > > > flesh
> > > > > >> out the
> > > > > >> > > > > > details of
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > migration
> > > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying
> > we
> > > > will
> > > > > >> support
> > > > > >> > > > > > online
> > > > > >> > > > > > > > > > > >> migrations.
> > > > > >> > > > > > > > > > > >> > Is
> > > > > >> > > > > > > > > > > >> > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this
> > in
> > > > > >> detail. It
> > > > > >> > > > could
> > > > > >> > > > > > also
> > > > > >> > > > > > > > > be done
> > > > > >> > > > > > > > > > > >> > as a
> > > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> > > support
> > > > > >> online
> > > > > >> > > > > > migrations
> > > > > >> > > > > > > > > for the
> > > > > >> > > > > > > > > > > >> > group.
> > > > > >> > > > > > > > > > > >> > > > > That
> > > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using
> > the
> > > > old
> > > > > >> protocol
> > > > > >> > > > > > will be
> > > > > >> > > > > > > > > able to
> > > > > >> > > > > > > > > > > >> > switch to
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> > > > explain
> > > > > >> how that
> > > > > >> > > > > > will work
> > > > > >> > > > > > > > > > > >> though.
> > > > > >> > > > > > > > > > > >> > It is
> > > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step
> > process:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster
> > > must
> > > > be
> > > > > >> > > > upgraded or
> > > > > >> > > > > > > > > rolled to a
> > > > > >> > > > > > > > > > > >> > software
> > > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > > coordinator.
> > > > > >> Both the
> > > > > >> > > > > > old and
> > > > > >> > > > > > > > > the new
> > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > > > > >> protocol and
> > > > > >> > > > rely
> > > > > >> > > > > > on the
> > > > > >> > > > > > > > > same
> > > > > >> > > > > > > > > > > >> > persisted
> > > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> > > > together.
> > > > > >> This
> > > > > >> > > > point
> > > > > >> > > > > > is an
> > > > > >> > > > > > > > > > > >> offline
> > > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > > > > >> because it
> > > > > >> > > > would
> > > > > >> > > > > > require
> > > > > >> > > > > > > > > > > >> > shutting down
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > > > > >> starting up the
> > > > > >> > > > > > new one
> > > > > >> > > > > > > > > and
> > > > > >> > > > > > > > > > > >> that
> > > > > >> > > > > > > > > > > >> > would
> > > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > unavailabilities.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > > > > >> metadata
> > > > > >> > > > > > version/IBP
> > > > > >> > > > > > > > > must be
> > > > > >> > > > > > > > > > > >> > upgraded
> > > > > >> > > > > > > > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> X
> > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the
> > new
> > > > > >> protocol.
> > > > > >> > > > This
> > > > > >> > > > > > cannot
> > > > > >> > > > > > > > > be done
> > > > > >> > > > > > > > > > > >> > before
> > > > > >> > > > > > > > > > > >> > > > > 1)
> > > > > >> > > > > > > > > > > >> > > > > >> is
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated
> > because
> > > > the
> > > > > >> old
> > > > > >> > > > > > coordinator
> > > > > >> > > > > > > > > doesn't
> > > > > >> > > > > > > > > > > >> > support
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers
> > > > must be
> > > > > >> > > > upgraded
> > > > > >> > > > > > to a
> > > > > >> > > > > > > > > version
> > > > > >> > > > > > > > > > > >> > > > > supporting
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must
> > > have
> > > > > >> KIP-792).
> > > > > >> > > > If
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > consumer
> > > > > >> > > > > > > > > > > >> is
> > > > > >> > > > > > > > > > > >> > > > > already
> > > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be
> > > > done at
> > > > > >> this
> > > > > >> > > > point.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers
> > > > must
> > > > > >> be rolled
> > > > > >> > > > > > with the
> > > > > >> > > > > > > > > > > >> feature
> > > > > >> > > > > > > > > > > >> > flag
> > > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group
> > is
> > > > > >> automatically
> > > > > >> > > > > > > > > converted when
> > > > > >> > > > > > > > > > > >> > the first
> > > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> > > > protocol
> > > > > >> joins
> > > > > >> > > > the
> > > > > >> > > > > > group.
> > > > > >> > > > > > > > > While
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > members
> > > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> > > > being
> > > > > >> > > > upgraded,
> > > > > >> > > > > > the old
> > > > > >> > > > > > > > > > > >> protocol
> > > > > >> > > > > > > > > > > >> > is
> > > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify
> > all
> > > > of
> > > > > >> this in
> > > > > >> > > > the
> > > > > >> > > > > > KIP.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy
> > > > that we
> > > > > >> are
> > > > > >> > > > > > pushing the
> > > > > >> > > > > > > > > pattern
> > > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems
> > > like
> > > > > >> there
> > > > > >> > > > could be
> > > > > >> > > > > > some
> > > > > >> > > > > > > > > tricky
> > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> > mechanism
> > > > for
> > > > > >> users to
> > > > > >> > > > > > detect
> > > > > >> > > > > > > > > that
> > > > > >> > > > > > > > > > > >> they
> > > > > >> > > > > > > > > > > >> > need to
> > > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> > switching
> > > > to
> > > > > >> the new
> > > > > >> > > > > > protocol?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I
> > am
> > > a
> > > > bit
> > > > > >> more
> > > > > >> > > > > > optimistic
> > > > > >> > > > > > > > > than
> > > > > >> > > > > > > > > > > >> you
> > > > > >> > > > > > > > > > > >> > on this
> > > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > > > > >> majority of
> > > > > >> > > > the
> > > > > >> > > > > > cases
> > > > > >> > > > > > > > > are
> > > > > >> > > > > > > > > > > >> simple
> > > > > >> > > > > > > > > > > >> > > > > regexes
> > > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the
> > new
> > > > > >> engine. The
> > > > > >> > > > > > > > > coordinator will
> > > > > >> > > > > > > > > > > >> > verify
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> > > > consumer
> > > > > >> if the
> > > > > >> > > > > > regex is
> > > > > >> > > > > > > > > not
> > > > > >> > > > > > > > > > > >> valid.
> > > > > >> > > > > > > > > > > >> > > > > Coming
> > > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path,
> > in
> > > > the
> > > > > >> worst
> > > > > >> > > > case,
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > first
> > > > > >> > > > > > > > > > > >> > upgraded
> > > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will
> > be
> > > > > >> rejected.
> > > > > >> > > > This
> > > > > >> > > > > > should
> > > > > >> > > > > > > > > be used
> > > > > >> > > > > > > > > > > >> > as the
> > > > > >> > > > > > > > > > > >> > > > > >> last
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > > > customers to
> > > > > >> > > > validate
> > > > > >> > > > > > their
> > > > > >> > > > > > > > > regex
> > > > > >> > > > > > > > > > > >> > before
> > > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to
> > > > test
> > > > > >> them with
> > > > > >> > > > > > another
> > > > > >> > > > > > > > > group.
> > > > > >> > > > > > > > > > > >> For
> > > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done
> > in a
> > > > > >> pre-prod
> > > > > >> > > > > > environment.
> > > > > >> > > > > > > > > > > >> Another
> > > > > >> > > > > > > > > > > >> > way
> > > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > > >> consumer-group tool
> > > > > >> > > > to
> > > > > >> > > > > > > > > provide a
> > > > > >> > > > > > > > > > > >> regex
> > > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would
> > > > this be
> > > > > >> > > > enough in
> > > > > >> > > > > > your
> > > > > >> > > > > > > > > > > >> opinion?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to
> > > the
> > > > last
> > > > > >> > > > question,
> > > > > >> > > > > > will
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> Java
> > > > > >> > > > > > > > > > > >> > client
> > > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > > > > >> current regex
> > > > > >> > > > > > engine for
> > > > > >> > > > > > > > > > > >> > compatibility
> > > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be
> > handy
> > > to
> > > > > >> keep using
> > > > > >> > > > > > client
> > > > > >> > > > > > > > > based
> > > > > >> > > > > > > > > > > >> > regex at
> > > > > >> > > > > > > > > > > >> > > > > >> first
> > > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations
> > > simple
> > > > and
> > > > > >> then
> > > > > >> > > > > > migrate to
> > > > > >> > > > > > > > > server
> > > > > >> > > > > > > > > > > >> > based
> > > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand
> > your
> > > > point
> > > > > >> but I
> > > > > >> > > > am
> > > > > >> > > > > > > > > concerned that
> > > > > >> > > > > > > > > > > >> > this
> > > > > >> > > > > > > > > > > >> > > > > would
> > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay
> > > in
> > > > > >> this mode.
> > > > > >> > > > That
> > > > > >> > > > > > > > > would go
> > > > > >> > > > > > > > > > > >> > against our
> > > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> > > > client
> > > > > >> because
> > > > > >> > > > we
> > > > > >> > > > > > would
> > > > > >> > > > > > > > > have to
> > > > > >> > > > > > > > > > > >> > continue
> > > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on
> > > the
> > > > > >> client
> > > > > >> > > > side. I
> > > > > >> > > > > > would
> > > > > >> > > > > > > > > rather
> > > > > >> > > > > > > > > > > >> > not do
> > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say
> > > > that
> > > > > >> the group
> > > > > >> > > > > > > > > coordinator
> > > > > >> > > > > > > > > > > >> will
> > > > > >> > > > > > > > > > > >> > be
> > > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > > configurations
> > > > > >> and that
> > > > > >> > > > the
> > > > > >> > > > > > > > > > > >> configurations
> > > > > >> > > > > > > > > > > >> > will be
> > > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > > deleted.
> > > > > >> Will a
> > > > > >> > > > > > transition to
> > > > > >> > > > > > > > > DEAD
> > > > > >> > > > > > > > > > > >> > trigger
> > > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > configurations?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right.
> > The
> > > > > >> > > > configurations
> > > > > >> > > > > > will be
> > > > > >> > > > > > > > > > > >> deleted
> > > > > >> > > > > > > > > > > >> > when
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > > > > >> together.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the
> > > > choice to
> > > > > >> store
> > > > > >> > > > the
> > > > > >> > > > > > > > > configs in
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > group
> > > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list
> > > all
> > > > > >> cluster
> > > > > >> > > > > > configs and
> > > > > >> > > > > > > > > their
> > > > > >> > > > > > > > > > > >> > values?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think
> > so.
> > > > The
> > > > > >> group
> > > > > >> > > > > > > > > configurations are
> > > > > >> > > > > > > > > > > >> > overrides
> > > > > >> > > > > > > > > > > >> > > > > >> of
> > > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you
> > > > want to
> > > > > >> know
> > > > > >> > > > all
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > overrides
> > > > > >> > > > > > > > > > > >> > though,
> > > > > >> > > > > > > > > > > >> > > > > you
> > > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the
> > > group
> > > > > >> > > > coordinators.
> > > > > >> > > > > > You
> > > > > >> > > > > > > > > cannot
> > > > > >> > > > > > > > > > > >> > rely on
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> > > > someone
> > > > > >> > > > configure a
> > > > > >> > > > > > group
> > > > > >> > > > > > > > > before
> > > > > >> > > > > > > > > > > >> > starting
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > > > > >> allowing the
> > > > > >> > > > > > explicit
> > > > > >> > > > > > > > > creation
> > > > > >> > > > > > > > > > > >> of
> > > > > >> > > > > > > > > > > >> > > > > groups?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could
> > be
> > > > > >> decoupled
> > > > > >> > > > from
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > group
> > > > > >> > > > > > > > > > > >> > lifecycle.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group
> > > will
> > > > be
> > > > > >> > > > > > automatically
> > > > > >> > > > > > > > > created in
> > > > > >> > > > > > > > > > > >> > this
> > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will
> > > be
> > > > > >> lost after
> > > > > >> > > > the
> > > > > >> > > > > > > > > retention
> > > > > >> > > > > > > > > > > >> > period of
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > > > >> > > > Consumer.subscribe
> > > > > >> > > > > > method
> > > > > >> > > > > > > > > for the
> > > > > >> > > > > > > > > > > >> > Java
> > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > >> `java.util.regex.Pattern` of
> > > > > >> > > > do
> > > > > >> > > > > > we
> > > > > >> > > > > > > > > have to
> > > > > >> > > > > > > > > > > >> > introduce
> > > > > >> > > > > > > > > > > >> > > > > an
> > > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very
> > > group
> > > > > >> question. I
> > > > > >> > > > > > forgot
> > > > > >> > > > > > > > > about
> > > > > >> > > > > > > > > > > >> that
> > > > > >> > > > > > > > > > > >> > one.
> > > > > >> > > > > > > > > > > >> > > > > As
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> `java.util.regex.Pattern` is
> > > > > >> > > > not
> > > > > >> > > > > > fully
> > > > > >> > > > > > > > > > > >> compatible
> > > > > >> > > > > > > > > > > >> > with
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use,
> > it
> > > > > >> might be
> > > > > >> > > > better
> > > > > >> > > > > > to
> > > > > >> > > > > > > > > deprecate
> > > > > >> > > > > > > > > > > >> > it and
> > > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a
> > > > string.
> > > > > >> We would
> > > > > >> > > > > > rely on
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> server
> > > > > >> > > > > > > > > > > >> > side
> > > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > > migration, I
> > > > > >> think
> > > > > >> > > > that
> > > > > >> > > > > > we
> > > > > >> > > > > > > > > could
> > > > > >> > > > > > > > > > > >> still
> > > > > >> > > > > > > > > > > >> > try to
> > > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use
> > it.
> > > > That
> > > > > >> should
> > > > > >> > > > > > work, I
> > > > > >> > > > > > > > > think, in
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree
> > with
> > > > > >> Justine
> > > > > >> > > > that we
> > > > > >> > > > > > > > > should be
> > > > > >> > > > > > > > > > > >> > clearer
> > > > > >> > > > > > > > > > > >> > > > > about
> > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > > >> > > > IBP/metadata.version
> > > > > >> > > > > > from the
> > > > > >> > > > > > > > > > > >> feature
> > > > > >> > > > > > > > > > > >> > flag.
> > > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > > > > >> default for
> > > > > >> > > > the
> > > > > >> > > > > > feature
> > > > > >> > > > > > > > > flag
> > > > > >> > > > > > > > > > > >> to
> > > > > >> > > > > > > > > > > >> > true
> > > > > >> > > > > > > > > > > >> > > > > >> based
> > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > metadata.version
> > > > once
> > > > > >> we
> > > > > >> > > > want to
> > > > > >> > > > > > make
> > > > > >> > > > > > > > > it the
> > > > > >> > > > > > > > > > > >> > default.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to
> > use
> > > > that
> > > > > >> feature
> > > > > >> > > > > > flag
> > > > > >> > > > > > > > > mainly
> > > > > >> > > > > > > > > > > >> > during the
> > > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should
> > > not
> > > > have
> > > > > >> > > > mentioned
> > > > > >> > > > > > it, I
> > > > > >> > > > > > > > > think,
> > > > > >> > > > > > > > > > > >> > because
> > > > > >> > > > > > > > > > > >> > > > > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal
> > config
> > > > for
> > > > > >> it.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of
> > the
> > > > > >> protocol APIs
> > > > > >> > > > > > don't
> > > > > >> > > > > > > > > mention
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > required
> > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add
> > that
> > > > for
> > > > > >> > > > consistency.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a
> > bit
> > > > odd
> > > > > >> that
> > > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > >> > > > > > > > > > > >> > > > > requires
> > > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it
> > seems
> > > > to do
> > > > > >> more
> > > > > >> > > > than
> > > > > >> > > > > > > > > reading.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is
> > > > how the
> > > > > >> > > > current
> > > > > >> > > > > > > > > protocol works
> > > > > >> > > > > > > > > > > >> > though.
> > > > > >> > > > > > > > > > > >> > > > > We
> > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read
> > > > Group" to
> > > > > >> join a
> > > > > >> > > > > > group. We
> > > > > >> > > > > > > > > could
> > > > > >> > > > > > > > > > > >> > consider
> > > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not
> > sure
> > > > that
> > > > > >> it is
> > > > > >> > > > worth
> > > > > >> > > > > > it.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is
> > topic
> > > > > >> recreation
> > > > > >> > > > > > handled by
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > consumer
> > > > > >> > > > > > > > > > > >> > > > > with
> > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It
> > > > would be
> > > > > >> good to
> > > > > >> > > > > > have a
> > > > > >> > > > > > > > > section
> > > > > >> > > > > > > > > > > >> on
> > > > > >> > > > > > > > > > > >> > this.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> > > > protocol
> > > > > >> > > > > > perspective, the
> > > > > >> > > > > > > > > new
> > > > > >> > > > > > > > > > > >> topic
> > > > > >> > > > > > > > > > > >> > will
> > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it
> > > will
> > > > > >> treat it
> > > > > >> > > > like a
> > > > > >> > > > > > > > > topic with
> > > > > >> > > > > > > > > > > >> a
> > > > > >> > > > > > > > > > > >> > > > > different
> > > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is
> > that
> > > > the
> > > > > >> > > > fetch/commit
> > > > > >> > > > > > > > > offsets
> > > > > >> > > > > > > > > > > >> APIs
> > > > > >> > > > > > > > > > > >> > do not
> > > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > > consumer
> > > > > >> would
> > > > > >> > > > reuse the
> > > > > >> > > > > > > > > offsets
> > > > > >> > > > > > > > > > > >> based
> > > > > >> > > > > > > > > > > >> > on the
> > > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should
> > > > update
> > > > > >> those
> > > > > >> > > > APIs
> > > > > >> > > > > > as well
> > > > > >> > > > > > > > > in
> > > > > >> > > > > > > > > > > >> order
> > > > > >> > > > > > > > > > > >> > to be
> > > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That
> > would
> > > > > >> strengthen
> > > > > >> > > > the
> > > > > >> > > > > > > > > semantics of
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> > > > mentions
> > > > > >> we will
> > > > > >> > > > > > write the
> > > > > >> > > > > > > > > new
> > > > > >> > > > > > > > > > > >> > coordinator
> > > > > >> > > > > > > > > > > >> > > > > >> in
> > > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > > > >> implementation
> > > > > >> > > > > > detail, do
> > > > > >> > > > > > > > > we plan
> > > > > >> > > > > > > > > > > >> to
> > > > > >> > > > > > > > > > > >> > have a
> > > > > >> > > > > > > > > > > >> > > > > >> new
> > > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we
> > have a
> > > > > >> scalability
> > > > > >> > > > > > goal when
> > > > > >> > > > > > > > > it
> > > > > >> > > > > > > > > > > >> comes
> > > > > >> > > > > > > > > > > >> > to how
> > > > > >> > > > > > > > > > > >> > > > > >> many
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol
> > > can
> > > > > >> support?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have
> > > > numbers
> > > > > >> at the
> > > > > >> > > > > > moment. The
> > > > > >> > > > > > > > > > > >> protocol
> > > > > >> > > > > > > > > > > >> > should
> > > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per
> > > > group.
> > > > > >> We will
> > > > > >> > > > > > measure
> > > > > >> > > > > > > > > this when
> > > > > >> > > > > > > > > > > >> > we have
> > > > > >> > > > > > > > > > > >> > > > > a
> > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation.
> > > Note
> > > > > >> that we
> > > > > >> > > > might
> > > > > >> > > > > > have
> > > > > >> > > > > > > > > other
> > > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g.
> > offset
> > > > > >> commits).
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > > > consider
> > > > > >> having
> > > > > >> > > > > > > > > SubscribedTopidIds
> > > > > >> > > > > > > > > > > >> > instead
> > > > > >> > > > > > > > > > > >> > > > > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > SubscribedTopicNames
> > > > > >> in
> > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have
> > to
> > > > > >> resolve the
> > > > > >> > > > > > regex on
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > server, we
> > > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic
> > name?
> > > > The
> > > > > >> > > > difference
> > > > > >> > > > > > is that
> > > > > >> > > > > > > > > > > >> sending
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas
> > sending
> > > > the
> > > > > >> topic
> > > > > >> > > > names
> > > > > >> > > > > > is less
> > > > > >> > > > > > > > > > > >> > efficient.
> > > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> > > > easier
> > > > > >> to
> > > > > >> > > > handle
> > > > > >> > > > > > if we
> > > > > >> > > > > > > > > have
> > > > > >> > > > > > > > > > > >> topic
> > > > > >> > > > > > > > > > > >> > ids.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > > > > >> consolidate the
> > > > > >> > > > > > metadata
> > > > > >> > > > > > > > > lookup
> > > > > >> > > > > > > > > > > >> > on the
> > > > > >> > > > > > > > > > > >> > > > > >> server
> > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do
> > > agree
> > > > > >> with your
> > > > > >> > > > > > point. As
> > > > > >> > > > > > > > > a
> > > > > >> > > > > > > > > > > >> second
> > > > > >> > > > > > > > > > > >> > > > > though,
> > > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be
> > better
> > > > here
> > > > > >> for the
> > > > > >> > > > > > delete and
> > > > > >> > > > > > > > > > > >> > recreation
> > > > > >> > > > > > > > > > > >> > > > > case.
> > > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we
> > > may
> > > > > >> allow
> > > > > >> > > > users to
> > > > > >> > > > > > > > > subscribe
> > > > > >> > > > > > > > > > > >> with
> > > > > >> > > > > > > > > > > >> > topic
> > > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future because
> > > that
> > > > is
> > > > > >> the only
> > > > > >> > > > > > way to be
> > > > > >> > > > > > > > > > > >> really
> > > > > >> > > > > > > > > > > >> > robust
> > > > > >> > > > > > > > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12,
> > > > 2022 at
> > > > > >> 1:38 PM
> > > > > >> > > > > > David
> > > > > >> > > > > > > > > Jacot <
> > > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for
> > your
> > > > > >> comments.
> > > > > >> > > > Please
> > > > > >> > > > > > find
> > > > > >> > > > > > > > > my
> > > > > >> > > > > > > > > > > >> answers
> > > > > >> > > > > > > > > > > >> > > > > below.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > > > > >> protocol
> > > > > >> > > > relies on
> > > > > >> > > > > > > > > topic IDs
> > > > > >> > > > > > > > > > > >> > with the
> > > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names
> > > based
> > > > in
> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > >> > > > > > > > > > > >> > > > > >> I
> > > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> > > > names
> > > > > >> is the
> > > > > >> > > > > > right call
> > > > > >> > > > > > > > > here.
> > > > > >> > > > > > > > > > > >> I
> > > > > >> > > > > > > > > > > >> > need to
> > > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > > > > >> Obviously, the
> > > > > >> > > > > > KIP does
> > > > > >> > > > > > > > > not
> > > > > >> > > > > > > > > > > >> > change the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs
> > to
> > > > use
> > > > > >> topic
> > > > > >> > > > IDs.
> > > > > >> > > > > > This
> > > > > >> > > > > > > > > may be
> > > > > >> > > > > > > > > > > >> > something
> > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though
> > > as
> > > > it
> > > > > >> would
> > > > > >> > > > give
> > > > > >> > > > > > better
> > > > > >> > > > > > > > > > > >> overall
> > > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're
> > right.
> > > I
> > > > > >> think that
> > > > > >> > > > I
> > > > > >> > > > > > should
> > > > > >> > > > > > > > > not
> > > > > >> > > > > > > > > > > >> have
> > > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will
> > > remove
> > > > > >> it. We can
> > > > > >> > > > > > use an
> > > > > >> > > > > > > > > internal
> > > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > > > > >> feature.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> > > > types
> > > > > >> will be
> > > > > >> > > > > > > > > supported. The
> > > > > >> > > > > > > > > > > >> > change is
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only
> > requirement
> > > > is
> > > > > >> that the
> > > > > >> > > > > > cluster
> > > > > >> > > > > > > > > uses
> > > > > >> > > > > > > > > > > >> topic
> > > > > >> > > > > > > > > > > >> > IDs.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul
> > 11,
> > > > 2022
> > > > > >> at 9:53
> > > > > >> > > > PM
> > > > > >> > > > > > > > > Guozhang
> > > > > >> > > > > > > > > > > >> Wang <
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for
> > the
> > > > > >> feedback.
> > > > > >> > > > Here
> > > > > >> > > > > > are
> > > > > >> > > > > > > > > some
> > > > > >> > > > > > > > > > > >> replies
> > > > > >> > > > > > > > > > > >> > > > > inlined
> > > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul
> > 9,
> > > > 2022
> > > > > >> at
> > > > > >> > > > 2:53 AM
> > > > > >> > > > > > > > > Ismael Juma
> > > > > >> > > > > > > > > > > >> <
> > > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for
> > > the
> > > > > >> KIP. This
> > > > > >> > > > has
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > > >> potential
> > > > > >> > > > > > > > > > > >> > to be a
> > > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A
> > few
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > > > >> > > > questions/comments:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think
> > > > it's
> > > > > >> > > > premature to
> > > > > >> > > > > > talk
> > > > > >> > > > > > > > > about
> > > > > >> > > > > > > > > > > >> > target
> > > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of
> > > the
> > > > > >> existing
> > > > > >> > > > group
> > > > > >> > > > > > > > > protocol.
> > > > > >> > > > > > > > > > > >> > Unlike
> > > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > > > protocol
> > > > > >> and hence
> > > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how
> > quickly
> > > > > >> applications
> > > > > >> > > > > > migrate
> > > > > >> > > > > > > > > to the
> > > > > >> > > > > > > > > > > >> new
> > > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree
> > > > with
> > > > > >> you. I
> > > > > >> > > > think
> > > > > >> > > > > > we can
> > > > > >> > > > > > > > > > > >> remove
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > `Compatibility,
> > > > > >> > > > Deprecation,
> > > > > >> > > > > > and
> > > > > >> > > > > > > > > Migration
> > > > > >> > > > > > > > > > > >> > Plan` and
> > > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> > > > decide
> > > > > >> in the
> > > > > >> > > > > > future
> > > > > >> > > > > > > > > about
> > > > > >> > > > > > > > > > > >> when
> > > > > >> > > > > > > > > > > >> > we
> > > > > >> > > > > > > > > > > >> > > > > would
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > > > > >> behaviors.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP
> > > > says we
> > > > > >> > > > intend to
> > > > > >> > > > > > > > > release this
> > > > > >> > > > > > > > > > > >> > in 4.x,
> > > > > >> > > > > > > > > > > >> > > > > >> but
> > > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> > > > added
> > > > > >> that as
> > > > > >> > > > a
> > > > > >> > > > > > way to
> > > > > >> > > > > > > > > > > >> estimate
> > > > > >> > > > > > > > > > > >> > when
> > > > > >> > > > > > > > > > > >> > > > > we'd
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > > > > >> protocol, I
> > > > > >> > > > also
> > > > > >> > > > > > suggest
> > > > > >> > > > > > > > > > > >> removing
> > > > > >> > > > > > > > > > > >> > this
> > > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> > that's
> > > > not
> > > > > >> > > > specifically
> > > > > >> > > > > > > > > related to
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but
> > it's
> > > > more
> > > > > >> for
> > > > > >> > > > client
> > > > > >> > > > > > > > > upgrades.
> > > > > >> > > > > > > > > > > >> I.e.
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > implementation
> > > > may
> > > > > >> be done
> > > > > >> > > > > > first,
> > > > > >> > > > > > > > > and then
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > client
> > > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it
> > > as
> > > > > >> "released"
> > > > > >> > > > by
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > time
> > > > > >> > > > > > > > > > > >> clients
> > > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time,
> > to
> > > > > >> enable the
> > > > > >> > > > > > feature the
> > > > > >> > > > > > > > > > > >> clients
> > > > > >> > > > > > > > > > > >> > need to
> > > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode
> > with
> > > a
> > > > > >> rolling
> > > > > >> > > > bounce
> > > > > >> > > > > > and
> > > > > >> > > > > > > > > then set
> > > > > >> > > > > > > > > > > >> > the flag
> > > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and
> > > > hence
> > > > > >> we feel
> > > > > >> > > > it's
> > > > > >> > > > > > > > > better to be
> > > > > >> > > > > > > > > > > >> > released
> > > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need
> > > to
> > > > > >> flesh out
> > > > > >> > > > the
> > > > > >> > > > > > > > > details of
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > migration
> > > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're
> > saying
> > > > we
> > > > > >> will
> > > > > >> > > > support
> > > > > >> > > > > > online
> > > > > >> > > > > > > > > > > >> > migrations. Is
> > > > > >> > > > > > > > > > > >> > > > > >> that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain
> > this
> > > > in
> > > > > >> detail.
> > > > > >> > > > It
> > > > > >> > > > > > could
> > > > > >> > > > > > > > > also be
> > > > > >> > > > > > > > > > > >> > done as a
> > > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > > > that's
> > > > > >> the
> > > > > >> > > > part we
> > > > > >> > > > > > can
> > > > > >> > > > > > > > > be more
> > > > > >> > > > > > > > > > > >> > concrete
> > > > > >> > > > > > > > > > > >> > > > > >> about
> > > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is
> > > related
> > > > to
> > > > > >> your
> > > > > >> > > > > > question 2)
> > > > > >> > > > > > > > > above).
> > > > > >> > > > > > > > > > > >> > We will
> > > > > >> > > > > > > > > > > >> > > > > >> work
> > > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > > > > >> parallel as we
> > > > > >> > > > > > solicit
> > > > > >> > > > > > > > > more
> > > > > >> > > > > > > > > > > >> > feedback.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am
> > > happy
> > > > > >> that we are
> > > > > >> > > > > > pushing
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > pattern
> > > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it
> > seems
> > > > like
> > > > > >> there
> > > > > >> > > > > > could be
> > > > > >> > > > > > > > > some
> > > > > >> > > > > > > > > > > >> tricky
> > > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > > > mechanism
> > > > > >> for
> > > > > >> > > > users to
> > > > > >> > > > > > > > > detect that
> > > > > >> > > > > > > > > > > >> > they
> > > > > >> > > > > > > > > > > >> > > > > need
> > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > > > switching
> > > > > >> to the
> > > > > >> > > > new
> > > > > >> > > > > > > > > protocol?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > we
> > > > need
> > > > > >> some
> > > > > >> > > > > > tooling for
> > > > > >> > > > > > > > > > > >> non-java
> > > > > >> > > > > > > > > > > >> > client
> > > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run"
> > the
> > > > > >> client before
> > > > > >> > > > > > > > > switching to
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > new
> > > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific
> > idea
> > > on
> > > > > >> top of my
> > > > > >> > > > head
> > > > > >> > > > > > > > > though,
> > > > > >> > > > > > > > > > > >> maybe
> > > > > >> > > > > > > > > > > >> > others
> > > > > >> > > > > > > > > > > >> > > > > >> like
> > > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in
> > here?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related
> > > to
> > > > the
> > > > > >> last
> > > > > >> > > > > > question,
> > > > > >> > > > > > > > > will the
> > > > > >> > > > > > > > > > > >> > Java
> > > > > >> > > > > > > > > > > >> > > > > >> client
> > > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with
> > > the
> > > > > >> current
> > > > > >> > > > regex
> > > > > >> > > > > > > > > engine for
> > > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be
> > > > handy
> > > > > >> to keep
> > > > > >> > > > using
> > > > > >> > > > > > > > > client
> > > > > >> > > > > > > > > > > >> based
> > > > > >> > > > > > > > > > > >> > regex
> > > > > >> > > > > > > > > > > >> > > > > at
> > > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> > > > simple
> > > > > >> and
> > > > > >> > > > then
> > > > > >> > > > > > migrate
> > > > > >> > > > > > > > > to
> > > > > >> > > > > > > > > > > >> > server based
> > > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I
> > > have
> > > > not
> > > > > >> thought
> > > > > >> > > > > > about
> > > > > >> > > > > > > > > that for
> > > > > >> > > > > > > > > > > >> > java
> > > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What
> > > kind
> > > > of
> > > > > >> > > > > > compatibility
> > > > > >> > > > > > > > > issues do
> > > > > >> > > > > > > > > > > >> > you have
> > > > > >> > > > > > > > > > > >> > > > > in
> > > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we
> > > say
> > > > > >> that the
> > > > > >> > > > group
> > > > > >> > > > > > > > > coordinator
> > > > > >> > > > > > > > > > > >> > will be
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > > > >> configurations and
> > > > > >> > > > that
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > > >> > configurations
> > > > > >> > > > > > > > > > > >> > > > > will
> > > > > >> > > > > > > > > > > >> > > > > >> be
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> > > > deleted.
> > > > > >> Will a
> > > > > >> > > > > > > > > transition to
> > > > > >> > > > > > > > > > > >> DEAD
> > > > > >> > > > > > > > > > > >> > > > > trigger
> > > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > configurations?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since
> > the
> > > > DEAD
> > > > > >> state
> > > > > >> > > > is an
> > > > > >> > > > > > > > > ending
> > > > > >> > > > > > > > > > > >> state
> > > > > >> > > > > > > > > > > >> > (we
> > > > > >> > > > > > > > > > > >> > > > > would
> > > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when
> > the
> > > > > >> group is
> > > > > >> > > > EMPTY
> > > > > >> > > > > > and
> > > > > >> > > > > > > > > also all
> > > > > >> > > > > > > > > > > >> of
> > > > > >> > > > > > > > > > > >> > its
> > > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's
> > transited
> > > > to
> > > > > >> DEAD this
> > > > > >> > > > > > group
> > > > > >> > > > > > > > > would
> > > > > >> > > > > > > > > > > >> never
> > > > > >> > > > > > > > > > > >> > be
> > > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will
> > the
> > > > > >> choice to
> > > > > >> > > > store
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > configs
> > > > > >> > > > > > > > > > > >> in
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to
> > > > list all
> > > > > >> > > > cluster
> > > > > >> > > > > > configs
> > > > > >> > > > > > > > > and
> > > > > >> > > > > > > > > > > >> their
> > > > > >> > > > > > > > > > > >> > > > > values?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a
> > good
> > > > > >> question,
> > > > > >> > > > and our
> > > > > >> > > > > > > > > thoughts
> > > > > >> > > > > > > > > > > >> are
> > > > > >> > > > > > > > > > > >> > that
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > configurations"
> > > > are
> > > > > >> > > > overrides
> > > > > >> > > > > > of the
> > > > > >> > > > > > > > > > > >> > cluster-level
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized
> > per
> > > > > >> group so
> > > > > >> > > > when an
> > > > > >> > > > > > > > > admin list
> > > > > >> > > > > > > > > > > >> > cluster
> > > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just
> > the
> > > > > >> cluster-level
> > > > > >> > > > > > > > > "defaults", not
> > > > > >> > > > > > > > > > > >> > showing
> > > > > >> > > > > > > > > > > >> > > > > >> any
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > customizations.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How
> > would
> > > > > >> someone
> > > > > >> > > > > > configure a
> > > > > >> > > > > > > > > group
> > > > > >> > > > > > > > > > > >> > before
> > > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > > considered
> > > > > >> allowing
> > > > > >> > > > the
> > > > > >> > > > > > explicit
> > > > > >> > > > > > > > > > > >> > creation of
> > > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively,
> > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs
> > > could
> > > > be
> > > > > >> > > > decoupled
> > > > > >> > > > > > from
> > > > > >> > > > > > > > > the group
> > > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs
> > > can
> > > > be
> > > > > >> created
> > > > > >> > > > > > before
> > > > > >> > > > > > > > > the group
> > > > > >> > > > > > > > > > > >> > itself
> > > > > >> > > > > > > > > > > >> > > > > as
> > > > > >> > > > > > > > > > > >> > > > > >> an
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of
> > course,
> > > > this
> > > > > >> > > > requires
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > > >> corresponding
> > > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > > > coordinator
> > > > > >> based on
> > > > > >> > > > the
> > > > > >> > > > > > group
> > > > > >> > > > > > > > > id ---
> > > > > >> > > > > > > > > > > >> > the only
> > > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the
> > group
> > > > > >> itself is
> > > > > >> > > > gone
> > > > > >> > > > > > we
> > > > > >> > > > > > > > > also check
> > > > > >> > > > > > > > > > > >> > if there
> > > > > >> > > > > > > > > > > >> > > > > >> are
> > > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > > > related to
> > > > > >> that
> > > > > >> > > > group
> > > > > >> > > > > > and
> > > > > >> > > > > > > > > delete
> > > > > >> > > > > > > > > > > >> as
> > > > > >> > > > > > > > > > > >> > well.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly
> > > this
> > > > > >> indeed
> > > > > >> > > > > > introduces an
> > > > > >> > > > > > > > > > > >> > asymmetry on
> > > > > >> > > > > > > > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles
> > of
> > > > the
> > > > > >> config
> > > > > >> > > > > > entities,
> > > > > >> > > > > > > > > and we
> > > > > >> > > > > > > > > > > >> > would like
> > > > > >> > > > > > > > > > > >> > > > > >> to
> > > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > > > whether we
> > > > > >> should
> > > > > >> > > > aim
> > > > > >> > > > > > for
> > > > > >> > > > > > > > > symmetry
> > > > > >> > > > > > > > > > > >> > i.e.
> > > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and
> > > > hence
> > > > > >> not
> > > > > >> > > > delete
> > > > > >> > > > > > them at
> > > > > >> > > > > > > > > all
> > > > > >> > > > > > > > > > > >> when
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> group
> > > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> > > explicit
> > > > > >> deletion
> > > > > >> > > > > > operations
> > > > > >> > > > > > > > > by
> > > > > >> > > > > > > > > > > >> > themselves.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will
> > the
> > > > > >> > > > > > Consumer.subscribe
> > > > > >> > > > > > > > > method for
> > > > > >> > > > > > > > > > > >> > the Java
> > > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > `java.util.regex.Pattern` of
> > > > > >> > > > > > do we
> > > > > >> > > > > > > > > have
> > > > > >> > > > > > > > > > > >> to
> > > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we
> > do
> > > > not
> > > > > >> need to
> > > > > >> > > > > > introduce
> > > > > >> > > > > > > > > an
> > > > > >> > > > > > > > > > > >> > overload, but
> > > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > > > > >> compatibility
> > > > > >> > > > > > issues
> > > > > >> > > > > > > > > that we
> > > > > >> > > > > > > > > > > >> may
> > > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I
> > agree
> > > > with
> > > > > >> Justine
> > > > > >> > > > > > that we
> > > > > >> > > > > > > > > should
> > > > > >> > > > > > > > > > > >> be
> > > > > >> > > > > > > > > > > >> > clearer
> > > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > > >> > > > > > IBP/metadata.version
> > > > > >> > > > > > > > > from the
> > > > > >> > > > > > > > > > > >> > feature
> > > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch
> > > the
> > > > > >> default
> > > > > >> > > > for
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > feature
> > > > > >> > > > > > > > > > > >> flag
> > > > > >> > > > > > > > > > > >> > to
> > > > > >> > > > > > > > > > > >> > > > > true
> > > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > metadata.version
> > > > > >> once we
> > > > > >> > > > > > want to
> > > > > >> > > > > > > > > make it
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of
> > > the
> > > > > >> protocol
> > > > > >> > > > APIs
> > > > > >> > > > > > don't
> > > > > >> > > > > > > > > mention
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > >> required
> > > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to
> > add
> > > > that
> > > > > >> for
> > > > > >> > > > > > consistency.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > > > > >> certainly do
> > > > > >> > > > that.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is
> > a
> > > > bit
> > > > > >> odd that
> > > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it
> > > > seems
> > > > > >> to do
> > > > > >> > > > more
> > > > > >> > > > > > than
> > > > > >> > > > > > > > > reading.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> > > > thought
> > > > > >> myself
> > > > > >> > > > as
> > > > > >> > > > > > well,
> > > > > >> > > > > > > > > but in
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > end we
> > > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > > alternative:
> > > > > >> adding
> > > > > >> > > > > > Write
> > > > > >> > > > > > > > > Group
> > > > > >> > > > > > > > > > > >> seems
> > > > > >> > > > > > > > > > > >> > an
> > > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > > > > >> elsewhere (we
> > > > > >> > > > only
> > > > > >> > > > > > have
> > > > > >> > > > > > > > > Read /
> > > > > >> > > > > > > > > > > >> > Delete
> > > > > >> > > > > > > > > > > >> > > > > and
> > > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would
> > > > like to
> > > > > >> hear
> > > > > >> > > > others
> > > > > >> > > > > > > > > thoughts.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> > > > topic
> > > > > >> > > > recreation
> > > > > >> > > > > > > > > handled by
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > consumer
> > > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol?
> > It
> > > > > >> would be
> > > > > >> > > > good
> > > > > >> > > > > > to have
> > > > > >> > > > > > > > > a
> > > > > >> > > > > > > > > > > >> > section on
> > > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean
> > with
> > > > regex
> > > > > >> > > > > > subscription
> > > > > >> > > > > > > > > right? Yes
> > > > > >> > > > > > > > > > > >> > we can
> > > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> > > > basically
> > > > > >> the
> > > > > >> > > > idea
> > > > > >> > > > > > is that
> > > > > >> > > > > > > > > > > >> consumer
> > > > > >> > > > > > > > > > > >> > would
> > > > > >> > > > > > > > > > > >> > > > > be
> > > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > > > protocol as
> > > > > >> it's
> > > > > >> > > > > > handled all
> > > > > >> > > > > > > > > by the
> > > > > >> > > > > > > > > > > >> > brokers.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The
> > KIP
> > > > > >> mentions we
> > > > > >> > > > will
> > > > > >> > > > > > write
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> new
> > > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > > > > >> implementation
> > > > > >> > > > > > detail,
> > > > > >> > > > > > > > > do we
> > > > > >> > > > > > > > > > > >> > plan to
> > > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> > > > thought
> > > > > >> about
> > > > > >> > > > > > that. But
> > > > > >> > > > > > > > > I think
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > answer
> > > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we
> > > > have a
> > > > > >> > > > scalability
> > > > > >> > > > > > goal
> > > > > >> > > > > > > > > when it
> > > > > >> > > > > > > > > > > >> > comes to
> > > > > >> > > > > > > > > > > >> > > > > >> how
> > > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > > > protocol can
> > > > > >> > > > support?
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> > > group,
> > > > I
> > > > > >> think we
> > > > > >> > > > > > should
> > > > > >> > > > > > > > > shoot for
> > > > > >> > > > > > > > > > > >> > 1000s of
> > > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability
> > > > goals
> > > > > >> also
> > > > > >> > > > depend
> > > > > >> > > > > > on the
> > > > > >> > > > > > > > > offset
> > > > > >> > > > > > > > > > > >> > > > > management
> > > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities
> > > of
> > > > the
> > > > > >> > > > coordinator
> > > > > >> > > > > > > > > which we
> > > > > >> > > > > > > > > > > >> did
> > > > > >> > > > > > > > > > > >> > not
> > > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to
> > give a
> > > > > >> number that
> > > > > >> > > > > > applies
> > > > > >> > > > > > > > > > > >> > universally.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > > > > >> consider
> > > > > >> > > > having
> > > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> SubscribedTopicNames in
> > > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we
> > > have
> > > > to
> > > > > >> resolve
> > > > > >> > > > the
> > > > > >> > > > > > regex
> > > > > >> > > > > > > > > on the
> > > > > >> > > > > > > > > > > >> > server,
> > > > > >> > > > > > > > > > > >> > > > > we
> > > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic
> > > > name?
> > > > > >> The
> > > > > >> > > > > > difference is
> > > > > >> > > > > > > > > that
> > > > > >> > > > > > > > > > > >> > sending the
> > > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas
> > > > sending
> > > > > >> the topic
> > > > > >> > > > > > names is
> > > > > >> > > > > > > > > less
> > > > > >> > > > > > > > > > > >> > efficient.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation
> > > is
> > > > > >> easier to
> > > > > >> > > > > > handle if
> > > > > >> > > > > > > > > we have
> > > > > >> > > > > > > > > > > >> > topic
> > > > > >> > > > > > > > > > > >> > > > > ids.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main
> > > reason
> > > > to
> > > > > >> still
> > > > > >> > > > let
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > clients
> > > > > >> > > > > > > > > > > >> send
> > > > > >> > > > > > > > > > > >> > names
> > > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> > > > names
> > > > > >> -> ids
> > > > > >> > > > on
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > broker /
> > > > > >> > > > > > > > > > > >> > admin
> > > > > >> > > > > > > > > > > >> > > > > client
> > > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we
> > > > added
> > > > > >> topic id
> > > > > >> > > > in
> > > > > >> > > > > > > > > KIP-516, we
> > > > > >> > > > > > > > > > > >> > never
> > > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > consumer/producers
> > > > > >> > > > leveraging
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > related
> > > > > >> > > > > > > > > > > >> > newer
> > > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> > > > topic
> > > > > >> id as
> > > > > >> > > > empty
> > > > > >> > > > > > UUID.
> > > > > >> > > > > > > > > We
> > > > > >> > > > > > > > > > > >> want
> > > > > >> > > > > > > > > > > >> > to keep
> > > > > >> > > > > > > > > > > >> > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > consumer/producer
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin
> > and
> > > > only
> > > > > >> > > > delegate
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > reasoning
> > > > > >> > > > > > > > > > > >> on
> > > > > >> > > > > > > > > > > >> > broker
> > > > > >> > > > > > > > > > > >> > > > > >> and
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed,
> > Jul
> > > 6,
> > > > > >> 2022 at
> > > > > >> > > > 10:45
> > > > > >> > > > > > AM
> > > > > >> > > > > > > > > David
> > > > > >> > > > > > > > > > > >> Jacot
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> <dj...@confluent.io.invalid>
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would
> > > > like to
> > > > > >> start a
> > > > > >> > > > > > > > > discussion
> > > > > >> > > > > > > > > > > >> thread
> > > > > >> > > > > > > > > > > >> > on
> > > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > Generation
> > > > of
> > > > > >> the
> > > > > >> > > > Consumer
> > > > > >> > > > > > > > > Rebalance
> > > > > >> > > > > > > > > > > >> > Protocol.
> > > > > >> > > > > > > > > > > >> > > > > >> With
> > > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make
> > > the
> > > > > >> rebalance
> > > > > >> > > > > > protocol
> > > > > >> > > > > > > > > (for
> > > > > >> > > > > > > > > > > >> > consumers)
> > > > > >> > > > > > > > > > > >> > > > > >> more
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > scalable,
> > > > > >> easier to
> > > > > >> > > > > > implement for
> > > > > >> > > > > > > > > > > >> > clients, and
> > > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > operators.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP
> > is
> > > > here:
> > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please
> > > take
> > > > a
> > > > > >> look and
> > > > > >> > > > let
> > > > > >> > > > > > me
> > > > > >> > > > > > > > > know what
> > > > > >> > > > > > > > > > > >> > you
> > > > > >> > > > > > > > > > > >> > > > > think.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I
> > will
> > > > be
> > > > > >> away from
> > > > > >> > > > > > July
> > > > > >> > > > > > > > > 18th to
> > > > > >> > > > > > > > > > > >> > August 8th.
> > > > > >> > > > > > > > > > > >> > > > > >> That
> > > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time
> > to
> > > > read
> > > > > >> and
> > > > > >> > > > digest
> > > > > >> > > > > > this
> > > > > >> > > > > > > > > long
> > > > > >> > > > > > > > > > > >> KIP.
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > > > >> >
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >> --
> > > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > >
> > > > > >> > > > > >
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> > > > > --
> > > > > >> > > > > -- Guozhang
> > > > > >> > > >
> > > > > >> > >
> > > > > >> > >
> > > > > >> > > --
> > > > > >> > > -- Guozhang
> > > > > >>
> > > > > >
> > > > > >
> > > > > > --
> > > > > > -- Guozhang
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > > -- Guozhang
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
> > >
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Luke Chen <sh...@gmail.com>]

Hi David,

Thanks for your patience for answering all these questions!

One more question from me:
1. When the selected consumer is going to be in client-side assigning, we
will return COMPUTE_ASSIGNMENT in heartbeat response. I'm wondering if we
could use a flag to indicate the assignment is required. The rationale is
when returning error, we lose a chance to catch up with the latest target
assignment or revoke partitions. Since we expected the record fetching is
not blocked by the rebalance/assignment computing, we might want to allow
the members to update the member assignment in each heartbeat.

Luke

On Tue, Sep 27, 2022 at 9:43 AM Jason Gustafson <ja...@confluent.io.invalid>
wrote:

> Hey David,
>
> Thanks for the response.
>
> > Do you think that we could have a case where two members with the same
> static id are active?
>
> I was just thinking about the handoff when we spin up a new node before the
> old one completes shutdown. It's probably ok to fence the old member
> abruptly. I do recall we had a few issues with this handoff in the past. It
> was due to the fact that the replaced node might have been the leader.
> Since we've removed the leader notion, perhaps this kind of issue won't
> resurface.
>
> In regard to the RE2 library. I understand the intent to pass this through
> as a string without any validation so that we don't need the dependence on
> the client. I do wonder if that will be restrictive in practice. For
> example, if a regex changes on the client side, would we need to validate
> against the current assignment to see which partitions to continue
> fetching? Or would we instead invalidate the complete current assignment?
> Perhaps this is a rare case and we can do the latter. There could be more
> situations though.
>
> As an alternative, I wonder if we could define a minimalistic regex syntax
> (a subset of RE2 say) that could map more easily to regex support that
> already exists in clients? For example, in this scenario, we are matching
> topics, so we only need to be able to represent topic characters and not
> the whole smorgasbord of whitespace and unicode options. Would it make
> sense to have a reduced syntax which includes only standard topic
> characters, wildcards, character classes, and repetitions?
>
> Best,
> Jason
>
>
>
>
> On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com> wrote:
>
> > Hi David, thanks for all the detailed explanations. I think they all make
> > sense. Just want to have a couple follow-ups here:
> >
> > > I don't really see the benefits here because server side assignors
> > don't have metadata at all. They only assign topic-partitions. They
> > are not supposed to generate metadata nor to receive metadata from the
> > members.
> >
> > Just to clarify I was asking about the `version` of the assignor (i.e. up
> > to what version that the client would support), and I do agree we would
> not
> > need metadata. What I have in mind is that, for some specific built-in
> > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > newer version we would have a hierarchical rack ID string format, like
> > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > would still be in old format. In this case, the broker then needs to
> choose
> > the old versioned assignor. I'm probably making something up here for
> rack
> > aware assignors, but I'm wondering if in general such an "auto-downgrade"
> > behavior would be needed still for broker-side assignor, and if yes would
> > "version" still be useful.
> >
> > > Yeah, that's right. Within a rebalance, `onAssignment` is called once
> > when the member transitions to a new epoch. This one contains the full
> > metadata provided by the client side assignor. Then, `onAssignment`
> > can be called max N times where N is the number of partitions pending
> > revocation by other members. Let me try to clarify this in the KIP.
> >
> > Okay, my understanding is that the calling ordering of these callbacks
> > would be like the following:
> >
> > ----------------------------------------
> > onPartitionsRevoked();   // just once, since we do not really need
> > to revoke incrementally.
> >
> > onAssignment();    // the first call, with epoch incremented
> > onPartitionsAssigned();   // paired with the onAssignment
> >
> > onAssignment();              // the first onAssignment would bump up the
> > epoch, and the metadata reflected.
> > onPartitionsAssigned();   // each time we get an additional assignment,
> we
> > call onAssignment and then paired with an onPartitionsAssigned
> > ...
> > onAssignment();
> > onPartitionsAssigned();   // on each of the onAssignment calls, the
> encoded
> > metadata would not change, only the incrementally added partitions be
> > reflected
> >
> > Is that the case?
> >
> > I'm wondering if we would still call onAssignment just once, that encodes
> > all the assignment for this rebalance, including all the partitions that
> > should be assigned to the member but not yet assigned since they have not
> > been revoked by others. In that case the call ordering would be:
> >
> > ----------------------------------------
> > onPartitionsRevoked();   // just once
> > onAssignment();    // just once, with epoch incremented, and metadata
> > encoded changed, the "assignment" field also reflect the final target
> > assignment
> > onPartitionsAssigned();   // multiple times, which represent
> incrementally
> > added partitions
> > ...
> > onPartitionsAssigned();
> >
> > The motivation from this is that, most users would only implement the
> > rebalance callback listeners and hence we'd definitely need to make sure
> > the semantics of that does not change much, and the time
> > `onPartitionsAssigned` indicate the time when the partitions are actually
> > assigned to it; while for assignors, the `onAssignment` is used to
> indicate
> > what decision is made regarding for this member, i.e. when the partitions
> > are decided to be given to it, but not necessarily meant that it has been
> > given, since that time should be determined by the time of
> > `onPartitionsAssigned`. The benefits though, would be that assignor
> > implementers would not need to reason which `onAssignment` would be the
> > last one for this epoch.
> >
> >
> > Guozhang
> >
> >
> >
> >
> > Guozhang
> >
> > On Thu, Sep 22, 2022 at 2:20 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > field,
> > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> the
> > > > rationale seems to be assuming that we should always be able to do
> the
> > > > assignment at the broker-side assignor without errors. Personally I
> > think
> > > > it's still potentially beneficial to add the Error field even for
> > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > topics are not recognized with the current broker's metadata. What do
> > you
> > > > think?
> > >
> > > Yeah, that seems reasonable. However, I wonder if it would be better
> > > to use an exception on the server side. This is what we usually do for
> > > server side plugins. On the client side, we use a field because the
> > > errors are not defined in advance.
> > >
> > > Your comment also makes me think about what we should do when the
> > > server side assignor fails. I suppose that we have to keep the current
> > > assignment until a new event occurs. For instance, in your example,
> > > the coordinator would have to trigger a rebalance when unrecognized
> > > topics are available. This would be part of the metadata monitoring.
> > >
> > > > 2) The client-side "GroupMember" has three additional fields
> > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > agree
> > > > that broker-side assignor would not need reason/metadata since they
> are
> > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> is
> > > it
> > > > possible that we evolve our broker-side built-in assignor but the old
> > > > versioned clients would not be able to work with the new version, in
> > > which
> > > > case we need to let the broker being aware of this and upgrade its
> > > behavior
> > > > to cooperate with the clients?
> > >
> > > I don't really see the benefits here because server side assignors
> > > don't have metadata at all. They only assign topic-partitions. They
> > > are not supposed to generate metadata nor to receive metadata from the
> > > members.
> > >
> > > > 3) Also related to 2) above, for the client-side "GroupMember",
> instead
> > > of
> > > > including these three fields, what about just adding the "Metadata"
> > field
> > > > class which has these three fields? Also, there are two "Metadata"
> > > > currently in the APIs, the first is a class that encodes
> > > > reason/version/metadata, and the second is just the encoded metadata
> > > bytes.
> > > > I'm wondering what about just naming the first as memberMetadata,
> which
> > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > bytebuffer field as metadataBytes?
> > >
> > > That's a good point. Let me try to rationalize this interface based on
> > > your suggestions.
> > >
> > > Best,
> > > David
> > >
> > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > >
> > > > Hello David,
> > > >
> > > > Just had a few more nit questions about the public APIs:
> > > >
> > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > field,
> > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> the
> > > > rationale seems to be assuming that we should always be able to do
> the
> > > > assignment at the broker-side assignor without errors. Personally I
> > think
> > > > it's still potentially beneficial to add the Error field even for
> > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > topics are not recognized with the current broker's metadata. What do
> > you
> > > > think?
> > > >
> > > > 2) The client-side "GroupMember" has three additional fields
> > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > agree
> > > > that broker-side assignor would not need reason/metadata since they
> are
> > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> is
> > > it
> > > > possible that we evolve our broker-side built-in assignor but the old
> > > > versioned clients would not be able to work with the new version, in
> > > which
> > > > case we need to let the broker being aware of this and upgrade its
> > > behavior
> > > > to cooperate with the clients?
> > > >
> > > > 3) Also related to 2) above, for the client-side "GroupMember",
> instead
> > > of
> > > > including these three fields, what about just adding the "Metadata"
> > field
> > > > class which has these three fields? Also, there are two "Metadata"
> > > > currently in the APIs, the first is a class that encodes
> > > > reason/version/metadata, and the second is just the encoded metadata
> > > bytes.
> > > > I'm wondering what about just naming the first as memberMetadata,
> which
> > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > bytebuffer field as metadataBytes?
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > >
> > > > > Hello David,
> > > > >
> > > > > Thanks for bringing this question up. I think the main benefits as
> > you
> > > > > listed is 2) above if it stays; just to clarify, we would only be
> > able
> > > to
> > > > > save one round trip if the rebalance is still triggered by the
> > broker;
> > > if
> > > > > the rebalance is triggered by the client then the num.round.trips
> are
> > > the
> > > > > same:
> > > > >
> > > > > 1) With GroupPrepareAssignment:
> > > > >
> > > > > T0: client decides to do a new assignment, suppose it has already
> > sent
> > > a
> > > > > HB and hence has to wait for it to return first since only one
> > request
> > > /
> > > > > response can be inflight with the coordinator's socket.
> > > > > T1: client receives the HB response, and then sends the
> > > > > GroupPrepareAssignment request.
> > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > T3: it calculates the new assignment, and sends a
> > > GroupInstallAssignment
> > > > > request.
> > > > >
> > > > > In total, two round trips.
> > > > >
> > > > > 2) Without GroupPrepareAssignment:
> > > > >
> > > > > T0: client decides to do a new assignment, suppose it has already
> > sent
> > > a
> > > > > HB and hence has to wait for it to return first since only one
> > request
> > > /
> > > > > response can be inflight with the coordinator's socket.
> > > > > T1: client receives the HB response, and then sends the new HB
> > request
> > > > > with the flag indicating a new rebalance needed..
> > > > > T2: the HB response with the optional member metadata map is
> > returned.
> > > > > T3: it calculates the new assignment, and sends a
> > > GroupInstallAssignment
> > > > > request.
> > > > >
> > > > > In total, two round trips as well.
> > > > >
> > > > > -----------------------------
> > > > >
> > > > > So to complete the full picture here, we'd need to modify both HB
> > > request
> > > > > and response so that the client can also indicate a new rebalance
> via
> > > the
> > > > > HB request as well, right?
> > > > >
> > > > > Assuming all above is true, I think it's okay to merge the
> > > > > GroupPrepareAssignment into HB given that we can make the
> additional
> > > fields
> > > > > encoding the full member (subscription) metadata and topic metadata
> > as
> > > > > optional fields.
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > <dj...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > >> Hi all,
> > > > >>
> > > > >> During an offline conversation, someone asked why we need the
> > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > instead
> > > > >> provide the group state in the heartbeat response. This has a few
> > > > >> advantages: 1) it does not require using a special error code to
> > > > >> signal that a new assignment is required as the signal would be
> the
> > > > >> provided group state; 2) it removes one round trip when a client
> > side
> > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > >> response's definition quite large. I recall that I went with the
> > > > >> current approach due to this.
> > > > >>
> > > > >> Providing the group state in the heartbeat response is appealing.
> > What
> > > > >> do you guys think?
> > > > >>
> > > > >> Best,
> > > > >> David
> > > > >>
> > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > > >> >
> > > > >> > Hi Guozhang,
> > > > >> >
> > > > >> > 1. I have added a reference to the relevant chapter instead of
> > > > >> > repeating the whole thing. Does that work for you?
> > > > >> >
> > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> about
> > > when
> > > > >> > a rebalance should be triggered for the non-upgraded members
> using
> > > the
> > > > >> > old protocol. The section mentions that a rebalance must be
> > > triggered
> > > > >> > when a new assignment is installed. This implies that the group
> > > epoch
> > > > >> > was updated either by a native member or a non-upgraded member.
> > For
> > > > >> > the latter, the JoinGroup request would be the trigger. I have
> > > added a
> > > > >> > reference to the relevant chapter in the "JoinGroup Handling"
> > > section
> > > > >> > as well. Does that make sense?
> > > > >> >
> > > > >> > Thanks,
> > > > >> > David
> > > > >> >
> > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> wangguoz@gmail.com
> > >
> > > > >> wrote:
> > > > >> > >
> > > > >> > > Hello David,
> > > > >> > >
> > > > >> > > Alright I think that's sufficient. Just to make that clear in
> > the
> > > doc,
> > > > >> > > could we update:
> > > > >> > >
> > > > >> > > 1) the heartbeat request handling section, stating when
> > > coordinator
> > > > >> will
> > > > >> > > trigger rebalance based on the HB's member metadata / reason?
> > > > >> > > 2) the "Rebalance Triggers" section to include what we
> described
> > > in
> > > > >> "Group
> > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > >> > >
> > > > >> > >
> > > > >> > > Guozhang
> > > > >> > >
> > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > wrote:
> > > > >> > >
> > > > >> > > > Hi Guozhang,
> > > > >> > > >
> > > > >> > > > I thought that the assignor will always be consulted when
> the
> > > next
> > > > >> > > > heartbeat request is constructed. In other words,
> > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > heartbeat.
> > > > >> This
> > > > >> > > > gives the opportunity for the assignor to enforce a
> rebalance
> > by
> > > > >> > > > setting the reason to a non-zero value or by changing the
> > > bytes. Do
> > > > >> > > > you think that this is not sufficient? Are you concerned by
> > the
> > > > >> delay?
> > > > >> > > >
> > > > >> > > > Best,
> > > > >> > > > David
> > > > >> > > >
> > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > >> wrote:
> > > > >> > > > >
> > > > >> > > > > Hello David,
> > > > >> > > > >
> > > > >> > > > > One of Jun's comments make me thinking:
> > > > >> > > > >
> > > > >> > > > > ```
> > > > >> > > > > In this case, a new assignment is triggered by the client
> > side
> > > > >> > > > > assignor. When constructing the HB, the consumer will
> always
> > > > >> consult
> > > > >> > > > > the client side assignor and propagate the information to
> > the
> > > > >> group
> > > > >> > > > > coordinator. In other words, we don't expect users to call
> > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > >> > > > > ```
> > > > >> > > > >
> > > > >> > > > > As I looked at the current PartitionAssignor's interface,
> we
> > > > >> actually do
> > > > >> > > > > not have a way yet to instruct how to construct the next
> HB
> > > > >> request, e.g.
> > > > >> > > > > when the assignor wants to enforce a new rebalance with a
> > new
> > > > >> assignment,
> > > > >> > > > > we'd need some customizable APIs inside the
> > PartitionAssignor
> > > to
> > > > >> indicate
> > > > >> > > > > the next HB telling broker about so. WDYT about adding
> such
> > an
> > > > >> API on the
> > > > >> > > > > PartitionAssignor?
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > Guozhang
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > > > wrote:
> > > > >> > > > >
> > > > >> > > > > > Hi Jun,
> > > > >> > > > > >
> > > > >> > > > > > I have updated the KIP to include your feedback. I have
> > also
> > > > >> tried to
> > > > >> > > > > > clarify the parts which were not cleared.
> > > > >> > > > > >
> > > > >> > > > > > Best,
> > > > >> > > > > > David
> > > > >> > > > > >
> > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > djacot@confluent.io
> > > > >> >
> > > > >> > > > wrote:
> > > > >> > > > > > >
> > > > >> > > > > > > Hi Jun,
> > > > >> > > > > > >
> > > > >> > > > > > > Thanks for your feedback. Let me start by answering
> your
> > > > >> questions
> > > > >> > > > > > > inline and I will update the KIP next week.
> > > > >> > > > > > >
> > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> the
> > > KIP
> > > > >> seem to
> > > > >> > > > be
> > > > >> > > > > > fewer
> > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > >> wildcard. A few
> > > > >> > > > > > > > comments below.
> > > > >> > > > > > >
> > > > >> > > > > > > I would also add that the KIP removes the global sync
> > > barrier
> > > > >> in the
> > > > >> > > > > > > protocol which is essential to improve group stability
> > and
> > > > >> > > > > > > scalability, and the KIP also simplifies the client by
> > > moving
> > > > >> most of
> > > > >> > > > > > > the logic to the server side.
> > > > >> > > > > > >
> > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > support
> > > > >> rolling
> > > > >> > > > > > changing
> > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > >> > > > > > >
> > > > >> > > > > > > Definitely. The group coordinator will use the
> assignor
> > > used
> > > > >> by a
> > > > >> > > > > > > majority of the members. This allows the group to move
> > > from
> > > > >> one
> > > > >> > > > > > > assignor to another by a roll. This is explained in
> the
> > > > >> Assignor
> > > > >> > > > > > > Selection chapter.
> > > > >> > > > > > >
> > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > >> required in
> > > > >> > > > every
> > > > >> > > > > > > > request or the scenarios when it needs to be filled?
> > For
> > > > >> example,
> > > > >> > > > it's
> > > > >> > > > > > not
> > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > >> > > > > > >
> > > > >> > > > > > > The client is expected to set those fields in case of
> a
> > > > >> connection
> > > > >> > > > > > > issue (e.g. timeout) or when the fields have changed
> > since
> > > > >> the last
> > > > >> > > > > > > HB. The server populates those fields as long as the
> > > member
> > > > >> is not
> > > > >> > > > > > > fully reconciled - the member should acknowledge that
> it
> > > has
> > > > >> the
> > > > >> > > > > > > expected epoch and assignment. I will clarify this in
> > the
> > > KIP.
> > > > >> > > > > > >
> > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > affinity
> > > > >> between the
> > > > >> > > > > > client
> > > > >> > > > > > > > and the broker is only considered during fetching,
> but
> > > not
> > > > >> during
> > > > >> > > > > > assigning
> > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > assignment
> > > is
> > > > >> made,
> > > > >> > > > there
> > > > >> > > > > > is
> > > > >> > > > > > > > no opportunity for read affinity because no replicas
> > of
> > > > >> assigned
> > > > >> > > > > > partitions
> > > > >> > > > > > > > are close to the member. I am wondering if we should
> > use
> > > > >> this
> > > > >> > > > > > opportunity
> > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > >> > > > > > >
> > > > >> > > > > > > That's an interesting idea. I don't see any issue with
> > > adding
> > > > >> the
> > > > >> > > > rack
> > > > >> > > > > > > to the members. I will do so.
> > > > >> > > > > > >
> > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> how
> > > busy
> > > > >> a group
> > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > > >> that we
> > > > >> > > > could
> > > > >> > > > > > add
> > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > event
> > > > >> loop is
> > > > >> > > > doing
> > > > >> > > > > > the
> > > > >> > > > > > > > actual work.
> > > > >> > > > > > >
> > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > >> > > > > > >
> > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > >> handling? For
> > > > >> > > > > > example,
> > > > >> > > > > > > > does it need to trigger the check if any group with
> > the
> > > > >> wildcard
> > > > >> > > > > > > > subscription now has a new matching topic?
> > > > >> > > > > > >
> > > > >> > > > > > > Sure. When the new group coordinator takes over, it
> has
> > > to:
> > > > >> > > > > > > * Setup the session timeouts.
> > > > >> > > > > > > * Trigger a new assignment if a client side assignor
> is
> > > used.
> > > > >> We
> > > > >> > > > don't
> > > > >> > > > > > > store the information about the member selected to run
> > the
> > > > >> assignment
> > > > >> > > > > > > so we have to start a new one.
> > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > >> subscriptions, and
> > > > >> > > > > > > trigger a rebalance if needed.
> > > > >> > > > > > >
> > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > > >> the epoch
> > > > >> > > > > > field
> > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > >> ConsumerGroupMetadataValue
> > > > >> > > > > > reflect
> > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > >> > > > > > >
> > > > >> > > > > > > Sure. I will clarify that but it is always the latest
> > > group
> > > > >> epoch.
> > > > >> > > > > > > When the group state is updated, the group epoch is
> > > bumped so
> > > > >> we use
> > > > >> > > > > > > that one for all the change records related to the
> > update.
> > > > >> > > > > > >
> > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > following
> > > > >> > > > invariants
> > > > >> > > > > > are
> > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > > >> not to
> > > > >> > > > get any
> > > > >> > > > > > > > assigned partitions, right?
> > > > >> > > > > > >
> > > > >> > > > > > > That's right. Here I meant that the members provided
> by
> > > the
> > > > >> assignor
> > > > >> > > > > > > in the assignment must exist in the group. The
> assignor
> > > can
> > > > >> not make
> > > > >> > > > > > > up new member ids.
> > > > >> > > > > > >
> > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > equals to
> > > > >> 0":
> > > > >> > > > When
> > > > >> > > > > > would
> > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > >> > > > > > >
> > > > >> > > > > > > A member is expected to abandon all its partitions and
> > > > >> rejoins when
> > > > >> > > > it
> > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case,
> > the
> > > > >> group
> > > > >> > > > > > > coordinator will have removed the member from the
> group.
> > > The
> > > > >> member
> > > > >> > > > > > > can rejoin the group with the same member id but with
> 0
> > as
> > > > >> epoch. Let
> > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > >> > > > > > >
> > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger a
> > > > >> > > > > > reassignment
> > > > >> > > > > > > > by either providing a non-zero reason or by updating
> > the
> > > > >> assignor
> > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> the
> > > > >> deprecation
> > > > >> > > > of
> > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > >> > > > > > >
> > > > >> > > > > > > In this case, a new assignment is triggered by the
> > client
> > > side
> > > > >> > > > > > > assignor. When constructing the HB, the consumer will
> > > always
> > > > >> consult
> > > > >> > > > > > > the client side assignor and propagate the information
> > to
> > > the
> > > > >> group
> > > > >> > > > > > > coordinator. In other words, we don't expect users to
> > call
> > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > >> > > > > > >
> > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > section
> > > > >> seems to
> > > > >> > > > have
> > > > >> > > > > > > > multiple typos.
> > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > immediately
> > > > >> gets into
> > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> incorrect.
> > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> to
> > > get
> > > > >> into
> > > > >> > > > > > epoch=3,
> > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > >> epoch=2,
> > > > >> > > > > > > > partitions=[foo-0].
> > > > >> > > > > > >
> > > > >> > > > > > > Sorry for that! I will revise them.
> > > > >> > > > > > >
> > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > upgrade
> > > > >> from
> > > > >> > > > any
> > > > >> > > > > > old
> > > > >> > > > > > > > version to new one?
> > > > >> > > > > > >
> > > > >> > > > > > > We will support upgrading from the consumer protocol
> > > version
> > > > >> 3,
> > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet
> so
> > > the
> > > > >> earliest
> > > > >> > > > > > > version is unknown at the moment. This is explained in
> > the
> > > > >> migration
> > > > >> > > > > > > plan chapter.
> > > > >> > > > > > >
> > > > >> > > > > > > Thanks again for your feedback, Jun. I will update the
> > KIP
> > > > >> based on
> > > > >> > > > it
> > > > >> > > > > > > next week.
> > > > >> > > > > > >
> > > > >> > > > > > > Best,
> > > > >> > > > > > > David
> > > > >> > > > > > >
> > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > >> <ju...@confluent.io.invalid>
> > > > >> > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > Hi, David,
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> the
> > > KIP
> > > > >> seem to
> > > > >> > > > be
> > > > >> > > > > > fewer
> > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > >> wildcard. A few
> > > > >> > > > > > > > comments below.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > support
> > > > >> rolling
> > > > >> > > > > > changing
> > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > >> required in
> > > > >> > > > every
> > > > >> > > > > > > > request or the scenarios when it needs to be filled?
> > For
> > > > >> example,
> > > > >> > > > it's
> > > > >> > > > > > not
> > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > affinity
> > > > >> between the
> > > > >> > > > > > client
> > > > >> > > > > > > > and the broker is only considered during fetching,
> but
> > > not
> > > > >> during
> > > > >> > > > > > assigning
> > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > assignment
> > > is
> > > > >> made,
> > > > >> > > > there
> > > > >> > > > > > is
> > > > >> > > > > > > > no opportunity for read affinity because no replicas
> > of
> > > > >> assigned
> > > > >> > > > > > partitions
> > > > >> > > > > > > > are close to the member. I am wondering if we should
> > use
> > > > >> this
> > > > >> > > > > > opportunity
> > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> how
> > > busy
> > > > >> a group
> > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > > >> that we
> > > > >> > > > could
> > > > >> > > > > > add
> > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > event
> > > > >> loop is
> > > > >> > > > doing
> > > > >> > > > > > the
> > > > >> > > > > > > > actual work.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > >> handling? For
> > > > >> > > > > > example,
> > > > >> > > > > > > > does it need to trigger the check if any group with
> > the
> > > > >> wildcard
> > > > >> > > > > > > > subscription now has a new matching topic?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > > >> the epoch
> > > > >> > > > > > field
> > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > >> ConsumerGroupMetadataValue
> > > > >> > > > > > reflect
> > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > following
> > > > >> > > > invariants
> > > > >> > > > > > are
> > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > > >> not to
> > > > >> > > > get any
> > > > >> > > > > > > > assigned partitions, right?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > equals to
> > > > >> 0":
> > > > >> > > > When
> > > > >> > > > > > would
> > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger a
> > > > >> > > > > > reassignment
> > > > >> > > > > > > > by either providing a non-zero reason or by updating
> > the
> > > > >> assignor
> > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> the
> > > > >> deprecation
> > > > >> > > > of
> > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > section
> > > > >> seems to
> > > > >> > > > have
> > > > >> > > > > > > > multiple typos.
> > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > immediately
> > > > >> gets into
> > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> incorrect.
> > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> to
> > > get
> > > > >> into
> > > > >> > > > > > epoch=3,
> > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > >> epoch=2,
> > > > >> > > > > > > > partitions=[foo-0].
> > > > >> > > > > > > >
> > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > upgrade
> > > > >> from
> > > > >> > > > any
> > > > >> > > > > > old
> > > > >> > > > > > > > version to new one?
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks,
> > > > >> > > > > > > >
> > > > >> > > > > > > > Jun
> > > > >> > > > > > > >
> > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > > Hi all,
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > coordinator
> > > > >> in
> > > > >> > > > Java. I
> > > > >> > > > > > > > > discussed this offline with a few folks and folks
> > are
> > > > >> concerned
> > > > >> > > > that
> > > > >> > > > > > > > > we could introduce many regressions in the old
> > > protocol
> > > > >> if we do
> > > > >> > > > so.
> > > > >> > > > > > > > > Therefore, I am going to remove this statement
> from
> > > the
> > > > >> KIP. It
> > > > >> > > > is an
> > > > >> > > > > > > > > implementation detail after all so it does not
> have
> > > to be
> > > > >> > > > decided at
> > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > refactor the
> > > > >> > > > current
> > > > >> > > > > > > > > implementation as a first step.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Cheers,
> > > > >> > > > > > > > > David
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > >> djacot@confluent.io
> > > > >> > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Hi Luke,
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > assigning,
> > > > >> > > > > > reconciling,
> > > > >> > > > > > > > > stable,
> > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > section,
> > > > >> right?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > This sentence does not refer to those group
> states
> > > but
> > > > >> rather
> > > > >> > > > to a
> > > > >> > > > > > > > > > state machine replication (SMR). This refers to
> > the
> > > > >> entire
> > > > >> > > > state of
> > > > >> > > > > > > > > > group coordinator which is replicated via the
> log
> > > > >> layer. I will
> > > > >> > > > > > > > > > clarify this in the KIP.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > modelled
> > > > >> as an
> > > > >> > > > event
> > > > >> > > > > > > > > loop"?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > The idea is to follow a model similar to the new
> > > quorum
> > > > >> > > > > > controller. We
> > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > >> __consumer_offsets
> > > > >> > > > > > > > > > partition is assigned to a unique thread and all
> > the
> > > > >> events
> > > > >> > > > (e.g.
> > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > > >> thread. This
> > > > >> > > > > > simplify
> > > > >> > > > > > > > > > concurrency and will enable us to do simulation
> > > testing
> > > > >> for the
> > > > >> > > > > > group
> > > > >> > > > > > > > > > coordinator.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > >> *__consumer_offsets*
> > > > >> > > > > > > > > partitions?
> > > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> > by
> > > a
> > > > >> group
> > > > >> > > > > > coordinator?
> > > > >> > > > > > > > > For
> > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > consumer
> > > > >> groups, and
> > > > >> > > > > > both
> > > > >> > > > > > > > > exist in
> > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > > >> for it,
> > > > >> > > > or
> > > > >> > > > > > 2?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> think.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> *is
> > > the
> > > > >> number
> > > > >> > > > of
> > > > >> > > > > > threads
> > > > >> > > > > > > > > used
> > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> if
> > > the
> > > > >> purpose
> > > > >> > > > of
> > > > >> > > > > > the
> > > > >> > > > > > > > > threads
> > > > >> > > > > > > > > > > is only to keep the state of each consumer
> group
> > > (or
> > > > >> > > > > > > > > *__consumer_offsets*
> > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > should we
> > > > >> need
> > > > >> > > > > > > > > multi-threads
> > > > >> > > > > > > > > > > here?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to shard
> > the
> > > > >> > > > processing as
> > > > >> > > > > > the
> > > > >> > > > > > > > > > computation could be heavy.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> the
> > > > >> default
> > > > >> > > > session
> > > > >> > > > > > > > > timeout not
> > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> thought
> > > the
> > > > >> min/max
> > > > >> > > > > > session
> > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > no?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > >> timeout to
> > > > >> > > > detect
> > > > >> > > > > > client
> > > > >> > > > > > > > > > > failures when using the consumer group
> protocol.
> > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > The
> > > > >> minimum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > The
> > > > >> maximum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > This is indeed a mistake. The default session
> > > timeout
> > > > >> should
> > > > >> > > > be 45s
> > > > >> > > > > > > > > > (the current default).
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > [range,
> > > > >> uniform],
> > > > >> > > > > > which means
> > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > know
> > > > >> why not
> > > > >> > > > > > uniform
> > > > >> > > > > > > > > one? I
> > > > >> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > > >> (former
> > > > >> > > > sticky
> > > > >> > > > > > > > > assinor)
> > > > >> > > > > > > > > > > for better evenly distribution. Any other
> reason
> > > we
> > > > >> choose
> > > > >> > > > range
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > as default?
> > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> The
> > > > >> server side
> > > > >> > > > > > assignors.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > The order on the server side has no influence
> > > because
> > > > >> the
> > > > >> > > > client
> > > > >> > > > > > must
> > > > >> > > > > > > > > > chose the selector that he wants to use. There
> is
> > no
> > > > >> default
> > > > >> > > > in the
> > > > >> > > > > > > > > > current proposal. If the assignor is not
> specified
> > > by
> > > > >> the
> > > > >> > > > client,
> > > > >> > > > > > the
> > > > >> > > > > > > > > > request is rejected. The default client value
> for
> > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > > that my
> > > > >> > > > answers
> > > > >> > > > > > help
> > > > >> > > > > > > > > > to clarify things. I will update the KIP as well
> > > based
> > > > >> on your
> > > > >> > > > > > > > > > feedback.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Cheers,
> > > > >> > > > > > > > > > David
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > >> showuon@gmail.com>
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Hi David,
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thanks for the update.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Some more questions:
> > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> mentioned:
> > > > >> > > > > > > > > > > > The new group coordinator will have a state
> > > machine
> > > > >> per
> > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where each
> > state
> > > > >> machine is
> > > > >> > > > > > modelled
> > > > >> > > > > > > > > as an
> > > > >> > > > > > > > > > > event loop. Those state machines will be
> > executed
> > > in
> > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > assigning,
> > > > >> > > > > > reconciling,
> > > > >> > > > > > > > > stable,
> > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > section,
> > > > >> right?
> > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > modelled
> > > > >> as an
> > > > >> > > > event
> > > > >> > > > > > > > > loop"?
> > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > >> *__consumer_offsets*
> > > > >> > > > > > > > > partitions?
> > > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> > by
> > > a
> > > > >> group
> > > > >> > > > > > coordinator?
> > > > >> > > > > > > > > For
> > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > consumer
> > > > >> groups, and
> > > > >> > > > > > both
> > > > >> > > > > > > > > exist in
> > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > > >> for it,
> > > > >> > > > or
> > > > >> > > > > > 2?
> > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> *is
> > > the
> > > > >> number
> > > > >> > > > of
> > > > >> > > > > > threads
> > > > >> > > > > > > > > used
> > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> if
> > > the
> > > > >> purpose
> > > > >> > > > of
> > > > >> > > > > > the
> > > > >> > > > > > > > > threads
> > > > >> > > > > > > > > > > is only to keep the state of each consumer
> group
> > > (or
> > > > >> > > > > > > > > *__consumer_offsets*
> > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > should we
> > > > >> need
> > > > >> > > > > > > > > multi-threads
> > > > >> > > > > > > > > > > here?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 2. For the default value in the new configs:
> > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> the
> > > > >> default
> > > > >> > > > session
> > > > >> > > > > > > > > timeout not
> > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> thought
> > > the
> > > > >> min/max
> > > > >> > > > > > session
> > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > no?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > >> timeout to
> > > > >> > > > detect
> > > > >> > > > > > client
> > > > >> > > > > > > > > > > failures when using the consumer group
> protocol.
> > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > The
> > > > >> minimum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > The
> > > > >> maximum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > [range,
> > > > >> uniform],
> > > > >> > > > > > which means
> > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > know
> > > > >> why not
> > > > >> > > > > > uniform
> > > > >> > > > > > > > > one? I
> > > > >> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > > >> (former
> > > > >> > > > sticky
> > > > >> > > > > > > > > assinor)
> > > > >> > > > > > > > > > > for better evenly distribution. Any other
> reason
> > > we
> > > > >> choose
> > > > >> > > > range
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > as default?
> > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> The
> > > > >> server side
> > > > >> > > > > > assignors.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thank you.
> > > > >> > > > > > > > > > > Luke
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > > >> showuon@gmail.com
> > > > >> > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > Hi Sagar,
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > >> integrating with
> > > > >> > > > > > KIP-848,
> > > > >> > > > > > > > > but I
> > > > >> > > > > > > > > > > > think we should have a separate discussion
> > > thread
> > > > >> for the
> > > > >> > > > Kafka
> > > > >> > > > > > > > > Connect
> > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > Consumer
> > > > >> Rebalance
> > > > >> > > > > > Protocol
> > > > >> > > > > > > > > [1],
> > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > consumer
> > > > >> rebalance
> > > > >> > > > > > protocol,
> > > > >> > > > > > > > > WDYT?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > [1]
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > Thank you.
> > > > >> > > > > > > > > > > > Luke
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > > Looks
> > > > >> like
> > > > >> > > > there's
> > > > >> > > > > > > > > agreement on
> > > > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> > > revisit
> > > > >> the doc
> > > > >> > > > and
> > > > >> > > > > > see
> > > > >> > > > > > > > > what
> > > > >> > > > > > > > > > > >> changes are to be made.
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> Thanks!
> > > > >> > > > > > > > > > > >> Sagar.
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > > > > >> wrote:
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > > That's
> > > > >> really
> > > > >> > > > > > helpful. I
> > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > > and the
> > > > >> > > > Consumer
> > > > >> > > > > > could
> > > > >> > > > > > > > > share
> > > > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > > > >> difference is
> > > > >> > > > that
> > > > >> > > > > > the
> > > > >> > > > > > > > > Consumer
> > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> whereas
> > > > >> Connect
> > > > >> > > > assigns
> > > > >> > > > > > tasks
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > support
> > > > >> different
> > > > >> > > > > > resource
> > > > >> > > > > > > > > types
> > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect.
> > The
> > > > >> dedicated
> > > > >> > > > APIs
> > > > >> > > > > > > > > would be
> > > > >> > > > > > > > > > > >> > similar to the new ones but different on
> > the
> > > > >> > > > > > content/resources and
> > > > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > > > >> coordinator
> > > > >> > > > side.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > > not a
> > > > >> fan of
> > > > >> > > > > > > > > overcharging
> > > > >> > > > > > > > > > > >> > APIs to serve different purposes. That
> > being
> > > > >> said, I am
> > > > >> > > > not
> > > > >> > > > > > > > > opposed to
> > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do
> it.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > I think that we can continue to discuss
> it
> > > here
> > > > >> for now
> > > > >> > > > in
> > > > >> > > > > > order
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible with
> > what
> > > we
> > > > >> will do
> > > > >> > > > for
> > > > >> > > > > > > > > Connect in
> > > > >> > > > > > > > > > > >> > the future.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Best,
> > > > >> > > > > > > > > > > >> > David
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> Jacot
> > <
> > > > >> > > > > > djacot@confluent.io>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > Hi all,
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > through
> > > and
> > > > >> address
> > > > >> > > > > > your
> > > > >> > > > > > > > > comments
> > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > feedback.
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > Cheers,
> > > > >> > > > > > > > > > > >> > > David
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > > Harris
> > > > >> <
> > > > >> > > > > > > > > gharris1727@gmail.com
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to
> > see
> > > > >> > > > cooperative
> > > > >> > > > > > > > > rebalancing
> > > > >> > > > > > > > > > > >> > making it
> > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> section
> > > > >> describes how
> > > > >> > > > > > member
> > > > >> > > > > > > > > epoch
> > > > >> > > > > > > > > > > >> > should
> > > > >> > > > > > > > > > > >> > > > advance in step with the group epoch
> > and
> > > > >> assignment
> > > > >> > > > > > epoch
> > > > >> > > > > > > > > values. I
> > > > >> > > > > > > > > > > >> > think
> > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > reasons
> > > > >> described
> > > > >> > > > in
> > > > >> > > > > > the
> > > > >> > > > > > > > > KIP. When
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > > >> partitions to a
> > > > >> > > > > > worker,
> > > > >> > > > > > > > > what
> > > > >> > > > > > > > > > > >> member
> > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> assignment
> > > use?
> > > > >> Are
> > > > >> > > > member
> > > > >> > > > > > epochs
> > > > >> > > > > > > > > > > >> re-used,
> > > > >> > > > > > > > > > > >> > and
> > > > >> > > > > > > > > > > >> > > > a single member epoch can correspond
> to
> > > > >> multiple
> > > > >> > > > > > different
> > > > >> > > > > > > > > > > >> > (monotonically
> > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > > opaque
> > > > >> to the
> > > > >> > > > group
> > > > >> > > > > > > > > > > >> coordinator? If
> > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> assignor
> > > > >> > > > implementations
> > > > >> > > > > > > > > interact
> > > > >> > > > > > > > > > > >> with
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> > > meaning
> > > > >> agreed
> > > > >> > > > > > upon? If
> > > > >> > > > > > > > > so, what
> > > > >> > > > > > > > > > > >> > is the
> > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field
> over
> > > > >> including
> > > > >> > > > such
> > > > >> > > > > > > > > functionality
> > > > >> > > > > > > > > > > >> > in the
> > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 3. The following is included in the
> > KIP:
> > > > >> "Thanks to
> > > > >> > > > > > this, the
> > > > >> > > > > > > > > input
> > > > >> > > > > > > > > > > >> of
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> driven
> > > by
> > > > >> the group
> > > > >> > > > > > > > > coordinator.
> > > > >> > > > > > > > > > > >> The
> > > > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > > >> maintaining
> > > > >> > > > any
> > > > >> > > > > > state
> > > > >> > > > > > > > > besides
> > > > >> > > > > > > > > > > >> its
> > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > > that the
> > > > >> > > > > > client-side
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > >> MAY
> > > > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> > state
> > > > >> (such as
> > > > >> > > > > > partition
> > > > >> > > > > > > > > > > >> > throughput,
> > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics,
> > etc),
> > > or
> > > > >> that
> > > > >> > > > > > additional
> > > > >> > > > > > > > > > > >> > non-Metadata
> > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > classes
> > > > >> > > > > > > > > > > >> > > > for
> > > > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > >> > > > > > > > > > > >> > > > and
> > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > >> > > > > > that
> > > > >> > > > > > > > > seem to
> > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it possible
> > for
> > > > >> these two
> > > > >> > > > > > > > > implementations
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> > > feature
> > > > >> parity
> > > > >> > > > of
> > > > >> > > > > > > > > server-side
> > > > >> > > > > > > > > > > >> and
> > > > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > > >> facilitate
> > > > >> > > > > > operational
> > > > >> > > > > > > > > > > >> > flexibility in
> > > > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > > >> server-side
> > > > >> > > > > > assignor
> > > > >> > > > > > > > > has some
> > > > >> > > > > > > > > > > >> > poor
> > > > >> > > > > > > > > > > >> > > > behavior and needs a patch, deploying
> > the
> > > > >> patched
> > > > >> > > > > > assignor to
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > > >> client-side
> > > > >> > > > > > assignor
> > > > >> > > > > > > > > may be
> > > > >> > > > > > > > > > > >> > faster
> > > > >> > > > > > > > > > > >> > > > and less risky than patching all of
> the
> > > > >> brokers.
> > > > >> > > > With
> > > > >> > > > > > the
> > > > >> > > > > > > > > currently
> > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > > >> > > > reimplementation
> > > > >> > > > > > would
> > > > >> > > > > > > > > have
> > > > >> > > > > > > > > > > >> to be
> > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > > diverged
> > > > >> > > > > > significantly,
> > > > >> > > > > > > > > then it
> > > > >> > > > > > > > > > > >> is
> > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> would
> > > not be
> > > > >> > > > possible.
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > --
> > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar
> <
> > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > discuss
> > > how
> > > > >> Connect
> > > > >> > > > > > would
> > > > >> > > > > > > > > need to
> > > > >> > > > > > > > > > > >> > change
> > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> protocol.
> > > > >> Here's the
> > > > >> > > > page:
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > > tried
> > > > >> to keep
> > > > >> > > > a
> > > > >> > > > > > format
> > > > >> > > > > > > > > > > >> similar to
> > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> think.
> > > Also,
> > > > >> do you
> > > > >> > > > > > think this
> > > > >> > > > > > > > > > > >> should
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> thread
> > > or
> > > > >> is this
> > > > >> > > > one
> > > > >> > > > > > fine?
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > Sagar <
> > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > > Kafka
> > > > >> Streams.
> > > > >> > > > > > Sure, I
> > > > >> > > > > > > > > would do
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > > >> document.
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > > Guozhang
> > > > >> Wang <
> > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> It would be great if you could
> > come
> > > > >> back with
> > > > >> > > > some
> > > > >> > > > > > > > > analysis on
> > > > >> > > > > > > > > > > >> > how to
> > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > integration
> > > > >> with
> > > > >> > > > the new
> > > > >> > > > > > > > > protocol;
> > > > >> > > > > > > > > > > >> so
> > > > >> > > > > > > > > > > >> > far
> > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > > "protocol
> > > > >> type"
> > > > >> > > > we
> > > > >> > > > > > did not
> > > > >> > > > > > > > > yet
> > > > >> > > > > > > > > > > >> think
> > > > >> > > > > > > > > > > >> > > > > through
> > > > >> > > > > > > > > > > >> > > > > >> the Connect side
> implementations.
> > > For
> > > > >> Streams
> > > > >> > > > > > here's a
> > > > >> > > > > > > > > draft of
> > > > >> > > > > > > > > > > >> > > > > >> integration
> > > > >> > > > > > > > > > > >> > > > > >> plan:
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> > > Connect.
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> > > Sagar <
> > > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > Thank you for your response.
> The
> > > > >> reason I
> > > > >> > > > thought
> > > > >> > > > > > > > > connect can
> > > > >> > > > > > > > > > > >> > also fit
> > > > >> > > > > > > > > > > >> > > > > >> into
> > > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that even
> > > today the
> > > > >> > > > connect
> > > > >> > > > > > uses a
> > > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > >> > > > > > > > > > > >> > > > > >> > extending from
> > > AbstractCoordinator to
> > > > >> empower
> > > > >> > > > > > > > > rebalances of
> > > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > > >> WorkerCoordinator sets
> > > > >> > > > the
> > > > >> > > > > > > > > > > >> protocolType()
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> connect
> > > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata() method
> > by
> > > > >> plumbing
> > > > >> > > > into
> > > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > > > >> connect would
> > > > >> > > > be
> > > > >> > > > > > > > > similar at a
> > > > >> > > > > > > > > > > >> > high
> > > > >> > > > > > > > > > > >> > > > > level
> > > > >> > > > > > > > > > > >> > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> > > because
> > > > >> of the
> > > > >> > > > > > Client
> > > > >> > > > > > > > > side
> > > > >> > > > > > > > > > > >> > assignors
> > > > >> > > > > > > > > > > >> > > > > being
> > > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> > implementation
> > > > >> level, we
> > > > >> > > > > > might
> > > > >> > > > > > > > > need to
> > > > >> > > > > > > > > > > >> make
> > > > >> > > > > > > > > > > >> > a lot
> > > > >> > > > > > > > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > > > >> assignment
> > > > >> > > > protocol
> > > > >> > > > > > like
> > > > >> > > > > > > > > > > >> enhancing
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > > > >> SyncGroup and
> > > > >> > > > > > using
> > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar lines
> > to
> > > > >> streams (or
> > > > >> > > > > > there
> > > > >> > > > > > > > > might be
> > > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > > >> > > > > > > > > > > >> > > > > >> > would try to perform a
> detailed
> > > > >> analysis of
> > > > >> > > > the
> > > > >> > > > > > same
> > > > >> > > > > > > > > and we
> > > > >> > > > > > > > > > > >> > can have
> > > > >> > > > > > > > > > > >> > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > separate discussion thread for
> > > that
> > > > >> as that
> > > > >> > > > would
> > > > >> > > > > > > > > derail this
> > > > >> > > > > > > > > > > >> > > > > discussion
> > > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if that
> > sounds
> > > > >> good to
> > > > >> > > > you.
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47
> PM
> > > David
> > > > >> Jacot
> > > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > > >> `Assignment#error`.
> > > > >> > > > > > Sure, I
> > > > >> > > > > > > > > can
> > > > >> > > > > > > > > > > >> > mention it.
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to
> transition C
> > > from
> > > > >> his
> > > > >> > > > current
> > > > >> > > > > > > > > assignment
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > his
> > > > >> > > > > > > > > > > >> > > > > >> > > target assignment when he
> can
> > > move
> > > > >> to
> > > > >> > > > epoch 3.
> > > > >> > > > > > When
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > happens,
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > member assignment is updated
> > and
> > > > >> persisted
> > > > >> > > > > > with all
> > > > >> > > > > > > > > its
> > > > >> > > > > > > > > > > >> > assigned
> > > > >> > > > > > > > > > > >> > > > > >> > > partitions even if they are
> > not
> > > all
> > > > >> revoked
> > > > >> > > > > > yet. In
> > > > >> > > > > > > > > other
> > > > >> > > > > > > > > > > >> > words, the
> > > > >> > > > > > > > > > > >> > > > > >> > > member assignment becomes
> the
> > > target
> > > > >> > > > > > assignment.
> > > > >> > > > > > > > > This is
> > > > >> > > > > > > > > > > >> > basically
> > > > >> > > > > > > > > > > >> > > > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid having
> > to
> > > > >> write all
> > > > >> > > > the
> > > > >> > > > > > > > > changes to
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > log.
> > > > >> > > > > > > > > > > >> > > > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > examples are based on the
> > > persisted
> > > > >> state
> > > > >> > > > so I
> > > > >> > > > > > > > > understand
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I
> can
> > > > >> improve
> > > > >> > > > this in
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> > description.
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it
> could
> > > > >> reuse the
> > > > >> > > > > > protocol
> > > > >> > > > > > > > > with a
> > > > >> > > > > > > > > > > >> > client side
> > > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > > > >> protocol. The
> > > > >> > > > > > assignment
> > > > >> > > > > > > > > is
> > > > >> > > > > > > > > > > >> about
> > > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions +
> metadata,
> > > could
> > > > >> > > > Connect
> > > > >> > > > > > fit
> > > > >> > > > > > > > > into this?
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > David
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55
> > PM
> > > > >> Sagar <
> > > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just
> > had
> > > > >> minor
> > > > >> > > > > > observations:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > > > >> section in
> > > > >> > > > Client
> > > > >> > > > > > Side
> > > > >> > > > > > > > > mode
> > > > >> > > > > > > > > > > >> > Assignment
> > > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this
> > > case,
> > > > >> the
> > > > >> > > > client
> > > > >> > > > > > side
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > >> can
> > > > >> > > > > > > > > > > >> > > > > return
> > > > >> > > > > > > > > > > >> > > > > >> an
> > > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > coordinator`.
> > > > >> In this
> > > > >> > > > > > case are
> > > > >> > > > > > > > > you
> > > > >> > > > > > > > > > > >> > referring to
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > > >> AssignmentError
> > > > >> > > > that's
> > > > >> > > > > > > > > listed down
> > > > >> > > > > > > > > > > >> > towards
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > end?
> > > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it
> > would
> > > > >> make sense
> > > > >> > > > to
> > > > >> > > > > > > > > mention this
> > > > >> > > > > > > > > > > >> > > > > explicitly
> > > > >> > > > > > > > > > > >> > > > > >> > here?
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > > section, I
> > > > >> have a
> > > > >> > > > > > slight
> > > > >> > > > > > > > > > > >> confusion,
> > > > >> > > > > > > > > > > >> > not
> > > > >> > > > > > > > > > > >> > > > > sure
> > > > >> > > > > > > > > > > >> > > > > >> if
> > > > >> > > > > > > > > > > >> > > > > >> > > > others have the same.
> > Consider
> > > > >> this step:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the
> group
> > > > >> coordinator
> > > > >> > > > > > > > > transitions him
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > epoch
> > > > >> > > > > > > > > > > >> > > > > 3
> > > > >> > > > > > > > > > > >> > > > > >> > > because
> > > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > > revoke. It
> > > > >> > > > persists
> > > > >> > > > > > the
> > > > >> > > > > > > > > change and
> > > > >> > > > > > > > > > > >> > reply.
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > > (epoch=3)
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> > partitions=[foo-0]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> > partitions=[foo-2]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> > partitions=[foo-1]
> > > > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > > >> partitions=[foo-0,
> > > > >> > > > > > foo-1]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > > >> partitions=[foo-2]
> > > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > > >> partitions=[foo-1]
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > > transitions
> > > > >> to
> > > > >> > > > epoch 3
> > > > >> > > > > > but
> > > > >> > > > > > > > > cannot
> > > > >> > > > > > > > > > > >> get
> > > > >> > > > > > > > > > > >> > foo-1
> > > > >> > > > > > > > > > > >> > > > > >> yet.
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that
> > > member C
> > > > >> can't
> > > > >> > > > get
> > > > >> > > > > > the
> > > > >> > > > > > > > > foo-1
> > > > >> > > > > > > > > > > >> > partition
> > > > >> > > > > > > > > > > >> > > > > yet,
> > > > >> > > > > > > > > > > >> > > > > >> > but
> > > > >> > > > > > > > > > > >> > > > > >> > > > based on the description
> > > above,
> > > > >> it seems
> > > > >> > > > it
> > > > >> > > > > > > > > already has
> > > > >> > > > > > > > > > > >> it.
> > > > >> > > > > > > > > > > >> > Do you
> > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > >> > > > > > > > > > > >> > > > > >> > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > would be better to remove
> it
> > > and
> > > > >> > > > populate it
> > > > >> > > > > > only
> > > > >> > > > > > > > > when it
> > > > >> > > > > > > > > > > >> > actually
> > > > >> > > > > > > > > > > >> > > > > >> gets
> > > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of
> other
> > > > >> places, so
> > > > >> > > > have
> > > > >> > > > > > I
> > > > >> > > > > > > > > > > >> understood it
> > > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it
> might
> > > be
> > > > >> out of
> > > > >> > > > scope
> > > > >> > > > > > of
> > > > >> > > > > > > > > this
> > > > >> > > > > > > > > > > >> > discussion,
> > > > >> > > > > > > > > > > >> > > > > but
> > > > >> > > > > > > > > > > >> > > > > >> > from
> > > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > > > >> probably be
> > > > >> > > > > > running in
> > > > >> > > > > > > > > client
> > > > >> > > > > > > > > > > >> > side
> > > > >> > > > > > > > > > > >> > > > > >> assignor
> > > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > > >> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > > > >> protocol as it
> > > > >> > > > has
> > > > >> > > > > > its
> > > > >> > > > > > > > > own
> > > > >> > > > > > > > > > > >> Custom
> > > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > >> > > > > > > > > > > >> > > > > >> > > > and
> IncrementalCooperative).
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at
> 5:00
> > > PM
> > > > >> David
> > > > >> > > > Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal
> is
> > > to
> > > > >> move
> > > > >> > > > forward
> > > > >> > > > > > with
> > > > >> > > > > > > > > > > >> > specialized API
> > > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying on
> one
> > > > >> generic API.
> > > > >> > > > For
> > > > >> > > > > > > > > Connect, we
> > > > >> > > > > > > > > > > >> > can apply
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > > > >> reuse/share the
> > > > >> > > > core
> > > > >> > > > > > > > > > > >> > implementation on
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the
> > schema
> > > > >> registry, I
> > > > >> > > > > > think
> > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > >> > should
> > > > >> > > > > > > > > > > >> > > > > >> consider
> > > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to
> > do
> > > > >> simple
> > > > >> > > > > > > > > membership/leader
> > > > >> > > > > > > > > > > >> > election.
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at
> > > 10:22
> > > > >> AM Ismael
> > > > >> > > > > > Juma <
> > > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > > >> java.util.regex.Pattern
> > > > >> > > > vs
> > > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > document the
> differences
> > > in
> > > > >> more
> > > > >> > > > detail
> > > > >> > > > > > before
> > > > >> > > > > > > > > > > >> deciding
> > > > >> > > > > > > > > > > >> > one
> > > > >> > > > > > > > > > > >> > > > > way
> > > > >> > > > > > > > > > > >> > > > > >> or
> > > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if people
> > pass
> > > > >> > > > > > > > > java.util.regex.Pattern,
> > > > >> > > > > > > > > > > >> they
> > > > >> > > > > > > > > > > >> > expect
> > > > >> > > > > > > > > > > >> > > > > >> > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be
> honored.
> > > If
> > > > >> we are
> > > > >> > > > doing
> > > > >> > > > > > > > > something
> > > > >> > > > > > > > > > > >> > different,
> > > > >> > > > > > > > > > > >> > > > > >> then
> > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider adding
> > an
> > > > >> overload
> > > > >> > > > with
> > > > >> > > > > > our own
> > > > >> > > > > > > > > > > >> Pattern
> > > > >> > > > > > > > > > > >> > class
> > > > >> > > > > > > > > > > >> > > > > (I
> > > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to
> > expose
> > > > >> re2j's at
> > > > >> > > > this
> > > > >> > > > > > > > > point).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic
> ids,
> > > any
> > > > >> major new
> > > > >> > > > > > protocol
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> > > > > integrate
> > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should handle
> the
> > > topic
> > > > >> > > > > > recreation case
> > > > >> > > > > > > > > > > >> > correctly.
> > > > >> > > > > > > > > > > >> > > > > That's
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > main
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to
> handle.
> > I
> > > > >> agree with
> > > > >> > > > > > David
> > > > >> > > > > > > > > that we'd
> > > > >> > > > > > > > > > > >> > want to
> > > > >> > > > > > > > > > > >> > > > > add
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> > protocols
> > > > >> that don't
> > > > >> > > > > > have it
> > > > >> > > > > > > > > yet and
> > > > >> > > > > > > > > > > >> > that we
> > > > >> > > > > > > > > > > >> > > > > can
> > > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> > > versus
> > > > >> adding
> > > > >> > > > new
> > > > >> > > > > > APIs
> > > > >> > > > > > > > > to the
> > > > >> > > > > > > > > > > >> > Java
> > > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that adding
> new
> > > APIs
> > > > >> is
> > > > >> > > > required
> > > > >> > > > > > for
> > > > >> > > > > > > > > > > >> reasonable
> > > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> > > about
> > > > >> the
> > > > >> > > > > > coordinator
> > > > >> > > > > > > > > storing
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> configs.
> > > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs
> to
> > be
> > > > >> > > > centralized
> > > > >> > > > > > in the
> > > > >> > > > > > > > > > > >> metadata
> > > > >> > > > > > > > > > > >> > log for
> > > > >> > > > > > > > > > > >> > > > > >> > > various
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > > > >> visibility,
> > > > >> > > > > > consistency,
> > > > >> > > > > > > > > > > >> etc.).
> > > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > >> > > > > > > > > > > >> > > > > >> > > am
> > > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about
> automatically
> > > > >> deleting
> > > > >> > > > > > configs in a
> > > > >> > > > > > > > > way
> > > > >> > > > > > > > > > > >> that
> > > > >> > > > > > > > > > > >> > they
> > > > >> > > > > > > > > > > >> > > > > >> cannot
> > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> > property
> > > > >> for modern
> > > > >> > > > > > systems
> > > > >> > > > > > > > > is to
> > > > >> > > > > > > > > > > >> > minimize
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > number
> > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data
> loss
> > > > >> scenarios.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022
> at
> > > 3:47
> > > > >> PM David
> > > > >> > > > > > Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> > > answers
> > > > >> are
> > > > >> > > > below:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > path,
> > > > >> especially
> > > > >> > > > > > the last
> > > > >> > > > > > > > > step
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > > > > clients
> > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > > >> protocol, in
> > > > >> > > > which
> > > > >> > > > > > we
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > window
> > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs
> and
> > > old
> > > > >> > > > protocols /
> > > > >> > > > > > rpcs
> > > > >> > > > > > > > > are
> > > > >> > > > > > > > > > > >> used
> > > > >> > > > > > > > > > > >> > by
> > > > >> > > > > > > > > > > >> > > > > >> members
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > >> coordinator could
> > > > >> > > > > > "mimic"
> > > > >> > > > > > > > > the old
> > > > >> > > > > > > > > > > >> > behavior
> > > > >> > > > > > > > > > > >> > > > > >> while
> > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > something
> > > > >> we
> > > > >> > > > need to
> > > > >> > > > > > > > > present
> > > > >> > > > > > > > > > > >> about.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > published
> > > a
> > > > >> new
> > > > >> > > > version
> > > > >> > > > > > of KIP
> > > > >> > > > > > > > > which
> > > > >> > > > > > > > > > > >> > includes
> > > > >> > > > > > > > > > > >> > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about this.
> > See
> > > the
> > > > >> > > > "Supporting
> > > > >> > > > > > > > > Online
> > > > >> > > > > > > > > > > >> > Consumer
> > > > >> > > > > > > > > > > >> > > > > Group
> > > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> > "Compatibility,
> > > > >> > > > Deprecation,
> > > > >> > > > > > and
> > > > >> > > > > > > > > Migration
> > > > >> > > > > > > > > > > >> > Plan". I
> > > > >> > > > > > > > > > > >> > > > > >> think
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> > through
> > > a
> > > > >> few cases
> > > > >> > > > > > now but
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > overall idea
> > > > >> > > > > > > > > > > >> > > > > >> and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > > > >> understandable.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> > topic
> > > > >> ids. So
> > > > >> > > > far as
> > > > >> > > > > > > > > KIP-516 the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > >> are
> > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > admin
> > > > >> client,
> > > > >> > > > but
> > > > >> > > > > > they
> > > > >> > > > > > > > > are not
> > > > >> > > > > > > > > > > >> > exposed
> > > > >> > > > > > > > > > > >> > > > > via
> > > > >> > > > > > > > > > > >> > > > > >> any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet.
> I
> > > think
> > > > >> the
> > > > >> > > > > > question is,
> > > > >> > > > > > > > > first
> > > > >> > > > > > > > > > > >> > should we
> > > > >> > > > > > > > > > > >> > > > > >> let
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > maintaining
> > > > >> the
> > > > >> > > > names
> > > > >> > > > > > -> ids
> > > > >> > > > > > > > > mapping
> > > > >> > > > > > > > > > > >> > itself
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> augmented
> > > > >> existing
> > > > >> > > > RPCs
> > > > >> > > > > > and the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> RPCs
> > > > >> > > > > > > > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should
> we
> > > ever
> > > > >> consider
> > > > >> > > > > > exposing
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> well
> > > (both
> > > > >> > > > > > > > > subscribe/assign, as
> > > > >> > > > > > > > > > > >> well
> > > > >> > > > > > > > > > > >> > as in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > > like
> > > > >> topic
> > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we
> > > would
> > > > >> include
> > > > >> > > > > > converting
> > > > >> > > > > > > > > all
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > offsets
> > > > >> > > > > > > > > > > >> > > > > >> > > related
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic
> > ids
> > > in
> > > > >> this
> > > > >> > > > KIP,
> > > > >> > > > > > the
> > > > >> > > > > > > > > consumer
> > > > >> > > > > > > > > > > >> > would be
> > > > >> > > > > > > > > > > >> > > > > >> able
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with
> > topic
> > > > >> ids. That
> > > > >> > > > > > being
> > > > >> > > > > > > > > said, it
> > > > >> > > > > > > > > > > >> > still has
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> > > various
> > > > >> APIs so
> > > > >> > > > > > having a
> > > > >> > > > > > > > > > > >> mapping
> > > > >> > > > > > > > > > > >> > in the
> > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to
> > me.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a
> > strong
> > > > >> opinion on
> > > > >> > > > > > this.
> > > > >> > > > > > > > > Here I
> > > > >> > > > > > > > > > > >> > wonder if
> > > > >> > > > > > > > > > > >> > > > > >> this
> > > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of
> > this
> > > > >> KIP. I
> > > > >> > > > would
> > > > >> > > > > > rather
> > > > >> > > > > > > > > focus
> > > > >> > > > > > > > > > > >> on
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > internals
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> > consider
> > > > >> this
> > > > >> > > > > > separately if
> > > > >> > > > > > > > > we see
> > > > >> > > > > > > > > > > >> > value in
> > > > >> > > > > > > > > > > >> > > > > >> doing
> > > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> > Ismael's
> > > > >> point about
> > > > >> > > > > > using
> > > > >> > > > > > > > > topic ids
> > > > >> > > > > > > > > > > >> > in the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> ConsumerGroupHeartbeatRequest, I
> > > > >> > > > think
> > > > >> > > > > > that
> > > > >> > > > > > > > > there
> > > > >> > > > > > > > > > > >> is
> > > > >> > > > > > > > > > > >> > one
> > > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > > >> > > > > > > > > > > >> > > > > >> > > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> > > consumer
> > > > >> will
> > > > >> > > > have
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> opportunity
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> validate
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists
> > before
> > > > >> passing
> > > > >> > > > them
> > > > >> > > > > > into
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> group
> > > > >> > > > > > > > > > > >> > > > > rebalance
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously,
> > the
> > > > >> > > > coordinator
> > > > >> > > > > > will
> > > > >> > > > > > > > > also
> > > > >> > > > > > > > > > > >> notice
> > > > >> > > > > > > > > > > >> > it but
> > > > >> > > > > > > > > > > >> > > > > >> it
> > > > >> > > > > > > > > > > >> > > > > >> > > does
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have a
> way
> > to
> > > > >> reject an
> > > > >> > > > > > invalid
> > > > >> > > > > > > > > topic in
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> response.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > > David
> > > > >> on all
> > > > >> > > > other
> > > > >> > > > > > minor
> > > > >> > > > > > > > > > > >> questions
> > > > >> > > > > > > > > > > >> > > > > except
> > > > >> > > > > > > > > > > >> > > > > >> for
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> `subscribe(Pattern)`
> > > > >> question:
> > > > >> > > > > > personally
> > > > >> > > > > > > > > I think
> > > > >> > > > > > > > > > > >> > it's not
> > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > subscribe
> > > > >> API with
> > > > >> > > > > > Pattern,
> > > > >> > > > > > > > > but
> > > > >> > > > > > > > > > > >> > instead we
> > > > >> > > > > > > > > > > >> > > > > >> still
> > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > documenting
> > > > >> that our
> > > > >> > > > > > > > > subscription may
> > > > >> > > > > > > > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > incompatible
> > > > >> case is a
> > > > >> > > > > > very rare
> > > > >> > > > > > > > > > > >> scenario
> > > > >> > > > > > > > > > > >> > I felt
> > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> `String`
> > > based
> > > > >> > > > > > subscription may
> > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I
> > have
> > > to
> > > > >> look at
> > > > >> > > > the
> > > > >> > > > > > > > > differences
> > > > >> > > > > > > > > > > >> > between
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > two
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > > > >> understand the
> > > > >> > > > > > potential
> > > > >> > > > > > > > > issues.
> > > > >> > > > > > > > > > > >> My
> > > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work for
> > all
> > > the
> > > > >> basic
> > > > >> > > > > > regular
> > > > >> > > > > > > > > > > >> > expressions. The
> > > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> > > mainly
> > > > >> about
> > > > >> > > > the
> > > > >> > > > > > various
> > > > >> > > > > > > > > > > >> character
> > > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other
> > people
> > > > >> think
> > > > >> > > > about
> > > > >> > > > > > this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022
> > at
> > > > >> 11:28 PM
> > > > >> > > > > > Guozhang
> > > > >> > > > > > > > > Wang <
> > > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > >> > > > > > > > > > > >> > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I
> > think
> > > on
> > > > >> the high
> > > > >> > > > > > level
> > > > >> > > > > > > > > there are
> > > > >> > > > > > > > > > > >> > two meta
> > > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a
> bit
> > > more:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > path,
> > > > >> especially
> > > > >> > > > > > the last
> > > > >> > > > > > > > > step
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > > > > clients
> > > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > > >> protocol, in
> > > > >> > > > which
> > > > >> > > > > > we
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > window
> > > > >> > > > > > > > > > > >> > > > > >> > where
> > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs
> and
> > > old
> > > > >> > > > protocols /
> > > > >> > > > > > rpcs
> > > > >> > > > > > > > > are
> > > > >> > > > > > > > > > > >> used
> > > > >> > > > > > > > > > > >> > by
> > > > >> > > > > > > > > > > >> > > > > >> members
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > > >> coordinator could
> > > > >> > > > > > "mimic"
> > > > >> > > > > > > > > the old
> > > > >> > > > > > > > > > > >> > behavior
> > > > >> > > > > > > > > > > >> > > > > >> while
> > > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > something
> > > > >> we
> > > > >> > > > need to
> > > > >> > > > > > > > > present
> > > > >> > > > > > > > > > > >> about.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> > topic
> > > > >> ids. So
> > > > >> > > > far as
> > > > >> > > > > > > > > KIP-516 the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > >> are
> > > > >> > > > > > > > > > > >> > > > > >> > > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > admin
> > > > >> client,
> > > > >> > > > but
> > > > >> > > > > > they
> > > > >> > > > > > > > > are not
> > > > >> > > > > > > > > > > >> > exposed
> > > > >> > > > > > > > > > > >> > > > > via
> > > > >> > > > > > > > > > > >> > > > > >> any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet.
> I
> > > think
> > > > >> the
> > > > >> > > > > > question is,
> > > > >> > > > > > > > > first
> > > > >> > > > > > > > > > > >> > should we
> > > > >> > > > > > > > > > > >> > > > > >> let
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > maintaining
> > > > >> the
> > > > >> > > > names
> > > > >> > > > > > -> ids
> > > > >> > > > > > > > > mapping
> > > > >> > > > > > > > > > > >> > itself
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > fully
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the
> augmented
> > > > >> existing
> > > > >> > > > RPCs
> > > > >> > > > > > and the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> RPCs
> > > > >> > > > > > > > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should
> we
> > > ever
> > > > >> consider
> > > > >> > > > > > exposing
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > topic ids
> > > > >> > > > > > > > > > > >> > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as
> well
> > > (both
> > > > >> > > > > > > > > subscribe/assign, as
> > > > >> > > > > > > > > > > >> well
> > > > >> > > > > > > > > > > >> > as in
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > > like
> > > > >> topic
> > > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > > David
> > > > >> on all
> > > > >> > > > other
> > > > >> > > > > > minor
> > > > >> > > > > > > > > > > >> questions
> > > > >> > > > > > > > > > > >> > > > > except
> > > > >> > > > > > > > > > > >> > > > > >> for
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> `subscribe(Pattern)`
> > > > >> question:
> > > > >> > > > > > personally
> > > > >> > > > > > > > > I think
> > > > >> > > > > > > > > > > >> > it's not
> > > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > > subscribe
> > > > >> API with
> > > > >> > > > > > Pattern,
> > > > >> > > > > > > > > but
> > > > >> > > > > > > > > > > >> > instead we
> > > > >> > > > > > > > > > > >> > > > > >> still
> > > > >> > > > > > > > > > > >> > > > > >> > > use
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > documenting
> > > > >> that our
> > > > >> > > > > > > > > subscription may
> > > > >> > > > > > > > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > incompatible
> > > > >> case is a
> > > > >> > > > > > very rare
> > > > >> > > > > > > > > > > >> scenario
> > > > >> > > > > > > > > > > >> > I felt
> > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > >> > > > > > > > > > > >> > > > > >> > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded
> `String`
> > > based
> > > > >> > > > > > subscription may
> > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12,
> 2022
> > > at
> > > > >> 5:23 AM
> > > > >> > > > > > David Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > <djacot@confluent.io.invalid
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > > > >> feedback. Let
> > > > >> > > > me
> > > > >> > > > > > answer
> > > > >> > > > > > > > > your
> > > > >> > > > > > > > > > > >> > questions
> > > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think
> it's
> > > > >> premature to
> > > > >> > > > > > talk about
> > > > >> > > > > > > > > > > >> target
> > > > >> > > > > > > > > > > >> > > > > versions
> > > > >> > > > > > > > > > > >> > > > > >> for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > > > >> existing group
> > > > >> > > > > > protocol.
> > > > >> > > > > > > > > > > >> Unlike
> > > > >> > > > > > > > > > > >> > KRaft,
> > > > >> > > > > > > > > > > >> > > > > >> this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client
> protocol
> > > and
> > > > >> hence
> > > > >> > > > > > > > > deprecation/removal
> > > > >> > > > > > > > > > > >> > will be
> > > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > > > >> applications
> > > > >> > > > migrate
> > > > >> > > > > > to
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes
> sense.
> > I
> > > will
> > > > >> > > > remove it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP
> says
> > we
> > > > >> intend to
> > > > >> > > > > > release
> > > > >> > > > > > > > > this in
> > > > >> > > > > > > > > > > >> > 4.x, but
> > > > >> > > > > > > > > > > >> > > > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we
> added
> > > that
> > > > >> as a
> > > > >> > > > way to
> > > > >> > > > > > > > > estimate
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > we'd
> > > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > > protocol, I
> > > > >> also
> > > > >> > > > > > suggest
> > > > >> > > > > > > > > removing
> > > > >> > > > > > > > > > > >> > this part.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain
> my
> > > > >> reasoning. As
> > > > >> > > > > > > > > explained, I
> > > > >> > > > > > > > > > > >> plan
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in
> > Java
> > > > >> while we
> > > > >> > > > > > implement
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> This
> > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the
> internals
> > > will
> > > > >> be
> > > > >> > > > slightly
> > > > >> > > > > > > > > different
> > > > >> > > > > > > > > > > >> > (e.g.
> > > > >> > > > > > > > > > > >> > > > > >> threading
> > > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I
> > wanted
> > > to
> > > > >> tighten
> > > > >> > > > the
> > > > >> > > > > > > > > switch from
> > > > >> > > > > > > > > > > >> > the old
> > > > >> > > > > > > > > > > >> > > > > >> group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to
> the
> > > new
> > > > >> group
> > > > >> > > > > > coordinator
> > > > >> > > > > > > > > to a
> > > > >> > > > > > > > > > > >> > major
> > > > >> > > > > > > > > > > >> > > > > >> release.
> > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative
> would
> > > be to
> > > > >> use a
> > > > >> > > > flag
> > > > >> > > > > > to do
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > switch
> > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> > > upgrade.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to
> > > flesh
> > > > >> out the
> > > > >> > > > > > details of
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > migration
> > > > >> > > > > > > > > > > >> > > > > >> story.
> > > > >> > > > > > > > > > > >> > > > > >> > > It
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying
> we
> > > will
> > > > >> support
> > > > >> > > > > > online
> > > > >> > > > > > > > > > > >> migrations.
> > > > >> > > > > > > > > > > >> > Is
> > > > >> > > > > > > > > > > >> > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this
> in
> > > > >> detail. It
> > > > >> > > > could
> > > > >> > > > > > also
> > > > >> > > > > > > > > be done
> > > > >> > > > > > > > > > > >> > as a
> > > > >> > > > > > > > > > > >> > > > > >> separate
> > > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> > support
> > > > >> online
> > > > >> > > > > > migrations
> > > > >> > > > > > > > > for the
> > > > >> > > > > > > > > > > >> > group.
> > > > >> > > > > > > > > > > >> > > > > That
> > > > >> > > > > > > > > > > >> > > > > >> > > means
> > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using
> the
> > > old
> > > > >> protocol
> > > > >> > > > > > will be
> > > > >> > > > > > > > > able to
> > > > >> > > > > > > > > > > >> > switch to
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> > > explain
> > > > >> how that
> > > > >> > > > > > will work
> > > > >> > > > > > > > > > > >> though.
> > > > >> > > > > > > > > > > >> > It is
> > > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step
> process:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster
> > must
> > > be
> > > > >> > > > upgraded or
> > > > >> > > > > > > > > rolled to a
> > > > >> > > > > > > > > > > >> > software
> > > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > coordinator.
> > > > >> Both the
> > > > >> > > > > > old and
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > > will
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > > > >> protocol and
> > > > >> > > > rely
> > > > >> > > > > > on the
> > > > >> > > > > > > > > same
> > > > >> > > > > > > > > > > >> > persisted
> > > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> > > together.
> > > > >> This
> > > > >> > > > point
> > > > >> > > > > > is an
> > > > >> > > > > > > > > > > >> offline
> > > > >> > > > > > > > > > > >> > > > > >> migration.
> > > > >> > > > > > > > > > > >> > > > > >> > We
> > > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > > > >> because it
> > > > >> > > > would
> > > > >> > > > > > require
> > > > >> > > > > > > > > > > >> > shutting down
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > current
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > > > >> starting up the
> > > > >> > > > > > new one
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> that
> > > > >> > > > > > > > > > > >> > would
> > > > >> > > > > > > > > > > >> > > > > >> cause
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> unavailabilities.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > > > >> metadata
> > > > >> > > > > > version/IBP
> > > > >> > > > > > > > > must be
> > > > >> > > > > > > > > > > >> > upgraded
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> X
> > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the
> new
> > > > >> protocol.
> > > > >> > > > This
> > > > >> > > > > > cannot
> > > > >> > > > > > > > > be done
> > > > >> > > > > > > > > > > >> > before
> > > > >> > > > > > > > > > > >> > > > > 1)
> > > > >> > > > > > > > > > > >> > > > > >> is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated
> because
> > > the
> > > > >> old
> > > > >> > > > > > coordinator
> > > > >> > > > > > > > > doesn't
> > > > >> > > > > > > > > > > >> > support
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers
> > > must be
> > > > >> > > > upgraded
> > > > >> > > > > > to a
> > > > >> > > > > > > > > version
> > > > >> > > > > > > > > > > >> > > > > supporting
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must
> > have
> > > > >> KIP-792).
> > > > >> > > > If
> > > > >> > > > > > the
> > > > >> > > > > > > > > consumer
> > > > >> > > > > > > > > > > >> is
> > > > >> > > > > > > > > > > >> > > > > already
> > > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be
> > > done at
> > > > >> this
> > > > >> > > > point.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers
> > > must
> > > > >> be rolled
> > > > >> > > > > > with the
> > > > >> > > > > > > > > > > >> feature
> > > > >> > > > > > > > > > > >> > flag
> > > > >> > > > > > > > > > > >> > > > > >> turned
> > > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group
> is
> > > > >> automatically
> > > > >> > > > > > > > > converted when
> > > > >> > > > > > > > > > > >> > the first
> > > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> > > protocol
> > > > >> joins
> > > > >> > > > the
> > > > >> > > > > > group.
> > > > >> > > > > > > > > While
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > members
> > > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> > > being
> > > > >> > > > upgraded,
> > > > >> > > > > > the old
> > > > >> > > > > > > > > > > >> protocol
> > > > >> > > > > > > > > > > >> > is
> > > > >> > > > > > > > > > > >> > > > > >> proxied
> > > > >> > > > > > > > > > > >> > > > > >> > > into
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify
> all
> > > of
> > > > >> this in
> > > > >> > > > the
> > > > >> > > > > > KIP.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy
> > > that we
> > > > >> are
> > > > >> > > > > > pushing the
> > > > >> > > > > > > > > pattern
> > > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems
> > like
> > > > >> there
> > > > >> > > > could be
> > > > >> > > > > > some
> > > > >> > > > > > > > > tricky
> > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a
> mechanism
> > > for
> > > > >> users to
> > > > >> > > > > > detect
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> they
> > > > >> > > > > > > > > > > >> > need to
> > > > >> > > > > > > > > > > >> > > > > >> > update
> > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before
> switching
> > > to
> > > > >> the new
> > > > >> > > > > > protocol?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I
> am
> > a
> > > bit
> > > > >> more
> > > > >> > > > > > optimistic
> > > > >> > > > > > > > > than
> > > > >> > > > > > > > > > > >> you
> > > > >> > > > > > > > > > > >> > on this
> > > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > > > >> majority of
> > > > >> > > > the
> > > > >> > > > > > cases
> > > > >> > > > > > > > > are
> > > > >> > > > > > > > > > > >> simple
> > > > >> > > > > > > > > > > >> > > > > regexes
> > > > >> > > > > > > > > > > >> > > > > >> > which
> > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the
> new
> > > > >> engine. The
> > > > >> > > > > > > > > coordinator will
> > > > >> > > > > > > > > > > >> > verify
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> > > consumer
> > > > >> if the
> > > > >> > > > > > regex is
> > > > >> > > > > > > > > not
> > > > >> > > > > > > > > > > >> valid.
> > > > >> > > > > > > > > > > >> > > > > Coming
> > > > >> > > > > > > > > > > >> > > > > >> > back
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path,
> in
> > > the
> > > > >> worst
> > > > >> > > > case,
> > > > >> > > > > > the
> > > > >> > > > > > > > > first
> > > > >> > > > > > > > > > > >> > upgraded
> > > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will
> be
> > > > >> rejected.
> > > > >> > > > This
> > > > >> > > > > > should
> > > > >> > > > > > > > > be used
> > > > >> > > > > > > > > > > >> > as the
> > > > >> > > > > > > > > > > >> > > > > >> last
> > > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > > customers to
> > > > >> > > > validate
> > > > >> > > > > > their
> > > > >> > > > > > > > > regex
> > > > >> > > > > > > > > > > >> > before
> > > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to
> > > test
> > > > >> them with
> > > > >> > > > > > another
> > > > >> > > > > > > > > group.
> > > > >> > > > > > > > > > > >> For
> > > > >> > > > > > > > > > > >> > > > > >> instance,
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done
> in a
> > > > >> pre-prod
> > > > >> > > > > > environment.
> > > > >> > > > > > > > > > > >> Another
> > > > >> > > > > > > > > > > >> > way
> > > > >> > > > > > > > > > > >> > > > > >> would be
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > > >> consumer-group tool
> > > > >> > > > to
> > > > >> > > > > > > > > provide a
> > > > >> > > > > > > > > > > >> regex
> > > > >> > > > > > > > > > > >> > > > > >> validation
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would
> > > this be
> > > > >> > > > enough in
> > > > >> > > > > > your
> > > > >> > > > > > > > > > > >> opinion?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to
> > the
> > > last
> > > > >> > > > question,
> > > > >> > > > > > will
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> Java
> > > > >> > > > > > > > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > > >> allow
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > > > >> current regex
> > > > >> > > > > > engine for
> > > > >> > > > > > > > > > > >> > compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be
> handy
> > to
> > > > >> keep using
> > > > >> > > > > > client
> > > > >> > > > > > > > > based
> > > > >> > > > > > > > > > > >> > regex at
> > > > >> > > > > > > > > > > >> > > > > >> first
> > > > >> > > > > > > > > > > >> > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations
> > simple
> > > and
> > > > >> then
> > > > >> > > > > > migrate to
> > > > >> > > > > > > > > server
> > > > >> > > > > > > > > > > >> > based
> > > > >> > > > > > > > > > > >> > > > > >> regexes
> > > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand
> your
> > > point
> > > > >> but I
> > > > >> > > > am
> > > > >> > > > > > > > > concerned that
> > > > >> > > > > > > > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay
> > in
> > > > >> this mode.
> > > > >> > > > That
> > > > >> > > > > > > > > would go
> > > > >> > > > > > > > > > > >> > against our
> > > > >> > > > > > > > > > > >> > > > > >> goal
> > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> > > client
> > > > >> because
> > > > >> > > > we
> > > > >> > > > > > would
> > > > >> > > > > > > > > have to
> > > > >> > > > > > > > > > > >> > continue
> > > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on
> > the
> > > > >> client
> > > > >> > > > side. I
> > > > >> > > > > > would
> > > > >> > > > > > > > > rather
> > > > >> > > > > > > > > > > >> > not do
> > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say
> > > that
> > > > >> the group
> > > > >> > > > > > > > > coordinator
> > > > >> > > > > > > > > > > >> will
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > configurations
> > > > >> and that
> > > > >> > > > the
> > > > >> > > > > > > > > > > >> configurations
> > > > >> > > > > > > > > > > >> > will be
> > > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> > deleted.
> > > > >> Will a
> > > > >> > > > > > transition to
> > > > >> > > > > > > > > DEAD
> > > > >> > > > > > > > > > > >> > trigger
> > > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > > >> > > > > > > > > > > >> > > > > >> > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> configurations?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right.
> The
> > > > >> > > > configurations
> > > > >> > > > > > will be
> > > > >> > > > > > > > > > > >> deleted
> > > > >> > > > > > > > > > > >> > when
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > > > >> together.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the
> > > choice to
> > > > >> store
> > > > >> > > > the
> > > > >> > > > > > > > > configs in
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > group
> > > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list
> > all
> > > > >> cluster
> > > > >> > > > > > configs and
> > > > >> > > > > > > > > their
> > > > >> > > > > > > > > > > >> > values?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think
> so.
> > > The
> > > > >> group
> > > > >> > > > > > > > > configurations are
> > > > >> > > > > > > > > > > >> > overrides
> > > > >> > > > > > > > > > > >> > > > > >> of
> > > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you
> > > want to
> > > > >> know
> > > > >> > > > all
> > > > >> > > > > > the
> > > > >> > > > > > > > > overrides
> > > > >> > > > > > > > > > > >> > though,
> > > > >> > > > > > > > > > > >> > > > > you
> > > > >> > > > > > > > > > > >> > > > > >> > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the
> > group
> > > > >> > > > coordinators.
> > > > >> > > > > > You
> > > > >> > > > > > > > > cannot
> > > > >> > > > > > > > > > > >> > rely on
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> > > someone
> > > > >> > > > configure a
> > > > >> > > > > > group
> > > > >> > > > > > > > > before
> > > > >> > > > > > > > > > > >> > starting
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > > > >> allowing the
> > > > >> > > > > > explicit
> > > > >> > > > > > > > > creation
> > > > >> > > > > > > > > > > >> of
> > > > >> > > > > > > > > > > >> > > > > groups?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could
> be
> > > > >> decoupled
> > > > >> > > > from
> > > > >> > > > > > the
> > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > >> > lifecycle.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group
> > will
> > > be
> > > > >> > > > > > automatically
> > > > >> > > > > > > > > created in
> > > > >> > > > > > > > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > case.
> > > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will
> > be
> > > > >> lost after
> > > > >> > > > the
> > > > >> > > > > > > > > retention
> > > > >> > > > > > > > > > > >> > period of
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > > >> > > > Consumer.subscribe
> > > > >> > > > > > method
> > > > >> > > > > > > > > for the
> > > > >> > > > > > > > > > > >> > Java
> > > > >> > > > > > > > > > > >> > > > > >> client
> > > > >> > > > > > > > > > > >> > > > > >> > > still
> > > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> `java.util.regex.Pattern` of
> > > > >> > > > do
> > > > >> > > > > > we
> > > > >> > > > > > > > > have to
> > > > >> > > > > > > > > > > >> > introduce
> > > > >> > > > > > > > > > > >> > > > > an
> > > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very
> > group
> > > > >> question. I
> > > > >> > > > > > forgot
> > > > >> > > > > > > > > about
> > > > >> > > > > > > > > > > >> that
> > > > >> > > > > > > > > > > >> > one.
> > > > >> > > > > > > > > > > >> > > > > As
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> `java.util.regex.Pattern` is
> > > > >> > > > not
> > > > >> > > > > > fully
> > > > >> > > > > > > > > > > >> compatible
> > > > >> > > > > > > > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use,
> it
> > > > >> might be
> > > > >> > > > better
> > > > >> > > > > > to
> > > > >> > > > > > > > > deprecate
> > > > >> > > > > > > > > > > >> > it and
> > > > >> > > > > > > > > > > >> > > > > >> use an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a
> > > string.
> > > > >> We would
> > > > >> > > > > > rely on
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> server
> > > > >> > > > > > > > > > > >> > side
> > > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > > migration, I
> > > > >> think
> > > > >> > > > that
> > > > >> > > > > > we
> > > > >> > > > > > > > > could
> > > > >> > > > > > > > > > > >> still
> > > > >> > > > > > > > > > > >> > try to
> > > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use
> it.
> > > That
> > > > >> should
> > > > >> > > > > > work, I
> > > > >> > > > > > > > > think, in
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> majority
> > > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree
> with
> > > > >> Justine
> > > > >> > > > that we
> > > > >> > > > > > > > > should be
> > > > >> > > > > > > > > > > >> > clearer
> > > > >> > > > > > > > > > > >> > > > > about
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > >> > > > IBP/metadata.version
> > > > >> > > > > > from the
> > > > >> > > > > > > > > > > >> feature
> > > > >> > > > > > > > > > > >> > flag.
> > > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > > >> > > > > > > > > > > >> > > > > >> > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > > > >> default for
> > > > >> > > > the
> > > > >> > > > > > feature
> > > > >> > > > > > > > > flag
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > true
> > > > >> > > > > > > > > > > >> > > > > >> based
> > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> metadata.version
> > > once
> > > > >> we
> > > > >> > > > want to
> > > > >> > > > > > make
> > > > >> > > > > > > > > it the
> > > > >> > > > > > > > > > > >> > default.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to
> use
> > > that
> > > > >> feature
> > > > >> > > > > > flag
> > > > >> > > > > > > > > mainly
> > > > >> > > > > > > > > > > >> > during the
> > > > >> > > > > > > > > > > >> > > > > >> > > development
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should
> > not
> > > have
> > > > >> > > > mentioned
> > > > >> > > > > > it, I
> > > > >> > > > > > > > > think,
> > > > >> > > > > > > > > > > >> > because
> > > > >> > > > > > > > > > > >> > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > could
> > > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal
> config
> > > for
> > > > >> it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of
> the
> > > > >> protocol APIs
> > > > >> > > > > > don't
> > > > >> > > > > > > > > mention
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > required
> > > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add
> that
> > > for
> > > > >> > > > consistency.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a
> bit
> > > odd
> > > > >> that
> > > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > >> > > > > > > > > > > >> > > > > requires
> > > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it
> seems
> > > to do
> > > > >> more
> > > > >> > > > than
> > > > >> > > > > > > > > reading.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is
> > > how the
> > > > >> > > > current
> > > > >> > > > > > > > > protocol works
> > > > >> > > > > > > > > > > >> > though.
> > > > >> > > > > > > > > > > >> > > > > We
> > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read
> > > Group" to
> > > > >> join a
> > > > >> > > > > > group. We
> > > > >> > > > > > > > > could
> > > > >> > > > > > > > > > > >> > consider
> > > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not
> sure
> > > that
> > > > >> it is
> > > > >> > > > worth
> > > > >> > > > > > it.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is
> topic
> > > > >> recreation
> > > > >> > > > > > handled by
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > consumer
> > > > >> > > > > > > > > > > >> > > > > with
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It
> > > would be
> > > > >> good to
> > > > >> > > > > > have a
> > > > >> > > > > > > > > section
> > > > >> > > > > > > > > > > >> on
> > > > >> > > > > > > > > > > >> > this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> > > protocol
> > > > >> > > > > > perspective, the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> topic
> > > > >> > > > > > > > > > > >> > will
> > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it
> > will
> > > > >> treat it
> > > > >> > > > like a
> > > > >> > > > > > > > > topic with
> > > > >> > > > > > > > > > > >> a
> > > > >> > > > > > > > > > > >> > > > > different
> > > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is
> that
> > > the
> > > > >> > > > fetch/commit
> > > > >> > > > > > > > > offsets
> > > > >> > > > > > > > > > > >> APIs
> > > > >> > > > > > > > > > > >> > do not
> > > > >> > > > > > > > > > > >> > > > > >> > support
> > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> > consumer
> > > > >> would
> > > > >> > > > reuse the
> > > > >> > > > > > > > > offsets
> > > > >> > > > > > > > > > > >> based
> > > > >> > > > > > > > > > > >> > on the
> > > > >> > > > > > > > > > > >> > > > > >> > same.
> > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should
> > > update
> > > > >> those
> > > > >> > > > APIs
> > > > >> > > > > > as well
> > > > >> > > > > > > > > in
> > > > >> > > > > > > > > > > >> order
> > > > >> > > > > > > > > > > >> > to be
> > > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That
> would
> > > > >> strengthen
> > > > >> > > > the
> > > > >> > > > > > > > > semantics of
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> > > mentions
> > > > >> we will
> > > > >> > > > > > write the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> > coordinator
> > > > >> > > > > > > > > > > >> > > > > >> in
> > > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > > >> implementation
> > > > >> > > > > > detail, do
> > > > >> > > > > > > > > we plan
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > have a
> > > > >> > > > > > > > > > > >> > > > > >> new
> > > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we
> have a
> > > > >> scalability
> > > > >> > > > > > goal when
> > > > >> > > > > > > > > it
> > > > >> > > > > > > > > > > >> comes
> > > > >> > > > > > > > > > > >> > to how
> > > > >> > > > > > > > > > > >> > > > > >> many
> > > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol
> > can
> > > > >> support?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have
> > > numbers
> > > > >> at the
> > > > >> > > > > > moment. The
> > > > >> > > > > > > > > > > >> protocol
> > > > >> > > > > > > > > > > >> > should
> > > > >> > > > > > > > > > > >> > > > > >> > > support
> > > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per
> > > group.
> > > > >> We will
> > > > >> > > > > > measure
> > > > >> > > > > > > > > this when
> > > > >> > > > > > > > > > > >> > we have
> > > > >> > > > > > > > > > > >> > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation.
> > Note
> > > > >> that we
> > > > >> > > > might
> > > > >> > > > > > have
> > > > >> > > > > > > > > other
> > > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > > >> > > > > > > > > > > >> > > > > >> > down
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g.
> offset
> > > > >> commits).
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > > consider
> > > > >> having
> > > > >> > > > > > > > > SubscribedTopidIds
> > > > >> > > > > > > > > > > >> > instead
> > > > >> > > > > > > > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > SubscribedTopicNames
> > > > >> in
> > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > >> > > > > > > > > > > >> > > > > >> Is
> > > > >> > > > > > > > > > > >> > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have
> to
> > > > >> resolve the
> > > > >> > > > > > regex on
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > server, we
> > > > >> > > > > > > > > > > >> > > > > >> can do
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic
> name?
> > > The
> > > > >> > > > difference
> > > > >> > > > > > is that
> > > > >> > > > > > > > > > > >> sending
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> regex is
> > > > >> > > > > > > > > > > >> > > > > >> > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas
> sending
> > > the
> > > > >> topic
> > > > >> > > > names
> > > > >> > > > > > is less
> > > > >> > > > > > > > > > > >> > efficient.
> > > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> > > easier
> > > > >> to
> > > > >> > > > handle
> > > > >> > > > > > if we
> > > > >> > > > > > > > > have
> > > > >> > > > > > > > > > > >> topic
> > > > >> > > > > > > > > > > >> > ids.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > > > >> consolidate the
> > > > >> > > > > > metadata
> > > > >> > > > > > > > > lookup
> > > > >> > > > > > > > > > > >> > on the
> > > > >> > > > > > > > > > > >> > > > > >> server
> > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do
> > agree
> > > > >> with your
> > > > >> > > > > > point. As
> > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > >> second
> > > > >> > > > > > > > > > > >> > > > > though,
> > > > >> > > > > > > > > > > >> > > > > >> > using
> > > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be
> better
> > > here
> > > > >> for the
> > > > >> > > > > > delete and
> > > > >> > > > > > > > > > > >> > recreation
> > > > >> > > > > > > > > > > >> > > > > case.
> > > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we
> > may
> > > > >> allow
> > > > >> > > > users to
> > > > >> > > > > > > > > subscribe
> > > > >> > > > > > > > > > > >> with
> > > > >> > > > > > > > > > > >> > topic
> > > > >> > > > > > > > > > > >> > > > > >> ids
> > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future because
> > that
> > > is
> > > > >> the only
> > > > >> > > > > > way to be
> > > > >> > > > > > > > > > > >> really
> > > > >> > > > > > > > > > > >> > robust
> > > > >> > > > > > > > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12,
> > > 2022 at
> > > > >> 1:38 PM
> > > > >> > > > > > David
> > > > >> > > > > > > > > Jacot <
> > > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for
> your
> > > > >> comments.
> > > > >> > > > Please
> > > > >> > > > > > find
> > > > >> > > > > > > > > my
> > > > >> > > > > > > > > > > >> answers
> > > > >> > > > > > > > > > > >> > > > > below.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > > > >> protocol
> > > > >> > > > relies on
> > > > >> > > > > > > > > topic IDs
> > > > >> > > > > > > > > > > >> > with the
> > > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names
> > based
> > > in
> > > > >> the
> > > > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > >> > > > > > > > > > > >> > > > > >> I
> > > > >> > > > > > > > > > > >> > > > > >> > am
> > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> > > names
> > > > >> is the
> > > > >> > > > > > right call
> > > > >> > > > > > > > > here.
> > > > >> > > > > > > > > > > >> I
> > > > >> > > > > > > > > > > >> > need to
> > > > >> > > > > > > > > > > >> > > > > >> > think
> > > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > > > >> Obviously, the
> > > > >> > > > > > KIP does
> > > > >> > > > > > > > > not
> > > > >> > > > > > > > > > > >> > change the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs
> to
> > > use
> > > > >> topic
> > > > >> > > > IDs.
> > > > >> > > > > > This
> > > > >> > > > > > > > > may be
> > > > >> > > > > > > > > > > >> > something
> > > > >> > > > > > > > > > > >> > > > > >> that
> > > > >> > > > > > > > > > > >> > > > > >> > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though
> > as
> > > it
> > > > >> would
> > > > >> > > > give
> > > > >> > > > > > better
> > > > >> > > > > > > > > > > >> overall
> > > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're
> right.
> > I
> > > > >> think that
> > > > >> > > > I
> > > > >> > > > > > should
> > > > >> > > > > > > > > not
> > > > >> > > > > > > > > > > >> have
> > > > >> > > > > > > > > > > >> > > > > mentioned
> > > > >> > > > > > > > > > > >> > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will
> > remove
> > > > >> it. We can
> > > > >> > > > > > use an
> > > > >> > > > > > > > > internal
> > > > >> > > > > > > > > > > >> > > > > >> configuration
> > > > >> > > > > > > > > > > >> > > > > >> > > while
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > > > >> feature.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> > > types
> > > > >> will be
> > > > >> > > > > > > > > supported. The
> > > > >> > > > > > > > > > > >> > change is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only
> requirement
> > > is
> > > > >> that the
> > > > >> > > > > > cluster
> > > > >> > > > > > > > > uses
> > > > >> > > > > > > > > > > >> topic
> > > > >> > > > > > > > > > > >> > IDs.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul
> 11,
> > > 2022
> > > > >> at 9:53
> > > > >> > > > PM
> > > > >> > > > > > > > > Guozhang
> > > > >> > > > > > > > > > > >> Wang <
> > > > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for
> the
> > > > >> feedback.
> > > > >> > > > Here
> > > > >> > > > > > are
> > > > >> > > > > > > > > some
> > > > >> > > > > > > > > > > >> replies
> > > > >> > > > > > > > > > > >> > > > > inlined
> > > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul
> 9,
> > > 2022
> > > > >> at
> > > > >> > > > 2:53 AM
> > > > >> > > > > > > > > Ismael Juma
> > > > >> > > > > > > > > > > >> <
> > > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for
> > the
> > > > >> KIP. This
> > > > >> > > > has
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> potential
> > > > >> > > > > > > > > > > >> > to be a
> > > > >> > > > > > > > > > > >> > > > > >> > great
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A
> few
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > > >> > > > questions/comments:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think
> > > it's
> > > > >> > > > premature to
> > > > >> > > > > > talk
> > > > >> > > > > > > > > about
> > > > >> > > > > > > > > > > >> > target
> > > > >> > > > > > > > > > > >> > > > > >> versions
> > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of
> > the
> > > > >> existing
> > > > >> > > > group
> > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > > > >> > Unlike
> > > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > > protocol
> > > > >> and hence
> > > > >> > > > > > > > > > > >> > deprecation/removal will
> > > > >> > > > > > > > > > > >> > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how
> quickly
> > > > >> applications
> > > > >> > > > > > migrate
> > > > >> > > > > > > > > to the
> > > > >> > > > > > > > > > > >> new
> > > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree
> > > with
> > > > >> you. I
> > > > >> > > > think
> > > > >> > > > > > we can
> > > > >> > > > > > > > > > > >> remove
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> proposed
> > > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > `Compatibility,
> > > > >> > > > Deprecation,
> > > > >> > > > > > and
> > > > >> > > > > > > > > Migration
> > > > >> > > > > > > > > > > >> > Plan` and
> > > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> > > decide
> > > > >> in the
> > > > >> > > > > > future
> > > > >> > > > > > > > > about
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > we
> > > > >> > > > > > > > > > > >> > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > > > >> behaviors.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP
> > > says we
> > > > >> > > > intend to
> > > > >> > > > > > > > > release this
> > > > >> > > > > > > > > > > >> > in 4.x,
> > > > >> > > > > > > > > > > >> > > > > >> but
> > > > >> > > > > > > > > > > >> > > > > >> > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> > > added
> > > > >> that as
> > > > >> > > > a
> > > > >> > > > > > way to
> > > > >> > > > > > > > > > > >> estimate
> > > > >> > > > > > > > > > > >> > when
> > > > >> > > > > > > > > > > >> > > > > we'd
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > > > >> protocol, I
> > > > >> > > > also
> > > > >> > > > > > suggest
> > > > >> > > > > > > > > > > >> removing
> > > > >> > > > > > > > > > > >> > this
> > > > >> > > > > > > > > > > >> > > > > >> part.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think
> that's
> > > not
> > > > >> > > > specifically
> > > > >> > > > > > > > > related to
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but
> it's
> > > more
> > > > >> for
> > > > >> > > > client
> > > > >> > > > > > > > > upgrades.
> > > > >> > > > > > > > > > > >> I.e.
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> implementation
> > > may
> > > > >> be done
> > > > >> > > > > > first,
> > > > >> > > > > > > > > and then
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > client
> > > > >> > > > > > > > > > > >> > > > > >> > side,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it
> > as
> > > > >> "released"
> > > > >> > > > by
> > > > >> > > > > > the
> > > > >> > > > > > > > > time
> > > > >> > > > > > > > > > > >> clients
> > > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time,
> to
> > > > >> enable the
> > > > >> > > > > > feature the
> > > > >> > > > > > > > > > > >> clients
> > > > >> > > > > > > > > > > >> > need to
> > > > >> > > > > > > > > > > >> > > > > >> > first
> > > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode
> with
> > a
> > > > >> rolling
> > > > >> > > > bounce
> > > > >> > > > > > and
> > > > >> > > > > > > > > then set
> > > > >> > > > > > > > > > > >> > the flag
> > > > >> > > > > > > > > > > >> > > > > >> > with a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and
> > > hence
> > > > >> we feel
> > > > >> > > > it's
> > > > >> > > > > > > > > better to be
> > > > >> > > > > > > > > > > >> > released
> > > > >> > > > > > > > > > > >> > > > > >> in a
> > > > >> > > > > > > > > > > >> > > > > >> > > major
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need
> > to
> > > > >> flesh out
> > > > >> > > > the
> > > > >> > > > > > > > > details of
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > migration
> > > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're
> saying
> > > we
> > > > >> will
> > > > >> > > > support
> > > > >> > > > > > online
> > > > >> > > > > > > > > > > >> > migrations. Is
> > > > >> > > > > > > > > > > >> > > > > >> that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain
> this
> > > in
> > > > >> detail.
> > > > >> > > > It
> > > > >> > > > > > could
> > > > >> > > > > > > > > also be
> > > > >> > > > > > > > > > > >> > done as a
> > > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > > that's
> > > > >> the
> > > > >> > > > part we
> > > > >> > > > > > can
> > > > >> > > > > > > > > be more
> > > > >> > > > > > > > > > > >> > concrete
> > > > >> > > > > > > > > > > >> > > > > >> about
> > > > >> > > > > > > > > > > >> > > > > >> > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is
> > related
> > > to
> > > > >> your
> > > > >> > > > > > question 2)
> > > > >> > > > > > > > > above).
> > > > >> > > > > > > > > > > >> > We will
> > > > >> > > > > > > > > > > >> > > > > >> work
> > > > >> > > > > > > > > > > >> > > > > >> > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > > > >> parallel as we
> > > > >> > > > > > solicit
> > > > >> > > > > > > > > more
> > > > >> > > > > > > > > > > >> > feedback.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am
> > happy
> > > > >> that we are
> > > > >> > > > > > pushing
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > pattern
> > > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it
> seems
> > > like
> > > > >> there
> > > > >> > > > > > could be
> > > > >> > > > > > > > > some
> > > > >> > > > > > > > > > > >> tricky
> > > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > > mechanism
> > > > >> for
> > > > >> > > > users to
> > > > >> > > > > > > > > detect that
> > > > >> > > > > > > > > > > >> > they
> > > > >> > > > > > > > > > > >> > > > > need
> > > > >> > > > > > > > > > > >> > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> > > update
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > > switching
> > > > >> to the
> > > > >> > > > new
> > > > >> > > > > > > > > protocol?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> we
> > > need
> > > > >> some
> > > > >> > > > > > tooling for
> > > > >> > > > > > > > > > > >> non-java
> > > > >> > > > > > > > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > > >> > users
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run"
> the
> > > > >> client before
> > > > >> > > > > > > > > switching to
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > new
> > > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > > >> > > > > > > > > > > >> > > > > >> > > I
> > > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific
> idea
> > on
> > > > >> top of my
> > > > >> > > > head
> > > > >> > > > > > > > > though,
> > > > >> > > > > > > > > > > >> maybe
> > > > >> > > > > > > > > > > >> > others
> > > > >> > > > > > > > > > > >> > > > > >> like
> > > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in
> here?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related
> > to
> > > the
> > > > >> last
> > > > >> > > > > > question,
> > > > >> > > > > > > > > will the
> > > > >> > > > > > > > > > > >> > Java
> > > > >> > > > > > > > > > > >> > > > > >> client
> > > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with
> > the
> > > > >> current
> > > > >> > > > regex
> > > > >> > > > > > > > > engine for
> > > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be
> > > handy
> > > > >> to keep
> > > > >> > > > using
> > > > >> > > > > > > > > client
> > > > >> > > > > > > > > > > >> based
> > > > >> > > > > > > > > > > >> > regex
> > > > >> > > > > > > > > > > >> > > > > at
> > > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> > > simple
> > > > >> and
> > > > >> > > > then
> > > > >> > > > > > migrate
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > server based
> > > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I
> > have
> > > not
> > > > >> thought
> > > > >> > > > > > about
> > > > >> > > > > > > > > that for
> > > > >> > > > > > > > > > > >> > java
> > > > >> > > > > > > > > > > >> > > > > >> clients,
> > > > >> > > > > > > > > > > >> > > > > >> > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What
> > kind
> > > of
> > > > >> > > > > > compatibility
> > > > >> > > > > > > > > issues do
> > > > >> > > > > > > > > > > >> > you have
> > > > >> > > > > > > > > > > >> > > > > in
> > > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we
> > say
> > > > >> that the
> > > > >> > > > group
> > > > >> > > > > > > > > coordinator
> > > > >> > > > > > > > > > > >> > will be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > > >> configurations and
> > > > >> > > > that
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> > configurations
> > > > >> > > > > > > > > > > >> > > > > will
> > > > >> > > > > > > > > > > >> > > > > >> be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> > > deleted.
> > > > >> Will a
> > > > >> > > > > > > > > transition to
> > > > >> > > > > > > > > > > >> DEAD
> > > > >> > > > > > > > > > > >> > > > > trigger
> > > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > configurations?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since
> the
> > > DEAD
> > > > >> state
> > > > >> > > > is an
> > > > >> > > > > > > > > ending
> > > > >> > > > > > > > > > > >> state
> > > > >> > > > > > > > > > > >> > (we
> > > > >> > > > > > > > > > > >> > > > > would
> > > > >> > > > > > > > > > > >> > > > > >> > only
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when
> the
> > > > >> group is
> > > > >> > > > EMPTY
> > > > >> > > > > > and
> > > > >> > > > > > > > > also all
> > > > >> > > > > > > > > > > >> of
> > > > >> > > > > > > > > > > >> > its
> > > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > > >> > > > > > > > > > > >> > > > > >> > > are
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's
> transited
> > > to
> > > > >> DEAD this
> > > > >> > > > > > group
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > >> never
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > > >> revived.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will
> the
> > > > >> choice to
> > > > >> > > > store
> > > > >> > > > > > the
> > > > >> > > > > > > > > configs
> > > > >> > > > > > > > > > > >> in
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to
> > > list all
> > > > >> > > > cluster
> > > > >> > > > > > configs
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > > >> their
> > > > >> > > > > > > > > > > >> > > > > values?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a
> good
> > > > >> question,
> > > > >> > > > and our
> > > > >> > > > > > > > > thoughts
> > > > >> > > > > > > > > > > >> are
> > > > >> > > > > > > > > > > >> > that
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > configurations"
> > > are
> > > > >> > > > overrides
> > > > >> > > > > > of the
> > > > >> > > > > > > > > > > >> > cluster-level
> > > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized
> per
> > > > >> group so
> > > > >> > > > when an
> > > > >> > > > > > > > > admin list
> > > > >> > > > > > > > > > > >> > cluster
> > > > >> > > > > > > > > > > >> > > > > >> > configs
> > > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just
> the
> > > > >> cluster-level
> > > > >> > > > > > > > > "defaults", not
> > > > >> > > > > > > > > > > >> > showing
> > > > >> > > > > > > > > > > >> > > > > >> any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > customizations.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How
> would
> > > > >> someone
> > > > >> > > > > > configure a
> > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > >> > before
> > > > >> > > > > > > > > > > >> > > > > >> starting
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> > considered
> > > > >> allowing
> > > > >> > > > the
> > > > >> > > > > > explicit
> > > > >> > > > > > > > > > > >> > creation of
> > > > >> > > > > > > > > > > >> > > > > >> groups?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively,
> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs
> > could
> > > be
> > > > >> > > > decoupled
> > > > >> > > > > > from
> > > > >> > > > > > > > > the group
> > > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs
> > can
> > > be
> > > > >> created
> > > > >> > > > > > before
> > > > >> > > > > > > > > the group
> > > > >> > > > > > > > > > > >> > itself
> > > > >> > > > > > > > > > > >> > > > > as
> > > > >> > > > > > > > > > > >> > > > > >> an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of
> course,
> > > this
> > > > >> > > > requires
> > > > >> > > > > > the
> > > > >> > > > > > > > > > > >> corresponding
> > > > >> > > > > > > > > > > >> > > > > >> request to
> > > > >> > > > > > > > > > > >> > > > > >> > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > > coordinator
> > > > >> based on
> > > > >> > > > the
> > > > >> > > > > > group
> > > > >> > > > > > > > > id ---
> > > > >> > > > > > > > > > > >> > the only
> > > > >> > > > > > > > > > > >> > > > > >> > thing
> > > > >> > > > > > > > > > > >> > > > > >> > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the
> group
> > > > >> itself is
> > > > >> > > > gone
> > > > >> > > > > > we
> > > > >> > > > > > > > > also check
> > > > >> > > > > > > > > > > >> > if there
> > > > >> > > > > > > > > > > >> > > > > >> are
> > > > >> > > > > > > > > > > >> > > > > >> > > any
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > > related to
> > > > >> that
> > > > >> > > > group
> > > > >> > > > > > and
> > > > >> > > > > > > > > delete
> > > > >> > > > > > > > > > > >> as
> > > > >> > > > > > > > > > > >> > well.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly
> > this
> > > > >> indeed
> > > > >> > > > > > introduces an
> > > > >> > > > > > > > > > > >> > asymmetry on
> > > > >> > > > > > > > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles
> of
> > > the
> > > > >> config
> > > > >> > > > > > entities,
> > > > >> > > > > > > > > and we
> > > > >> > > > > > > > > > > >> > would like
> > > > >> > > > > > > > > > > >> > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > > whether we
> > > > >> should
> > > > >> > > > aim
> > > > >> > > > > > for
> > > > >> > > > > > > > > symmetry
> > > > >> > > > > > > > > > > >> > i.e.
> > > > >> > > > > > > > > > > >> > > > > >> totally
> > > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and
> > > hence
> > > > >> not
> > > > >> > > > delete
> > > > >> > > > > > them at
> > > > >> > > > > > > > > all
> > > > >> > > > > > > > > > > >> when
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> group
> > > > >> > > > > > > > > > > >> > > > > >> > is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> > explicit
> > > > >> deletion
> > > > >> > > > > > operations
> > > > >> > > > > > > > > by
> > > > >> > > > > > > > > > > >> > themselves.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will
> the
> > > > >> > > > > > Consumer.subscribe
> > > > >> > > > > > > > > method for
> > > > >> > > > > > > > > > > >> > the Java
> > > > >> > > > > > > > > > > >> > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > `java.util.regex.Pattern` of
> > > > >> > > > > > do we
> > > > >> > > > > > > > > have
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we
> do
> > > not
> > > > >> need to
> > > > >> > > > > > introduce
> > > > >> > > > > > > > > an
> > > > >> > > > > > > > > > > >> > overload, but
> > > > >> > > > > > > > > > > >> > > > > >> I'm
> > > > >> > > > > > > > > > > >> > > > > >> > > all
> > > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > > > >> compatibility
> > > > >> > > > > > issues
> > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > >> may
> > > > >> > > > > > > > > > > >> > > > > overlook.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I
> agree
> > > with
> > > > >> Justine
> > > > >> > > > > > that we
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > >> be
> > > > >> > > > > > > > > > > >> > clearer
> > > > >> > > > > > > > > > > >> > > > > >> > about
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > >> > > > > > IBP/metadata.version
> > > > >> > > > > > > > > from the
> > > > >> > > > > > > > > > > >> > feature
> > > > >> > > > > > > > > > > >> > > > > >> flag.
> > > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch
> > the
> > > > >> default
> > > > >> > > > for
> > > > >> > > > > > the
> > > > >> > > > > > > > > feature
> > > > >> > > > > > > > > > > >> flag
> > > > >> > > > > > > > > > > >> > to
> > > > >> > > > > > > > > > > >> > > > > true
> > > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > metadata.version
> > > > >> once we
> > > > >> > > > > > want to
> > > > >> > > > > > > > > make it
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> default.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of
> > the
> > > > >> protocol
> > > > >> > > > APIs
> > > > >> > > > > > don't
> > > > >> > > > > > > > > mention
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > >> required
> > > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to
> add
> > > that
> > > > >> for
> > > > >> > > > > > consistency.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > > > >> certainly do
> > > > >> > > > that.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is
> a
> > > bit
> > > > >> odd that
> > > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > >> > > > > > > > > > > >> > > > > >> > requires
> > > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it
> > > seems
> > > > >> to do
> > > > >> > > > more
> > > > >> > > > > > than
> > > > >> > > > > > > > > reading.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> > > thought
> > > > >> myself
> > > > >> > > > as
> > > > >> > > > > > well,
> > > > >> > > > > > > > > but in
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > end we
> > > > >> > > > > > > > > > > >> > > > > >> > could
> > > > >> > > > > > > > > > > >> > > > > >> > > not
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > alternative:
> > > > >> adding
> > > > >> > > > > > Write
> > > > >> > > > > > > > > Group
> > > > >> > > > > > > > > > > >> seems
> > > > >> > > > > > > > > > > >> > an
> > > > >> > > > > > > > > > > >> > > > > >> overkill
> > > > >> > > > > > > > > > > >> > > > > >> > > here
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > > > >> elsewhere (we
> > > > >> > > > only
> > > > >> > > > > > have
> > > > >> > > > > > > > > Read /
> > > > >> > > > > > > > > > > >> > Delete
> > > > >> > > > > > > > > > > >> > > > > and
> > > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would
> > > like to
> > > > >> hear
> > > > >> > > > others
> > > > >> > > > > > > > > thoughts.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> > > topic
> > > > >> > > > recreation
> > > > >> > > > > > > > > handled by
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > consumer
> > > > >> > > > > > > > > > > >> > > > > >> > with
> > > > >> > > > > > > > > > > >> > > > > >> > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol?
> It
> > > > >> would be
> > > > >> > > > good
> > > > >> > > > > > to have
> > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > >> > section on
> > > > >> > > > > > > > > > > >> > > > > >> this.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean
> with
> > > regex
> > > > >> > > > > > subscription
> > > > >> > > > > > > > > right? Yes
> > > > >> > > > > > > > > > > >> > we can
> > > > >> > > > > > > > > > > >> > > > > >> add a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> > > basically
> > > > >> the
> > > > >> > > > idea
> > > > >> > > > > > is that
> > > > >> > > > > > > > > > > >> consumer
> > > > >> > > > > > > > > > > >> > would
> > > > >> > > > > > > > > > > >> > > > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > > protocol as
> > > > >> it's
> > > > >> > > > > > handled all
> > > > >> > > > > > > > > by the
> > > > >> > > > > > > > > > > >> > brokers.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The
> KIP
> > > > >> mentions we
> > > > >> > > > will
> > > > >> > > > > > write
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> new
> > > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > > >> > > > > > > > > > > >> > > > > >> > in
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > > > >> implementation
> > > > >> > > > > > detail,
> > > > >> > > > > > > > > do we
> > > > >> > > > > > > > > > > >> > plan to
> > > > >> > > > > > > > > > > >> > > > > >> have a
> > > > >> > > > > > > > > > > >> > > > > >> > > new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> > > thought
> > > > >> about
> > > > >> > > > > > that. But
> > > > >> > > > > > > > > I think
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > answer
> > > > >> > > > > > > > > > > >> > > > > >> > > should
> > > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we
> > > have a
> > > > >> > > > scalability
> > > > >> > > > > > goal
> > > > >> > > > > > > > > when it
> > > > >> > > > > > > > > > > >> > comes to
> > > > >> > > > > > > > > > > >> > > > > >> how
> > > > >> > > > > > > > > > > >> > > > > >> > > many
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > > protocol can
> > > > >> > > > support?
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> > group,
> > > I
> > > > >> think we
> > > > >> > > > > > should
> > > > >> > > > > > > > > shoot for
> > > > >> > > > > > > > > > > >> > 1000s of
> > > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability
> > > goals
> > > > >> also
> > > > >> > > > depend
> > > > >> > > > > > on the
> > > > >> > > > > > > > > offset
> > > > >> > > > > > > > > > > >> > > > > management
> > > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities
> > of
> > > the
> > > > >> > > > coordinator
> > > > >> > > > > > > > > which we
> > > > >> > > > > > > > > > > >> did
> > > > >> > > > > > > > > > > >> > not
> > > > >> > > > > > > > > > > >> > > > > >> cover in
> > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to
> give a
> > > > >> number that
> > > > >> > > > > > applies
> > > > >> > > > > > > > > > > >> > universally.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > > > >> consider
> > > > >> > > > having
> > > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > > >> > > > > > > > > > > >> > > > > >> instead
> > > > >> > > > > > > > > > > >> > > > > >> > of
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> SubscribedTopicNames in
> > > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we
> > have
> > > to
> > > > >> resolve
> > > > >> > > > the
> > > > >> > > > > > regex
> > > > >> > > > > > > > > on the
> > > > >> > > > > > > > > > > >> > server,
> > > > >> > > > > > > > > > > >> > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > can
> > > > >> > > > > > > > > > > >> > > > > >> > > do
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic
> > > name?
> > > > >> The
> > > > >> > > > > > difference is
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > >> > sending the
> > > > >> > > > > > > > > > > >> > > > > >> > regex
> > > > >> > > > > > > > > > > >> > > > > >> > > is
> > > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas
> > > sending
> > > > >> the topic
> > > > >> > > > > > names is
> > > > >> > > > > > > > > less
> > > > >> > > > > > > > > > > >> > efficient.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation
> > is
> > > > >> easier to
> > > > >> > > > > > handle if
> > > > >> > > > > > > > > we have
> > > > >> > > > > > > > > > > >> > topic
> > > > >> > > > > > > > > > > >> > > > > ids.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main
> > reason
> > > to
> > > > >> still
> > > > >> > > > let
> > > > >> > > > > > the
> > > > >> > > > > > > > > clients
> > > > >> > > > > > > > > > > >> send
> > > > >> > > > > > > > > > > >> > names
> > > > >> > > > > > > > > > > >> > > > > >> is to
> > > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> > > names
> > > > >> -> ids
> > > > >> > > > on
> > > > >> > > > > > the
> > > > >> > > > > > > > > broker /
> > > > >> > > > > > > > > > > >> > admin
> > > > >> > > > > > > > > > > >> > > > > client
> > > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we
> > > added
> > > > >> topic id
> > > > >> > > > in
> > > > >> > > > > > > > > KIP-516, we
> > > > >> > > > > > > > > > > >> > never
> > > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > consumer/producers
> > > > >> > > > leveraging
> > > > >> > > > > > the
> > > > >> > > > > > > > > related
> > > > >> > > > > > > > > > > >> > newer
> > > > >> > > > > > > > > > > >> > > > > >> versioned
> > > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> > > topic
> > > > >> id as
> > > > >> > > > empty
> > > > >> > > > > > UUID.
> > > > >> > > > > > > > > We
> > > > >> > > > > > > > > > > >> want
> > > > >> > > > > > > > > > > >> > to keep
> > > > >> > > > > > > > > > > >> > > > > >> the
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> consumer/producer
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin
> and
> > > only
> > > > >> > > > delegate
> > > > >> > > > > > the
> > > > >> > > > > > > > > reasoning
> > > > >> > > > > > > > > > > >> on
> > > > >> > > > > > > > > > > >> > broker
> > > > >> > > > > > > > > > > >> > > > > >> and
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed,
> Jul
> > 6,
> > > > >> 2022 at
> > > > >> > > > 10:45
> > > > >> > > > > > AM
> > > > >> > > > > > > > > David
> > > > >> > > > > > > > > > > >> Jacot
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would
> > > like to
> > > > >> start a
> > > > >> > > > > > > > > discussion
> > > > >> > > > > > > > > > > >> thread
> > > > >> > > > > > > > > > > >> > on
> > > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > >> > > > > > > > > > > >> > > > > >> > > The
> > > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> Generation
> > > of
> > > > >> the
> > > > >> > > > Consumer
> > > > >> > > > > > > > > Rebalance
> > > > >> > > > > > > > > > > >> > Protocol.
> > > > >> > > > > > > > > > > >> > > > > >> With
> > > > >> > > > > > > > > > > >> > > > > >> > > this
> > > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make
> > the
> > > > >> rebalance
> > > > >> > > > > > protocol
> > > > >> > > > > > > > > (for
> > > > >> > > > > > > > > > > >> > consumers)
> > > > >> > > > > > > > > > > >> > > > > >> more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> scalable,
> > > > >> easier to
> > > > >> > > > > > implement for
> > > > >> > > > > > > > > > > >> > clients, and
> > > > >> > > > > > > > > > > >> > > > > >> easier
> > > > >> > > > > > > > > > > >> > > > > >> > > to
> > > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> operators.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP
> is
> > > here:
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please
> > take
> > > a
> > > > >> look and
> > > > >> > > > let
> > > > >> > > > > > me
> > > > >> > > > > > > > > know what
> > > > >> > > > > > > > > > > >> > you
> > > > >> > > > > > > > > > > >> > > > > think.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I
> will
> > > be
> > > > >> away from
> > > > >> > > > > > July
> > > > >> > > > > > > > > 18th to
> > > > >> > > > > > > > > > > >> > August 8th.
> > > > >> > > > > > > > > > > >> > > > > >> That
> > > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time
> to
> > > read
> > > > >> and
> > > > >> > > > digest
> > > > >> > > > > > this
> > > > >> > > > > > > > > long
> > > > >> > > > > > > > > > > >> KIP.
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> > >
> > > > >> > > > > > > > > > > >> > > > > >> >
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> --
> > > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > --
> > > > >> > > > > -- Guozhang
> > > > >> > > >
> > > > >> > >
> > > > >> > >
> > > > >> > > --
> > > > >> > > -- Guozhang
> > > > >>
> > > > >
> > > > >
> > > > > --
> > > > > -- Guozhang
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
> >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Jason Gustafson <ja...@confluent.io.INVALID>]

Hey David,

Thanks for the response.

> Do you think that we could have a case where two members with the same
static id are active?

I was just thinking about the handoff when we spin up a new node before the
old one completes shutdown. It's probably ok to fence the old member
abruptly. I do recall we had a few issues with this handoff in the past. It
was due to the fact that the replaced node might have been the leader.
Since we've removed the leader notion, perhaps this kind of issue won't
resurface.

In regard to the RE2 library. I understand the intent to pass this through
as a string without any validation so that we don't need the dependence on
the client. I do wonder if that will be restrictive in practice. For
example, if a regex changes on the client side, would we need to validate
against the current assignment to see which partitions to continue
fetching? Or would we instead invalidate the complete current assignment?
Perhaps this is a rare case and we can do the latter. There could be more
situations though.

As an alternative, I wonder if we could define a minimalistic regex syntax
(a subset of RE2 say) that could map more easily to regex support that
already exists in clients? For example, in this scenario, we are matching
topics, so we only need to be able to represent topic characters and not
the whole smorgasbord of whitespace and unicode options. Would it make
sense to have a reduced syntax which includes only standard topic
characters, wildcards, character classes, and repetitions?

Best,
Jason




On Thu, Sep 22, 2022 at 9:02 AM Guozhang Wang <wa...@gmail.com> wrote:

> Hi David, thanks for all the detailed explanations. I think they all make
> sense. Just want to have a couple follow-ups here:
>
> > I don't really see the benefits here because server side assignors
> don't have metadata at all. They only assign topic-partitions. They
> are not supposed to generate metadata nor to receive metadata from the
> members.
>
> Just to clarify I was asking about the `version` of the assignor (i.e. up
> to what version that the client would support), and I do agree we would not
> need metadata. What I have in mind is that, for some specific built-in
> broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> newer version we would have a hierarchical rack ID string format, like
> "tier1-tier2" etc, but if some client has not upgraded their rack ID
> would still be in old format. In this case, the broker then needs to choose
> the old versioned assignor. I'm probably making something up here for rack
> aware assignors, but I'm wondering if in general such an "auto-downgrade"
> behavior would be needed still for broker-side assignor, and if yes would
> "version" still be useful.
>
> > Yeah, that's right. Within a rebalance, `onAssignment` is called once
> when the member transitions to a new epoch. This one contains the full
> metadata provided by the client side assignor. Then, `onAssignment`
> can be called max N times where N is the number of partitions pending
> revocation by other members. Let me try to clarify this in the KIP.
>
> Okay, my understanding is that the calling ordering of these callbacks
> would be like the following:
>
> ----------------------------------------
> onPartitionsRevoked();   // just once, since we do not really need
> to revoke incrementally.
>
> onAssignment();    // the first call, with epoch incremented
> onPartitionsAssigned();   // paired with the onAssignment
>
> onAssignment();              // the first onAssignment would bump up the
> epoch, and the metadata reflected.
> onPartitionsAssigned();   // each time we get an additional assignment, we
> call onAssignment and then paired with an onPartitionsAssigned
> ...
> onAssignment();
> onPartitionsAssigned();   // on each of the onAssignment calls, the encoded
> metadata would not change, only the incrementally added partitions be
> reflected
>
> Is that the case?
>
> I'm wondering if we would still call onAssignment just once, that encodes
> all the assignment for this rebalance, including all the partitions that
> should be assigned to the member but not yet assigned since they have not
> been revoked by others. In that case the call ordering would be:
>
> ----------------------------------------
> onPartitionsRevoked();   // just once
> onAssignment();    // just once, with epoch incremented, and metadata
> encoded changed, the "assignment" field also reflect the final target
> assignment
> onPartitionsAssigned();   // multiple times, which represent incrementally
> added partitions
> ...
> onPartitionsAssigned();
>
> The motivation from this is that, most users would only implement the
> rebalance callback listeners and hence we'd definitely need to make sure
> the semantics of that does not change much, and the time
> `onPartitionsAssigned` indicate the time when the partitions are actually
> assigned to it; while for assignors, the `onAssignment` is used to indicate
> what decision is made regarding for this member, i.e. when the partitions
> are decided to be given to it, but not necessarily meant that it has been
> given, since that time should be determined by the time of
> `onPartitionsAssigned`. The benefits though, would be that assignor
> implementers would not need to reason which `onAssignment` would be the
> last one for this epoch.
>
>
> Guozhang
>
>
>
>
> Guozhang
>
> On Thu, Sep 22, 2022 at 2:20 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Guozhang,
> >
> > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > field,
> > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > rationale seems to be assuming that we should always be able to do the
> > > assignment at the broker-side assignor without errors. Personally I
> think
> > > it's still potentially beneficial to add the Error field even for
> > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > topics are not recognized with the current broker's metadata. What do
> you
> > > think?
> >
> > Yeah, that seems reasonable. However, I wonder if it would be better
> > to use an exception on the server side. This is what we usually do for
> > server side plugins. On the client side, we use a field because the
> > errors are not defined in advance.
> >
> > Your comment also makes me think about what we should do when the
> > server side assignor fails. I suppose that we have to keep the current
> > assignment until a new event occurs. For instance, in your example,
> > the coordinator would have to trigger a rebalance when unrecognized
> > topics are available. This would be part of the metadata monitoring.
> >
> > > 2) The client-side "GroupMember" has three additional fields
> > > reason/version/metadata compared with the broker-side GroupMember. I
> > agree
> > > that broker-side assignor would not need reason/metadata since they are
> > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > it
> > > possible that we evolve our broker-side built-in assignor but the old
> > > versioned clients would not be able to work with the new version, in
> > which
> > > case we need to let the broker being aware of this and upgrade its
> > behavior
> > > to cooperate with the clients?
> >
> > I don't really see the benefits here because server side assignors
> > don't have metadata at all. They only assign topic-partitions. They
> > are not supposed to generate metadata nor to receive metadata from the
> > members.
> >
> > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > of
> > > including these three fields, what about just adding the "Metadata"
> field
> > > class which has these three fields? Also, there are two "Metadata"
> > > currently in the APIs, the first is a class that encodes
> > > reason/version/metadata, and the second is just the encoded metadata
> > bytes.
> > > I'm wondering what about just naming the first as memberMetadata, which
> > > then has a bytebuffer field Metadata, or instead naming the second
> > > bytebuffer field as metadataBytes?
> >
> > That's a good point. Let me try to rationalize this interface based on
> > your suggestions.
> >
> > Best,
> > David
> >
> > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> > >
> > > Hello David,
> > >
> > > Just had a few more nit questions about the public APIs:
> > >
> > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > field,
> > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > rationale seems to be assuming that we should always be able to do the
> > > assignment at the broker-side assignor without errors. Personally I
> think
> > > it's still potentially beneficial to add the Error field even for
> > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > topics are not recognized with the current broker's metadata. What do
> you
> > > think?
> > >
> > > 2) The client-side "GroupMember" has three additional fields
> > > reason/version/metadata compared with the broker-side GroupMember. I
> > agree
> > > that broker-side assignor would not need reason/metadata since they are
> > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > it
> > > possible that we evolve our broker-side built-in assignor but the old
> > > versioned clients would not be able to work with the new version, in
> > which
> > > case we need to let the broker being aware of this and upgrade its
> > behavior
> > > to cooperate with the clients?
> > >
> > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > of
> > > including these three fields, what about just adding the "Metadata"
> field
> > > class which has these three fields? Also, there are two "Metadata"
> > > currently in the APIs, the first is a class that encodes
> > > reason/version/metadata, and the second is just the encoded metadata
> > bytes.
> > > I'm wondering what about just naming the first as memberMetadata, which
> > > then has a bytebuffer field Metadata, or instead naming the second
> > > bytebuffer field as metadataBytes?
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > > > Hello David,
> > > >
> > > > Thanks for bringing this question up. I think the main benefits as
> you
> > > > listed is 2) above if it stays; just to clarify, we would only be
> able
> > to
> > > > save one round trip if the rebalance is still triggered by the
> broker;
> > if
> > > > the rebalance is triggered by the client then the num.round.trips are
> > the
> > > > same:
> > > >
> > > > 1) With GroupPrepareAssignment:
> > > >
> > > > T0: client decides to do a new assignment, suppose it has already
> sent
> > a
> > > > HB and hence has to wait for it to return first since only one
> request
> > /
> > > > response can be inflight with the coordinator's socket.
> > > > T1: client receives the HB response, and then sends the
> > > > GroupPrepareAssignment request.
> > > > T2: the GroupPrepareAssignment response is returned.
> > > > T3: it calculates the new assignment, and sends a
> > GroupInstallAssignment
> > > > request.
> > > >
> > > > In total, two round trips.
> > > >
> > > > 2) Without GroupPrepareAssignment:
> > > >
> > > > T0: client decides to do a new assignment, suppose it has already
> sent
> > a
> > > > HB and hence has to wait for it to return first since only one
> request
> > /
> > > > response can be inflight with the coordinator's socket.
> > > > T1: client receives the HB response, and then sends the new HB
> request
> > > > with the flag indicating a new rebalance needed..
> > > > T2: the HB response with the optional member metadata map is
> returned.
> > > > T3: it calculates the new assignment, and sends a
> > GroupInstallAssignment
> > > > request.
> > > >
> > > > In total, two round trips as well.
> > > >
> > > > -----------------------------
> > > >
> > > > So to complete the full picture here, we'd need to modify both HB
> > request
> > > > and response so that the client can also indicate a new rebalance via
> > the
> > > > HB request as well, right?
> > > >
> > > > Assuming all above is true, I think it's okay to merge the
> > > > GroupPrepareAssignment into HB given that we can make the additional
> > fields
> > > > encoding the full member (subscription) metadata and topic metadata
> as
> > > > optional fields.
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > >> Hi all,
> > > >>
> > > >> During an offline conversation, someone asked why we need the
> > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> instead
> > > >> provide the group state in the heartbeat response. This has a few
> > > >> advantages: 1) it does not require using a special error code to
> > > >> signal that a new assignment is required as the signal would be the
> > > >> provided group state; 2) it removes one round trip when a client
> side
> > > >> assignor is used. The downside is that it makes the heartbeat
> > > >> response's definition quite large. I recall that I went with the
> > > >> current approach due to this.
> > > >>
> > > >> Providing the group state in the heartbeat response is appealing.
> What
> > > >> do you guys think?
> > > >>
> > > >> Best,
> > > >> David
> > > >>
> > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > wrote:
> > > >> >
> > > >> > Hi Guozhang,
> > > >> >
> > > >> > 1. I have added a reference to the relevant chapter instead of
> > > >> > repeating the whole thing. Does that work for you?
> > > >> >
> > > >> > 2. The "Rebalance Triggers" section you are referring to is about
> > when
> > > >> > a rebalance should be triggered for the non-upgraded members using
> > the
> > > >> > old protocol. The section mentions that a rebalance must be
> > triggered
> > > >> > when a new assignment is installed. This implies that the group
> > epoch
> > > >> > was updated either by a native member or a non-upgraded member.
> For
> > > >> > the latter, the JoinGroup request would be the trigger. I have
> > added a
> > > >> > reference to the relevant chapter in the "JoinGroup Handling"
> > section
> > > >> > as well. Does that make sense?
> > > >> >
> > > >> > Thanks,
> > > >> > David
> > > >> >
> > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wangguoz@gmail.com
> >
> > > >> wrote:
> > > >> > >
> > > >> > > Hello David,
> > > >> > >
> > > >> > > Alright I think that's sufficient. Just to make that clear in
> the
> > doc,
> > > >> > > could we update:
> > > >> > >
> > > >> > > 1) the heartbeat request handling section, stating when
> > coordinator
> > > >> will
> > > >> > > trigger rebalance based on the HB's member metadata / reason?
> > > >> > > 2) the "Rebalance Triggers" section to include what we described
> > in
> > > >> "Group
> > > >> > > Epoch - Trigger a rebalance" section as well?
> > > >> > >
> > > >> > >
> > > >> > > Guozhang
> > > >> > >
> > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > >> <dj...@confluent.io.invalid>
> > > >> > > wrote:
> > > >> > >
> > > >> > > > Hi Guozhang,
> > > >> > > >
> > > >> > > > I thought that the assignor will always be consulted when the
> > next
> > > >> > > > heartbeat request is constructed. In other words,
> > > >> > > > `PartitionAssignor#metadata` will be called for every
> heartbeat.
> > > >> This
> > > >> > > > gives the opportunity for the assignor to enforce a rebalance
> by
> > > >> > > > setting the reason to a non-zero value or by changing the
> > bytes. Do
> > > >> > > > you think that this is not sufficient? Are you concerned by
> the
> > > >> delay?
> > > >> > > >
> > > >> > > > Best,
> > > >> > > > David
> > > >> > > >
> > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > wangguoz@gmail.com>
> > > >> wrote:
> > > >> > > > >
> > > >> > > > > Hello David,
> > > >> > > > >
> > > >> > > > > One of Jun's comments make me thinking:
> > > >> > > > >
> > > >> > > > > ```
> > > >> > > > > In this case, a new assignment is triggered by the client
> side
> > > >> > > > > assignor. When constructing the HB, the consumer will always
> > > >> consult
> > > >> > > > > the client side assignor and propagate the information to
> the
> > > >> group
> > > >> > > > > coordinator. In other words, we don't expect users to call
> > > >> > > > > Consumer#enforceRebalance anymore.
> > > >> > > > > ```
> > > >> > > > >
> > > >> > > > > As I looked at the current PartitionAssignor's interface, we
> > > >> actually do
> > > >> > > > > not have a way yet to instruct how to construct the next HB
> > > >> request, e.g.
> > > >> > > > > when the assignor wants to enforce a new rebalance with a
> new
> > > >> assignment,
> > > >> > > > > we'd need some customizable APIs inside the
> PartitionAssignor
> > to
> > > >> indicate
> > > >> > > > > the next HB telling broker about so. WDYT about adding such
> an
> > > >> API on the
> > > >> > > > > PartitionAssignor?
> > > >> > > > >
> > > >> > > > >
> > > >> > > > > Guozhang
> > > >> > > > >
> > > >> > > > >
> > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > >> <dj...@confluent.io.invalid>
> > > >> > > > > wrote:
> > > >> > > > >
> > > >> > > > > > Hi Jun,
> > > >> > > > > >
> > > >> > > > > > I have updated the KIP to include your feedback. I have
> also
> > > >> tried to
> > > >> > > > > > clarify the parts which were not cleared.
> > > >> > > > > >
> > > >> > > > > > Best,
> > > >> > > > > > David
> > > >> > > > > >
> > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > djacot@confluent.io
> > > >> >
> > > >> > > > wrote:
> > > >> > > > > > >
> > > >> > > > > > > Hi Jun,
> > > >> > > > > > >
> > > >> > > > > > > Thanks for your feedback. Let me start by answering your
> > > >> questions
> > > >> > > > > > > inline and I will update the KIP next week.
> > > >> > > > > > >
> > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > KIP
> > > >> seem to
> > > >> > > > be
> > > >> > > > > > fewer
> > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > >> wildcard. A few
> > > >> > > > > > > > comments below.
> > > >> > > > > > >
> > > >> > > > > > > I would also add that the KIP removes the global sync
> > barrier
> > > >> in the
> > > >> > > > > > > protocol which is essential to improve group stability
> and
> > > >> > > > > > > scalability, and the KIP also simplifies the client by
> > moving
> > > >> most of
> > > >> > > > > > > the logic to the server side.
> > > >> > > > > > >
> > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > support
> > > >> rolling
> > > >> > > > > > changing
> > > >> > > > > > > > of the partition assignor in the consumers?
> > > >> > > > > > >
> > > >> > > > > > > Definitely. The group coordinator will use the assignor
> > used
> > > >> by a
> > > >> > > > > > > majority of the members. This allows the group to move
> > from
> > > >> one
> > > >> > > > > > > assignor to another by a roll. This is explained in the
> > > >> Assignor
> > > >> > > > > > > Selection chapter.
> > > >> > > > > > >
> > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > >> required in
> > > >> > > > every
> > > >> > > > > > > > request or the scenarios when it needs to be filled?
> For
> > > >> example,
> > > >> > > > it's
> > > >> > > > > > not
> > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > >> > > > > > >
> > > >> > > > > > > The client is expected to set those fields in case of a
> > > >> connection
> > > >> > > > > > > issue (e.g. timeout) or when the fields have changed
> since
> > > >> the last
> > > >> > > > > > > HB. The server populates those fields as long as the
> > member
> > > >> is not
> > > >> > > > > > > fully reconciled - the member should acknowledge that it
> > has
> > > >> the
> > > >> > > > > > > expected epoch and assignment. I will clarify this in
> the
> > KIP.
> > > >> > > > > > >
> > > >> > > > > > > > 31. In the current consumer protocol, the rack
> affinity
> > > >> between the
> > > >> > > > > > client
> > > >> > > > > > > > and the broker is only considered during fetching, but
> > not
> > > >> during
> > > >> > > > > > assigning
> > > >> > > > > > > > partitions to consumers. Sometimes, once the
> assignment
> > is
> > > >> made,
> > > >> > > > there
> > > >> > > > > > is
> > > >> > > > > > > > no opportunity for read affinity because no replicas
> of
> > > >> assigned
> > > >> > > > > > partitions
> > > >> > > > > > > > are close to the member. I am wondering if we should
> use
> > > >> this
> > > >> > > > > > opportunity
> > > >> > > > > > > > to address this by including rack in GroupMember.
> > > >> > > > > > >
> > > >> > > > > > > That's an interesting idea. I don't see any issue with
> > adding
> > > >> the
> > > >> > > > rack
> > > >> > > > > > > to the members. I will do so.
> > > >> > > > > > >
> > > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> > busy
> > > >> a group
> > > >> > > > > > > > coordinator is. By moving the event loop model, it
> seems
> > > >> that we
> > > >> > > > could
> > > >> > > > > > add
> > > >> > > > > > > > a metric that tracks the fraction of the time the
> event
> > > >> loop is
> > > >> > > > doing
> > > >> > > > > > the
> > > >> > > > > > > > actual work.
> > > >> > > > > > >
> > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > >> > > > > > >
> > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > >> handling? For
> > > >> > > > > > example,
> > > >> > > > > > > > does it need to trigger the check if any group with
> the
> > > >> wildcard
> > > >> > > > > > > > subscription now has a new matching topic?
> > > >> > > > > > >
> > > >> > > > > > > Sure. When the new group coordinator takes over, it has
> > to:
> > > >> > > > > > > * Setup the session timeouts.
> > > >> > > > > > > * Trigger a new assignment if a client side assignor is
> > used.
> > > >> We
> > > >> > > > don't
> > > >> > > > > > > store the information about the member selected to run
> the
> > > >> assignment
> > > >> > > > > > > so we have to start a new one.
> > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > >> subscriptions, and
> > > >> > > > > > > trigger a rebalance if needed.
> > > >> > > > > > >
> > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> what
> > > >> the epoch
> > > >> > > > > > field
> > > >> > > > > > > > reflects? For example, does the epoch in
> > > >> ConsumerGroupMetadataValue
> > > >> > > > > > reflect
> > > >> > > > > > > > the latest group epoch? What about the one in
> > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > >> > > > > > >
> > > >> > > > > > > Sure. I will clarify that but it is always the latest
> > group
> > > >> epoch.
> > > >> > > > > > > When the group state is updated, the group epoch is
> > bumped so
> > > >> we use
> > > >> > > > > > > that one for all the change records related to the
> update.
> > > >> > > > > > >
> > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > following
> > > >> > > > invariants
> > > >> > > > > > are
> > > >> > > > > > > > met: ... All members exists." It's possible for a
> member
> > > >> not to
> > > >> > > > get any
> > > >> > > > > > > > assigned partitions, right?
> > > >> > > > > > >
> > > >> > > > > > > That's right. Here I meant that the members provided by
> > the
> > > >> assignor
> > > >> > > > > > > in the assignment must exist in the group. The assignor
> > can
> > > >> not make
> > > >> > > > > > > up new member ids.
> > > >> > > > > > >
> > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > equals to
> > > >> 0":
> > > >> > > > When
> > > >> > > > > > would
> > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > >> > > > > > >
> > > >> > > > > > > A member is expected to abandon all its partitions and
> > > >> rejoins when
> > > >> > > > it
> > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case,
> the
> > > >> group
> > > >> > > > > > > coordinator will have removed the member from the group.
> > The
> > > >> member
> > > >> > > > > > > can rejoin the group with the same member id but with 0
> as
> > > >> epoch. Let
> > > >> > > > > > > me see if I can clarify this in the KIP.
> > > >> > > > > > >
> > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger a
> > > >> > > > > > reassignment
> > > >> > > > > > > > by either providing a non-zero reason or by updating
> the
> > > >> assignor
> > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > >> deprecation
> > > >> > > > of
> > > >> > > > > > > > Consumer#enforeRebalance.
> > > >> > > > > > >
> > > >> > > > > > > In this case, a new assignment is triggered by the
> client
> > side
> > > >> > > > > > > assignor. When constructing the HB, the consumer will
> > always
> > > >> consult
> > > >> > > > > > > the client side assignor and propagate the information
> to
> > the
> > > >> group
> > > >> > > > > > > coordinator. In other words, we don't expect users to
> call
> > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > >> > > > > > >
> > > >> > > > > > > > 38. The reassignment examples are nice. But the
> section
> > > >> seems to
> > > >> > > > have
> > > >> > > > > > > > multiple typos.
> > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > immediately
> > > >> gets into
> > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > > >> into
> > > >> > > > > > epoch=3,
> > > >> > > > > > > > partitions=[foo-1] too early.
> > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > >> epoch=2,
> > > >> > > > > > > > partitions=[foo-0].
> > > >> > > > > > >
> > > >> > > > > > > Sorry for that! I will revise them.
> > > >> > > > > > >
> > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > upgrade
> > > >> from
> > > >> > > > any
> > > >> > > > > > old
> > > >> > > > > > > > version to new one?
> > > >> > > > > > >
> > > >> > > > > > > We will support upgrading from the consumer protocol
> > version
> > > >> 3,
> > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> > the
> > > >> earliest
> > > >> > > > > > > version is unknown at the moment. This is explained in
> the
> > > >> migration
> > > >> > > > > > > plan chapter.
> > > >> > > > > > >
> > > >> > > > > > > Thanks again for your feedback, Jun. I will update the
> KIP
> > > >> based on
> > > >> > > > it
> > > >> > > > > > > next week.
> > > >> > > > > > >
> > > >> > > > > > > Best,
> > > >> > > > > > > David
> > > >> > > > > > >
> > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > >> <ju...@confluent.io.invalid>
> > > >> > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > Hi, David,
> > > >> > > > > > > >
> > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > KIP
> > > >> seem to
> > > >> > > > be
> > > >> > > > > > fewer
> > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > >> wildcard. A few
> > > >> > > > > > > > comments below.
> > > >> > > > > > > >
> > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > support
> > > >> rolling
> > > >> > > > > > changing
> > > >> > > > > > > > of the partition assignor in the consumers?
> > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > >> required in
> > > >> > > > every
> > > >> > > > > > > > request or the scenarios when it needs to be filled?
> For
> > > >> example,
> > > >> > > > it's
> > > >> > > > > > not
> > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > >> > > > > > > >
> > > >> > > > > > > > 31. In the current consumer protocol, the rack
> affinity
> > > >> between the
> > > >> > > > > > client
> > > >> > > > > > > > and the broker is only considered during fetching, but
> > not
> > > >> during
> > > >> > > > > > assigning
> > > >> > > > > > > > partitions to consumers. Sometimes, once the
> assignment
> > is
> > > >> made,
> > > >> > > > there
> > > >> > > > > > is
> > > >> > > > > > > > no opportunity for read affinity because no replicas
> of
> > > >> assigned
> > > >> > > > > > partitions
> > > >> > > > > > > > are close to the member. I am wondering if we should
> use
> > > >> this
> > > >> > > > > > opportunity
> > > >> > > > > > > > to address this by including rack in GroupMember.
> > > >> > > > > > > >
> > > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> > busy
> > > >> a group
> > > >> > > > > > > > coordinator is. By moving the event loop model, it
> seems
> > > >> that we
> > > >> > > > could
> > > >> > > > > > add
> > > >> > > > > > > > a metric that tracks the fraction of the time the
> event
> > > >> loop is
> > > >> > > > doing
> > > >> > > > > > the
> > > >> > > > > > > > actual work.
> > > >> > > > > > > >
> > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > >> handling? For
> > > >> > > > > > example,
> > > >> > > > > > > > does it need to trigger the check if any group with
> the
> > > >> wildcard
> > > >> > > > > > > > subscription now has a new matching topic?
> > > >> > > > > > > >
> > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> what
> > > >> the epoch
> > > >> > > > > > field
> > > >> > > > > > > > reflects? For example, does the epoch in
> > > >> ConsumerGroupMetadataValue
> > > >> > > > > > reflect
> > > >> > > > > > > > the latest group epoch? What about the one in
> > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > >> > > > > > > >
> > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > following
> > > >> > > > invariants
> > > >> > > > > > are
> > > >> > > > > > > > met: ... All members exists." It's possible for a
> member
> > > >> not to
> > > >> > > > get any
> > > >> > > > > > > > assigned partitions, right?
> > > >> > > > > > > >
> > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > equals to
> > > >> 0":
> > > >> > > > When
> > > >> > > > > > would
> > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > >> > > > > > > >
> > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger a
> > > >> > > > > > reassignment
> > > >> > > > > > > > by either providing a non-zero reason or by updating
> the
> > > >> assignor
> > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > >> deprecation
> > > >> > > > of
> > > >> > > > > > > > Consumer#enforeRebalance.
> > > >> > > > > > > >
> > > >> > > > > > > > 38. The reassignment examples are nice. But the
> section
> > > >> seems to
> > > >> > > > have
> > > >> > > > > > > > multiple typos.
> > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > immediately
> > > >> gets into
> > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > > >> into
> > > >> > > > > > epoch=3,
> > > >> > > > > > > > partitions=[foo-1] too early.
> > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > >> epoch=2,
> > > >> > > > > > > > partitions=[foo-0].
> > > >> > > > > > > >
> > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > upgrade
> > > >> from
> > > >> > > > any
> > > >> > > > > > old
> > > >> > > > > > > > version to new one?
> > > >> > > > > > > >
> > > >> > > > > > > > Thanks,
> > > >> > > > > > > >
> > > >> > > > > > > > Jun
> > > >> > > > > > > >
> > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > >> > > > > > <dj...@confluent.io.invalid>
> > > >> > > > > > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > > Hi all,
> > > >> > > > > > > > >
> > > >> > > > > > > > > The KIP states that we will re-implement the
> > coordinator
> > > >> in
> > > >> > > > Java. I
> > > >> > > > > > > > > discussed this offline with a few folks and folks
> are
> > > >> concerned
> > > >> > > > that
> > > >> > > > > > > > > we could introduce many regressions in the old
> > protocol
> > > >> if we do
> > > >> > > > so.
> > > >> > > > > > > > > Therefore, I am going to remove this statement from
> > the
> > > >> KIP. It
> > > >> > > > is an
> > > >> > > > > > > > > implementation detail after all so it does not have
> > to be
> > > >> > > > decided at
> > > >> > > > > > > > > this stage. We will likely start by trying to
> > refactor the
> > > >> > > > current
> > > >> > > > > > > > > implementation as a first step.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Cheers,
> > > >> > > > > > > > > David
> > > >> > > > > > > > >
> > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > >> djacot@confluent.io
> > > >> > > > >
> > > >> > > > > > wrote:
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Hi Luke,
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > assigning,
> > > >> > > > > > reconciling,
> > > >> > > > > > > > > stable,
> > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> section,
> > > >> right?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > This sentence does not refer to those group states
> > but
> > > >> rather
> > > >> > > > to a
> > > >> > > > > > > > > > state machine replication (SMR). This refers to
> the
> > > >> entire
> > > >> > > > state of
> > > >> > > > > > > > > > group coordinator which is replicated via the log
> > > >> layer. I will
> > > >> > > > > > > > > > clarify this in the KIP.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > modelled
> > > >> as an
> > > >> > > > event
> > > >> > > > > > > > > loop"?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > The idea is to follow a model similar to the new
> > quorum
> > > >> > > > > > controller. We
> > > >> > > > > > > > > > will have N threads to process events. Each
> > > >> __consumer_offsets
> > > >> > > > > > > > > > partition is assigned to a unique thread and all
> the
> > > >> events
> > > >> > > > (e.g.
> > > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > >> thread. This
> > > >> > > > > > simplify
> > > >> > > > > > > > > > concurrency and will enable us to do simulation
> > testing
> > > >> for the
> > > >> > > > > > group
> > > >> > > > > > > > > > coordinator.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > >> *__consumer_offsets*
> > > >> > > > > > > > > partitions?
> > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> by
> > a
> > > >> group
> > > >> > > > > > coordinator?
> > > >> > > > > > > > > For
> > > >> > > > > > > > > > > example, if one group coordinator owns 2
> consumer
> > > >> groups, and
> > > >> > > > > > both
> > > >> > > > > > > > > exist in
> > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > > >> for it,
> > > >> > > > or
> > > >> > > > > > 2?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > > >> number
> > > >> > > > of
> > > >> > > > > > threads
> > > >> > > > > > > > > used
> > > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> > the
> > > >> purpose
> > > >> > > > of
> > > >> > > > > > the
> > > >> > > > > > > > > threads
> > > >> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > > >> > > > > > > > > *__consumer_offsets*
> > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > should we
> > > >> need
> > > >> > > > > > > > > multi-threads
> > > >> > > > > > > > > > > here?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > See 1.2. The idea is to have an ability to shard
> the
> > > >> > > > processing as
> > > >> > > > > > the
> > > >> > > > > > > > > > computation could be heavy.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > >> default
> > > >> > > > session
> > > >> > > > > > > > > timeout not
> > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > > >> min/max
> > > >> > > > > > session
> > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> no?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > >> timeout to
> > > >> > > > detect
> > > >> > > > > > client
> > > >> > > > > > > > > > > failures when using the consumer group protocol.
> > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> The
> > > >> minimum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> The
> > > >> maximum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > This is indeed a mistake. The default session
> > timeout
> > > >> should
> > > >> > > > be 45s
> > > >> > > > > > > > > > (the current default).
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 2.2. The default server side assignor are
> [range,
> > > >> uniform],
> > > >> > > > > > which means
> > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > know
> > > >> why not
> > > >> > > > > > uniform
> > > >> > > > > > > > > one? I
> > > >> > > > > > > > > > > thought usually users will choose uniform
> assignor
> > > >> (former
> > > >> > > > sticky
> > > >> > > > > > > > > assinor)
> > > >> > > > > > > > > > > for better evenly distribution. Any other reason
> > we
> > > >> choose
> > > >> > > > range
> > > >> > > > > > > > > assignor
> > > >> > > > > > > > > > > as default?
> > > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > >> server side
> > > >> > > > > > assignors.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > The order on the server side has no influence
> > because
> > > >> the
> > > >> > > > client
> > > >> > > > > > must
> > > >> > > > > > > > > > chose the selector that he wants to use. There is
> no
> > > >> default
> > > >> > > > in the
> > > >> > > > > > > > > > current proposal. If the assignor is not specified
> > by
> > > >> the
> > > >> > > > client,
> > > >> > > > > > the
> > > >> > > > > > > > > > request is rejected. The default client value for
> > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > that my
> > > >> > > > answers
> > > >> > > > > > help
> > > >> > > > > > > > > > to clarify things. I will update the KIP as well
> > based
> > > >> on your
> > > >> > > > > > > > > > feedback.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Cheers,
> > > >> > > > > > > > > > David
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > >> showuon@gmail.com>
> > > >> > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Hi David,
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Thanks for the update.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Some more questions:
> > > >> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > >> > > > > > > > > > > > The new group coordinator will have a state
> > machine
> > > >> per
> > > >> > > > > > > > > > > *__consumer_offsets* partitions, where each
> state
> > > >> machine is
> > > >> > > > > > modelled
> > > >> > > > > > > > > as an
> > > >> > > > > > > > > > > event loop. Those state machines will be
> executed
> > in
> > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > assigning,
> > > >> > > > > > reconciling,
> > > >> > > > > > > > > stable,
> > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> section,
> > > >> right?
> > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > modelled
> > > >> as an
> > > >> > > > event
> > > >> > > > > > > > > loop"?
> > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > >> *__consumer_offsets*
> > > >> > > > > > > > > partitions?
> > > >> > > > > > > > > > > Not a state machine "per consumer group" owned
> by
> > a
> > > >> group
> > > >> > > > > > coordinator?
> > > >> > > > > > > > > For
> > > >> > > > > > > > > > > example, if one group coordinator owns 2
> consumer
> > > >> groups, and
> > > >> > > > > > both
> > > >> > > > > > > > > exist in
> > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > > >> for it,
> > > >> > > > or
> > > >> > > > > > 2?
> > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > > >> number
> > > >> > > > of
> > > >> > > > > > threads
> > > >> > > > > > > > > used
> > > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> > the
> > > >> purpose
> > > >> > > > of
> > > >> > > > > > the
> > > >> > > > > > > > > threads
> > > >> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > > >> > > > > > > > > *__consumer_offsets*
> > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > should we
> > > >> need
> > > >> > > > > > > > > multi-threads
> > > >> > > > > > > > > > > here?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 2. For the default value in the new configs:
> > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > >> default
> > > >> > > > session
> > > >> > > > > > > > > timeout not
> > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > > >> min/max
> > > >> > > > > > session
> > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> no?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > >> timeout to
> > > >> > > > detect
> > > >> > > > > > client
> > > >> > > > > > > > > > > failures when using the consumer group protocol.
> > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> The
> > > >> minimum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> The
> > > >> maximum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 2.2. The default server side assignor are
> [range,
> > > >> uniform],
> > > >> > > > > > which means
> > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > know
> > > >> why not
> > > >> > > > > > uniform
> > > >> > > > > > > > > one? I
> > > >> > > > > > > > > > > thought usually users will choose uniform
> assignor
> > > >> (former
> > > >> > > > sticky
> > > >> > > > > > > > > assinor)
> > > >> > > > > > > > > > > for better evenly distribution. Any other reason
> > we
> > > >> choose
> > > >> > > > range
> > > >> > > > > > > > > assignor
> > > >> > > > > > > > > > > as default?
> > > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > >> server side
> > > >> > > > > > assignors.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Thank you.
> > > >> > > > > > > > > > > Luke
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > >> showuon@gmail.com
> > > >> > > > >
> > > >> > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > Hi Sagar,
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > >> integrating with
> > > >> > > > > > KIP-848,
> > > >> > > > > > > > > but I
> > > >> > > > > > > > > > > > think we should have a separate discussion
> > thread
> > > >> for the
> > > >> > > > Kafka
> > > >> > > > > > > > > Connect
> > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> Consumer
> > > >> Rebalance
> > > >> > > > > > Protocol
> > > >> > > > > > > > > [1],
> > > >> > > > > > > > > > > > and let this discussion thread focus on
> consumer
> > > >> rebalance
> > > >> > > > > > protocol,
> > > >> > > > > > > > > WDYT?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > [1]
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > >
> > > >> > > > > >
> > > >> > > >
> > > >>
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > Thank you.
> > > >> > > > > > > > > > > > Luke
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > >> > > > > > sagarmeansocean@gmail.com>
> > > >> > > > > > > > > wrote:
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > Looks
> > > >> like
> > > >> > > > there's
> > > >> > > > > > > > > agreement on
> > > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> > revisit
> > > >> the doc
> > > >> > > > and
> > > >> > > > > > see
> > > >> > > > > > > > > what
> > > >> > > > > > > > > > > >> changes are to be made.
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > > > > >> Thanks!
> > > >> > > > > > > > > > > >> Sagar.
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > >> > > > > > > > > > > >> wrote:
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > > > > >> > Hi Sagar,
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > That's
> > > >> really
> > > >> > > > > > helpful. I
> > > >> > > > > > > > > > > >> > will take a look at it.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > and the
> > > >> > > > Consumer
> > > >> > > > > > could
> > > >> > > > > > > > > share
> > > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > > >> difference is
> > > >> > > > that
> > > >> > > > > > the
> > > >> > > > > > > > > Consumer
> > > >> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > > >> Connect
> > > >> > > > assigns
> > > >> > > > > > tasks
> > > >> > > > > > > > > to
> > > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> support
> > > >> different
> > > >> > > > > > resource
> > > >> > > > > > > > > types
> > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect.
> The
> > > >> dedicated
> > > >> > > > APIs
> > > >> > > > > > > > > would be
> > > >> > > > > > > > > > > >> > similar to the new ones but different on
> the
> > > >> > > > > > content/resources and
> > > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > > >> coordinator
> > > >> > > > side.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > not a
> > > >> fan of
> > > >> > > > > > > > > overcharging
> > > >> > > > > > > > > > > >> > APIs to serve different purposes. That
> being
> > > >> said, I am
> > > >> > > > not
> > > >> > > > > > > > > opposed to
> > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > I think that we can continue to discuss it
> > here
> > > >> for now
> > > >> > > > in
> > > >> > > > > > order
> > > >> > > > > > > > > to
> > > >> > > > > > > > > > > >> > ensure that this KIP is compatible with
> what
> > we
> > > >> will do
> > > >> > > > for
> > > >> > > > > > > > > Connect in
> > > >> > > > > > > > > > > >> > the future.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > Best,
> > > >> > > > > > > > > > > >> > David
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot
> <
> > > >> > > > > > djacot@confluent.io>
> > > >> > > > > > > > > wrote:
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > Hi all,
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> through
> > and
> > > >> address
> > > >> > > > > > your
> > > >> > > > > > > > > comments
> > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > feedback.
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > Cheers,
> > > >> > > > > > > > > > > >> > > David
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > Harris
> > > >> <
> > > >> > > > > > > > > gharris1727@gmail.com
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > wrote:
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > Hey All!
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to
> see
> > > >> > > > cooperative
> > > >> > > > > > > > > rebalancing
> > > >> > > > > > > > > > > >> > making it
> > > >> > > > > > > > > > > >> > > > down the stack!
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > > >> describes how
> > > >> > > > > > member
> > > >> > > > > > > > > epoch
> > > >> > > > > > > > > > > >> > should
> > > >> > > > > > > > > > > >> > > > advance in step with the group epoch
> and
> > > >> assignment
> > > >> > > > > > epoch
> > > >> > > > > > > > > values. I
> > > >> > > > > > > > > > > >> > think
> > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> reasons
> > > >> described
> > > >> > > > in
> > > >> > > > > > the
> > > >> > > > > > > > > KIP. When
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > >> partitions to a
> > > >> > > > > > worker,
> > > >> > > > > > > > > what
> > > >> > > > > > > > > > > >> member
> > > >> > > > > > > > > > > >> > > > epoch does each incremental assignment
> > use?
> > > >> Are
> > > >> > > > member
> > > >> > > > > > epochs
> > > >> > > > > > > > > > > >> re-used,
> > > >> > > > > > > > > > > >> > and
> > > >> > > > > > > > > > > >> > > > a single member epoch can correspond to
> > > >> multiple
> > > >> > > > > > different
> > > >> > > > > > > > > > > >> > (monotonically
> > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > opaque
> > > >> to the
> > > >> > > > group
> > > >> > > > > > > > > > > >> coordinator? If
> > > >> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > > >> > > > implementations
> > > >> > > > > > > > > interact
> > > >> > > > > > > > > > > >> with
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> > meaning
> > > >> agreed
> > > >> > > > > > upon? If
> > > >> > > > > > > > > so, what
> > > >> > > > > > > > > > > >> > is the
> > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > > >> including
> > > >> > > > such
> > > >> > > > > > > > > functionality
> > > >> > > > > > > > > > > >> > in the
> > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 3. The following is included in the
> KIP:
> > > >> "Thanks to
> > > >> > > > > > this, the
> > > >> > > > > > > > > input
> > > >> > > > > > > > > > > >> of
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > client side assignor is entirely driven
> > by
> > > >> the group
> > > >> > > > > > > > > coordinator.
> > > >> > > > > > > > > > > >> The
> > > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > >> maintaining
> > > >> > > > any
> > > >> > > > > > state
> > > >> > > > > > > > > besides
> > > >> > > > > > > > > > > >> its
> > > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > that the
> > > >> > > > > > client-side
> > > >> > > > > > > > > assignor
> > > >> > > > > > > > > > > >> MAY
> > > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> state
> > > >> (such as
> > > >> > > > > > partition
> > > >> > > > > > > > > > > >> > throughput,
> > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics,
> etc),
> > or
> > > >> that
> > > >> > > > > > additional
> > > >> > > > > > > > > > > >> > non-Metadata
> > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> classes
> > > >> > > > > > > > > > > >> > > > for
> > > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > >> > > > > > > > > > > >> > > > and
> > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > >> > > > > > that
> > > >> > > > > > > > > seem to
> > > >> > > > > > > > > > > >> > > > overlap significantly. Is it possible
> for
> > > >> these two
> > > >> > > > > > > > > implementations
> > > >> > > > > > > > > > > >> to
> > > >> > > > > > > > > > > >> > be
> > > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> > feature
> > > >> parity
> > > >> > > > of
> > > >> > > > > > > > > server-side
> > > >> > > > > > > > > > > >> and
> > > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > >> facilitate
> > > >> > > > > > operational
> > > >> > > > > > > > > > > >> > flexibility in
> > > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > >> server-side
> > > >> > > > > > assignor
> > > >> > > > > > > > > has some
> > > >> > > > > > > > > > > >> > poor
> > > >> > > > > > > > > > > >> > > > behavior and needs a patch, deploying
> the
> > > >> patched
> > > >> > > > > > assignor to
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > client
> > > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > >> client-side
> > > >> > > > > > assignor
> > > >> > > > > > > > > may be
> > > >> > > > > > > > > > > >> > faster
> > > >> > > > > > > > > > > >> > > > and less risky than patching all of the
> > > >> brokers.
> > > >> > > > With
> > > >> > > > > > the
> > > >> > > > > > > > > currently
> > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > >> > > > reimplementation
> > > >> > > > > > would
> > > >> > > > > > > > > have
> > > >> > > > > > > > > > > >> to be
> > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > diverged
> > > >> > > > > > significantly,
> > > >> > > > > > > > > then it
> > > >> > > > > > > > > > > >> is
> > > >> > > > > > > > > > > >> > > > possible that a reimplementation would
> > not be
> > > >> > > > possible.
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > --
> > > >> > > > > > > > > > > >> > > > Greg Harris
> > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > >> > > > > > > > > > > >> > wrote:
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> discuss
> > how
> > > >> Connect
> > > >> > > > > > would
> > > >> > > > > > > > > need to
> > > >> > > > > > > > > > > >> > change
> > > >> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> > > >> Here's the
> > > >> > > > page:
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > >
> > > >> > > > > >
> > > >> > > >
> > > >>
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > tried
> > > >> to keep
> > > >> > > > a
> > > >> > > > > > format
> > > >> > > > > > > > > > > >> similar to
> > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> > Also,
> > > >> do you
> > > >> > > > > > think this
> > > >> > > > > > > > > > > >> should
> > > >> > > > > > > > > > > >> > be
> > > >> > > > > > > > > > > >> > > > > moved to a separate discussion thread
> > or
> > > >> is this
> > > >> > > > one
> > > >> > > > > > fine?
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > Thanks!
> > > >> > > > > > > > > > > >> > > > > Sagar.
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> Sagar <
> > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > >> > > > > > > > > > > >> > wrote:
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > Kafka
> > > >> Streams.
> > > >> > > > > > Sure, I
> > > >> > > > > > > > > would do
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > >> document.
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > Guozhang
> > > >> Wang <
> > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > >> > > > > > > > > > > >> > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >> It would be great if you could
> come
> > > >> back with
> > > >> > > > some
> > > >> > > > > > > > > analysis on
> > > >> > > > > > > > > > > >> > how to
> > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > integration
> > > >> with
> > > >> > > > the new
> > > >> > > > > > > > > protocol;
> > > >> > > > > > > > > > > >> so
> > > >> > > > > > > > > > > >> > far
> > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > "protocol
> > > >> type"
> > > >> > > > we
> > > >> > > > > > did not
> > > >> > > > > > > > > yet
> > > >> > > > > > > > > > > >> think
> > > >> > > > > > > > > > > >> > > > > through
> > > >> > > > > > > > > > > >> > > > > >> the Connect side implementations.
> > For
> > > >> Streams
> > > >> > > > > > here's a
> > > >> > > > > > > > > draft of
> > > >> > > > > > > > > > > >> > > > > >> integration
> > > >> > > > > > > > > > > >> > > > > >> plan:
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > >
> > > >> > > > > >
> > > >> > > >
> > > >>
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> > Connect.
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> > Sagar <
> > > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> > > >> reason I
> > > >> > > > thought
> > > >> > > > > > > > > connect can
> > > >> > > > > > > > > > > >> > also fit
> > > >> > > > > > > > > > > >> > > > > >> into
> > > >> > > > > > > > > > > >> > > > > >> > this new scheme is that even
> > today the
> > > >> > > > connect
> > > >> > > > > > uses a
> > > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > >> > > > > > > > > > > >> > > > > >> > extending from
> > AbstractCoordinator to
> > > >> empower
> > > >> > > > > > > > > rebalances of
> > > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > >> WorkerCoordinator sets
> > > >> > > > the
> > > >> > > > > > > > > > > >> protocolType()
> > > >> > > > > > > > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > >> connect
> > > >> > > > > > > > > > > >> > > > > >> > and uses the metadata() method
> by
> > > >> plumbing
> > > >> > > > into
> > > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > > >> connect would
> > > >> > > > be
> > > >> > > > > > > > > similar at a
> > > >> > > > > > > > > > > >> > high
> > > >> > > > > > > > > > > >> > > > > level
> > > >> > > > > > > > > > > >> > > > > >> to
> > > >> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> > because
> > > >> of the
> > > >> > > > > > Client
> > > >> > > > > > > > > side
> > > >> > > > > > > > > > > >> > assignors
> > > >> > > > > > > > > > > >> > > > > being
> > > >> > > > > > > > > > > >> > > > > >> > used in both. At an
> implementation
> > > >> level, we
> > > >> > > > > > might
> > > >> > > > > > > > > need to
> > > >> > > > > > > > > > > >> make
> > > >> > > > > > > > > > > >> > a lot
> > > >> > > > > > > > > > > >> > > > > of
> > > >> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > > >> assignment
> > > >> > > > protocol
> > > >> > > > > > like
> > > >> > > > > > > > > > > >> enhancing
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > > >> SyncGroup and
> > > >> > > > > > using
> > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > >> > > > > > > > > > > >> > > > > >> > API etc again on similar lines
> to
> > > >> streams (or
> > > >> > > > > > there
> > > >> > > > > > > > > might be
> > > >> > > > > > > > > > > >> > > > > >> deviations). I
> > > >> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> > > >> analysis of
> > > >> > > > the
> > > >> > > > > > same
> > > >> > > > > > > > > and we
> > > >> > > > > > > > > > > >> > can have
> > > >> > > > > > > > > > > >> > > > > a
> > > >> > > > > > > > > > > >> > > > > >> > separate discussion thread for
> > that
> > > >> as that
> > > >> > > > would
> > > >> > > > > > > > > derail this
> > > >> > > > > > > > > > > >> > > > > discussion
> > > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if that
> sounds
> > > >> good to
> > > >> > > > you.
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> > David
> > > >> Jacot
> > > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> > wrote:
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > >> `Assignment#error`.
> > > >> > > > > > Sure, I
> > > >> > > > > > > > > can
> > > >> > > > > > > > > > > >> > mention it.
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> > from
> > > >> his
> > > >> > > > current
> > > >> > > > > > > > > assignment
> > > >> > > > > > > > > > > >> to
> > > >> > > > > > > > > > > >> > his
> > > >> > > > > > > > > > > >> > > > > >> > > target assignment when he can
> > move
> > > >> to
> > > >> > > > epoch 3.
> > > >> > > > > > When
> > > >> > > > > > > > > that
> > > >> > > > > > > > > > > >> > happens,
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > member assignment is updated
> and
> > > >> persisted
> > > >> > > > > > with all
> > > >> > > > > > > > > its
> > > >> > > > > > > > > > > >> > assigned
> > > >> > > > > > > > > > > >> > > > > >> > > partitions even if they are
> not
> > all
> > > >> revoked
> > > >> > > > > > yet. In
> > > >> > > > > > > > > other
> > > >> > > > > > > > > > > >> > words, the
> > > >> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> > target
> > > >> > > > > > assignment.
> > > >> > > > > > > > > This is
> > > >> > > > > > > > > > > >> > basically
> > > >> > > > > > > > > > > >> > > > > an
> > > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid having
> to
> > > >> write all
> > > >> > > > the
> > > >> > > > > > > > > changes to
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > log.
> > > >> > > > > > > > > > > >> > > > > The
> > > >> > > > > > > > > > > >> > > > > >> > > examples are based on the
> > persisted
> > > >> state
> > > >> > > > so I
> > > >> > > > > > > > > understand
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> > > >> improve
> > > >> > > > this in
> > > >> > > > > > the
> > > >> > > > > > > > > > > >> > description.
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> > > >> reuse the
> > > >> > > > > > protocol
> > > >> > > > > > > > > with a
> > > >> > > > > > > > > > > >> > client side
> > > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > > >> protocol. The
> > > >> > > > > > assignment
> > > >> > > > > > > > > is
> > > >> > > > > > > > > > > >> about
> > > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> > could
> > > >> > > > Connect
> > > >> > > > > > fit
> > > >> > > > > > > > > into this?
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> > > Best,
> > > >> > > > > > > > > > > >> > > > > >> > > David
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55
> PM
> > > >> Sagar <
> > > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > >> > > > > > > > > > > >> > > > > >> wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just
> had
> > > >> minor
> > > >> > > > > > observations:
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > > >> section in
> > > >> > > > Client
> > > >> > > > > > Side
> > > >> > > > > > > > > mode
> > > >> > > > > > > > > > > >> > Assignment
> > > >> > > > > > > > > > > >> > > > > >> > > process,
> > > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this
> > case,
> > > >> the
> > > >> > > > client
> > > >> > > > > > side
> > > >> > > > > > > > > assignor
> > > >> > > > > > > > > > > >> can
> > > >> > > > > > > > > > > >> > > > > return
> > > >> > > > > > > > > > > >> > > > > >> an
> > > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> > coordinator`.
> > > >> In this
> > > >> > > > > > case are
> > > >> > > > > > > > > you
> > > >> > > > > > > > > > > >> > referring to
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > >> AssignmentError
> > > >> > > > that's
> > > >> > > > > > > > > listed down
> > > >> > > > > > > > > > > >> > towards
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > end?
> > > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it
> would
> > > >> make sense
> > > >> > > > to
> > > >> > > > > > > > > mention this
> > > >> > > > > > > > > > > >> > > > > explicitly
> > > >> > > > > > > > > > > >> > > > > >> > here?
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > section, I
> > > >> have a
> > > >> > > > > > slight
> > > >> > > > > > > > > > > >> confusion,
> > > >> > > > > > > > > > > >> > not
> > > >> > > > > > > > > > > >> > > > > sure
> > > >> > > > > > > > > > > >> > > > > >> if
> > > >> > > > > > > > > > > >> > > > > >> > > > others have the same.
> Consider
> > > >> this step:
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> > > >> coordinator
> > > >> > > > > > > > > transitions him
> > > >> > > > > > > > > > > >> to
> > > >> > > > > > > > > > > >> > epoch
> > > >> > > > > > > > > > > >> > > > > 3
> > > >> > > > > > > > > > > >> > > > > >> > > because
> > > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > revoke. It
> > > >> > > > persists
> > > >> > > > > > the
> > > >> > > > > > > > > change and
> > > >> > > > > > > > > > > >> > reply.
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > (epoch=3)
> > > >> > > > > > > > > > > >> > > > > >> > > >       - A -
> partitions=[foo-0]
> > > >> > > > > > > > > > > >> > > > > >> > > >       - B -
> partitions=[foo-2]
> > > >> > > > > > > > > > > >> > > > > >> > > >       - C -
> partitions=[foo-1]
> > > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > >> partitions=[foo-0,
> > > >> > > > > > foo-1]
> > > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > >> partitions=[foo-2]
> > > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > >> partitions=[foo-1]
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > transitions
> > > >> to
> > > >> > > > epoch 3
> > > >> > > > > > but
> > > >> > > > > > > > > cannot
> > > >> > > > > > > > > > > >> get
> > > >> > > > > > > > > > > >> > foo-1
> > > >> > > > > > > > > > > >> > > > > >> yet.
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that
> > member C
> > > >> can't
> > > >> > > > get
> > > >> > > > > > the
> > > >> > > > > > > > > foo-1
> > > >> > > > > > > > > > > >> > partition
> > > >> > > > > > > > > > > >> > > > > yet,
> > > >> > > > > > > > > > > >> > > > > >> > but
> > > >> > > > > > > > > > > >> > > > > >> > > > based on the description
> > above,
> > > >> it seems
> > > >> > > > it
> > > >> > > > > > > > > already has
> > > >> > > > > > > > > > > >> it.
> > > >> > > > > > > > > > > >> > Do you
> > > >> > > > > > > > > > > >> > > > > >> > think
> > > >> > > > > > > > > > > >> > > > > >> > > it
> > > >> > > > > > > > > > > >> > > > > >> > > > would be better to remove it
> > and
> > > >> > > > populate it
> > > >> > > > > > only
> > > >> > > > > > > > > when it
> > > >> > > > > > > > > > > >> > actually
> > > >> > > > > > > > > > > >> > > > > >> gets
> > > >> > > > > > > > > > > >> > > > > >> > > it?
> > > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> > > >> places, so
> > > >> > > > have
> > > >> > > > > > I
> > > >> > > > > > > > > > > >> understood it
> > > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > > >> > > > > > > > > > > >> > > > > >> > > ?
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might
> > be
> > > >> out of
> > > >> > > > scope
> > > >> > > > > > of
> > > >> > > > > > > > > this
> > > >> > > > > > > > > > > >> > discussion,
> > > >> > > > > > > > > > > >> > > > > but
> > > >> > > > > > > > > > > >> > > > > >> > from
> > > >> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > > >> probably be
> > > >> > > > > > running in
> > > >> > > > > > > > > client
> > > >> > > > > > > > > > > >> > side
> > > >> > > > > > > > > > > >> > > > > >> assignor
> > > >> > > > > > > > > > > >> > > > > >> > > mode
> > > >> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > > >> protocol as it
> > > >> > > > has
> > > >> > > > > > its
> > > >> > > > > > > > > own
> > > >> > > > > > > > > > > >> Custom
> > > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > >> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00
> > PM
> > > >> David
> > > >> > > > Jacot
> > > >> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is
> > to
> > > >> move
> > > >> > > > forward
> > > >> > > > > > with
> > > >> > > > > > > > > > > >> > specialized API
> > > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> > > >> generic API.
> > > >> > > > For
> > > >> > > > > > > > > Connect, we
> > > >> > > > > > > > > > > >> > can apply
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > > >> reuse/share the
> > > >> > > > core
> > > >> > > > > > > > > > > >> > implementation on
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the
> schema
> > > >> registry, I
> > > >> > > > > > think
> > > >> > > > > > > > > that we
> > > >> > > > > > > > > > > >> > should
> > > >> > > > > > > > > > > >> > > > > >> consider
> > > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to
> do
> > > >> simple
> > > >> > > > > > > > > membership/leader
> > > >> > > > > > > > > > > >> > election.
> > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > >> > > > > > > > > > > >> > > > > >> > > > > David
> > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at
> > 10:22
> > > >> AM Ismael
> > > >> > > > > > Juma <
> > > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > >> java.util.regex.Pattern
> > > >> > > > vs
> > > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > >> > > > > > > > > > > >> > > > > >> > > > > > document the differences
> > in
> > > >> more
> > > >> > > > detail
> > > >> > > > > > before
> > > >> > > > > > > > > > > >> deciding
> > > >> > > > > > > > > > > >> > one
> > > >> > > > > > > > > > > >> > > > > way
> > > >> > > > > > > > > > > >> > > > > >> or
> > > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if people
> pass
> > > >> > > > > > > > > java.util.regex.Pattern,
> > > >> > > > > > > > > > > >> they
> > > >> > > > > > > > > > > >> > expect
> > > >> > > > > > > > > > > >> > > > > >> > their
> > > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored.
> > If
> > > >> we are
> > > >> > > > doing
> > > >> > > > > > > > > something
> > > >> > > > > > > > > > > >> > different,
> > > >> > > > > > > > > > > >> > > > > >> then
> > > >> > > > > > > > > > > >> > > > > >> > we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > should consider adding
> an
> > > >> overload
> > > >> > > > with
> > > >> > > > > > our own
> > > >> > > > > > > > > > > >> Pattern
> > > >> > > > > > > > > > > >> > class
> > > >> > > > > > > > > > > >> > > > > (I
> > > >> > > > > > > > > > > >> > > > > >> > > don't
> > > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to
> expose
> > > >> re2j's at
> > > >> > > > this
> > > >> > > > > > > > > point).
> > > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids,
> > any
> > > >> major new
> > > >> > > > > > protocol
> > > >> > > > > > > > > should
> > > >> > > > > > > > > > > >> > > > > integrate
> > > >> > > > > > > > > > > >> > > > > >> > fully
> > > >> > > > > > > > > > > >> > > > > >> > > > > with
> > > >> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> > topic
> > > >> > > > > > recreation case
> > > >> > > > > > > > > > > >> > correctly.
> > > >> > > > > > > > > > > >> > > > > That's
> > > >> > > > > > > > > > > >> > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> > > main
> > > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle.
> I
> > > >> agree with
> > > >> > > > > > David
> > > >> > > > > > > > > that we'd
> > > >> > > > > > > > > > > >> > want to
> > > >> > > > > > > > > > > >> > > > > add
> > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant
> protocols
> > > >> that don't
> > > >> > > > > > have it
> > > >> > > > > > > > > yet and
> > > >> > > > > > > > > > > >> > that we
> > > >> > > > > > > > > > > >> > > > > can
> > > >> > > > > > > > > > > >> > > > > >> > > probably
> > > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> > versus
> > > >> adding
> > > >> > > > new
> > > >> > > > > > APIs
> > > >> > > > > > > > > to the
> > > >> > > > > > > > > > > >> > Java
> > > >> > > > > > > > > > > >> > > > > >> Consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > >> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new
> > APIs
> > > >> is
> > > >> > > > required
> > > >> > > > > > for
> > > >> > > > > > > > > > > >> reasonable
> > > >> > > > > > > > > > > >> > > > > >> semantics).
> > > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> > about
> > > >> the
> > > >> > > > > > coordinator
> > > >> > > > > > > > > storing
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> configs.
> > > >> > > > > > > > > > > >> > > > > >> > > It's
> > > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to
> be
> > > >> > > > centralized
> > > >> > > > > > in the
> > > >> > > > > > > > > > > >> metadata
> > > >> > > > > > > > > > > >> > log for
> > > >> > > > > > > > > > > >> > > > > >> > > various
> > > >> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > > >> visibility,
> > > >> > > > > > consistency,
> > > >> > > > > > > > > > > >> etc.).
> > > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > > >> > > > > > > > > > > >> > > > > >> > > am
> > > >> > > > > > > > > > > >> > > > > >> > > > > not
> > > >> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> > > >> deleting
> > > >> > > > > > configs in a
> > > >> > > > > > > > > way
> > > >> > > > > > > > > > > >> that
> > > >> > > > > > > > > > > >> > they
> > > >> > > > > > > > > > > >> > > > > >> cannot
> > > >> > > > > > > > > > > >> > > > > >> > > be
> > > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> property
> > > >> for modern
> > > >> > > > > > systems
> > > >> > > > > > > > > is to
> > > >> > > > > > > > > > > >> > minimize
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > number
> > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> > > >> scenarios.
> > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at
> > 3:47
> > > >> PM David
> > > >> > > > > > Jacot
> > > >> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> > answers
> > > >> are
> > > >> > > > below:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> path,
> > > >> especially
> > > >> > > > > > the last
> > > >> > > > > > > > > step
> > > >> > > > > > > > > > > >> when
> > > >> > > > > > > > > > > >> > > > > clients
> > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > >> protocol, in
> > > >> > > > which
> > > >> > > > > > we
> > > >> > > > > > > > > would
> > > >> > > > > > > > > > > >> have a
> > > >> > > > > > > > > > > >> > > > > window
> > > >> > > > > > > > > > > >> > > > > >> > where
> > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > old
> > > >> > > > protocols /
> > > >> > > > > > rpcs
> > > >> > > > > > > > > are
> > > >> > > > > > > > > > > >> used
> > > >> > > > > > > > > > > >> > by
> > > >> > > > > > > > > > > >> > > > > >> members
> > > >> > > > > > > > > > > >> > > > > >> > of
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > >> coordinator could
> > > >> > > > > > "mimic"
> > > >> > > > > > > > > the old
> > > >> > > > > > > > > > > >> > behavior
> > > >> > > > > > > > > > > >> > > > > >> while
> > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > something
> > > >> we
> > > >> > > > need to
> > > >> > > > > > > > > present
> > > >> > > > > > > > > > > >> about.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> published
> > a
> > > >> new
> > > >> > > > version
> > > >> > > > > > of KIP
> > > >> > > > > > > > > which
> > > >> > > > > > > > > > > >> > includes
> > > >> > > > > > > > > > > >> > > > > >> more
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > details about this.
> See
> > the
> > > >> > > > "Supporting
> > > >> > > > > > > > > Online
> > > >> > > > > > > > > > > >> > Consumer
> > > >> > > > > > > > > > > >> > > > > Group
> > > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > and the
> "Compatibility,
> > > >> > > > Deprecation,
> > > >> > > > > > and
> > > >> > > > > > > > > Migration
> > > >> > > > > > > > > > > >> > Plan". I
> > > >> > > > > > > > > > > >> > > > > >> think
> > > >> > > > > > > > > > > >> > > > > >> > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> through
> > a
> > > >> few cases
> > > >> > > > > > now but
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > overall idea
> > > >> > > > > > > > > > > >> > > > > >> and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > > >> understandable.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> topic
> > > >> ids. So
> > > >> > > > far as
> > > >> > > > > > > > > KIP-516 the
> > > >> > > > > > > > > > > >> > topic ids
> > > >> > > > > > > > > > > >> > > > > >> are
> > > >> > > > > > > > > > > >> > > > > >> > > only
> > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> admin
> > > >> client,
> > > >> > > > but
> > > >> > > > > > they
> > > >> > > > > > > > > are not
> > > >> > > > > > > > > > > >> > exposed
> > > >> > > > > > > > > > > >> > > > > via
> > > >> > > > > > > > > > > >> > > > > >> any
> > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > think
> > > >> the
> > > >> > > > > > question is,
> > > >> > > > > > > > > first
> > > >> > > > > > > > > > > >> > should we
> > > >> > > > > > > > > > > >> > > > > >> let
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > maintaining
> > > >> the
> > > >> > > > names
> > > >> > > > > > -> ids
> > > >> > > > > > > > > mapping
> > > >> > > > > > > > > > > >> > itself
> > > >> > > > > > > > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > fully
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > > >> existing
> > > >> > > > RPCs
> > > >> > > > > > and the
> > > >> > > > > > > > > new
> > > >> > > > > > > > > > > >> RPCs
> > > >> > > > > > > > > > > >> > with
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > ever
> > > >> consider
> > > >> > > > > > exposing
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > topic ids
> > > >> > > > > > > > > > > >> > > > > in
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > (both
> > > >> > > > > > > > > subscribe/assign, as
> > > >> > > > > > > > > > > >> well
> > > >> > > > > > > > > > > >> > as in
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > like
> > > >> topic
> > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we
> > would
> > > >> include
> > > >> > > > > > converting
> > > >> > > > > > > > > all
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > offsets
> > > >> > > > > > > > > > > >> > > > > >> > > related
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic
> ids
> > in
> > > >> this
> > > >> > > > KIP,
> > > >> > > > > > the
> > > >> > > > > > > > > consumer
> > > >> > > > > > > > > > > >> > would be
> > > >> > > > > > > > > > > >> > > > > >> able
> > > >> > > > > > > > > > > >> > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with
> topic
> > > >> ids. That
> > > >> > > > > > being
> > > >> > > > > > > > > said, it
> > > >> > > > > > > > > > > >> > still has
> > > >> > > > > > > > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > > provide
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> > various
> > > >> APIs so
> > > >> > > > > > having a
> > > >> > > > > > > > > > > >> mapping
> > > >> > > > > > > > > > > >> > in the
> > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to
> me.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a
> strong
> > > >> opinion on
> > > >> > > > > > this.
> > > >> > > > > > > > > Here I
> > > >> > > > > > > > > > > >> > wonder if
> > > >> > > > > > > > > > > >> > > > > >> this
> > > >> > > > > > > > > > > >> > > > > >> > > goes
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of
> this
> > > >> KIP. I
> > > >> > > > would
> > > >> > > > > > rather
> > > >> > > > > > > > > focus
> > > >> > > > > > > > > > > >> on
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> > internals
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> consider
> > > >> this
> > > >> > > > > > separately if
> > > >> > > > > > > > > we see
> > > >> > > > > > > > > > > >> > value in
> > > >> > > > > > > > > > > >> > > > > >> doing
> > > >> > > > > > > > > > > >> > > > > >> > > it.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to
> Ismael's
> > > >> point about
> > > >> > > > > > using
> > > >> > > > > > > > > topic ids
> > > >> > > > > > > > > > > >> > in the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> ConsumerGroupHeartbeatRequest, I
> > > >> > > > think
> > > >> > > > > > that
> > > >> > > > > > > > > there
> > > >> > > > > > > > > > > >> is
> > > >> > > > > > > > > > > >> > one
> > > >> > > > > > > > > > > >> > > > > >> > advantage
> > > >> > > > > > > > > > > >> > > > > >> > > in
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> > consumer
> > > >> will
> > > >> > > > have
> > > >> > > > > > the
> > > >> > > > > > > > > > > >> opportunity
> > > >> > > > > > > > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > >> validate
> > > >> > > > > > > > > > > >> > > > > >> > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists
> before
> > > >> passing
> > > >> > > > them
> > > >> > > > > > into
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> group
> > > >> > > > > > > > > > > >> > > > > rebalance
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously,
> the
> > > >> > > > coordinator
> > > >> > > > > > will
> > > >> > > > > > > > > also
> > > >> > > > > > > > > > > >> notice
> > > >> > > > > > > > > > > >> > it but
> > > >> > > > > > > > > > > >> > > > > >> it
> > > >> > > > > > > > > > > >> > > > > >> > > does
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way
> to
> > > >> reject an
> > > >> > > > > > invalid
> > > >> > > > > > > > > topic in
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> response.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > David
> > > >> on all
> > > >> > > > other
> > > >> > > > > > minor
> > > >> > > > > > > > > > > >> questions
> > > >> > > > > > > > > > > >> > > > > except
> > > >> > > > > > > > > > > >> > > > > >> for
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > > >> question:
> > > >> > > > > > personally
> > > >> > > > > > > > > I think
> > > >> > > > > > > > > > > >> > it's not
> > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > subscribe
> > > >> API with
> > > >> > > > > > Pattern,
> > > >> > > > > > > > > but
> > > >> > > > > > > > > > > >> > instead we
> > > >> > > > > > > > > > > >> > > > > >> still
> > > >> > > > > > > > > > > >> > > > > >> > > use
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> documenting
> > > >> that our
> > > >> > > > > > > > > subscription may
> > > >> > > > > > > > > > > >> be
> > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> incompatible
> > > >> case is a
> > > >> > > > > > very rare
> > > >> > > > > > > > > > > >> scenario
> > > >> > > > > > > > > > > >> > I felt
> > > >> > > > > > > > > > > >> > > > > >> > using
> > > >> > > > > > > > > > > >> > > > > >> > > an
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > based
> > > >> > > > > > subscription may
> > > >> > > > > > > > > be
> > > >> > > > > > > > > > > >> more
> > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > >> > > > > > > > > > > >> > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I
> have
> > to
> > > >> look at
> > > >> > > > the
> > > >> > > > > > > > > differences
> > > >> > > > > > > > > > > >> > between
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > two
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > > >> understand the
> > > >> > > > > > potential
> > > >> > > > > > > > > issues.
> > > >> > > > > > > > > > > >> My
> > > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work for
> all
> > the
> > > >> basic
> > > >> > > > > > regular
> > > >> > > > > > > > > > > >> > expressions. The
> > > >> > > > > > > > > > > >> > > > > >> > > differences
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> > mainly
> > > >> about
> > > >> > > > the
> > > >> > > > > > various
> > > >> > > > > > > > > > > >> character
> > > >> > > > > > > > > > > >> > > > > >> classes. I
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other
> people
> > > >> think
> > > >> > > > about
> > > >> > > > > > this.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022
> at
> > > >> 11:28 PM
> > > >> > > > > > Guozhang
> > > >> > > > > > > > > Wang <
> > > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > >> > > > > > > > > > > >> > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I
> think
> > on
> > > >> the high
> > > >> > > > > > level
> > > >> > > > > > > > > there are
> > > >> > > > > > > > > > > >> > two meta
> > > >> > > > > > > > > > > >> > > > > >> > > points we
> > > >> > > > > > > > > > > >> > > > > >> > > > > need
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit
> > more:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> path,
> > > >> especially
> > > >> > > > > > the last
> > > >> > > > > > > > > step
> > > >> > > > > > > > > > > >> when
> > > >> > > > > > > > > > > >> > > > > clients
> > > >> > > > > > > > > > > >> > > > > >> > > flip the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > >> protocol, in
> > > >> > > > which
> > > >> > > > > > we
> > > >> > > > > > > > > would
> > > >> > > > > > > > > > > >> have a
> > > >> > > > > > > > > > > >> > > > > window
> > > >> > > > > > > > > > > >> > > > > >> > where
> > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > old
> > > >> > > > protocols /
> > > >> > > > > > rpcs
> > > >> > > > > > > > > are
> > > >> > > > > > > > > > > >> used
> > > >> > > > > > > > > > > >> > by
> > > >> > > > > > > > > > > >> > > > > >> members
> > > >> > > > > > > > > > > >> > > > > >> > of
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > >> coordinator could
> > > >> > > > > > "mimic"
> > > >> > > > > > > > > the old
> > > >> > > > > > > > > > > >> > behavior
> > > >> > > > > > > > > > > >> > > > > >> while
> > > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > something
> > > >> we
> > > >> > > > need to
> > > >> > > > > > > > > present
> > > >> > > > > > > > > > > >> about.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of
> topic
> > > >> ids. So
> > > >> > > > far as
> > > >> > > > > > > > > KIP-516 the
> > > >> > > > > > > > > > > >> > topic ids
> > > >> > > > > > > > > > > >> > > > > >> are
> > > >> > > > > > > > > > > >> > > > > >> > > only
> > > >> > > > > > > > > > > >> > > > > >> > > > > used
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> admin
> > > >> client,
> > > >> > > > but
> > > >> > > > > > they
> > > >> > > > > > > > > are not
> > > >> > > > > > > > > > > >> > exposed
> > > >> > > > > > > > > > > >> > > > > via
> > > >> > > > > > > > > > > >> > > > > >> any
> > > >> > > > > > > > > > > >> > > > > >> > > > > public
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > think
> > > >> the
> > > >> > > > > > question is,
> > > >> > > > > > > > > first
> > > >> > > > > > > > > > > >> > should we
> > > >> > > > > > > > > > > >> > > > > >> let
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > maintaining
> > > >> the
> > > >> > > > names
> > > >> > > > > > -> ids
> > > >> > > > > > > > > mapping
> > > >> > > > > > > > > > > >> > itself
> > > >> > > > > > > > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > fully
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > > >> existing
> > > >> > > > RPCs
> > > >> > > > > > and the
> > > >> > > > > > > > > new
> > > >> > > > > > > > > > > >> RPCs
> > > >> > > > > > > > > > > >> > with
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > ever
> > > >> consider
> > > >> > > > > > exposing
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > topic ids
> > > >> > > > > > > > > > > >> > > > > in
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > (both
> > > >> > > > > > > > > subscribe/assign, as
> > > >> > > > > > > > > > > >> well
> > > >> > > > > > > > > > > >> > as in
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > like
> > > >> topic
> > > >> > > > > > > > > > > >> > deletion-and-recreation).
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > David
> > > >> on all
> > > >> > > > other
> > > >> > > > > > minor
> > > >> > > > > > > > > > > >> questions
> > > >> > > > > > > > > > > >> > > > > except
> > > >> > > > > > > > > > > >> > > > > >> for
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > > >> question:
> > > >> > > > > > personally
> > > >> > > > > > > > > I think
> > > >> > > > > > > > > > > >> > it's not
> > > >> > > > > > > > > > > >> > > > > >> > > necessary
> > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > subscribe
> > > >> API with
> > > >> > > > > > Pattern,
> > > >> > > > > > > > > but
> > > >> > > > > > > > > > > >> > instead we
> > > >> > > > > > > > > > > >> > > > > >> still
> > > >> > > > > > > > > > > >> > > > > >> > > use
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> documenting
> > > >> that our
> > > >> > > > > > > > > subscription may
> > > >> > > > > > > > > > > >> be
> > > >> > > > > > > > > > > >> > > > > >> rejected by
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> incompatible
> > > >> case is a
> > > >> > > > > > very rare
> > > >> > > > > > > > > > > >> scenario
> > > >> > > > > > > > > > > >> > I felt
> > > >> > > > > > > > > > > >> > > > > >> > using
> > > >> > > > > > > > > > > >> > > > > >> > > an
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > based
> > > >> > > > > > subscription may
> > > >> > > > > > > > > be
> > > >> > > > > > > > > > > >> more
> > > >> > > > > > > > > > > >> > > > > >> vulnerable
> > > >> > > > > > > > > > > >> > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > various
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022
> > at
> > > >> 5:23 AM
> > > >> > > > > > David Jacot
> > > >> > > > > > > > > > > >> > > > > >> > > > >
> <djacot@confluent.io.invalid
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > > >> feedback. Let
> > > >> > > > me
> > > >> > > > > > answer
> > > >> > > > > > > > > your
> > > >> > > > > > > > > > > >> > questions
> > > >> > > > > > > > > > > >> > > > > >> > inline.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> > > >> premature to
> > > >> > > > > > talk about
> > > >> > > > > > > > > > > >> target
> > > >> > > > > > > > > > > >> > > > > versions
> > > >> > > > > > > > > > > >> > > > > >> for
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > > >> existing group
> > > >> > > > > > protocol.
> > > >> > > > > > > > > > > >> Unlike
> > > >> > > > > > > > > > > >> > KRaft,
> > > >> > > > > > > > > > > >> > > > > >> this
> > > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol
> > and
> > > >> hence
> > > >> > > > > > > > > deprecation/removal
> > > >> > > > > > > > > > > >> > will be
> > > >> > > > > > > > > > > >> > > > > >> > heavily
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > > >> applications
> > > >> > > > migrate
> > > >> > > > > > to
> > > >> > > > > > > > > the new
> > > >> > > > > > > > > > > >> > protocol.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense.
> I
> > will
> > > >> > > > remove it.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says
> we
> > > >> intend to
> > > >> > > > > > release
> > > >> > > > > > > > > this in
> > > >> > > > > > > > > > > >> > 4.x, but
> > > >> > > > > > > > > > > >> > > > > it
> > > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > > >> > > > > > > > > > > >> > > > > >> > > > > made
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> > that
> > > >> as a
> > > >> > > > way to
> > > >> > > > > > > > > estimate
> > > >> > > > > > > > > > > >> when
> > > >> > > > > > > > > > > >> > we'd
> > > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > > >> > > > > > > > > > > >> > > > > >> > > > > and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > protocol, I
> > > >> also
> > > >> > > > > > suggest
> > > >> > > > > > > > > removing
> > > >> > > > > > > > > > > >> > this part.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> > > >> reasoning. As
> > > >> > > > > > > > > explained, I
> > > >> > > > > > > > > > > >> plan
> > > >> > > > > > > > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > >> rewrite
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in
> Java
> > > >> while we
> > > >> > > > > > implement
> > > >> > > > > > > > > the new
> > > >> > > > > > > > > > > >> > protocol.
> > > >> > > > > > > > > > > >> > > > > >> This
> > > >> > > > > > > > > > > >> > > > > >> > > means
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> > will
> > > >> be
> > > >> > > > slightly
> > > >> > > > > > > > > different
> > > >> > > > > > > > > > > >> > (e.g.
> > > >> > > > > > > > > > > >> > > > > >> threading
> > > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I
> wanted
> > to
> > > >> tighten
> > > >> > > > the
> > > >> > > > > > > > > switch from
> > > >> > > > > > > > > > > >> > the old
> > > >> > > > > > > > > > > >> > > > > >> group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the
> > new
> > > >> group
> > > >> > > > > > coordinator
> > > >> > > > > > > > > to a
> > > >> > > > > > > > > > > >> > major
> > > >> > > > > > > > > > > >> > > > > >> release.
> > > >> > > > > > > > > > > >> > > > > >> > > The
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would
> > be to
> > > >> use a
> > > >> > > > flag
> > > >> > > > > > to do
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > switch
> > > >> > > > > > > > > > > >> > > > > >> instead
> > > >> > > > > > > > > > > >> > > > > >> > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> > upgrade.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to
> > flesh
> > > >> out the
> > > >> > > > > > details of
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > migration
> > > >> > > > > > > > > > > >> > > > > >> story.
> > > >> > > > > > > > > > > >> > > > > >> > > It
> > > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we
> > will
> > > >> support
> > > >> > > > > > online
> > > >> > > > > > > > > > > >> migrations.
> > > >> > > > > > > > > > > >> > Is
> > > >> > > > > > > > > > > >> > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > correct?
> > > >> > > > > > > > > > > >> > > > > >> > > > > We
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> > > >> detail. It
> > > >> > > > could
> > > >> > > > > > also
> > > >> > > > > > > > > be done
> > > >> > > > > > > > > > > >> > as a
> > > >> > > > > > > > > > > >> > > > > >> separate
> > > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > > >> > > > > > > > > > > >> > > > > >> > > > > if
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> support
> > > >> online
> > > >> > > > > > migrations
> > > >> > > > > > > > > for the
> > > >> > > > > > > > > > > >> > group.
> > > >> > > > > > > > > > > >> > > > > That
> > > >> > > > > > > > > > > >> > > > > >> > > means
> > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the
> > old
> > > >> protocol
> > > >> > > > > > will be
> > > >> > > > > > > > > able to
> > > >> > > > > > > > > > > >> > switch to
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> > explain
> > > >> how that
> > > >> > > > > > will work
> > > >> > > > > > > > > > > >> though.
> > > >> > > > > > > > > > > >> > It is
> > > >> > > > > > > > > > > >> > > > > >> > > basically a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster
> must
> > be
> > > >> > > > upgraded or
> > > >> > > > > > > > > rolled to a
> > > >> > > > > > > > > > > >> > software
> > > >> > > > > > > > > > > >> > > > > >> > > supporting
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > coordinator.
> > > >> Both the
> > > >> > > > > > old and
> > > >> > > > > > > > > the new
> > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > >> > > > > > > > > > > >> > > > > >> > > will
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > > >> protocol and
> > > >> > > > rely
> > > >> > > > > > on the
> > > >> > > > > > > > > same
> > > >> > > > > > > > > > > >> > persisted
> > > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> > together.
> > > >> This
> > > >> > > > point
> > > >> > > > > > is an
> > > >> > > > > > > > > > > >> offline
> > > >> > > > > > > > > > > >> > > > > >> migration.
> > > >> > > > > > > > > > > >> > > > > >> > We
> > > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > > >> because it
> > > >> > > > would
> > > >> > > > > > require
> > > >> > > > > > > > > > > >> > shutting down
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > current
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > > >> starting up the
> > > >> > > > > > new one
> > > >> > > > > > > > > and
> > > >> > > > > > > > > > > >> that
> > > >> > > > > > > > > > > >> > would
> > > >> > > > > > > > > > > >> > > > > >> cause
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > > >> metadata
> > > >> > > > > > version/IBP
> > > >> > > > > > > > > must be
> > > >> > > > > > > > > > > >> > upgraded
> > > >> > > > > > > > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> X
> > > >> > > > > > > > > > > >> > > > > >> > in
> > > >> > > > > > > > > > > >> > > > > >> > > > > order
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> > > >> protocol.
> > > >> > > > This
> > > >> > > > > > cannot
> > > >> > > > > > > > > be done
> > > >> > > > > > > > > > > >> > before
> > > >> > > > > > > > > > > >> > > > > 1)
> > > >> > > > > > > > > > > >> > > > > >> is
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because
> > the
> > > >> old
> > > >> > > > > > coordinator
> > > >> > > > > > > > > doesn't
> > > >> > > > > > > > > > > >> > support
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers
> > must be
> > > >> > > > upgraded
> > > >> > > > > > to a
> > > >> > > > > > > > > version
> > > >> > > > > > > > > > > >> > > > > supporting
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > online
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must
> have
> > > >> KIP-792).
> > > >> > > > If
> > > >> > > > > > the
> > > >> > > > > > > > > consumer
> > > >> > > > > > > > > > > >> is
> > > >> > > > > > > > > > > >> > > > > already
> > > >> > > > > > > > > > > >> > > > > >> > > there.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be
> > done at
> > > >> this
> > > >> > > > point.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers
> > must
> > > >> be rolled
> > > >> > > > > > with the
> > > >> > > > > > > > > > > >> feature
> > > >> > > > > > > > > > > >> > flag
> > > >> > > > > > > > > > > >> > > > > >> turned
> > > >> > > > > > > > > > > >> > > > > >> > > on.
> > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> > > >> automatically
> > > >> > > > > > > > > converted when
> > > >> > > > > > > > > > > >> > the first
> > > >> > > > > > > > > > > >> > > > > >> > > consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> > protocol
> > > >> joins
> > > >> > > > the
> > > >> > > > > > group.
> > > >> > > > > > > > > While
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > members
> > > >> > > > > > > > > > > >> > > > > >> > > using the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> > being
> > > >> > > > upgraded,
> > > >> > > > > > the old
> > > >> > > > > > > > > > > >> protocol
> > > >> > > > > > > > > > > >> > is
> > > >> > > > > > > > > > > >> > > > > >> proxied
> > > >> > > > > > > > > > > >> > > > > >> > > into
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all
> > of
> > > >> this in
> > > >> > > > the
> > > >> > > > > > KIP.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy
> > that we
> > > >> are
> > > >> > > > > > pushing the
> > > >> > > > > > > > > pattern
> > > >> > > > > > > > > > > >> > > > > >> subscriptions
> > > >> > > > > > > > > > > >> > > > > >> > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems
> like
> > > >> there
> > > >> > > > could be
> > > >> > > > > > some
> > > >> > > > > > > > > tricky
> > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism
> > for
> > > >> users to
> > > >> > > > > > detect
> > > >> > > > > > > > > that
> > > >> > > > > > > > > > > >> they
> > > >> > > > > > > > > > > >> > need to
> > > >> > > > > > > > > > > >> > > > > >> > update
> > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching
> > to
> > > >> the new
> > > >> > > > > > protocol?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am
> a
> > bit
> > > >> more
> > > >> > > > > > optimistic
> > > >> > > > > > > > > than
> > > >> > > > > > > > > > > >> you
> > > >> > > > > > > > > > > >> > on this
> > > >> > > > > > > > > > > >> > > > > >> > > point. I
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > > >> majority of
> > > >> > > > the
> > > >> > > > > > cases
> > > >> > > > > > > > > are
> > > >> > > > > > > > > > > >> simple
> > > >> > > > > > > > > > > >> > > > > regexes
> > > >> > > > > > > > > > > >> > > > > >> > which
> > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> > > >> engine. The
> > > >> > > > > > > > > coordinator will
> > > >> > > > > > > > > > > >> > verify
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > regex
> > > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> > consumer
> > > >> if the
> > > >> > > > > > regex is
> > > >> > > > > > > > > not
> > > >> > > > > > > > > > > >> valid.
> > > >> > > > > > > > > > > >> > > > > Coming
> > > >> > > > > > > > > > > >> > > > > >> > back
> > > >> > > > > > > > > > > >> > > > > >> > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in
> > the
> > > >> worst
> > > >> > > > case,
> > > >> > > > > > the
> > > >> > > > > > > > > first
> > > >> > > > > > > > > > > >> > upgraded
> > > >> > > > > > > > > > > >> > > > > >> > consumer
> > > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> > > >> rejected.
> > > >> > > > This
> > > >> > > > > > should
> > > >> > > > > > > > > be used
> > > >> > > > > > > > > > > >> > as the
> > > >> > > > > > > > > > > >> > > > > >> last
> > > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > customers to
> > > >> > > > validate
> > > >> > > > > > their
> > > >> > > > > > > > > regex
> > > >> > > > > > > > > > > >> > before
> > > >> > > > > > > > > > > >> > > > > >> > upgrading
> > > >> > > > > > > > > > > >> > > > > >> > > > > their
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to
> > test
> > > >> them with
> > > >> > > > > > another
> > > >> > > > > > > > > group.
> > > >> > > > > > > > > > > >> For
> > > >> > > > > > > > > > > >> > > > > >> instance,
> > > >> > > > > > > > > > > >> > > > > >> > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> > > >> pre-prod
> > > >> > > > > > environment.
> > > >> > > > > > > > > > > >> Another
> > > >> > > > > > > > > > > >> > way
> > > >> > > > > > > > > > > >> > > > > >> would be
> > > >> > > > > > > > > > > >> > > > > >> > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > >> consumer-group tool
> > > >> > > > to
> > > >> > > > > > > > > provide a
> > > >> > > > > > > > > > > >> regex
> > > >> > > > > > > > > > > >> > > > > >> validation
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would
> > this be
> > > >> > > > enough in
> > > >> > > > > > your
> > > >> > > > > > > > > > > >> opinion?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to
> the
> > last
> > > >> > > > question,
> > > >> > > > > > will
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> Java
> > > >> > > > > > > > > > > >> > client
> > > >> > > > > > > > > > > >> > > > > >> allow
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > > >> current regex
> > > >> > > > > > engine for
> > > >> > > > > > > > > > > >> > compatibility
> > > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > > >> > > > > > > > > > > >> > > > > >> > > > > For
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy
> to
> > > >> keep using
> > > >> > > > > > client
> > > >> > > > > > > > > based
> > > >> > > > > > > > > > > >> > regex at
> > > >> > > > > > > > > > > >> > > > > >> first
> > > >> > > > > > > > > > > >> > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations
> simple
> > and
> > > >> then
> > > >> > > > > > migrate to
> > > >> > > > > > > > > server
> > > >> > > > > > > > > > > >> > based
> > > >> > > > > > > > > > > >> > > > > >> regexes
> > > >> > > > > > > > > > > >> > > > > >> > > as a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> > point
> > > >> but I
> > > >> > > > am
> > > >> > > > > > > > > concerned that
> > > >> > > > > > > > > > > >> > this
> > > >> > > > > > > > > > > >> > > > > would
> > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > >> > > > > > > > > > > >> > > > > >> > > > > users
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay
> in
> > > >> this mode.
> > > >> > > > That
> > > >> > > > > > > > > would go
> > > >> > > > > > > > > > > >> > against our
> > > >> > > > > > > > > > > >> > > > > >> goal
> > > >> > > > > > > > > > > >> > > > > >> > > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> > client
> > > >> because
> > > >> > > > we
> > > >> > > > > > would
> > > >> > > > > > > > > have to
> > > >> > > > > > > > > > > >> > continue
> > > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on
> the
> > > >> client
> > > >> > > > side. I
> > > >> > > > > > would
> > > >> > > > > > > > > rather
> > > >> > > > > > > > > > > >> > not do
> > > >> > > > > > > > > > > >> > > > > >> this.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say
> > that
> > > >> the group
> > > >> > > > > > > > > coordinator
> > > >> > > > > > > > > > > >> will
> > > >> > > > > > > > > > > >> > be
> > > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> configurations
> > > >> and that
> > > >> > > > the
> > > >> > > > > > > > > > > >> configurations
> > > >> > > > > > > > > > > >> > will be
> > > >> > > > > > > > > > > >> > > > > >> > > deleted
> > > >> > > > > > > > > > > >> > > > > >> > > > > when
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is
> deleted.
> > > >> Will a
> > > >> > > > > > transition to
> > > >> > > > > > > > > DEAD
> > > >> > > > > > > > > > > >> > trigger
> > > >> > > > > > > > > > > >> > > > > >> > deletion
> > > >> > > > > > > > > > > >> > > > > >> > > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > >> > > > configurations
> > > >> > > > > > will be
> > > >> > > > > > > > > > > >> deleted
> > > >> > > > > > > > > > > >> > when
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > group is
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > > >> together.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the
> > choice to
> > > >> store
> > > >> > > > the
> > > >> > > > > > > > > configs in
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > group
> > > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list
> all
> > > >> cluster
> > > >> > > > > > configs and
> > > >> > > > > > > > > their
> > > >> > > > > > > > > > > >> > values?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so.
> > The
> > > >> group
> > > >> > > > > > > > > configurations are
> > > >> > > > > > > > > > > >> > overrides
> > > >> > > > > > > > > > > >> > > > > >> of
> > > >> > > > > > > > > > > >> > > > > >> > > cluster
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you
> > want to
> > > >> know
> > > >> > > > all
> > > >> > > > > > the
> > > >> > > > > > > > > overrides
> > > >> > > > > > > > > > > >> > though,
> > > >> > > > > > > > > > > >> > > > > you
> > > >> > > > > > > > > > > >> > > > > >> > > would
> > > >> > > > > > > > > > > >> > > > > >> > > > > have
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the
> group
> > > >> > > > coordinators.
> > > >> > > > > > You
> > > >> > > > > > > > > cannot
> > > >> > > > > > > > > > > >> > rely on
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > metadata
> > > >> > > > > > > > > > > >> > > > > >> > > > > log
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> > someone
> > > >> > > > configure a
> > > >> > > > > > group
> > > >> > > > > > > > > before
> > > >> > > > > > > > > > > >> > starting
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > > >> allowing the
> > > >> > > > > > explicit
> > > >> > > > > > > > > creation
> > > >> > > > > > > > > > > >> of
> > > >> > > > > > > > > > > >> > > > > groups?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> > > >> decoupled
> > > >> > > > from
> > > >> > > > > > the
> > > >> > > > > > > > > group
> > > >> > > > > > > > > > > >> > lifecycle.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group
> will
> > be
> > > >> > > > > > automatically
> > > >> > > > > > > > > created in
> > > >> > > > > > > > > > > >> > this
> > > >> > > > > > > > > > > >> > > > > case.
> > > >> > > > > > > > > > > >> > > > > >> > > However,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will
> be
> > > >> lost after
> > > >> > > > the
> > > >> > > > > > > > > retention
> > > >> > > > > > > > > > > >> > period of
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > >> > > > Consumer.subscribe
> > > >> > > > > > method
> > > >> > > > > > > > > for the
> > > >> > > > > > > > > > > >> > Java
> > > >> > > > > > > > > > > >> > > > > >> client
> > > >> > > > > > > > > > > >> > > > > >> > > still
> > > >> > > > > > > > > > > >> > > > > >> > > > > take
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >> `java.util.regex.Pattern` of
> > > >> > > > do
> > > >> > > > > > we
> > > >> > > > > > > > > have to
> > > >> > > > > > > > > > > >> > introduce
> > > >> > > > > > > > > > > >> > > > > an
> > > >> > > > > > > > > > > >> > > > > >> > > overload?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very
> group
> > > >> question. I
> > > >> > > > > > forgot
> > > >> > > > > > > > > about
> > > >> > > > > > > > > > > >> that
> > > >> > > > > > > > > > > >> > one.
> > > >> > > > > > > > > > > >> > > > > As
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> `java.util.regex.Pattern` is
> > > >> > > > not
> > > >> > > > > > fully
> > > >> > > > > > > > > > > >> compatible
> > > >> > > > > > > > > > > >> > with
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > engine
> > > >> > > > > > > > > > > >> > > > > >> > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> > > >> might be
> > > >> > > > better
> > > >> > > > > > to
> > > >> > > > > > > > > deprecate
> > > >> > > > > > > > > > > >> > it and
> > > >> > > > > > > > > > > >> > > > > >> use an
> > > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a
> > string.
> > > >> We would
> > > >> > > > > > rely on
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> server
> > > >> > > > > > > > > > > >> > side
> > > >> > > > > > > > > > > >> > > > > >> > > validation.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > migration, I
> > > >> think
> > > >> > > > that
> > > >> > > > > > we
> > > >> > > > > > > > > could
> > > >> > > > > > > > > > > >> still
> > > >> > > > > > > > > > > >> > try to
> > > >> > > > > > > > > > > >> > > > > >> > > toString
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it.
> > That
> > > >> should
> > > >> > > > > > work, I
> > > >> > > > > > > > > think, in
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> majority
> > > >> > > > > > > > > > > >> > > > > >> > > of the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> > > >> Justine
> > > >> > > > that we
> > > >> > > > > > > > > should be
> > > >> > > > > > > > > > > >> > clearer
> > > >> > > > > > > > > > > >> > > > > about
> > > >> > > > > > > > > > > >> > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > >> > > > IBP/metadata.version
> > > >> > > > > > from the
> > > >> > > > > > > > > > > >> feature
> > > >> > > > > > > > > > > >> > flag.
> > > >> > > > > > > > > > > >> > > > > >> Maybe
> > > >> > > > > > > > > > > >> > > > > >> > > we
> > > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > > >> default for
> > > >> > > > the
> > > >> > > > > > feature
> > > >> > > > > > > > > flag
> > > >> > > > > > > > > > > >> to
> > > >> > > > > > > > > > > >> > true
> > > >> > > > > > > > > > > >> > > > > >> based
> > > >> > > > > > > > > > > >> > > > > >> > on
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> > once
> > > >> we
> > > >> > > > want to
> > > >> > > > > > make
> > > >> > > > > > > > > it the
> > > >> > > > > > > > > > > >> > default.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> > that
> > > >> feature
> > > >> > > > > > flag
> > > >> > > > > > > > > mainly
> > > >> > > > > > > > > > > >> > during the
> > > >> > > > > > > > > > > >> > > > > >> > > development
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should
> not
> > have
> > > >> > > > mentioned
> > > >> > > > > > it, I
> > > >> > > > > > > > > think,
> > > >> > > > > > > > > > > >> > because
> > > >> > > > > > > > > > > >> > > > > we
> > > >> > > > > > > > > > > >> > > > > >> > > could
> > > >> > > > > > > > > > > >> > > > > >> > > > > use
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config
> > for
> > > >> it.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> > > >> protocol APIs
> > > >> > > > > > don't
> > > >> > > > > > > > > mention
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > required
> > > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that
> > for
> > > >> > > > consistency.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit
> > odd
> > > >> that
> > > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > >> > > > > > > > > > > >> > > > > requires
> > > >> > > > > > > > > > > >> > > > > >> > > "Read
> > > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems
> > to do
> > > >> more
> > > >> > > > than
> > > >> > > > > > > > > reading.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is
> > how the
> > > >> > > > current
> > > >> > > > > > > > > protocol works
> > > >> > > > > > > > > > > >> > though.
> > > >> > > > > > > > > > > >> > > > > We
> > > >> > > > > > > > > > > >> > > > > >> > only
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read
> > Group" to
> > > >> join a
> > > >> > > > > > group. We
> > > >> > > > > > > > > could
> > > >> > > > > > > > > > > >> > consider
> > > >> > > > > > > > > > > >> > > > > >> > > changing
> > > >> > > > > > > > > > > >> > > > > >> > > > > this
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure
> > that
> > > >> it is
> > > >> > > > worth
> > > >> > > > > > it.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> > > >> recreation
> > > >> > > > > > handled by
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > consumer
> > > >> > > > > > > > > > > >> > > > > with
> > > >> > > > > > > > > > > >> > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It
> > would be
> > > >> good to
> > > >> > > > > > have a
> > > >> > > > > > > > > section
> > > >> > > > > > > > > > > >> on
> > > >> > > > > > > > > > > >> > this.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> > protocol
> > > >> > > > > > perspective, the
> > > >> > > > > > > > > new
> > > >> > > > > > > > > > > >> topic
> > > >> > > > > > > > > > > >> > will
> > > >> > > > > > > > > > > >> > > > > >> have a
> > > >> > > > > > > > > > > >> > > > > >> > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it
> will
> > > >> treat it
> > > >> > > > like a
> > > >> > > > > > > > > topic with
> > > >> > > > > > > > > > > >> a
> > > >> > > > > > > > > > > >> > > > > different
> > > >> > > > > > > > > > > >> > > > > >> > > name.
> > > >> > > > > > > > > > > >> > > > > >> > > > > The
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that
> > the
> > > >> > > > fetch/commit
> > > >> > > > > > > > > offsets
> > > >> > > > > > > > > > > >> APIs
> > > >> > > > > > > > > > > >> > do not
> > > >> > > > > > > > > > > >> > > > > >> > support
> > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the
> consumer
> > > >> would
> > > >> > > > reuse the
> > > >> > > > > > > > > offsets
> > > >> > > > > > > > > > > >> based
> > > >> > > > > > > > > > > >> > on the
> > > >> > > > > > > > > > > >> > > > > >> > same.
> > > >> > > > > > > > > > > >> > > > > >> > > I
> > > >> > > > > > > > > > > >> > > > > >> > > > > think
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should
> > update
> > > >> those
> > > >> > > > APIs
> > > >> > > > > > as well
> > > >> > > > > > > > > in
> > > >> > > > > > > > > > > >> order
> > > >> > > > > > > > > > > >> > to be
> > > >> > > > > > > > > > > >> > > > > >> > > consistent
> > > >> > > > > > > > > > > >> > > > > >> > > > > end
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> > > >> strengthen
> > > >> > > > the
> > > >> > > > > > > > > semantics of
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> consumer.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> > mentions
> > > >> we will
> > > >> > > > > > write the
> > > >> > > > > > > > > new
> > > >> > > > > > > > > > > >> > coordinator
> > > >> > > > > > > > > > > >> > > > > >> in
> > > >> > > > > > > > > > > >> > > > > >> > > Java.
> > > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > >> implementation
> > > >> > > > > > detail, do
> > > >> > > > > > > > > we plan
> > > >> > > > > > > > > > > >> to
> > > >> > > > > > > > > > > >> > have a
> > > >> > > > > > > > > > > >> > > > > >> new
> > > >> > > > > > > > > > > >> > > > > >> > > gradle
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> > > >> scalability
> > > >> > > > > > goal when
> > > >> > > > > > > > > it
> > > >> > > > > > > > > > > >> comes
> > > >> > > > > > > > > > > >> > to how
> > > >> > > > > > > > > > > >> > > > > >> many
> > > >> > > > > > > > > > > >> > > > > >> > > > > members
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol
> can
> > > >> support?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have
> > numbers
> > > >> at the
> > > >> > > > > > moment. The
> > > >> > > > > > > > > > > >> protocol
> > > >> > > > > > > > > > > >> > should
> > > >> > > > > > > > > > > >> > > > > >> > > support
> > > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per
> > group.
> > > >> We will
> > > >> > > > > > measure
> > > >> > > > > > > > > this when
> > > >> > > > > > > > > > > >> > we have
> > > >> > > > > > > > > > > >> > > > > a
> > > >> > > > > > > > > > > >> > > > > >> > first
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation.
> Note
> > > >> that we
> > > >> > > > might
> > > >> > > > > > have
> > > >> > > > > > > > > other
> > > >> > > > > > > > > > > >> > > > > bottlenecks
> > > >> > > > > > > > > > > >> > > > > >> > down
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> > > >> commits).
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > consider
> > > >> having
> > > >> > > > > > > > > SubscribedTopidIds
> > > >> > > > > > > > > > > >> > instead
> > > >> > > > > > > > > > > >> > > > > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > SubscribedTopicNames
> > > >> in
> > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > >> > > > > > > > > > > >> > > > > >> Is
> > > >> > > > > > > > > > > >> > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> > > >> resolve the
> > > >> > > > > > regex on
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > server, we
> > > >> > > > > > > > > > > >> > > > > >> can do
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > same
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name?
> > The
> > > >> > > > difference
> > > >> > > > > > is that
> > > >> > > > > > > > > > > >> sending
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> regex is
> > > >> > > > > > > > > > > >> > > > > >> > > more
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending
> > the
> > > >> topic
> > > >> > > > names
> > > >> > > > > > is less
> > > >> > > > > > > > > > > >> > efficient.
> > > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> > easier
> > > >> to
> > > >> > > > handle
> > > >> > > > > > if we
> > > >> > > > > > > > > have
> > > >> > > > > > > > > > > >> topic
> > > >> > > > > > > > > > > >> > ids.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > > >> consolidate the
> > > >> > > > > > metadata
> > > >> > > > > > > > > lookup
> > > >> > > > > > > > > > > >> > on the
> > > >> > > > > > > > > > > >> > > > > >> server
> > > >> > > > > > > > > > > >> > > > > >> > > for
> > > >> > > > > > > > > > > >> > > > > >> > > > > both
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do
> agree
> > > >> with your
> > > >> > > > > > point. As
> > > >> > > > > > > > > a
> > > >> > > > > > > > > > > >> second
> > > >> > > > > > > > > > > >> > > > > though,
> > > >> > > > > > > > > > > >> > > > > >> > using
> > > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better
> > here
> > > >> for the
> > > >> > > > > > delete and
> > > >> > > > > > > > > > > >> > recreation
> > > >> > > > > > > > > > > >> > > > > case.
> > > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we
> may
> > > >> allow
> > > >> > > > users to
> > > >> > > > > > > > > subscribe
> > > >> > > > > > > > > > > >> with
> > > >> > > > > > > > > > > >> > topic
> > > >> > > > > > > > > > > >> > > > > >> ids
> > > >> > > > > > > > > > > >> > > > > >> > in
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future because
> that
> > is
> > > >> the only
> > > >> > > > > > way to be
> > > >> > > > > > > > > > > >> really
> > > >> > > > > > > > > > > >> > robust
> > > >> > > > > > > > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > > topic
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12,
> > 2022 at
> > > >> 1:38 PM
> > > >> > > > > > David
> > > >> > > > > > > > > Jacot <
> > > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> > > >> comments.
> > > >> > > > Please
> > > >> > > > > > find
> > > >> > > > > > > > > my
> > > >> > > > > > > > > > > >> answers
> > > >> > > > > > > > > > > >> > > > > below.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > > >> protocol
> > > >> > > > relies on
> > > >> > > > > > > > > topic IDs
> > > >> > > > > > > > > > > >> > with the
> > > >> > > > > > > > > > > >> > > > > >> > > exception
> > > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names
> based
> > in
> > > >> the
> > > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > >> > > > > > > > > > > >> > > > > >> I
> > > >> > > > > > > > > > > >> > > > > >> > am
> > > >> > > > > > > > > > > >> > > > > >> > > not
> > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> > names
> > > >> is the
> > > >> > > > > > right call
> > > >> > > > > > > > > here.
> > > >> > > > > > > > > > > >> I
> > > >> > > > > > > > > > > >> > need to
> > > >> > > > > > > > > > > >> > > > > >> > think
> > > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > > >> Obviously, the
> > > >> > > > > > KIP does
> > > >> > > > > > > > > not
> > > >> > > > > > > > > > > >> > change the
> > > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to
> > use
> > > >> topic
> > > >> > > > IDs.
> > > >> > > > > > This
> > > >> > > > > > > > > may be
> > > >> > > > > > > > > > > >> > something
> > > >> > > > > > > > > > > >> > > > > >> that
> > > >> > > > > > > > > > > >> > > > > >> > we
> > > >> > > > > > > > > > > >> > > > > >> > > > > should
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though
> as
> > it
> > > >> would
> > > >> > > > give
> > > >> > > > > > better
> > > >> > > > > > > > > > > >> overall
> > > >> > > > > > > > > > > >> > > > > >> guarantee in
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right.
> I
> > > >> think that
> > > >> > > > I
> > > >> > > > > > should
> > > >> > > > > > > > > not
> > > >> > > > > > > > > > > >> have
> > > >> > > > > > > > > > > >> > > > > mentioned
> > > >> > > > > > > > > > > >> > > > > >> > this
> > > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will
> remove
> > > >> it. We can
> > > >> > > > > > use an
> > > >> > > > > > > > > internal
> > > >> > > > > > > > > > > >> > > > > >> configuration
> > > >> > > > > > > > > > > >> > > > > >> > > while
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > > >> feature.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> > types
> > > >> will be
> > > >> > > > > > > > > supported. The
> > > >> > > > > > > > > > > >> > change is
> > > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement
> > is
> > > >> that the
> > > >> > > > > > cluster
> > > >> > > > > > > > > uses
> > > >> > > > > > > > > > > >> topic
> > > >> > > > > > > > > > > >> > IDs.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11,
> > 2022
> > > >> at 9:53
> > > >> > > > PM
> > > >> > > > > > > > > Guozhang
> > > >> > > > > > > > > > > >> Wang <
> > > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> > > >> feedback.
> > > >> > > > Here
> > > >> > > > > > are
> > > >> > > > > > > > > some
> > > >> > > > > > > > > > > >> replies
> > > >> > > > > > > > > > > >> > > > > inlined
> > > >> > > > > > > > > > > >> > > > > >> > > below:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> > 2022
> > > >> at
> > > >> > > > 2:53 AM
> > > >> > > > > > > > > Ismael Juma
> > > >> > > > > > > > > > > >> <
> > > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for
> the
> > > >> KIP. This
> > > >> > > > has
> > > >> > > > > > the
> > > >> > > > > > > > > > > >> potential
> > > >> > > > > > > > > > > >> > to be a
> > > >> > > > > > > > > > > >> > > > > >> > great
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > >> > > > questions/comments:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think
> > it's
> > > >> > > > premature to
> > > >> > > > > > talk
> > > >> > > > > > > > > about
> > > >> > > > > > > > > > > >> > target
> > > >> > > > > > > > > > > >> > > > > >> versions
> > > >> > > > > > > > > > > >> > > > > >> > > for
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of
> the
> > > >> existing
> > > >> > > > group
> > > >> > > > > > > > > protocol.
> > > >> > > > > > > > > > > >> > Unlike
> > > >> > > > > > > > > > > >> > > > > >> KRaft,
> > > >> > > > > > > > > > > >> > > > > >> > > this
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > protocol
> > > >> and hence
> > > >> > > > > > > > > > > >> > deprecation/removal will
> > > >> > > > > > > > > > > >> > > > > >> be
> > > >> > > > > > > > > > > >> > > > > >> > > heavily
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> > > >> applications
> > > >> > > > > > migrate
> > > >> > > > > > > > > to the
> > > >> > > > > > > > > > > >> new
> > > >> > > > > > > > > > > >> > > > > >> protocol.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree
> > with
> > > >> you. I
> > > >> > > > think
> > > >> > > > > > we can
> > > >> > > > > > > > > > > >> remove
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> proposed
> > > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> `Compatibility,
> > > >> > > > Deprecation,
> > > >> > > > > > and
> > > >> > > > > > > > > Migration
> > > >> > > > > > > > > > > >> > Plan` and
> > > >> > > > > > > > > > > >> > > > > >> > > instead
> > > >> > > > > > > > > > > >> > > > > >> > > > > just
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> > decide
> > > >> in the
> > > >> > > > > > future
> > > >> > > > > > > > > about
> > > >> > > > > > > > > > > >> when
> > > >> > > > > > > > > > > >> > we
> > > >> > > > > > > > > > > >> > > > > would
> > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > > >> behaviors.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP
> > says we
> > > >> > > > intend to
> > > >> > > > > > > > > release this
> > > >> > > > > > > > > > > >> > in 4.x,
> > > >> > > > > > > > > > > >> > > > > >> but
> > > >> > > > > > > > > > > >> > > > > >> > it
> > > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> > added
> > > >> that as
> > > >> > > > a
> > > >> > > > > > way to
> > > >> > > > > > > > > > > >> estimate
> > > >> > > > > > > > > > > >> > when
> > > >> > > > > > > > > > > >> > > > > we'd
> > > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > > >> protocol, I
> > > >> > > > also
> > > >> > > > > > suggest
> > > >> > > > > > > > > > > >> removing
> > > >> > > > > > > > > > > >> > this
> > > >> > > > > > > > > > > >> > > > > >> part.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's
> > not
> > > >> > > > specifically
> > > >> > > > > > > > > related to
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> > more
> > > >> for
> > > >> > > > client
> > > >> > > > > > > > > upgrades.
> > > >> > > > > > > > > > > >> I.e.
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation
> > may
> > > >> be done
> > > >> > > > > > first,
> > > >> > > > > > > > > and then
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > client
> > > >> > > > > > > > > > > >> > > > > >> > side,
> > > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it
> as
> > > >> "released"
> > > >> > > > by
> > > >> > > > > > the
> > > >> > > > > > > > > time
> > > >> > > > > > > > > > > >> clients
> > > >> > > > > > > > > > > >> > > > > >> > > implementations
> > > >> > > > > > > > > > > >> > > > > >> > > > > are
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> > > >> enable the
> > > >> > > > > > feature the
> > > >> > > > > > > > > > > >> clients
> > > >> > > > > > > > > > > >> > need to
> > > >> > > > > > > > > > > >> > > > > >> > first
> > > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with
> a
> > > >> rolling
> > > >> > > > bounce
> > > >> > > > > > and
> > > >> > > > > > > > > then set
> > > >> > > > > > > > > > > >> > the flag
> > > >> > > > > > > > > > > >> > > > > >> > with a
> > > >> > > > > > > > > > > >> > > > > >> > > > > second
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and
> > hence
> > > >> we feel
> > > >> > > > it's
> > > >> > > > > > > > > better to be
> > > >> > > > > > > > > > > >> > released
> > > >> > > > > > > > > > > >> > > > > >> in a
> > > >> > > > > > > > > > > >> > > > > >> > > major
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need
> to
> > > >> flesh out
> > > >> > > > the
> > > >> > > > > > > > > details of
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > migration
> > > >> > > > > > > > > > > >> > > > > >> > > story.
> > > >> > > > > > > > > > > >> > > > > >> > > > > It
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying
> > we
> > > >> will
> > > >> > > > support
> > > >> > > > > > online
> > > >> > > > > > > > > > > >> > migrations. Is
> > > >> > > > > > > > > > > >> > > > > >> that
> > > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this
> > in
> > > >> detail.
> > > >> > > > It
> > > >> > > > > > could
> > > >> > > > > > > > > also be
> > > >> > > > > > > > > > > >> > done as a
> > > >> > > > > > > > > > > >> > > > > >> > > separate
> > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > that's
> > > >> the
> > > >> > > > part we
> > > >> > > > > > can
> > > >> > > > > > > > > be more
> > > >> > > > > > > > > > > >> > concrete
> > > >> > > > > > > > > > > >> > > > > >> about
> > > >> > > > > > > > > > > >> > > > > >> > > for
> > > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is
> related
> > to
> > > >> your
> > > >> > > > > > question 2)
> > > >> > > > > > > > > above).
> > > >> > > > > > > > > > > >> > We will
> > > >> > > > > > > > > > > >> > > > > >> work
> > > >> > > > > > > > > > > >> > > > > >> > on
> > > >> > > > > > > > > > > >> > > > > >> > > > > making
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > > >> parallel as we
> > > >> > > > > > solicit
> > > >> > > > > > > > > more
> > > >> > > > > > > > > > > >> > feedback.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am
> happy
> > > >> that we are
> > > >> > > > > > pushing
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > pattern
> > > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > >> > > > > > > > > > > >> > > > > >> > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> > like
> > > >> there
> > > >> > > > > > could be
> > > >> > > > > > > > > some
> > > >> > > > > > > > > > > >> tricky
> > > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > mechanism
> > > >> for
> > > >> > > > users to
> > > >> > > > > > > > > detect that
> > > >> > > > > > > > > > > >> > they
> > > >> > > > > > > > > > > >> > > > > need
> > > >> > > > > > > > > > > >> > > > > >> to
> > > >> > > > > > > > > > > >> > > > > >> > > update
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > switching
> > > >> to the
> > > >> > > > new
> > > >> > > > > > > > > protocol?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> > need
> > > >> some
> > > >> > > > > > tooling for
> > > >> > > > > > > > > > > >> non-java
> > > >> > > > > > > > > > > >> > client
> > > >> > > > > > > > > > > >> > > > > >> > users
> > > >> > > > > > > > > > > >> > > > > >> > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> > > >> client before
> > > >> > > > > > > > > switching to
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > new
> > > >> > > > > > > > > > > >> > > > > >> > protocol.
> > > >> > > > > > > > > > > >> > > > > >> > > I
> > > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea
> on
> > > >> top of my
> > > >> > > > head
> > > >> > > > > > > > > though,
> > > >> > > > > > > > > > > >> maybe
> > > >> > > > > > > > > > > >> > others
> > > >> > > > > > > > > > > >> > > > > >> like
> > > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related
> to
> > the
> > > >> last
> > > >> > > > > > question,
> > > >> > > > > > > > > will the
> > > >> > > > > > > > > > > >> > Java
> > > >> > > > > > > > > > > >> > > > > >> client
> > > >> > > > > > > > > > > >> > > > > >> > > allow
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with
> the
> > > >> current
> > > >> > > > regex
> > > >> > > > > > > > > engine for
> > > >> > > > > > > > > > > >> > > > > >> compatibility
> > > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be
> > handy
> > > >> to keep
> > > >> > > > using
> > > >> > > > > > > > > client
> > > >> > > > > > > > > > > >> based
> > > >> > > > > > > > > > > >> > regex
> > > >> > > > > > > > > > > >> > > > > at
> > > >> > > > > > > > > > > >> > > > > >> > > first to
> > > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> > simple
> > > >> and
> > > >> > > > then
> > > >> > > > > > migrate
> > > >> > > > > > > > > to
> > > >> > > > > > > > > > > >> > server based
> > > >> > > > > > > > > > > >> > > > > >> > > regexes
> > > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I
> have
> > not
> > > >> thought
> > > >> > > > > > about
> > > >> > > > > > > > > that for
> > > >> > > > > > > > > > > >> > java
> > > >> > > > > > > > > > > >> > > > > >> clients,
> > > >> > > > > > > > > > > >> > > > > >> > > and
> > > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What
> kind
> > of
> > > >> > > > > > compatibility
> > > >> > > > > > > > > issues do
> > > >> > > > > > > > > > > >> > you have
> > > >> > > > > > > > > > > >> > > > > in
> > > >> > > > > > > > > > > >> > > > > >> > > mind?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we
> say
> > > >> that the
> > > >> > > > group
> > > >> > > > > > > > > coordinator
> > > >> > > > > > > > > > > >> > will be
> > > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > >> configurations and
> > > >> > > > that
> > > >> > > > > > the
> > > >> > > > > > > > > > > >> > configurations
> > > >> > > > > > > > > > > >> > > > > will
> > > >> > > > > > > > > > > >> > > > > >> be
> > > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> > deleted.
> > > >> Will a
> > > >> > > > > > > > > transition to
> > > >> > > > > > > > > > > >> DEAD
> > > >> > > > > > > > > > > >> > > > > trigger
> > > >> > > > > > > > > > > >> > > > > >> > > deletion
> > > >> > > > > > > > > > > >> > > > > >> > > > > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > configurations?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> > DEAD
> > > >> state
> > > >> > > > is an
> > > >> > > > > > > > > ending
> > > >> > > > > > > > > > > >> state
> > > >> > > > > > > > > > > >> > (we
> > > >> > > > > > > > > > > >> > > > > would
> > > >> > > > > > > > > > > >> > > > > >> > only
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> > > >> group is
> > > >> > > > EMPTY
> > > >> > > > > > and
> > > >> > > > > > > > > also all
> > > >> > > > > > > > > > > >> of
> > > >> > > > > > > > > > > >> > its
> > > >> > > > > > > > > > > >> > > > > >> > metadata
> > > >> > > > > > > > > > > >> > > > > >> > > are
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited
> > to
> > > >> DEAD this
> > > >> > > > > > group
> > > >> > > > > > > > > would
> > > >> > > > > > > > > > > >> never
> > > >> > > > > > > > > > > >> > be
> > > >> > > > > > > > > > > >> > > > > >> revived.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> > > >> choice to
> > > >> > > > store
> > > >> > > > > > the
> > > >> > > > > > > > > configs
> > > >> > > > > > > > > > > >> in
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > group
> > > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to
> > list all
> > > >> > > > cluster
> > > >> > > > > > configs
> > > >> > > > > > > > > and
> > > >> > > > > > > > > > > >> their
> > > >> > > > > > > > > > > >> > > > > values?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> > > >> question,
> > > >> > > > and our
> > > >> > > > > > > > > thoughts
> > > >> > > > > > > > > > > >> are
> > > >> > > > > > > > > > > >> > that
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > so-called
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> configurations"
> > are
> > > >> > > > overrides
> > > >> > > > > > of the
> > > >> > > > > > > > > > > >> > cluster-level
> > > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> > > >> group so
> > > >> > > > when an
> > > >> > > > > > > > > admin list
> > > >> > > > > > > > > > > >> > cluster
> > > >> > > > > > > > > > > >> > > > > >> > configs
> > > >> > > > > > > > > > > >> > > > > >> > > it's
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> > > >> cluster-level
> > > >> > > > > > > > > "defaults", not
> > > >> > > > > > > > > > > >> > showing
> > > >> > > > > > > > > > > >> > > > > >> any
> > > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> customizations.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> > > >> someone
> > > >> > > > > > configure a
> > > >> > > > > > > > > group
> > > >> > > > > > > > > > > >> > before
> > > >> > > > > > > > > > > >> > > > > >> starting
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we
> considered
> > > >> allowing
> > > >> > > > the
> > > >> > > > > > explicit
> > > >> > > > > > > > > > > >> > creation of
> > > >> > > > > > > > > > > >> > > > > >> groups?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs
> could
> > be
> > > >> > > > decoupled
> > > >> > > > > > from
> > > >> > > > > > > > > the group
> > > >> > > > > > > > > > > >> > > > > lifecycle.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs
> can
> > be
> > > >> created
> > > >> > > > > > before
> > > >> > > > > > > > > the group
> > > >> > > > > > > > > > > >> > itself
> > > >> > > > > > > > > > > >> > > > > as
> > > >> > > > > > > > > > > >> > > > > >> an
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course,
> > this
> > > >> > > > requires
> > > >> > > > > > the
> > > >> > > > > > > > > > > >> corresponding
> > > >> > > > > > > > > > > >> > > > > >> request to
> > > >> > > > > > > > > > > >> > > > > >> > > be
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > coordinator
> > > >> based on
> > > >> > > > the
> > > >> > > > > > group
> > > >> > > > > > > > > id ---
> > > >> > > > > > > > > > > >> > the only
> > > >> > > > > > > > > > > >> > > > > >> > thing
> > > >> > > > > > > > > > > >> > > > > >> > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> > > >> itself is
> > > >> > > > gone
> > > >> > > > > > we
> > > >> > > > > > > > > also check
> > > >> > > > > > > > > > > >> > if there
> > > >> > > > > > > > > > > >> > > > > >> are
> > > >> > > > > > > > > > > >> > > > > >> > > any
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > related to
> > > >> that
> > > >> > > > group
> > > >> > > > > > and
> > > >> > > > > > > > > delete
> > > >> > > > > > > > > > > >> as
> > > >> > > > > > > > > > > >> > well.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly
> this
> > > >> indeed
> > > >> > > > > > introduces an
> > > >> > > > > > > > > > > >> > asymmetry on
> > > >> > > > > > > > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > creation
> > > >> > > > > > > > > > > >> > > > > >> > > > > /
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of
> > the
> > > >> config
> > > >> > > > > > entities,
> > > >> > > > > > > > > and we
> > > >> > > > > > > > > > > >> > would like
> > > >> > > > > > > > > > > >> > > > > >> to
> > > >> > > > > > > > > > > >> > > > > >> > > hear
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > whether we
> > > >> should
> > > >> > > > aim
> > > >> > > > > > for
> > > >> > > > > > > > > symmetry
> > > >> > > > > > > > > > > >> > i.e.
> > > >> > > > > > > > > > > >> > > > > >> totally
> > > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and
> > hence
> > > >> not
> > > >> > > > delete
> > > >> > > > > > them at
> > > >> > > > > > > > > all
> > > >> > > > > > > > > > > >> when
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> group
> > > >> > > > > > > > > > > >> > > > > >> > is
> > > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> explicit
> > > >> deletion
> > > >> > > > > > operations
> > > >> > > > > > > > > by
> > > >> > > > > > > > > > > >> > themselves.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > >> > > > > > Consumer.subscribe
> > > >> > > > > > > > > method for
> > > >> > > > > > > > > > > >> > the Java
> > > >> > > > > > > > > > > >> > > > > >> > client
> > > >> > > > > > > > > > > >> > > > > >> > > > > still
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > `java.util.regex.Pattern` of
> > > >> > > > > > do we
> > > >> > > > > > > > > have
> > > >> > > > > > > > > > > >> to
> > > >> > > > > > > > > > > >> > > > > >> introduce an
> > > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do
> > not
> > > >> need to
> > > >> > > > > > introduce
> > > >> > > > > > > > > an
> > > >> > > > > > > > > > > >> > overload, but
> > > >> > > > > > > > > > > >> > > > > >> I'm
> > > >> > > > > > > > > > > >> > > > > >> > > all
> > > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > > >> compatibility
> > > >> > > > > > issues
> > > >> > > > > > > > > that we
> > > >> > > > > > > > > > > >> may
> > > >> > > > > > > > > > > >> > > > > overlook.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree
> > with
> > > >> Justine
> > > >> > > > > > that we
> > > >> > > > > > > > > should
> > > >> > > > > > > > > > > >> be
> > > >> > > > > > > > > > > >> > clearer
> > > >> > > > > > > > > > > >> > > > > >> > about
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > >> > > > > > IBP/metadata.version
> > > >> > > > > > > > > from the
> > > >> > > > > > > > > > > >> > feature
> > > >> > > > > > > > > > > >> > > > > >> flag.
> > > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > > >> > > > > > > > > > > >> > > > > >> > > > > we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch
> the
> > > >> default
> > > >> > > > for
> > > >> > > > > > the
> > > >> > > > > > > > > feature
> > > >> > > > > > > > > > > >> flag
> > > >> > > > > > > > > > > >> > to
> > > >> > > > > > > > > > > >> > > > > true
> > > >> > > > > > > > > > > >> > > > > >> > > based on
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > metadata.version
> > > >> once we
> > > >> > > > > > want to
> > > >> > > > > > > > > make it
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> default.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of
> the
> > > >> protocol
> > > >> > > > APIs
> > > >> > > > > > don't
> > > >> > > > > > > > > mention
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > >> required
> > > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > >> > > > > > > > > > > >> > > > > >> > > > > it
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add
> > that
> > > >> for
> > > >> > > > > > consistency.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > > >> certainly do
> > > >> > > > that.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a
> > bit
> > > >> odd that
> > > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > >> > > > > > > > > > > >> > > > > >> > requires
> > > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it
> > seems
> > > >> to do
> > > >> > > > more
> > > >> > > > > > than
> > > >> > > > > > > > > reading.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> > thought
> > > >> myself
> > > >> > > > as
> > > >> > > > > > well,
> > > >> > > > > > > > > but in
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > end we
> > > >> > > > > > > > > > > >> > > > > >> > could
> > > >> > > > > > > > > > > >> > > > > >> > > not
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > alternative:
> > > >> adding
> > > >> > > > > > Write
> > > >> > > > > > > > > Group
> > > >> > > > > > > > > > > >> seems
> > > >> > > > > > > > > > > >> > an
> > > >> > > > > > > > > > > >> > > > > >> overkill
> > > >> > > > > > > > > > > >> > > > > >> > > here
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > > >> elsewhere (we
> > > >> > > > only
> > > >> > > > > > have
> > > >> > > > > > > > > Read /
> > > >> > > > > > > > > > > >> > Delete
> > > >> > > > > > > > > > > >> > > > > and
> > > >> > > > > > > > > > > >> > > > > >> > > Describe
> > > >> > > > > > > > > > > >> > > > > >> > > > > on
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would
> > like to
> > > >> hear
> > > >> > > > others
> > > >> > > > > > > > > thoughts.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> > topic
> > > >> > > > recreation
> > > >> > > > > > > > > handled by
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > > consumer
> > > >> > > > > > > > > > > >> > > > > >> > with
> > > >> > > > > > > > > > > >> > > > > >> > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> > > >> would be
> > > >> > > > good
> > > >> > > > > > to have
> > > >> > > > > > > > > a
> > > >> > > > > > > > > > > >> > section on
> > > >> > > > > > > > > > > >> > > > > >> this.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with
> > regex
> > > >> > > > > > subscription
> > > >> > > > > > > > > right? Yes
> > > >> > > > > > > > > > > >> > we can
> > > >> > > > > > > > > > > >> > > > > >> add a
> > > >> > > > > > > > > > > >> > > > > >> > > > > section
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> > basically
> > > >> the
> > > >> > > > idea
> > > >> > > > > > is that
> > > >> > > > > > > > > > > >> consumer
> > > >> > > > > > > > > > > >> > would
> > > >> > > > > > > > > > > >> > > > > be
> > > >> > > > > > > > > > > >> > > > > >> > > totally
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > protocol as
> > > >> it's
> > > >> > > > > > handled all
> > > >> > > > > > > > > by the
> > > >> > > > > > > > > > > >> > brokers.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> > > >> mentions we
> > > >> > > > will
> > > >> > > > > > write
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> new
> > > >> > > > > > > > > > > >> > > > > >> coordinator
> > > >> > > > > > > > > > > >> > > > > >> > in
> > > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > > >> implementation
> > > >> > > > > > detail,
> > > >> > > > > > > > > do we
> > > >> > > > > > > > > > > >> > plan to
> > > >> > > > > > > > > > > >> > > > > >> have a
> > > >> > > > > > > > > > > >> > > > > >> > > new
> > > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> > thought
> > > >> about
> > > >> > > > > > that. But
> > > >> > > > > > > > > I think
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > answer
> > > >> > > > > > > > > > > >> > > > > >> > > should
> > > >> > > > > > > > > > > >> > > > > >> > > > > be
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we
> > have a
> > > >> > > > scalability
> > > >> > > > > > goal
> > > >> > > > > > > > > when it
> > > >> > > > > > > > > > > >> > comes to
> > > >> > > > > > > > > > > >> > > > > >> how
> > > >> > > > > > > > > > > >> > > > > >> > > many
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > protocol can
> > > >> > > > support?
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> group,
> > I
> > > >> think we
> > > >> > > > > > should
> > > >> > > > > > > > > shoot for
> > > >> > > > > > > > > > > >> > 1000s of
> > > >> > > > > > > > > > > >> > > > > >> > > members.
> > > >> > > > > > > > > > > >> > > > > >> > > > > But
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability
> > goals
> > > >> also
> > > >> > > > depend
> > > >> > > > > > on the
> > > >> > > > > > > > > offset
> > > >> > > > > > > > > > > >> > > > > management
> > > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities
> of
> > the
> > > >> > > > coordinator
> > > >> > > > > > > > > which we
> > > >> > > > > > > > > > > >> did
> > > >> > > > > > > > > > > >> > not
> > > >> > > > > > > > > > > >> > > > > >> cover in
> > > >> > > > > > > > > > > >> > > > > >> > > this
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> > > >> number that
> > > >> > > > > > applies
> > > >> > > > > > > > > > > >> > universally.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > > >> consider
> > > >> > > > having
> > > >> > > > > > > > > > > >> > SubscribedTopidIds
> > > >> > > > > > > > > > > >> > > > > >> instead
> > > >> > > > > > > > > > > >> > > > > >> > of
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> SubscribedTopicNames in
> > > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > >> > > > > > > > > > > >> > > > > >> > > Is the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we
> have
> > to
> > > >> resolve
> > > >> > > > the
> > > >> > > > > > regex
> > > >> > > > > > > > > on the
> > > >> > > > > > > > > > > >> > server,
> > > >> > > > > > > > > > > >> > > > > we
> > > >> > > > > > > > > > > >> > > > > >> > can
> > > >> > > > > > > > > > > >> > > > > >> > > do
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic
> > name?
> > > >> The
> > > >> > > > > > difference is
> > > >> > > > > > > > > that
> > > >> > > > > > > > > > > >> > sending the
> > > >> > > > > > > > > > > >> > > > > >> > regex
> > > >> > > > > > > > > > > >> > > > > >> > > is
> > > >> > > > > > > > > > > >> > > > > >> > > > > more
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas
> > sending
> > > >> the topic
> > > >> > > > > > names is
> > > >> > > > > > > > > less
> > > >> > > > > > > > > > > >> > efficient.
> > > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation
> is
> > > >> easier to
> > > >> > > > > > handle if
> > > >> > > > > > > > > we have
> > > >> > > > > > > > > > > >> > topic
> > > >> > > > > > > > > > > >> > > > > ids.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main
> reason
> > to
> > > >> still
> > > >> > > > let
> > > >> > > > > > the
> > > >> > > > > > > > > clients
> > > >> > > > > > > > > > > >> send
> > > >> > > > > > > > > > > >> > names
> > > >> > > > > > > > > > > >> > > > > >> is to
> > > >> > > > > > > > > > > >> > > > > >> > > keep
> > > >> > > > > > > > > > > >> > > > > >> > > > > the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> > names
> > > >> -> ids
> > > >> > > > on
> > > >> > > > > > the
> > > >> > > > > > > > > broker /
> > > >> > > > > > > > > > > >> > admin
> > > >> > > > > > > > > > > >> > > > > client
> > > >> > > > > > > > > > > >> > > > > >> > > only.
> > > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we
> > added
> > > >> topic id
> > > >> > > > in
> > > >> > > > > > > > > KIP-516, we
> > > >> > > > > > > > > > > >> > never
> > > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > consumer/producers
> > > >> > > > leveraging
> > > >> > > > > > the
> > > >> > > > > > > > > related
> > > >> > > > > > > > > > > >> > newer
> > > >> > > > > > > > > > > >> > > > > >> versioned
> > > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> > topic
> > > >> id as
> > > >> > > > empty
> > > >> > > > > > UUID.
> > > >> > > > > > > > > We
> > > >> > > > > > > > > > > >> want
> > > >> > > > > > > > > > > >> > to keep
> > > >> > > > > > > > > > > >> > > > > >> the
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and
> > only
> > > >> > > > delegate
> > > >> > > > > > the
> > > >> > > > > > > > > reasoning
> > > >> > > > > > > > > > > >> on
> > > >> > > > > > > > > > > >> > broker
> > > >> > > > > > > > > > > >> > > > > >> and
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul
> 6,
> > > >> 2022 at
> > > >> > > > 10:45
> > > >> > > > > > AM
> > > >> > > > > > > > > David
> > > >> > > > > > > > > > > >> Jacot
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> <dj...@confluent.io.invalid>
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would
> > like to
> > > >> start a
> > > >> > > > > > > > > discussion
> > > >> > > > > > > > > > > >> thread
> > > >> > > > > > > > > > > >> > on
> > > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > > >> > > > > > > > > > > >> > > > > >> > > The
> > > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation
> > of
> > > >> the
> > > >> > > > Consumer
> > > >> > > > > > > > > Rebalance
> > > >> > > > > > > > > > > >> > Protocol.
> > > >> > > > > > > > > > > >> > > > > >> With
> > > >> > > > > > > > > > > >> > > > > >> > > this
> > > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make
> the
> > > >> rebalance
> > > >> > > > > > protocol
> > > >> > > > > > > > > (for
> > > >> > > > > > > > > > > >> > consumers)
> > > >> > > > > > > > > > > >> > > > > >> more
> > > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> > > >> easier to
> > > >> > > > > > implement for
> > > >> > > > > > > > > > > >> > clients, and
> > > >> > > > > > > > > > > >> > > > > >> easier
> > > >> > > > > > > > > > > >> > > > > >> > > to
> > > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> > here:
> > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please
> take
> > a
> > > >> look and
> > > >> > > > let
> > > >> > > > > > me
> > > >> > > > > > > > > know what
> > > >> > > > > > > > > > > >> > you
> > > >> > > > > > > > > > > >> > > > > think.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will
> > be
> > > >> away from
> > > >> > > > > > July
> > > >> > > > > > > > > 18th to
> > > >> > > > > > > > > > > >> > August 8th.
> > > >> > > > > > > > > > > >> > > > > >> That
> > > >> > > > > > > > > > > >> > > > > >> > > gives
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> > read
> > > >> and
> > > >> > > > digest
> > > >> > > > > > this
> > > >> > > > > > > > > long
> > > >> > > > > > > > > > > >> KIP.
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >> > > > > > > > > > > >> > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> > >
> > > >> > > > > > > > > > > >> > > > > >> >
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >> --
> > > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > >
> > > >> > > > > >
> > > >> > > > >
> > > >> > > > >
> > > >> > > > > --
> > > >> > > > > -- Guozhang
> > > >> > > >
> > > >> > >
> > > >> > >
> > > >> > > --
> > > >> > > -- Guozhang
> > > >>
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
> >
>
>
> --
> -- Guozhang
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Re versions: fair enough, I think it's okay to keep it as strings. Just to
clarify the concern I had is that if we do want to augment it in the
future, it will be harder to change a `string` field into a `struct` :P

Re onAssignment: actually even with the current proposal, since we are
giving the final encoded metadata upon the first `onAssignment` call for
the generation already, Streams would infer all the standby tasks that need
are migrated in / out, so for Streams' purposes, we'd need to handle such
decoupling anyways, and it's actually the opposite: if the first
`onAssignment` does not include those yet-to-be-assigned then Streams would
not know if a migrating out standby would really be promoting to active
later or should it be completely removed, until later when next
`onAssignment` is called. And that's where I mentioned "we'd need to figure
out which onAssignment is the final call" etc. If we have all the
partitions on the `onAssignment`, then we can infer such actions and decide
whether we should close a standby task immediately or just recycle it and
wait for the active task to be assigned eventually.

On the other hand, if we call onAssignment incrementally similar to
onPartitionsAssigned, in which we include both assigned and
yet-to-be-assigned partitions, then for people who implement both
interfaces, they can just ignore the `onPartitionsAssigned` since they have
all knowledge they need from onAssignment already. And that's what I'm
trying to avoid by making the functionality of the two more separated.


Guozhang




On Mon, Sep 26, 2022 at 11:30 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Regarding the version, I would rather add this later when we have a
> clear use case for it. It seems to me that we are speculating here. I
> understand your point but I am not fully convinced by the solution at
> the moment. Would you agree with this?
>
> Regarding the onAssignment, I was thinking about the case where a task
> is promoted from standby to active. In this case, having onAssignment
> with the combined partitions could make it difficult, no? I am
> thinking that the assignor will have to remove the standby based on
> the metadata but it won't know what to do after that. If the partition
> is assigned directly, it can convert it to an active task. On the
> other hand, if the partition is not available yet, it would have to
> keep it as a standby until the partition is really assigned. It seems
> to be that the assignor won't have the information to make this
> decision, no? In this case, decouple the "why" from the "when" seems
> to make things harder. I am not so familiar with Streams though so my
> intuition could be wrong here.
>
> David
>
> On Mon, Sep 26, 2022 at 7:26 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Regarding the version, what I was thinking is that in the HB request, for
> > "serverAssignor" field, instead of just having it as a single string
> field,
> > maybe we could consider also making it a structure that includes: name,
> > minimumVersion, maximumVersion. Where the minimumVersion/maximumVersion
> > means the versions of the server assignor that the client work best with.
> > That being said, I agree with you that such information may also be
> > inferred elsewhere e.g. by looking into the "rackId" field, and see if it
> > contains a hyphen or not etc. All I was wondering is that, if such
> version
> > information would be useful for the server assignors to determine its
> > actual assignment logic. I do not feel very strong about this one though
> > --- even if we do not add it now, we can potentially add later, it's just
> > that changing a single string field to a structure would be hard for
> > compatibility and we'd then probably have to add top-level fields.
> >
> > Regarding the `onAssignment` logic, again my train of thoughts is that,
> if
> > users want to know exactly when a partition is assigned / revoked, they
> > would be leveraging on the rebalance callbacks, as that's what people
> > should rely on to determine "when" partitions are assigned. The
> > `onAssignment` should be used for getting "why" such partition assignment
> > decision is made, and hence returning `combined partitions` would be
> okay.
> > Streams e.g. implement both rebalance callbacks and the assignors, and it
> > gets the "when" from the former (and create/close active tasks
> accordingly)
> > and the "why" from the latter (and update its global info bookkeeping as
> > well as standby maintenance accordingly). Most users would be just
> > interested in the rebalance callback, and not implement their own
> assignor
> > at all if they do not care about "why" as they trust the server assignors
> > would take good care of those, and only about "when". So if we did build
> > such two types of APIs from scratch, I'd indeed feel that not providing
> the
> > partitions but only the metadata for `onAssignment` may be less confusing
> > and push users to separate the usage of these two more clearly, but since
> > we already introduced partitions in `onAssignment` for compatibility I'm
> > less keen on removing them.
> >
> >
> > Guozhang
> >
> > On Mon, Sep 26, 2022 at 6:55 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > Regarding the version, my understanding is that the version would be
> > > either the client software version or the request version, is this
> > > correct? If so, we could indeed pass this information down to the
> > > assignor via the interface. One way would be to pass a "server
> > > context" to the assignor and that context would include that
> > > information (and perhaps more). Is this what you are looking for?
> > >
> > > Regarding the onAssignment, I think that I understand your point. I
> > > suppose that the assignor could also be clever and keep track of the
> > > last metadata to decide whether it has to do something or not. One
> > > question that is still not clear to me is whether the assignor needs
> > > to know all the assigned partitions upfront regardless of whether they
> > > are already revoked or not. Do you think that we need this as well?
> > >
> > > From an API perspective, we could have something like
> > > onAssignment(Metadata(version, reason, metadata, assigned partitions,
> > > pending partitions)). Where the assigned partitions are the partitions
> > > ready to be used and the pending partitions are the one assigned to
> > > the member but not revoked yet. I find it a bit weird that this method
> > > would be called only once because the assignor would not know when the
> > > pending partitions changes. That does not look like a clean API. An
> > > alternative would be to use onAssignment(Metadata(version, reason,
> > > metadata, combined partitions)) but this seems error prone because it
> > > is not clear whether a partition is usable or not. Or do you think
> > > that we should not provide the partitions but only the metadata?
> > >
> > > Best,
> > > David
> > >
> > > On Fri, Sep 23, 2022 at 9:40 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > >
> > > > Hello David,
> > > >
> > > > On Fri, Sep 23, 2022 at 2:00 AM David Jacot
> <djacot@confluent.io.invalid
> > > >
> > > > wrote:
> > > >
> > > > > Hey,
> > > > >
> > > > > > Just to clarify I was asking about the `version` of the assignor
> > > (i.e. up
> > > > > to what version that the client would support), and I do agree we
> > > would not
> > > > > need metadata. What I have in mind is that, for some specific
> built-in
> > > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> in a
> > > > > newer version we would have a hierarchical rack ID string format,
> like
> > > > > "tier1-tier2" etc, but if some client has not upgraded their rack
> ID
> > > > > would still be in old format. In this case, the broker then needs
> to
> > > choose
> > > > > the old versioned assignor. I'm probably making something up here
> for
> > > rack
> > > > > aware assignors, but I'm wondering if in general such an
> > > "auto-downgrade"
> > > > > behavior would be needed still for broker-side assignor, and if yes
> > > would
> > > > > "version" still be useful.
> > > > >
> > > > > Got it. That's an interesting thought. I think that the issue is
> that
> > > > > the client will never tell you which version of the server-side
> > > > > assignor should be used. Do you think that the coordinator would
> > > > > downgrade the version if the assignment fails with a higher
> version? I
> > > > > tend to believe that this should be handled within the assignor
> > > > > itself. In the example that you mentioned, the assignor would have
> to
> > > > > handle all the cases. I am not really convinced that we need this
> at
> > > > > the moment.
> > > > >
> > > > > The version from the client side would not be indicating the broker
> > > which
> > > > version to use, but rather which version the client would "work best
> > > with".
> > > > Such a "version" field would not be settible by the users, since they
> > > will
> > > > be hard-codedly bumped when the Kafka byte code version bumped.
> > > > Back to the rack aware assignor example, if the older versioned
> client
> > > does
> > > > not have a hierarchical rack ID, however if the assignment returned
> to
> > > them
> > > > is assuming a hierarchical rack structure, it may not reflect the
> best
> > > > workload balance among those new and old versioned clients. That
> means,
> > > > when receiving the members subscriptions at the server side, if the
> > > > versions from all these members are different, the broker's assignor
> may
> > > > need to consider using the lower version logic to do the assignment.
> So
> > > yes
> > > > the assignor would indeed have to handle all such cases, but it
> needs to
> > > do
> > > > so such that if there are clients who would not work with certain new
> > > > logic, it would then handle such cases automatically by e.g. still
> using
> > > an
> > > > older versioned logic.
> > > >
> > > >
> > > >
> > > > > > Okay, my understanding is that the calling ordering of these
> > > callbacks
> > > > > would be like the following:
> > > > >
> > > > > Yes, your examples look right.
> > > > >
> > > > > > I'm wondering if we would still call onAssignment just once, that
> > > encodes
> > > > > all the assignment for this rebalance, including all the partitions
> > > that
> > > > > should be assigned to the member but not yet assigned since they
> have
> > > not
> > > > > been revoked by others. In that case the call ordering would be:
> > > > >
> > > > > Interesting. Is there a case for Streams where having the full
> > > > > assignment is beneficial? For instance, I can think of the
> following
> > > > > case. When a standby task is promoted to an active task, the
> metadata
> > > > > would not contain the standby task anymore and the assignment may
> not
> > > > > have the partition yet. In this case, Streams would stop the
> standby
> > > > > tasks but not have the active task yet if my understanding of
> Streams
> > > > > is correct. So knowing the full assignment could be helpful here.
> > > > >
> > > > > If we want to do this, we could structure the assignment given to
> the
> > > > > member as follow: version, error, metadata, assigned partitions,
> > > > > pending partitions, where the pending partitions would be the one
> > > > > assigned to this member but not yet available. What do you think?
> > > > >
> > > > > Regarding onAssignment being called only once, I am not sure to
> fully
> > > > > grasp the benefit yet. Does the assignor really care about this? In
> > > > > the end, the epoch does not really matter for the assignor because
> it
> > > > > has to converge its state to the desired state anyway.
> > > > >
> > > > > Here's my rationale (maybe rephased a bit :P ): the implementers of
> > > > rebalance listener and assignor are two groups of people, and most
> users
> > > > fall into the former group, while only very few people fall into the
> > > later
> > > > group. For rebalance listener implementers, they just want to know
> when a
> > > > partition is actually revoked or assigned to the consumer and reacts
> to
> > > it,
> > > > for this purpose, `onPartitionsRevoked` and `onPartitionsAssigned`
> would
> > > be
> > > > triggered interleavingly upon `poll` calls across rebalances. The
> usual
> > > > logic for such rebalance listeners are metrics reporting, committing
> > > > offsets (if they do not use Kafka for that), etc. They would not care
> > > which
> > > > calls are from which rebalances --- in the past with eager
> rebalance, it
> > > > maybe that each rebalance is associated with exactly a
> > > > `onPartitionsRevoked` first and then a `onPartitionsAssigned`, but it
> > > would
> > > > no longer the cases now.
> > > >
> > > > The implementers of the assignor though, would care about "how the
> > > > assignment was made", that includes from which rebalance a certain
> > > > revoke/assign decision was made, based on what metadata such
> assignment
> > > is
> > > > made, etc. And that's the whole point of the `onAssignment` function
> > > since
> > > > otherwise they can just rely on the listeners. They usually
> > > implementation
> > > > logic of this callback is to e.g. bookkeep the assignment decision
> > > driving
> > > > factors a.k.a. the metadata, global information that needs to be
> > > propagated
> > > > to all members, etc. Take Streams as an example, the active
> processing
> > > > tasks go along with the assigned partitions, and we can always just
> > > > incrementally create / close them upon each rebalance listener
> triggers,
> > > > when certain partitions are revoked or assigned together; standby
> tasks
> > > > however are encoded with the metadata, and we can only know which
> standby
> > > > tasks should we get / drop based on the `onAssignment` function, and
> in
> > > > fact the creation of such tasks as a result of the metadata
> bookkeeping
> > > > does not need to wait until all the partitions that are yet-assigned
> have
> > > > been completely assigned to the member. Such information may not
> always
> > > be
> > > > updatable in an incremental manner as the partitions-revoked /
> > > > partitions-assigned. In such a case, it's better to just trigger this
> > > > function "once per decision made" i.e. once per rebalance generation.
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Thu, Sep 22, 2022 at 6:01 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > > > >
> > > > > > Hi David, thanks for all the detailed explanations. I think they
> all
> > > make
> > > > > > sense. Just want to have a couple follow-ups here:
> > > > > >
> > > > > > > I don't really see the benefits here because server side
> assignors
> > > > > > don't have metadata at all. They only assign topic-partitions.
> They
> > > > > > are not supposed to generate metadata nor to receive metadata
> from
> > > the
> > > > > > members.
> > > > > >
> > > > > > Just to clarify I was asking about the `version` of the assignor
> > > (i.e. up
> > > > > > to what version that the client would support), and I do agree we
> > > would
> > > > > not
> > > > > > need metadata. What I have in mind is that, for some specific
> > > built-in
> > > > > > broker-assignors, e.g. rack-aware assignors, if it's possible
> that
> > > in a
> > > > > > newer version we would have a hierarchical rack ID string format,
> > > like
> > > > > > "tier1-tier2" etc, but if some client has not upgraded their
> rack ID
> > > > > > would still be in old format. In this case, the broker then
> needs to
> > > > > choose
> > > > > > the old versioned assignor. I'm probably making something up
> here for
> > > > > rack
> > > > > > aware assignors, but I'm wondering if in general such an
> > > "auto-downgrade"
> > > > > > behavior would be needed still for broker-side assignor, and if
> yes
> > > would
> > > > > > "version" still be useful.
> > > > > >
> > > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is
> called
> > > once
> > > > > > when the member transitions to a new epoch. This one contains the
> > > full
> > > > > > metadata provided by the client side assignor. Then,
> `onAssignment`
> > > > > > can be called max N times where N is the number of partitions
> pending
> > > > > > revocation by other members. Let me try to clarify this in the
> KIP.
> > > > > >
> > > > > > Okay, my understanding is that the calling ordering of these
> > > callbacks
> > > > > > would be like the following:
> > > > > >
> > > > > > ----------------------------------------
> > > > > > onPartitionsRevoked();   // just once, since we do not really
> need
> > > > > > to revoke incrementally.
> > > > > >
> > > > > > onAssignment();    // the first call, with epoch incremented
> > > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > > >
> > > > > > onAssignment();              // the first onAssignment would
> bump up
> > > the
> > > > > > epoch, and the metadata reflected.
> > > > > > onPartitionsAssigned();   // each time we get an additional
> > > assignment,
> > > > > we
> > > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > > ...
> > > > > > onAssignment();
> > > > > > onPartitionsAssigned();   // on each of the onAssignment calls,
> the
> > > > > encoded
> > > > > > metadata would not change, only the incrementally added
> partitions be
> > > > > > reflected
> > > > > >
> > > > > > Is that the case?
> > > > > >
> > > > > > I'm wondering if we would still call onAssignment just once, that
> > > encodes
> > > > > > all the assignment for this rebalance, including all the
> partitions
> > > that
> > > > > > should be assigned to the member but not yet assigned since they
> > > have not
> > > > > > been revoked by others. In that case the call ordering would be:
> > > > > >
> > > > > > ----------------------------------------
> > > > > > onPartitionsRevoked();   // just once
> > > > > > onAssignment();    // just once, with epoch incremented, and
> metadata
> > > > > > encoded changed, the "assignment" field also reflect the final
> target
> > > > > > assignment
> > > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > > incrementally
> > > > > > added partitions
> > > > > > ...
> > > > > > onPartitionsAssigned();
> > > > > >
> > > > > > The motivation from this is that, most users would only
> implement the
> > > > > > rebalance callback listeners and hence we'd definitely need to
> make
> > > sure
> > > > > > the semantics of that does not change much, and the time
> > > > > > `onPartitionsAssigned` indicate the time when the partitions are
> > > actually
> > > > > > assigned to it; while for assignors, the `onAssignment` is used
> to
> > > > > indicate
> > > > > > what decision is made regarding for this member, i.e. when the
> > > partitions
> > > > > > are decided to be given to it, but not necessarily meant that it
> has
> > > been
> > > > > > given, since that time should be determined by the time of
> > > > > > `onPartitionsAssigned`. The benefits though, would be that
> assignor
> > > > > > implementers would not need to reason which `onAssignment` would
> be
> > > the
> > > > > > last one for this epoch.
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > > <djacot@confluent.io.invalid
> > > > > >
> > > > > > wrote:
> > > > > >
> > > > > > > Hi Guozhang,
> > > > > > >
> > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> Error
> > > byte
> > > > > > > field,
> > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> not.
> > > And
> > > > > the
> > > > > > > > rationale seems to be assuming that we should always be able
> to
> > > do
> > > > > the
> > > > > > > > assignment at the broker-side assignor without errors.
> > > Personally I
> > > > > think
> > > > > > > > it's still potentially beneficial to add the Error field
> even for
> > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > subscribed
> > > > > > > > topics are not recognized with the current broker's metadata.
> > > What
> > > > > do you
> > > > > > > > think?
> > > > > > >
> > > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > > better
> > > > > > > to use an exception on the server side. This is what we
> usually do
> > > for
> > > > > > > server side plugins. On the client side, we use a field
> because the
> > > > > > > errors are not defined in advance.
> > > > > > >
> > > > > > > Your comment also makes me think about what we should do when
> the
> > > > > > > server side assignor fails. I suppose that we have to keep the
> > > current
> > > > > > > assignment until a new event occurs. For instance, in your
> example,
> > > > > > > the coordinator would have to trigger a rebalance when
> unrecognized
> > > > > > > topics are available. This would be part of the metadata
> > > monitoring.
> > > > > > >
> > > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > > reason/version/metadata compared with the broker-side
> > > GroupMember. I
> > > > > > > agree
> > > > > > > > that broker-side assignor would not need reason/metadata
> since
> > > they
> > > > > are
> > > > > > > > blackbox strings/bytes to the assignor, but what about
> version?
> > > E.g.
> > > > > is
> > > > > > > it
> > > > > > > > possible that we evolve our broker-side built-in assignor but
> > > the old
> > > > > > > > versioned clients would not be able to work with the new
> > > version, in
> > > > > > > which
> > > > > > > > case we need to let the broker being aware of this and
> upgrade
> > > its
> > > > > > > behavior
> > > > > > > > to cooperate with the clients?
> > > > > > >
> > > > > > > I don't really see the benefits here because server side
> assignors
> > > > > > > don't have metadata at all. They only assign topic-partitions.
> They
> > > > > > > are not supposed to generate metadata nor to receive metadata
> from
> > > the
> > > > > > > members.
> > > > > > >
> > > > > > > > 3) Also related to 2) above, for the client-side
> "GroupMember",
> > > > > instead
> > > > > > > of
> > > > > > > > including these three fields, what about just adding the
> > > "Metadata"
> > > > > field
> > > > > > > > class which has these three fields? Also, there are two
> > > "Metadata"
> > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > reason/version/metadata, and the second is just the encoded
> > > metadata
> > > > > > > bytes.
> > > > > > > > I'm wondering what about just naming the first as
> memberMetadata,
> > > > > which
> > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > second
> > > > > > > > bytebuffer field as metadataBytes?
> > > > > > >
> > > > > > > That's a good point. Let me try to rationalize this interface
> > > based on
> > > > > > > your suggestions.
> > > > > > >
> > > > > > > Best,
> > > > > > > David
> > > > > > >
> > > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <
> wangguoz@gmail.com>
> > > > > wrote:
> > > > > > > >
> > > > > > > > Hello David,
> > > > > > > >
> > > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > > >
> > > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an
> Error
> > > byte
> > > > > > > field,
> > > > > > > > while the broker-side "PartitionAssignor#Assignment" does
> not.
> > > And
> > > > > the
> > > > > > > > rationale seems to be assuming that we should always be able
> to
> > > do
> > > > > the
> > > > > > > > assignment at the broker-side assignor without errors.
> > > Personally I
> > > > > think
> > > > > > > > it's still potentially beneficial to add the Error field
> even for
> > > > > > > > broker-side assignors, e.g. for some edge cases where some
> > > subscribed
> > > > > > > > topics are not recognized with the current broker's metadata.
> > > What
> > > > > do you
> > > > > > > > think?
> > > > > > > >
> > > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > > reason/version/metadata compared with the broker-side
> > > GroupMember. I
> > > > > > > agree
> > > > > > > > that broker-side assignor would not need reason/metadata
> since
> > > they
> > > > > are
> > > > > > > > blackbox strings/bytes to the assignor, but what about
> version?
> > > E.g.
> > > > > is
> > > > > > > it
> > > > > > > > possible that we evolve our broker-side built-in assignor but
> > > the old
> > > > > > > > versioned clients would not be able to work with the new
> > > version, in
> > > > > > > which
> > > > > > > > case we need to let the broker being aware of this and
> upgrade
> > > its
> > > > > > > behavior
> > > > > > > > to cooperate with the clients?
> > > > > > > >
> > > > > > > > 3) Also related to 2) above, for the client-side
> "GroupMember",
> > > > > instead
> > > > > > > of
> > > > > > > > including these three fields, what about just adding the
> > > "Metadata"
> > > > > field
> > > > > > > > class which has these three fields? Also, there are two
> > > "Metadata"
> > > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > > reason/version/metadata, and the second is just the encoded
> > > metadata
> > > > > > > bytes.
> > > > > > > > I'm wondering what about just naming the first as
> memberMetadata,
> > > > > which
> > > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > > second
> > > > > > > > bytebuffer field as metadataBytes?
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hello David,
> > > > > > > > >
> > > > > > > > > Thanks for bringing this question up. I think the main
> > > benefits as
> > > > > you
> > > > > > > > > listed is 2) above if it stays; just to clarify, we would
> only
> > > be
> > > > > able
> > > > > > > to
> > > > > > > > > save one round trip if the rebalance is still triggered by
> the
> > > > > broker;
> > > > > > > if
> > > > > > > > > the rebalance is triggered by the client then the
> > > num.round.trips
> > > > > are
> > > > > > > the
> > > > > > > > > same:
> > > > > > > > >
> > > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > > >
> > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > already
> > > > > sent
> > > > > > > a
> > > > > > > > > HB and hence has to wait for it to return first since only
> one
> > > > > request
> > > > > > > /
> > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > > GroupPrepareAssignment request.
> > > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > GroupInstallAssignment
> > > > > > > > > request.
> > > > > > > > >
> > > > > > > > > In total, two round trips.
> > > > > > > > >
> > > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > > >
> > > > > > > > > T0: client decides to do a new assignment, suppose it has
> > > already
> > > > > sent
> > > > > > > a
> > > > > > > > > HB and hence has to wait for it to return first since only
> one
> > > > > request
> > > > > > > /
> > > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > > T1: client receives the HB response, and then sends the
> new HB
> > > > > request
> > > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > > T2: the HB response with the optional member metadata map
> is
> > > > > returned.
> > > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > > GroupInstallAssignment
> > > > > > > > > request.
> > > > > > > > >
> > > > > > > > > In total, two round trips as well.
> > > > > > > > >
> > > > > > > > > -----------------------------
> > > > > > > > >
> > > > > > > > > So to complete the full picture here, we'd need to modify
> both
> > > HB
> > > > > > > request
> > > > > > > > > and response so that the client can also indicate a new
> > > rebalance
> > > > > via
> > > > > > > the
> > > > > > > > > HB request as well, right?
> > > > > > > > >
> > > > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > > additional
> > > > > > > fields
> > > > > > > > > encoding the full member (subscription) metadata and topic
> > > > > metadata as
> > > > > > > > > optional fields.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Guozhang
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > >> Hi all,
> > > > > > > > >>
> > > > > > > > >> During an offline conversation, someone asked why we need
> the
> > > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we
> could
> > > > > instead
> > > > > > > > >> provide the group state in the heartbeat response. This
> has a
> > > few
> > > > > > > > >> advantages: 1) it does not require using a special error
> code
> > > to
> > > > > > > > >> signal that a new assignment is required as the signal
> would
> > > be
> > > > > the
> > > > > > > > >> provided group state; 2) it removes one round trip when a
> > > client
> > > > > side
> > > > > > > > >> assignor is used. The downside is that it makes the
> heartbeat
> > > > > > > > >> response's definition quite large. I recall that I went
> with
> > > the
> > > > > > > > >> current approach due to this.
> > > > > > > > >>
> > > > > > > > >> Providing the group state in the heartbeat response is
> > > appealing.
> > > > > What
> > > > > > > > >> do you guys think?
> > > > > > > > >>
> > > > > > > > >> Best,
> > > > > > > > >> David
> > > > > > > > >>
> > > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > > djacot@confluent.io>
> > > > > > > wrote:
> > > > > > > > >> >
> > > > > > > > >> > Hi Guozhang,
> > > > > > > > >> >
> > > > > > > > >> > 1. I have added a reference to the relevant chapter
> instead
> > > of
> > > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > > >> >
> > > > > > > > >> > 2. The "Rebalance Triggers" section you are referring
> to is
> > > > > about
> > > > > > > when
> > > > > > > > >> > a rebalance should be triggered for the non-upgraded
> members
> > > > > using
> > > > > > > the
> > > > > > > > >> > old protocol. The section mentions that a rebalance
> must be
> > > > > > > triggered
> > > > > > > > >> > when a new assignment is installed. This implies that
> the
> > > group
> > > > > > > epoch
> > > > > > > > >> > was updated either by a native member or a non-upgraded
> > > member.
> > > > > For
> > > > > > > > >> > the latter, the JoinGroup request would be the trigger.
> I
> > > have
> > > > > > > added a
> > > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > > Handling"
> > > > > > > section
> > > > > > > > >> > as well. Does that make sense?
> > > > > > > > >> >
> > > > > > > > >> > Thanks,
> > > > > > > > >> > David
> > > > > > > > >> >
> > > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > > > >> wrote:
> > > > > > > > >> > >
> > > > > > > > >> > > Hello David,
> > > > > > > > >> > >
> > > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > > clear in
> > > > > the
> > > > > > > doc,
> > > > > > > > >> > > could we update:
> > > > > > > > >> > >
> > > > > > > > >> > > 1) the heartbeat request handling section, stating
> when
> > > > > > > coordinator
> > > > > > > > >> will
> > > > > > > > >> > > trigger rebalance based on the HB's member metadata /
> > > reason?
> > > > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > > > described
> > > > > > > in
> > > > > > > > >> "Group
> > > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > > >> > >
> > > > > > > > >> > >
> > > > > > > > >> > > Guozhang
> > > > > > > > >> > >
> > > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > >> > > wrote:
> > > > > > > > >> > >
> > > > > > > > >> > > > Hi Guozhang,
> > > > > > > > >> > > >
> > > > > > > > >> > > > I thought that the assignor will always be consulted
> > > when
> > > > > the
> > > > > > > next
> > > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > > >> > > > `PartitionAssignor#metadata` will be called for
> every
> > > > > heartbeat.
> > > > > > > > >> This
> > > > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > > > rebalance by
> > > > > > > > >> > > > setting the reason to a non-zero value or by
> changing
> > > the
> > > > > > > bytes. Do
> > > > > > > > >> > > > you think that this is not sufficient? Are you
> > > concerned by
> > > > > the
> > > > > > > > >> delay?
> > > > > > > > >> > > >
> > > > > > > > >> > > > Best,
> > > > > > > > >> > > > David
> > > > > > > > >> > > >
> > > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > > wangguoz@gmail.com>
> > > > > > > > >> wrote:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > Hello David,
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > ```
> > > > > > > > >> > > > > In this case, a new assignment is triggered by the
> > > client
> > > > > side
> > > > > > > > >> > > > > assignor. When constructing the HB, the consumer
> will
> > > > > always
> > > > > > > > >> consult
> > > > > > > > >> > > > > the client side assignor and propagate the
> > > information to
> > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > coordinator. In other words, we don't expect
> users to
> > > call
> > > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > > >> > > > > ```
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > > interface,
> > > > > we
> > > > > > > > >> actually do
> > > > > > > > >> > > > > not have a way yet to instruct how to construct
> the
> > > next
> > > > > HB
> > > > > > > > >> request, e.g.
> > > > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> > > with a
> > > > > new
> > > > > > > > >> assignment,
> > > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > > PartitionAssignor
> > > > > > > to
> > > > > > > > >> indicate
> > > > > > > > >> > > > > the next HB telling broker about so. WDYT about
> adding
> > > > > such an
> > > > > > > > >> API on the
> > > > > > > > >> > > > > PartitionAssignor?
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > Guozhang
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > Hi Jun,
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > I have updated the KIP to include your
> feedback. I
> > > have
> > > > > also
> > > > > > > > >> tried to
> > > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > Best,
> > > > > > > > >> > > > > > David
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > > djacot@confluent.io
> > > > > > > > >> >
> > > > > > > > >> > > > wrote:
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Hi Jun,
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > > answering
> > > > > your
> > > > > > > > >> questions
> > > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> benefits
> > > of
> > > > > the
> > > > > > > KIP
> > > > > > > > >> seem to
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > fewer
> > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > support of
> > > > > > > > >> wildcard. A few
> > > > > > > > >> > > > > > > > comments below.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > I would also add that the KIP removes the
> global
> > > sync
> > > > > > > barrier
> > > > > > > > >> in the
> > > > > > > > >> > > > > > > protocol which is essential to improve group
> > > > > stability and
> > > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > > client by
> > > > > > > moving
> > > > > > > > >> most of
> > > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> plan
> > > to
> > > > > > > support
> > > > > > > > >> rolling
> > > > > > > > >> > > > > > changing
> > > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > > > assignor
> > > > > > > used
> > > > > > > > >> by a
> > > > > > > > >> > > > > > > majority of the members. This allows the
> group to
> > > move
> > > > > > > from
> > > > > > > > >> one
> > > > > > > > >> > > > > > > assignor to another by a roll. This is
> explained
> > > in
> > > > > the
> > > > > > > > >> Assignor
> > > > > > > > >> > > > > > > Selection chapter.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> whether
> > > it's
> > > > > > > > >> required in
> > > > > > > > >> > > > every
> > > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > > > filled? For
> > > > > > > > >> example,
> > > > > > > > >> > > > it's
> > > > > > > > >> > > > > > not
> > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > > filled.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > The client is expected to set those fields in
> > > case of
> > > > > a
> > > > > > > > >> connection
> > > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > > changed
> > > > > since
> > > > > > > > >> the last
> > > > > > > > >> > > > > > > HB. The server populates those fields as long
> as
> > > the
> > > > > > > member
> > > > > > > > >> is not
> > > > > > > > >> > > > > > > fully reconciled - the member should
> acknowledge
> > > that
> > > > > it
> > > > > > > has
> > > > > > > > >> the
> > > > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> > > this in
> > > > > the
> > > > > > > KIP.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> rack
> > > > > affinity
> > > > > > > > >> between the
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > and the broker is only considered during
> > > fetching,
> > > > > but
> > > > > > > not
> > > > > > > > >> during
> > > > > > > > >> > > > > > assigning
> > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > assignment
> > > > > > > is
> > > > > > > > >> made,
> > > > > > > > >> > > > there
> > > > > > > > >> > > > > > is
> > > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > > > replicas of
> > > > > > > > >> assigned
> > > > > > > > >> > > > > > partitions
> > > > > > > > >> > > > > > > > are close to the member. I am wondering if
> we
> > > > > should use
> > > > > > > > >> this
> > > > > > > > >> > > > > > opportunity
> > > > > > > > >> > > > > > > > to address this by including rack in
> > > GroupMember.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's an interesting idea. I don't see any
> issue
> > > with
> > > > > > > adding
> > > > > > > > >> the
> > > > > > > > >> > > > rack
> > > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful
> to
> > > know
> > > > > how
> > > > > > > busy
> > > > > > > > >> a group
> > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> model,
> > > it
> > > > > seems
> > > > > > > > >> that we
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > add
> > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> time
> > > the
> > > > > event
> > > > > > > > >> loop is
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > actual work.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > failover
> > > > > > > > >> handling? For
> > > > > > > > >> > > > > > example,
> > > > > > > > >> > > > > > > > does it need to trigger the check if any
> group
> > > with
> > > > > the
> > > > > > > > >> wildcard
> > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sure. When the new group coordinator takes
> over,
> > > it
> > > > > has
> > > > > > > to:
> > > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > > assignor
> > > > > is
> > > > > > > used.
> > > > > > > > >> We
> > > > > > > > >> > > > don't
> > > > > > > > >> > > > > > > store the information about the member
> selected to
> > > > > run the
> > > > > > > > >> assignment
> > > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > > >> > > > > > > * Update the topics metadata, verify the
> wildcard
> > > > > > > > >> subscriptions, and
> > > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > document
> > > > > what
> > > > > > > > >> the epoch
> > > > > > > > >> > > > > > field
> > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > >> > > > > > reflect
> > > > > > > > >> > > > > > > > the latest group epoch? What about the one
> in
> > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> > > latest
> > > > > > > group
> > > > > > > > >> epoch.
> > > > > > > > >> > > > > > > When the group state is updated, the group
> epoch
> > > is
> > > > > > > bumped so
> > > > > > > > >> we use
> > > > > > > > >> > > > > > > that one for all the change records related
> to the
> > > > > update.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> the
> > > > > > > following
> > > > > > > > >> > > > invariants
> > > > > > > > >> > > > > > are
> > > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> for
> > > a
> > > > > member
> > > > > > > > >> not to
> > > > > > > > >> > > > get any
> > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > That's right. Here I meant that the members
> > > provided
> > > > > by
> > > > > > > the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > > > assignor
> > > > > > > can
> > > > > > > > >> not make
> > > > > > > > >> > > > > > > up new member ids.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > > epoch
> > > > > > > equals to
> > > > > > > > >> 0":
> > > > > > > > >> > > > When
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > a consumer rejoin and what member id would
> be
> > > used?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > A member is expected to abandon all its
> > > partitions and
> > > > > > > > >> rejoins when
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In
> this
> > > case,
> > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > > > coordinator will have removed the member from
> the
> > > > > group.
> > > > > > > The
> > > > > > > > >> member
> > > > > > > > >> > > > > > > can rejoin the group with the same member id
> but
> > > with
> > > > > 0 as
> > > > > > > > >> epoch. Let
> > > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> ability
> > > to
> > > > > > > trigger a
> > > > > > > > >> > > > > > reassignment
> > > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > > > updating the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > > with
> > > > > the
> > > > > > > > >> deprecation
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > In this case, a new assignment is triggered
> by the
> > > > > client
> > > > > > > side
> > > > > > > > >> > > > > > > assignor. When constructing the HB, the
> consumer
> > > will
> > > > > > > always
> > > > > > > > >> consult
> > > > > > > > >> > > > > > > the client side assignor and propagate the
> > > > > information to
> > > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > > users to
> > > > > call
> > > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> the
> > > > > section
> > > > > > > > >> seems to
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2,
> B
> > > > > > > immediately
> > > > > > > > >> gets into
> > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > incorrect.
> > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3,
> C
> > > seems
> > > > > to
> > > > > > > get
> > > > > > > > >> into
> > > > > > > > >> > > > > > epoch=3,
> > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > > has A -
> > > > > > > > >> epoch=2,
> > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> support
> > > the
> > > > > > > upgrade
> > > > > > > > >> from
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > old
> > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > We will support upgrading from the consumer
> > > protocol
> > > > > > > version
> > > > > > > > >> 3,
> > > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not
> implemented
> > > yet
> > > > > so
> > > > > > > the
> > > > > > > > >> earliest
> > > > > > > > >> > > > > > > version is unknown at the moment. This is
> > > explained
> > > > > in the
> > > > > > > > >> migration
> > > > > > > > >> > > > > > > plan chapter.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will
> update
> > > > > the KIP
> > > > > > > > >> based on
> > > > > > > > >> > > > it
> > > > > > > > >> > > > > > > next week.
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > Best,
> > > > > > > > >> > > > > > > David
> > > > > > > > >> > > > > > >
> > > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > > >> > > > wrote:
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Hi, David,
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main
> benefits
> > > of
> > > > > the
> > > > > > > KIP
> > > > > > > > >> seem to
> > > > > > > > >> > > > be
> > > > > > > > >> > > > > > fewer
> > > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > > support of
> > > > > > > > >> wildcard. A few
> > > > > > > > >> > > > > > > > comments below.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we
> plan
> > > to
> > > > > > > support
> > > > > > > > >> rolling
> > > > > > > > >> > > > > > changing
> > > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > > >> > > > > > > > 30.2 For each field, could you explain
> whether
> > > it's
> > > > > > > > >> required in
> > > > > > > > >> > > > every
> > > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > > > filled? For
> > > > > > > > >> example,
> > > > > > > > >> > > > it's
> > > > > > > > >> > > > > > not
> > > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > > filled.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 31. In the current consumer protocol, the
> rack
> > > > > affinity
> > > > > > > > >> between the
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > and the broker is only considered during
> > > fetching,
> > > > > but
> > > > > > > not
> > > > > > > > >> during
> > > > > > > > >> > > > > > assigning
> > > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > > assignment
> > > > > > > is
> > > > > > > > >> made,
> > > > > > > > >> > > > there
> > > > > > > > >> > > > > > is
> > > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > > > replicas of
> > > > > > > > >> assigned
> > > > > > > > >> > > > > > partitions
> > > > > > > > >> > > > > > > > are close to the member. I am wondering if
> we
> > > > > should use
> > > > > > > > >> this
> > > > > > > > >> > > > > > opportunity
> > > > > > > > >> > > > > > > > to address this by including rack in
> > > GroupMember.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful
> to
> > > know
> > > > > how
> > > > > > > busy
> > > > > > > > >> a group
> > > > > > > > >> > > > > > > > coordinator is. By moving the event loop
> model,
> > > it
> > > > > seems
> > > > > > > > >> that we
> > > > > > > > >> > > > could
> > > > > > > > >> > > > > > add
> > > > > > > > >> > > > > > > > a metric that tracks the fraction of the
> time
> > > the
> > > > > event
> > > > > > > > >> loop is
> > > > > > > > >> > > > doing
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > actual work.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > > failover
> > > > > > > > >> handling? For
> > > > > > > > >> > > > > > example,
> > > > > > > > >> > > > > > > > does it need to trigger the check if any
> group
> > > with
> > > > > the
> > > > > > > > >> wildcard
> > > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > > document
> > > > > what
> > > > > > > > >> the epoch
> > > > > > > > >> > > > > > field
> > > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > > >> > > > > > reflect
> > > > > > > > >> > > > > > > > the latest group epoch? What about the one
> in
> > > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that
> the
> > > > > > > following
> > > > > > > > >> > > > invariants
> > > > > > > > >> > > > > > are
> > > > > > > > >> > > > > > > > met: ... All members exists." It's possible
> for
> > > a
> > > > > member
> > > > > > > > >> not to
> > > > > > > > >> > > > get any
> > > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > > epoch
> > > > > > > equals to
> > > > > > > > >> 0":
> > > > > > > > >> > > > When
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > a consumer rejoin and what member id would
> be
> > > used?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 37. "Instead, power users will have the
> ability
> > > to
> > > > > > > trigger a
> > > > > > > > >> > > > > > reassignment
> > > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > > > updating the
> > > > > > > > >> assignor
> > > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > > with
> > > > > the
> > > > > > > > >> deprecation
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But
> the
> > > > > section
> > > > > > > > >> seems to
> > > > > > > > >> > > > have
> > > > > > > > >> > > > > > > > multiple typos.
> > > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2,
> B
> > > > > > > immediately
> > > > > > > > >> gets into
> > > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > > incorrect.
> > > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3,
> C
> > > seems
> > > > > to
> > > > > > > get
> > > > > > > > >> into
> > > > > > > > >> > > > > > epoch=3,
> > > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > > has A -
> > > > > > > > >> epoch=2,
> > > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we
> support
> > > the
> > > > > > > upgrade
> > > > > > > > >> from
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > old
> > > > > > > > >> > > > > > > > version to new one?
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Thanks,
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > Jun
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > > > > wrote:
> > > > > > > > >> > > > > > > >
> > > > > > > > >> > > > > > > > > Hi all,
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > The KIP states that we will re-implement
> the
> > > > > > > coordinator
> > > > > > > > >> in
> > > > > > > > >> > > > Java. I
> > > > > > > > >> > > > > > > > > discussed this offline with a few folks
> and
> > > folks
> > > > > are
> > > > > > > > >> concerned
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > > > > we could introduce many regressions in
> the old
> > > > > > > protocol
> > > > > > > > >> if we do
> > > > > > > > >> > > > so.
> > > > > > > > >> > > > > > > > > Therefore, I am going to remove this
> statement
> > > > > from
> > > > > > > the
> > > > > > > > >> KIP. It
> > > > > > > > >> > > > is an
> > > > > > > > >> > > > > > > > > implementation detail after all so it
> does not
> > > > > have
> > > > > > > to be
> > > > > > > > >> > > > decided at
> > > > > > > > >> > > > > > > > > this stage. We will likely start by
> trying to
> > > > > > > refactor the
> > > > > > > > >> > > > current
> > > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > Cheers,
> > > > > > > > >> > > > > > > > > David
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David
> Jacot <
> > > > > > > > >> djacot@confluent.io
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > "Empty,
> > > > > > > assigning,
> > > > > > > > >> > > > > > reconciling,
> > > > > > > > >> > > > > > > > > stable,
> > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> States
> > > > > section,
> > > > > > > > >> right?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > This sentence does not refer to those
> group
> > > > > states
> > > > > > > but
> > > > > > > > >> rather
> > > > > > > > >> > > > to a
> > > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > > refers to
> > > > > the
> > > > > > > > >> entire
> > > > > > > > >> > > > state of
> > > > > > > > >> > > > > > > > > > group coordinator which is replicated
> via
> > > the
> > > > > log
> > > > > > > > >> layer. I will
> > > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> machine
> > > is
> > > > > > > modelled
> > > > > > > > >> as an
> > > > > > > > >> > > > event
> > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> > > the new
> > > > > > > quorum
> > > > > > > > >> > > > > > controller. We
> > > > > > > > >> > > > > > > > > > will have N threads to process events.
> Each
> > > > > > > > >> __consumer_offsets
> > > > > > > > >> > > > > > > > > > partition is assigned to a unique
> thread and
> > > > > all the
> > > > > > > > >> events
> > > > > > > > >> > > > (e.g.
> > > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are
> processed by
> > > this
> > > > > > > > >> thread. This
> > > > > > > > >> > > > > > simplify
> > > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > > simulation
> > > > > > > testing
> > > > > > > > >> for the
> > > > > > > > >> > > > > > group
> > > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> per
> > > > > > > > >> *__consumer_offsets*
> > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> group"
> > > > > owned by
> > > > > > > a
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator?
> > > > > > > > >> > > > > > > > > For
> > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> owns 2
> > > > > consumer
> > > > > > > > >> groups, and
> > > > > > > > >> > > > > > both
> > > > > > > > >> > > > > > > > > exist in
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > > state
> > > > > > > machine
> > > > > > > > >> for it,
> > > > > > > > >> > > > or
> > > > > > > > >> > > > > > 2?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from
> there, I
> > > > > think.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > "*group.coordinator.threads"
> > > > > *is
> > > > > > > the
> > > > > > > > >> number
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > threads
> > > > > > > > >> > > > > > > > > used
> > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > wondering
> > > > > if
> > > > > > > the
> > > > > > > > >> purpose
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > threads
> > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> consumer
> > > > > group
> > > > > > > (or
> > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> computation,
> > > why
> > > > > > > should we
> > > > > > > > >> need
> > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability
> to
> > > > > shard the
> > > > > > > > >> > > > processing as
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > > does
> > > > > the
> > > > > > > > >> default
> > > > > > > > >> > > > session
> > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> max(60s)? I
> > > > > thought
> > > > > > > the
> > > > > > > > >> min/max
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> bound of
> > > it,
> > > > > no?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > > 30s The
> > > > > > > > >> timeout to
> > > > > > > > >> > > > detect
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > > protocol.
> > > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> int
> > > > > 45s The
> > > > > > > > >> minimum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> int
> > > > > 60s The
> > > > > > > > >> maximum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > > session
> > > > > > > timeout
> > > > > > > > >> should
> > > > > > > > >> > > > be 45s
> > > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> are
> > > > > [range,
> > > > > > > > >> uniform],
> > > > > > > > >> > > > > > which means
> > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > > like to
> > > > > > > know
> > > > > > > > >> why not
> > > > > > > > >> > > > > > uniform
> > > > > > > > >> > > > > > > > > one? I
> > > > > > > > >> > > > > > > > > > > thought usually users will choose
> uniform
> > > > > assignor
> > > > > > > > >> (former
> > > > > > > > >> > > > sticky
> > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> other
> > > > > reason
> > > > > > > we
> > > > > > > > >> choose
> > > > > > > > >> > > > range
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > uniform
> > > > > The
> > > > > > > > >> server side
> > > > > > > > >> > > > > > assignors.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > The order on the server side has no
> > > influence
> > > > > > > because
> > > > > > > > >> the
> > > > > > > > >> > > > client
> > > > > > > > >> > > > > > must
> > > > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> > > There
> > > > > is no
> > > > > > > > >> default
> > > > > > > > >> > > > in the
> > > > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > > > specified
> > > > > > > by
> > > > > > > > >> the
> > > > > > > > >> > > > client,
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > > request is rejected. The default client
> > > value
> > > > > for
> > > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform`
> though.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Thanks for your very good comments,
> Luke. I
> > > hope
> > > > > > > that my
> > > > > > > > >> > > > answers
> > > > > > > > >> > > > > > help
> > > > > > > > >> > > > > > > > > > to clarify things. I will update the
> KIP as
> > > well
> > > > > > > based
> > > > > > > > >> on your
> > > > > > > > >> > > > > > > > > > feedback.
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > > >> > > > > > > > > > David
> > > > > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke
> Chen <
> > > > > > > > >> showuon@gmail.com>
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > > mentioned:
> > > > > > > > >> > > > > > > > > > > > The new group coordinator will have
> a
> > > state
> > > > > > > machine
> > > > > > > > >> per
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> > > each
> > > > > state
> > > > > > > > >> machine is
> > > > > > > > >> > > > > > modelled
> > > > > > > > >> > > > > > > > > as an
> > > > > > > > >> > > > > > > > > > > event loop. Those state machines will
> be
> > > > > executed
> > > > > > > in
> > > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > > "Empty,
> > > > > > > assigning,
> > > > > > > > >> > > > > > reconciling,
> > > > > > > > >> > > > > > > > > stable,
> > > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group
> States
> > > > > section,
> > > > > > > > >> right?
> > > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state
> machine
> > > is
> > > > > > > modelled
> > > > > > > > >> as an
> > > > > > > > >> > > > event
> > > > > > > > >> > > > > > > > > loop"?
> > > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine
> per
> > > > > > > > >> *__consumer_offsets*
> > > > > > > > >> > > > > > > > > partitions?
> > > > > > > > >> > > > > > > > > > > Not a state machine "per consumer
> group"
> > > > > owned by
> > > > > > > a
> > > > > > > > >> group
> > > > > > > > >> > > > > > coordinator?
> > > > > > > > >> > > > > > > > > For
> > > > > > > > >> > > > > > > > > > > example, if one group coordinator
> owns 2
> > > > > consumer
> > > > > > > > >> groups, and
> > > > > > > > >> > > > > > both
> > > > > > > > >> > > > > > > > > exist in
> > > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > > state
> > > > > > > machine
> > > > > > > > >> for it,
> > > > > > > > >> > > > or
> > > > > > > > >> > > > > > 2?
> > > > > > > > >> > > > > > > > > > > 1.4. I know the
> > > "*group.coordinator.threads"
> > > > > *is
> > > > > > > the
> > > > > > > > >> number
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > threads
> > > > > > > > >> > > > > > > > > used
> > > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > > wondering
> > > > > if
> > > > > > > the
> > > > > > > > >> purpose
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > threads
> > > > > > > > >> > > > > > > > > > > is only to keep the state of each
> consumer
> > > > > group
> > > > > > > (or
> > > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > > >> > > > > > > > > > > partitions?), and no heavy
> computation,
> > > why
> > > > > > > should we
> > > > > > > > >> need
> > > > > > > > >> > > > > > > > > multi-threads
> > > > > > > > >> > > > > > > > > > > here?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > > configs:
> > > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > > does
> > > > > the
> > > > > > > > >> default
> > > > > > > > >> > > > session
> > > > > > > > >> > > > > > > > > timeout not
> > > > > > > > >> > > > > > > > > > > locate between min (45s) and
> max(60s)? I
> > > > > thought
> > > > > > > the
> > > > > > > > >> min/max
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > > > timeout is to define lower/upper
> bound of
> > > it,
> > > > > no?
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > > 30s The
> > > > > > > > >> timeout to
> > > > > > > > >> > > > detect
> > > > > > > > >> > > > > > client
> > > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > > protocol.
> > > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms
> int
> > > > > 45s The
> > > > > > > > >> minimum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms
> int
> > > > > 60s The
> > > > > > > > >> maximum
> > > > > > > > >> > > > > > session
> > > > > > > > >> > > > > > > > > timeout.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor
> are
> > > > > [range,
> > > > > > > > >> uniform],
> > > > > > > > >> > > > > > which means
> > > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > > like to
> > > > > > > know
> > > > > > > > >> why not
> > > > > > > > >> > > > > > uniform
> > > > > > > > >> > > > > > > > > one? I
> > > > > > > > >> > > > > > > > > > > thought usually users will choose
> uniform
> > > > > assignor
> > > > > > > > >> (former
> > > > > > > > >> > > > sticky
> > > > > > > > >> > > > > > > > > assinor)
> > > > > > > > >> > > > > > > > > > > for better evenly distribution. Any
> other
> > > > > reason
> > > > > > > we
> > > > > > > > >> choose
> > > > > > > > >> > > > range
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > as default?
> > > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > > uniform
> > > > > The
> > > > > > > > >> server side
> > > > > > > > >> > > > > > assignors.
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > > >> > > > > > > > > > > Luke
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke
> Chen
> > > <
> > > > > > > > >> showuon@gmail.com
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka
> Connect
> > > > > > > > >> integrating with
> > > > > > > > >> > > > > > KIP-848,
> > > > > > > > >> > > > > > > > > but I
> > > > > > > > >> > > > > > > > > > > > think we should have a separate
> > > discussion
> > > > > > > thread
> > > > > > > > >> for the
> > > > > > > > >> > > > Kafka
> > > > > > > > >> > > > > > > > > Connect
> > > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With
> New
> > > > > Consumer
> > > > > > > > >> Rebalance
> > > > > > > > >> > > > > > Protocol
> > > > > > > > >> > > > > > > > > [1],
> > > > > > > > >> > > > > > > > > > > > and let this discussion thread
> focus on
> > > > > consumer
> > > > > > > > >> rebalance
> > > > > > > > >> > > > > > protocol,
> > > > > > > > >> > > > > > > > > WDYT?
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM
> Sagar <
> > > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > > feedback.
> > > > > > > Looks
> > > > > > > > >> like
> > > > > > > > >> > > > there's
> > > > > > > > >> > > > > > > > > agreement on
> > > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> > > would
> > > > > > > revisit
> > > > > > > > >> the doc
> > > > > > > > >> > > > and
> > > > > > > > >> > > > > > see
> > > > > > > > >> > > > > > > > > what
> > > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM
> David
> > > Jacot
> > > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > > document.
> > > > > > > That's
> > > > > > > > >> really
> > > > > > > > >> > > > > > helpful. I
> > > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> > > Connect
> > > > > > > and the
> > > > > > > > >> > > > Consumer
> > > > > > > > >> > > > > > could
> > > > > > > > >> > > > > > > > > share
> > > > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> > > main
> > > > > > > > >> difference is
> > > > > > > > >> > > > that
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > Consumer
> > > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to
> members
> > > > > whereas
> > > > > > > > >> Connect
> > > > > > > > >> > > > assigns
> > > > > > > > >> > > > > > tasks
> > > > > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > > forward:
> > > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed
> APIs to
> > > > > support
> > > > > > > > >> different
> > > > > > > > >> > > > > > resource
> > > > > > > > >> > > > > > > > > types
> > > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.);
> or
> > > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > > > > Connect. The
> > > > > > > > >> dedicated
> > > > > > > > >> > > > APIs
> > > > > > > > >> > > > > > > > > would be
> > > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > > different on
> > > > > the
> > > > > > > > >> > > > > > content/resources and
> > > > > > > > >> > > > > > > > > > > >> > they would rely on the same
> engine
> > > on the
> > > > > > > > >> coordinator
> > > > > > > > >> > > > side.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2)
> because
> > > I am
> > > > > > > not a
> > > > > > > > >> fan of
> > > > > > > > >> > > > > > > > > overcharging
> > > > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> > > That
> > > > > being
> > > > > > > > >> said, I am
> > > > > > > > >> > > > not
> > > > > > > > >> > > > > > > > > opposed to
> > > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way
> to
> > > do
> > > > > it.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > > discuss
> > > > > it
> > > > > > > here
> > > > > > > > >> for now
> > > > > > > > >> > > > in
> > > > > > > > >> > > > > > order
> > > > > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is
> compatible
> > > with
> > > > > what
> > > > > > > we
> > > > > > > > >> will do
> > > > > > > > >> > > > for
> > > > > > > > >> > > > > > > > > Connect in
> > > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM
> David
> > > > > Jacot <
> > > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I
> will go
> > > > > through
> > > > > > > and
> > > > > > > > >> address
> > > > > > > > >> > > > > > your
> > > > > > > > >> > > > > > > > > comments
> > > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for
> your
> > > > > > > feedback.
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> > > Gregory
> > > > > > > Harris
> > > > > > > > >> <
> > > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > > wonderful
> > > > > to see
> > > > > > > > >> > > > cooperative
> > > > > > > > >> > > > > > > > > rebalancing
> > > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected
> Alternatives'
> > > > > section
> > > > > > > > >> describes how
> > > > > > > > >> > > > > > member
> > > > > > > > >> > > > > > > > > epoch
> > > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > > >> > > > > > > > > > > >> > > > advance in step with the
> group
> > > epoch
> > > > > and
> > > > > > > > >> assignment
> > > > > > > > >> > > > > > epoch
> > > > > > > > >> > > > > > > > > values. I
> > > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for
> the
> > > > > reasons
> > > > > > > > >> described
> > > > > > > > >> > > > in
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > KIP. When
> > > > > > > > >> > > > > > > > > > > >> the
> > > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > > assigning
> > > > > > > > >> partitions to a
> > > > > > > > >> > > > > > worker,
> > > > > > > > >> > > > > > > > > what
> > > > > > > > >> > > > > > > > > > > >> member
> > > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > > assignment
> > > > > > > use?
> > > > > > > > >> Are
> > > > > > > > >> > > > member
> > > > > > > > >> > > > > > epochs
> > > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > > correspond
> > > > > to
> > > > > > > > >> multiple
> > > > > > > > >> > > > > > different
> > > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> > > field
> > > > > > > opaque
> > > > > > > > >> to the
> > > > > > > > >> > > > group
> > > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > > >> > > > > > > > > > > >> > > > not, should custom
> client-side
> > > > > assignor
> > > > > > > > >> > > > implementations
> > > > > > > > >> > > > > > > > > interact
> > > > > > > > >> > > > > > > > > > > >> with
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> > > common
> > > > > > > meaning
> > > > > > > > >> agreed
> > > > > > > > >> > > > > > upon? If
> > > > > > > > >> > > > > > > > > so, what
> > > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> > > field
> > > > > over
> > > > > > > > >> including
> > > > > > > > >> > > > such
> > > > > > > > >> > > > > > > > > functionality
> > > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 3. The following is included
> in
> > > the
> > > > > KIP:
> > > > > > > > >> "Thanks to
> > > > > > > > >> > > > > > this, the
> > > > > > > > >> > > > > > > > > input
> > > > > > > > >> > > > > > > > > > > >> of
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > client side assignor is
> entirely
> > > > > driven
> > > > > > > by
> > > > > > > > >> the group
> > > > > > > > >> > > > > > > > > coordinator.
> > > > > > > > >> > > > > > > > > > > >> The
> > > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > > responsible for
> > > > > > > > >> maintaining
> > > > > > > > >> > > > any
> > > > > > > > >> > > > > > state
> > > > > > > > >> > > > > > > > > besides
> > > > > > > > >> > > > > > > > > > > >> its
> > > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does
> this
> > > mean
> > > > > > > that the
> > > > > > > > >> > > > > > client-side
> > > > > > > > >> > > > > > > > > assignor
> > > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > > non-Metadata
> > > > > state
> > > > > > > > >> (such as
> > > > > > > > >> > > > > > partition
> > > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > > topics,
> > > > > etc),
> > > > > > > or
> > > > > > > > >> that
> > > > > > > > >> > > > > > additional
> > > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are
> separate
> > > > > classes
> > > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > > >> > > > > >
> > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > > >> > > > > > that
> > > > > > > > >> > > > > > > > > seem to
> > > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > > > > possible for
> > > > > > > > >> these two
> > > > > > > > >> > > > > > > > > implementations
> > > > > > > > >> > > > > > > > > > > >> to
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> > > promote
> > > > > > > feature
> > > > > > > > >> parity
> > > > > > > > >> > > > of
> > > > > > > > >> > > > > > > > > server-side
> > > > > > > > >> > > > > > > > > > > >> and
> > > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and
> would
> > > also
> > > > > > > > >> facilitate
> > > > > > > > >> > > > > > operational
> > > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > > >> > > > > > > > > > > >> > > > certain situations. For
> example,
> > > if a
> > > > > > > > >> server-side
> > > > > > > > >> > > > > > assignor
> > > > > > > > >> > > > > > > > > has some
> > > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > > > deploying the
> > > > > > > > >> patched
> > > > > > > > >> > > > > > assignor to
> > > > > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer
> group
> > > to a
> > > > > > > > >> client-side
> > > > > > > > >> > > > > > assignor
> > > > > > > > >> > > > > > > > > may be
> > > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > > >> > > > > > > > > > > >> > > > and less risky than patching
> all
> > > of
> > > > > the
> > > > > > > > >> brokers.
> > > > > > > > >> > > > With
> > > > > > > > >> > > > > > the
> > > > > > > > >> > > > > > > > > currently
> > > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > > non-trivial
> > > > > > > > >> > > > reimplementation
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > have
> > > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two
> APIs
> > > have
> > > > > > > diverged
> > > > > > > > >> > > > > > significantly,
> > > > > > > > >> > > > > > > > > then it
> > > > > > > > >> > > > > > > > > > > >> is
> > > > > > > > >> > > > > > > > > > > >> > > > possible that a
> reimplementation
> > > > > would
> > > > > > > not be
> > > > > > > > >> > > > possible.
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39
> AM
> > > Sagar
> > > > > <
> > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence
> page to
> > > > > discuss
> > > > > > > how
> > > > > > > > >> Connect
> > > > > > > > >> > > > > > would
> > > > > > > > >> > > > > > > > > need to
> > > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > > > protocol.
> > > > > > > > >> Here's the
> > > > > > > > >> > > > page:
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> >
> > > > > > > > >> > > > > > > > > > > >>
> > > > > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >
> > > > > > > > >> > > >
> > > > > > > > >>
> > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish
> and I
> > > have
> > > > > > > tried
> > > > > > > > >> to keep
> > > > > > > > >> > > > a
> > > > > > > > >> > > > > > format
> > > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what
> you
> > > > > think.
> > > > > > > Also,
> > > > > > > > >> do you
> > > > > > > > >> > > > > > think this
> > > > > > > > >> > > > > > > > > > > >> should
> > > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate
> discussion
> > > > > thread
> > > > > > > or
> > > > > > > > >> is this
> > > > > > > > >> > > > one
> > > > > > > > >> > > > > > fine?
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at
> 7:37 AM
> > > > > Sagar <
> > > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> > > doc on
> > > > > > > Kafka
> > > > > > > > >> Streams.
> > > > > > > > >> > > > > > Sure, I
> > > > > > > > >> > > > > > > > > would do
> > > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> > > such a
> > > > > > > > >> document.
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at
> 4:47
> > > AM
> > > > > > > Guozhang
> > > > > > > > >> Wang <
> > > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> > > could
> > > > > come
> > > > > > > > >> back with
> > > > > > > > >> > > > some
> > > > > > > > >> > > > > > > > > analysis on
> > > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect
> side
> > > > > > > integration
> > > > > > > > >> with
> > > > > > > > >> > > > the new
> > > > > > > > >> > > > > > > > > protocol;
> > > > > > > > >> > > > > > > > > > > >> so
> > > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on
> the
> > > new
> > > > > > > "protocol
> > > > > > > > >> type"
> > > > > > > > >> > > > we
> > > > > > > > >> > > > > > did not
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Regarding the version, I would rather add this later when we have a
clear use case for it. It seems to me that we are speculating here. I
understand your point but I am not fully convinced by the solution at
the moment. Would you agree with this?

Regarding the onAssignment, I was thinking about the case where a task
is promoted from standby to active. In this case, having onAssignment
with the combined partitions could make it difficult, no? I am
thinking that the assignor will have to remove the standby based on
the metadata but it won't know what to do after that. If the partition
is assigned directly, it can convert it to an active task. On the
other hand, if the partition is not available yet, it would have to
keep it as a standby until the partition is really assigned. It seems
to be that the assignor won't have the information to make this
decision, no? In this case, decouple the "why" from the "when" seems
to make things harder. I am not so familiar with Streams though so my
intuition could be wrong here.

David

On Mon, Sep 26, 2022 at 7:26 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Regarding the version, what I was thinking is that in the HB request, for
> "serverAssignor" field, instead of just having it as a single string field,
> maybe we could consider also making it a structure that includes: name,
> minimumVersion, maximumVersion. Where the minimumVersion/maximumVersion
> means the versions of the server assignor that the client work best with.
> That being said, I agree with you that such information may also be
> inferred elsewhere e.g. by looking into the "rackId" field, and see if it
> contains a hyphen or not etc. All I was wondering is that, if such version
> information would be useful for the server assignors to determine its
> actual assignment logic. I do not feel very strong about this one though
> --- even if we do not add it now, we can potentially add later, it's just
> that changing a single string field to a structure would be hard for
> compatibility and we'd then probably have to add top-level fields.
>
> Regarding the `onAssignment` logic, again my train of thoughts is that, if
> users want to know exactly when a partition is assigned / revoked, they
> would be leveraging on the rebalance callbacks, as that's what people
> should rely on to determine "when" partitions are assigned. The
> `onAssignment` should be used for getting "why" such partition assignment
> decision is made, and hence returning `combined partitions` would be okay.
> Streams e.g. implement both rebalance callbacks and the assignors, and it
> gets the "when" from the former (and create/close active tasks accordingly)
> and the "why" from the latter (and update its global info bookkeeping as
> well as standby maintenance accordingly). Most users would be just
> interested in the rebalance callback, and not implement their own assignor
> at all if they do not care about "why" as they trust the server assignors
> would take good care of those, and only about "when". So if we did build
> such two types of APIs from scratch, I'd indeed feel that not providing the
> partitions but only the metadata for `onAssignment` may be less confusing
> and push users to separate the usage of these two more clearly, but since
> we already introduced partitions in `onAssignment` for compatibility I'm
> less keen on removing them.
>
>
> Guozhang
>
> On Mon, Sep 26, 2022 at 6:55 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Guozhang,
> >
> > Regarding the version, my understanding is that the version would be
> > either the client software version or the request version, is this
> > correct? If so, we could indeed pass this information down to the
> > assignor via the interface. One way would be to pass a "server
> > context" to the assignor and that context would include that
> > information (and perhaps more). Is this what you are looking for?
> >
> > Regarding the onAssignment, I think that I understand your point. I
> > suppose that the assignor could also be clever and keep track of the
> > last metadata to decide whether it has to do something or not. One
> > question that is still not clear to me is whether the assignor needs
> > to know all the assigned partitions upfront regardless of whether they
> > are already revoked or not. Do you think that we need this as well?
> >
> > From an API perspective, we could have something like
> > onAssignment(Metadata(version, reason, metadata, assigned partitions,
> > pending partitions)). Where the assigned partitions are the partitions
> > ready to be used and the pending partitions are the one assigned to
> > the member but not revoked yet. I find it a bit weird that this method
> > would be called only once because the assignor would not know when the
> > pending partitions changes. That does not look like a clean API. An
> > alternative would be to use onAssignment(Metadata(version, reason,
> > metadata, combined partitions)) but this seems error prone because it
> > is not clear whether a partition is usable or not. Or do you think
> > that we should not provide the partitions but only the metadata?
> >
> > Best,
> > David
> >
> > On Fri, Sep 23, 2022 at 9:40 PM Guozhang Wang <wa...@gmail.com> wrote:
> > >
> > > Hello David,
> > >
> > > On Fri, Sep 23, 2022 at 2:00 AM David Jacot <djacot@confluent.io.invalid
> > >
> > > wrote:
> > >
> > > > Hey,
> > > >
> > > > > Just to clarify I was asking about the `version` of the assignor
> > (i.e. up
> > > > to what version that the client would support), and I do agree we
> > would not
> > > > need metadata. What I have in mind is that, for some specific built-in
> > > > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > > > newer version we would have a hierarchical rack ID string format, like
> > > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > > would still be in old format. In this case, the broker then needs to
> > choose
> > > > the old versioned assignor. I'm probably making something up here for
> > rack
> > > > aware assignors, but I'm wondering if in general such an
> > "auto-downgrade"
> > > > behavior would be needed still for broker-side assignor, and if yes
> > would
> > > > "version" still be useful.
> > > >
> > > > Got it. That's an interesting thought. I think that the issue is that
> > > > the client will never tell you which version of the server-side
> > > > assignor should be used. Do you think that the coordinator would
> > > > downgrade the version if the assignment fails with a higher version? I
> > > > tend to believe that this should be handled within the assignor
> > > > itself. In the example that you mentioned, the assignor would have to
> > > > handle all the cases. I am not really convinced that we need this at
> > > > the moment.
> > > >
> > > > The version from the client side would not be indicating the broker
> > which
> > > version to use, but rather which version the client would "work best
> > with".
> > > Such a "version" field would not be settible by the users, since they
> > will
> > > be hard-codedly bumped when the Kafka byte code version bumped.
> > > Back to the rack aware assignor example, if the older versioned client
> > does
> > > not have a hierarchical rack ID, however if the assignment returned to
> > them
> > > is assuming a hierarchical rack structure, it may not reflect the best
> > > workload balance among those new and old versioned clients. That means,
> > > when receiving the members subscriptions at the server side, if the
> > > versions from all these members are different, the broker's assignor may
> > > need to consider using the lower version logic to do the assignment. So
> > yes
> > > the assignor would indeed have to handle all such cases, but it needs to
> > do
> > > so such that if there are clients who would not work with certain new
> > > logic, it would then handle such cases automatically by e.g. still using
> > an
> > > older versioned logic.
> > >
> > >
> > >
> > > > > Okay, my understanding is that the calling ordering of these
> > callbacks
> > > > would be like the following:
> > > >
> > > > Yes, your examples look right.
> > > >
> > > > > I'm wondering if we would still call onAssignment just once, that
> > encodes
> > > > all the assignment for this rebalance, including all the partitions
> > that
> > > > should be assigned to the member but not yet assigned since they have
> > not
> > > > been revoked by others. In that case the call ordering would be:
> > > >
> > > > Interesting. Is there a case for Streams where having the full
> > > > assignment is beneficial? For instance, I can think of the following
> > > > case. When a standby task is promoted to an active task, the metadata
> > > > would not contain the standby task anymore and the assignment may not
> > > > have the partition yet. In this case, Streams would stop the standby
> > > > tasks but not have the active task yet if my understanding of Streams
> > > > is correct. So knowing the full assignment could be helpful here.
> > > >
> > > > If we want to do this, we could structure the assignment given to the
> > > > member as follow: version, error, metadata, assigned partitions,
> > > > pending partitions, where the pending partitions would be the one
> > > > assigned to this member but not yet available. What do you think?
> > > >
> > > > Regarding onAssignment being called only once, I am not sure to fully
> > > > grasp the benefit yet. Does the assignor really care about this? In
> > > > the end, the epoch does not really matter for the assignor because it
> > > > has to converge its state to the desired state anyway.
> > > >
> > > > Here's my rationale (maybe rephased a bit :P ): the implementers of
> > > rebalance listener and assignor are two groups of people, and most users
> > > fall into the former group, while only very few people fall into the
> > later
> > > group. For rebalance listener implementers, they just want to know when a
> > > partition is actually revoked or assigned to the consumer and reacts to
> > it,
> > > for this purpose, `onPartitionsRevoked` and `onPartitionsAssigned` would
> > be
> > > triggered interleavingly upon `poll` calls across rebalances. The usual
> > > logic for such rebalance listeners are metrics reporting, committing
> > > offsets (if they do not use Kafka for that), etc. They would not care
> > which
> > > calls are from which rebalances --- in the past with eager rebalance, it
> > > maybe that each rebalance is associated with exactly a
> > > `onPartitionsRevoked` first and then a `onPartitionsAssigned`, but it
> > would
> > > no longer the cases now.
> > >
> > > The implementers of the assignor though, would care about "how the
> > > assignment was made", that includes from which rebalance a certain
> > > revoke/assign decision was made, based on what metadata such assignment
> > is
> > > made, etc. And that's the whole point of the `onAssignment` function
> > since
> > > otherwise they can just rely on the listeners. They usually
> > implementation
> > > logic of this callback is to e.g. bookkeep the assignment decision
> > driving
> > > factors a.k.a. the metadata, global information that needs to be
> > propagated
> > > to all members, etc. Take Streams as an example, the active processing
> > > tasks go along with the assigned partitions, and we can always just
> > > incrementally create / close them upon each rebalance listener triggers,
> > > when certain partitions are revoked or assigned together; standby tasks
> > > however are encoded with the metadata, and we can only know which standby
> > > tasks should we get / drop based on the `onAssignment` function, and in
> > > fact the creation of such tasks as a result of the metadata bookkeeping
> > > does not need to wait until all the partitions that are yet-assigned have
> > > been completely assigned to the member. Such information may not always
> > be
> > > updatable in an incremental manner as the partitions-revoked /
> > > partitions-assigned. In such a case, it's better to just trigger this
> > > function "once per decision made" i.e. once per rebalance generation.
> > >
> > >
> > > Guozhang
> > >
> > >
> > >
> > > > Best,
> > > > David
> > > >
> > > > On Thu, Sep 22, 2022 at 6:01 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > > >
> > > > > Hi David, thanks for all the detailed explanations. I think they all
> > make
> > > > > sense. Just want to have a couple follow-ups here:
> > > > >
> > > > > > I don't really see the benefits here because server side assignors
> > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > are not supposed to generate metadata nor to receive metadata from
> > the
> > > > > members.
> > > > >
> > > > > Just to clarify I was asking about the `version` of the assignor
> > (i.e. up
> > > > > to what version that the client would support), and I do agree we
> > would
> > > > not
> > > > > need metadata. What I have in mind is that, for some specific
> > built-in
> > > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> > in a
> > > > > newer version we would have a hierarchical rack ID string format,
> > like
> > > > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > > > would still be in old format. In this case, the broker then needs to
> > > > choose
> > > > > the old versioned assignor. I'm probably making something up here for
> > > > rack
> > > > > aware assignors, but I'm wondering if in general such an
> > "auto-downgrade"
> > > > > behavior would be needed still for broker-side assignor, and if yes
> > would
> > > > > "version" still be useful.
> > > > >
> > > > > > Yeah, that's right. Within a rebalance, `onAssignment` is called
> > once
> > > > > when the member transitions to a new epoch. This one contains the
> > full
> > > > > metadata provided by the client side assignor. Then, `onAssignment`
> > > > > can be called max N times where N is the number of partitions pending
> > > > > revocation by other members. Let me try to clarify this in the KIP.
> > > > >
> > > > > Okay, my understanding is that the calling ordering of these
> > callbacks
> > > > > would be like the following:
> > > > >
> > > > > ----------------------------------------
> > > > > onPartitionsRevoked();   // just once, since we do not really need
> > > > > to revoke incrementally.
> > > > >
> > > > > onAssignment();    // the first call, with epoch incremented
> > > > > onPartitionsAssigned();   // paired with the onAssignment
> > > > >
> > > > > onAssignment();              // the first onAssignment would bump up
> > the
> > > > > epoch, and the metadata reflected.
> > > > > onPartitionsAssigned();   // each time we get an additional
> > assignment,
> > > > we
> > > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > > ...
> > > > > onAssignment();
> > > > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > > > encoded
> > > > > metadata would not change, only the incrementally added partitions be
> > > > > reflected
> > > > >
> > > > > Is that the case?
> > > > >
> > > > > I'm wondering if we would still call onAssignment just once, that
> > encodes
> > > > > all the assignment for this rebalance, including all the partitions
> > that
> > > > > should be assigned to the member but not yet assigned since they
> > have not
> > > > > been revoked by others. In that case the call ordering would be:
> > > > >
> > > > > ----------------------------------------
> > > > > onPartitionsRevoked();   // just once
> > > > > onAssignment();    // just once, with epoch incremented, and metadata
> > > > > encoded changed, the "assignment" field also reflect the final target
> > > > > assignment
> > > > > onPartitionsAssigned();   // multiple times, which represent
> > > > incrementally
> > > > > added partitions
> > > > > ...
> > > > > onPartitionsAssigned();
> > > > >
> > > > > The motivation from this is that, most users would only implement the
> > > > > rebalance callback listeners and hence we'd definitely need to make
> > sure
> > > > > the semantics of that does not change much, and the time
> > > > > `onPartitionsAssigned` indicate the time when the partitions are
> > actually
> > > > > assigned to it; while for assignors, the `onAssignment` is used to
> > > > indicate
> > > > > what decision is made regarding for this member, i.e. when the
> > partitions
> > > > > are decided to be given to it, but not necessarily meant that it has
> > been
> > > > > given, since that time should be determined by the time of
> > > > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > > > implementers would not need to reason which `onAssignment` would be
> > the
> > > > > last one for this epoch.
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> > <djacot@confluent.io.invalid
> > > > >
> > > > > wrote:
> > > > >
> > > > > > Hi Guozhang,
> > > > > >
> > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > byte
> > > > > > field,
> > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > And
> > > > the
> > > > > > > rationale seems to be assuming that we should always be able to
> > do
> > > > the
> > > > > > > assignment at the broker-side assignor without errors.
> > Personally I
> > > > think
> > > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > > broker-side assignors, e.g. for some edge cases where some
> > subscribed
> > > > > > > topics are not recognized with the current broker's metadata.
> > What
> > > > do you
> > > > > > > think?
> > > > > >
> > > > > > Yeah, that seems reasonable. However, I wonder if it would be
> > better
> > > > > > to use an exception on the server side. This is what we usually do
> > for
> > > > > > server side plugins. On the client side, we use a field because the
> > > > > > errors are not defined in advance.
> > > > > >
> > > > > > Your comment also makes me think about what we should do when the
> > > > > > server side assignor fails. I suppose that we have to keep the
> > current
> > > > > > assignment until a new event occurs. For instance, in your example,
> > > > > > the coordinator would have to trigger a rebalance when unrecognized
> > > > > > topics are available. This would be part of the metadata
> > monitoring.
> > > > > >
> > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > reason/version/metadata compared with the broker-side
> > GroupMember. I
> > > > > > agree
> > > > > > > that broker-side assignor would not need reason/metadata since
> > they
> > > > are
> > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > E.g.
> > > > is
> > > > > > it
> > > > > > > possible that we evolve our broker-side built-in assignor but
> > the old
> > > > > > > versioned clients would not be able to work with the new
> > version, in
> > > > > > which
> > > > > > > case we need to let the broker being aware of this and upgrade
> > its
> > > > > > behavior
> > > > > > > to cooperate with the clients?
> > > > > >
> > > > > > I don't really see the benefits here because server side assignors
> > > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > > are not supposed to generate metadata nor to receive metadata from
> > the
> > > > > > members.
> > > > > >
> > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > instead
> > > > > > of
> > > > > > > including these three fields, what about just adding the
> > "Metadata"
> > > > field
> > > > > > > class which has these three fields? Also, there are two
> > "Metadata"
> > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > reason/version/metadata, and the second is just the encoded
> > metadata
> > > > > > bytes.
> > > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > > which
> > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > second
> > > > > > > bytebuffer field as metadataBytes?
> > > > > >
> > > > > > That's a good point. Let me try to rationalize this interface
> > based on
> > > > > > your suggestions.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > > > wrote:
> > > > > > >
> > > > > > > Hello David,
> > > > > > >
> > > > > > > Just had a few more nit questions about the public APIs:
> > > > > > >
> > > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> > byte
> > > > > > field,
> > > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> > And
> > > > the
> > > > > > > rationale seems to be assuming that we should always be able to
> > do
> > > > the
> > > > > > > assignment at the broker-side assignor without errors.
> > Personally I
> > > > think
> > > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > > broker-side assignors, e.g. for some edge cases where some
> > subscribed
> > > > > > > topics are not recognized with the current broker's metadata.
> > What
> > > > do you
> > > > > > > think?
> > > > > > >
> > > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > > reason/version/metadata compared with the broker-side
> > GroupMember. I
> > > > > > agree
> > > > > > > that broker-side assignor would not need reason/metadata since
> > they
> > > > are
> > > > > > > blackbox strings/bytes to the assignor, but what about version?
> > E.g.
> > > > is
> > > > > > it
> > > > > > > possible that we evolve our broker-side built-in assignor but
> > the old
> > > > > > > versioned clients would not be able to work with the new
> > version, in
> > > > > > which
> > > > > > > case we need to let the broker being aware of this and upgrade
> > its
> > > > > > behavior
> > > > > > > to cooperate with the clients?
> > > > > > >
> > > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > > instead
> > > > > > of
> > > > > > > including these three fields, what about just adding the
> > "Metadata"
> > > > field
> > > > > > > class which has these three fields? Also, there are two
> > "Metadata"
> > > > > > > currently in the APIs, the first is a class that encodes
> > > > > > > reason/version/metadata, and the second is just the encoded
> > metadata
> > > > > > bytes.
> > > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > > which
> > > > > > > then has a bytebuffer field Metadata, or instead naming the
> > second
> > > > > > > bytebuffer field as metadataBytes?
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > > wrote:
> > > > > > >
> > > > > > > > Hello David,
> > > > > > > >
> > > > > > > > Thanks for bringing this question up. I think the main
> > benefits as
> > > > you
> > > > > > > > listed is 2) above if it stays; just to clarify, we would only
> > be
> > > > able
> > > > > > to
> > > > > > > > save one round trip if the rebalance is still triggered by the
> > > > broker;
> > > > > > if
> > > > > > > > the rebalance is triggered by the client then the
> > num.round.trips
> > > > are
> > > > > > the
> > > > > > > > same:
> > > > > > > >
> > > > > > > > 1) With GroupPrepareAssignment:
> > > > > > > >
> > > > > > > > T0: client decides to do a new assignment, suppose it has
> > already
> > > > sent
> > > > > > a
> > > > > > > > HB and hence has to wait for it to return first since only one
> > > > request
> > > > > > /
> > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > > GroupPrepareAssignment request.
> > > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > GroupInstallAssignment
> > > > > > > > request.
> > > > > > > >
> > > > > > > > In total, two round trips.
> > > > > > > >
> > > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > > >
> > > > > > > > T0: client decides to do a new assignment, suppose it has
> > already
> > > > sent
> > > > > > a
> > > > > > > > HB and hence has to wait for it to return first since only one
> > > > request
> > > > > > /
> > > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > > T1: client receives the HB response, and then sends the new HB
> > > > request
> > > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > > T2: the HB response with the optional member metadata map is
> > > > returned.
> > > > > > > > T3: it calculates the new assignment, and sends a
> > > > > > GroupInstallAssignment
> > > > > > > > request.
> > > > > > > >
> > > > > > > > In total, two round trips as well.
> > > > > > > >
> > > > > > > > -----------------------------
> > > > > > > >
> > > > > > > > So to complete the full picture here, we'd need to modify both
> > HB
> > > > > > request
> > > > > > > > and response so that the client can also indicate a new
> > rebalance
> > > > via
> > > > > > the
> > > > > > > > HB request as well, right?
> > > > > > > >
> > > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > > additional
> > > > > > fields
> > > > > > > > encoding the full member (subscription) metadata and topic
> > > > metadata as
> > > > > > > > optional fields.
> > > > > > > >
> > > > > > > >
> > > > > > > > Guozhang
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > >> Hi all,
> > > > > > > >>
> > > > > > > >> During an offline conversation, someone asked why we need the
> > > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > > > instead
> > > > > > > >> provide the group state in the heartbeat response. This has a
> > few
> > > > > > > >> advantages: 1) it does not require using a special error code
> > to
> > > > > > > >> signal that a new assignment is required as the signal would
> > be
> > > > the
> > > > > > > >> provided group state; 2) it removes one round trip when a
> > client
> > > > side
> > > > > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > > > > >> response's definition quite large. I recall that I went with
> > the
> > > > > > > >> current approach due to this.
> > > > > > > >>
> > > > > > > >> Providing the group state in the heartbeat response is
> > appealing.
> > > > What
> > > > > > > >> do you guys think?
> > > > > > > >>
> > > > > > > >> Best,
> > > > > > > >> David
> > > > > > > >>
> > > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> > djacot@confluent.io>
> > > > > > wrote:
> > > > > > > >> >
> > > > > > > >> > Hi Guozhang,
> > > > > > > >> >
> > > > > > > >> > 1. I have added a reference to the relevant chapter instead
> > of
> > > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > > >> >
> > > > > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> > > > about
> > > > > > when
> > > > > > > >> > a rebalance should be triggered for the non-upgraded members
> > > > using
> > > > > > the
> > > > > > > >> > old protocol. The section mentions that a rebalance must be
> > > > > > triggered
> > > > > > > >> > when a new assignment is installed. This implies that the
> > group
> > > > > > epoch
> > > > > > > >> > was updated either by a native member or a non-upgraded
> > member.
> > > > For
> > > > > > > >> > the latter, the JoinGroup request would be the trigger. I
> > have
> > > > > > added a
> > > > > > > >> > reference to the relevant chapter in the "JoinGroup
> > Handling"
> > > > > > section
> > > > > > > >> > as well. Does that make sense?
> > > > > > > >> >
> > > > > > > >> > Thanks,
> > > > > > > >> > David
> > > > > > > >> >
> > > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > > wangguoz@gmail.com>
> > > > > > > >> wrote:
> > > > > > > >> > >
> > > > > > > >> > > Hello David,
> > > > > > > >> > >
> > > > > > > >> > > Alright I think that's sufficient. Just to make that
> > clear in
> > > > the
> > > > > > doc,
> > > > > > > >> > > could we update:
> > > > > > > >> > >
> > > > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > > > coordinator
> > > > > > > >> will
> > > > > > > >> > > trigger rebalance based on the HB's member metadata /
> > reason?
> > > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > > described
> > > > > > in
> > > > > > > >> "Group
> > > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > > >> > >
> > > > > > > >> > >
> > > > > > > >> > > Guozhang
> > > > > > > >> > >
> > > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > wrote:
> > > > > > > >> > >
> > > > > > > >> > > > Hi Guozhang,
> > > > > > > >> > > >
> > > > > > > >> > > > I thought that the assignor will always be consulted
> > when
> > > > the
> > > > > > next
> > > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > > heartbeat.
> > > > > > > >> This
> > > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > > rebalance by
> > > > > > > >> > > > setting the reason to a non-zero value or by changing
> > the
> > > > > > bytes. Do
> > > > > > > >> > > > you think that this is not sufficient? Are you
> > concerned by
> > > > the
> > > > > > > >> delay?
> > > > > > > >> > > >
> > > > > > > >> > > > Best,
> > > > > > > >> > > > David
> > > > > > > >> > > >
> > > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > > wangguoz@gmail.com>
> > > > > > > >> wrote:
> > > > > > > >> > > > >
> > > > > > > >> > > > > Hello David,
> > > > > > > >> > > > >
> > > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > > >> > > > >
> > > > > > > >> > > > > ```
> > > > > > > >> > > > > In this case, a new assignment is triggered by the
> > client
> > > > side
> > > > > > > >> > > > > assignor. When constructing the HB, the consumer will
> > > > always
> > > > > > > >> consult
> > > > > > > >> > > > > the client side assignor and propagate the
> > information to
> > > > the
> > > > > > > >> group
> > > > > > > >> > > > > coordinator. In other words, we don't expect users to
> > call
> > > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > > >> > > > > ```
> > > > > > > >> > > > >
> > > > > > > >> > > > > As I looked at the current PartitionAssignor's
> > interface,
> > > > we
> > > > > > > >> actually do
> > > > > > > >> > > > > not have a way yet to instruct how to construct the
> > next
> > > > HB
> > > > > > > >> request, e.g.
> > > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> > with a
> > > > new
> > > > > > > >> assignment,
> > > > > > > >> > > > > we'd need some customizable APIs inside the
> > > > PartitionAssignor
> > > > > > to
> > > > > > > >> indicate
> > > > > > > >> > > > > the next HB telling broker about so. WDYT about adding
> > > > such an
> > > > > > > >> API on the
> > > > > > > >> > > > > PartitionAssignor?
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > Guozhang
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > > >> <dj...@confluent.io.invalid>
> > > > > > > >> > > > > wrote:
> > > > > > > >> > > > >
> > > > > > > >> > > > > > Hi Jun,
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > I have updated the KIP to include your feedback. I
> > have
> > > > also
> > > > > > > >> tried to
> > > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > Best,
> > > > > > > >> > > > > > David
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > > djacot@confluent.io
> > > > > > > >> >
> > > > > > > >> > > > wrote:
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Hi Jun,
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> > answering
> > > > your
> > > > > > > >> questions
> > > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > of
> > > > the
> > > > > > KIP
> > > > > > > >> seem to
> > > > > > > >> > > > be
> > > > > > > >> > > > > > fewer
> > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > support of
> > > > > > > >> wildcard. A few
> > > > > > > >> > > > > > > > comments below.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > I would also add that the KIP removes the global
> > sync
> > > > > > barrier
> > > > > > > >> in the
> > > > > > > >> > > > > > > protocol which is essential to improve group
> > > > stability and
> > > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> > client by
> > > > > > moving
> > > > > > > >> most of
> > > > > > > >> > > > > > > the logic to the server side.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > to
> > > > > > support
> > > > > > > >> rolling
> > > > > > > >> > > > > > changing
> > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > > assignor
> > > > > > used
> > > > > > > >> by a
> > > > > > > >> > > > > > > majority of the members. This allows the group to
> > move
> > > > > > from
> > > > > > > >> one
> > > > > > > >> > > > > > > assignor to another by a roll. This is explained
> > in
> > > > the
> > > > > > > >> Assignor
> > > > > > > >> > > > > > > Selection chapter.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > it's
> > > > > > > >> required in
> > > > > > > >> > > > every
> > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > > filled? For
> > > > > > > >> example,
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > not
> > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > filled.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > The client is expected to set those fields in
> > case of
> > > > a
> > > > > > > >> connection
> > > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> > changed
> > > > since
> > > > > > > >> the last
> > > > > > > >> > > > > > > HB. The server populates those fields as long as
> > the
> > > > > > member
> > > > > > > >> is not
> > > > > > > >> > > > > > > fully reconciled - the member should acknowledge
> > that
> > > > it
> > > > > > has
> > > > > > > >> the
> > > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> > this in
> > > > the
> > > > > > KIP.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > affinity
> > > > > > > >> between the
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > and the broker is only considered during
> > fetching,
> > > > but
> > > > > > not
> > > > > > > >> during
> > > > > > > >> > > > > > assigning
> > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > assignment
> > > > > > is
> > > > > > > >> made,
> > > > > > > >> > > > there
> > > > > > > >> > > > > > is
> > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > > replicas of
> > > > > > > >> assigned
> > > > > > > >> > > > > > partitions
> > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > > > should use
> > > > > > > >> this
> > > > > > > >> > > > > > opportunity
> > > > > > > >> > > > > > > > to address this by including rack in
> > GroupMember.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's an interesting idea. I don't see any issue
> > with
> > > > > > adding
> > > > > > > >> the
> > > > > > > >> > > > rack
> > > > > > > >> > > > > > > to the members. I will do so.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > know
> > > > how
> > > > > > busy
> > > > > > > >> a group
> > > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> > it
> > > > seems
> > > > > > > >> that we
> > > > > > > >> > > > could
> > > > > > > >> > > > > > add
> > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > the
> > > > event
> > > > > > > >> loop is
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > actual work.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > failover
> > > > > > > >> handling? For
> > > > > > > >> > > > > > example,
> > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > with
> > > > the
> > > > > > > >> wildcard
> > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sure. When the new group coordinator takes over,
> > it
> > > > has
> > > > > > to:
> > > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > > >> > > > > > > * Trigger a new assignment if a client side
> > assignor
> > > > is
> > > > > > used.
> > > > > > > >> We
> > > > > > > >> > > > don't
> > > > > > > >> > > > > > > store the information about the member selected to
> > > > run the
> > > > > > > >> assignment
> > > > > > > >> > > > > > > so we have to start a new one.
> > > > > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > > > > >> subscriptions, and
> > > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > document
> > > > what
> > > > > > > >> the epoch
> > > > > > > >> > > > > > field
> > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > >> > > > > > reflect
> > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> > latest
> > > > > > group
> > > > > > > >> epoch.
> > > > > > > >> > > > > > > When the group state is updated, the group epoch
> > is
> > > > > > bumped so
> > > > > > > >> we use
> > > > > > > >> > > > > > > that one for all the change records related to the
> > > > update.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > > following
> > > > > > > >> > > > invariants
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> > a
> > > > member
> > > > > > > >> not to
> > > > > > > >> > > > get any
> > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > That's right. Here I meant that the members
> > provided
> > > > by
> > > > > > the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > > assignor
> > > > > > can
> > > > > > > >> not make
> > > > > > > >> > > > > > > up new member ids.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > epoch
> > > > > > equals to
> > > > > > > >> 0":
> > > > > > > >> > > > When
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > used?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > A member is expected to abandon all its
> > partitions and
> > > > > > > >> rejoins when
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this
> > case,
> > > > the
> > > > > > > >> group
> > > > > > > >> > > > > > > coordinator will have removed the member from the
> > > > group.
> > > > > > The
> > > > > > > >> member
> > > > > > > >> > > > > > > can rejoin the group with the same member id but
> > with
> > > > 0 as
> > > > > > > >> epoch. Let
> > > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> > to
> > > > > > trigger a
> > > > > > > >> > > > > > reassignment
> > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > > updating the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > with
> > > > the
> > > > > > > >> deprecation
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > In this case, a new assignment is triggered by the
> > > > client
> > > > > > side
> > > > > > > >> > > > > > > assignor. When constructing the HB, the consumer
> > will
> > > > > > always
> > > > > > > >> consult
> > > > > > > >> > > > > > > the client side assignor and propagate the
> > > > information to
> > > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > > > coordinator. In other words, we don't expect
> > users to
> > > > call
> > > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > section
> > > > > > > >> seems to
> > > > > > > >> > > > have
> > > > > > > >> > > > > > > > multiple typos.
> > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > immediately
> > > > > > > >> gets into
> > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > incorrect.
> > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > seems
> > > > to
> > > > > > get
> > > > > > > >> into
> > > > > > > >> > > > > > epoch=3,
> > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > has A -
> > > > > > > >> epoch=2,
> > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> > the
> > > > > > upgrade
> > > > > > > >> from
> > > > > > > >> > > > any
> > > > > > > >> > > > > > old
> > > > > > > >> > > > > > > > version to new one?
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > We will support upgrading from the consumer
> > protocol
> > > > > > version
> > > > > > > >> 3,
> > > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented
> > yet
> > > > so
> > > > > > the
> > > > > > > >> earliest
> > > > > > > >> > > > > > > version is unknown at the moment. This is
> > explained
> > > > in the
> > > > > > > >> migration
> > > > > > > >> > > > > > > plan chapter.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will update
> > > > the KIP
> > > > > > > >> based on
> > > > > > > >> > > > it
> > > > > > > >> > > > > > > next week.
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > Best,
> > > > > > > >> > > > > > > David
> > > > > > > >> > > > > > >
> > > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > > >> <ju...@confluent.io.invalid>
> > > > > > > >> > > > wrote:
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Hi, David,
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> > of
> > > > the
> > > > > > KIP
> > > > > > > >> seem to
> > > > > > > >> > > > be
> > > > > > > >> > > > > > fewer
> > > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> > support of
> > > > > > > >> wildcard. A few
> > > > > > > >> > > > > > > > comments below.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> > to
> > > > > > support
> > > > > > > >> rolling
> > > > > > > >> > > > > > changing
> > > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> > it's
> > > > > > > >> required in
> > > > > > > >> > > > every
> > > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > > filled? For
> > > > > > > >> example,
> > > > > > > >> > > > it's
> > > > > > > >> > > > > > not
> > > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> > filled.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > > affinity
> > > > > > > >> between the
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > and the broker is only considered during
> > fetching,
> > > > but
> > > > > > not
> > > > > > > >> during
> > > > > > > >> > > > > > assigning
> > > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > > assignment
> > > > > > is
> > > > > > > >> made,
> > > > > > > >> > > > there
> > > > > > > >> > > > > > is
> > > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > > replicas of
> > > > > > > >> assigned
> > > > > > > >> > > > > > partitions
> > > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > > > should use
> > > > > > > >> this
> > > > > > > >> > > > > > opportunity
> > > > > > > >> > > > > > > > to address this by including rack in
> > GroupMember.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> > know
> > > > how
> > > > > > busy
> > > > > > > >> a group
> > > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> > it
> > > > seems
> > > > > > > >> that we
> > > > > > > >> > > > could
> > > > > > > >> > > > > > add
> > > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> > the
> > > > event
> > > > > > > >> loop is
> > > > > > > >> > > > doing
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > actual work.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> > failover
> > > > > > > >> handling? For
> > > > > > > >> > > > > > example,
> > > > > > > >> > > > > > > > does it need to trigger the check if any group
> > with
> > > > the
> > > > > > > >> wildcard
> > > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> > document
> > > > what
> > > > > > > >> the epoch
> > > > > > > >> > > > > > field
> > > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > > >> ConsumerGroupMetadataValue
> > > > > > > >> > > > > > reflect
> > > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > > following
> > > > > > > >> > > > invariants
> > > > > > > >> > > > > > are
> > > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> > a
> > > > member
> > > > > > > >> not to
> > > > > > > >> > > > get any
> > > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> > epoch
> > > > > > equals to
> > > > > > > >> 0":
> > > > > > > >> > > > When
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> > used?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> > to
> > > > > > trigger a
> > > > > > > >> > > > > > reassignment
> > > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > > updating the
> > > > > > > >> assignor
> > > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> > with
> > > > the
> > > > > > > >> deprecation
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > > section
> > > > > > > >> seems to
> > > > > > > >> > > > have
> > > > > > > >> > > > > > > > multiple typos.
> > > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > > immediately
> > > > > > > >> gets into
> > > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > > incorrect.
> > > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> > seems
> > > > to
> > > > > > get
> > > > > > > >> into
> > > > > > > >> > > > > > epoch=3,
> > > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> > has A -
> > > > > > > >> epoch=2,
> > > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> > the
> > > > > > upgrade
> > > > > > > >> from
> > > > > > > >> > > > any
> > > > > > > >> > > > > > old
> > > > > > > >> > > > > > > > version to new one?
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Thanks,
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > Jun
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > wrote:
> > > > > > > >> > > > > > > >
> > > > > > > >> > > > > > > > > Hi all,
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > > > coordinator
> > > > > > > >> in
> > > > > > > >> > > > Java. I
> > > > > > > >> > > > > > > > > discussed this offline with a few folks and
> > folks
> > > > are
> > > > > > > >> concerned
> > > > > > > >> > > > that
> > > > > > > >> > > > > > > > > we could introduce many regressions in the old
> > > > > > protocol
> > > > > > > >> if we do
> > > > > > > >> > > > so.
> > > > > > > >> > > > > > > > > Therefore, I am going to remove this statement
> > > > from
> > > > > > the
> > > > > > > >> KIP. It
> > > > > > > >> > > > is an
> > > > > > > >> > > > > > > > > implementation detail after all so it does not
> > > > have
> > > > > > to be
> > > > > > > >> > > > decided at
> > > > > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > > > > refactor the
> > > > > > > >> > > > current
> > > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > Cheers,
> > > > > > > >> > > > > > > > > David
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > > > > >> djacot@confluent.io
> > > > > > > >> > > > >
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > "Empty,
> > > > > > assigning,
> > > > > > > >> > > > > > reconciling,
> > > > > > > >> > > > > > > > > stable,
> > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > section,
> > > > > > > >> right?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > This sentence does not refer to those group
> > > > states
> > > > > > but
> > > > > > > >> rather
> > > > > > > >> > > > to a
> > > > > > > >> > > > > > > > > > state machine replication (SMR). This
> > refers to
> > > > the
> > > > > > > >> entire
> > > > > > > >> > > > state of
> > > > > > > >> > > > > > > > > > group coordinator which is replicated via
> > the
> > > > log
> > > > > > > >> layer. I will
> > > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> > is
> > > > > > modelled
> > > > > > > >> as an
> > > > > > > >> > > > event
> > > > > > > >> > > > > > > > > loop"?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> > the new
> > > > > > quorum
> > > > > > > >> > > > > > controller. We
> > > > > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > > > > >> __consumer_offsets
> > > > > > > >> > > > > > > > > > partition is assigned to a unique thread and
> > > > all the
> > > > > > > >> events
> > > > > > > >> > > > (e.g.
> > > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by
> > this
> > > > > > > >> thread. This
> > > > > > > >> > > > > > simplify
> > > > > > > >> > > > > > > > > > concurrency and will enable us to do
> > simulation
> > > > > > testing
> > > > > > > >> for the
> > > > > > > >> > > > > > group
> > > > > > > >> > > > > > > > > > coordinator.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > >> *__consumer_offsets*
> > > > > > > >> > > > > > > > > partitions?
> > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > > > owned by
> > > > > > a
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator?
> > > > > > > >> > > > > > > > > For
> > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > consumer
> > > > > > > >> groups, and
> > > > > > > >> > > > > > both
> > > > > > > >> > > > > > > > > exist in
> > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > state
> > > > > > machine
> > > > > > > >> for it,
> > > > > > > >> > > > or
> > > > > > > >> > > > > > 2?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > > > think.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.4. I know the
> > "*group.coordinator.threads"
> > > > *is
> > > > > > the
> > > > > > > >> number
> > > > > > > >> > > > of
> > > > > > > >> > > > > > threads
> > > > > > > >> > > > > > > > > used
> > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > wondering
> > > > if
> > > > > > the
> > > > > > > >> purpose
> > > > > > > >> > > > of
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > threads
> > > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > > group
> > > > > > (or
> > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > why
> > > > > > should we
> > > > > > > >> need
> > > > > > > >> > > > > > > > > multi-threads
> > > > > > > >> > > > > > > > > > > here?
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> > > > shard the
> > > > > > > >> > > > processing as
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > does
> > > > the
> > > > > > > >> default
> > > > > > > >> > > > session
> > > > > > > >> > > > > > > > > timeout not
> > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > thought
> > > > > > the
> > > > > > > >> min/max
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> > it,
> > > > no?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > 30s The
> > > > > > > >> timeout to
> > > > > > > >> > > > detect
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > protocol.
> > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> > > > 45s The
> > > > > > > >> minimum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> > > > 60s The
> > > > > > > >> maximum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> > session
> > > > > > timeout
> > > > > > > >> should
> > > > > > > >> > > > be 45s
> > > > > > > >> > > > > > > > > > (the current default).
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > [range,
> > > > > > > >> uniform],
> > > > > > > >> > > > > > which means
> > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > like to
> > > > > > know
> > > > > > > >> why not
> > > > > > > >> > > > > > uniform
> > > > > > > >> > > > > > > > > one? I
> > > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > assignor
> > > > > > > >> (former
> > > > > > > >> > > > sticky
> > > > > > > >> > > > > > > > > assinor)
> > > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > > reason
> > > > > > we
> > > > > > > >> choose
> > > > > > > >> > > > range
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > as default?
> > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > uniform
> > > > The
> > > > > > > >> server side
> > > > > > > >> > > > > > assignors.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > The order on the server side has no
> > influence
> > > > > > because
> > > > > > > >> the
> > > > > > > >> > > > client
> > > > > > > >> > > > > > must
> > > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> > There
> > > > is no
> > > > > > > >> default
> > > > > > > >> > > > in the
> > > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > > specified
> > > > > > by
> > > > > > > >> the
> > > > > > > >> > > > client,
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > > request is rejected. The default client
> > value
> > > > for
> > > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I
> > hope
> > > > > > that my
> > > > > > > >> > > > answers
> > > > > > > >> > > > > > help
> > > > > > > >> > > > > > > > > > to clarify things. I will update the KIP as
> > well
> > > > > > based
> > > > > > > >> on your
> > > > > > > >> > > > > > > > > > feedback.
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > Cheers,
> > > > > > > >> > > > > > > > > > David
> > > > > > > >> > > > > > > > > >
> > > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > > > > >> showuon@gmail.com>
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Hi David,
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > > mentioned:
> > > > > > > >> > > > > > > > > > > > The new group coordinator will have a
> > state
> > > > > > machine
> > > > > > > >> per
> > > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> > each
> > > > state
> > > > > > > >> machine is
> > > > > > > >> > > > > > modelled
> > > > > > > >> > > > > > > > > as an
> > > > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > > executed
> > > > > > in
> > > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> > "Empty,
> > > > > > assigning,
> > > > > > > >> > > > > > reconciling,
> > > > > > > >> > > > > > > > > stable,
> > > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > > section,
> > > > > > > >> right?
> > > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> > is
> > > > > > modelled
> > > > > > > >> as an
> > > > > > > >> > > > event
> > > > > > > >> > > > > > > > > loop"?
> > > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > > >> *__consumer_offsets*
> > > > > > > >> > > > > > > > > partitions?
> > > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > > > owned by
> > > > > > a
> > > > > > > >> group
> > > > > > > >> > > > > > coordinator?
> > > > > > > >> > > > > > > > > For
> > > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > > consumer
> > > > > > > >> groups, and
> > > > > > > >> > > > > > both
> > > > > > > >> > > > > > > > > exist in
> > > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> > state
> > > > > > machine
> > > > > > > >> for it,
> > > > > > > >> > > > or
> > > > > > > >> > > > > > 2?
> > > > > > > >> > > > > > > > > > > 1.4. I know the
> > "*group.coordinator.threads"
> > > > *is
> > > > > > the
> > > > > > > >> number
> > > > > > > >> > > > of
> > > > > > > >> > > > > > threads
> > > > > > > >> > > > > > > > > used
> > > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> > wondering
> > > > if
> > > > > > the
> > > > > > > >> purpose
> > > > > > > >> > > > of
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > threads
> > > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > > group
> > > > > > (or
> > > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> > why
> > > > > > should we
> > > > > > > >> need
> > > > > > > >> > > > > > > > > multi-threads
> > > > > > > >> > > > > > > > > > > here?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2. For the default value in the new
> > configs:
> > > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> > does
> > > > the
> > > > > > > >> default
> > > > > > > >> > > > session
> > > > > > > >> > > > > > > > > timeout not
> > > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > > thought
> > > > > > the
> > > > > > > >> min/max
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> > it,
> > > > no?
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> > 30s The
> > > > > > > >> timeout to
> > > > > > > >> > > > detect
> > > > > > > >> > > > > > client
> > > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > > protocol.
> > > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> > > > 45s The
> > > > > > > >> minimum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> > > > 60s The
> > > > > > > >> maximum
> > > > > > > >> > > > > > session
> > > > > > > >> > > > > > > > > timeout.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > > [range,
> > > > > > > >> uniform],
> > > > > > > >> > > > > > which means
> > > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> > like to
> > > > > > know
> > > > > > > >> why not
> > > > > > > >> > > > > > uniform
> > > > > > > >> > > > > > > > > one? I
> > > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > > assignor
> > > > > > > >> (former
> > > > > > > >> > > > sticky
> > > > > > > >> > > > > > > > > assinor)
> > > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > > reason
> > > > > > we
> > > > > > > >> choose
> > > > > > > >> > > > range
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > as default?
> > > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> > uniform
> > > > The
> > > > > > > >> server side
> > > > > > > >> > > > > > assignors.
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > Thank you.
> > > > > > > >> > > > > > > > > > > Luke
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen
> > <
> > > > > > > >> showuon@gmail.com
> > > > > > > >> > > > >
> > > > > > > >> > > > > > wrote:
> > > > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > > > > >> integrating with
> > > > > > > >> > > > > > KIP-848,
> > > > > > > >> > > > > > > > > but I
> > > > > > > >> > > > > > > > > > > > think we should have a separate
> > discussion
> > > > > > thread
> > > > > > > >> for the
> > > > > > > >> > > > Kafka
> > > > > > > >> > > > > > > > > Connect
> > > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > > > Consumer
> > > > > > > >> Rebalance
> > > > > > > >> > > > > > Protocol
> > > > > > > >> > > > > > > > > [1],
> > > > > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > > > consumer
> > > > > > > >> rebalance
> > > > > > > >> > > > > > protocol,
> > > > > > > >> > > > > > > > > WDYT?
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > [1]
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > > >> > > > > > > > > > > > Luke
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> > feedback.
> > > > > > Looks
> > > > > > > >> like
> > > > > > > >> > > > there's
> > > > > > > >> > > > > > > > > agreement on
> > > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> > would
> > > > > > revisit
> > > > > > > >> the doc
> > > > > > > >> > > > and
> > > > > > > >> > > > > > see
> > > > > > > >> > > > > > > > > what
> > > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David
> > Jacot
> > > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> > document.
> > > > > > That's
> > > > > > > >> really
> > > > > > > >> > > > > > helpful. I
> > > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> > Connect
> > > > > > and the
> > > > > > > >> > > > Consumer
> > > > > > > >> > > > > > could
> > > > > > > >> > > > > > > > > share
> > > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> > main
> > > > > > > >> difference is
> > > > > > > >> > > > that
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > Consumer
> > > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > > > whereas
> > > > > > > >> Connect
> > > > > > > >> > > > assigns
> > > > > > > >> > > > > > tasks
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> > forward:
> > > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > > > support
> > > > > > > >> different
> > > > > > > >> > > > > > resource
> > > > > > > >> > > > > > > > > types
> > > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > > > Connect. The
> > > > > > > >> dedicated
> > > > > > > >> > > > APIs
> > > > > > > >> > > > > > > > > would be
> > > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> > different on
> > > > the
> > > > > > > >> > > > > > content/resources and
> > > > > > > >> > > > > > > > > > > >> > they would rely on the same engine
> > on the
> > > > > > > >> coordinator
> > > > > > > >> > > > side.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > I personally lean towards 2) because
> > I am
> > > > > > not a
> > > > > > > >> fan of
> > > > > > > >> > > > > > > > > overcharging
> > > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> > That
> > > > being
> > > > > > > >> said, I am
> > > > > > > >> > > > not
> > > > > > > >> > > > > > > > > opposed to
> > > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to
> > do
> > > > it.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> > discuss
> > > > it
> > > > > > here
> > > > > > > >> for now
> > > > > > > >> > > > in
> > > > > > > >> > > > > > order
> > > > > > > >> > > > > > > > > to
> > > > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible
> > with
> > > > what
> > > > > > we
> > > > > > > >> will do
> > > > > > > >> > > > for
> > > > > > > >> > > > > > > > > Connect in
> > > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > > >> > > > > > > > > > > >> > David
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> > > > Jacot <
> > > > > > > >> > > > > > djacot@confluent.io>
> > > > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > > > through
> > > > > > and
> > > > > > > >> address
> > > > > > > >> > > > > > your
> > > > > > > >> > > > > > > > > comments
> > > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > > > > feedback.
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > > >> > > > > > > > > > > >> > > David
> > > > > > > >> > > > > > > > > > > >> > >
> > > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> > Gregory
> > > > > > Harris
> > > > > > > >> <
> > > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> > wonderful
> > > > to see
> > > > > > > >> > > > cooperative
> > > > > > > >> > > > > > > > > rebalancing
> > > > > > > >> > > > > > > > > > > >> > making it
> > > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > > > section
> > > > > > > >> describes how
> > > > > > > >> > > > > > member
> > > > > > > >> > > > > > > > > epoch
> > > > > > > >> > > > > > > > > > > >> > should
> > > > > > > >> > > > > > > > > > > >> > > > advance in step with the group
> > epoch
> > > > and
> > > > > > > >> assignment
> > > > > > > >> > > > > > epoch
> > > > > > > >> > > > > > > > > values. I
> > > > > > > >> > > > > > > > > > > >> > think
> > > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > > > reasons
> > > > > > > >> described
> > > > > > > >> > > > in
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > KIP. When
> > > > > > > >> > > > > > > > > > > >> the
> > > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> > assigning
> > > > > > > >> partitions to a
> > > > > > > >> > > > > > worker,
> > > > > > > >> > > > > > > > > what
> > > > > > > >> > > > > > > > > > > >> member
> > > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > > assignment
> > > > > > use?
> > > > > > > >> Are
> > > > > > > >> > > > member
> > > > > > > >> > > > > > epochs
> > > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > > >> > > > > > > > > > > >> > and
> > > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> > correspond
> > > > to
> > > > > > > >> multiple
> > > > > > > >> > > > > > different
> > > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> > field
> > > > > > opaque
> > > > > > > >> to the
> > > > > > > >> > > > group
> > > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > > > assignor
> > > > > > > >> > > > implementations
> > > > > > > >> > > > > > > > > interact
> > > > > > > >> > > > > > > > > > > >> with
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> > common
> > > > > > meaning
> > > > > > > >> agreed
> > > > > > > >> > > > > > upon? If
> > > > > > > >> > > > > > > > > so, what
> > > > > > > >> > > > > > > > > > > >> > is the
> > > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> > field
> > > > over
> > > > > > > >> including
> > > > > > > >> > > > such
> > > > > > > >> > > > > > > > > functionality
> > > > > > > >> > > > > > > > > > > >> > in the
> > > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 3. The following is included in
> > the
> > > > KIP:
> > > > > > > >> "Thanks to
> > > > > > > >> > > > > > this, the
> > > > > > > >> > > > > > > > > input
> > > > > > > >> > > > > > > > > > > >> of
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> > > > driven
> > > > > > by
> > > > > > > >> the group
> > > > > > > >> > > > > > > > > coordinator.
> > > > > > > >> > > > > > > > > > > >> The
> > > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> > responsible for
> > > > > > > >> maintaining
> > > > > > > >> > > > any
> > > > > > > >> > > > > > state
> > > > > > > >> > > > > > > > > besides
> > > > > > > >> > > > > > > > > > > >> its
> > > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this
> > mean
> > > > > > that the
> > > > > > > >> > > > > > client-side
> > > > > > > >> > > > > > > > > assignor
> > > > > > > >> > > > > > > > > > > >> MAY
> > > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> > non-Metadata
> > > > state
> > > > > > > >> (such as
> > > > > > > >> > > > > > partition
> > > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> > topics,
> > > > etc),
> > > > > > or
> > > > > > > >> that
> > > > > > > >> > > > > > additional
> > > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > > > classes
> > > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > > >> > > > > >
> > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > > >> > > > > > that
> > > > > > > >> > > > > > > > > seem to
> > > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > > > possible for
> > > > > > > >> these two
> > > > > > > >> > > > > > > > > implementations
> > > > > > > >> > > > > > > > > > > >> to
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> > promote
> > > > > > feature
> > > > > > > >> parity
> > > > > > > >> > > > of
> > > > > > > >> > > > > > > > > server-side
> > > > > > > >> > > > > > > > > > > >> and
> > > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and would
> > also
> > > > > > > >> facilitate
> > > > > > > >> > > > > > operational
> > > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > > >> > > > > > > > > > > >> > > > certain situations. For example,
> > if a
> > > > > > > >> server-side
> > > > > > > >> > > > > > assignor
> > > > > > > >> > > > > > > > > has some
> > > > > > > >> > > > > > > > > > > >> > poor
> > > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > > deploying the
> > > > > > > >> patched
> > > > > > > >> > > > > > assignor to
> > > > > > > >> > > > > > > > > the
> > > > > > > >> > > > > > > > > > > >> > client
> > > > > > > >> > > > > > > > > > > >> > > > and switching one consumer group
> > to a
> > > > > > > >> client-side
> > > > > > > >> > > > > > assignor
> > > > > > > >> > > > > > > > > may be
> > > > > > > >> > > > > > > > > > > >> > faster
> > > > > > > >> > > > > > > > > > > >> > > > and less risky than patching all
> > of
> > > > the
> > > > > > > >> brokers.
> > > > > > > >> > > > With
> > > > > > > >> > > > > > the
> > > > > > > >> > > > > > > > > currently
> > > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> > non-trivial
> > > > > > > >> > > > reimplementation
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > have
> > > > > > > >> > > > > > > > > > > >> to be
> > > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs
> > have
> > > > > > diverged
> > > > > > > >> > > > > > significantly,
> > > > > > > >> > > > > > > > > then it
> > > > > > > >> > > > > > > > > > > >> is
> > > > > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> > > > would
> > > > > > not be
> > > > > > > >> > > > possible.
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM
> > Sagar
> > > > <
> > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > >
> > > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > > > discuss
> > > > > > how
> > > > > > > >> Connect
> > > > > > > >> > > > > > would
> > > > > > > >> > > > > > > > > need to
> > > > > > > >> > > > > > > > > > > >> > change
> > > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > > protocol.
> > > > > > > >> Here's the
> > > > > > > >> > > > page:
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> >
> > > > > > > >> > > > > > > > > > > >>
> > > > > > > >> > > > > > > > >
> > > > > > > >> > > > > >
> > > > > > > >> > > >
> > > > > > > >>
> > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I
> > have
> > > > > > tried
> > > > > > > >> to keep
> > > > > > > >> > > > a
> > > > > > > >> > > > > > format
> > > > > > > >> > > > > > > > > > > >> similar to
> > > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > > > think.
> > > > > > Also,
> > > > > > > >> do you
> > > > > > > >> > > > > > think this
> > > > > > > >> > > > > > > > > > > >> should
> > > > > > > >> > > > > > > > > > > >> > be
> > > > > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> > > > thread
> > > > > > or
> > > > > > > >> is this
> > > > > > > >> > > > one
> > > > > > > >> > > > > > fine?
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > > > Sagar <
> > > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> > doc on
> > > > > > Kafka
> > > > > > > >> Streams.
> > > > > > > >> > > > > > Sure, I
> > > > > > > >> > > > > > > > > would do
> > > > > > > >> > > > > > > > > > > >> > the
> > > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> > such a
> > > > > > > >> document.
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47
> > AM
> > > > > > Guozhang
> > > > > > > >> Wang <
> > > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> > could
> > > > come
> > > > > > > >> back with
> > > > > > > >> > > > some
> > > > > > > >> > > > > > > > > analysis on
> > > > > > > >> > > > > > > > > > > >> > how to
> > > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > > > integration
> > > > > > > >> with
> > > > > > > >> > > > the new
> > > > > > > >> > > > > > > > > protocol;
> > > > > > > >> > > > > > > > > > > >> so
> > > > > > > >> > > > > > > > > > > >> > far
> > > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the
> > new
> > > > > > "protocol
> > > > > > > >> type"
> > > > > > > >> > > > we
> > > > > > > >> > > > > > did not
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Regarding the version, what I was thinking is that in the HB request, for
"serverAssignor" field, instead of just having it as a single string field,
maybe we could consider also making it a structure that includes: name,
minimumVersion, maximumVersion. Where the minimumVersion/maximumVersion
means the versions of the server assignor that the client work best with.
That being said, I agree with you that such information may also be
inferred elsewhere e.g. by looking into the "rackId" field, and see if it
contains a hyphen or not etc. All I was wondering is that, if such version
information would be useful for the server assignors to determine its
actual assignment logic. I do not feel very strong about this one though
--- even if we do not add it now, we can potentially add later, it's just
that changing a single string field to a structure would be hard for
compatibility and we'd then probably have to add top-level fields.

Regarding the `onAssignment` logic, again my train of thoughts is that, if
users want to know exactly when a partition is assigned / revoked, they
would be leveraging on the rebalance callbacks, as that's what people
should rely on to determine "when" partitions are assigned. The
`onAssignment` should be used for getting "why" such partition assignment
decision is made, and hence returning `combined partitions` would be okay.
Streams e.g. implement both rebalance callbacks and the assignors, and it
gets the "when" from the former (and create/close active tasks accordingly)
and the "why" from the latter (and update its global info bookkeeping as
well as standby maintenance accordingly). Most users would be just
interested in the rebalance callback, and not implement their own assignor
at all if they do not care about "why" as they trust the server assignors
would take good care of those, and only about "when". So if we did build
such two types of APIs from scratch, I'd indeed feel that not providing the
partitions but only the metadata for `onAssignment` may be less confusing
and push users to separate the usage of these two more clearly, but since
we already introduced partitions in `onAssignment` for compatibility I'm
less keen on removing them.


Guozhang

On Mon, Sep 26, 2022 at 6:55 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Guozhang,
>
> Regarding the version, my understanding is that the version would be
> either the client software version or the request version, is this
> correct? If so, we could indeed pass this information down to the
> assignor via the interface. One way would be to pass a "server
> context" to the assignor and that context would include that
> information (and perhaps more). Is this what you are looking for?
>
> Regarding the onAssignment, I think that I understand your point. I
> suppose that the assignor could also be clever and keep track of the
> last metadata to decide whether it has to do something or not. One
> question that is still not clear to me is whether the assignor needs
> to know all the assigned partitions upfront regardless of whether they
> are already revoked or not. Do you think that we need this as well?
>
> From an API perspective, we could have something like
> onAssignment(Metadata(version, reason, metadata, assigned partitions,
> pending partitions)). Where the assigned partitions are the partitions
> ready to be used and the pending partitions are the one assigned to
> the member but not revoked yet. I find it a bit weird that this method
> would be called only once because the assignor would not know when the
> pending partitions changes. That does not look like a clean API. An
> alternative would be to use onAssignment(Metadata(version, reason,
> metadata, combined partitions)) but this seems error prone because it
> is not clear whether a partition is usable or not. Or do you think
> that we should not provide the partitions but only the metadata?
>
> Best,
> David
>
> On Fri, Sep 23, 2022 at 9:40 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello David,
> >
> > On Fri, Sep 23, 2022 at 2:00 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hey,
> > >
> > > > Just to clarify I was asking about the `version` of the assignor
> (i.e. up
> > > to what version that the client would support), and I do agree we
> would not
> > > need metadata. What I have in mind is that, for some specific built-in
> > > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > > newer version we would have a hierarchical rack ID string format, like
> > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > would still be in old format. In this case, the broker then needs to
> choose
> > > the old versioned assignor. I'm probably making something up here for
> rack
> > > aware assignors, but I'm wondering if in general such an
> "auto-downgrade"
> > > behavior would be needed still for broker-side assignor, and if yes
> would
> > > "version" still be useful.
> > >
> > > Got it. That's an interesting thought. I think that the issue is that
> > > the client will never tell you which version of the server-side
> > > assignor should be used. Do you think that the coordinator would
> > > downgrade the version if the assignment fails with a higher version? I
> > > tend to believe that this should be handled within the assignor
> > > itself. In the example that you mentioned, the assignor would have to
> > > handle all the cases. I am not really convinced that we need this at
> > > the moment.
> > >
> > > The version from the client side would not be indicating the broker
> which
> > version to use, but rather which version the client would "work best
> with".
> > Such a "version" field would not be settible by the users, since they
> will
> > be hard-codedly bumped when the Kafka byte code version bumped.
> > Back to the rack aware assignor example, if the older versioned client
> does
> > not have a hierarchical rack ID, however if the assignment returned to
> them
> > is assuming a hierarchical rack structure, it may not reflect the best
> > workload balance among those new and old versioned clients. That means,
> > when receiving the members subscriptions at the server side, if the
> > versions from all these members are different, the broker's assignor may
> > need to consider using the lower version logic to do the assignment. So
> yes
> > the assignor would indeed have to handle all such cases, but it needs to
> do
> > so such that if there are clients who would not work with certain new
> > logic, it would then handle such cases automatically by e.g. still using
> an
> > older versioned logic.
> >
> >
> >
> > > > Okay, my understanding is that the calling ordering of these
> callbacks
> > > would be like the following:
> > >
> > > Yes, your examples look right.
> > >
> > > > I'm wondering if we would still call onAssignment just once, that
> encodes
> > > all the assignment for this rebalance, including all the partitions
> that
> > > should be assigned to the member but not yet assigned since they have
> not
> > > been revoked by others. In that case the call ordering would be:
> > >
> > > Interesting. Is there a case for Streams where having the full
> > > assignment is beneficial? For instance, I can think of the following
> > > case. When a standby task is promoted to an active task, the metadata
> > > would not contain the standby task anymore and the assignment may not
> > > have the partition yet. In this case, Streams would stop the standby
> > > tasks but not have the active task yet if my understanding of Streams
> > > is correct. So knowing the full assignment could be helpful here.
> > >
> > > If we want to do this, we could structure the assignment given to the
> > > member as follow: version, error, metadata, assigned partitions,
> > > pending partitions, where the pending partitions would be the one
> > > assigned to this member but not yet available. What do you think?
> > >
> > > Regarding onAssignment being called only once, I am not sure to fully
> > > grasp the benefit yet. Does the assignor really care about this? In
> > > the end, the epoch does not really matter for the assignor because it
> > > has to converge its state to the desired state anyway.
> > >
> > > Here's my rationale (maybe rephased a bit :P ): the implementers of
> > rebalance listener and assignor are two groups of people, and most users
> > fall into the former group, while only very few people fall into the
> later
> > group. For rebalance listener implementers, they just want to know when a
> > partition is actually revoked or assigned to the consumer and reacts to
> it,
> > for this purpose, `onPartitionsRevoked` and `onPartitionsAssigned` would
> be
> > triggered interleavingly upon `poll` calls across rebalances. The usual
> > logic for such rebalance listeners are metrics reporting, committing
> > offsets (if they do not use Kafka for that), etc. They would not care
> which
> > calls are from which rebalances --- in the past with eager rebalance, it
> > maybe that each rebalance is associated with exactly a
> > `onPartitionsRevoked` first and then a `onPartitionsAssigned`, but it
> would
> > no longer the cases now.
> >
> > The implementers of the assignor though, would care about "how the
> > assignment was made", that includes from which rebalance a certain
> > revoke/assign decision was made, based on what metadata such assignment
> is
> > made, etc. And that's the whole point of the `onAssignment` function
> since
> > otherwise they can just rely on the listeners. They usually
> implementation
> > logic of this callback is to e.g. bookkeep the assignment decision
> driving
> > factors a.k.a. the metadata, global information that needs to be
> propagated
> > to all members, etc. Take Streams as an example, the active processing
> > tasks go along with the assigned partitions, and we can always just
> > incrementally create / close them upon each rebalance listener triggers,
> > when certain partitions are revoked or assigned together; standby tasks
> > however are encoded with the metadata, and we can only know which standby
> > tasks should we get / drop based on the `onAssignment` function, and in
> > fact the creation of such tasks as a result of the metadata bookkeeping
> > does not need to wait until all the partitions that are yet-assigned have
> > been completely assigned to the member. Such information may not always
> be
> > updatable in an incremental manner as the partitions-revoked /
> > partitions-assigned. In such a case, it's better to just trigger this
> > function "once per decision made" i.e. once per rebalance generation.
> >
> >
> > Guozhang
> >
> >
> >
> > > Best,
> > > David
> > >
> > > On Thu, Sep 22, 2022 at 6:01 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > >
> > > > Hi David, thanks for all the detailed explanations. I think they all
> make
> > > > sense. Just want to have a couple follow-ups here:
> > > >
> > > > > I don't really see the benefits here because server side assignors
> > > > don't have metadata at all. They only assign topic-partitions. They
> > > > are not supposed to generate metadata nor to receive metadata from
> the
> > > > members.
> > > >
> > > > Just to clarify I was asking about the `version` of the assignor
> (i.e. up
> > > > to what version that the client would support), and I do agree we
> would
> > > not
> > > > need metadata. What I have in mind is that, for some specific
> built-in
> > > > broker-assignors, e.g. rack-aware assignors, if it's possible that
> in a
> > > > newer version we would have a hierarchical rack ID string format,
> like
> > > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > > would still be in old format. In this case, the broker then needs to
> > > choose
> > > > the old versioned assignor. I'm probably making something up here for
> > > rack
> > > > aware assignors, but I'm wondering if in general such an
> "auto-downgrade"
> > > > behavior would be needed still for broker-side assignor, and if yes
> would
> > > > "version" still be useful.
> > > >
> > > > > Yeah, that's right. Within a rebalance, `onAssignment` is called
> once
> > > > when the member transitions to a new epoch. This one contains the
> full
> > > > metadata provided by the client side assignor. Then, `onAssignment`
> > > > can be called max N times where N is the number of partitions pending
> > > > revocation by other members. Let me try to clarify this in the KIP.
> > > >
> > > > Okay, my understanding is that the calling ordering of these
> callbacks
> > > > would be like the following:
> > > >
> > > > ----------------------------------------
> > > > onPartitionsRevoked();   // just once, since we do not really need
> > > > to revoke incrementally.
> > > >
> > > > onAssignment();    // the first call, with epoch incremented
> > > > onPartitionsAssigned();   // paired with the onAssignment
> > > >
> > > > onAssignment();              // the first onAssignment would bump up
> the
> > > > epoch, and the metadata reflected.
> > > > onPartitionsAssigned();   // each time we get an additional
> assignment,
> > > we
> > > > call onAssignment and then paired with an onPartitionsAssigned
> > > > ...
> > > > onAssignment();
> > > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > > encoded
> > > > metadata would not change, only the incrementally added partitions be
> > > > reflected
> > > >
> > > > Is that the case?
> > > >
> > > > I'm wondering if we would still call onAssignment just once, that
> encodes
> > > > all the assignment for this rebalance, including all the partitions
> that
> > > > should be assigned to the member but not yet assigned since they
> have not
> > > > been revoked by others. In that case the call ordering would be:
> > > >
> > > > ----------------------------------------
> > > > onPartitionsRevoked();   // just once
> > > > onAssignment();    // just once, with epoch incremented, and metadata
> > > > encoded changed, the "assignment" field also reflect the final target
> > > > assignment
> > > > onPartitionsAssigned();   // multiple times, which represent
> > > incrementally
> > > > added partitions
> > > > ...
> > > > onPartitionsAssigned();
> > > >
> > > > The motivation from this is that, most users would only implement the
> > > > rebalance callback listeners and hence we'd definitely need to make
> sure
> > > > the semantics of that does not change much, and the time
> > > > `onPartitionsAssigned` indicate the time when the partitions are
> actually
> > > > assigned to it; while for assignors, the `onAssignment` is used to
> > > indicate
> > > > what decision is made regarding for this member, i.e. when the
> partitions
> > > > are decided to be given to it, but not necessarily meant that it has
> been
> > > > given, since that time should be determined by the time of
> > > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > > implementers would not need to reason which `onAssignment` would be
> the
> > > > last one for this epoch.
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > > Guozhang
> > > >
> > > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot
> <djacot@confluent.io.invalid
> > > >
> > > > wrote:
> > > >
> > > > > Hi Guozhang,
> > > > >
> > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> byte
> > > > > field,
> > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> And
> > > the
> > > > > > rationale seems to be assuming that we should always be able to
> do
> > > the
> > > > > > assignment at the broker-side assignor without errors.
> Personally I
> > > think
> > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > broker-side assignors, e.g. for some edge cases where some
> subscribed
> > > > > > topics are not recognized with the current broker's metadata.
> What
> > > do you
> > > > > > think?
> > > > >
> > > > > Yeah, that seems reasonable. However, I wonder if it would be
> better
> > > > > to use an exception on the server side. This is what we usually do
> for
> > > > > server side plugins. On the client side, we use a field because the
> > > > > errors are not defined in advance.
> > > > >
> > > > > Your comment also makes me think about what we should do when the
> > > > > server side assignor fails. I suppose that we have to keep the
> current
> > > > > assignment until a new event occurs. For instance, in your example,
> > > > > the coordinator would have to trigger a rebalance when unrecognized
> > > > > topics are available. This would be part of the metadata
> monitoring.
> > > > >
> > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > reason/version/metadata compared with the broker-side
> GroupMember. I
> > > > > agree
> > > > > > that broker-side assignor would not need reason/metadata since
> they
> > > are
> > > > > > blackbox strings/bytes to the assignor, but what about version?
> E.g.
> > > is
> > > > > it
> > > > > > possible that we evolve our broker-side built-in assignor but
> the old
> > > > > > versioned clients would not be able to work with the new
> version, in
> > > > > which
> > > > > > case we need to let the broker being aware of this and upgrade
> its
> > > > > behavior
> > > > > > to cooperate with the clients?
> > > > >
> > > > > I don't really see the benefits here because server side assignors
> > > > > don't have metadata at all. They only assign topic-partitions. They
> > > > > are not supposed to generate metadata nor to receive metadata from
> the
> > > > > members.
> > > > >
> > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > instead
> > > > > of
> > > > > > including these three fields, what about just adding the
> "Metadata"
> > > field
> > > > > > class which has these three fields? Also, there are two
> "Metadata"
> > > > > > currently in the APIs, the first is a class that encodes
> > > > > > reason/version/metadata, and the second is just the encoded
> metadata
> > > > > bytes.
> > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > which
> > > > > > then has a bytebuffer field Metadata, or instead naming the
> second
> > > > > > bytebuffer field as metadataBytes?
> > > > >
> > > > > That's a good point. Let me try to rationalize this interface
> based on
> > > > > your suggestions.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > > > >
> > > > > > Hello David,
> > > > > >
> > > > > > Just had a few more nit questions about the public APIs:
> > > > > >
> > > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error
> byte
> > > > > field,
> > > > > > while the broker-side "PartitionAssignor#Assignment" does not.
> And
> > > the
> > > > > > rationale seems to be assuming that we should always be able to
> do
> > > the
> > > > > > assignment at the broker-side assignor without errors.
> Personally I
> > > think
> > > > > > it's still potentially beneficial to add the Error field even for
> > > > > > broker-side assignors, e.g. for some edge cases where some
> subscribed
> > > > > > topics are not recognized with the current broker's metadata.
> What
> > > do you
> > > > > > think?
> > > > > >
> > > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > > reason/version/metadata compared with the broker-side
> GroupMember. I
> > > > > agree
> > > > > > that broker-side assignor would not need reason/metadata since
> they
> > > are
> > > > > > blackbox strings/bytes to the assignor, but what about version?
> E.g.
> > > is
> > > > > it
> > > > > > possible that we evolve our broker-side built-in assignor but
> the old
> > > > > > versioned clients would not be able to work with the new
> version, in
> > > > > which
> > > > > > case we need to let the broker being aware of this and upgrade
> its
> > > > > behavior
> > > > > > to cooperate with the clients?
> > > > > >
> > > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > > instead
> > > > > of
> > > > > > including these three fields, what about just adding the
> "Metadata"
> > > field
> > > > > > class which has these three fields? Also, there are two
> "Metadata"
> > > > > > currently in the APIs, the first is a class that encodes
> > > > > > reason/version/metadata, and the second is just the encoded
> metadata
> > > > > bytes.
> > > > > > I'm wondering what about just naming the first as memberMetadata,
> > > which
> > > > > > then has a bytebuffer field Metadata, or instead naming the
> second
> > > > > > bytebuffer field as metadataBytes?
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <
> wangguoz@gmail.com>
> > > > > wrote:
> > > > > >
> > > > > > > Hello David,
> > > > > > >
> > > > > > > Thanks for bringing this question up. I think the main
> benefits as
> > > you
> > > > > > > listed is 2) above if it stays; just to clarify, we would only
> be
> > > able
> > > > > to
> > > > > > > save one round trip if the rebalance is still triggered by the
> > > broker;
> > > > > if
> > > > > > > the rebalance is triggered by the client then the
> num.round.trips
> > > are
> > > > > the
> > > > > > > same:
> > > > > > >
> > > > > > > 1) With GroupPrepareAssignment:
> > > > > > >
> > > > > > > T0: client decides to do a new assignment, suppose it has
> already
> > > sent
> > > > > a
> > > > > > > HB and hence has to wait for it to return first since only one
> > > request
> > > > > /
> > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > T1: client receives the HB response, and then sends the
> > > > > > > GroupPrepareAssignment request.
> > > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > > T3: it calculates the new assignment, and sends a
> > > > > GroupInstallAssignment
> > > > > > > request.
> > > > > > >
> > > > > > > In total, two round trips.
> > > > > > >
> > > > > > > 2) Without GroupPrepareAssignment:
> > > > > > >
> > > > > > > T0: client decides to do a new assignment, suppose it has
> already
> > > sent
> > > > > a
> > > > > > > HB and hence has to wait for it to return first since only one
> > > request
> > > > > /
> > > > > > > response can be inflight with the coordinator's socket.
> > > > > > > T1: client receives the HB response, and then sends the new HB
> > > request
> > > > > > > with the flag indicating a new rebalance needed..
> > > > > > > T2: the HB response with the optional member metadata map is
> > > returned.
> > > > > > > T3: it calculates the new assignment, and sends a
> > > > > GroupInstallAssignment
> > > > > > > request.
> > > > > > >
> > > > > > > In total, two round trips as well.
> > > > > > >
> > > > > > > -----------------------------
> > > > > > >
> > > > > > > So to complete the full picture here, we'd need to modify both
> HB
> > > > > request
> > > > > > > and response so that the client can also indicate a new
> rebalance
> > > via
> > > > > the
> > > > > > > HB request as well, right?
> > > > > > >
> > > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > > GroupPrepareAssignment into HB given that we can make the
> > > additional
> > > > > fields
> > > > > > > encoding the full member (subscription) metadata and topic
> > > metadata as
> > > > > > > optional fields.
> > > > > > >
> > > > > > >
> > > > > > > Guozhang
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > >> Hi all,
> > > > > > >>
> > > > > > >> During an offline conversation, someone asked why we need the
> > > > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > > instead
> > > > > > >> provide the group state in the heartbeat response. This has a
> few
> > > > > > >> advantages: 1) it does not require using a special error code
> to
> > > > > > >> signal that a new assignment is required as the signal would
> be
> > > the
> > > > > > >> provided group state; 2) it removes one round trip when a
> client
> > > side
> > > > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > > > >> response's definition quite large. I recall that I went with
> the
> > > > > > >> current approach due to this.
> > > > > > >>
> > > > > > >> Providing the group state in the heartbeat response is
> appealing.
> > > What
> > > > > > >> do you guys think?
> > > > > > >>
> > > > > > >> Best,
> > > > > > >> David
> > > > > > >>
> > > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <
> djacot@confluent.io>
> > > > > wrote:
> > > > > > >> >
> > > > > > >> > Hi Guozhang,
> > > > > > >> >
> > > > > > >> > 1. I have added a reference to the relevant chapter instead
> of
> > > > > > >> > repeating the whole thing. Does that work for you?
> > > > > > >> >
> > > > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> > > about
> > > > > when
> > > > > > >> > a rebalance should be triggered for the non-upgraded members
> > > using
> > > > > the
> > > > > > >> > old protocol. The section mentions that a rebalance must be
> > > > > triggered
> > > > > > >> > when a new assignment is installed. This implies that the
> group
> > > > > epoch
> > > > > > >> > was updated either by a native member or a non-upgraded
> member.
> > > For
> > > > > > >> > the latter, the JoinGroup request would be the trigger. I
> have
> > > > > added a
> > > > > > >> > reference to the relevant chapter in the "JoinGroup
> Handling"
> > > > > section
> > > > > > >> > as well. Does that make sense?
> > > > > > >> >
> > > > > > >> > Thanks,
> > > > > > >> > David
> > > > > > >> >
> > > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > > > >> wrote:
> > > > > > >> > >
> > > > > > >> > > Hello David,
> > > > > > >> > >
> > > > > > >> > > Alright I think that's sufficient. Just to make that
> clear in
> > > the
> > > > > doc,
> > > > > > >> > > could we update:
> > > > > > >> > >
> > > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > > coordinator
> > > > > > >> will
> > > > > > >> > > trigger rebalance based on the HB's member metadata /
> reason?
> > > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > > described
> > > > > in
> > > > > > >> "Group
> > > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > > >> > >
> > > > > > >> > >
> > > > > > >> > > Guozhang
> > > > > > >> > >
> > > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > wrote:
> > > > > > >> > >
> > > > > > >> > > > Hi Guozhang,
> > > > > > >> > > >
> > > > > > >> > > > I thought that the assignor will always be consulted
> when
> > > the
> > > > > next
> > > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > > heartbeat.
> > > > > > >> This
> > > > > > >> > > > gives the opportunity for the assignor to enforce a
> > > rebalance by
> > > > > > >> > > > setting the reason to a non-zero value or by changing
> the
> > > > > bytes. Do
> > > > > > >> > > > you think that this is not sufficient? Are you
> concerned by
> > > the
> > > > > > >> delay?
> > > > > > >> > > >
> > > > > > >> > > > Best,
> > > > > > >> > > > David
> > > > > > >> > > >
> > > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > > wangguoz@gmail.com>
> > > > > > >> wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > Hello David,
> > > > > > >> > > > >
> > > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > > >> > > > >
> > > > > > >> > > > > ```
> > > > > > >> > > > > In this case, a new assignment is triggered by the
> client
> > > side
> > > > > > >> > > > > assignor. When constructing the HB, the consumer will
> > > always
> > > > > > >> consult
> > > > > > >> > > > > the client side assignor and propagate the
> information to
> > > the
> > > > > > >> group
> > > > > > >> > > > > coordinator. In other words, we don't expect users to
> call
> > > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > > >> > > > > ```
> > > > > > >> > > > >
> > > > > > >> > > > > As I looked at the current PartitionAssignor's
> interface,
> > > we
> > > > > > >> actually do
> > > > > > >> > > > > not have a way yet to instruct how to construct the
> next
> > > HB
> > > > > > >> request, e.g.
> > > > > > >> > > > > when the assignor wants to enforce a new rebalance
> with a
> > > new
> > > > > > >> assignment,
> > > > > > >> > > > > we'd need some customizable APIs inside the
> > > PartitionAssignor
> > > > > to
> > > > > > >> indicate
> > > > > > >> > > > > the next HB telling broker about so. WDYT about adding
> > > such an
> > > > > > >> API on the
> > > > > > >> > > > > PartitionAssignor?
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > Guozhang
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > > >> <dj...@confluent.io.invalid>
> > > > > > >> > > > > wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > > Hi Jun,
> > > > > > >> > > > > >
> > > > > > >> > > > > > I have updated the KIP to include your feedback. I
> have
> > > also
> > > > > > >> tried to
> > > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > > >> > > > > >
> > > > > > >> > > > > > Best,
> > > > > > >> > > > > > David
> > > > > > >> > > > > >
> > > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > > djacot@confluent.io
> > > > > > >> >
> > > > > > >> > > > wrote:
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Hi Jun,
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Thanks for your feedback. Let me start by
> answering
> > > your
> > > > > > >> questions
> > > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> of
> > > the
> > > > > KIP
> > > > > > >> seem to
> > > > > > >> > > > be
> > > > > > >> > > > > > fewer
> > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> support of
> > > > > > >> wildcard. A few
> > > > > > >> > > > > > > > comments below.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > I would also add that the KIP removes the global
> sync
> > > > > barrier
> > > > > > >> in the
> > > > > > >> > > > > > > protocol which is essential to improve group
> > > stability and
> > > > > > >> > > > > > > scalability, and the KIP also simplifies the
> client by
> > > > > moving
> > > > > > >> most of
> > > > > > >> > > > > > > the logic to the server side.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> to
> > > > > support
> > > > > > >> rolling
> > > > > > >> > > > > > changing
> > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Definitely. The group coordinator will use the
> > > assignor
> > > > > used
> > > > > > >> by a
> > > > > > >> > > > > > > majority of the members. This allows the group to
> move
> > > > > from
> > > > > > >> one
> > > > > > >> > > > > > > assignor to another by a roll. This is explained
> in
> > > the
> > > > > > >> Assignor
> > > > > > >> > > > > > > Selection chapter.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> it's
> > > > > > >> required in
> > > > > > >> > > > every
> > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > filled? For
> > > > > > >> example,
> > > > > > >> > > > it's
> > > > > > >> > > > > > not
> > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> filled.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > The client is expected to set those fields in
> case of
> > > a
> > > > > > >> connection
> > > > > > >> > > > > > > issue (e.g. timeout) or when the fields have
> changed
> > > since
> > > > > > >> the last
> > > > > > >> > > > > > > HB. The server populates those fields as long as
> the
> > > > > member
> > > > > > >> is not
> > > > > > >> > > > > > > fully reconciled - the member should acknowledge
> that
> > > it
> > > > > has
> > > > > > >> the
> > > > > > >> > > > > > > expected epoch and assignment. I will clarify
> this in
> > > the
> > > > > KIP.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > affinity
> > > > > > >> between the
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > and the broker is only considered during
> fetching,
> > > but
> > > > > not
> > > > > > >> during
> > > > > > >> > > > > > assigning
> > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > assignment
> > > > > is
> > > > > > >> made,
> > > > > > >> > > > there
> > > > > > >> > > > > > is
> > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > replicas of
> > > > > > >> assigned
> > > > > > >> > > > > > partitions
> > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > > should use
> > > > > > >> this
> > > > > > >> > > > > > opportunity
> > > > > > >> > > > > > > > to address this by including rack in
> GroupMember.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's an interesting idea. I don't see any issue
> with
> > > > > adding
> > > > > > >> the
> > > > > > >> > > > rack
> > > > > > >> > > > > > > to the members. I will do so.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> know
> > > how
> > > > > busy
> > > > > > >> a group
> > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> it
> > > seems
> > > > > > >> that we
> > > > > > >> > > > could
> > > > > > >> > > > > > add
> > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> the
> > > event
> > > > > > >> loop is
> > > > > > >> > > > doing
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > actual work.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> failover
> > > > > > >> handling? For
> > > > > > >> > > > > > example,
> > > > > > >> > > > > > > > does it need to trigger the check if any group
> with
> > > the
> > > > > > >> wildcard
> > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sure. When the new group coordinator takes over,
> it
> > > has
> > > > > to:
> > > > > > >> > > > > > > * Setup the session timeouts.
> > > > > > >> > > > > > > * Trigger a new assignment if a client side
> assignor
> > > is
> > > > > used.
> > > > > > >> We
> > > > > > >> > > > don't
> > > > > > >> > > > > > > store the information about the member selected to
> > > run the
> > > > > > >> assignment
> > > > > > >> > > > > > > so we have to start a new one.
> > > > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > > > >> subscriptions, and
> > > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> document
> > > what
> > > > > > >> the epoch
> > > > > > >> > > > > > field
> > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > >> ConsumerGroupMetadataValue
> > > > > > >> > > > > > reflect
> > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sure. I will clarify that but it is always the
> latest
> > > > > group
> > > > > > >> epoch.
> > > > > > >> > > > > > > When the group state is updated, the group epoch
> is
> > > > > bumped so
> > > > > > >> we use
> > > > > > >> > > > > > > that one for all the change records related to the
> > > update.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > following
> > > > > > >> > > > invariants
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> a
> > > member
> > > > > > >> not to
> > > > > > >> > > > get any
> > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > That's right. Here I meant that the members
> provided
> > > by
> > > > > the
> > > > > > >> assignor
> > > > > > >> > > > > > > in the assignment must exist in the group. The
> > > assignor
> > > > > can
> > > > > > >> not make
> > > > > > >> > > > > > > up new member ids.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> epoch
> > > > > equals to
> > > > > > >> 0":
> > > > > > >> > > > When
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> used?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > A member is expected to abandon all its
> partitions and
> > > > > > >> rejoins when
> > > > > > >> > > > it
> > > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this
> case,
> > > the
> > > > > > >> group
> > > > > > >> > > > > > > coordinator will have removed the member from the
> > > group.
> > > > > The
> > > > > > >> member
> > > > > > >> > > > > > > can rejoin the group with the same member id but
> with
> > > 0 as
> > > > > > >> epoch. Let
> > > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> to
> > > > > trigger a
> > > > > > >> > > > > > reassignment
> > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > updating the
> > > > > > >> assignor
> > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> with
> > > the
> > > > > > >> deprecation
> > > > > > >> > > > of
> > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > In this case, a new assignment is triggered by the
> > > client
> > > > > side
> > > > > > >> > > > > > > assignor. When constructing the HB, the consumer
> will
> > > > > always
> > > > > > >> consult
> > > > > > >> > > > > > > the client side assignor and propagate the
> > > information to
> > > > > the
> > > > > > >> group
> > > > > > >> > > > > > > coordinator. In other words, we don't expect
> users to
> > > call
> > > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > section
> > > > > > >> seems to
> > > > > > >> > > > have
> > > > > > >> > > > > > > > multiple typos.
> > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > immediately
> > > > > > >> gets into
> > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > incorrect.
> > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> seems
> > > to
> > > > > get
> > > > > > >> into
> > > > > > >> > > > > > epoch=3,
> > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> has A -
> > > > > > >> epoch=2,
> > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> the
> > > > > upgrade
> > > > > > >> from
> > > > > > >> > > > any
> > > > > > >> > > > > > old
> > > > > > >> > > > > > > > version to new one?
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > We will support upgrading from the consumer
> protocol
> > > > > version
> > > > > > >> 3,
> > > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented
> yet
> > > so
> > > > > the
> > > > > > >> earliest
> > > > > > >> > > > > > > version is unknown at the moment. This is
> explained
> > > in the
> > > > > > >> migration
> > > > > > >> > > > > > > plan chapter.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Thanks again for your feedback, Jun. I will update
> > > the KIP
> > > > > > >> based on
> > > > > > >> > > > it
> > > > > > >> > > > > > > next week.
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > Best,
> > > > > > >> > > > > > > David
> > > > > > >> > > > > > >
> > > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > > >> <ju...@confluent.io.invalid>
> > > > > > >> > > > wrote:
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Hi, David,
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits
> of
> > > the
> > > > > KIP
> > > > > > >> seem to
> > > > > > >> > > > be
> > > > > > >> > > > > > fewer
> > > > > > >> > > > > > > > RPCs during rebalance and more efficient
> support of
> > > > > > >> wildcard. A few
> > > > > > >> > > > > > > > comments below.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan
> to
> > > > > support
> > > > > > >> rolling
> > > > > > >> > > > > > changing
> > > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > > >> > > > > > > > 30.2 For each field, could you explain whether
> it's
> > > > > > >> required in
> > > > > > >> > > > every
> > > > > > >> > > > > > > > request or the scenarios when it needs to be
> > > filled? For
> > > > > > >> example,
> > > > > > >> > > > it's
> > > > > > >> > > > > > not
> > > > > > >> > > > > > > > clear to me when TopicPartitions needs to be
> filled.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > > affinity
> > > > > > >> between the
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > and the broker is only considered during
> fetching,
> > > but
> > > > > not
> > > > > > >> during
> > > > > > >> > > > > > assigning
> > > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > > assignment
> > > > > is
> > > > > > >> made,
> > > > > > >> > > > there
> > > > > > >> > > > > > is
> > > > > > >> > > > > > > > no opportunity for read affinity because no
> > > replicas of
> > > > > > >> assigned
> > > > > > >> > > > > > partitions
> > > > > > >> > > > > > > > are close to the member. I am wondering if we
> > > should use
> > > > > > >> this
> > > > > > >> > > > > > opportunity
> > > > > > >> > > > > > > > to address this by including rack in
> GroupMember.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 32. On the metric side, often, it's useful to
> know
> > > how
> > > > > busy
> > > > > > >> a group
> > > > > > >> > > > > > > > coordinator is. By moving the event loop model,
> it
> > > seems
> > > > > > >> that we
> > > > > > >> > > > could
> > > > > > >> > > > > > add
> > > > > > >> > > > > > > > a metric that tracks the fraction of the time
> the
> > > event
> > > > > > >> loop is
> > > > > > >> > > > doing
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > actual work.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 33. Could we add a section on coordinator
> failover
> > > > > > >> handling? For
> > > > > > >> > > > > > example,
> > > > > > >> > > > > > > > does it need to trigger the check if any group
> with
> > > the
> > > > > > >> wildcard
> > > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we
> document
> > > what
> > > > > > >> the epoch
> > > > > > >> > > > > > field
> > > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > > >> ConsumerGroupMetadataValue
> > > > > > >> > > > > > reflect
> > > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > > following
> > > > > > >> > > > invariants
> > > > > > >> > > > > > are
> > > > > > >> > > > > > > > met: ... All members exists." It's possible for
> a
> > > member
> > > > > > >> not to
> > > > > > >> > > > get any
> > > > > > >> > > > > > > > assigned partitions, right?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 36. "He can rejoins the group with a member
> epoch
> > > > > equals to
> > > > > > >> 0":
> > > > > > >> > > > When
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > a consumer rejoin and what member id would be
> used?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 37. "Instead, power users will have the ability
> to
> > > > > trigger a
> > > > > > >> > > > > > reassignment
> > > > > > >> > > > > > > > by either providing a non-zero reason or by
> > > updating the
> > > > > > >> assignor
> > > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting
> with
> > > the
> > > > > > >> deprecation
> > > > > > >> > > > of
> > > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > > section
> > > > > > >> seems to
> > > > > > >> > > > have
> > > > > > >> > > > > > > > multiple typos.
> > > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > > immediately
> > > > > > >> gets into
> > > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > > incorrect.
> > > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C
> seems
> > > to
> > > > > get
> > > > > > >> into
> > > > > > >> > > > > > epoch=3,
> > > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still
> has A -
> > > > > > >> epoch=2,
> > > > > > >> > > > > > > > partitions=[foo-0].
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support
> the
> > > > > upgrade
> > > > > > >> from
> > > > > > >> > > > any
> > > > > > >> > > > > > old
> > > > > > >> > > > > > > > version to new one?
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Thanks,
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > Jun
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > wrote:
> > > > > > >> > > > > > > >
> > > > > > >> > > > > > > > > Hi all,
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > > coordinator
> > > > > > >> in
> > > > > > >> > > > Java. I
> > > > > > >> > > > > > > > > discussed this offline with a few folks and
> folks
> > > are
> > > > > > >> concerned
> > > > > > >> > > > that
> > > > > > >> > > > > > > > > we could introduce many regressions in the old
> > > > > protocol
> > > > > > >> if we do
> > > > > > >> > > > so.
> > > > > > >> > > > > > > > > Therefore, I am going to remove this statement
> > > from
> > > > > the
> > > > > > >> KIP. It
> > > > > > >> > > > is an
> > > > > > >> > > > > > > > > implementation detail after all so it does not
> > > have
> > > > > to be
> > > > > > >> > > > decided at
> > > > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > > > refactor the
> > > > > > >> > > > current
> > > > > > >> > > > > > > > > implementation as a first step.
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > Cheers,
> > > > > > >> > > > > > > > > David
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > > > >> djacot@confluent.io
> > > > > > >> > > > >
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Hi Luke,
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> "Empty,
> > > > > assigning,
> > > > > > >> > > > > > reconciling,
> > > > > > >> > > > > > > > > stable,
> > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > section,
> > > > > > >> right?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > This sentence does not refer to those group
> > > states
> > > > > but
> > > > > > >> rather
> > > > > > >> > > > to a
> > > > > > >> > > > > > > > > > state machine replication (SMR). This
> refers to
> > > the
> > > > > > >> entire
> > > > > > >> > > > state of
> > > > > > >> > > > > > > > > > group coordinator which is replicated via
> the
> > > log
> > > > > > >> layer. I will
> > > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> is
> > > > > modelled
> > > > > > >> as an
> > > > > > >> > > > event
> > > > > > >> > > > > > > > > loop"?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > The idea is to follow a model similar to
> the new
> > > > > quorum
> > > > > > >> > > > > > controller. We
> > > > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > > > >> __consumer_offsets
> > > > > > >> > > > > > > > > > partition is assigned to a unique thread and
> > > all the
> > > > > > >> events
> > > > > > >> > > > (e.g.
> > > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by
> this
> > > > > > >> thread. This
> > > > > > >> > > > > > simplify
> > > > > > >> > > > > > > > > > concurrency and will enable us to do
> simulation
> > > > > testing
> > > > > > >> for the
> > > > > > >> > > > > > group
> > > > > > >> > > > > > > > > > coordinator.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > >> *__consumer_offsets*
> > > > > > >> > > > > > > > > partitions?
> > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > > owned by
> > > > > a
> > > > > > >> group
> > > > > > >> > > > > > coordinator?
> > > > > > >> > > > > > > > > For
> > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > consumer
> > > > > > >> groups, and
> > > > > > >> > > > > > both
> > > > > > >> > > > > > > > > exist in
> > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> state
> > > > > machine
> > > > > > >> for it,
> > > > > > >> > > > or
> > > > > > >> > > > > > 2?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > > think.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.4. I know the
> "*group.coordinator.threads"
> > > *is
> > > > > the
> > > > > > >> number
> > > > > > >> > > > of
> > > > > > >> > > > > > threads
> > > > > > >> > > > > > > > > used
> > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> wondering
> > > if
> > > > > the
> > > > > > >> purpose
> > > > > > >> > > > of
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > threads
> > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > group
> > > > > (or
> > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> why
> > > > > should we
> > > > > > >> need
> > > > > > >> > > > > > > > > multi-threads
> > > > > > >> > > > > > > > > > > here?
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> > > shard the
> > > > > > >> > > > processing as
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > computation could be heavy.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> does
> > > the
> > > > > > >> default
> > > > > > >> > > > session
> > > > > > >> > > > > > > > > timeout not
> > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > thought
> > > > > the
> > > > > > >> min/max
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> it,
> > > no?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> 30s The
> > > > > > >> timeout to
> > > > > > >> > > > detect
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > protocol.
> > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> > > 45s The
> > > > > > >> minimum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> > > 60s The
> > > > > > >> maximum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > This is indeed a mistake. The default
> session
> > > > > timeout
> > > > > > >> should
> > > > > > >> > > > be 45s
> > > > > > >> > > > > > > > > > (the current default).
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > [range,
> > > > > > >> uniform],
> > > > > > >> > > > > > which means
> > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> like to
> > > > > know
> > > > > > >> why not
> > > > > > >> > > > > > uniform
> > > > > > >> > > > > > > > > one? I
> > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > assignor
> > > > > > >> (former
> > > > > > >> > > > sticky
> > > > > > >> > > > > > > > > assinor)
> > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > reason
> > > > > we
> > > > > > >> choose
> > > > > > >> > > > range
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > as default?
> > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> uniform
> > > The
> > > > > > >> server side
> > > > > > >> > > > > > assignors.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > The order on the server side has no
> influence
> > > > > because
> > > > > > >> the
> > > > > > >> > > > client
> > > > > > >> > > > > > must
> > > > > > >> > > > > > > > > > chose the selector that he wants to use.
> There
> > > is no
> > > > > > >> default
> > > > > > >> > > > in the
> > > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > > specified
> > > > > by
> > > > > > >> the
> > > > > > >> > > > client,
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > > request is rejected. The default client
> value
> > > for
> > > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I
> hope
> > > > > that my
> > > > > > >> > > > answers
> > > > > > >> > > > > > help
> > > > > > >> > > > > > > > > > to clarify things. I will update the KIP as
> well
> > > > > based
> > > > > > >> on your
> > > > > > >> > > > > > > > > > feedback.
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > Cheers,
> > > > > > >> > > > > > > > > > David
> > > > > > >> > > > > > > > > >
> > > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > > > >> showuon@gmail.com>
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Hi David,
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Some more questions:
> > > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > > mentioned:
> > > > > > >> > > > > > > > > > > > The new group coordinator will have a
> state
> > > > > machine
> > > > > > >> per
> > > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where
> each
> > > state
> > > > > > >> machine is
> > > > > > >> > > > > > modelled
> > > > > > >> > > > > > > > > as an
> > > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > > executed
> > > > > in
> > > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 1.1. I think the state machine are:
> "Empty,
> > > > > assigning,
> > > > > > >> > > > > > reconciling,
> > > > > > >> > > > > > > > > stable,
> > > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > > section,
> > > > > > >> right?
> > > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine
> is
> > > > > modelled
> > > > > > >> as an
> > > > > > >> > > > event
> > > > > > >> > > > > > > > > loop"?
> > > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > > >> *__consumer_offsets*
> > > > > > >> > > > > > > > > partitions?
> > > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > > owned by
> > > > > a
> > > > > > >> group
> > > > > > >> > > > > > coordinator?
> > > > > > >> > > > > > > > > For
> > > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > > consumer
> > > > > > >> groups, and
> > > > > > >> > > > > > both
> > > > > > >> > > > > > > > > exist in
> > > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1
> state
> > > > > machine
> > > > > > >> for it,
> > > > > > >> > > > or
> > > > > > >> > > > > > 2?
> > > > > > >> > > > > > > > > > > 1.4. I know the
> "*group.coordinator.threads"
> > > *is
> > > > > the
> > > > > > >> number
> > > > > > >> > > > of
> > > > > > >> > > > > > threads
> > > > > > >> > > > > > > > > used
> > > > > > >> > > > > > > > > > > to run the state machines. But I'm
> wondering
> > > if
> > > > > the
> > > > > > >> purpose
> > > > > > >> > > > of
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > threads
> > > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > > group
> > > > > (or
> > > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > > >> > > > > > > > > > > partitions?), and no heavy computation,
> why
> > > > > should we
> > > > > > >> need
> > > > > > >> > > > > > > > > multi-threads
> > > > > > >> > > > > > > > > > > here?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2. For the default value in the new
> configs:
> > > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why
> does
> > > the
> > > > > > >> default
> > > > > > >> > > > session
> > > > > > >> > > > > > > > > timeout not
> > > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > > thought
> > > > > the
> > > > > > >> min/max
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of
> it,
> > > no?
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int
> 30s The
> > > > > > >> timeout to
> > > > > > >> > > > detect
> > > > > > >> > > > > > client
> > > > > > >> > > > > > > > > > > failures when using the consumer group
> > > protocol.
> > > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> > > 45s The
> > > > > > >> minimum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> > > 60s The
> > > > > > >> maximum
> > > > > > >> > > > > > session
> > > > > > >> > > > > > > > > timeout.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > > [range,
> > > > > > >> uniform],
> > > > > > >> > > > > > which means
> > > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd
> like to
> > > > > know
> > > > > > >> why not
> > > > > > >> > > > > > uniform
> > > > > > >> > > > > > > > > one? I
> > > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > > assignor
> > > > > > >> (former
> > > > > > >> > > > sticky
> > > > > > >> > > > > > > > > assinor)
> > > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > > reason
> > > > > we
> > > > > > >> choose
> > > > > > >> > > > range
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > as default?
> > > > > > >> > > > > > > > > > > group.consumer.assignors List range,
> uniform
> > > The
> > > > > > >> server side
> > > > > > >> > > > > > assignors.
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > Thank you.
> > > > > > >> > > > > > > > > > > Luke
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen
> <
> > > > > > >> showuon@gmail.com
> > > > > > >> > > > >
> > > > > > >> > > > > > wrote:
> > > > > > >> > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > > > >> integrating with
> > > > > > >> > > > > > KIP-848,
> > > > > > >> > > > > > > > > but I
> > > > > > >> > > > > > > > > > > > think we should have a separate
> discussion
> > > > > thread
> > > > > > >> for the
> > > > > > >> > > > Kafka
> > > > > > >> > > > > > > > > Connect
> > > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > > Consumer
> > > > > > >> Rebalance
> > > > > > >> > > > > > Protocol
> > > > > > >> > > > > > > > > [1],
> > > > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > > consumer
> > > > > > >> rebalance
> > > > > > >> > > > > > protocol,
> > > > > > >> > > > > > > > > WDYT?
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > [1]
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > Thank you.
> > > > > > >> > > > > > > > > > > > Luke
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the
> feedback.
> > > > > Looks
> > > > > > >> like
> > > > > > >> > > > there's
> > > > > > >> > > > > > > > > agreement on
> > > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I
> would
> > > > > revisit
> > > > > > >> the doc
> > > > > > >> > > > and
> > > > > > >> > > > > > see
> > > > > > >> > > > > > > > > what
> > > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> Thanks!
> > > > > > >> > > > > > > > > > > >> Sagar.
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David
> Jacot
> > > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > >> > > > > > > > > > > >> wrote:
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the
> document.
> > > > > That's
> > > > > > >> really
> > > > > > >> > > > > > helpful. I
> > > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both
> Connect
> > > > > and the
> > > > > > >> > > > Consumer
> > > > > > >> > > > > > could
> > > > > > >> > > > > > > > > share
> > > > > > >> > > > > > > > > > > >> > the same underlying "engine". The
> main
> > > > > > >> difference is
> > > > > > >> > > > that
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > Consumer
> > > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > > whereas
> > > > > > >> Connect
> > > > > > >> > > > assigns
> > > > > > >> > > > > > tasks
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > workers. I see two ways to move
> forward:
> > > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > > support
> > > > > > >> different
> > > > > > >> > > > > > resource
> > > > > > >> > > > > > > > > types
> > > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > > Connect. The
> > > > > > >> dedicated
> > > > > > >> > > > APIs
> > > > > > >> > > > > > > > > would be
> > > > > > >> > > > > > > > > > > >> > similar to the new ones but
> different on
> > > the
> > > > > > >> > > > > > content/resources and
> > > > > > >> > > > > > > > > > > >> > they would rely on the same engine
> on the
> > > > > > >> coordinator
> > > > > > >> > > > side.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > I personally lean towards 2) because
> I am
> > > > > not a
> > > > > > >> fan of
> > > > > > >> > > > > > > > > overcharging
> > > > > > >> > > > > > > > > > > >> > APIs to serve different purposes.
> That
> > > being
> > > > > > >> said, I am
> > > > > > >> > > > not
> > > > > > >> > > > > > > > > opposed to
> > > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to
> do
> > > it.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > I think that we can continue to
> discuss
> > > it
> > > > > here
> > > > > > >> for now
> > > > > > >> > > > in
> > > > > > >> > > > > > order
> > > > > > >> > > > > > > > > to
> > > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible
> with
> > > what
> > > > > we
> > > > > > >> will do
> > > > > > >> > > > for
> > > > > > >> > > > > > > > > Connect in
> > > > > > >> > > > > > > > > > > >> > the future.
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > Best,
> > > > > > >> > > > > > > > > > > >> > David
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> > > Jacot <
> > > > > > >> > > > > > djacot@confluent.io>
> > > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > > through
> > > > > and
> > > > > > >> address
> > > > > > >> > > > > > your
> > > > > > >> > > > > > > > > comments
> > > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > > > feedback.
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > > >> > > > > > > > > > > >> > > David
> > > > > > >> > > > > > > > > > > >> > >
> > > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM
> Gregory
> > > > > Harris
> > > > > > >> <
> > > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's
> wonderful
> > > to see
> > > > > > >> > > > cooperative
> > > > > > >> > > > > > > > > rebalancing
> > > > > > >> > > > > > > > > > > >> > making it
> > > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > > section
> > > > > > >> describes how
> > > > > > >> > > > > > member
> > > > > > >> > > > > > > > > epoch
> > > > > > >> > > > > > > > > > > >> > should
> > > > > > >> > > > > > > > > > > >> > > > advance in step with the group
> epoch
> > > and
> > > > > > >> assignment
> > > > > > >> > > > > > epoch
> > > > > > >> > > > > > > > > values. I
> > > > > > >> > > > > > > > > > > >> > think
> > > > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > > reasons
> > > > > > >> described
> > > > > > >> > > > in
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > KIP. When
> > > > > > >> > > > > > > > > > > >> the
> > > > > > >> > > > > > > > > > > >> > > > protocol is incrementally
> assigning
> > > > > > >> partitions to a
> > > > > > >> > > > > > worker,
> > > > > > >> > > > > > > > > what
> > > > > > >> > > > > > > > > > > >> member
> > > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > > assignment
> > > > > use?
> > > > > > >> Are
> > > > > > >> > > > member
> > > > > > >> > > > > > epochs
> > > > > > >> > > > > > > > > > > >> re-used,
> > > > > > >> > > > > > > > > > > >> > and
> > > > > > >> > > > > > > > > > > >> > > > a single member epoch can
> correspond
> > > to
> > > > > > >> multiple
> > > > > > >> > > > > > different
> > > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason'
> field
> > > > > opaque
> > > > > > >> to the
> > > > > > >> > > > group
> > > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > > assignor
> > > > > > >> > > > implementations
> > > > > > >> > > > > > > > > interact
> > > > > > >> > > > > > > > > > > >> with
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its
> common
> > > > > meaning
> > > > > > >> agreed
> > > > > > >> > > > > > upon? If
> > > > > > >> > > > > > > > > so, what
> > > > > > >> > > > > > > > > > > >> > is the
> > > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason
> field
> > > over
> > > > > > >> including
> > > > > > >> > > > such
> > > > > > >> > > > > > > > > functionality
> > > > > > >> > > > > > > > > > > >> > in the
> > > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 3. The following is included in
> the
> > > KIP:
> > > > > > >> "Thanks to
> > > > > > >> > > > > > this, the
> > > > > > >> > > > > > > > > input
> > > > > > >> > > > > > > > > > > >> of
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> > > driven
> > > > > by
> > > > > > >> the group
> > > > > > >> > > > > > > > > coordinator.
> > > > > > >> > > > > > > > > > > >> The
> > > > > > >> > > > > > > > > > > >> > > > consumer is no longer
> responsible for
> > > > > > >> maintaining
> > > > > > >> > > > any
> > > > > > >> > > > > > state
> > > > > > >> > > > > > > > > besides
> > > > > > >> > > > > > > > > > > >> its
> > > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this
> mean
> > > > > that the
> > > > > > >> > > > > > client-side
> > > > > > >> > > > > > > > > assignor
> > > > > > >> > > > > > > > > > > >> MAY
> > > > > > >> > > > > > > > > > > >> > > > incorporate additional
> non-Metadata
> > > state
> > > > > > >> (such as
> > > > > > >> > > > > > partition
> > > > > > >> > > > > > > > > > > >> > throughput,
> > > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config
> topics,
> > > etc),
> > > > > or
> > > > > > >> that
> > > > > > >> > > > > > additional
> > > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > > classes
> > > > > > >> > > > > > > > > > > >> > > > for
> > > > > > >> > > > > >
> org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > >> > > > > > > > > > > >> > > > and
> > > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > >> > > > > > that
> > > > > > >> > > > > > > > > seem to
> > > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > > possible for
> > > > > > >> these two
> > > > > > >> > > > > > > > > implementations
> > > > > > >> > > > > > > > > > > >> to
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > unified? This would serve to
> promote
> > > > > feature
> > > > > > >> parity
> > > > > > >> > > > of
> > > > > > >> > > > > > > > > server-side
> > > > > > >> > > > > > > > > > > >> and
> > > > > > >> > > > > > > > > > > >> > > > client-side assignors, and would
> also
> > > > > > >> facilitate
> > > > > > >> > > > > > operational
> > > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > > >> > > > > > > > > > > >> > > > certain situations. For example,
> if a
> > > > > > >> server-side
> > > > > > >> > > > > > assignor
> > > > > > >> > > > > > > > > has some
> > > > > > >> > > > > > > > > > > >> > poor
> > > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > > deploying the
> > > > > > >> patched
> > > > > > >> > > > > > assignor to
> > > > > > >> > > > > > > > > the
> > > > > > >> > > > > > > > > > > >> > client
> > > > > > >> > > > > > > > > > > >> > > > and switching one consumer group
> to a
> > > > > > >> client-side
> > > > > > >> > > > > > assignor
> > > > > > >> > > > > > > > > may be
> > > > > > >> > > > > > > > > > > >> > faster
> > > > > > >> > > > > > > > > > > >> > > > and less risky than patching all
> of
> > > the
> > > > > > >> brokers.
> > > > > > >> > > > With
> > > > > > >> > > > > > the
> > > > > > >> > > > > > > > > currently
> > > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a
> non-trivial
> > > > > > >> > > > reimplementation
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > have
> > > > > > >> > > > > > > > > > > >> to be
> > > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs
> have
> > > > > diverged
> > > > > > >> > > > > > significantly,
> > > > > > >> > > > > > > > > then it
> > > > > > >> > > > > > > > > > > >> is
> > > > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> > > would
> > > > > not be
> > > > > > >> > > > possible.
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > --
> > > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM
> Sagar
> > > <
> > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > >
> > > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > > discuss
> > > > > how
> > > > > > >> Connect
> > > > > > >> > > > > > would
> > > > > > >> > > > > > > > > need to
> > > > > > >> > > > > > > > > > > >> > change
> > > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > > protocol.
> > > > > > >> Here's the
> > > > > > >> > > > page:
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> >
> > > > > > >> > > > > > > > > > > >>
> > > > > > >> > > > > > > > >
> > > > > > >> > > > > >
> > > > > > >> > > >
> > > > > > >>
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I
> have
> > > > > tried
> > > > > > >> to keep
> > > > > > >> > > > a
> > > > > > >> > > > > > format
> > > > > > >> > > > > > > > > > > >> similar to
> > > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > > think.
> > > > > Also,
> > > > > > >> do you
> > > > > > >> > > > > > think this
> > > > > > >> > > > > > > > > > > >> should
> > > > > > >> > > > > > > > > > > >> > be
> > > > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> > > thread
> > > > > or
> > > > > > >> is this
> > > > > > >> > > > one
> > > > > > >> > > > > > fine?
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > > Sagar <
> > > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > >> > > > > > > > > > > >> > wrote:
> > > > > > >> > > > > > > > > > > >> > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the
> doc on
> > > > > Kafka
> > > > > > >> Streams.
> > > > > > >> > > > > > Sure, I
> > > > > > >> > > > > > > > > would do
> > > > > > >> > > > > > > > > > > >> > the
> > > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with
> such a
> > > > > > >> document.
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47
> AM
> > > > > Guozhang
> > > > > > >> Wang <
> > > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > > >> > > > > > > > > > > >> > > > > >
> > > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you
> could
> > > come
> > > > > > >> back with
> > > > > > >> > > > some
> > > > > > >> > > > > > > > > analysis on
> > > > > > >> > > > > > > > > > > >> > how to
> > > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > > integration
> > > > > > >> with
> > > > > > >> > > > the new
> > > > > > >> > > > > > > > > protocol;
> > > > > > >> > > > > > > > > > > >> so
> > > > > > >> > > > > > > > > > > >> > far
> > > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the
> new
> > > > > "protocol
> > > > > > >> type"
> > > > > > >> > > > we
> > > > > > >> > > > > > did not
> > >
> >
> >
> > --
> > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Guozhang,

Regarding the version, my understanding is that the version would be
either the client software version or the request version, is this
correct? If so, we could indeed pass this information down to the
assignor via the interface. One way would be to pass a "server
context" to the assignor and that context would include that
information (and perhaps more). Is this what you are looking for?

Regarding the onAssignment, I think that I understand your point. I
suppose that the assignor could also be clever and keep track of the
last metadata to decide whether it has to do something or not. One
question that is still not clear to me is whether the assignor needs
to know all the assigned partitions upfront regardless of whether they
are already revoked or not. Do you think that we need this as well?

From an API perspective, we could have something like
onAssignment(Metadata(version, reason, metadata, assigned partitions,
pending partitions)). Where the assigned partitions are the partitions
ready to be used and the pending partitions are the one assigned to
the member but not revoked yet. I find it a bit weird that this method
would be called only once because the assignor would not know when the
pending partitions changes. That does not look like a clean API. An
alternative would be to use onAssignment(Metadata(version, reason,
metadata, combined partitions)) but this seems error prone because it
is not clear whether a partition is usable or not. Or do you think
that we should not provide the partitions but only the metadata?

Best,
David

On Fri, Sep 23, 2022 at 9:40 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hello David,
>
> On Fri, Sep 23, 2022 at 2:00 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hey,
> >
> > > Just to clarify I was asking about the `version` of the assignor (i.e. up
> > to what version that the client would support), and I do agree we would not
> > need metadata. What I have in mind is that, for some specific built-in
> > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > newer version we would have a hierarchical rack ID string format, like
> > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > would still be in old format. In this case, the broker then needs to choose
> > the old versioned assignor. I'm probably making something up here for rack
> > aware assignors, but I'm wondering if in general such an "auto-downgrade"
> > behavior would be needed still for broker-side assignor, and if yes would
> > "version" still be useful.
> >
> > Got it. That's an interesting thought. I think that the issue is that
> > the client will never tell you which version of the server-side
> > assignor should be used. Do you think that the coordinator would
> > downgrade the version if the assignment fails with a higher version? I
> > tend to believe that this should be handled within the assignor
> > itself. In the example that you mentioned, the assignor would have to
> > handle all the cases. I am not really convinced that we need this at
> > the moment.
> >
> > The version from the client side would not be indicating the broker which
> version to use, but rather which version the client would "work best with".
> Such a "version" field would not be settible by the users, since they will
> be hard-codedly bumped when the Kafka byte code version bumped.
> Back to the rack aware assignor example, if the older versioned client does
> not have a hierarchical rack ID, however if the assignment returned to them
> is assuming a hierarchical rack structure, it may not reflect the best
> workload balance among those new and old versioned clients. That means,
> when receiving the members subscriptions at the server side, if the
> versions from all these members are different, the broker's assignor may
> need to consider using the lower version logic to do the assignment. So yes
> the assignor would indeed have to handle all such cases, but it needs to do
> so such that if there are clients who would not work with certain new
> logic, it would then handle such cases automatically by e.g. still using an
> older versioned logic.
>
>
>
> > > Okay, my understanding is that the calling ordering of these callbacks
> > would be like the following:
> >
> > Yes, your examples look right.
> >
> > > I'm wondering if we would still call onAssignment just once, that encodes
> > all the assignment for this rebalance, including all the partitions that
> > should be assigned to the member but not yet assigned since they have not
> > been revoked by others. In that case the call ordering would be:
> >
> > Interesting. Is there a case for Streams where having the full
> > assignment is beneficial? For instance, I can think of the following
> > case. When a standby task is promoted to an active task, the metadata
> > would not contain the standby task anymore and the assignment may not
> > have the partition yet. In this case, Streams would stop the standby
> > tasks but not have the active task yet if my understanding of Streams
> > is correct. So knowing the full assignment could be helpful here.
> >
> > If we want to do this, we could structure the assignment given to the
> > member as follow: version, error, metadata, assigned partitions,
> > pending partitions, where the pending partitions would be the one
> > assigned to this member but not yet available. What do you think?
> >
> > Regarding onAssignment being called only once, I am not sure to fully
> > grasp the benefit yet. Does the assignor really care about this? In
> > the end, the epoch does not really matter for the assignor because it
> > has to converge its state to the desired state anyway.
> >
> > Here's my rationale (maybe rephased a bit :P ): the implementers of
> rebalance listener and assignor are two groups of people, and most users
> fall into the former group, while only very few people fall into the later
> group. For rebalance listener implementers, they just want to know when a
> partition is actually revoked or assigned to the consumer and reacts to it,
> for this purpose, `onPartitionsRevoked` and `onPartitionsAssigned` would be
> triggered interleavingly upon `poll` calls across rebalances. The usual
> logic for such rebalance listeners are metrics reporting, committing
> offsets (if they do not use Kafka for that), etc. They would not care which
> calls are from which rebalances --- in the past with eager rebalance, it
> maybe that each rebalance is associated with exactly a
> `onPartitionsRevoked` first and then a `onPartitionsAssigned`, but it would
> no longer the cases now.
>
> The implementers of the assignor though, would care about "how the
> assignment was made", that includes from which rebalance a certain
> revoke/assign decision was made, based on what metadata such assignment is
> made, etc. And that's the whole point of the `onAssignment` function since
> otherwise they can just rely on the listeners. They usually implementation
> logic of this callback is to e.g. bookkeep the assignment decision driving
> factors a.k.a. the metadata, global information that needs to be propagated
> to all members, etc. Take Streams as an example, the active processing
> tasks go along with the assigned partitions, and we can always just
> incrementally create / close them upon each rebalance listener triggers,
> when certain partitions are revoked or assigned together; standby tasks
> however are encoded with the metadata, and we can only know which standby
> tasks should we get / drop based on the `onAssignment` function, and in
> fact the creation of such tasks as a result of the metadata bookkeeping
> does not need to wait until all the partitions that are yet-assigned have
> been completely assigned to the member. Such information may not always be
> updatable in an incremental manner as the partitions-revoked /
> partitions-assigned. In such a case, it's better to just trigger this
> function "once per decision made" i.e. once per rebalance generation.
>
>
> Guozhang
>
>
>
> > Best,
> > David
> >
> > On Thu, Sep 22, 2022 at 6:01 PM Guozhang Wang <wa...@gmail.com> wrote:
> > >
> > > Hi David, thanks for all the detailed explanations. I think they all make
> > > sense. Just want to have a couple follow-ups here:
> > >
> > > > I don't really see the benefits here because server side assignors
> > > don't have metadata at all. They only assign topic-partitions. They
> > > are not supposed to generate metadata nor to receive metadata from the
> > > members.
> > >
> > > Just to clarify I was asking about the `version` of the assignor (i.e. up
> > > to what version that the client would support), and I do agree we would
> > not
> > > need metadata. What I have in mind is that, for some specific built-in
> > > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > > newer version we would have a hierarchical rack ID string format, like
> > > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > > would still be in old format. In this case, the broker then needs to
> > choose
> > > the old versioned assignor. I'm probably making something up here for
> > rack
> > > aware assignors, but I'm wondering if in general such an "auto-downgrade"
> > > behavior would be needed still for broker-side assignor, and if yes would
> > > "version" still be useful.
> > >
> > > > Yeah, that's right. Within a rebalance, `onAssignment` is called once
> > > when the member transitions to a new epoch. This one contains the full
> > > metadata provided by the client side assignor. Then, `onAssignment`
> > > can be called max N times where N is the number of partitions pending
> > > revocation by other members. Let me try to clarify this in the KIP.
> > >
> > > Okay, my understanding is that the calling ordering of these callbacks
> > > would be like the following:
> > >
> > > ----------------------------------------
> > > onPartitionsRevoked();   // just once, since we do not really need
> > > to revoke incrementally.
> > >
> > > onAssignment();    // the first call, with epoch incremented
> > > onPartitionsAssigned();   // paired with the onAssignment
> > >
> > > onAssignment();              // the first onAssignment would bump up the
> > > epoch, and the metadata reflected.
> > > onPartitionsAssigned();   // each time we get an additional assignment,
> > we
> > > call onAssignment and then paired with an onPartitionsAssigned
> > > ...
> > > onAssignment();
> > > onPartitionsAssigned();   // on each of the onAssignment calls, the
> > encoded
> > > metadata would not change, only the incrementally added partitions be
> > > reflected
> > >
> > > Is that the case?
> > >
> > > I'm wondering if we would still call onAssignment just once, that encodes
> > > all the assignment for this rebalance, including all the partitions that
> > > should be assigned to the member but not yet assigned since they have not
> > > been revoked by others. In that case the call ordering would be:
> > >
> > > ----------------------------------------
> > > onPartitionsRevoked();   // just once
> > > onAssignment();    // just once, with epoch incremented, and metadata
> > > encoded changed, the "assignment" field also reflect the final target
> > > assignment
> > > onPartitionsAssigned();   // multiple times, which represent
> > incrementally
> > > added partitions
> > > ...
> > > onPartitionsAssigned();
> > >
> > > The motivation from this is that, most users would only implement the
> > > rebalance callback listeners and hence we'd definitely need to make sure
> > > the semantics of that does not change much, and the time
> > > `onPartitionsAssigned` indicate the time when the partitions are actually
> > > assigned to it; while for assignors, the `onAssignment` is used to
> > indicate
> > > what decision is made regarding for this member, i.e. when the partitions
> > > are decided to be given to it, but not necessarily meant that it has been
> > > given, since that time should be determined by the time of
> > > `onPartitionsAssigned`. The benefits though, would be that assignor
> > > implementers would not need to reason which `onAssignment` would be the
> > > last one for this epoch.
> > >
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > > Guozhang
> > >
> > > On Thu, Sep 22, 2022 at 2:20 AM David Jacot <djacot@confluent.io.invalid
> > >
> > > wrote:
> > >
> > > > Hi Guozhang,
> > > >
> > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > > field,
> > > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> > the
> > > > > rationale seems to be assuming that we should always be able to do
> > the
> > > > > assignment at the broker-side assignor without errors. Personally I
> > think
> > > > > it's still potentially beneficial to add the Error field even for
> > > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > > topics are not recognized with the current broker's metadata. What
> > do you
> > > > > think?
> > > >
> > > > Yeah, that seems reasonable. However, I wonder if it would be better
> > > > to use an exception on the server side. This is what we usually do for
> > > > server side plugins. On the client side, we use a field because the
> > > > errors are not defined in advance.
> > > >
> > > > Your comment also makes me think about what we should do when the
> > > > server side assignor fails. I suppose that we have to keep the current
> > > > assignment until a new event occurs. For instance, in your example,
> > > > the coordinator would have to trigger a rebalance when unrecognized
> > > > topics are available. This would be part of the metadata monitoring.
> > > >
> > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > > agree
> > > > > that broker-side assignor would not need reason/metadata since they
> > are
> > > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> > is
> > > > it
> > > > > possible that we evolve our broker-side built-in assignor but the old
> > > > > versioned clients would not be able to work with the new version, in
> > > > which
> > > > > case we need to let the broker being aware of this and upgrade its
> > > > behavior
> > > > > to cooperate with the clients?
> > > >
> > > > I don't really see the benefits here because server side assignors
> > > > don't have metadata at all. They only assign topic-partitions. They
> > > > are not supposed to generate metadata nor to receive metadata from the
> > > > members.
> > > >
> > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > instead
> > > > of
> > > > > including these three fields, what about just adding the "Metadata"
> > field
> > > > > class which has these three fields? Also, there are two "Metadata"
> > > > > currently in the APIs, the first is a class that encodes
> > > > > reason/version/metadata, and the second is just the encoded metadata
> > > > bytes.
> > > > > I'm wondering what about just naming the first as memberMetadata,
> > which
> > > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > > bytebuffer field as metadataBytes?
> > > >
> > > > That's a good point. Let me try to rationalize this interface based on
> > > > your suggestions.
> > > >
> > > > Best,
> > > > David
> > > >
> > > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > > > >
> > > > > Hello David,
> > > > >
> > > > > Just had a few more nit questions about the public APIs:
> > > > >
> > > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > > field,
> > > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> > the
> > > > > rationale seems to be assuming that we should always be able to do
> > the
> > > > > assignment at the broker-side assignor without errors. Personally I
> > think
> > > > > it's still potentially beneficial to add the Error field even for
> > > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > > topics are not recognized with the current broker's metadata. What
> > do you
> > > > > think?
> > > > >
> > > > > 2) The client-side "GroupMember" has three additional fields
> > > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > > agree
> > > > > that broker-side assignor would not need reason/metadata since they
> > are
> > > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> > is
> > > > it
> > > > > possible that we evolve our broker-side built-in assignor but the old
> > > > > versioned clients would not be able to work with the new version, in
> > > > which
> > > > > case we need to let the broker being aware of this and upgrade its
> > > > behavior
> > > > > to cooperate with the clients?
> > > > >
> > > > > 3) Also related to 2) above, for the client-side "GroupMember",
> > instead
> > > > of
> > > > > including these three fields, what about just adding the "Metadata"
> > field
> > > > > class which has these three fields? Also, there are two "Metadata"
> > > > > currently in the APIs, the first is a class that encodes
> > > > > reason/version/metadata, and the second is just the encoded metadata
> > > > bytes.
> > > > > I'm wondering what about just naming the first as memberMetadata,
> > which
> > > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > > bytebuffer field as metadataBytes?
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > > > wrote:
> > > > >
> > > > > > Hello David,
> > > > > >
> > > > > > Thanks for bringing this question up. I think the main benefits as
> > you
> > > > > > listed is 2) above if it stays; just to clarify, we would only be
> > able
> > > > to
> > > > > > save one round trip if the rebalance is still triggered by the
> > broker;
> > > > if
> > > > > > the rebalance is triggered by the client then the num.round.trips
> > are
> > > > the
> > > > > > same:
> > > > > >
> > > > > > 1) With GroupPrepareAssignment:
> > > > > >
> > > > > > T0: client decides to do a new assignment, suppose it has already
> > sent
> > > > a
> > > > > > HB and hence has to wait for it to return first since only one
> > request
> > > > /
> > > > > > response can be inflight with the coordinator's socket.
> > > > > > T1: client receives the HB response, and then sends the
> > > > > > GroupPrepareAssignment request.
> > > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > > T3: it calculates the new assignment, and sends a
> > > > GroupInstallAssignment
> > > > > > request.
> > > > > >
> > > > > > In total, two round trips.
> > > > > >
> > > > > > 2) Without GroupPrepareAssignment:
> > > > > >
> > > > > > T0: client decides to do a new assignment, suppose it has already
> > sent
> > > > a
> > > > > > HB and hence has to wait for it to return first since only one
> > request
> > > > /
> > > > > > response can be inflight with the coordinator's socket.
> > > > > > T1: client receives the HB response, and then sends the new HB
> > request
> > > > > > with the flag indicating a new rebalance needed..
> > > > > > T2: the HB response with the optional member metadata map is
> > returned.
> > > > > > T3: it calculates the new assignment, and sends a
> > > > GroupInstallAssignment
> > > > > > request.
> > > > > >
> > > > > > In total, two round trips as well.
> > > > > >
> > > > > > -----------------------------
> > > > > >
> > > > > > So to complete the full picture here, we'd need to modify both HB
> > > > request
> > > > > > and response so that the client can also indicate a new rebalance
> > via
> > > > the
> > > > > > HB request as well, right?
> > > > > >
> > > > > > Assuming all above is true, I think it's okay to merge the
> > > > > > GroupPrepareAssignment into HB given that we can make the
> > additional
> > > > fields
> > > > > > encoding the full member (subscription) metadata and topic
> > metadata as
> > > > > > optional fields.
> > > > > >
> > > > > >
> > > > > > Guozhang
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > > <dj...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > >> Hi all,
> > > > > >>
> > > > > >> During an offline conversation, someone asked why we need the
> > > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> > instead
> > > > > >> provide the group state in the heartbeat response. This has a few
> > > > > >> advantages: 1) it does not require using a special error code to
> > > > > >> signal that a new assignment is required as the signal would be
> > the
> > > > > >> provided group state; 2) it removes one round trip when a client
> > side
> > > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > > >> response's definition quite large. I recall that I went with the
> > > > > >> current approach due to this.
> > > > > >>
> > > > > >> Providing the group state in the heartbeat response is appealing.
> > What
> > > > > >> do you guys think?
> > > > > >>
> > > > > >> Best,
> > > > > >> David
> > > > > >>
> > > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > > > wrote:
> > > > > >> >
> > > > > >> > Hi Guozhang,
> > > > > >> >
> > > > > >> > 1. I have added a reference to the relevant chapter instead of
> > > > > >> > repeating the whole thing. Does that work for you?
> > > > > >> >
> > > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> > about
> > > > when
> > > > > >> > a rebalance should be triggered for the non-upgraded members
> > using
> > > > the
> > > > > >> > old protocol. The section mentions that a rebalance must be
> > > > triggered
> > > > > >> > when a new assignment is installed. This implies that the group
> > > > epoch
> > > > > >> > was updated either by a native member or a non-upgraded member.
> > For
> > > > > >> > the latter, the JoinGroup request would be the trigger. I have
> > > > added a
> > > > > >> > reference to the relevant chapter in the "JoinGroup Handling"
> > > > section
> > > > > >> > as well. Does that make sense?
> > > > > >> >
> > > > > >> > Thanks,
> > > > > >> > David
> > > > > >> >
> > > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> > wangguoz@gmail.com>
> > > > > >> wrote:
> > > > > >> > >
> > > > > >> > > Hello David,
> > > > > >> > >
> > > > > >> > > Alright I think that's sufficient. Just to make that clear in
> > the
> > > > doc,
> > > > > >> > > could we update:
> > > > > >> > >
> > > > > >> > > 1) the heartbeat request handling section, stating when
> > > > coordinator
> > > > > >> will
> > > > > >> > > trigger rebalance based on the HB's member metadata / reason?
> > > > > >> > > 2) the "Rebalance Triggers" section to include what we
> > described
> > > > in
> > > > > >> "Group
> > > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > > >> > >
> > > > > >> > >
> > > > > >> > > Guozhang
> > > > > >> > >
> > > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > > >> <dj...@confluent.io.invalid>
> > > > > >> > > wrote:
> > > > > >> > >
> > > > > >> > > > Hi Guozhang,
> > > > > >> > > >
> > > > > >> > > > I thought that the assignor will always be consulted when
> > the
> > > > next
> > > > > >> > > > heartbeat request is constructed. In other words,
> > > > > >> > > > `PartitionAssignor#metadata` will be called for every
> > heartbeat.
> > > > > >> This
> > > > > >> > > > gives the opportunity for the assignor to enforce a
> > rebalance by
> > > > > >> > > > setting the reason to a non-zero value or by changing the
> > > > bytes. Do
> > > > > >> > > > you think that this is not sufficient? Are you concerned by
> > the
> > > > > >> delay?
> > > > > >> > > >
> > > > > >> > > > Best,
> > > > > >> > > > David
> > > > > >> > > >
> > > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > > wangguoz@gmail.com>
> > > > > >> wrote:
> > > > > >> > > > >
> > > > > >> > > > > Hello David,
> > > > > >> > > > >
> > > > > >> > > > > One of Jun's comments make me thinking:
> > > > > >> > > > >
> > > > > >> > > > > ```
> > > > > >> > > > > In this case, a new assignment is triggered by the client
> > side
> > > > > >> > > > > assignor. When constructing the HB, the consumer will
> > always
> > > > > >> consult
> > > > > >> > > > > the client side assignor and propagate the information to
> > the
> > > > > >> group
> > > > > >> > > > > coordinator. In other words, we don't expect users to call
> > > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > > >> > > > > ```
> > > > > >> > > > >
> > > > > >> > > > > As I looked at the current PartitionAssignor's interface,
> > we
> > > > > >> actually do
> > > > > >> > > > > not have a way yet to instruct how to construct the next
> > HB
> > > > > >> request, e.g.
> > > > > >> > > > > when the assignor wants to enforce a new rebalance with a
> > new
> > > > > >> assignment,
> > > > > >> > > > > we'd need some customizable APIs inside the
> > PartitionAssignor
> > > > to
> > > > > >> indicate
> > > > > >> > > > > the next HB telling broker about so. WDYT about adding
> > such an
> > > > > >> API on the
> > > > > >> > > > > PartitionAssignor?
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> > > > > Guozhang
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > > >> <dj...@confluent.io.invalid>
> > > > > >> > > > > wrote:
> > > > > >> > > > >
> > > > > >> > > > > > Hi Jun,
> > > > > >> > > > > >
> > > > > >> > > > > > I have updated the KIP to include your feedback. I have
> > also
> > > > > >> tried to
> > > > > >> > > > > > clarify the parts which were not cleared.
> > > > > >> > > > > >
> > > > > >> > > > > > Best,
> > > > > >> > > > > > David
> > > > > >> > > > > >
> > > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > > djacot@confluent.io
> > > > > >> >
> > > > > >> > > > wrote:
> > > > > >> > > > > > >
> > > > > >> > > > > > > Hi Jun,
> > > > > >> > > > > > >
> > > > > >> > > > > > > Thanks for your feedback. Let me start by answering
> > your
> > > > > >> questions
> > > > > >> > > > > > > inline and I will update the KIP next week.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> > the
> > > > KIP
> > > > > >> seem to
> > > > > >> > > > be
> > > > > >> > > > > > fewer
> > > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > > >> wildcard. A few
> > > > > >> > > > > > > > comments below.
> > > > > >> > > > > > >
> > > > > >> > > > > > > I would also add that the KIP removes the global sync
> > > > barrier
> > > > > >> in the
> > > > > >> > > > > > > protocol which is essential to improve group
> > stability and
> > > > > >> > > > > > > scalability, and the KIP also simplifies the client by
> > > > moving
> > > > > >> most of
> > > > > >> > > > > > > the logic to the server side.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > > support
> > > > > >> rolling
> > > > > >> > > > > > changing
> > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > >> > > > > > >
> > > > > >> > > > > > > Definitely. The group coordinator will use the
> > assignor
> > > > used
> > > > > >> by a
> > > > > >> > > > > > > majority of the members. This allows the group to move
> > > > from
> > > > > >> one
> > > > > >> > > > > > > assignor to another by a roll. This is explained in
> > the
> > > > > >> Assignor
> > > > > >> > > > > > > Selection chapter.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > > >> required in
> > > > > >> > > > every
> > > > > >> > > > > > > > request or the scenarios when it needs to be
> > filled? For
> > > > > >> example,
> > > > > >> > > > it's
> > > > > >> > > > > > not
> > > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >> > > > > > >
> > > > > >> > > > > > > The client is expected to set those fields in case of
> > a
> > > > > >> connection
> > > > > >> > > > > > > issue (e.g. timeout) or when the fields have changed
> > since
> > > > > >> the last
> > > > > >> > > > > > > HB. The server populates those fields as long as the
> > > > member
> > > > > >> is not
> > > > > >> > > > > > > fully reconciled - the member should acknowledge that
> > it
> > > > has
> > > > > >> the
> > > > > >> > > > > > > expected epoch and assignment. I will clarify this in
> > the
> > > > KIP.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > affinity
> > > > > >> between the
> > > > > >> > > > > > client
> > > > > >> > > > > > > > and the broker is only considered during fetching,
> > but
> > > > not
> > > > > >> during
> > > > > >> > > > > > assigning
> > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > assignment
> > > > is
> > > > > >> made,
> > > > > >> > > > there
> > > > > >> > > > > > is
> > > > > >> > > > > > > > no opportunity for read affinity because no
> > replicas of
> > > > > >> assigned
> > > > > >> > > > > > partitions
> > > > > >> > > > > > > > are close to the member. I am wondering if we
> > should use
> > > > > >> this
> > > > > >> > > > > > opportunity
> > > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > > >> > > > > > >
> > > > > >> > > > > > > That's an interesting idea. I don't see any issue with
> > > > adding
> > > > > >> the
> > > > > >> > > > rack
> > > > > >> > > > > > > to the members. I will do so.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> > how
> > > > busy
> > > > > >> a group
> > > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > > > >> that we
> > > > > >> > > > could
> > > > > >> > > > > > add
> > > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > event
> > > > > >> loop is
> > > > > >> > > > doing
> > > > > >> > > > > > the
> > > > > >> > > > > > > > actual work.
> > > > > >> > > > > > >
> > > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > > >> handling? For
> > > > > >> > > > > > example,
> > > > > >> > > > > > > > does it need to trigger the check if any group with
> > the
> > > > > >> wildcard
> > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > >> > > > > > >
> > > > > >> > > > > > > Sure. When the new group coordinator takes over, it
> > has
> > > > to:
> > > > > >> > > > > > > * Setup the session timeouts.
> > > > > >> > > > > > > * Trigger a new assignment if a client side assignor
> > is
> > > > used.
> > > > > >> We
> > > > > >> > > > don't
> > > > > >> > > > > > > store the information about the member selected to
> > run the
> > > > > >> assignment
> > > > > >> > > > > > > so we have to start a new one.
> > > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > > >> subscriptions, and
> > > > > >> > > > > > > trigger a rebalance if needed.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > > > >> the epoch
> > > > > >> > > > > > field
> > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > >> ConsumerGroupMetadataValue
> > > > > >> > > > > > reflect
> > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > >> > > > > > >
> > > > > >> > > > > > > Sure. I will clarify that but it is always the latest
> > > > group
> > > > > >> epoch.
> > > > > >> > > > > > > When the group state is updated, the group epoch is
> > > > bumped so
> > > > > >> we use
> > > > > >> > > > > > > that one for all the change records related to the
> > update.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > following
> > > > > >> > > > invariants
> > > > > >> > > > > > are
> > > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > > > >> not to
> > > > > >> > > > get any
> > > > > >> > > > > > > > assigned partitions, right?
> > > > > >> > > > > > >
> > > > > >> > > > > > > That's right. Here I meant that the members provided
> > by
> > > > the
> > > > > >> assignor
> > > > > >> > > > > > > in the assignment must exist in the group. The
> > assignor
> > > > can
> > > > > >> not make
> > > > > >> > > > > > > up new member ids.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > > equals to
> > > > > >> 0":
> > > > > >> > > > When
> > > > > >> > > > > > would
> > > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > > >> > > > > > >
> > > > > >> > > > > > > A member is expected to abandon all its partitions and
> > > > > >> rejoins when
> > > > > >> > > > it
> > > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case,
> > the
> > > > > >> group
> > > > > >> > > > > > > coordinator will have removed the member from the
> > group.
> > > > The
> > > > > >> member
> > > > > >> > > > > > > can rejoin the group with the same member id but with
> > 0 as
> > > > > >> epoch. Let
> > > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > > trigger a
> > > > > >> > > > > > reassignment
> > > > > >> > > > > > > > by either providing a non-zero reason or by
> > updating the
> > > > > >> assignor
> > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> > the
> > > > > >> deprecation
> > > > > >> > > > of
> > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > >> > > > > > >
> > > > > >> > > > > > > In this case, a new assignment is triggered by the
> > client
> > > > side
> > > > > >> > > > > > > assignor. When constructing the HB, the consumer will
> > > > always
> > > > > >> consult
> > > > > >> > > > > > > the client side assignor and propagate the
> > information to
> > > > the
> > > > > >> group
> > > > > >> > > > > > > coordinator. In other words, we don't expect users to
> > call
> > > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > section
> > > > > >> seems to
> > > > > >> > > > have
> > > > > >> > > > > > > > multiple typos.
> > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > immediately
> > > > > >> gets into
> > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > incorrect.
> > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> > to
> > > > get
> > > > > >> into
> > > > > >> > > > > > epoch=3,
> > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > > >> epoch=2,
> > > > > >> > > > > > > > partitions=[foo-0].
> > > > > >> > > > > > >
> > > > > >> > > > > > > Sorry for that! I will revise them.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > > upgrade
> > > > > >> from
> > > > > >> > > > any
> > > > > >> > > > > > old
> > > > > >> > > > > > > > version to new one?
> > > > > >> > > > > > >
> > > > > >> > > > > > > We will support upgrading from the consumer protocol
> > > > version
> > > > > >> 3,
> > > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet
> > so
> > > > the
> > > > > >> earliest
> > > > > >> > > > > > > version is unknown at the moment. This is explained
> > in the
> > > > > >> migration
> > > > > >> > > > > > > plan chapter.
> > > > > >> > > > > > >
> > > > > >> > > > > > > Thanks again for your feedback, Jun. I will update
> > the KIP
> > > > > >> based on
> > > > > >> > > > it
> > > > > >> > > > > > > next week.
> > > > > >> > > > > > >
> > > > > >> > > > > > > Best,
> > > > > >> > > > > > > David
> > > > > >> > > > > > >
> > > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > > >> <ju...@confluent.io.invalid>
> > > > > >> > > > wrote:
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Hi, David,
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> > the
> > > > KIP
> > > > > >> seem to
> > > > > >> > > > be
> > > > > >> > > > > > fewer
> > > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > > >> wildcard. A few
> > > > > >> > > > > > > > comments below.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > > support
> > > > > >> rolling
> > > > > >> > > > > > changing
> > > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > > >> required in
> > > > > >> > > > every
> > > > > >> > > > > > > > request or the scenarios when it needs to be
> > filled? For
> > > > > >> example,
> > > > > >> > > > it's
> > > > > >> > > > > > not
> > > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> > affinity
> > > > > >> between the
> > > > > >> > > > > > client
> > > > > >> > > > > > > > and the broker is only considered during fetching,
> > but
> > > > not
> > > > > >> during
> > > > > >> > > > > > assigning
> > > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> > assignment
> > > > is
> > > > > >> made,
> > > > > >> > > > there
> > > > > >> > > > > > is
> > > > > >> > > > > > > > no opportunity for read affinity because no
> > replicas of
> > > > > >> assigned
> > > > > >> > > > > > partitions
> > > > > >> > > > > > > > are close to the member. I am wondering if we
> > should use
> > > > > >> this
> > > > > >> > > > > > opportunity
> > > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> > how
> > > > busy
> > > > > >> a group
> > > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > > > >> that we
> > > > > >> > > > could
> > > > > >> > > > > > add
> > > > > >> > > > > > > > a metric that tracks the fraction of the time the
> > event
> > > > > >> loop is
> > > > > >> > > > doing
> > > > > >> > > > > > the
> > > > > >> > > > > > > > actual work.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > > >> handling? For
> > > > > >> > > > > > example,
> > > > > >> > > > > > > > does it need to trigger the check if any group with
> > the
> > > > > >> wildcard
> > > > > >> > > > > > > > subscription now has a new matching topic?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > > > >> the epoch
> > > > > >> > > > > > field
> > > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > > >> ConsumerGroupMetadataValue
> > > > > >> > > > > > reflect
> > > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > > following
> > > > > >> > > > invariants
> > > > > >> > > > > > are
> > > > > >> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > > > >> not to
> > > > > >> > > > get any
> > > > > >> > > > > > > > assigned partitions, right?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > > equals to
> > > > > >> 0":
> > > > > >> > > > When
> > > > > >> > > > > > would
> > > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > > trigger a
> > > > > >> > > > > > reassignment
> > > > > >> > > > > > > > by either providing a non-zero reason or by
> > updating the
> > > > > >> assignor
> > > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> > the
> > > > > >> deprecation
> > > > > >> > > > of
> > > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> > section
> > > > > >> seems to
> > > > > >> > > > have
> > > > > >> > > > > > > > multiple typos.
> > > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > > immediately
> > > > > >> gets into
> > > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> > incorrect.
> > > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> > to
> > > > get
> > > > > >> into
> > > > > >> > > > > > epoch=3,
> > > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > > >> epoch=2,
> > > > > >> > > > > > > > partitions=[foo-0].
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > > upgrade
> > > > > >> from
> > > > > >> > > > any
> > > > > >> > > > > > old
> > > > > >> > > > > > > > version to new one?
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Thanks,
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Jun
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > > >> > > > > > > > wrote:
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > Hi all,
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > > coordinator
> > > > > >> in
> > > > > >> > > > Java. I
> > > > > >> > > > > > > > > discussed this offline with a few folks and folks
> > are
> > > > > >> concerned
> > > > > >> > > > that
> > > > > >> > > > > > > > > we could introduce many regressions in the old
> > > > protocol
> > > > > >> if we do
> > > > > >> > > > so.
> > > > > >> > > > > > > > > Therefore, I am going to remove this statement
> > from
> > > > the
> > > > > >> KIP. It
> > > > > >> > > > is an
> > > > > >> > > > > > > > > implementation detail after all so it does not
> > have
> > > > to be
> > > > > >> > > > decided at
> > > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > > refactor the
> > > > > >> > > > current
> > > > > >> > > > > > > > > implementation as a first step.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Cheers,
> > > > > >> > > > > > > > > David
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > > >> djacot@confluent.io
> > > > > >> > > > >
> > > > > >> > > > > > wrote:
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Hi Luke,
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > > assigning,
> > > > > >> > > > > > reconciling,
> > > > > >> > > > > > > > > stable,
> > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > section,
> > > > > >> right?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > This sentence does not refer to those group
> > states
> > > > but
> > > > > >> rather
> > > > > >> > > > to a
> > > > > >> > > > > > > > > > state machine replication (SMR). This refers to
> > the
> > > > > >> entire
> > > > > >> > > > state of
> > > > > >> > > > > > > > > > group coordinator which is replicated via the
> > log
> > > > > >> layer. I will
> > > > > >> > > > > > > > > > clarify this in the KIP.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > > modelled
> > > > > >> as an
> > > > > >> > > > event
> > > > > >> > > > > > > > > loop"?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > The idea is to follow a model similar to the new
> > > > quorum
> > > > > >> > > > > > controller. We
> > > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > > >> __consumer_offsets
> > > > > >> > > > > > > > > > partition is assigned to a unique thread and
> > all the
> > > > > >> events
> > > > > >> > > > (e.g.
> > > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > > > >> thread. This
> > > > > >> > > > > > simplify
> > > > > >> > > > > > > > > > concurrency and will enable us to do simulation
> > > > testing
> > > > > >> for the
> > > > > >> > > > > > group
> > > > > >> > > > > > > > > > coordinator.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > >> *__consumer_offsets*
> > > > > >> > > > > > > > > partitions?
> > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > owned by
> > > > a
> > > > > >> group
> > > > > >> > > > > > coordinator?
> > > > > >> > > > > > > > > For
> > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > consumer
> > > > > >> groups, and
> > > > > >> > > > > > both
> > > > > >> > > > > > > > > exist in
> > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > > machine
> > > > > >> for it,
> > > > > >> > > > or
> > > > > >> > > > > > 2?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> > think.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> > *is
> > > > the
> > > > > >> number
> > > > > >> > > > of
> > > > > >> > > > > > threads
> > > > > >> > > > > > > > > used
> > > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> > if
> > > > the
> > > > > >> purpose
> > > > > >> > > > of
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > threads
> > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > group
> > > > (or
> > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > > should we
> > > > > >> need
> > > > > >> > > > > > > > > multi-threads
> > > > > >> > > > > > > > > > > here?
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> > shard the
> > > > > >> > > > processing as
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > computation could be heavy.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> > the
> > > > > >> default
> > > > > >> > > > session
> > > > > >> > > > > > > > > timeout not
> > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > thought
> > > > the
> > > > > >> min/max
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > no?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > > >> timeout to
> > > > > >> > > > detect
> > > > > >> > > > > > client
> > > > > >> > > > > > > > > > > failures when using the consumer group
> > protocol.
> > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> > 45s The
> > > > > >> minimum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> > 60s The
> > > > > >> maximum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > This is indeed a mistake. The default session
> > > > timeout
> > > > > >> should
> > > > > >> > > > be 45s
> > > > > >> > > > > > > > > > (the current default).
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > [range,
> > > > > >> uniform],
> > > > > >> > > > > > which means
> > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > > know
> > > > > >> why not
> > > > > >> > > > > > uniform
> > > > > >> > > > > > > > > one? I
> > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > > > >> (former
> > > > > >> > > > sticky
> > > > > >> > > > > > > > > assinor)
> > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > reason
> > > > we
> > > > > >> choose
> > > > > >> > > > range
> > > > > >> > > > > > > > > assignor
> > > > > >> > > > > > > > > > > as default?
> > > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> > The
> > > > > >> server side
> > > > > >> > > > > > assignors.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > The order on the server side has no influence
> > > > because
> > > > > >> the
> > > > > >> > > > client
> > > > > >> > > > > > must
> > > > > >> > > > > > > > > > chose the selector that he wants to use. There
> > is no
> > > > > >> default
> > > > > >> > > > in the
> > > > > >> > > > > > > > > > current proposal. If the assignor is not
> > specified
> > > > by
> > > > > >> the
> > > > > >> > > > client,
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > > request is rejected. The default client value
> > for
> > > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > > > that my
> > > > > >> > > > answers
> > > > > >> > > > > > help
> > > > > >> > > > > > > > > > to clarify things. I will update the KIP as well
> > > > based
> > > > > >> on your
> > > > > >> > > > > > > > > > feedback.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Cheers,
> > > > > >> > > > > > > > > > David
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > > >> showuon@gmail.com>
> > > > > >> > > > > > wrote:
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Hi David,
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Thanks for the update.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Some more questions:
> > > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> > mentioned:
> > > > > >> > > > > > > > > > > > The new group coordinator will have a state
> > > > machine
> > > > > >> per
> > > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where each
> > state
> > > > > >> machine is
> > > > > >> > > > > > modelled
> > > > > >> > > > > > > > > as an
> > > > > >> > > > > > > > > > > event loop. Those state machines will be
> > executed
> > > > in
> > > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > > assigning,
> > > > > >> > > > > > reconciling,
> > > > > >> > > > > > > > > stable,
> > > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> > section,
> > > > > >> right?
> > > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > > modelled
> > > > > >> as an
> > > > > >> > > > event
> > > > > >> > > > > > > > > loop"?
> > > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > > >> *__consumer_offsets*
> > > > > >> > > > > > > > > partitions?
> > > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> > owned by
> > > > a
> > > > > >> group
> > > > > >> > > > > > coordinator?
> > > > > >> > > > > > > > > For
> > > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> > consumer
> > > > > >> groups, and
> > > > > >> > > > > > both
> > > > > >> > > > > > > > > exist in
> > > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > > machine
> > > > > >> for it,
> > > > > >> > > > or
> > > > > >> > > > > > 2?
> > > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> > *is
> > > > the
> > > > > >> number
> > > > > >> > > > of
> > > > > >> > > > > > threads
> > > > > >> > > > > > > > > used
> > > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> > if
> > > > the
> > > > > >> purpose
> > > > > >> > > > of
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > threads
> > > > > >> > > > > > > > > > > is only to keep the state of each consumer
> > group
> > > > (or
> > > > > >> > > > > > > > > *__consumer_offsets*
> > > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > > should we
> > > > > >> need
> > > > > >> > > > > > > > > multi-threads
> > > > > >> > > > > > > > > > > here?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > 2. For the default value in the new configs:
> > > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> > the
> > > > > >> default
> > > > > >> > > > session
> > > > > >> > > > > > > > > timeout not
> > > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> > thought
> > > > the
> > > > > >> min/max
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> > no?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > > >> timeout to
> > > > > >> > > > detect
> > > > > >> > > > > > client
> > > > > >> > > > > > > > > > > failures when using the consumer group
> > protocol.
> > > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> > 45s The
> > > > > >> minimum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> > 60s The
> > > > > >> maximum
> > > > > >> > > > > > session
> > > > > >> > > > > > > > > timeout.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> > [range,
> > > > > >> uniform],
> > > > > >> > > > > > which means
> > > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > > know
> > > > > >> why not
> > > > > >> > > > > > uniform
> > > > > >> > > > > > > > > one? I
> > > > > >> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > > > >> (former
> > > > > >> > > > sticky
> > > > > >> > > > > > > > > assinor)
> > > > > >> > > > > > > > > > > for better evenly distribution. Any other
> > reason
> > > > we
> > > > > >> choose
> > > > > >> > > > range
> > > > > >> > > > > > > > > assignor
> > > > > >> > > > > > > > > > > as default?
> > > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> > The
> > > > > >> server side
> > > > > >> > > > > > assignors.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Thank you.
> > > > > >> > > > > > > > > > > Luke
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > > > >> showuon@gmail.com
> > > > > >> > > > >
> > > > > >> > > > > > wrote:
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > Hi Sagar,
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > > >> integrating with
> > > > > >> > > > > > KIP-848,
> > > > > >> > > > > > > > > but I
> > > > > >> > > > > > > > > > > > think we should have a separate discussion
> > > > thread
> > > > > >> for the
> > > > > >> > > > Kafka
> > > > > >> > > > > > > > > Connect
> > > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > Consumer
> > > > > >> Rebalance
> > > > > >> > > > > > Protocol
> > > > > >> > > > > > > > > [1],
> > > > > >> > > > > > > > > > > > and let this discussion thread focus on
> > consumer
> > > > > >> rebalance
> > > > > >> > > > > > protocol,
> > > > > >> > > > > > > > > WDYT?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > [1]
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > >
> > > > > >> > > > > >
> > > > > >> > > >
> > > > > >>
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > Thank you.
> > > > > >> > > > > > > > > > > > Luke
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > > >> > > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > > > Looks
> > > > > >> like
> > > > > >> > > > there's
> > > > > >> > > > > > > > > agreement on
> > > > > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> > > > revisit
> > > > > >> the doc
> > > > > >> > > > and
> > > > > >> > > > > > see
> > > > > >> > > > > > > > > what
> > > > > >> > > > > > > > > > > >> changes are to be made.
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > > > > >> Thanks!
> > > > > >> > > > > > > > > > > >> Sagar.
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > >> > > > > > > > > > > >> wrote:
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > > > That's
> > > > > >> really
> > > > > >> > > > > > helpful. I
> > > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > > > and the
> > > > > >> > > > Consumer
> > > > > >> > > > > > could
> > > > > >> > > > > > > > > share
> > > > > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > > > > >> difference is
> > > > > >> > > > that
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > Consumer
> > > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> > whereas
> > > > > >> Connect
> > > > > >> > > > assigns
> > > > > >> > > > > > tasks
> > > > > >> > > > > > > > > to
> > > > > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > support
> > > > > >> different
> > > > > >> > > > > > resource
> > > > > >> > > > > > > > > types
> > > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> > Connect. The
> > > > > >> dedicated
> > > > > >> > > > APIs
> > > > > >> > > > > > > > > would be
> > > > > >> > > > > > > > > > > >> > similar to the new ones but different on
> > the
> > > > > >> > > > > > content/resources and
> > > > > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > > > > >> coordinator
> > > > > >> > > > side.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > > > not a
> > > > > >> fan of
> > > > > >> > > > > > > > > overcharging
> > > > > >> > > > > > > > > > > >> > APIs to serve different purposes. That
> > being
> > > > > >> said, I am
> > > > > >> > > > not
> > > > > >> > > > > > > > > opposed to
> > > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do
> > it.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > I think that we can continue to discuss
> > it
> > > > here
> > > > > >> for now
> > > > > >> > > > in
> > > > > >> > > > > > order
> > > > > >> > > > > > > > > to
> > > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible with
> > what
> > > > we
> > > > > >> will do
> > > > > >> > > > for
> > > > > >> > > > > > > > > Connect in
> > > > > >> > > > > > > > > > > >> > the future.
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > Best,
> > > > > >> > > > > > > > > > > >> > David
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> > Jacot <
> > > > > >> > > > > > djacot@confluent.io>
> > > > > >> > > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > Hi all,
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> > through
> > > > and
> > > > > >> address
> > > > > >> > > > > > your
> > > > > >> > > > > > > > > comments
> > > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > > feedback.
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > Cheers,
> > > > > >> > > > > > > > > > > >> > > David
> > > > > >> > > > > > > > > > > >> > >
> > > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > > > Harris
> > > > > >> <
> > > > > >> > > > > > > > > gharris1727@gmail.com
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >> > wrote:
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful
> > to see
> > > > > >> > > > cooperative
> > > > > >> > > > > > > > > rebalancing
> > > > > >> > > > > > > > > > > >> > making it
> > > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> > section
> > > > > >> describes how
> > > > > >> > > > > > member
> > > > > >> > > > > > > > > epoch
> > > > > >> > > > > > > > > > > >> > should
> > > > > >> > > > > > > > > > > >> > > > advance in step with the group epoch
> > and
> > > > > >> assignment
> > > > > >> > > > > > epoch
> > > > > >> > > > > > > > > values. I
> > > > > >> > > > > > > > > > > >> > think
> > > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> > reasons
> > > > > >> described
> > > > > >> > > > in
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > KIP. When
> > > > > >> > > > > > > > > > > >> the
> > > > > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > > > >> partitions to a
> > > > > >> > > > > > worker,
> > > > > >> > > > > > > > > what
> > > > > >> > > > > > > > > > > >> member
> > > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> > assignment
> > > > use?
> > > > > >> Are
> > > > > >> > > > member
> > > > > >> > > > > > epochs
> > > > > >> > > > > > > > > > > >> re-used,
> > > > > >> > > > > > > > > > > >> > and
> > > > > >> > > > > > > > > > > >> > > > a single member epoch can correspond
> > to
> > > > > >> multiple
> > > > > >> > > > > > different
> > > > > >> > > > > > > > > > > >> > (monotonically
> > > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > > > opaque
> > > > > >> to the
> > > > > >> > > > group
> > > > > >> > > > > > > > > > > >> coordinator? If
> > > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> > assignor
> > > > > >> > > > implementations
> > > > > >> > > > > > > > > interact
> > > > > >> > > > > > > > > > > >> with
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> > > > meaning
> > > > > >> agreed
> > > > > >> > > > > > upon? If
> > > > > >> > > > > > > > > so, what
> > > > > >> > > > > > > > > > > >> > is the
> > > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field
> > over
> > > > > >> including
> > > > > >> > > > such
> > > > > >> > > > > > > > > functionality
> > > > > >> > > > > > > > > > > >> > in the
> > > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 3. The following is included in the
> > KIP:
> > > > > >> "Thanks to
> > > > > >> > > > > > this, the
> > > > > >> > > > > > > > > input
> > > > > >> > > > > > > > > > > >> of
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> > driven
> > > > by
> > > > > >> the group
> > > > > >> > > > > > > > > coordinator.
> > > > > >> > > > > > > > > > > >> The
> > > > > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > > > >> maintaining
> > > > > >> > > > any
> > > > > >> > > > > > state
> > > > > >> > > > > > > > > besides
> > > > > >> > > > > > > > > > > >> its
> > > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > > > that the
> > > > > >> > > > > > client-side
> > > > > >> > > > > > > > > assignor
> > > > > >> > > > > > > > > > > >> MAY
> > > > > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> > state
> > > > > >> (such as
> > > > > >> > > > > > partition
> > > > > >> > > > > > > > > > > >> > throughput,
> > > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics,
> > etc),
> > > > or
> > > > > >> that
> > > > > >> > > > > > additional
> > > > > >> > > > > > > > > > > >> > non-Metadata
> > > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> > classes
> > > > > >> > > > > > > > > > > >> > > > for
> > > > > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > >> > > > > > > > > > > >> > > > and
> > > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > >> > > > > > that
> > > > > >> > > > > > > > > seem to
> > > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> > possible for
> > > > > >> these two
> > > > > >> > > > > > > > > implementations
> > > > > >> > > > > > > > > > > >> to
> > > > > >> > > > > > > > > > > >> > be
> > > > > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> > > > feature
> > > > > >> parity
> > > > > >> > > > of
> > > > > >> > > > > > > > > server-side
> > > > > >> > > > > > > > > > > >> and
> > > > > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > > > >> facilitate
> > > > > >> > > > > > operational
> > > > > >> > > > > > > > > > > >> > flexibility in
> > > > > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > > > >> server-side
> > > > > >> > > > > > assignor
> > > > > >> > > > > > > > > has some
> > > > > >> > > > > > > > > > > >> > poor
> > > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> > deploying the
> > > > > >> patched
> > > > > >> > > > > > assignor to
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > >> > client
> > > > > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > > > >> client-side
> > > > > >> > > > > > assignor
> > > > > >> > > > > > > > > may be
> > > > > >> > > > > > > > > > > >> > faster
> > > > > >> > > > > > > > > > > >> > > > and less risky than patching all of
> > the
> > > > > >> brokers.
> > > > > >> > > > With
> > > > > >> > > > > > the
> > > > > >> > > > > > > > > currently
> > > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > > > >> > > > reimplementation
> > > > > >> > > > > > would
> > > > > >> > > > > > > > > have
> > > > > >> > > > > > > > > > > >> to be
> > > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > > > diverged
> > > > > >> > > > > > significantly,
> > > > > >> > > > > > > > > then it
> > > > > >> > > > > > > > > > > >> is
> > > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> > would
> > > > not be
> > > > > >> > > > possible.
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > --
> > > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar
> > <
> > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > >> > > > > > > > > > > >> > wrote:
> > > > > >> > > > > > > > > > > >> > > >
> > > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> > discuss
> > > > how
> > > > > >> Connect
> > > > > >> > > > > > would
> > > > > >> > > > > > > > > need to
> > > > > >> > > > > > > > > > > >> > change
> > > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> > protocol.
> > > > > >> Here's the
> > > > > >> > > > page:
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> >
> > > > > >> > > > > > > > > > > >>
> > > > > >> > > > > > > > >
> > > > > >> > > > > >
> > > > > >> > > >
> > > > > >>
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > > > tried
> > > > > >> to keep
> > > > > >> > > > a
> > > > > >> > > > > > format
> > > > > >> > > > > > > > > > > >> similar to
> > > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> > think.
> > > > Also,
> > > > > >> do you
> > > > > >> > > > > > think this
> > > > > >> > > > > > > > > > > >> should
> > > > > >> > > > > > > > > > > >> > be
> > > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> > thread
> > > > or
> > > > > >> is this
> > > > > >> > > > one
> > > > > >> > > > > > fine?
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> > Sagar <
> > > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > >> > > > > > > > > > > >> > wrote:
> > > > > >> > > > > > > > > > > >> > > > >
> > > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > > > Kafka
> > > > > >> Streams.
> > > > > >> > > > > > Sure, I
> > > > > >> > > > > > > > > would do
> > > > > >> > > > > > > > > > > >> > the
> > > > > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > > > >> document.
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > > > Guozhang
> > > > > >> Wang <
> > > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > > >> > > > > > > > > > > >> > > > > >
> > > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > >> > > > > > > > > > > >> > > > > >>
> > > > > >> > > > > > > > > > > >> > > > > >> It would be great if you could
> > come
> > > > > >> back with
> > > > > >> > > > some
> > > > > >> > > > > > > > > analysis on
> > > > > >> > > > > > > > > > > >> > how to
> > > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > > integration
> > > > > >> with
> > > > > >> > > > the new
> > > > > >> > > > > > > > > protocol;
> > > > > >> > > > > > > > > > > >> so
> > > > > >> > > > > > > > > > > >> > far
> > > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > > > "protocol
> > > > > >> type"
> > > > > >> > > > we
> > > > > >> > > > > > did not
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hello David,

On Fri, Sep 23, 2022 at 2:00 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hey,
>
> > Just to clarify I was asking about the `version` of the assignor (i.e. up
> to what version that the client would support), and I do agree we would not
> need metadata. What I have in mind is that, for some specific built-in
> broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> newer version we would have a hierarchical rack ID string format, like
> "tier1-tier2" etc, but if some client has not upgraded their rack ID
> would still be in old format. In this case, the broker then needs to choose
> the old versioned assignor. I'm probably making something up here for rack
> aware assignors, but I'm wondering if in general such an "auto-downgrade"
> behavior would be needed still for broker-side assignor, and if yes would
> "version" still be useful.
>
> Got it. That's an interesting thought. I think that the issue is that
> the client will never tell you which version of the server-side
> assignor should be used. Do you think that the coordinator would
> downgrade the version if the assignment fails with a higher version? I
> tend to believe that this should be handled within the assignor
> itself. In the example that you mentioned, the assignor would have to
> handle all the cases. I am not really convinced that we need this at
> the moment.
>
> The version from the client side would not be indicating the broker which
version to use, but rather which version the client would "work best with".
Such a "version" field would not be settible by the users, since they will
be hard-codedly bumped when the Kafka byte code version bumped.
Back to the rack aware assignor example, if the older versioned client does
not have a hierarchical rack ID, however if the assignment returned to them
is assuming a hierarchical rack structure, it may not reflect the best
workload balance among those new and old versioned clients. That means,
when receiving the members subscriptions at the server side, if the
versions from all these members are different, the broker's assignor may
need to consider using the lower version logic to do the assignment. So yes
the assignor would indeed have to handle all such cases, but it needs to do
so such that if there are clients who would not work with certain new
logic, it would then handle such cases automatically by e.g. still using an
older versioned logic.



> > Okay, my understanding is that the calling ordering of these callbacks
> would be like the following:
>
> Yes, your examples look right.
>
> > I'm wondering if we would still call onAssignment just once, that encodes
> all the assignment for this rebalance, including all the partitions that
> should be assigned to the member but not yet assigned since they have not
> been revoked by others. In that case the call ordering would be:
>
> Interesting. Is there a case for Streams where having the full
> assignment is beneficial? For instance, I can think of the following
> case. When a standby task is promoted to an active task, the metadata
> would not contain the standby task anymore and the assignment may not
> have the partition yet. In this case, Streams would stop the standby
> tasks but not have the active task yet if my understanding of Streams
> is correct. So knowing the full assignment could be helpful here.
>
> If we want to do this, we could structure the assignment given to the
> member as follow: version, error, metadata, assigned partitions,
> pending partitions, where the pending partitions would be the one
> assigned to this member but not yet available. What do you think?
>
> Regarding onAssignment being called only once, I am not sure to fully
> grasp the benefit yet. Does the assignor really care about this? In
> the end, the epoch does not really matter for the assignor because it
> has to converge its state to the desired state anyway.
>
> Here's my rationale (maybe rephased a bit :P ): the implementers of
rebalance listener and assignor are two groups of people, and most users
fall into the former group, while only very few people fall into the later
group. For rebalance listener implementers, they just want to know when a
partition is actually revoked or assigned to the consumer and reacts to it,
for this purpose, `onPartitionsRevoked` and `onPartitionsAssigned` would be
triggered interleavingly upon `poll` calls across rebalances. The usual
logic for such rebalance listeners are metrics reporting, committing
offsets (if they do not use Kafka for that), etc. They would not care which
calls are from which rebalances --- in the past with eager rebalance, it
maybe that each rebalance is associated with exactly a
`onPartitionsRevoked` first and then a `onPartitionsAssigned`, but it would
no longer the cases now.

The implementers of the assignor though, would care about "how the
assignment was made", that includes from which rebalance a certain
revoke/assign decision was made, based on what metadata such assignment is
made, etc. And that's the whole point of the `onAssignment` function since
otherwise they can just rely on the listeners. They usually implementation
logic of this callback is to e.g. bookkeep the assignment decision driving
factors a.k.a. the metadata, global information that needs to be propagated
to all members, etc. Take Streams as an example, the active processing
tasks go along with the assigned partitions, and we can always just
incrementally create / close them upon each rebalance listener triggers,
when certain partitions are revoked or assigned together; standby tasks
however are encoded with the metadata, and we can only know which standby
tasks should we get / drop based on the `onAssignment` function, and in
fact the creation of such tasks as a result of the metadata bookkeeping
does not need to wait until all the partitions that are yet-assigned have
been completely assigned to the member. Such information may not always be
updatable in an incremental manner as the partitions-revoked /
partitions-assigned. In such a case, it's better to just trigger this
function "once per decision made" i.e. once per rebalance generation.


Guozhang



> Best,
> David
>
> On Thu, Sep 22, 2022 at 6:01 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hi David, thanks for all the detailed explanations. I think they all make
> > sense. Just want to have a couple follow-ups here:
> >
> > > I don't really see the benefits here because server side assignors
> > don't have metadata at all. They only assign topic-partitions. They
> > are not supposed to generate metadata nor to receive metadata from the
> > members.
> >
> > Just to clarify I was asking about the `version` of the assignor (i.e. up
> > to what version that the client would support), and I do agree we would
> not
> > need metadata. What I have in mind is that, for some specific built-in
> > broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> > newer version we would have a hierarchical rack ID string format, like
> > "tier1-tier2" etc, but if some client has not upgraded their rack ID
> > would still be in old format. In this case, the broker then needs to
> choose
> > the old versioned assignor. I'm probably making something up here for
> rack
> > aware assignors, but I'm wondering if in general such an "auto-downgrade"
> > behavior would be needed still for broker-side assignor, and if yes would
> > "version" still be useful.
> >
> > > Yeah, that's right. Within a rebalance, `onAssignment` is called once
> > when the member transitions to a new epoch. This one contains the full
> > metadata provided by the client side assignor. Then, `onAssignment`
> > can be called max N times where N is the number of partitions pending
> > revocation by other members. Let me try to clarify this in the KIP.
> >
> > Okay, my understanding is that the calling ordering of these callbacks
> > would be like the following:
> >
> > ----------------------------------------
> > onPartitionsRevoked();   // just once, since we do not really need
> > to revoke incrementally.
> >
> > onAssignment();    // the first call, with epoch incremented
> > onPartitionsAssigned();   // paired with the onAssignment
> >
> > onAssignment();              // the first onAssignment would bump up the
> > epoch, and the metadata reflected.
> > onPartitionsAssigned();   // each time we get an additional assignment,
> we
> > call onAssignment and then paired with an onPartitionsAssigned
> > ...
> > onAssignment();
> > onPartitionsAssigned();   // on each of the onAssignment calls, the
> encoded
> > metadata would not change, only the incrementally added partitions be
> > reflected
> >
> > Is that the case?
> >
> > I'm wondering if we would still call onAssignment just once, that encodes
> > all the assignment for this rebalance, including all the partitions that
> > should be assigned to the member but not yet assigned since they have not
> > been revoked by others. In that case the call ordering would be:
> >
> > ----------------------------------------
> > onPartitionsRevoked();   // just once
> > onAssignment();    // just once, with epoch incremented, and metadata
> > encoded changed, the "assignment" field also reflect the final target
> > assignment
> > onPartitionsAssigned();   // multiple times, which represent
> incrementally
> > added partitions
> > ...
> > onPartitionsAssigned();
> >
> > The motivation from this is that, most users would only implement the
> > rebalance callback listeners and hence we'd definitely need to make sure
> > the semantics of that does not change much, and the time
> > `onPartitionsAssigned` indicate the time when the partitions are actually
> > assigned to it; while for assignors, the `onAssignment` is used to
> indicate
> > what decision is made regarding for this member, i.e. when the partitions
> > are decided to be given to it, but not necessarily meant that it has been
> > given, since that time should be determined by the time of
> > `onPartitionsAssigned`. The benefits though, would be that assignor
> > implementers would not need to reason which `onAssignment` would be the
> > last one for this epoch.
> >
> >
> > Guozhang
> >
> >
> >
> >
> > Guozhang
> >
> > On Thu, Sep 22, 2022 at 2:20 AM David Jacot <djacot@confluent.io.invalid
> >
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > field,
> > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> the
> > > > rationale seems to be assuming that we should always be able to do
> the
> > > > assignment at the broker-side assignor without errors. Personally I
> think
> > > > it's still potentially beneficial to add the Error field even for
> > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > topics are not recognized with the current broker's metadata. What
> do you
> > > > think?
> > >
> > > Yeah, that seems reasonable. However, I wonder if it would be better
> > > to use an exception on the server side. This is what we usually do for
> > > server side plugins. On the client side, we use a field because the
> > > errors are not defined in advance.
> > >
> > > Your comment also makes me think about what we should do when the
> > > server side assignor fails. I suppose that we have to keep the current
> > > assignment until a new event occurs. For instance, in your example,
> > > the coordinator would have to trigger a rebalance when unrecognized
> > > topics are available. This would be part of the metadata monitoring.
> > >
> > > > 2) The client-side "GroupMember" has three additional fields
> > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > agree
> > > > that broker-side assignor would not need reason/metadata since they
> are
> > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> is
> > > it
> > > > possible that we evolve our broker-side built-in assignor but the old
> > > > versioned clients would not be able to work with the new version, in
> > > which
> > > > case we need to let the broker being aware of this and upgrade its
> > > behavior
> > > > to cooperate with the clients?
> > >
> > > I don't really see the benefits here because server side assignors
> > > don't have metadata at all. They only assign topic-partitions. They
> > > are not supposed to generate metadata nor to receive metadata from the
> > > members.
> > >
> > > > 3) Also related to 2) above, for the client-side "GroupMember",
> instead
> > > of
> > > > including these three fields, what about just adding the "Metadata"
> field
> > > > class which has these three fields? Also, there are two "Metadata"
> > > > currently in the APIs, the first is a class that encodes
> > > > reason/version/metadata, and the second is just the encoded metadata
> > > bytes.
> > > > I'm wondering what about just naming the first as memberMetadata,
> which
> > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > bytebuffer field as metadataBytes?
> > >
> > > That's a good point. Let me try to rationalize this interface based on
> > > your suggestions.
> > >
> > > Best,
> > > David
> > >
> > > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > >
> > > > Hello David,
> > > >
> > > > Just had a few more nit questions about the public APIs:
> > > >
> > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > field,
> > > > while the broker-side "PartitionAssignor#Assignment" does not. And
> the
> > > > rationale seems to be assuming that we should always be able to do
> the
> > > > assignment at the broker-side assignor without errors. Personally I
> think
> > > > it's still potentially beneficial to add the Error field even for
> > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > topics are not recognized with the current broker's metadata. What
> do you
> > > > think?
> > > >
> > > > 2) The client-side "GroupMember" has three additional fields
> > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > agree
> > > > that broker-side assignor would not need reason/metadata since they
> are
> > > > blackbox strings/bytes to the assignor, but what about version? E.g.
> is
> > > it
> > > > possible that we evolve our broker-side built-in assignor but the old
> > > > versioned clients would not be able to work with the new version, in
> > > which
> > > > case we need to let the broker being aware of this and upgrade its
> > > behavior
> > > > to cooperate with the clients?
> > > >
> > > > 3) Also related to 2) above, for the client-side "GroupMember",
> instead
> > > of
> > > > including these three fields, what about just adding the "Metadata"
> field
> > > > class which has these three fields? Also, there are two "Metadata"
> > > > currently in the APIs, the first is a class that encodes
> > > > reason/version/metadata, and the second is just the encoded metadata
> > > bytes.
> > > > I'm wondering what about just naming the first as memberMetadata,
> which
> > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > bytebuffer field as metadataBytes?
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > >
> > > > > Hello David,
> > > > >
> > > > > Thanks for bringing this question up. I think the main benefits as
> you
> > > > > listed is 2) above if it stays; just to clarify, we would only be
> able
> > > to
> > > > > save one round trip if the rebalance is still triggered by the
> broker;
> > > if
> > > > > the rebalance is triggered by the client then the num.round.trips
> are
> > > the
> > > > > same:
> > > > >
> > > > > 1) With GroupPrepareAssignment:
> > > > >
> > > > > T0: client decides to do a new assignment, suppose it has already
> sent
> > > a
> > > > > HB and hence has to wait for it to return first since only one
> request
> > > /
> > > > > response can be inflight with the coordinator's socket.
> > > > > T1: client receives the HB response, and then sends the
> > > > > GroupPrepareAssignment request.
> > > > > T2: the GroupPrepareAssignment response is returned.
> > > > > T3: it calculates the new assignment, and sends a
> > > GroupInstallAssignment
> > > > > request.
> > > > >
> > > > > In total, two round trips.
> > > > >
> > > > > 2) Without GroupPrepareAssignment:
> > > > >
> > > > > T0: client decides to do a new assignment, suppose it has already
> sent
> > > a
> > > > > HB and hence has to wait for it to return first since only one
> request
> > > /
> > > > > response can be inflight with the coordinator's socket.
> > > > > T1: client receives the HB response, and then sends the new HB
> request
> > > > > with the flag indicating a new rebalance needed..
> > > > > T2: the HB response with the optional member metadata map is
> returned.
> > > > > T3: it calculates the new assignment, and sends a
> > > GroupInstallAssignment
> > > > > request.
> > > > >
> > > > > In total, two round trips as well.
> > > > >
> > > > > -----------------------------
> > > > >
> > > > > So to complete the full picture here, we'd need to modify both HB
> > > request
> > > > > and response so that the client can also indicate a new rebalance
> via
> > > the
> > > > > HB request as well, right?
> > > > >
> > > > > Assuming all above is true, I think it's okay to merge the
> > > > > GroupPrepareAssignment into HB given that we can make the
> additional
> > > fields
> > > > > encoding the full member (subscription) metadata and topic
> metadata as
> > > > > optional fields.
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > > <dj...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > >> Hi all,
> > > > >>
> > > > >> During an offline conversation, someone asked why we need the
> > > > >> ConsumerGroupPrepareAssignment API and suggested that we could
> instead
> > > > >> provide the group state in the heartbeat response. This has a few
> > > > >> advantages: 1) it does not require using a special error code to
> > > > >> signal that a new assignment is required as the signal would be
> the
> > > > >> provided group state; 2) it removes one round trip when a client
> side
> > > > >> assignor is used. The downside is that it makes the heartbeat
> > > > >> response's definition quite large. I recall that I went with the
> > > > >> current approach due to this.
> > > > >>
> > > > >> Providing the group state in the heartbeat response is appealing.
> What
> > > > >> do you guys think?
> > > > >>
> > > > >> Best,
> > > > >> David
> > > > >>
> > > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > > >> >
> > > > >> > Hi Guozhang,
> > > > >> >
> > > > >> > 1. I have added a reference to the relevant chapter instead of
> > > > >> > repeating the whole thing. Does that work for you?
> > > > >> >
> > > > >> > 2. The "Rebalance Triggers" section you are referring to is
> about
> > > when
> > > > >> > a rebalance should be triggered for the non-upgraded members
> using
> > > the
> > > > >> > old protocol. The section mentions that a rebalance must be
> > > triggered
> > > > >> > when a new assignment is installed. This implies that the group
> > > epoch
> > > > >> > was updated either by a native member or a non-upgraded member.
> For
> > > > >> > the latter, the JoinGroup request would be the trigger. I have
> > > added a
> > > > >> > reference to the relevant chapter in the "JoinGroup Handling"
> > > section
> > > > >> > as well. Does that make sense?
> > > > >> >
> > > > >> > Thanks,
> > > > >> > David
> > > > >> >
> > > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <
> wangguoz@gmail.com>
> > > > >> wrote:
> > > > >> > >
> > > > >> > > Hello David,
> > > > >> > >
> > > > >> > > Alright I think that's sufficient. Just to make that clear in
> the
> > > doc,
> > > > >> > > could we update:
> > > > >> > >
> > > > >> > > 1) the heartbeat request handling section, stating when
> > > coordinator
> > > > >> will
> > > > >> > > trigger rebalance based on the HB's member metadata / reason?
> > > > >> > > 2) the "Rebalance Triggers" section to include what we
> described
> > > in
> > > > >> "Group
> > > > >> > > Epoch - Trigger a rebalance" section as well?
> > > > >> > >
> > > > >> > >
> > > > >> > > Guozhang
> > > > >> > >
> > > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > wrote:
> > > > >> > >
> > > > >> > > > Hi Guozhang,
> > > > >> > > >
> > > > >> > > > I thought that the assignor will always be consulted when
> the
> > > next
> > > > >> > > > heartbeat request is constructed. In other words,
> > > > >> > > > `PartitionAssignor#metadata` will be called for every
> heartbeat.
> > > > >> This
> > > > >> > > > gives the opportunity for the assignor to enforce a
> rebalance by
> > > > >> > > > setting the reason to a non-zero value or by changing the
> > > bytes. Do
> > > > >> > > > you think that this is not sufficient? Are you concerned by
> the
> > > > >> delay?
> > > > >> > > >
> > > > >> > > > Best,
> > > > >> > > > David
> > > > >> > > >
> > > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > > wangguoz@gmail.com>
> > > > >> wrote:
> > > > >> > > > >
> > > > >> > > > > Hello David,
> > > > >> > > > >
> > > > >> > > > > One of Jun's comments make me thinking:
> > > > >> > > > >
> > > > >> > > > > ```
> > > > >> > > > > In this case, a new assignment is triggered by the client
> side
> > > > >> > > > > assignor. When constructing the HB, the consumer will
> always
> > > > >> consult
> > > > >> > > > > the client side assignor and propagate the information to
> the
> > > > >> group
> > > > >> > > > > coordinator. In other words, we don't expect users to call
> > > > >> > > > > Consumer#enforceRebalance anymore.
> > > > >> > > > > ```
> > > > >> > > > >
> > > > >> > > > > As I looked at the current PartitionAssignor's interface,
> we
> > > > >> actually do
> > > > >> > > > > not have a way yet to instruct how to construct the next
> HB
> > > > >> request, e.g.
> > > > >> > > > > when the assignor wants to enforce a new rebalance with a
> new
> > > > >> assignment,
> > > > >> > > > > we'd need some customizable APIs inside the
> PartitionAssignor
> > > to
> > > > >> indicate
> > > > >> > > > > the next HB telling broker about so. WDYT about adding
> such an
> > > > >> API on the
> > > > >> > > > > PartitionAssignor?
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > Guozhang
> > > > >> > > > >
> > > > >> > > > >
> > > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > > >> <dj...@confluent.io.invalid>
> > > > >> > > > > wrote:
> > > > >> > > > >
> > > > >> > > > > > Hi Jun,
> > > > >> > > > > >
> > > > >> > > > > > I have updated the KIP to include your feedback. I have
> also
> > > > >> tried to
> > > > >> > > > > > clarify the parts which were not cleared.
> > > > >> > > > > >
> > > > >> > > > > > Best,
> > > > >> > > > > > David
> > > > >> > > > > >
> > > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > djacot@confluent.io
> > > > >> >
> > > > >> > > > wrote:
> > > > >> > > > > > >
> > > > >> > > > > > > Hi Jun,
> > > > >> > > > > > >
> > > > >> > > > > > > Thanks for your feedback. Let me start by answering
> your
> > > > >> questions
> > > > >> > > > > > > inline and I will update the KIP next week.
> > > > >> > > > > > >
> > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> the
> > > KIP
> > > > >> seem to
> > > > >> > > > be
> > > > >> > > > > > fewer
> > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > >> wildcard. A few
> > > > >> > > > > > > > comments below.
> > > > >> > > > > > >
> > > > >> > > > > > > I would also add that the KIP removes the global sync
> > > barrier
> > > > >> in the
> > > > >> > > > > > > protocol which is essential to improve group
> stability and
> > > > >> > > > > > > scalability, and the KIP also simplifies the client by
> > > moving
> > > > >> most of
> > > > >> > > > > > > the logic to the server side.
> > > > >> > > > > > >
> > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > support
> > > > >> rolling
> > > > >> > > > > > changing
> > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > >> > > > > > >
> > > > >> > > > > > > Definitely. The group coordinator will use the
> assignor
> > > used
> > > > >> by a
> > > > >> > > > > > > majority of the members. This allows the group to move
> > > from
> > > > >> one
> > > > >> > > > > > > assignor to another by a roll. This is explained in
> the
> > > > >> Assignor
> > > > >> > > > > > > Selection chapter.
> > > > >> > > > > > >
> > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > >> required in
> > > > >> > > > every
> > > > >> > > > > > > > request or the scenarios when it needs to be
> filled? For
> > > > >> example,
> > > > >> > > > it's
> > > > >> > > > > > not
> > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > >> > > > > > >
> > > > >> > > > > > > The client is expected to set those fields in case of
> a
> > > > >> connection
> > > > >> > > > > > > issue (e.g. timeout) or when the fields have changed
> since
> > > > >> the last
> > > > >> > > > > > > HB. The server populates those fields as long as the
> > > member
> > > > >> is not
> > > > >> > > > > > > fully reconciled - the member should acknowledge that
> it
> > > has
> > > > >> the
> > > > >> > > > > > > expected epoch and assignment. I will clarify this in
> the
> > > KIP.
> > > > >> > > > > > >
> > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> affinity
> > > > >> between the
> > > > >> > > > > > client
> > > > >> > > > > > > > and the broker is only considered during fetching,
> but
> > > not
> > > > >> during
> > > > >> > > > > > assigning
> > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> assignment
> > > is
> > > > >> made,
> > > > >> > > > there
> > > > >> > > > > > is
> > > > >> > > > > > > > no opportunity for read affinity because no
> replicas of
> > > > >> assigned
> > > > >> > > > > > partitions
> > > > >> > > > > > > > are close to the member. I am wondering if we
> should use
> > > > >> this
> > > > >> > > > > > opportunity
> > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > >> > > > > > >
> > > > >> > > > > > > That's an interesting idea. I don't see any issue with
> > > adding
> > > > >> the
> > > > >> > > > rack
> > > > >> > > > > > > to the members. I will do so.
> > > > >> > > > > > >
> > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> how
> > > busy
> > > > >> a group
> > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> seems
> > > > >> that we
> > > > >> > > > could
> > > > >> > > > > > add
> > > > >> > > > > > > > a metric that tracks the fraction of the time the
> event
> > > > >> loop is
> > > > >> > > > doing
> > > > >> > > > > > the
> > > > >> > > > > > > > actual work.
> > > > >> > > > > > >
> > > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > > >> > > > > > >
> > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > >> handling? For
> > > > >> > > > > > example,
> > > > >> > > > > > > > does it need to trigger the check if any group with
> the
> > > > >> wildcard
> > > > >> > > > > > > > subscription now has a new matching topic?
> > > > >> > > > > > >
> > > > >> > > > > > > Sure. When the new group coordinator takes over, it
> has
> > > to:
> > > > >> > > > > > > * Setup the session timeouts.
> > > > >> > > > > > > * Trigger a new assignment if a client side assignor
> is
> > > used.
> > > > >> We
> > > > >> > > > don't
> > > > >> > > > > > > store the information about the member selected to
> run the
> > > > >> assignment
> > > > >> > > > > > > so we have to start a new one.
> > > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > > >> subscriptions, and
> > > > >> > > > > > > trigger a rebalance if needed.
> > > > >> > > > > > >
> > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> what
> > > > >> the epoch
> > > > >> > > > > > field
> > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > >> ConsumerGroupMetadataValue
> > > > >> > > > > > reflect
> > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > >> > > > > > >
> > > > >> > > > > > > Sure. I will clarify that but it is always the latest
> > > group
> > > > >> epoch.
> > > > >> > > > > > > When the group state is updated, the group epoch is
> > > bumped so
> > > > >> we use
> > > > >> > > > > > > that one for all the change records related to the
> update.
> > > > >> > > > > > >
> > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > following
> > > > >> > > > invariants
> > > > >> > > > > > are
> > > > >> > > > > > > > met: ... All members exists." It's possible for a
> member
> > > > >> not to
> > > > >> > > > get any
> > > > >> > > > > > > > assigned partitions, right?
> > > > >> > > > > > >
> > > > >> > > > > > > That's right. Here I meant that the members provided
> by
> > > the
> > > > >> assignor
> > > > >> > > > > > > in the assignment must exist in the group. The
> assignor
> > > can
> > > > >> not make
> > > > >> > > > > > > up new member ids.
> > > > >> > > > > > >
> > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > equals to
> > > > >> 0":
> > > > >> > > > When
> > > > >> > > > > > would
> > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > >> > > > > > >
> > > > >> > > > > > > A member is expected to abandon all its partitions and
> > > > >> rejoins when
> > > > >> > > > it
> > > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case,
> the
> > > > >> group
> > > > >> > > > > > > coordinator will have removed the member from the
> group.
> > > The
> > > > >> member
> > > > >> > > > > > > can rejoin the group with the same member id but with
> 0 as
> > > > >> epoch. Let
> > > > >> > > > > > > me see if I can clarify this in the KIP.
> > > > >> > > > > > >
> > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger a
> > > > >> > > > > > reassignment
> > > > >> > > > > > > > by either providing a non-zero reason or by
> updating the
> > > > >> assignor
> > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> the
> > > > >> deprecation
> > > > >> > > > of
> > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > >> > > > > > >
> > > > >> > > > > > > In this case, a new assignment is triggered by the
> client
> > > side
> > > > >> > > > > > > assignor. When constructing the HB, the consumer will
> > > always
> > > > >> consult
> > > > >> > > > > > > the client side assignor and propagate the
> information to
> > > the
> > > > >> group
> > > > >> > > > > > > coordinator. In other words, we don't expect users to
> call
> > > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > > >> > > > > > >
> > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> section
> > > > >> seems to
> > > > >> > > > have
> > > > >> > > > > > > > multiple typos.
> > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > immediately
> > > > >> gets into
> > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> incorrect.
> > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> to
> > > get
> > > > >> into
> > > > >> > > > > > epoch=3,
> > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > >> epoch=2,
> > > > >> > > > > > > > partitions=[foo-0].
> > > > >> > > > > > >
> > > > >> > > > > > > Sorry for that! I will revise them.
> > > > >> > > > > > >
> > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > upgrade
> > > > >> from
> > > > >> > > > any
> > > > >> > > > > > old
> > > > >> > > > > > > > version to new one?
> > > > >> > > > > > >
> > > > >> > > > > > > We will support upgrading from the consumer protocol
> > > version
> > > > >> 3,
> > > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet
> so
> > > the
> > > > >> earliest
> > > > >> > > > > > > version is unknown at the moment. This is explained
> in the
> > > > >> migration
> > > > >> > > > > > > plan chapter.
> > > > >> > > > > > >
> > > > >> > > > > > > Thanks again for your feedback, Jun. I will update
> the KIP
> > > > >> based on
> > > > >> > > > it
> > > > >> > > > > > > next week.
> > > > >> > > > > > >
> > > > >> > > > > > > Best,
> > > > >> > > > > > > David
> > > > >> > > > > > >
> > > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > > >> <ju...@confluent.io.invalid>
> > > > >> > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > Hi, David,
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of
> the
> > > KIP
> > > > >> seem to
> > > > >> > > > be
> > > > >> > > > > > fewer
> > > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > > >> wildcard. A few
> > > > >> > > > > > > > comments below.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > > support
> > > > >> rolling
> > > > >> > > > > > changing
> > > > >> > > > > > > > of the partition assignor in the consumers?
> > > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > > >> required in
> > > > >> > > > every
> > > > >> > > > > > > > request or the scenarios when it needs to be
> filled? For
> > > > >> example,
> > > > >> > > > it's
> > > > >> > > > > > not
> > > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 31. In the current consumer protocol, the rack
> affinity
> > > > >> between the
> > > > >> > > > > > client
> > > > >> > > > > > > > and the broker is only considered during fetching,
> but
> > > not
> > > > >> during
> > > > >> > > > > > assigning
> > > > >> > > > > > > > partitions to consumers. Sometimes, once the
> assignment
> > > is
> > > > >> made,
> > > > >> > > > there
> > > > >> > > > > > is
> > > > >> > > > > > > > no opportunity for read affinity because no
> replicas of
> > > > >> assigned
> > > > >> > > > > > partitions
> > > > >> > > > > > > > are close to the member. I am wondering if we
> should use
> > > > >> this
> > > > >> > > > > > opportunity
> > > > >> > > > > > > > to address this by including rack in GroupMember.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 32. On the metric side, often, it's useful to know
> how
> > > busy
> > > > >> a group
> > > > >> > > > > > > > coordinator is. By moving the event loop model, it
> seems
> > > > >> that we
> > > > >> > > > could
> > > > >> > > > > > add
> > > > >> > > > > > > > a metric that tracks the fraction of the time the
> event
> > > > >> loop is
> > > > >> > > > doing
> > > > >> > > > > > the
> > > > >> > > > > > > > actual work.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > > >> handling? For
> > > > >> > > > > > example,
> > > > >> > > > > > > > does it need to trigger the check if any group with
> the
> > > > >> wildcard
> > > > >> > > > > > > > subscription now has a new matching topic?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> what
> > > > >> the epoch
> > > > >> > > > > > field
> > > > >> > > > > > > > reflects? For example, does the epoch in
> > > > >> ConsumerGroupMetadataValue
> > > > >> > > > > > reflect
> > > > >> > > > > > > > the latest group epoch? What about the one in
> > > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > > following
> > > > >> > > > invariants
> > > > >> > > > > > are
> > > > >> > > > > > > > met: ... All members exists." It's possible for a
> member
> > > > >> not to
> > > > >> > > > get any
> > > > >> > > > > > > > assigned partitions, right?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > > equals to
> > > > >> 0":
> > > > >> > > > When
> > > > >> > > > > > would
> > > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > > >> > > > > > > >
> > > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger a
> > > > >> > > > > > reassignment
> > > > >> > > > > > > > by either providing a non-zero reason or by
> updating the
> > > > >> assignor
> > > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with
> the
> > > > >> deprecation
> > > > >> > > > of
> > > > >> > > > > > > > Consumer#enforeRebalance.
> > > > >> > > > > > > >
> > > > >> > > > > > > > 38. The reassignment examples are nice. But the
> section
> > > > >> seems to
> > > > >> > > > have
> > > > >> > > > > > > > multiple typos.
> > > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > > immediately
> > > > >> gets into
> > > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems
> incorrect.
> > > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems
> to
> > > get
> > > > >> into
> > > > >> > > > > > epoch=3,
> > > > >> > > > > > > > partitions=[foo-1] too early.
> > > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > > >> epoch=2,
> > > > >> > > > > > > > partitions=[foo-0].
> > > > >> > > > > > > >
> > > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > upgrade
> > > > >> from
> > > > >> > > > any
> > > > >> > > > > > old
> > > > >> > > > > > > > version to new one?
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks,
> > > > >> > > > > > > >
> > > > >> > > > > > > > Jun
> > > > >> > > > > > > >
> > > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > >> > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > > Hi all,
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > The KIP states that we will re-implement the
> > > coordinator
> > > > >> in
> > > > >> > > > Java. I
> > > > >> > > > > > > > > discussed this offline with a few folks and folks
> are
> > > > >> concerned
> > > > >> > > > that
> > > > >> > > > > > > > > we could introduce many regressions in the old
> > > protocol
> > > > >> if we do
> > > > >> > > > so.
> > > > >> > > > > > > > > Therefore, I am going to remove this statement
> from
> > > the
> > > > >> KIP. It
> > > > >> > > > is an
> > > > >> > > > > > > > > implementation detail after all so it does not
> have
> > > to be
> > > > >> > > > decided at
> > > > >> > > > > > > > > this stage. We will likely start by trying to
> > > refactor the
> > > > >> > > > current
> > > > >> > > > > > > > > implementation as a first step.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Cheers,
> > > > >> > > > > > > > > David
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > > >> djacot@confluent.io
> > > > >> > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Hi Luke,
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > assigning,
> > > > >> > > > > > reconciling,
> > > > >> > > > > > > > > stable,
> > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> section,
> > > > >> right?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > This sentence does not refer to those group
> states
> > > but
> > > > >> rather
> > > > >> > > > to a
> > > > >> > > > > > > > > > state machine replication (SMR). This refers to
> the
> > > > >> entire
> > > > >> > > > state of
> > > > >> > > > > > > > > > group coordinator which is replicated via the
> log
> > > > >> layer. I will
> > > > >> > > > > > > > > > clarify this in the KIP.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > modelled
> > > > >> as an
> > > > >> > > > event
> > > > >> > > > > > > > > loop"?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > The idea is to follow a model similar to the new
> > > quorum
> > > > >> > > > > > controller. We
> > > > >> > > > > > > > > > will have N threads to process events. Each
> > > > >> __consumer_offsets
> > > > >> > > > > > > > > > partition is assigned to a unique thread and
> all the
> > > > >> events
> > > > >> > > > (e.g.
> > > > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > > >> thread. This
> > > > >> > > > > > simplify
> > > > >> > > > > > > > > > concurrency and will enable us to do simulation
> > > testing
> > > > >> for the
> > > > >> > > > > > group
> > > > >> > > > > > > > > > coordinator.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > >> *__consumer_offsets*
> > > > >> > > > > > > > > partitions?
> > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> owned by
> > > a
> > > > >> group
> > > > >> > > > > > coordinator?
> > > > >> > > > > > > > > For
> > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> consumer
> > > > >> groups, and
> > > > >> > > > > > both
> > > > >> > > > > > > > > exist in
> > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > > >> for it,
> > > > >> > > > or
> > > > >> > > > > > 2?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > See 1.1. The confusion comes from there, I
> think.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> *is
> > > the
> > > > >> number
> > > > >> > > > of
> > > > >> > > > > > threads
> > > > >> > > > > > > > > used
> > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> if
> > > the
> > > > >> purpose
> > > > >> > > > of
> > > > >> > > > > > the
> > > > >> > > > > > > > > threads
> > > > >> > > > > > > > > > > is only to keep the state of each consumer
> group
> > > (or
> > > > >> > > > > > > > > *__consumer_offsets*
> > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > should we
> > > > >> need
> > > > >> > > > > > > > > multi-threads
> > > > >> > > > > > > > > > > here?
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > See 1.2. The idea is to have an ability to
> shard the
> > > > >> > > > processing as
> > > > >> > > > > > the
> > > > >> > > > > > > > > > computation could be heavy.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> the
> > > > >> default
> > > > >> > > > session
> > > > >> > > > > > > > > timeout not
> > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> thought
> > > the
> > > > >> min/max
> > > > >> > > > > > session
> > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> no?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > >> timeout to
> > > > >> > > > detect
> > > > >> > > > > > client
> > > > >> > > > > > > > > > > failures when using the consumer group
> protocol.
> > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> 45s The
> > > > >> minimum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> 60s The
> > > > >> maximum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > This is indeed a mistake. The default session
> > > timeout
> > > > >> should
> > > > >> > > > be 45s
> > > > >> > > > > > > > > > (the current default).
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> [range,
> > > > >> uniform],
> > > > >> > > > > > which means
> > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > know
> > > > >> why not
> > > > >> > > > > > uniform
> > > > >> > > > > > > > > one? I
> > > > >> > > > > > > > > > > thought usually users will choose uniform
> assignor
> > > > >> (former
> > > > >> > > > sticky
> > > > >> > > > > > > > > assinor)
> > > > >> > > > > > > > > > > for better evenly distribution. Any other
> reason
> > > we
> > > > >> choose
> > > > >> > > > range
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > as default?
> > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> The
> > > > >> server side
> > > > >> > > > > > assignors.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > The order on the server side has no influence
> > > because
> > > > >> the
> > > > >> > > > client
> > > > >> > > > > > must
> > > > >> > > > > > > > > > chose the selector that he wants to use. There
> is no
> > > > >> default
> > > > >> > > > in the
> > > > >> > > > > > > > > > current proposal. If the assignor is not
> specified
> > > by
> > > > >> the
> > > > >> > > > client,
> > > > >> > > > > > the
> > > > >> > > > > > > > > > request is rejected. The default client value
> for
> > > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > > that my
> > > > >> > > > answers
> > > > >> > > > > > help
> > > > >> > > > > > > > > > to clarify things. I will update the KIP as well
> > > based
> > > > >> on your
> > > > >> > > > > > > > > > feedback.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Cheers,
> > > > >> > > > > > > > > > David
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > > >> showuon@gmail.com>
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Hi David,
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thanks for the update.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Some more questions:
> > > > >> > > > > > > > > > > 1. In Group Coordinator section, you
> mentioned:
> > > > >> > > > > > > > > > > > The new group coordinator will have a state
> > > machine
> > > > >> per
> > > > >> > > > > > > > > > > *__consumer_offsets* partitions, where each
> state
> > > > >> machine is
> > > > >> > > > > > modelled
> > > > >> > > > > > > > > as an
> > > > >> > > > > > > > > > > event loop. Those state machines will be
> executed
> > > in
> > > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > assigning,
> > > > >> > > > > > reconciling,
> > > > >> > > > > > > > > stable,
> > > > >> > > > > > > > > > > dead" mentioned in Consumer Group States
> section,
> > > > >> right?
> > > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > modelled
> > > > >> as an
> > > > >> > > > event
> > > > >> > > > > > > > > loop"?
> > > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > > >> *__consumer_offsets*
> > > > >> > > > > > > > > partitions?
> > > > >> > > > > > > > > > > Not a state machine "per consumer group"
> owned by
> > > a
> > > > >> group
> > > > >> > > > > > coordinator?
> > > > >> > > > > > > > > For
> > > > >> > > > > > > > > > > example, if one group coordinator owns 2
> consumer
> > > > >> groups, and
> > > > >> > > > > > both
> > > > >> > > > > > > > > exist in
> > > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > > >> for it,
> > > > >> > > > or
> > > > >> > > > > > 2?
> > > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads"
> *is
> > > the
> > > > >> number
> > > > >> > > > of
> > > > >> > > > > > threads
> > > > >> > > > > > > > > used
> > > > >> > > > > > > > > > > to run the state machines. But I'm wondering
> if
> > > the
> > > > >> purpose
> > > > >> > > > of
> > > > >> > > > > > the
> > > > >> > > > > > > > > threads
> > > > >> > > > > > > > > > > is only to keep the state of each consumer
> group
> > > (or
> > > > >> > > > > > > > > *__consumer_offsets*
> > > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > > should we
> > > > >> need
> > > > >> > > > > > > > > multi-threads
> > > > >> > > > > > > > > > > here?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 2. For the default value in the new configs:
> > > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does
> the
> > > > >> default
> > > > >> > > > session
> > > > >> > > > > > > > > timeout not
> > > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I
> thought
> > > the
> > > > >> min/max
> > > > >> > > > > > session
> > > > >> > > > > > > > > > > timeout is to define lower/upper bound of it,
> no?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > > >> timeout to
> > > > >> > > > detect
> > > > >> > > > > > client
> > > > >> > > > > > > > > > > failures when using the consumer group
> protocol.
> > > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int
> 45s The
> > > > >> minimum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int
> 60s The
> > > > >> maximum
> > > > >> > > > > > session
> > > > >> > > > > > > > > timeout.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 2.2. The default server side assignor are
> [range,
> > > > >> uniform],
> > > > >> > > > > > which means
> > > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > > know
> > > > >> why not
> > > > >> > > > > > uniform
> > > > >> > > > > > > > > one? I
> > > > >> > > > > > > > > > > thought usually users will choose uniform
> assignor
> > > > >> (former
> > > > >> > > > sticky
> > > > >> > > > > > > > > assinor)
> > > > >> > > > > > > > > > > for better evenly distribution. Any other
> reason
> > > we
> > > > >> choose
> > > > >> > > > range
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > as default?
> > > > >> > > > > > > > > > > group.consumer.assignors List range, uniform
> The
> > > > >> server side
> > > > >> > > > > > assignors.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thank you.
> > > > >> > > > > > > > > > > Luke
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > > >> showuon@gmail.com
> > > > >> > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > Hi Sagar,
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > > >> integrating with
> > > > >> > > > > > KIP-848,
> > > > >> > > > > > > > > but I
> > > > >> > > > > > > > > > > > think we should have a separate discussion
> > > thread
> > > > >> for the
> > > > >> > > > Kafka
> > > > >> > > > > > > > > Connect
> > > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> Consumer
> > > > >> Rebalance
> > > > >> > > > > > Protocol
> > > > >> > > > > > > > > [1],
> > > > >> > > > > > > > > > > > and let this discussion thread focus on
> consumer
> > > > >> rebalance
> > > > >> > > > > > protocol,
> > > > >> > > > > > > > > WDYT?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > [1]
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > Thank you.
> > > > >> > > > > > > > > > > > Luke
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > >> > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > > Looks
> > > > >> like
> > > > >> > > > there's
> > > > >> > > > > > > > > agreement on
> > > > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> > > revisit
> > > > >> the doc
> > > > >> > > > and
> > > > >> > > > > > see
> > > > >> > > > > > > > > what
> > > > >> > > > > > > > > > > >> changes are to be made.
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> Thanks!
> > > > >> > > > > > > > > > > >> Sagar.
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > > > > >> wrote:
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > > > > >> > Hi Sagar,
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > > That's
> > > > >> really
> > > > >> > > > > > helpful. I
> > > > >> > > > > > > > > > > >> > will take a look at it.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > > and the
> > > > >> > > > Consumer
> > > > >> > > > > > could
> > > > >> > > > > > > > > share
> > > > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > > > >> difference is
> > > > >> > > > that
> > > > >> > > > > > the
> > > > >> > > > > > > > > Consumer
> > > > >> > > > > > > > > > > >> > assigns topic-partitions to members
> whereas
> > > > >> Connect
> > > > >> > > > assigns
> > > > >> > > > > > tasks
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> support
> > > > >> different
> > > > >> > > > > > resource
> > > > >> > > > > > > > > types
> > > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for
> Connect. The
> > > > >> dedicated
> > > > >> > > > APIs
> > > > >> > > > > > > > > would be
> > > > >> > > > > > > > > > > >> > similar to the new ones but different on
> the
> > > > >> > > > > > content/resources and
> > > > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > > > >> coordinator
> > > > >> > > > side.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > > not a
> > > > >> fan of
> > > > >> > > > > > > > > overcharging
> > > > >> > > > > > > > > > > >> > APIs to serve different purposes. That
> being
> > > > >> said, I am
> > > > >> > > > not
> > > > >> > > > > > > > > opposed to
> > > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do
> it.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > I think that we can continue to discuss
> it
> > > here
> > > > >> for now
> > > > >> > > > in
> > > > >> > > > > > order
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > >> > ensure that this KIP is compatible with
> what
> > > we
> > > > >> will do
> > > > >> > > > for
> > > > >> > > > > > > > > Connect in
> > > > >> > > > > > > > > > > >> > the future.
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > Best,
> > > > >> > > > > > > > > > > >> > David
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David
> Jacot <
> > > > >> > > > > > djacot@confluent.io>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > Hi all,
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > I am back from vacation. I will go
> through
> > > and
> > > > >> address
> > > > >> > > > > > your
> > > > >> > > > > > > > > comments
> > > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > feedback.
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > Cheers,
> > > > >> > > > > > > > > > > >> > > David
> > > > >> > > > > > > > > > > >> > >
> > > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > > Harris
> > > > >> <
> > > > >> > > > > > > > > gharris1727@gmail.com
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > Hey All!
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful
> to see
> > > > >> > > > cooperative
> > > > >> > > > > > > > > rebalancing
> > > > >> > > > > > > > > > > >> > making it
> > > > >> > > > > > > > > > > >> > > > down the stack!
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives'
> section
> > > > >> describes how
> > > > >> > > > > > member
> > > > >> > > > > > > > > epoch
> > > > >> > > > > > > > > > > >> > should
> > > > >> > > > > > > > > > > >> > > > advance in step with the group epoch
> and
> > > > >> assignment
> > > > >> > > > > > epoch
> > > > >> > > > > > > > > values. I
> > > > >> > > > > > > > > > > >> > think
> > > > >> > > > > > > > > > > >> > > > that this is a good idea for the
> reasons
> > > > >> described
> > > > >> > > > in
> > > > >> > > > > > the
> > > > >> > > > > > > > > KIP. When
> > > > >> > > > > > > > > > > >> the
> > > > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > > >> partitions to a
> > > > >> > > > > > worker,
> > > > >> > > > > > > > > what
> > > > >> > > > > > > > > > > >> member
> > > > >> > > > > > > > > > > >> > > > epoch does each incremental
> assignment
> > > use?
> > > > >> Are
> > > > >> > > > member
> > > > >> > > > > > epochs
> > > > >> > > > > > > > > > > >> re-used,
> > > > >> > > > > > > > > > > >> > and
> > > > >> > > > > > > > > > > >> > > > a single member epoch can correspond
> to
> > > > >> multiple
> > > > >> > > > > > different
> > > > >> > > > > > > > > > > >> > (monotonically
> > > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > > opaque
> > > > >> to the
> > > > >> > > > group
> > > > >> > > > > > > > > > > >> coordinator? If
> > > > >> > > > > > > > > > > >> > > > not, should custom client-side
> assignor
> > > > >> > > > implementations
> > > > >> > > > > > > > > interact
> > > > >> > > > > > > > > > > >> with
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> > > meaning
> > > > >> agreed
> > > > >> > > > > > upon? If
> > > > >> > > > > > > > > so, what
> > > > >> > > > > > > > > > > >> > is the
> > > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field
> over
> > > > >> including
> > > > >> > > > such
> > > > >> > > > > > > > > functionality
> > > > >> > > > > > > > > > > >> > in the
> > > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 3. The following is included in the
> KIP:
> > > > >> "Thanks to
> > > > >> > > > > > this, the
> > > > >> > > > > > > > > input
> > > > >> > > > > > > > > > > >> of
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > client side assignor is entirely
> driven
> > > by
> > > > >> the group
> > > > >> > > > > > > > > coordinator.
> > > > >> > > > > > > > > > > >> The
> > > > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > > >> maintaining
> > > > >> > > > any
> > > > >> > > > > > state
> > > > >> > > > > > > > > besides
> > > > >> > > > > > > > > > > >> its
> > > > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > > that the
> > > > >> > > > > > client-side
> > > > >> > > > > > > > > assignor
> > > > >> > > > > > > > > > > >> MAY
> > > > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> state
> > > > >> (such as
> > > > >> > > > > > partition
> > > > >> > > > > > > > > > > >> > throughput,
> > > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics,
> etc),
> > > or
> > > > >> that
> > > > >> > > > > > additional
> > > > >> > > > > > > > > > > >> > non-Metadata
> > > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > 4. I see that there are separate
> classes
> > > > >> > > > > > > > > > > >> > > > for
> > > > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > >> > > > > > > > > > > >> > > > and
> > > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > >> > > > > > that
> > > > >> > > > > > > > > seem to
> > > > >> > > > > > > > > > > >> > > > overlap significantly. Is it
> possible for
> > > > >> these two
> > > > >> > > > > > > > > implementations
> > > > >> > > > > > > > > > > >> to
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> > > feature
> > > > >> parity
> > > > >> > > > of
> > > > >> > > > > > > > > server-side
> > > > >> > > > > > > > > > > >> and
> > > > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > > >> facilitate
> > > > >> > > > > > operational
> > > > >> > > > > > > > > > > >> > flexibility in
> > > > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > > >> server-side
> > > > >> > > > > > assignor
> > > > >> > > > > > > > > has some
> > > > >> > > > > > > > > > > >> > poor
> > > > >> > > > > > > > > > > >> > > > behavior and needs a patch,
> deploying the
> > > > >> patched
> > > > >> > > > > > assignor to
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > >> > client
> > > > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > > >> client-side
> > > > >> > > > > > assignor
> > > > >> > > > > > > > > may be
> > > > >> > > > > > > > > > > >> > faster
> > > > >> > > > > > > > > > > >> > > > and less risky than patching all of
> the
> > > > >> brokers.
> > > > >> > > > With
> > > > >> > > > > > the
> > > > >> > > > > > > > > currently
> > > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > > >> > > > reimplementation
> > > > >> > > > > > would
> > > > >> > > > > > > > > have
> > > > >> > > > > > > > > > > >> to be
> > > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > > diverged
> > > > >> > > > > > significantly,
> > > > >> > > > > > > > > then it
> > > > >> > > > > > > > > > > >> is
> > > > >> > > > > > > > > > > >> > > > possible that a reimplementation
> would
> > > not be
> > > > >> > > > possible.
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > --
> > > > >> > > > > > > > > > > >> > > > Greg Harris
> > > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar
> <
> > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > >
> > > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > I created a confluence page to
> discuss
> > > how
> > > > >> Connect
> > > > >> > > > > > would
> > > > >> > > > > > > > > need to
> > > > >> > > > > > > > > > > >> > change
> > > > >> > > > > > > > > > > >> > > > > based on the new rebalance
> protocol.
> > > > >> Here's the
> > > > >> > > > page:
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> >
> > > > >> > > > > > > > > > > >>
> > > > >> > > > > > > > >
> > > > >> > > > > >
> > > > >> > > >
> > > > >>
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > > tried
> > > > >> to keep
> > > > >> > > > a
> > > > >> > > > > > format
> > > > >> > > > > > > > > > > >> similar to
> > > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you
> think.
> > > Also,
> > > > >> do you
> > > > >> > > > > > think this
> > > > >> > > > > > > > > > > >> should
> > > > >> > > > > > > > > > > >> > be
> > > > >> > > > > > > > > > > >> > > > > moved to a separate discussion
> thread
> > > or
> > > > >> is this
> > > > >> > > > one
> > > > >> > > > > > fine?
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM
> Sagar <
> > > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > > >> > > > > > > > > > > >> > wrote:
> > > > >> > > > > > > > > > > >> > > > >
> > > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > > Kafka
> > > > >> Streams.
> > > > >> > > > > > Sure, I
> > > > >> > > > > > > > > would do
> > > > >> > > > > > > > > > > >> > the
> > > > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > > >> document.
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > > Guozhang
> > > > >> Wang <
> > > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > >> > > > > > > > > > > >> > > > > wrote:
> > > > >> > > > > > > > > > > >> > > > > >
> > > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > >> > > > > > > > > > > >> > > > > >>
> > > > >> > > > > > > > > > > >> > > > > >> It would be great if you could
> come
> > > > >> back with
> > > > >> > > > some
> > > > >> > > > > > > > > analysis on
> > > > >> > > > > > > > > > > >> > how to
> > > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > integration
> > > > >> with
> > > > >> > > > the new
> > > > >> > > > > > > > > protocol;
> > > > >> > > > > > > > > > > >> so
> > > > >> > > > > > > > > > > >> > far
> > > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > > "protocol
> > > > >> type"
> > > > >> > > > we
> > > > >> > > > > > did not
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hey,

> Just to clarify I was asking about the `version` of the assignor (i.e. up
to what version that the client would support), and I do agree we would not
need metadata. What I have in mind is that, for some specific built-in
broker-assignors, e.g. rack-aware assignors, if it's possible that in a
newer version we would have a hierarchical rack ID string format, like
"tier1-tier2" etc, but if some client has not upgraded their rack ID
would still be in old format. In this case, the broker then needs to choose
the old versioned assignor. I'm probably making something up here for rack
aware assignors, but I'm wondering if in general such an "auto-downgrade"
behavior would be needed still for broker-side assignor, and if yes would
"version" still be useful.

Got it. That's an interesting thought. I think that the issue is that
the client will never tell you which version of the server-side
assignor should be used. Do you think that the coordinator would
downgrade the version if the assignment fails with a higher version? I
tend to believe that this should be handled within the assignor
itself. In the example that you mentioned, the assignor would have to
handle all the cases. I am not really convinced that we need this at
the moment.

> Okay, my understanding is that the calling ordering of these callbacks
would be like the following:

Yes, your examples look right.

> I'm wondering if we would still call onAssignment just once, that encodes
all the assignment for this rebalance, including all the partitions that
should be assigned to the member but not yet assigned since they have not
been revoked by others. In that case the call ordering would be:

Interesting. Is there a case for Streams where having the full
assignment is beneficial? For instance, I can think of the following
case. When a standby task is promoted to an active task, the metadata
would not contain the standby task anymore and the assignment may not
have the partition yet. In this case, Streams would stop the standby
tasks but not have the active task yet if my understanding of Streams
is correct. So knowing the full assignment could be helpful here.

If we want to do this, we could structure the assignment given to the
member as follow: version, error, metadata, assigned partitions,
pending partitions, where the pending partitions would be the one
assigned to this member but not yet available. What do you think?

Regarding onAssignment being called only once, I am not sure to fully
grasp the benefit yet. Does the assignor really care about this? In
the end, the epoch does not really matter for the assignor because it
has to converge its state to the desired state anyway.

Best,
David

On Thu, Sep 22, 2022 at 6:01 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hi David, thanks for all the detailed explanations. I think they all make
> sense. Just want to have a couple follow-ups here:
>
> > I don't really see the benefits here because server side assignors
> don't have metadata at all. They only assign topic-partitions. They
> are not supposed to generate metadata nor to receive metadata from the
> members.
>
> Just to clarify I was asking about the `version` of the assignor (i.e. up
> to what version that the client would support), and I do agree we would not
> need metadata. What I have in mind is that, for some specific built-in
> broker-assignors, e.g. rack-aware assignors, if it's possible that in a
> newer version we would have a hierarchical rack ID string format, like
> "tier1-tier2" etc, but if some client has not upgraded their rack ID
> would still be in old format. In this case, the broker then needs to choose
> the old versioned assignor. I'm probably making something up here for rack
> aware assignors, but I'm wondering if in general such an "auto-downgrade"
> behavior would be needed still for broker-side assignor, and if yes would
> "version" still be useful.
>
> > Yeah, that's right. Within a rebalance, `onAssignment` is called once
> when the member transitions to a new epoch. This one contains the full
> metadata provided by the client side assignor. Then, `onAssignment`
> can be called max N times where N is the number of partitions pending
> revocation by other members. Let me try to clarify this in the KIP.
>
> Okay, my understanding is that the calling ordering of these callbacks
> would be like the following:
>
> ----------------------------------------
> onPartitionsRevoked();   // just once, since we do not really need
> to revoke incrementally.
>
> onAssignment();    // the first call, with epoch incremented
> onPartitionsAssigned();   // paired with the onAssignment
>
> onAssignment();              // the first onAssignment would bump up the
> epoch, and the metadata reflected.
> onPartitionsAssigned();   // each time we get an additional assignment, we
> call onAssignment and then paired with an onPartitionsAssigned
> ...
> onAssignment();
> onPartitionsAssigned();   // on each of the onAssignment calls, the encoded
> metadata would not change, only the incrementally added partitions be
> reflected
>
> Is that the case?
>
> I'm wondering if we would still call onAssignment just once, that encodes
> all the assignment for this rebalance, including all the partitions that
> should be assigned to the member but not yet assigned since they have not
> been revoked by others. In that case the call ordering would be:
>
> ----------------------------------------
> onPartitionsRevoked();   // just once
> onAssignment();    // just once, with epoch incremented, and metadata
> encoded changed, the "assignment" field also reflect the final target
> assignment
> onPartitionsAssigned();   // multiple times, which represent incrementally
> added partitions
> ...
> onPartitionsAssigned();
>
> The motivation from this is that, most users would only implement the
> rebalance callback listeners and hence we'd definitely need to make sure
> the semantics of that does not change much, and the time
> `onPartitionsAssigned` indicate the time when the partitions are actually
> assigned to it; while for assignors, the `onAssignment` is used to indicate
> what decision is made regarding for this member, i.e. when the partitions
> are decided to be given to it, but not necessarily meant that it has been
> given, since that time should be determined by the time of
> `onPartitionsAssigned`. The benefits though, would be that assignor
> implementers would not need to reason which `onAssignment` would be the
> last one for this epoch.
>
>
> Guozhang
>
>
>
>
> Guozhang
>
> On Thu, Sep 22, 2022 at 2:20 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Guozhang,
> >
> > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > field,
> > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > rationale seems to be assuming that we should always be able to do the
> > > assignment at the broker-side assignor without errors. Personally I think
> > > it's still potentially beneficial to add the Error field even for
> > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > topics are not recognized with the current broker's metadata. What do you
> > > think?
> >
> > Yeah, that seems reasonable. However, I wonder if it would be better
> > to use an exception on the server side. This is what we usually do for
> > server side plugins. On the client side, we use a field because the
> > errors are not defined in advance.
> >
> > Your comment also makes me think about what we should do when the
> > server side assignor fails. I suppose that we have to keep the current
> > assignment until a new event occurs. For instance, in your example,
> > the coordinator would have to trigger a rebalance when unrecognized
> > topics are available. This would be part of the metadata monitoring.
> >
> > > 2) The client-side "GroupMember" has three additional fields
> > > reason/version/metadata compared with the broker-side GroupMember. I
> > agree
> > > that broker-side assignor would not need reason/metadata since they are
> > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > it
> > > possible that we evolve our broker-side built-in assignor but the old
> > > versioned clients would not be able to work with the new version, in
> > which
> > > case we need to let the broker being aware of this and upgrade its
> > behavior
> > > to cooperate with the clients?
> >
> > I don't really see the benefits here because server side assignors
> > don't have metadata at all. They only assign topic-partitions. They
> > are not supposed to generate metadata nor to receive metadata from the
> > members.
> >
> > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > of
> > > including these three fields, what about just adding the "Metadata" field
> > > class which has these three fields? Also, there are two "Metadata"
> > > currently in the APIs, the first is a class that encodes
> > > reason/version/metadata, and the second is just the encoded metadata
> > bytes.
> > > I'm wondering what about just naming the first as memberMetadata, which
> > > then has a bytebuffer field Metadata, or instead naming the second
> > > bytebuffer field as metadataBytes?
> >
> > That's a good point. Let me try to rationalize this interface based on
> > your suggestions.
> >
> > Best,
> > David
> >
> > On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com> wrote:
> > >
> > > Hello David,
> > >
> > > Just had a few more nit questions about the public APIs:
> > >
> > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > field,
> > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > rationale seems to be assuming that we should always be able to do the
> > > assignment at the broker-side assignor without errors. Personally I think
> > > it's still potentially beneficial to add the Error field even for
> > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > topics are not recognized with the current broker's metadata. What do you
> > > think?
> > >
> > > 2) The client-side "GroupMember" has three additional fields
> > > reason/version/metadata compared with the broker-side GroupMember. I
> > agree
> > > that broker-side assignor would not need reason/metadata since they are
> > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > it
> > > possible that we evolve our broker-side built-in assignor but the old
> > > versioned clients would not be able to work with the new version, in
> > which
> > > case we need to let the broker being aware of this and upgrade its
> > behavior
> > > to cooperate with the clients?
> > >
> > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > of
> > > including these three fields, what about just adding the "Metadata" field
> > > class which has these three fields? Also, there are two "Metadata"
> > > currently in the APIs, the first is a class that encodes
> > > reason/version/metadata, and the second is just the encoded metadata
> > bytes.
> > > I'm wondering what about just naming the first as memberMetadata, which
> > > then has a bytebuffer field Metadata, or instead naming the second
> > > bytebuffer field as metadataBytes?
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > > > Hello David,
> > > >
> > > > Thanks for bringing this question up. I think the main benefits as you
> > > > listed is 2) above if it stays; just to clarify, we would only be able
> > to
> > > > save one round trip if the rebalance is still triggered by the broker;
> > if
> > > > the rebalance is triggered by the client then the num.round.trips are
> > the
> > > > same:
> > > >
> > > > 1) With GroupPrepareAssignment:
> > > >
> > > > T0: client decides to do a new assignment, suppose it has already sent
> > a
> > > > HB and hence has to wait for it to return first since only one request
> > /
> > > > response can be inflight with the coordinator's socket.
> > > > T1: client receives the HB response, and then sends the
> > > > GroupPrepareAssignment request.
> > > > T2: the GroupPrepareAssignment response is returned.
> > > > T3: it calculates the new assignment, and sends a
> > GroupInstallAssignment
> > > > request.
> > > >
> > > > In total, two round trips.
> > > >
> > > > 2) Without GroupPrepareAssignment:
> > > >
> > > > T0: client decides to do a new assignment, suppose it has already sent
> > a
> > > > HB and hence has to wait for it to return first since only one request
> > /
> > > > response can be inflight with the coordinator's socket.
> > > > T1: client receives the HB response, and then sends the new HB request
> > > > with the flag indicating a new rebalance needed..
> > > > T2: the HB response with the optional member metadata map is returned.
> > > > T3: it calculates the new assignment, and sends a
> > GroupInstallAssignment
> > > > request.
> > > >
> > > > In total, two round trips as well.
> > > >
> > > > -----------------------------
> > > >
> > > > So to complete the full picture here, we'd need to modify both HB
> > request
> > > > and response so that the client can also indicate a new rebalance via
> > the
> > > > HB request as well, right?
> > > >
> > > > Assuming all above is true, I think it's okay to merge the
> > > > GroupPrepareAssignment into HB given that we can make the additional
> > fields
> > > > encoding the full member (subscription) metadata and topic metadata as
> > > > optional fields.
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > >> Hi all,
> > > >>
> > > >> During an offline conversation, someone asked why we need the
> > > >> ConsumerGroupPrepareAssignment API and suggested that we could instead
> > > >> provide the group state in the heartbeat response. This has a few
> > > >> advantages: 1) it does not require using a special error code to
> > > >> signal that a new assignment is required as the signal would be the
> > > >> provided group state; 2) it removes one round trip when a client side
> > > >> assignor is used. The downside is that it makes the heartbeat
> > > >> response's definition quite large. I recall that I went with the
> > > >> current approach due to this.
> > > >>
> > > >> Providing the group state in the heartbeat response is appealing. What
> > > >> do you guys think?
> > > >>
> > > >> Best,
> > > >> David
> > > >>
> > > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > wrote:
> > > >> >
> > > >> > Hi Guozhang,
> > > >> >
> > > >> > 1. I have added a reference to the relevant chapter instead of
> > > >> > repeating the whole thing. Does that work for you?
> > > >> >
> > > >> > 2. The "Rebalance Triggers" section you are referring to is about
> > when
> > > >> > a rebalance should be triggered for the non-upgraded members using
> > the
> > > >> > old protocol. The section mentions that a rebalance must be
> > triggered
> > > >> > when a new assignment is installed. This implies that the group
> > epoch
> > > >> > was updated either by a native member or a non-upgraded member. For
> > > >> > the latter, the JoinGroup request would be the trigger. I have
> > added a
> > > >> > reference to the relevant chapter in the "JoinGroup Handling"
> > section
> > > >> > as well. Does that make sense?
> > > >> >
> > > >> > Thanks,
> > > >> > David
> > > >> >
> > > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> > > >> wrote:
> > > >> > >
> > > >> > > Hello David,
> > > >> > >
> > > >> > > Alright I think that's sufficient. Just to make that clear in the
> > doc,
> > > >> > > could we update:
> > > >> > >
> > > >> > > 1) the heartbeat request handling section, stating when
> > coordinator
> > > >> will
> > > >> > > trigger rebalance based on the HB's member metadata / reason?
> > > >> > > 2) the "Rebalance Triggers" section to include what we described
> > in
> > > >> "Group
> > > >> > > Epoch - Trigger a rebalance" section as well?
> > > >> > >
> > > >> > >
> > > >> > > Guozhang
> > > >> > >
> > > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > >> <dj...@confluent.io.invalid>
> > > >> > > wrote:
> > > >> > >
> > > >> > > > Hi Guozhang,
> > > >> > > >
> > > >> > > > I thought that the assignor will always be consulted when the
> > next
> > > >> > > > heartbeat request is constructed. In other words,
> > > >> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
> > > >> This
> > > >> > > > gives the opportunity for the assignor to enforce a rebalance by
> > > >> > > > setting the reason to a non-zero value or by changing the
> > bytes. Do
> > > >> > > > you think that this is not sufficient? Are you concerned by the
> > > >> delay?
> > > >> > > >
> > > >> > > > Best,
> > > >> > > > David
> > > >> > > >
> > > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > wangguoz@gmail.com>
> > > >> wrote:
> > > >> > > > >
> > > >> > > > > Hello David,
> > > >> > > > >
> > > >> > > > > One of Jun's comments make me thinking:
> > > >> > > > >
> > > >> > > > > ```
> > > >> > > > > In this case, a new assignment is triggered by the client side
> > > >> > > > > assignor. When constructing the HB, the consumer will always
> > > >> consult
> > > >> > > > > the client side assignor and propagate the information to the
> > > >> group
> > > >> > > > > coordinator. In other words, we don't expect users to call
> > > >> > > > > Consumer#enforceRebalance anymore.
> > > >> > > > > ```
> > > >> > > > >
> > > >> > > > > As I looked at the current PartitionAssignor's interface, we
> > > >> actually do
> > > >> > > > > not have a way yet to instruct how to construct the next HB
> > > >> request, e.g.
> > > >> > > > > when the assignor wants to enforce a new rebalance with a new
> > > >> assignment,
> > > >> > > > > we'd need some customizable APIs inside the PartitionAssignor
> > to
> > > >> indicate
> > > >> > > > > the next HB telling broker about so. WDYT about adding such an
> > > >> API on the
> > > >> > > > > PartitionAssignor?
> > > >> > > > >
> > > >> > > > >
> > > >> > > > > Guozhang
> > > >> > > > >
> > > >> > > > >
> > > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > >> <dj...@confluent.io.invalid>
> > > >> > > > > wrote:
> > > >> > > > >
> > > >> > > > > > Hi Jun,
> > > >> > > > > >
> > > >> > > > > > I have updated the KIP to include your feedback. I have also
> > > >> tried to
> > > >> > > > > > clarify the parts which were not cleared.
> > > >> > > > > >
> > > >> > > > > > Best,
> > > >> > > > > > David
> > > >> > > > > >
> > > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > djacot@confluent.io
> > > >> >
> > > >> > > > wrote:
> > > >> > > > > > >
> > > >> > > > > > > Hi Jun,
> > > >> > > > > > >
> > > >> > > > > > > Thanks for your feedback. Let me start by answering your
> > > >> questions
> > > >> > > > > > > inline and I will update the KIP next week.
> > > >> > > > > > >
> > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > KIP
> > > >> seem to
> > > >> > > > be
> > > >> > > > > > fewer
> > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > >> wildcard. A few
> > > >> > > > > > > > comments below.
> > > >> > > > > > >
> > > >> > > > > > > I would also add that the KIP removes the global sync
> > barrier
> > > >> in the
> > > >> > > > > > > protocol which is essential to improve group stability and
> > > >> > > > > > > scalability, and the KIP also simplifies the client by
> > moving
> > > >> most of
> > > >> > > > > > > the logic to the server side.
> > > >> > > > > > >
> > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > support
> > > >> rolling
> > > >> > > > > > changing
> > > >> > > > > > > > of the partition assignor in the consumers?
> > > >> > > > > > >
> > > >> > > > > > > Definitely. The group coordinator will use the assignor
> > used
> > > >> by a
> > > >> > > > > > > majority of the members. This allows the group to move
> > from
> > > >> one
> > > >> > > > > > > assignor to another by a roll. This is explained in the
> > > >> Assignor
> > > >> > > > > > > Selection chapter.
> > > >> > > > > > >
> > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > >> required in
> > > >> > > > every
> > > >> > > > > > > > request or the scenarios when it needs to be filled? For
> > > >> example,
> > > >> > > > it's
> > > >> > > > > > not
> > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > >> > > > > > >
> > > >> > > > > > > The client is expected to set those fields in case of a
> > > >> connection
> > > >> > > > > > > issue (e.g. timeout) or when the fields have changed since
> > > >> the last
> > > >> > > > > > > HB. The server populates those fields as long as the
> > member
> > > >> is not
> > > >> > > > > > > fully reconciled - the member should acknowledge that it
> > has
> > > >> the
> > > >> > > > > > > expected epoch and assignment. I will clarify this in the
> > KIP.
> > > >> > > > > > >
> > > >> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > > >> between the
> > > >> > > > > > client
> > > >> > > > > > > > and the broker is only considered during fetching, but
> > not
> > > >> during
> > > >> > > > > > assigning
> > > >> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > is
> > > >> made,
> > > >> > > > there
> > > >> > > > > > is
> > > >> > > > > > > > no opportunity for read affinity because no replicas of
> > > >> assigned
> > > >> > > > > > partitions
> > > >> > > > > > > > are close to the member. I am wondering if we should use
> > > >> this
> > > >> > > > > > opportunity
> > > >> > > > > > > > to address this by including rack in GroupMember.
> > > >> > > > > > >
> > > >> > > > > > > That's an interesting idea. I don't see any issue with
> > adding
> > > >> the
> > > >> > > > rack
> > > >> > > > > > > to the members. I will do so.
> > > >> > > > > > >
> > > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> > busy
> > > >> a group
> > > >> > > > > > > > coordinator is. By moving the event loop model, it seems
> > > >> that we
> > > >> > > > could
> > > >> > > > > > add
> > > >> > > > > > > > a metric that tracks the fraction of the time the event
> > > >> loop is
> > > >> > > > doing
> > > >> > > > > > the
> > > >> > > > > > > > actual work.
> > > >> > > > > > >
> > > >> > > > > > > That's a great idea. I will add it. Thanks.
> > > >> > > > > > >
> > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > >> handling? For
> > > >> > > > > > example,
> > > >> > > > > > > > does it need to trigger the check if any group with the
> > > >> wildcard
> > > >> > > > > > > > subscription now has a new matching topic?
> > > >> > > > > > >
> > > >> > > > > > > Sure. When the new group coordinator takes over, it has
> > to:
> > > >> > > > > > > * Setup the session timeouts.
> > > >> > > > > > > * Trigger a new assignment if a client side assignor is
> > used.
> > > >> We
> > > >> > > > don't
> > > >> > > > > > > store the information about the member selected to run the
> > > >> assignment
> > > >> > > > > > > so we have to start a new one.
> > > >> > > > > > > * Update the topics metadata, verify the wildcard
> > > >> subscriptions, and
> > > >> > > > > > > trigger a rebalance if needed.
> > > >> > > > > > >
> > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> > > >> the epoch
> > > >> > > > > > field
> > > >> > > > > > > > reflects? For example, does the epoch in
> > > >> ConsumerGroupMetadataValue
> > > >> > > > > > reflect
> > > >> > > > > > > > the latest group epoch? What about the one in
> > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > >> > > > > > >
> > > >> > > > > > > Sure. I will clarify that but it is always the latest
> > group
> > > >> epoch.
> > > >> > > > > > > When the group state is updated, the group epoch is
> > bumped so
> > > >> we use
> > > >> > > > > > > that one for all the change records related to the update.
> > > >> > > > > > >
> > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > following
> > > >> > > > invariants
> > > >> > > > > > are
> > > >> > > > > > > > met: ... All members exists." It's possible for a member
> > > >> not to
> > > >> > > > get any
> > > >> > > > > > > > assigned partitions, right?
> > > >> > > > > > >
> > > >> > > > > > > That's right. Here I meant that the members provided by
> > the
> > > >> assignor
> > > >> > > > > > > in the assignment must exist in the group. The assignor
> > can
> > > >> not make
> > > >> > > > > > > up new member ids.
> > > >> > > > > > >
> > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > equals to
> > > >> 0":
> > > >> > > > When
> > > >> > > > > > would
> > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > >> > > > > > >
> > > >> > > > > > > A member is expected to abandon all its partitions and
> > > >> rejoins when
> > > >> > > > it
> > > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> > > >> group
> > > >> > > > > > > coordinator will have removed the member from the group.
> > The
> > > >> member
> > > >> > > > > > > can rejoin the group with the same member id but with 0 as
> > > >> epoch. Let
> > > >> > > > > > > me see if I can clarify this in the KIP.
> > > >> > > > > > >
> > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger a
> > > >> > > > > > reassignment
> > > >> > > > > > > > by either providing a non-zero reason or by updating the
> > > >> assignor
> > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > >> deprecation
> > > >> > > > of
> > > >> > > > > > > > Consumer#enforeRebalance.
> > > >> > > > > > >
> > > >> > > > > > > In this case, a new assignment is triggered by the client
> > side
> > > >> > > > > > > assignor. When constructing the HB, the consumer will
> > always
> > > >> consult
> > > >> > > > > > > the client side assignor and propagate the information to
> > the
> > > >> group
> > > >> > > > > > > coordinator. In other words, we don't expect users to call
> > > >> > > > > > > Consumer#enforceRebalance anymore.
> > > >> > > > > > >
> > > >> > > > > > > > 38. The reassignment examples are nice. But the section
> > > >> seems to
> > > >> > > > have
> > > >> > > > > > > > multiple typos.
> > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > immediately
> > > >> gets into
> > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > > >> into
> > > >> > > > > > epoch=3,
> > > >> > > > > > > > partitions=[foo-1] too early.
> > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > >> epoch=2,
> > > >> > > > > > > > partitions=[foo-0].
> > > >> > > > > > >
> > > >> > > > > > > Sorry for that! I will revise them.
> > > >> > > > > > >
> > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > upgrade
> > > >> from
> > > >> > > > any
> > > >> > > > > > old
> > > >> > > > > > > > version to new one?
> > > >> > > > > > >
> > > >> > > > > > > We will support upgrading from the consumer protocol
> > version
> > > >> 3,
> > > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> > the
> > > >> earliest
> > > >> > > > > > > version is unknown at the moment. This is explained in the
> > > >> migration
> > > >> > > > > > > plan chapter.
> > > >> > > > > > >
> > > >> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> > > >> based on
> > > >> > > > it
> > > >> > > > > > > next week.
> > > >> > > > > > >
> > > >> > > > > > > Best,
> > > >> > > > > > > David
> > > >> > > > > > >
> > > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > >> <ju...@confluent.io.invalid>
> > > >> > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > Hi, David,
> > > >> > > > > > > >
> > > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > KIP
> > > >> seem to
> > > >> > > > be
> > > >> > > > > > fewer
> > > >> > > > > > > > RPCs during rebalance and more efficient support of
> > > >> wildcard. A few
> > > >> > > > > > > > comments below.
> > > >> > > > > > > >
> > > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > support
> > > >> rolling
> > > >> > > > > > changing
> > > >> > > > > > > > of the partition assignor in the consumers?
> > > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > > >> required in
> > > >> > > > every
> > > >> > > > > > > > request or the scenarios when it needs to be filled? For
> > > >> example,
> > > >> > > > it's
> > > >> > > > > > not
> > > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > >> > > > > > > >
> > > >> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > > >> between the
> > > >> > > > > > client
> > > >> > > > > > > > and the broker is only considered during fetching, but
> > not
> > > >> during
> > > >> > > > > > assigning
> > > >> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > is
> > > >> made,
> > > >> > > > there
> > > >> > > > > > is
> > > >> > > > > > > > no opportunity for read affinity because no replicas of
> > > >> assigned
> > > >> > > > > > partitions
> > > >> > > > > > > > are close to the member. I am wondering if we should use
> > > >> this
> > > >> > > > > > opportunity
> > > >> > > > > > > > to address this by including rack in GroupMember.
> > > >> > > > > > > >
> > > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> > busy
> > > >> a group
> > > >> > > > > > > > coordinator is. By moving the event loop model, it seems
> > > >> that we
> > > >> > > > could
> > > >> > > > > > add
> > > >> > > > > > > > a metric that tracks the fraction of the time the event
> > > >> loop is
> > > >> > > > doing
> > > >> > > > > > the
> > > >> > > > > > > > actual work.
> > > >> > > > > > > >
> > > >> > > > > > > > 33. Could we add a section on coordinator failover
> > > >> handling? For
> > > >> > > > > > example,
> > > >> > > > > > > > does it need to trigger the check if any group with the
> > > >> wildcard
> > > >> > > > > > > > subscription now has a new matching topic?
> > > >> > > > > > > >
> > > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > >> > > > ConsumerGroupPartitionMetadataValue,
> > > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> > > >> the epoch
> > > >> > > > > > field
> > > >> > > > > > > > reflects? For example, does the epoch in
> > > >> ConsumerGroupMetadataValue
> > > >> > > > > > reflect
> > > >> > > > > > > > the latest group epoch? What about the one in
> > > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > >> > > > > > ConsumerGroupMemberMetadataValue?
> > > >> > > > > > > >
> > > >> > > > > > > > 35. "the group coordinator will ensure that the
> > following
> > > >> > > > invariants
> > > >> > > > > > are
> > > >> > > > > > > > met: ... All members exists." It's possible for a member
> > > >> not to
> > > >> > > > get any
> > > >> > > > > > > > assigned partitions, right?
> > > >> > > > > > > >
> > > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> > equals to
> > > >> 0":
> > > >> > > > When
> > > >> > > > > > would
> > > >> > > > > > > > a consumer rejoin and what member id would be used?
> > > >> > > > > > > >
> > > >> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger a
> > > >> > > > > > reassignment
> > > >> > > > > > > > by either providing a non-zero reason or by updating the
> > > >> assignor
> > > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > >> deprecation
> > > >> > > > of
> > > >> > > > > > > > Consumer#enforeRebalance.
> > > >> > > > > > > >
> > > >> > > > > > > > 38. The reassignment examples are nice. But the section
> > > >> seems to
> > > >> > > > have
> > > >> > > > > > > > multiple typos.
> > > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > immediately
> > > >> gets into
> > > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > > >> into
> > > >> > > > > > epoch=3,
> > > >> > > > > > > > partitions=[foo-1] too early.
> > > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > >> epoch=2,
> > > >> > > > > > > > partitions=[foo-0].
> > > >> > > > > > > >
> > > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > upgrade
> > > >> from
> > > >> > > > any
> > > >> > > > > > old
> > > >> > > > > > > > version to new one?
> > > >> > > > > > > >
> > > >> > > > > > > > Thanks,
> > > >> > > > > > > >
> > > >> > > > > > > > Jun
> > > >> > > > > > > >
> > > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > >> > > > > > <dj...@confluent.io.invalid>
> > > >> > > > > > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > > Hi all,
> > > >> > > > > > > > >
> > > >> > > > > > > > > The KIP states that we will re-implement the
> > coordinator
> > > >> in
> > > >> > > > Java. I
> > > >> > > > > > > > > discussed this offline with a few folks and folks are
> > > >> concerned
> > > >> > > > that
> > > >> > > > > > > > > we could introduce many regressions in the old
> > protocol
> > > >> if we do
> > > >> > > > so.
> > > >> > > > > > > > > Therefore, I am going to remove this statement from
> > the
> > > >> KIP. It
> > > >> > > > is an
> > > >> > > > > > > > > implementation detail after all so it does not have
> > to be
> > > >> > > > decided at
> > > >> > > > > > > > > this stage. We will likely start by trying to
> > refactor the
> > > >> > > > current
> > > >> > > > > > > > > implementation as a first step.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Cheers,
> > > >> > > > > > > > > David
> > > >> > > > > > > > >
> > > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > >> djacot@confluent.io
> > > >> > > > >
> > > >> > > > > > wrote:
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Hi Luke,
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > assigning,
> > > >> > > > > > reconciling,
> > > >> > > > > > > > > stable,
> > > >> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > > >> right?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > This sentence does not refer to those group states
> > but
> > > >> rather
> > > >> > > > to a
> > > >> > > > > > > > > > state machine replication (SMR). This refers to the
> > > >> entire
> > > >> > > > state of
> > > >> > > > > > > > > > group coordinator which is replicated via the log
> > > >> layer. I will
> > > >> > > > > > > > > > clarify this in the KIP.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > modelled
> > > >> as an
> > > >> > > > event
> > > >> > > > > > > > > loop"?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > The idea is to follow a model similar to the new
> > quorum
> > > >> > > > > > controller. We
> > > >> > > > > > > > > > will have N threads to process events. Each
> > > >> __consumer_offsets
> > > >> > > > > > > > > > partition is assigned to a unique thread and all the
> > > >> events
> > > >> > > > (e.g.
> > > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > >> thread. This
> > > >> > > > > > simplify
> > > >> > > > > > > > > > concurrency and will enable us to do simulation
> > testing
> > > >> for the
> > > >> > > > > > group
> > > >> > > > > > > > > > coordinator.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > >> *__consumer_offsets*
> > > >> > > > > > > > > partitions?
> > > >> > > > > > > > > > > Not a state machine "per consumer group" owned by
> > a
> > > >> group
> > > >> > > > > > coordinator?
> > > >> > > > > > > > > For
> > > >> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > > >> groups, and
> > > >> > > > > > both
> > > >> > > > > > > > > exist in
> > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > > >> for it,
> > > >> > > > or
> > > >> > > > > > 2?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > > >> number
> > > >> > > > of
> > > >> > > > > > threads
> > > >> > > > > > > > > used
> > > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> > the
> > > >> purpose
> > > >> > > > of
> > > >> > > > > > the
> > > >> > > > > > > > > threads
> > > >> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > > >> > > > > > > > > *__consumer_offsets*
> > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > should we
> > > >> need
> > > >> > > > > > > > > multi-threads
> > > >> > > > > > > > > > > here?
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > > >> > > > processing as
> > > >> > > > > > the
> > > >> > > > > > > > > > computation could be heavy.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > >> default
> > > >> > > > session
> > > >> > > > > > > > > timeout not
> > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > > >> min/max
> > > >> > > > > > session
> > > >> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > >> timeout to
> > > >> > > > detect
> > > >> > > > > > client
> > > >> > > > > > > > > > > failures when using the consumer group protocol.
> > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > > >> minimum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > > >> maximum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > This is indeed a mistake. The default session
> > timeout
> > > >> should
> > > >> > > > be 45s
> > > >> > > > > > > > > > (the current default).
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > > >> uniform],
> > > >> > > > > > which means
> > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > know
> > > >> why not
> > > >> > > > > > uniform
> > > >> > > > > > > > > one? I
> > > >> > > > > > > > > > > thought usually users will choose uniform assignor
> > > >> (former
> > > >> > > > sticky
> > > >> > > > > > > > > assinor)
> > > >> > > > > > > > > > > for better evenly distribution. Any other reason
> > we
> > > >> choose
> > > >> > > > range
> > > >> > > > > > > > > assignor
> > > >> > > > > > > > > > > as default?
> > > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > >> server side
> > > >> > > > > > assignors.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > The order on the server side has no influence
> > because
> > > >> the
> > > >> > > > client
> > > >> > > > > > must
> > > >> > > > > > > > > > chose the selector that he wants to use. There is no
> > > >> default
> > > >> > > > in the
> > > >> > > > > > > > > > current proposal. If the assignor is not specified
> > by
> > > >> the
> > > >> > > > client,
> > > >> > > > > > the
> > > >> > > > > > > > > > request is rejected. The default client value for
> > > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > that my
> > > >> > > > answers
> > > >> > > > > > help
> > > >> > > > > > > > > > to clarify things. I will update the KIP as well
> > based
> > > >> on your
> > > >> > > > > > > > > > feedback.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Cheers,
> > > >> > > > > > > > > > David
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > >> showuon@gmail.com>
> > > >> > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Hi David,
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Thanks for the update.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Some more questions:
> > > >> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > >> > > > > > > > > > > > The new group coordinator will have a state
> > machine
> > > >> per
> > > >> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> > > >> machine is
> > > >> > > > > > modelled
> > > >> > > > > > > > > as an
> > > >> > > > > > > > > > > event loop. Those state machines will be executed
> > in
> > > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > assigning,
> > > >> > > > > > reconciling,
> > > >> > > > > > > > > stable,
> > > >> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > > >> right?
> > > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > modelled
> > > >> as an
> > > >> > > > event
> > > >> > > > > > > > > loop"?
> > > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > >> *__consumer_offsets*
> > > >> > > > > > > > > partitions?
> > > >> > > > > > > > > > > Not a state machine "per consumer group" owned by
> > a
> > > >> group
> > > >> > > > > > coordinator?
> > > >> > > > > > > > > For
> > > >> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > > >> groups, and
> > > >> > > > > > both
> > > >> > > > > > > > > exist in
> > > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > > >> for it,
> > > >> > > > or
> > > >> > > > > > 2?
> > > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > > >> number
> > > >> > > > of
> > > >> > > > > > threads
> > > >> > > > > > > > > used
> > > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> > the
> > > >> purpose
> > > >> > > > of
> > > >> > > > > > the
> > > >> > > > > > > > > threads
> > > >> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > > >> > > > > > > > > *__consumer_offsets*
> > > >> > > > > > > > > > > partitions?), and no heavy computation, why
> > should we
> > > >> need
> > > >> > > > > > > > > multi-threads
> > > >> > > > > > > > > > > here?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 2. For the default value in the new configs:
> > > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > >> default
> > > >> > > > session
> > > >> > > > > > > > > timeout not
> > > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > > >> min/max
> > > >> > > > > > session
> > > >> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > >> timeout to
> > > >> > > > detect
> > > >> > > > > > client
> > > >> > > > > > > > > > > failures when using the consumer group protocol.
> > > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > > >> minimum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > > >> maximum
> > > >> > > > > > session
> > > >> > > > > > > > > timeout.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > > >> uniform],
> > > >> > > > > > which means
> > > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > know
> > > >> why not
> > > >> > > > > > uniform
> > > >> > > > > > > > > one? I
> > > >> > > > > > > > > > > thought usually users will choose uniform assignor
> > > >> (former
> > > >> > > > sticky
> > > >> > > > > > > > > assinor)
> > > >> > > > > > > > > > > for better evenly distribution. Any other reason
> > we
> > > >> choose
> > > >> > > > range
> > > >> > > > > > > > > assignor
> > > >> > > > > > > > > > > as default?
> > > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > >> server side
> > > >> > > > > > assignors.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Thank you.
> > > >> > > > > > > > > > > Luke
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > >> showuon@gmail.com
> > > >> > > > >
> > > >> > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > Hi Sagar,
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > >> integrating with
> > > >> > > > > > KIP-848,
> > > >> > > > > > > > > but I
> > > >> > > > > > > > > > > > think we should have a separate discussion
> > thread
> > > >> for the
> > > >> > > > Kafka
> > > >> > > > > > > > > Connect
> > > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> > > >> Rebalance
> > > >> > > > > > Protocol
> > > >> > > > > > > > > [1],
> > > >> > > > > > > > > > > > and let this discussion thread focus on consumer
> > > >> rebalance
> > > >> > > > > > protocol,
> > > >> > > > > > > > > WDYT?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > [1]
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > >
> > > >> > > > > >
> > > >> > > >
> > > >>
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > Thank you.
> > > >> > > > > > > > > > > > Luke
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > >> > > > > > sagarmeansocean@gmail.com>
> > > >> > > > > > > > > wrote:
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > Looks
> > > >> like
> > > >> > > > there's
> > > >> > > > > > > > > agreement on
> > > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> > revisit
> > > >> the doc
> > > >> > > > and
> > > >> > > > > > see
> > > >> > > > > > > > > what
> > > >> > > > > > > > > > > >> changes are to be made.
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > > > > >> Thanks!
> > > >> > > > > > > > > > > >> Sagar.
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > >> > > > > > > > > > > >> wrote:
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > > > > >> > Hi Sagar,
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > That's
> > > >> really
> > > >> > > > > > helpful. I
> > > >> > > > > > > > > > > >> > will take a look at it.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > and the
> > > >> > > > Consumer
> > > >> > > > > > could
> > > >> > > > > > > > > share
> > > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > > >> difference is
> > > >> > > > that
> > > >> > > > > > the
> > > >> > > > > > > > > Consumer
> > > >> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > > >> Connect
> > > >> > > > assigns
> > > >> > > > > > tasks
> > > >> > > > > > > > > to
> > > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> > > >> different
> > > >> > > > > > resource
> > > >> > > > > > > > > types
> > > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> > > >> dedicated
> > > >> > > > APIs
> > > >> > > > > > > > > would be
> > > >> > > > > > > > > > > >> > similar to the new ones but different on the
> > > >> > > > > > content/resources and
> > > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > > >> coordinator
> > > >> > > > side.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > not a
> > > >> fan of
> > > >> > > > > > > > > overcharging
> > > >> > > > > > > > > > > >> > APIs to serve different purposes. That being
> > > >> said, I am
> > > >> > > > not
> > > >> > > > > > > > > opposed to
> > > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > I think that we can continue to discuss it
> > here
> > > >> for now
> > > >> > > > in
> > > >> > > > > > order
> > > >> > > > > > > > > to
> > > >> > > > > > > > > > > >> > ensure that this KIP is compatible with what
> > we
> > > >> will do
> > > >> > > > for
> > > >> > > > > > > > > Connect in
> > > >> > > > > > > > > > > >> > the future.
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > Best,
> > > >> > > > > > > > > > > >> > David
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > >> > > > > > djacot@confluent.io>
> > > >> > > > > > > > > wrote:
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > Hi all,
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > I am back from vacation. I will go through
> > and
> > > >> address
> > > >> > > > > > your
> > > >> > > > > > > > > comments
> > > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > feedback.
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > Cheers,
> > > >> > > > > > > > > > > >> > > David
> > > >> > > > > > > > > > > >> > >
> > > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > Harris
> > > >> <
> > > >> > > > > > > > > gharris1727@gmail.com
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >> > wrote:
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > Hey All!
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > > >> > > > cooperative
> > > >> > > > > > > > > rebalancing
> > > >> > > > > > > > > > > >> > making it
> > > >> > > > > > > > > > > >> > > > down the stack!
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > I had a few questions:
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > > >> describes how
> > > >> > > > > > member
> > > >> > > > > > > > > epoch
> > > >> > > > > > > > > > > >> > should
> > > >> > > > > > > > > > > >> > > > advance in step with the group epoch and
> > > >> assignment
> > > >> > > > > > epoch
> > > >> > > > > > > > > values. I
> > > >> > > > > > > > > > > >> > think
> > > >> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> > > >> described
> > > >> > > > in
> > > >> > > > > > the
> > > >> > > > > > > > > KIP. When
> > > >> > > > > > > > > > > >> the
> > > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > >> partitions to a
> > > >> > > > > > worker,
> > > >> > > > > > > > > what
> > > >> > > > > > > > > > > >> member
> > > >> > > > > > > > > > > >> > > > epoch does each incremental assignment
> > use?
> > > >> Are
> > > >> > > > member
> > > >> > > > > > epochs
> > > >> > > > > > > > > > > >> re-used,
> > > >> > > > > > > > > > > >> > and
> > > >> > > > > > > > > > > >> > > > a single member epoch can correspond to
> > > >> multiple
> > > >> > > > > > different
> > > >> > > > > > > > > > > >> > (monotonically
> > > >> > > > > > > > > > > >> > > > larger) assignments?
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > opaque
> > > >> to the
> > > >> > > > group
> > > >> > > > > > > > > > > >> coordinator? If
> > > >> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > > >> > > > implementations
> > > >> > > > > > > > > interact
> > > >> > > > > > > > > > > >> with
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> > meaning
> > > >> agreed
> > > >> > > > > > upon? If
> > > >> > > > > > > > > so, what
> > > >> > > > > > > > > > > >> > is the
> > > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > > >> including
> > > >> > > > such
> > > >> > > > > > > > > functionality
> > > >> > > > > > > > > > > >> > in the
> > > >> > > > > > > > > > > >> > > > opaque metadata?
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> > > >> "Thanks to
> > > >> > > > > > this, the
> > > >> > > > > > > > > input
> > > >> > > > > > > > > > > >> of
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > client side assignor is entirely driven
> > by
> > > >> the group
> > > >> > > > > > > > > coordinator.
> > > >> > > > > > > > > > > >> The
> > > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > >> maintaining
> > > >> > > > any
> > > >> > > > > > state
> > > >> > > > > > > > > besides
> > > >> > > > > > > > > > > >> its
> > > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > that the
> > > >> > > > > > client-side
> > > >> > > > > > > > > assignor
> > > >> > > > > > > > > > > >> MAY
> > > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> > > >> (such as
> > > >> > > > > > partition
> > > >> > > > > > > > > > > >> > throughput,
> > > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc),
> > or
> > > >> that
> > > >> > > > > > additional
> > > >> > > > > > > > > > > >> > non-Metadata
> > > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> > > >> > > > > > > > > > > >> > > > for
> > > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > >> > > > > > > > > > > >> > > > and
> > > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > >> > > > > > that
> > > >> > > > > > > > > seem to
> > > >> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> > > >> these two
> > > >> > > > > > > > > implementations
> > > >> > > > > > > > > > > >> to
> > > >> > > > > > > > > > > >> > be
> > > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> > feature
> > > >> parity
> > > >> > > > of
> > > >> > > > > > > > > server-side
> > > >> > > > > > > > > > > >> and
> > > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > >> facilitate
> > > >> > > > > > operational
> > > >> > > > > > > > > > > >> > flexibility in
> > > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > >> server-side
> > > >> > > > > > assignor
> > > >> > > > > > > > > has some
> > > >> > > > > > > > > > > >> > poor
> > > >> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> > > >> patched
> > > >> > > > > > assignor to
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > >> > client
> > > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > >> client-side
> > > >> > > > > > assignor
> > > >> > > > > > > > > may be
> > > >> > > > > > > > > > > >> > faster
> > > >> > > > > > > > > > > >> > > > and less risky than patching all of the
> > > >> brokers.
> > > >> > > > With
> > > >> > > > > > the
> > > >> > > > > > > > > currently
> > > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > >> > > > reimplementation
> > > >> > > > > > would
> > > >> > > > > > > > > have
> > > >> > > > > > > > > > > >> to be
> > > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > diverged
> > > >> > > > > > significantly,
> > > >> > > > > > > > > then it
> > > >> > > > > > > > > > > >> is
> > > >> > > > > > > > > > > >> > > > possible that a reimplementation would
> > not be
> > > >> > > > possible.
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > --
> > > >> > > > > > > > > > > >> > > > Greg Harris
> > > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > >> > > > > > > > > > > >> > wrote:
> > > >> > > > > > > > > > > >> > > >
> > > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > I created a confluence page to discuss
> > how
> > > >> Connect
> > > >> > > > > > would
> > > >> > > > > > > > > need to
> > > >> > > > > > > > > > > >> > change
> > > >> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> > > >> Here's the
> > > >> > > > page:
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> >
> > > >> > > > > > > > > > > >>
> > > >> > > > > > > > >
> > > >> > > > > >
> > > >> > > >
> > > >>
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > tried
> > > >> to keep
> > > >> > > > a
> > > >> > > > > > format
> > > >> > > > > > > > > > > >> similar to
> > > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> > Also,
> > > >> do you
> > > >> > > > > > think this
> > > >> > > > > > > > > > > >> should
> > > >> > > > > > > > > > > >> > be
> > > >> > > > > > > > > > > >> > > > > moved to a separate discussion thread
> > or
> > > >> is this
> > > >> > > > one
> > > >> > > > > > fine?
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > Thanks!
> > > >> > > > > > > > > > > >> > > > > Sagar.
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > >> > > > > > > > > sagarmeansocean@gmail.com>
> > > >> > > > > > > > > > > >> > wrote:
> > > >> > > > > > > > > > > >> > > > >
> > > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > Kafka
> > > >> Streams.
> > > >> > > > > > Sure, I
> > > >> > > > > > > > > would do
> > > >> > > > > > > > > > > >> > the
> > > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > >> document.
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > > Thanks!
> > > >> > > > > > > > > > > >> > > > > > Sagar.
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > Guozhang
> > > >> Wang <
> > > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > > >> > > > > > > > > > > >> > > > > wrote:
> > > >> > > > > > > > > > > >> > > > > >
> > > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > >> > > > > > > > > > > >> > > > > >>
> > > >> > > > > > > > > > > >> > > > > >> It would be great if you could come
> > > >> back with
> > > >> > > > some
> > > >> > > > > > > > > analysis on
> > > >> > > > > > > > > > > >> > how to
> > > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> > integration
> > > >> with
> > > >> > > > the new
> > > >> > > > > > > > > protocol;
> > > >> > > > > > > > > > > >> so
> > > >> > > > > > > > > > > >> > far
> > > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > "protocol
> > > >> type"
> > > >> > > > we
> > > >> > > > > > did not

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hi David, thanks for all the detailed explanations. I think they all make
sense. Just want to have a couple follow-ups here:

> I don't really see the benefits here because server side assignors
don't have metadata at all. They only assign topic-partitions. They
are not supposed to generate metadata nor to receive metadata from the
members.

Just to clarify I was asking about the `version` of the assignor (i.e. up
to what version that the client would support), and I do agree we would not
need metadata. What I have in mind is that, for some specific built-in
broker-assignors, e.g. rack-aware assignors, if it's possible that in a
newer version we would have a hierarchical rack ID string format, like
"tier1-tier2" etc, but if some client has not upgraded their rack ID
would still be in old format. In this case, the broker then needs to choose
the old versioned assignor. I'm probably making something up here for rack
aware assignors, but I'm wondering if in general such an "auto-downgrade"
behavior would be needed still for broker-side assignor, and if yes would
"version" still be useful.

> Yeah, that's right. Within a rebalance, `onAssignment` is called once
when the member transitions to a new epoch. This one contains the full
metadata provided by the client side assignor. Then, `onAssignment`
can be called max N times where N is the number of partitions pending
revocation by other members. Let me try to clarify this in the KIP.

Okay, my understanding is that the calling ordering of these callbacks
would be like the following:

----------------------------------------
onPartitionsRevoked();   // just once, since we do not really need
to revoke incrementally.

onAssignment();    // the first call, with epoch incremented
onPartitionsAssigned();   // paired with the onAssignment

onAssignment();              // the first onAssignment would bump up the
epoch, and the metadata reflected.
onPartitionsAssigned();   // each time we get an additional assignment, we
call onAssignment and then paired with an onPartitionsAssigned
...
onAssignment();
onPartitionsAssigned();   // on each of the onAssignment calls, the encoded
metadata would not change, only the incrementally added partitions be
reflected

Is that the case?

I'm wondering if we would still call onAssignment just once, that encodes
all the assignment for this rebalance, including all the partitions that
should be assigned to the member but not yet assigned since they have not
been revoked by others. In that case the call ordering would be:

----------------------------------------
onPartitionsRevoked();   // just once
onAssignment();    // just once, with epoch incremented, and metadata
encoded changed, the "assignment" field also reflect the final target
assignment
onPartitionsAssigned();   // multiple times, which represent incrementally
added partitions
...
onPartitionsAssigned();

The motivation from this is that, most users would only implement the
rebalance callback listeners and hence we'd definitely need to make sure
the semantics of that does not change much, and the time
`onPartitionsAssigned` indicate the time when the partitions are actually
assigned to it; while for assignors, the `onAssignment` is used to indicate
what decision is made regarding for this member, i.e. when the partitions
are decided to be given to it, but not necessarily meant that it has been
given, since that time should be determined by the time of
`onPartitionsAssigned`. The benefits though, would be that assignor
implementers would not need to reason which `onAssignment` would be the
last one for this epoch.


Guozhang




Guozhang

On Thu, Sep 22, 2022 at 2:20 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Guozhang,
>
> > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> field,
> > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > rationale seems to be assuming that we should always be able to do the
> > assignment at the broker-side assignor without errors. Personally I think
> > it's still potentially beneficial to add the Error field even for
> > broker-side assignors, e.g. for some edge cases where some subscribed
> > topics are not recognized with the current broker's metadata. What do you
> > think?
>
> Yeah, that seems reasonable. However, I wonder if it would be better
> to use an exception on the server side. This is what we usually do for
> server side plugins. On the client side, we use a field because the
> errors are not defined in advance.
>
> Your comment also makes me think about what we should do when the
> server side assignor fails. I suppose that we have to keep the current
> assignment until a new event occurs. For instance, in your example,
> the coordinator would have to trigger a rebalance when unrecognized
> topics are available. This would be part of the metadata monitoring.
>
> > 2) The client-side "GroupMember" has three additional fields
> > reason/version/metadata compared with the broker-side GroupMember. I
> agree
> > that broker-side assignor would not need reason/metadata since they are
> > blackbox strings/bytes to the assignor, but what about version? E.g. is
> it
> > possible that we evolve our broker-side built-in assignor but the old
> > versioned clients would not be able to work with the new version, in
> which
> > case we need to let the broker being aware of this and upgrade its
> behavior
> > to cooperate with the clients?
>
> I don't really see the benefits here because server side assignors
> don't have metadata at all. They only assign topic-partitions. They
> are not supposed to generate metadata nor to receive metadata from the
> members.
>
> > 3) Also related to 2) above, for the client-side "GroupMember", instead
> of
> > including these three fields, what about just adding the "Metadata" field
> > class which has these three fields? Also, there are two "Metadata"
> > currently in the APIs, the first is a class that encodes
> > reason/version/metadata, and the second is just the encoded metadata
> bytes.
> > I'm wondering what about just naming the first as memberMetadata, which
> > then has a bytebuffer field Metadata, or instead naming the second
> > bytebuffer field as metadataBytes?
>
> That's a good point. Let me try to rationalize this interface based on
> your suggestions.
>
> Best,
> David
>
> On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello David,
> >
> > Just had a few more nit questions about the public APIs:
> >
> > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> field,
> > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > rationale seems to be assuming that we should always be able to do the
> > assignment at the broker-side assignor without errors. Personally I think
> > it's still potentially beneficial to add the Error field even for
> > broker-side assignors, e.g. for some edge cases where some subscribed
> > topics are not recognized with the current broker's metadata. What do you
> > think?
> >
> > 2) The client-side "GroupMember" has three additional fields
> > reason/version/metadata compared with the broker-side GroupMember. I
> agree
> > that broker-side assignor would not need reason/metadata since they are
> > blackbox strings/bytes to the assignor, but what about version? E.g. is
> it
> > possible that we evolve our broker-side built-in assignor but the old
> > versioned clients would not be able to work with the new version, in
> which
> > case we need to let the broker being aware of this and upgrade its
> behavior
> > to cooperate with the clients?
> >
> > 3) Also related to 2) above, for the client-side "GroupMember", instead
> of
> > including these three fields, what about just adding the "Metadata" field
> > class which has these three fields? Also, there are two "Metadata"
> > currently in the APIs, the first is a class that encodes
> > reason/version/metadata, and the second is just the encoded metadata
> bytes.
> > I'm wondering what about just naming the first as memberMetadata, which
> > then has a bytebuffer field Metadata, or instead naming the second
> > bytebuffer field as metadataBytes?
> >
> > Guozhang
> >
> >
> >
> >
> >
> >
> >
> > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> >
> > > Hello David,
> > >
> > > Thanks for bringing this question up. I think the main benefits as you
> > > listed is 2) above if it stays; just to clarify, we would only be able
> to
> > > save one round trip if the rebalance is still triggered by the broker;
> if
> > > the rebalance is triggered by the client then the num.round.trips are
> the
> > > same:
> > >
> > > 1) With GroupPrepareAssignment:
> > >
> > > T0: client decides to do a new assignment, suppose it has already sent
> a
> > > HB and hence has to wait for it to return first since only one request
> /
> > > response can be inflight with the coordinator's socket.
> > > T1: client receives the HB response, and then sends the
> > > GroupPrepareAssignment request.
> > > T2: the GroupPrepareAssignment response is returned.
> > > T3: it calculates the new assignment, and sends a
> GroupInstallAssignment
> > > request.
> > >
> > > In total, two round trips.
> > >
> > > 2) Without GroupPrepareAssignment:
> > >
> > > T0: client decides to do a new assignment, suppose it has already sent
> a
> > > HB and hence has to wait for it to return first since only one request
> /
> > > response can be inflight with the coordinator's socket.
> > > T1: client receives the HB response, and then sends the new HB request
> > > with the flag indicating a new rebalance needed..
> > > T2: the HB response with the optional member metadata map is returned.
> > > T3: it calculates the new assignment, and sends a
> GroupInstallAssignment
> > > request.
> > >
> > > In total, two round trips as well.
> > >
> > > -----------------------------
> > >
> > > So to complete the full picture here, we'd need to modify both HB
> request
> > > and response so that the client can also indicate a new rebalance via
> the
> > > HB request as well, right?
> > >
> > > Assuming all above is true, I think it's okay to merge the
> > > GroupPrepareAssignment into HB given that we can make the additional
> fields
> > > encoding the full member (subscription) metadata and topic metadata as
> > > optional fields.
> > >
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> <dj...@confluent.io.invalid>
> > > wrote:
> > >
> > >> Hi all,
> > >>
> > >> During an offline conversation, someone asked why we need the
> > >> ConsumerGroupPrepareAssignment API and suggested that we could instead
> > >> provide the group state in the heartbeat response. This has a few
> > >> advantages: 1) it does not require using a special error code to
> > >> signal that a new assignment is required as the signal would be the
> > >> provided group state; 2) it removes one round trip when a client side
> > >> assignor is used. The downside is that it makes the heartbeat
> > >> response's definition quite large. I recall that I went with the
> > >> current approach due to this.
> > >>
> > >> Providing the group state in the heartbeat response is appealing. What
> > >> do you guys think?
> > >>
> > >> Best,
> > >> David
> > >>
> > >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> wrote:
> > >> >
> > >> > Hi Guozhang,
> > >> >
> > >> > 1. I have added a reference to the relevant chapter instead of
> > >> > repeating the whole thing. Does that work for you?
> > >> >
> > >> > 2. The "Rebalance Triggers" section you are referring to is about
> when
> > >> > a rebalance should be triggered for the non-upgraded members using
> the
> > >> > old protocol. The section mentions that a rebalance must be
> triggered
> > >> > when a new assignment is installed. This implies that the group
> epoch
> > >> > was updated either by a native member or a non-upgraded member. For
> > >> > the latter, the JoinGroup request would be the trigger. I have
> added a
> > >> > reference to the relevant chapter in the "JoinGroup Handling"
> section
> > >> > as well. Does that make sense?
> > >> >
> > >> > Thanks,
> > >> > David
> > >> >
> > >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> > >> wrote:
> > >> > >
> > >> > > Hello David,
> > >> > >
> > >> > > Alright I think that's sufficient. Just to make that clear in the
> doc,
> > >> > > could we update:
> > >> > >
> > >> > > 1) the heartbeat request handling section, stating when
> coordinator
> > >> will
> > >> > > trigger rebalance based on the HB's member metadata / reason?
> > >> > > 2) the "Rebalance Triggers" section to include what we described
> in
> > >> "Group
> > >> > > Epoch - Trigger a rebalance" section as well?
> > >> > >
> > >> > >
> > >> > > Guozhang
> > >> > >
> > >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > >> <dj...@confluent.io.invalid>
> > >> > > wrote:
> > >> > >
> > >> > > > Hi Guozhang,
> > >> > > >
> > >> > > > I thought that the assignor will always be consulted when the
> next
> > >> > > > heartbeat request is constructed. In other words,
> > >> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
> > >> This
> > >> > > > gives the opportunity for the assignor to enforce a rebalance by
> > >> > > > setting the reason to a non-zero value or by changing the
> bytes. Do
> > >> > > > you think that this is not sufficient? Are you concerned by the
> > >> delay?
> > >> > > >
> > >> > > > Best,
> > >> > > > David
> > >> > > >
> > >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> wangguoz@gmail.com>
> > >> wrote:
> > >> > > > >
> > >> > > > > Hello David,
> > >> > > > >
> > >> > > > > One of Jun's comments make me thinking:
> > >> > > > >
> > >> > > > > ```
> > >> > > > > In this case, a new assignment is triggered by the client side
> > >> > > > > assignor. When constructing the HB, the consumer will always
> > >> consult
> > >> > > > > the client side assignor and propagate the information to the
> > >> group
> > >> > > > > coordinator. In other words, we don't expect users to call
> > >> > > > > Consumer#enforceRebalance anymore.
> > >> > > > > ```
> > >> > > > >
> > >> > > > > As I looked at the current PartitionAssignor's interface, we
> > >> actually do
> > >> > > > > not have a way yet to instruct how to construct the next HB
> > >> request, e.g.
> > >> > > > > when the assignor wants to enforce a new rebalance with a new
> > >> assignment,
> > >> > > > > we'd need some customizable APIs inside the PartitionAssignor
> to
> > >> indicate
> > >> > > > > the next HB telling broker about so. WDYT about adding such an
> > >> API on the
> > >> > > > > PartitionAssignor?
> > >> > > > >
> > >> > > > >
> > >> > > > > Guozhang
> > >> > > > >
> > >> > > > >
> > >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > >> <dj...@confluent.io.invalid>
> > >> > > > > wrote:
> > >> > > > >
> > >> > > > > > Hi Jun,
> > >> > > > > >
> > >> > > > > > I have updated the KIP to include your feedback. I have also
> > >> tried to
> > >> > > > > > clarify the parts which were not cleared.
> > >> > > > > >
> > >> > > > > > Best,
> > >> > > > > > David
> > >> > > > > >
> > >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> djacot@confluent.io
> > >> >
> > >> > > > wrote:
> > >> > > > > > >
> > >> > > > > > > Hi Jun,
> > >> > > > > > >
> > >> > > > > > > Thanks for your feedback. Let me start by answering your
> > >> questions
> > >> > > > > > > inline and I will update the KIP next week.
> > >> > > > > > >
> > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> KIP
> > >> seem to
> > >> > > > be
> > >> > > > > > fewer
> > >> > > > > > > > RPCs during rebalance and more efficient support of
> > >> wildcard. A few
> > >> > > > > > > > comments below.
> > >> > > > > > >
> > >> > > > > > > I would also add that the KIP removes the global sync
> barrier
> > >> in the
> > >> > > > > > > protocol which is essential to improve group stability and
> > >> > > > > > > scalability, and the KIP also simplifies the client by
> moving
> > >> most of
> > >> > > > > > > the logic to the server side.
> > >> > > > > > >
> > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> support
> > >> rolling
> > >> > > > > > changing
> > >> > > > > > > > of the partition assignor in the consumers?
> > >> > > > > > >
> > >> > > > > > > Definitely. The group coordinator will use the assignor
> used
> > >> by a
> > >> > > > > > > majority of the members. This allows the group to move
> from
> > >> one
> > >> > > > > > > assignor to another by a roll. This is explained in the
> > >> Assignor
> > >> > > > > > > Selection chapter.
> > >> > > > > > >
> > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > >> required in
> > >> > > > every
> > >> > > > > > > > request or the scenarios when it needs to be filled? For
> > >> example,
> > >> > > > it's
> > >> > > > > > not
> > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > >> > > > > > >
> > >> > > > > > > The client is expected to set those fields in case of a
> > >> connection
> > >> > > > > > > issue (e.g. timeout) or when the fields have changed since
> > >> the last
> > >> > > > > > > HB. The server populates those fields as long as the
> member
> > >> is not
> > >> > > > > > > fully reconciled - the member should acknowledge that it
> has
> > >> the
> > >> > > > > > > expected epoch and assignment. I will clarify this in the
> KIP.
> > >> > > > > > >
> > >> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > >> between the
> > >> > > > > > client
> > >> > > > > > > > and the broker is only considered during fetching, but
> not
> > >> during
> > >> > > > > > assigning
> > >> > > > > > > > partitions to consumers. Sometimes, once the assignment
> is
> > >> made,
> > >> > > > there
> > >> > > > > > is
> > >> > > > > > > > no opportunity for read affinity because no replicas of
> > >> assigned
> > >> > > > > > partitions
> > >> > > > > > > > are close to the member. I am wondering if we should use
> > >> this
> > >> > > > > > opportunity
> > >> > > > > > > > to address this by including rack in GroupMember.
> > >> > > > > > >
> > >> > > > > > > That's an interesting idea. I don't see any issue with
> adding
> > >> the
> > >> > > > rack
> > >> > > > > > > to the members. I will do so.
> > >> > > > > > >
> > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> busy
> > >> a group
> > >> > > > > > > > coordinator is. By moving the event loop model, it seems
> > >> that we
> > >> > > > could
> > >> > > > > > add
> > >> > > > > > > > a metric that tracks the fraction of the time the event
> > >> loop is
> > >> > > > doing
> > >> > > > > > the
> > >> > > > > > > > actual work.
> > >> > > > > > >
> > >> > > > > > > That's a great idea. I will add it. Thanks.
> > >> > > > > > >
> > >> > > > > > > > 33. Could we add a section on coordinator failover
> > >> handling? For
> > >> > > > > > example,
> > >> > > > > > > > does it need to trigger the check if any group with the
> > >> wildcard
> > >> > > > > > > > subscription now has a new matching topic?
> > >> > > > > > >
> > >> > > > > > > Sure. When the new group coordinator takes over, it has
> to:
> > >> > > > > > > * Setup the session timeouts.
> > >> > > > > > > * Trigger a new assignment if a client side assignor is
> used.
> > >> We
> > >> > > > don't
> > >> > > > > > > store the information about the member selected to run the
> > >> assignment
> > >> > > > > > > so we have to start a new one.
> > >> > > > > > > * Update the topics metadata, verify the wildcard
> > >> subscriptions, and
> > >> > > > > > > trigger a rebalance if needed.
> > >> > > > > > >
> > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > >> > > > ConsumerGroupPartitionMetadataValue,
> > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> > >> the epoch
> > >> > > > > > field
> > >> > > > > > > > reflects? For example, does the epoch in
> > >> ConsumerGroupMetadataValue
> > >> > > > > > reflect
> > >> > > > > > > > the latest group epoch? What about the one in
> > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > >> > > > > > ConsumerGroupMemberMetadataValue?
> > >> > > > > > >
> > >> > > > > > > Sure. I will clarify that but it is always the latest
> group
> > >> epoch.
> > >> > > > > > > When the group state is updated, the group epoch is
> bumped so
> > >> we use
> > >> > > > > > > that one for all the change records related to the update.
> > >> > > > > > >
> > >> > > > > > > > 35. "the group coordinator will ensure that the
> following
> > >> > > > invariants
> > >> > > > > > are
> > >> > > > > > > > met: ... All members exists." It's possible for a member
> > >> not to
> > >> > > > get any
> > >> > > > > > > > assigned partitions, right?
> > >> > > > > > >
> > >> > > > > > > That's right. Here I meant that the members provided by
> the
> > >> assignor
> > >> > > > > > > in the assignment must exist in the group. The assignor
> can
> > >> not make
> > >> > > > > > > up new member ids.
> > >> > > > > > >
> > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> equals to
> > >> 0":
> > >> > > > When
> > >> > > > > > would
> > >> > > > > > > > a consumer rejoin and what member id would be used?
> > >> > > > > > >
> > >> > > > > > > A member is expected to abandon all its partitions and
> > >> rejoins when
> > >> > > > it
> > >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> > >> group
> > >> > > > > > > coordinator will have removed the member from the group.
> The
> > >> member
> > >> > > > > > > can rejoin the group with the same member id but with 0 as
> > >> epoch. Let
> > >> > > > > > > me see if I can clarify this in the KIP.
> > >> > > > > > >
> > >> > > > > > > > 37. "Instead, power users will have the ability to
> trigger a
> > >> > > > > > reassignment
> > >> > > > > > > > by either providing a non-zero reason or by updating the
> > >> assignor
> > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > >> deprecation
> > >> > > > of
> > >> > > > > > > > Consumer#enforeRebalance.
> > >> > > > > > >
> > >> > > > > > > In this case, a new assignment is triggered by the client
> side
> > >> > > > > > > assignor. When constructing the HB, the consumer will
> always
> > >> consult
> > >> > > > > > > the client side assignor and propagate the information to
> the
> > >> group
> > >> > > > > > > coordinator. In other words, we don't expect users to call
> > >> > > > > > > Consumer#enforceRebalance anymore.
> > >> > > > > > >
> > >> > > > > > > > 38. The reassignment examples are nice. But the section
> > >> seems to
> > >> > > > have
> > >> > > > > > > > multiple typos.
> > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> immediately
> > >> gets into
> > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> get
> > >> into
> > >> > > > > > epoch=3,
> > >> > > > > > > > partitions=[foo-1] too early.
> > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > >> epoch=2,
> > >> > > > > > > > partitions=[foo-0].
> > >> > > > > > >
> > >> > > > > > > Sorry for that! I will revise them.
> > >> > > > > > >
> > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> upgrade
> > >> from
> > >> > > > any
> > >> > > > > > old
> > >> > > > > > > > version to new one?
> > >> > > > > > >
> > >> > > > > > > We will support upgrading from the consumer protocol
> version
> > >> 3,
> > >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> the
> > >> earliest
> > >> > > > > > > version is unknown at the moment. This is explained in the
> > >> migration
> > >> > > > > > > plan chapter.
> > >> > > > > > >
> > >> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> > >> based on
> > >> > > > it
> > >> > > > > > > next week.
> > >> > > > > > >
> > >> > > > > > > Best,
> > >> > > > > > > David
> > >> > > > > > >
> > >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > >> <ju...@confluent.io.invalid>
> > >> > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > Hi, David,
> > >> > > > > > > >
> > >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> KIP
> > >> seem to
> > >> > > > be
> > >> > > > > > fewer
> > >> > > > > > > > RPCs during rebalance and more efficient support of
> > >> wildcard. A few
> > >> > > > > > > > comments below.
> > >> > > > > > > >
> > >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> support
> > >> rolling
> > >> > > > > > changing
> > >> > > > > > > > of the partition assignor in the consumers?
> > >> > > > > > > > 30.2 For each field, could you explain whether it's
> > >> required in
> > >> > > > every
> > >> > > > > > > > request or the scenarios when it needs to be filled? For
> > >> example,
> > >> > > > it's
> > >> > > > > > not
> > >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > >> > > > > > > >
> > >> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > >> between the
> > >> > > > > > client
> > >> > > > > > > > and the broker is only considered during fetching, but
> not
> > >> during
> > >> > > > > > assigning
> > >> > > > > > > > partitions to consumers. Sometimes, once the assignment
> is
> > >> made,
> > >> > > > there
> > >> > > > > > is
> > >> > > > > > > > no opportunity for read affinity because no replicas of
> > >> assigned
> > >> > > > > > partitions
> > >> > > > > > > > are close to the member. I am wondering if we should use
> > >> this
> > >> > > > > > opportunity
> > >> > > > > > > > to address this by including rack in GroupMember.
> > >> > > > > > > >
> > >> > > > > > > > 32. On the metric side, often, it's useful to know how
> busy
> > >> a group
> > >> > > > > > > > coordinator is. By moving the event loop model, it seems
> > >> that we
> > >> > > > could
> > >> > > > > > add
> > >> > > > > > > > a metric that tracks the fraction of the time the event
> > >> loop is
> > >> > > > doing
> > >> > > > > > the
> > >> > > > > > > > actual work.
> > >> > > > > > > >
> > >> > > > > > > > 33. Could we add a section on coordinator failover
> > >> handling? For
> > >> > > > > > example,
> > >> > > > > > > > does it need to trigger the check if any group with the
> > >> wildcard
> > >> > > > > > > > subscription now has a new matching topic?
> > >> > > > > > > >
> > >> > > > > > > > 34. ConsumerGroupMetadataValue,
> > >> > > > ConsumerGroupPartitionMetadataValue,
> > >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> > >> the epoch
> > >> > > > > > field
> > >> > > > > > > > reflects? For example, does the epoch in
> > >> ConsumerGroupMetadataValue
> > >> > > > > > reflect
> > >> > > > > > > > the latest group epoch? What about the one in
> > >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > >> > > > > > ConsumerGroupMemberMetadataValue?
> > >> > > > > > > >
> > >> > > > > > > > 35. "the group coordinator will ensure that the
> following
> > >> > > > invariants
> > >> > > > > > are
> > >> > > > > > > > met: ... All members exists." It's possible for a member
> > >> not to
> > >> > > > get any
> > >> > > > > > > > assigned partitions, right?
> > >> > > > > > > >
> > >> > > > > > > > 36. "He can rejoins the group with a member epoch
> equals to
> > >> 0":
> > >> > > > When
> > >> > > > > > would
> > >> > > > > > > > a consumer rejoin and what member id would be used?
> > >> > > > > > > >
> > >> > > > > > > > 37. "Instead, power users will have the ability to
> trigger a
> > >> > > > > > reassignment
> > >> > > > > > > > by either providing a non-zero reason or by updating the
> > >> assignor
> > >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > >> deprecation
> > >> > > > of
> > >> > > > > > > > Consumer#enforeRebalance.
> > >> > > > > > > >
> > >> > > > > > > > 38. The reassignment examples are nice. But the section
> > >> seems to
> > >> > > > have
> > >> > > > > > > > multiple typos.
> > >> > > > > > > > 38.1 When the group transitions to epoch 2, B
> immediately
> > >> gets into
> > >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> get
> > >> into
> > >> > > > > > epoch=3,
> > >> > > > > > > > partitions=[foo-1] too early.
> > >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > >> epoch=2,
> > >> > > > > > > > partitions=[foo-0].
> > >> > > > > > > >
> > >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> upgrade
> > >> from
> > >> > > > any
> > >> > > > > > old
> > >> > > > > > > > version to new one?
> > >> > > > > > > >
> > >> > > > > > > > Thanks,
> > >> > > > > > > >
> > >> > > > > > > > Jun
> > >> > > > > > > >
> > >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > >> > > > > > <dj...@confluent.io.invalid>
> > >> > > > > > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > > Hi all,
> > >> > > > > > > > >
> > >> > > > > > > > > The KIP states that we will re-implement the
> coordinator
> > >> in
> > >> > > > Java. I
> > >> > > > > > > > > discussed this offline with a few folks and folks are
> > >> concerned
> > >> > > > that
> > >> > > > > > > > > we could introduce many regressions in the old
> protocol
> > >> if we do
> > >> > > > so.
> > >> > > > > > > > > Therefore, I am going to remove this statement from
> the
> > >> KIP. It
> > >> > > > is an
> > >> > > > > > > > > implementation detail after all so it does not have
> to be
> > >> > > > decided at
> > >> > > > > > > > > this stage. We will likely start by trying to
> refactor the
> > >> > > > current
> > >> > > > > > > > > implementation as a first step.
> > >> > > > > > > > >
> > >> > > > > > > > > Cheers,
> > >> > > > > > > > > David
> > >> > > > > > > > >
> > >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > >> djacot@confluent.io
> > >> > > > >
> > >> > > > > > wrote:
> > >> > > > > > > > > >
> > >> > > > > > > > > > Hi Luke,
> > >> > > > > > > > > >
> > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> assigning,
> > >> > > > > > reconciling,
> > >> > > > > > > > > stable,
> > >> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > >> right?
> > >> > > > > > > > > >
> > >> > > > > > > > > > This sentence does not refer to those group states
> but
> > >> rather
> > >> > > > to a
> > >> > > > > > > > > > state machine replication (SMR). This refers to the
> > >> entire
> > >> > > > state of
> > >> > > > > > > > > > group coordinator which is replicated via the log
> > >> layer. I will
> > >> > > > > > > > > > clarify this in the KIP.
> > >> > > > > > > > > >
> > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> modelled
> > >> as an
> > >> > > > event
> > >> > > > > > > > > loop"?
> > >> > > > > > > > > >
> > >> > > > > > > > > > The idea is to follow a model similar to the new
> quorum
> > >> > > > > > controller. We
> > >> > > > > > > > > > will have N threads to process events. Each
> > >> __consumer_offsets
> > >> > > > > > > > > > partition is assigned to a unique thread and all the
> > >> events
> > >> > > > (e.g.
> > >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > >> thread. This
> > >> > > > > > simplify
> > >> > > > > > > > > > concurrency and will enable us to do simulation
> testing
> > >> for the
> > >> > > > > > group
> > >> > > > > > > > > > coordinator.
> > >> > > > > > > > > >
> > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > >> *__consumer_offsets*
> > >> > > > > > > > > partitions?
> > >> > > > > > > > > > > Not a state machine "per consumer group" owned by
> a
> > >> group
> > >> > > > > > coordinator?
> > >> > > > > > > > > For
> > >> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > >> groups, and
> > >> > > > > > both
> > >> > > > > > > > > exist in
> > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> machine
> > >> for it,
> > >> > > > or
> > >> > > > > > 2?
> > >> > > > > > > > > >
> > >> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > >> > > > > > > > > >
> > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> the
> > >> number
> > >> > > > of
> > >> > > > > > threads
> > >> > > > > > > > > used
> > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> the
> > >> purpose
> > >> > > > of
> > >> > > > > > the
> > >> > > > > > > > > threads
> > >> > > > > > > > > > > is only to keep the state of each consumer group
> (or
> > >> > > > > > > > > *__consumer_offsets*
> > >> > > > > > > > > > > partitions?), and no heavy computation, why
> should we
> > >> need
> > >> > > > > > > > > multi-threads
> > >> > > > > > > > > > > here?
> > >> > > > > > > > > >
> > >> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > >> > > > processing as
> > >> > > > > > the
> > >> > > > > > > > > > computation could be heavy.
> > >> > > > > > > > > >
> > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > >> default
> > >> > > > session
> > >> > > > > > > > > timeout not
> > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> the
> > >> min/max
> > >> > > > > > session
> > >> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > >> timeout to
> > >> > > > detect
> > >> > > > > > client
> > >> > > > > > > > > > > failures when using the consumer group protocol.
> > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > >> minimum
> > >> > > > > > session
> > >> > > > > > > > > timeout.
> > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > >> maximum
> > >> > > > > > session
> > >> > > > > > > > > timeout.
> > >> > > > > > > > > >
> > >> > > > > > > > > > This is indeed a mistake. The default session
> timeout
> > >> should
> > >> > > > be 45s
> > >> > > > > > > > > > (the current default).
> > >> > > > > > > > > >
> > >> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > >> uniform],
> > >> > > > > > which means
> > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> know
> > >> why not
> > >> > > > > > uniform
> > >> > > > > > > > > one? I
> > >> > > > > > > > > > > thought usually users will choose uniform assignor
> > >> (former
> > >> > > > sticky
> > >> > > > > > > > > assinor)
> > >> > > > > > > > > > > for better evenly distribution. Any other reason
> we
> > >> choose
> > >> > > > range
> > >> > > > > > > > > assignor
> > >> > > > > > > > > > > as default?
> > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > >> server side
> > >> > > > > > assignors.
> > >> > > > > > > > > >
> > >> > > > > > > > > > The order on the server side has no influence
> because
> > >> the
> > >> > > > client
> > >> > > > > > must
> > >> > > > > > > > > > chose the selector that he wants to use. There is no
> > >> default
> > >> > > > in the
> > >> > > > > > > > > > current proposal. If the assignor is not specified
> by
> > >> the
> > >> > > > client,
> > >> > > > > > the
> > >> > > > > > > > > > request is rejected. The default client value for
> > >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > >> > > > > > > > > >
> > >> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> that my
> > >> > > > answers
> > >> > > > > > help
> > >> > > > > > > > > > to clarify things. I will update the KIP as well
> based
> > >> on your
> > >> > > > > > > > > > feedback.
> > >> > > > > > > > > >
> > >> > > > > > > > > > Cheers,
> > >> > > > > > > > > > David
> > >> > > > > > > > > >
> > >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > >> showuon@gmail.com>
> > >> > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Hi David,
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Thanks for the update.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Some more questions:
> > >> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > >> > > > > > > > > > > > The new group coordinator will have a state
> machine
> > >> per
> > >> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> > >> machine is
> > >> > > > > > modelled
> > >> > > > > > > > > as an
> > >> > > > > > > > > > > event loop. Those state machines will be executed
> in
> > >> > > > > > > > > > > *group.coordinator.threads* threads.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> assigning,
> > >> > > > > > reconciling,
> > >> > > > > > > > > stable,
> > >> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > >> right?
> > >> > > > > > > > > > > 1.2. What do you mean "each state machine is
> modelled
> > >> as an
> > >> > > > event
> > >> > > > > > > > > loop"?
> > >> > > > > > > > > > > 1.3. Why do we need a state machine per
> > >> *__consumer_offsets*
> > >> > > > > > > > > partitions?
> > >> > > > > > > > > > > Not a state machine "per consumer group" owned by
> a
> > >> group
> > >> > > > > > coordinator?
> > >> > > > > > > > > For
> > >> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > >> groups, and
> > >> > > > > > both
> > >> > > > > > > > > exist in
> > >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> machine
> > >> for it,
> > >> > > > or
> > >> > > > > > 2?
> > >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> the
> > >> number
> > >> > > > of
> > >> > > > > > threads
> > >> > > > > > > > > used
> > >> > > > > > > > > > > to run the state machines. But I'm wondering if
> the
> > >> purpose
> > >> > > > of
> > >> > > > > > the
> > >> > > > > > > > > threads
> > >> > > > > > > > > > > is only to keep the state of each consumer group
> (or
> > >> > > > > > > > > *__consumer_offsets*
> > >> > > > > > > > > > > partitions?), and no heavy computation, why
> should we
> > >> need
> > >> > > > > > > > > multi-threads
> > >> > > > > > > > > > > here?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > 2. For the default value in the new configs:
> > >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > >> default
> > >> > > > session
> > >> > > > > > > > > timeout not
> > >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> the
> > >> min/max
> > >> > > > > > session
> > >> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > >> timeout to
> > >> > > > detect
> > >> > > > > > client
> > >> > > > > > > > > > > failures when using the consumer group protocol.
> > >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > >> minimum
> > >> > > > > > session
> > >> > > > > > > > > timeout.
> > >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > >> maximum
> > >> > > > > > session
> > >> > > > > > > > > timeout.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > >> uniform],
> > >> > > > > > which means
> > >> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> know
> > >> why not
> > >> > > > > > uniform
> > >> > > > > > > > > one? I
> > >> > > > > > > > > > > thought usually users will choose uniform assignor
> > >> (former
> > >> > > > sticky
> > >> > > > > > > > > assinor)
> > >> > > > > > > > > > > for better evenly distribution. Any other reason
> we
> > >> choose
> > >> > > > range
> > >> > > > > > > > > assignor
> > >> > > > > > > > > > > as default?
> > >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > >> server side
> > >> > > > > > assignors.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Thank you.
> > >> > > > > > > > > > > Luke
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > >> showuon@gmail.com
> > >> > > > >
> > >> > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > Hi Sagar,
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > >> integrating with
> > >> > > > > > KIP-848,
> > >> > > > > > > > > but I
> > >> > > > > > > > > > > > think we should have a separate discussion
> thread
> > >> for the
> > >> > > > Kafka
> > >> > > > > > > > > Connect
> > >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> > >> Rebalance
> > >> > > > > > Protocol
> > >> > > > > > > > > [1],
> > >> > > > > > > > > > > > and let this discussion thread focus on consumer
> > >> rebalance
> > >> > > > > > protocol,
> > >> > > > > > > > > WDYT?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > [1]
> > >> > > > > > > > > > > >
> > >> > > > > > > > >
> > >> > > > > >
> > >> > > >
> > >>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > Thank you.
> > >> > > > > > > > > > > > Luke
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > >> > > > > > sagarmeansocean@gmail.com>
> > >> > > > > > > > > wrote:
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> Looks
> > >> like
> > >> > > > there's
> > >> > > > > > > > > agreement on
> > >> > > > > > > > > > > >> using separate APIs for Connect. I would
> revisit
> > >> the doc
> > >> > > > and
> > >> > > > > > see
> > >> > > > > > > > > what
> > >> > > > > > > > > > > >> changes are to be made.
> > >> > > > > > > > > > > >>
> > >> > > > > > > > > > > >> Thanks!
> > >> > > > > > > > > > > >> Sagar.
> > >> > > > > > > > > > > >>
> > >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > >> > > > > > > > > <dj...@confluent.io.invalid>
> > >> > > > > > > > > > > >> wrote:
> > >> > > > > > > > > > > >>
> > >> > > > > > > > > > > >> > Hi Sagar,
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >> > Thanks for the feedback and the document.
> That's
> > >> really
> > >> > > > > > helpful. I
> > >> > > > > > > > > > > >> > will take a look at it.
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> and the
> > >> > > > Consumer
> > >> > > > > > could
> > >> > > > > > > > > share
> > >> > > > > > > > > > > >> > the same underlying "engine". The main
> > >> difference is
> > >> > > > that
> > >> > > > > > the
> > >> > > > > > > > > Consumer
> > >> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > >> Connect
> > >> > > > assigns
> > >> > > > > > tasks
> > >> > > > > > > > > to
> > >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> > >> different
> > >> > > > > > resource
> > >> > > > > > > > > types
> > >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > >> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> > >> dedicated
> > >> > > > APIs
> > >> > > > > > > > > would be
> > >> > > > > > > > > > > >> > similar to the new ones but different on the
> > >> > > > > > content/resources and
> > >> > > > > > > > > > > >> > they would rely on the same engine on the
> > >> coordinator
> > >> > > > side.
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >> > I personally lean towards 2) because I am
> not a
> > >> fan of
> > >> > > > > > > > > overcharging
> > >> > > > > > > > > > > >> > APIs to serve different purposes. That being
> > >> said, I am
> > >> > > > not
> > >> > > > > > > > > opposed to
> > >> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >> > I think that we can continue to discuss it
> here
> > >> for now
> > >> > > > in
> > >> > > > > > order
> > >> > > > > > > > > to
> > >> > > > > > > > > > > >> > ensure that this KIP is compatible with what
> we
> > >> will do
> > >> > > > for
> > >> > > > > > > > > Connect in
> > >> > > > > > > > > > > >> > the future.
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >> > Best,
> > >> > > > > > > > > > > >> > David
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > >> > > > > > djacot@confluent.io>
> > >> > > > > > > > > wrote:
> > >> > > > > > > > > > > >> > >
> > >> > > > > > > > > > > >> > > Hi all,
> > >> > > > > > > > > > > >> > >
> > >> > > > > > > > > > > >> > > I am back from vacation. I will go through
> and
> > >> address
> > >> > > > > > your
> > >> > > > > > > > > comments
> > >> > > > > > > > > > > >> > > in the coming days. Thanks for your
> feedback.
> > >> > > > > > > > > > > >> > >
> > >> > > > > > > > > > > >> > > Cheers,
> > >> > > > > > > > > > > >> > > David
> > >> > > > > > > > > > > >> > >
> > >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> Harris
> > >> <
> > >> > > > > > > > > gharris1727@gmail.com
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >> > wrote:
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > Hey All!
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > >> > > > cooperative
> > >> > > > > > > > > rebalancing
> > >> > > > > > > > > > > >> > making it
> > >> > > > > > > > > > > >> > > > down the stack!
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > I had a few questions:
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > >> describes how
> > >> > > > > > member
> > >> > > > > > > > > epoch
> > >> > > > > > > > > > > >> > should
> > >> > > > > > > > > > > >> > > > advance in step with the group epoch and
> > >> assignment
> > >> > > > > > epoch
> > >> > > > > > > > > values. I
> > >> > > > > > > > > > > >> > think
> > >> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> > >> described
> > >> > > > in
> > >> > > > > > the
> > >> > > > > > > > > KIP. When
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > >> partitions to a
> > >> > > > > > worker,
> > >> > > > > > > > > what
> > >> > > > > > > > > > > >> member
> > >> > > > > > > > > > > >> > > > epoch does each incremental assignment
> use?
> > >> Are
> > >> > > > member
> > >> > > > > > epochs
> > >> > > > > > > > > > > >> re-used,
> > >> > > > > > > > > > > >> > and
> > >> > > > > > > > > > > >> > > > a single member epoch can correspond to
> > >> multiple
> > >> > > > > > different
> > >> > > > > > > > > > > >> > (monotonically
> > >> > > > > > > > > > > >> > > > larger) assignments?
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> opaque
> > >> to the
> > >> > > > group
> > >> > > > > > > > > > > >> coordinator? If
> > >> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > >> > > > implementations
> > >> > > > > > > > > interact
> > >> > > > > > > > > > > >> with
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > Reason field, and how is its common
> meaning
> > >> agreed
> > >> > > > > > upon? If
> > >> > > > > > > > > so, what
> > >> > > > > > > > > > > >> > is the
> > >> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > >> including
> > >> > > > such
> > >> > > > > > > > > functionality
> > >> > > > > > > > > > > >> > in the
> > >> > > > > > > > > > > >> > > > opaque metadata?
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> > >> "Thanks to
> > >> > > > > > this, the
> > >> > > > > > > > > input
> > >> > > > > > > > > > > >> of
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > client side assignor is entirely driven
> by
> > >> the group
> > >> > > > > > > > > coordinator.
> > >> > > > > > > > > > > >> The
> > >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > >> maintaining
> > >> > > > any
> > >> > > > > > state
> > >> > > > > > > > > besides
> > >> > > > > > > > > > > >> its
> > >> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> that the
> > >> > > > > > client-side
> > >> > > > > > > > > assignor
> > >> > > > > > > > > > > >> MAY
> > >> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> > >> (such as
> > >> > > > > > partition
> > >> > > > > > > > > > > >> > throughput,
> > >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc),
> or
> > >> that
> > >> > > > > > additional
> > >> > > > > > > > > > > >> > non-Metadata
> > >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> > >> > > > > > > > > > > >> > > > for
> > >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > >> > > > > > > > > > > >> > > > and
> > >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > >> > > > > > that
> > >> > > > > > > > > seem to
> > >> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> > >> these two
> > >> > > > > > > > > implementations
> > >> > > > > > > > > > > >> to
> > >> > > > > > > > > > > >> > be
> > >> > > > > > > > > > > >> > > > unified? This would serve to promote
> feature
> > >> parity
> > >> > > > of
> > >> > > > > > > > > server-side
> > >> > > > > > > > > > > >> and
> > >> > > > > > > > > > > >> > > > client-side assignors, and would also
> > >> facilitate
> > >> > > > > > operational
> > >> > > > > > > > > > > >> > flexibility in
> > >> > > > > > > > > > > >> > > > certain situations. For example, if a
> > >> server-side
> > >> > > > > > assignor
> > >> > > > > > > > > has some
> > >> > > > > > > > > > > >> > poor
> > >> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> > >> patched
> > >> > > > > > assignor to
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > client
> > >> > > > > > > > > > > >> > > > and switching one consumer group to a
> > >> client-side
> > >> > > > > > assignor
> > >> > > > > > > > > may be
> > >> > > > > > > > > > > >> > faster
> > >> > > > > > > > > > > >> > > > and less risky than patching all of the
> > >> brokers.
> > >> > > > With
> > >> > > > > > the
> > >> > > > > > > > > currently
> > >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > >> > > > reimplementation
> > >> > > > > > would
> > >> > > > > > > > > have
> > >> > > > > > > > > > > >> to be
> > >> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> diverged
> > >> > > > > > significantly,
> > >> > > > > > > > > then it
> > >> > > > > > > > > > > >> is
> > >> > > > > > > > > > > >> > > > possible that a reimplementation would
> not be
> > >> > > > possible.
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > --
> > >> > > > > > > > > > > >> > > > Greg Harris
> > >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > >> > > > > > > > > > > >> > > > github.com/gharris1727
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > >> > > > > > > > > sagarmeansocean@gmail.com>
> > >> > > > > > > > > > > >> > wrote:
> > >> > > > > > > > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > I created a confluence page to discuss
> how
> > >> Connect
> > >> > > > > > would
> > >> > > > > > > > > need to
> > >> > > > > > > > > > > >> > change
> > >> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> > >> Here's the
> > >> > > > page:
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >>
> > >> > > > > > > > >
> > >> > > > > >
> > >> > > >
> > >>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> tried
> > >> to keep
> > >> > > > a
> > >> > > > > > format
> > >> > > > > > > > > > > >> similar to
> > >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> Also,
> > >> do you
> > >> > > > > > think this
> > >> > > > > > > > > > > >> should
> > >> > > > > > > > > > > >> > be
> > >> > > > > > > > > > > >> > > > > moved to a separate discussion thread
> or
> > >> is this
> > >> > > > one
> > >> > > > > > fine?
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > Thanks!
> > >> > > > > > > > > > > >> > > > > Sagar.
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > >> > > > > > > > > sagarmeansocean@gmail.com>
> > >> > > > > > > > > > > >> > wrote:
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > >> > > > > > > > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> Kafka
> > >> Streams.
> > >> > > > > > Sure, I
> > >> > > > > > > > > would do
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > >> document.
> > >> > > > > > > > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > > Thanks!
> > >> > > > > > > > > > > >> > > > > > Sagar.
> > >> > > > > > > > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> Guozhang
> > >> Wang <
> > >> > > > > > > > > > > >> wangguoz@gmail.com>
> > >> > > > > > > > > > > >> > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > > >> It would be great if you could come
> > >> back with
> > >> > > > some
> > >> > > > > > > > > analysis on
> > >> > > > > > > > > > > >> > how to
> > >> > > > > > > > > > > >> > > > > >> implement the Connect side
> integration
> > >> with
> > >> > > > the new
> > >> > > > > > > > > protocol;
> > >> > > > > > > > > > > >> so
> > >> > > > > > > > > > > >> > far
> > >> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> "protocol
> > >> type"
> > >> > > > we
> > >> > > > > > did not
> > >> > > > > > > > > yet
> > >> > > > > > > > > > > >> think
> > >> > > > > > > > > > > >> > > > > through
> > >> > > > > > > > > > > >> > > > > >> the Connect side implementations.
> For
> > >> Streams
> > >> > > > > > here's a
> > >> > > > > > > > > draft of
> > >> > > > > > > > > > > >> > > > > >> integration
> > >> > > > > > > > > > > >> > > > > >> plan:
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >>
> > >> > > > > > > > >
> > >> > > > > >
> > >> > > >
> > >>
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > >> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> Connect.
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> Sagar <
> > >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > >> > > > > > > > > > > >> > >
> > >> > > > > > > > > > > >> > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > > >> > Hi David,
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> > >> reason I
> > >> > > > thought
> > >> > > > > > > > > connect can
> > >> > > > > > > > > > > >> > also fit
> > >> > > > > > > > > > > >> > > > > >> into
> > >> > > > > > > > > > > >> > > > > >> > this new scheme is that even
> today the
> > >> > > > connect
> > >> > > > > > uses a
> > >> > > > > > > > > > > >> > > > > WorkerCoordinator
> > >> > > > > > > > > > > >> > > > > >> > extending from
> AbstractCoordinator to
> > >> empower
> > >> > > > > > > > > rebalances of
> > >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > >> WorkerCoordinator sets
> > >> > > > the
> > >> > > > > > > > > > > >> protocolType()
> > >> > > > > > > > > > > >> > to
> > >> > > > > > > > > > > >> > > > > >> connect
> > >> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> > >> plumbing
> > >> > > > into
> > >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > >> connect would
> > >> > > > be
> > >> > > > > > > > > similar at a
> > >> > > > > > > > > > > >> > high
> > >> > > > > > > > > > > >> > > > > level
> > >> > > > > > > > > > > >> > > > > >> to
> > >> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> because
> > >> of the
> > >> > > > > > Client
> > >> > > > > > > > > side
> > >> > > > > > > > > > > >> > assignors
> > >> > > > > > > > > > > >> > > > > being
> > >> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
> > >> level, we
> > >> > > > > > might
> > >> > > > > > > > > need to
> > >> > > > > > > > > > > >> make
> > >> > > > > > > > > > > >> > a lot
> > >> > > > > > > > > > > >> > > > > of
> > >> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > >> assignment
> > >> > > > protocol
> > >> > > > > > like
> > >> > > > > > > > > > > >> enhancing
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > >> SyncGroup and
> > >> > > > > > using
> > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > >> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> > >> streams (or
> > >> > > > > > there
> > >> > > > > > > > > might be
> > >> > > > > > > > > > > >> > > > > >> deviations). I
> > >> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> > >> analysis of
> > >> > > > the
> > >> > > > > > same
> > >> > > > > > > > > and we
> > >> > > > > > > > > > > >> > can have
> > >> > > > > > > > > > > >> > > > > a
> > >> > > > > > > > > > > >> > > > > >> > separate discussion thread for
> that
> > >> as that
> > >> > > > would
> > >> > > > > > > > > derail this
> > >> > > > > > > > > > > >> > > > > discussion
> > >> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
> > >> good to
> > >> > > > you.
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> > Thanks!
> > >> > > > > > > > > > > >> > > > > >> > Sagar.
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> David
> > >> Jacot
> > >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> > wrote:
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > >> `Assignment#error`.
> > >> > > > > > Sure, I
> > >> > > > > > > > > can
> > >> > > > > > > > > > > >> > mention it.
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> from
> > >> his
> > >> > > > current
> > >> > > > > > > > > assignment
> > >> > > > > > > > > > > >> to
> > >> > > > > > > > > > > >> > his
> > >> > > > > > > > > > > >> > > > > >> > > target assignment when he can
> move
> > >> to
> > >> > > > epoch 3.
> > >> > > > > > When
> > >> > > > > > > > > that
> > >> > > > > > > > > > > >> > happens,
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
> > >> persisted
> > >> > > > > > with all
> > >> > > > > > > > > its
> > >> > > > > > > > > > > >> > assigned
> > >> > > > > > > > > > > >> > > > > >> > > partitions even if they are not
> all
> > >> revoked
> > >> > > > > > yet. In
> > >> > > > > > > > > other
> > >> > > > > > > > > > > >> > words, the
> > >> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> target
> > >> > > > > > assignment.
> > >> > > > > > > > > This is
> > >> > > > > > > > > > > >> > basically
> > >> > > > > > > > > > > >> > > > > an
> > >> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> > >> write all
> > >> > > > the
> > >> > > > > > > > > changes to
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > log.
> > >> > > > > > > > > > > >> > > > > The
> > >> > > > > > > > > > > >> > > > > >> > > examples are based on the
> persisted
> > >> state
> > >> > > > so I
> > >> > > > > > > > > understand
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> > >> improve
> > >> > > > this in
> > >> > > > > > the
> > >> > > > > > > > > > > >> > description.
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> > >> reuse the
> > >> > > > > > protocol
> > >> > > > > > > > > with a
> > >> > > > > > > > > > > >> > client side
> > >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > >> protocol. The
> > >> > > > > > assignment
> > >> > > > > > > > > is
> > >> > > > > > > > > > > >> about
> > >> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> could
> > >> > > > Connect
> > >> > > > > > fit
> > >> > > > > > > > > into this?
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> > > Best,
> > >> > > > > > > > > > > >> > > > > >> > > David
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> > >> Sagar <
> > >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > >> > > > > > > > > > > >> > > > > >> wrote:
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
> > >> minor
> > >> > > > > > observations:
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > >> section in
> > >> > > > Client
> > >> > > > > > Side
> > >> > > > > > > > > mode
> > >> > > > > > > > > > > >> > Assignment
> > >> > > > > > > > > > > >> > > > > >> > > process,
> > >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this
> case,
> > >> the
> > >> > > > client
> > >> > > > > > side
> > >> > > > > > > > > assignor
> > >> > > > > > > > > > > >> can
> > >> > > > > > > > > > > >> > > > > return
> > >> > > > > > > > > > > >> > > > > >> an
> > >> > > > > > > > > > > >> > > > > >> > > > error to the group
> coordinator`.
> > >> In this
> > >> > > > > > case are
> > >> > > > > > > > > you
> > >> > > > > > > > > > > >> > referring to
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > >> AssignmentError
> > >> > > > that's
> > >> > > > > > > > > listed down
> > >> > > > > > > > > > > >> > towards
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > end?
> > >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
> > >> make sense
> > >> > > > to
> > >> > > > > > > > > mention this
> > >> > > > > > > > > > > >> > > > > explicitly
> > >> > > > > > > > > > > >> > > > > >> > here?
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> section, I
> > >> have a
> > >> > > > > > slight
> > >> > > > > > > > > > > >> confusion,
> > >> > > > > > > > > > > >> > not
> > >> > > > > > > > > > > >> > > > > sure
> > >> > > > > > > > > > > >> > > > > >> if
> > >> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
> > >> this step:
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> > >> coordinator
> > >> > > > > > > > > transitions him
> > >> > > > > > > > > > > >> to
> > >> > > > > > > > > > > >> > epoch
> > >> > > > > > > > > > > >> > > > > 3
> > >> > > > > > > > > > > >> > > > > >> > > because
> > >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> revoke. It
> > >> > > > persists
> > >> > > > > > the
> > >> > > > > > > > > change and
> > >> > > > > > > > > > > >> > reply.
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > >> > > > > > > > > > > >> > > > > >> > > >       - A
> > >> > > > > > > > > > > >> > > > > >> > > >       - B
> > >> > > > > > > > > > > >> > > > > >> > > >       - C
> > >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> (epoch=3)
> > >> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > >> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > >> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > >> partitions=[foo-0,
> > >> > > > > > foo-1]
> > >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > >> partitions=[foo-2]
> > >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > >> partitions=[foo-1]
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> transitions
> > >> to
> > >> > > > epoch 3
> > >> > > > > > but
> > >> > > > > > > > > cannot
> > >> > > > > > > > > > > >> get
> > >> > > > > > > > > > > >> > foo-1
> > >> > > > > > > > > > > >> > > > > >> yet.
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that
> member C
> > >> can't
> > >> > > > get
> > >> > > > > > the
> > >> > > > > > > > > foo-1
> > >> > > > > > > > > > > >> > partition
> > >> > > > > > > > > > > >> > > > > yet,
> > >> > > > > > > > > > > >> > > > > >> > but
> > >> > > > > > > > > > > >> > > > > >> > > > based on the description
> above,
> > >> it seems
> > >> > > > it
> > >> > > > > > > > > already has
> > >> > > > > > > > > > > >> it.
> > >> > > > > > > > > > > >> > Do you
> > >> > > > > > > > > > > >> > > > > >> > think
> > >> > > > > > > > > > > >> > > > > >> > > it
> > >> > > > > > > > > > > >> > > > > >> > > > would be better to remove it
> and
> > >> > > > populate it
> > >> > > > > > only
> > >> > > > > > > > > when it
> > >> > > > > > > > > > > >> > actually
> > >> > > > > > > > > > > >> > > > > >> gets
> > >> > > > > > > > > > > >> > > > > >> > > it?
> > >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> > >> places, so
> > >> > > > have
> > >> > > > > > I
> > >> > > > > > > > > > > >> understood it
> > >> > > > > > > > > > > >> > > > > >> > incorrectly
> > >> > > > > > > > > > > >> > > > > >> > > ?
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might
> be
> > >> out of
> > >> > > > scope
> > >> > > > > > of
> > >> > > > > > > > > this
> > >> > > > > > > > > > > >> > discussion,
> > >> > > > > > > > > > > >> > > > > but
> > >> > > > > > > > > > > >> > > > > >> > from
> > >> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > >> probably be
> > >> > > > > > running in
> > >> > > > > > > > > client
> > >> > > > > > > > > > > >> > side
> > >> > > > > > > > > > > >> > > > > >> assignor
> > >> > > > > > > > > > > >> > > > > >> > > mode
> > >> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > >> protocol as it
> > >> > > > has
> > >> > > > > > its
> > >> > > > > > > > > own
> > >> > > > > > > > > > > >> Custom
> > >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > >> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00
> PM
> > >> David
> > >> > > > Jacot
> > >> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is
> to
> > >> move
> > >> > > > forward
> > >> > > > > > with
> > >> > > > > > > > > > > >> > specialized API
> > >> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> > >> generic API.
> > >> > > > For
> > >> > > > > > > > > Connect, we
> > >> > > > > > > > > > > >> > can apply
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > >> reuse/share the
> > >> > > > core
> > >> > > > > > > > > > > >> > implementation on
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> > >> registry, I
> > >> > > > > > think
> > >> > > > > > > > > that we
> > >> > > > > > > > > > > >> > should
> > >> > > > > > > > > > > >> > > > > >> consider
> > >> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> > >> simple
> > >> > > > > > > > > membership/leader
> > >> > > > > > > > > > > >> > election.
> > >> > > > > > > > > > > >> > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > Best,
> > >> > > > > > > > > > > >> > > > > >> > > > > David
> > >> > > > > > > > > > > >> > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at
> 10:22
> > >> AM Ismael
> > >> > > > > > Juma <
> > >> > > > > > > > > > > >> > ismael@juma.me.uk
> > >> > > > > > > > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > >> java.util.regex.Pattern
> > >> > > > vs
> > >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > >> > > > > > > > > > > >> > > > > >> > > > > should
> > >> > > > > > > > > > > >> > > > > >> > > > > > document the differences
> in
> > >> more
> > >> > > > detail
> > >> > > > > > before
> > >> > > > > > > > > > > >> deciding
> > >> > > > > > > > > > > >> > one
> > >> > > > > > > > > > > >> > > > > way
> > >> > > > > > > > > > > >> > > > > >> or
> > >> > > > > > > > > > > >> > > > > >> > > > > another.
> > >> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > >> > > > > > > > > java.util.regex.Pattern,
> > >> > > > > > > > > > > >> they
> > >> > > > > > > > > > > >> > expect
> > >> > > > > > > > > > > >> > > > > >> > their
> > >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored.
> If
> > >> we are
> > >> > > > doing
> > >> > > > > > > > > something
> > >> > > > > > > > > > > >> > different,
> > >> > > > > > > > > > > >> > > > > >> then
> > >> > > > > > > > > > > >> > > > > >> > we
> > >> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> > >> overload
> > >> > > > with
> > >> > > > > > our own
> > >> > > > > > > > > > > >> Pattern
> > >> > > > > > > > > > > >> > class
> > >> > > > > > > > > > > >> > > > > (I
> > >> > > > > > > > > > > >> > > > > >> > > don't
> > >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> > >> re2j's at
> > >> > > > this
> > >> > > > > > > > > point).
> > >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids,
> any
> > >> major new
> > >> > > > > > protocol
> > >> > > > > > > > > should
> > >> > > > > > > > > > > >> > > > > integrate
> > >> > > > > > > > > > > >> > > > > >> > fully
> > >> > > > > > > > > > > >> > > > > >> > > > > with
> > >> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> topic
> > >> > > > > > recreation case
> > >> > > > > > > > > > > >> > correctly.
> > >> > > > > > > > > > > >> > > > > That's
> > >> > > > > > > > > > > >> > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> > > main
> > >> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> > >> agree with
> > >> > > > > > David
> > >> > > > > > > > > that we'd
> > >> > > > > > > > > > > >> > want to
> > >> > > > > > > > > > > >> > > > > add
> > >> > > > > > > > > > > >> > > > > >> > > topic
> > >> > > > > > > > > > > >> > > > > >> > > > > ids
> > >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
> > >> that don't
> > >> > > > > > have it
> > >> > > > > > > > > yet and
> > >> > > > > > > > > > > >> > that we
> > >> > > > > > > > > > > >> > > > > can
> > >> > > > > > > > > > > >> > > > > >> > > probably
> > >> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> versus
> > >> adding
> > >> > > > new
> > >> > > > > > APIs
> > >> > > > > > > > > to the
> > >> > > > > > > > > > > >> > Java
> > >> > > > > > > > > > > >> > > > > >> Consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > (unless
> > >> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new
> APIs
> > >> is
> > >> > > > required
> > >> > > > > > for
> > >> > > > > > > > > > > >> reasonable
> > >> > > > > > > > > > > >> > > > > >> semantics).
> > >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> about
> > >> the
> > >> > > > > > coordinator
> > >> > > > > > > > > storing
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> configs.
> > >> > > > > > > > > > > >> > > > > >> > > It's
> > >> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > >> > > > centralized
> > >> > > > > > in the
> > >> > > > > > > > > > > >> metadata
> > >> > > > > > > > > > > >> > log for
> > >> > > > > > > > > > > >> > > > > >> > > various
> > >> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > >> visibility,
> > >> > > > > > consistency,
> > >> > > > > > > > > > > >> etc.).
> > >> > > > > > > > > > > >> > > > > >> Similarly, I
> > >> > > > > > > > > > > >> > > > > >> > > am
> > >> > > > > > > > > > > >> > > > > >> > > > > not
> > >> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> > >> deleting
> > >> > > > > > configs in a
> > >> > > > > > > > > way
> > >> > > > > > > > > > > >> that
> > >> > > > > > > > > > > >> > they
> > >> > > > > > > > > > > >> > > > > >> cannot
> > >> > > > > > > > > > > >> > > > > >> > > be
> > >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property
> > >> for modern
> > >> > > > > > systems
> > >> > > > > > > > > is to
> > >> > > > > > > > > > > >> > minimize
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > number
> > >> > > > > > > > > > > >> > > > > >> > > > > of
> > >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> > >> scenarios.
> > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at
> 3:47
> > >> PM David
> > >> > > > > > Jacot
> > >> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> answers
> > >> are
> > >> > > > below:
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> > >> especially
> > >> > > > > > the last
> > >> > > > > > > > > step
> > >> > > > > > > > > > > >> when
> > >> > > > > > > > > > > >> > > > > clients
> > >> > > > > > > > > > > >> > > > > >> > > flip the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > >> protocol, in
> > >> > > > which
> > >> > > > > > we
> > >> > > > > > > > > would
> > >> > > > > > > > > > > >> have a
> > >> > > > > > > > > > > >> > > > > window
> > >> > > > > > > > > > > >> > > > > >> > where
> > >> > > > > > > > > > > >> > > > > >> > > > > both
> > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> old
> > >> > > > protocols /
> > >> > > > > > rpcs
> > >> > > > > > > > > are
> > >> > > > > > > > > > > >> used
> > >> > > > > > > > > > > >> > by
> > >> > > > > > > > > > > >> > > > > >> members
> > >> > > > > > > > > > > >> > > > > >> > of
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > same
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > >> coordinator could
> > >> > > > > > "mimic"
> > >> > > > > > > > > the old
> > >> > > > > > > > > > > >> > behavior
> > >> > > > > > > > > > > >> > > > > >> while
> > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> something
> > >> we
> > >> > > > need to
> > >> > > > > > > > > present
> > >> > > > > > > > > > > >> about.
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published
> a
> > >> new
> > >> > > > version
> > >> > > > > > of KIP
> > >> > > > > > > > > which
> > >> > > > > > > > > > > >> > includes
> > >> > > > > > > > > > > >> > > > > >> more
> > >> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See
> the
> > >> > > > "Supporting
> > >> > > > > > > > > Online
> > >> > > > > > > > > > > >> > Consumer
> > >> > > > > > > > > > > >> > > > > Group
> > >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > >> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > >> > > > Deprecation,
> > >> > > > > > and
> > >> > > > > > > > > Migration
> > >> > > > > > > > > > > >> > Plan". I
> > >> > > > > > > > > > > >> > > > > >> think
> > >> > > > > > > > > > > >> > > > > >> > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through
> a
> > >> few cases
> > >> > > > > > now but
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > overall idea
> > >> > > > > > > > > > > >> > > > > >> and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > >> understandable.
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > >> ids. So
> > >> > > > far as
> > >> > > > > > > > > KIP-516 the
> > >> > > > > > > > > > > >> > topic ids
> > >> > > > > > > > > > > >> > > > > >> are
> > >> > > > > > > > > > > >> > > > > >> > > only
> > >> > > > > > > > > > > >> > > > > >> > > > > used
> > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> > >> client,
> > >> > > > but
> > >> > > > > > they
> > >> > > > > > > > > are not
> > >> > > > > > > > > > > >> > exposed
> > >> > > > > > > > > > > >> > > > > via
> > >> > > > > > > > > > > >> > > > > >> any
> > >> > > > > > > > > > > >> > > > > >> > > > > public
> > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> think
> > >> the
> > >> > > > > > question is,
> > >> > > > > > > > > first
> > >> > > > > > > > > > > >> > should we
> > >> > > > > > > > > > > >> > > > > >> let
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> maintaining
> > >> the
> > >> > > > names
> > >> > > > > > -> ids
> > >> > > > > > > > > mapping
> > >> > > > > > > > > > > >> > itself
> > >> > > > > > > > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> > fully
> > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > >> existing
> > >> > > > RPCs
> > >> > > > > > and the
> > >> > > > > > > > > new
> > >> > > > > > > > > > > >> RPCs
> > >> > > > > > > > > > > >> > with
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > topic
> > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> ever
> > >> consider
> > >> > > > > > exposing
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > topic ids
> > >> > > > > > > > > > > >> > > > > in
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> (both
> > >> > > > > > > > > subscribe/assign, as
> > >> > > > > > > > > > > >> well
> > >> > > > > > > > > > > >> > as in
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> like
> > >> topic
> > >> > > > > > > > > > > >> > deletion-and-recreation).
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we
> would
> > >> include
> > >> > > > > > converting
> > >> > > > > > > > > all
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > offsets
> > >> > > > > > > > > > > >> > > > > >> > > related
> > >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids
> in
> > >> this
> > >> > > > KIP,
> > >> > > > > > the
> > >> > > > > > > > > consumer
> > >> > > > > > > > > > > >> > would be
> > >> > > > > > > > > > > >> > > > > >> able
> > >> > > > > > > > > > > >> > > > > >> > to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
> > >> ids. That
> > >> > > > > > being
> > >> > > > > > > > > said, it
> > >> > > > > > > > > > > >> > still has
> > >> > > > > > > > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> > > provide
> > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> various
> > >> APIs so
> > >> > > > > > having a
> > >> > > > > > > > > > > >> mapping
> > >> > > > > > > > > > > >> > in the
> > >> > > > > > > > > > > >> > > > > >> > > consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> > >> opinion on
> > >> > > > > > this.
> > >> > > > > > > > > Here I
> > >> > > > > > > > > > > >> > wonder if
> > >> > > > > > > > > > > >> > > > > >> this
> > >> > > > > > > > > > > >> > > > > >> > > goes
> > >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
> > >> KIP. I
> > >> > > > would
> > >> > > > > > rather
> > >> > > > > > > > > focus
> > >> > > > > > > > > > > >> on
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> > internals
> > >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider
> > >> this
> > >> > > > > > separately if
> > >> > > > > > > > > we see
> > >> > > > > > > > > > > >> > value in
> > >> > > > > > > > > > > >> > > > > >> doing
> > >> > > > > > > > > > > >> > > > > >> > > it.
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> > >> point about
> > >> > > > > > using
> > >> > > > > > > > > topic ids
> > >> > > > > > > > > > > >> > in the
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> ConsumerGroupHeartbeatRequest, I
> > >> > > > think
> > >> > > > > > that
> > >> > > > > > > > > there
> > >> > > > > > > > > > > >> is
> > >> > > > > > > > > > > >> > one
> > >> > > > > > > > > > > >> > > > > >> > advantage
> > >> > > > > > > > > > > >> > > > > >> > > in
> > >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> consumer
> > >> will
> > >> > > > have
> > >> > > > > > the
> > >> > > > > > > > > > > >> opportunity
> > >> > > > > > > > > > > >> > to
> > >> > > > > > > > > > > >> > > > > >> validate
> > >> > > > > > > > > > > >> > > > > >> > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> > >> passing
> > >> > > > them
> > >> > > > > > into
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> group
> > >> > > > > > > > > > > >> > > > > rebalance
> > >> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > >> > > > coordinator
> > >> > > > > > will
> > >> > > > > > > > > also
> > >> > > > > > > > > > > >> notice
> > >> > > > > > > > > > > >> > it but
> > >> > > > > > > > > > > >> > > > > >> it
> > >> > > > > > > > > > > >> > > > > >> > > does
> > >> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> > >> reject an
> > >> > > > > > invalid
> > >> > > > > > > > > topic in
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> response.
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> David
> > >> on all
> > >> > > > other
> > >> > > > > > minor
> > >> > > > > > > > > > > >> questions
> > >> > > > > > > > > > > >> > > > > except
> > >> > > > > > > > > > > >> > > > > >> for
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > >> question:
> > >> > > > > > personally
> > >> > > > > > > > > I think
> > >> > > > > > > > > > > >> > it's not
> > >> > > > > > > > > > > >> > > > > >> > > necessary
> > >> > > > > > > > > > > >> > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> subscribe
> > >> API with
> > >> > > > > > Pattern,
> > >> > > > > > > > > but
> > >> > > > > > > > > > > >> > instead we
> > >> > > > > > > > > > > >> > > > > >> still
> > >> > > > > > > > > > > >> > > > > >> > > use
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> > >> that our
> > >> > > > > > > > > subscription may
> > >> > > > > > > > > > > >> be
> > >> > > > > > > > > > > >> > > > > >> rejected by
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> > >> case is a
> > >> > > > > > very rare
> > >> > > > > > > > > > > >> scenario
> > >> > > > > > > > > > > >> > I felt
> > >> > > > > > > > > > > >> > > > > >> > using
> > >> > > > > > > > > > > >> > > > > >> > > an
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> based
> > >> > > > > > subscription may
> > >> > > > > > > > > be
> > >> > > > > > > > > > > >> more
> > >> > > > > > > > > > > >> > > > > >> vulnerable
> > >> > > > > > > > > > > >> > > > > >> > to
> > >> > > > > > > > > > > >> > > > > >> > > > > various
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have
> to
> > >> look at
> > >> > > > the
> > >> > > > > > > > > differences
> > >> > > > > > > > > > > >> > between
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > two
> > >> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > >> understand the
> > >> > > > > > potential
> > >> > > > > > > > > issues.
> > >> > > > > > > > > > > >> My
> > >> > > > > > > > > > > >> > > > > >> > > understanding is
> > >> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all
> the
> > >> basic
> > >> > > > > > regular
> > >> > > > > > > > > > > >> > expressions. The
> > >> > > > > > > > > > > >> > > > > >> > > differences
> > >> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> mainly
> > >> about
> > >> > > > the
> > >> > > > > > various
> > >> > > > > > > > > > > >> character
> > >> > > > > > > > > > > >> > > > > >> classes. I
> > >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
> > >> think
> > >> > > > about
> > >> > > > > > this.
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > David
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> > >> 11:28 PM
> > >> > > > > > Guozhang
> > >> > > > > > > > > Wang <
> > >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > >> > > > > > > > > > > >> > > > > >> > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think
> on
> > >> the high
> > >> > > > > > level
> > >> > > > > > > > > there are
> > >> > > > > > > > > > > >> > two meta
> > >> > > > > > > > > > > >> > > > > >> > > points we
> > >> > > > > > > > > > > >> > > > > >> > > > > need
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit
> more:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> > >> especially
> > >> > > > > > the last
> > >> > > > > > > > > step
> > >> > > > > > > > > > > >> when
> > >> > > > > > > > > > > >> > > > > clients
> > >> > > > > > > > > > > >> > > > > >> > > flip the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > >> protocol, in
> > >> > > > which
> > >> > > > > > we
> > >> > > > > > > > > would
> > >> > > > > > > > > > > >> have a
> > >> > > > > > > > > > > >> > > > > window
> > >> > > > > > > > > > > >> > > > > >> > where
> > >> > > > > > > > > > > >> > > > > >> > > > > both
> > >> > > > > > > > > > > >> > > > > >> > > > > > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> old
> > >> > > > protocols /
> > >> > > > > > rpcs
> > >> > > > > > > > > are
> > >> > > > > > > > > > > >> used
> > >> > > > > > > > > > > >> > by
> > >> > > > > > > > > > > >> > > > > >> members
> > >> > > > > > > > > > > >> > > > > >> > of
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > same
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > >> coordinator could
> > >> > > > > > "mimic"
> > >> > > > > > > > > the old
> > >> > > > > > > > > > > >> > behavior
> > >> > > > > > > > > > > >> > > > > >> while
> > >> > > > > > > > > > > >> > > > > >> > > > > using the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> something
> > >> we
> > >> > > > need to
> > >> > > > > > > > > present
> > >> > > > > > > > > > > >> about.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > >> ids. So
> > >> > > > far as
> > >> > > > > > > > > KIP-516 the
> > >> > > > > > > > > > > >> > topic ids
> > >> > > > > > > > > > > >> > > > > >> are
> > >> > > > > > > > > > > >> > > > > >> > > only
> > >> > > > > > > > > > > >> > > > > >> > > > > used
> > >> > > > > > > > > > > >> > > > > >> > > > > > > as
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> > >> client,
> > >> > > > but
> > >> > > > > > they
> > >> > > > > > > > > are not
> > >> > > > > > > > > > > >> > exposed
> > >> > > > > > > > > > > >> > > > > via
> > >> > > > > > > > > > > >> > > > > >> any
> > >> > > > > > > > > > > >> > > > > >> > > > > public
> > >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> think
> > >> the
> > >> > > > > > question is,
> > >> > > > > > > > > first
> > >> > > > > > > > > > > >> > should we
> > >> > > > > > > > > > > >> > > > > >> let
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> maintaining
> > >> the
> > >> > > > names
> > >> > > > > > -> ids
> > >> > > > > > > > > mapping
> > >> > > > > > > > > > > >> > itself
> > >> > > > > > > > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> > fully
> > >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > >> existing
> > >> > > > RPCs
> > >> > > > > > and the
> > >> > > > > > > > > new
> > >> > > > > > > > > > > >> RPCs
> > >> > > > > > > > > > > >> > with
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > topic
> > >> > > > > > > > > > > >> > > > > >> > > > > ids;
> > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> ever
> > >> consider
> > >> > > > > > exposing
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > topic ids
> > >> > > > > > > > > > > >> > > > > in
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> (both
> > >> > > > > > > > > subscribe/assign, as
> > >> > > > > > > > > > > >> well
> > >> > > > > > > > > > > >> > as in
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> like
> > >> topic
> > >> > > > > > > > > > > >> > deletion-and-recreation).
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> David
> > >> on all
> > >> > > > other
> > >> > > > > > minor
> > >> > > > > > > > > > > >> questions
> > >> > > > > > > > > > > >> > > > > except
> > >> > > > > > > > > > > >> > > > > >> for
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > >> question:
> > >> > > > > > personally
> > >> > > > > > > > > I think
> > >> > > > > > > > > > > >> > it's not
> > >> > > > > > > > > > > >> > > > > >> > > necessary
> > >> > > > > > > > > > > >> > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> subscribe
> > >> API with
> > >> > > > > > Pattern,
> > >> > > > > > > > > but
> > >> > > > > > > > > > > >> > instead we
> > >> > > > > > > > > > > >> > > > > >> still
> > >> > > > > > > > > > > >> > > > > >> > > use
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> > >> that our
> > >> > > > > > > > > subscription may
> > >> > > > > > > > > > > >> be
> > >> > > > > > > > > > > >> > > > > >> rejected by
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> > >> case is a
> > >> > > > > > very rare
> > >> > > > > > > > > > > >> scenario
> > >> > > > > > > > > > > >> > I felt
> > >> > > > > > > > > > > >> > > > > >> > using
> > >> > > > > > > > > > > >> > > > > >> > > an
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> based
> > >> > > > > > subscription may
> > >> > > > > > > > > be
> > >> > > > > > > > > > > >> more
> > >> > > > > > > > > > > >> > > > > >> vulnerable
> > >> > > > > > > > > > > >> > > > > >> > to
> > >> > > > > > > > > > > >> > > > > >> > > > > various
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022
> at
> > >> 5:23 AM
> > >> > > > > > David Jacot
> > >> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > >> feedback. Let
> > >> > > > me
> > >> > > > > > answer
> > >> > > > > > > > > your
> > >> > > > > > > > > > > >> > questions
> > >> > > > > > > > > > > >> > > > > >> > inline.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> > >> premature to
> > >> > > > > > talk about
> > >> > > > > > > > > > > >> target
> > >> > > > > > > > > > > >> > > > > versions
> > >> > > > > > > > > > > >> > > > > >> for
> > >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > >> existing group
> > >> > > > > > protocol.
> > >> > > > > > > > > > > >> Unlike
> > >> > > > > > > > > > > >> > KRaft,
> > >> > > > > > > > > > > >> > > > > >> this
> > >> > > > > > > > > > > >> > > > > >> > > > > affects a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > core
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol
> and
> > >> hence
> > >> > > > > > > > > deprecation/removal
> > >> > > > > > > > > > > >> > will be
> > >> > > > > > > > > > > >> > > > > >> > heavily
> > >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > >> applications
> > >> > > > migrate
> > >> > > > > > to
> > >> > > > > > > > > the new
> > >> > > > > > > > > > > >> > protocol.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I
> will
> > >> > > > remove it.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> > >> intend to
> > >> > > > > > release
> > >> > > > > > > > > this in
> > >> > > > > > > > > > > >> > 4.x, but
> > >> > > > > > > > > > > >> > > > > it
> > >> > > > > > > > > > > >> > > > > >> > > wasn't
> > >> > > > > > > > > > > >> > > > > >> > > > > made
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> that
> > >> as a
> > >> > > > way to
> > >> > > > > > > > > estimate
> > >> > > > > > > > > > > >> when
> > >> > > > > > > > > > > >> > we'd
> > >> > > > > > > > > > > >> > > > > >> > > deprecate
> > >> > > > > > > > > > > >> > > > > >> > > > > and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> protocol, I
> > >> also
> > >> > > > > > suggest
> > >> > > > > > > > > removing
> > >> > > > > > > > > > > >> > this part.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> > >> reasoning. As
> > >> > > > > > > > > explained, I
> > >> > > > > > > > > > > >> plan
> > >> > > > > > > > > > > >> > to
> > >> > > > > > > > > > > >> > > > > >> rewrite
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> > >> while we
> > >> > > > > > implement
> > >> > > > > > > > > the new
> > >> > > > > > > > > > > >> > protocol.
> > >> > > > > > > > > > > >> > > > > >> This
> > >> > > > > > > > > > > >> > > > > >> > > means
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> will
> > >> be
> > >> > > > slightly
> > >> > > > > > > > > different
> > >> > > > > > > > > > > >> > (e.g.
> > >> > > > > > > > > > > >> > > > > >> threading
> > >> > > > > > > > > > > >> > > > > >> > > > > model).
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted
> to
> > >> tighten
> > >> > > > the
> > >> > > > > > > > > switch from
> > >> > > > > > > > > > > >> > the old
> > >> > > > > > > > > > > >> > > > > >> group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the
> new
> > >> group
> > >> > > > > > coordinator
> > >> > > > > > > > > to a
> > >> > > > > > > > > > > >> > major
> > >> > > > > > > > > > > >> > > > > >> release.
> > >> > > > > > > > > > > >> > > > > >> > > The
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would
> be to
> > >> use a
> > >> > > > flag
> > >> > > > > > to do
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > switch
> > >> > > > > > > > > > > >> > > > > >> instead
> > >> > > > > > > > > > > >> > > > > >> > of
> > >> > > > > > > > > > > >> > > > > >> > > > > relying
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> upgrade.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to
> flesh
> > >> out the
> > >> > > > > > details of
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > migration
> > >> > > > > > > > > > > >> > > > > >> story.
> > >> > > > > > > > > > > >> > > > > >> > > It
> > >> > > > > > > > > > > >> > > > > >> > > > > sounds
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we
> will
> > >> support
> > >> > > > > > online
> > >> > > > > > > > > > > >> migrations.
> > >> > > > > > > > > > > >> > Is
> > >> > > > > > > > > > > >> > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > correct?
> > >> > > > > > > > > > > >> > > > > >> > > > > We
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> > >> detail. It
> > >> > > > could
> > >> > > > > > also
> > >> > > > > > > > > be done
> > >> > > > > > > > > > > >> > as a
> > >> > > > > > > > > > > >> > > > > >> separate
> > >> > > > > > > > > > > >> > > > > >> > > KIP,
> > >> > > > > > > > > > > >> > > > > >> > > > > if
> > >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
> > >> online
> > >> > > > > > migrations
> > >> > > > > > > > > for the
> > >> > > > > > > > > > > >> > group.
> > >> > > > > > > > > > > >> > > > > That
> > >> > > > > > > > > > > >> > > > > >> > > means
> > >> > > > > > > > > > > >> > > > > >> > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the
> old
> > >> protocol
> > >> > > > > > will be
> > >> > > > > > > > > able to
> > >> > > > > > > > > > > >> > switch to
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> explain
> > >> how that
> > >> > > > > > will work
> > >> > > > > > > > > > > >> though.
> > >> > > > > > > > > > > >> > It is
> > >> > > > > > > > > > > >> > > > > >> > > basically a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must
> be
> > >> > > > upgraded or
> > >> > > > > > > > > rolled to a
> > >> > > > > > > > > > > >> > software
> > >> > > > > > > > > > > >> > > > > >> > > supporting
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> coordinator.
> > >> Both the
> > >> > > > > > old and
> > >> > > > > > > > > the new
> > >> > > > > > > > > > > >> > > > > >> coordinator
> > >> > > > > > > > > > > >> > > > > >> > > will
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > >> protocol and
> > >> > > > rely
> > >> > > > > > on the
> > >> > > > > > > > > same
> > >> > > > > > > > > > > >> > persisted
> > >> > > > > > > > > > > >> > > > > >> > > metadata so
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> together.
> > >> This
> > >> > > > point
> > >> > > > > > is an
> > >> > > > > > > > > > > >> offline
> > >> > > > > > > > > > > >> > > > > >> migration.
> > >> > > > > > > > > > > >> > > > > >> > We
> > >> > > > > > > > > > > >> > > > > >> > > > > cannot
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > >> because it
> > >> > > > would
> > >> > > > > > require
> > >> > > > > > > > > > > >> > shutting down
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > current
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > >> starting up the
> > >> > > > > > new one
> > >> > > > > > > > > and
> > >> > > > > > > > > > > >> that
> > >> > > > > > > > > > > >> > would
> > >> > > > > > > > > > > >> > > > > >> cause
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > >> metadata
> > >> > > > > > version/IBP
> > >> > > > > > > > > must be
> > >> > > > > > > > > > > >> > upgraded
> > >> > > > > > > > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> X
> > >> > > > > > > > > > > >> > > > > >> > in
> > >> > > > > > > > > > > >> > > > > >> > > > > order
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> > >> protocol.
> > >> > > > This
> > >> > > > > > cannot
> > >> > > > > > > > > be done
> > >> > > > > > > > > > > >> > before
> > >> > > > > > > > > > > >> > > > > 1)
> > >> > > > > > > > > > > >> > > > > >> is
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because
> the
> > >> old
> > >> > > > > > coordinator
> > >> > > > > > > > > doesn't
> > >> > > > > > > > > > > >> > support
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers
> must be
> > >> > > > upgraded
> > >> > > > > > to a
> > >> > > > > > > > > version
> > >> > > > > > > > > > > >> > > > > supporting
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > online
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> > >> KIP-792).
> > >> > > > If
> > >> > > > > > the
> > >> > > > > > > > > consumer
> > >> > > > > > > > > > > >> is
> > >> > > > > > > > > > > >> > > > > already
> > >> > > > > > > > > > > >> > > > > >> > > there.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be
> done at
> > >> this
> > >> > > > point.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers
> must
> > >> be rolled
> > >> > > > > > with the
> > >> > > > > > > > > > > >> feature
> > >> > > > > > > > > > > >> > flag
> > >> > > > > > > > > > > >> > > > > >> turned
> > >> > > > > > > > > > > >> > > > > >> > > on.
> > >> > > > > > > > > > > >> > > > > >> > > > > The
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> > >> automatically
> > >> > > > > > > > > converted when
> > >> > > > > > > > > > > >> > the first
> > >> > > > > > > > > > > >> > > > > >> > > consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> protocol
> > >> joins
> > >> > > > the
> > >> > > > > > group.
> > >> > > > > > > > > While
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > members
> > >> > > > > > > > > > > >> > > > > >> > > using the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> being
> > >> > > > upgraded,
> > >> > > > > > the old
> > >> > > > > > > > > > > >> protocol
> > >> > > > > > > > > > > >> > is
> > >> > > > > > > > > > > >> > > > > >> proxied
> > >> > > > > > > > > > > >> > > > > >> > > into
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all
> of
> > >> this in
> > >> > > > the
> > >> > > > > > KIP.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy
> that we
> > >> are
> > >> > > > > > pushing the
> > >> > > > > > > > > pattern
> > >> > > > > > > > > > > >> > > > > >> subscriptions
> > >> > > > > > > > > > > >> > > > > >> > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
> > >> there
> > >> > > > could be
> > >> > > > > > some
> > >> > > > > > > > > tricky
> > >> > > > > > > > > > > >> > > > > >> compatibility
> > >> > > > > > > > > > > >> > > > > >> > > > > issues.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism
> for
> > >> users to
> > >> > > > > > detect
> > >> > > > > > > > > that
> > >> > > > > > > > > > > >> they
> > >> > > > > > > > > > > >> > need to
> > >> > > > > > > > > > > >> > > > > >> > update
> > >> > > > > > > > > > > >> > > > > >> > > > > their
> > >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching
> to
> > >> the new
> > >> > > > > > protocol?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a
> bit
> > >> more
> > >> > > > > > optimistic
> > >> > > > > > > > > than
> > >> > > > > > > > > > > >> you
> > >> > > > > > > > > > > >> > on this
> > >> > > > > > > > > > > >> > > > > >> > > point. I
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > >> majority of
> > >> > > > the
> > >> > > > > > cases
> > >> > > > > > > > > are
> > >> > > > > > > > > > > >> simple
> > >> > > > > > > > > > > >> > > > > regexes
> > >> > > > > > > > > > > >> > > > > >> > which
> > >> > > > > > > > > > > >> > > > > >> > > > > should
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> > >> engine. The
> > >> > > > > > > > > coordinator will
> > >> > > > > > > > > > > >> > verify
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > regex
> > >> > > > > > > > > > > >> > > > > >> > > > > anyway
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> consumer
> > >> if the
> > >> > > > > > regex is
> > >> > > > > > > > > not
> > >> > > > > > > > > > > >> valid.
> > >> > > > > > > > > > > >> > > > > Coming
> > >> > > > > > > > > > > >> > > > > >> > back
> > >> > > > > > > > > > > >> > > > > >> > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in
> the
> > >> worst
> > >> > > > case,
> > >> > > > > > the
> > >> > > > > > > > > first
> > >> > > > > > > > > > > >> > upgraded
> > >> > > > > > > > > > > >> > > > > >> > consumer
> > >> > > > > > > > > > > >> > > > > >> > > > > joining
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> > >> rejected.
> > >> > > > This
> > >> > > > > > should
> > >> > > > > > > > > be used
> > >> > > > > > > > > > > >> > as the
> > >> > > > > > > > > > > >> > > > > >> last
> > >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> customers to
> > >> > > > validate
> > >> > > > > > their
> > >> > > > > > > > > regex
> > >> > > > > > > > > > > >> > before
> > >> > > > > > > > > > > >> > > > > >> > upgrading
> > >> > > > > > > > > > > >> > > > > >> > > > > their
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to
> test
> > >> them with
> > >> > > > > > another
> > >> > > > > > > > > group.
> > >> > > > > > > > > > > >> For
> > >> > > > > > > > > > > >> > > > > >> instance,
> > >> > > > > > > > > > > >> > > > > >> > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> > >> pre-prod
> > >> > > > > > environment.
> > >> > > > > > > > > > > >> Another
> > >> > > > > > > > > > > >> > way
> > >> > > > > > > > > > > >> > > > > >> would be
> > >> > > > > > > > > > > >> > > > > >> > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > >> consumer-group tool
> > >> > > > to
> > >> > > > > > > > > provide a
> > >> > > > > > > > > > > >> regex
> > >> > > > > > > > > > > >> > > > > >> validation
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would
> this be
> > >> > > > enough in
> > >> > > > > > your
> > >> > > > > > > > > > > >> opinion?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the
> last
> > >> > > > question,
> > >> > > > > > will
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> Java
> > >> > > > > > > > > > > >> > client
> > >> > > > > > > > > > > >> > > > > >> allow
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > >> current regex
> > >> > > > > > engine for
> > >> > > > > > > > > > > >> > compatibility
> > >> > > > > > > > > > > >> > > > > >> > > reasons?
> > >> > > > > > > > > > > >> > > > > >> > > > > For
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
> > >> keep using
> > >> > > > > > client
> > >> > > > > > > > > based
> > >> > > > > > > > > > > >> > regex at
> > >> > > > > > > > > > > >> > > > > >> first
> > >> > > > > > > > > > > >> > > > > >> > to
> > >> > > > > > > > > > > >> > > > > >> > > keep
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple
> and
> > >> then
> > >> > > > > > migrate to
> > >> > > > > > > > > server
> > >> > > > > > > > > > > >> > based
> > >> > > > > > > > > > > >> > > > > >> regexes
> > >> > > > > > > > > > > >> > > > > >> > > as a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > second
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> point
> > >> but I
> > >> > > > am
> > >> > > > > > > > > concerned that
> > >> > > > > > > > > > > >> > this
> > >> > > > > > > > > > > >> > > > > would
> > >> > > > > > > > > > > >> > > > > >> > > allow
> > >> > > > > > > > > > > >> > > > > >> > > > > users
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
> > >> this mode.
> > >> > > > That
> > >> > > > > > > > > would go
> > >> > > > > > > > > > > >> > against our
> > >> > > > > > > > > > > >> > > > > >> goal
> > >> > > > > > > > > > > >> > > > > >> > > of
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> client
> > >> because
> > >> > > > we
> > >> > > > > > would
> > >> > > > > > > > > have to
> > >> > > > > > > > > > > >> > continue
> > >> > > > > > > > > > > >> > > > > >> > > monitoring
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> > >> client
> > >> > > > side. I
> > >> > > > > > would
> > >> > > > > > > > > rather
> > >> > > > > > > > > > > >> > not do
> > >> > > > > > > > > > > >> > > > > >> this.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say
> that
> > >> the group
> > >> > > > > > > > > coordinator
> > >> > > > > > > > > > > >> will
> > >> > > > > > > > > > > >> > be
> > >> > > > > > > > > > > >> > > > > >> > > responsible for
> > >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations
> > >> and that
> > >> > > > the
> > >> > > > > > > > > > > >> configurations
> > >> > > > > > > > > > > >> > will be
> > >> > > > > > > > > > > >> > > > > >> > > deleted
> > >> > > > > > > > > > > >> > > > > >> > > > > when
> > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
> > >> Will a
> > >> > > > > > transition to
> > >> > > > > > > > > DEAD
> > >> > > > > > > > > > > >> > trigger
> > >> > > > > > > > > > > >> > > > > >> > deletion
> > >> > > > > > > > > > > >> > > > > >> > > of
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > >> > > > configurations
> > >> > > > > > will be
> > >> > > > > > > > > > > >> deleted
> > >> > > > > > > > > > > >> > when
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > group is
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > >> together.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the
> choice to
> > >> store
> > >> > > > the
> > >> > > > > > > > > configs in
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > group
> > >> > > > > > > > > > > >> > > > > >> > > coordinator
> > >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> > >> cluster
> > >> > > > > > configs and
> > >> > > > > > > > > their
> > >> > > > > > > > > > > >> > values?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so.
> The
> > >> group
> > >> > > > > > > > > configurations are
> > >> > > > > > > > > > > >> > overrides
> > >> > > > > > > > > > > >> > > > > >> of
> > >> > > > > > > > > > > >> > > > > >> > > cluster
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you
> want to
> > >> know
> > >> > > > all
> > >> > > > > > the
> > >> > > > > > > > > overrides
> > >> > > > > > > > > > > >> > though,
> > >> > > > > > > > > > > >> > > > > you
> > >> > > > > > > > > > > >> > > > > >> > > would
> > >> > > > > > > > > > > >> > > > > >> > > > > have
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > >> > > > coordinators.
> > >> > > > > > You
> > >> > > > > > > > > cannot
> > >> > > > > > > > > > > >> > rely on
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > metadata
> > >> > > > > > > > > > > >> > > > > >> > > > > log
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> someone
> > >> > > > configure a
> > >> > > > > > group
> > >> > > > > > > > > before
> > >> > > > > > > > > > > >> > starting
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > >> allowing the
> > >> > > > > > explicit
> > >> > > > > > > > > creation
> > >> > > > > > > > > > > >> of
> > >> > > > > > > > > > > >> > > > > groups?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> > >> decoupled
> > >> > > > from
> > >> > > > > > the
> > >> > > > > > > > > group
> > >> > > > > > > > > > > >> > lifecycle.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will
> be
> > >> > > > > > automatically
> > >> > > > > > > > > created in
> > >> > > > > > > > > > > >> > this
> > >> > > > > > > > > > > >> > > > > case.
> > >> > > > > > > > > > > >> > > > > >> > > However,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
> > >> lost after
> > >> > > > the
> > >> > > > > > > > > retention
> > >> > > > > > > > > > > >> > period of
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > >> > > > Consumer.subscribe
> > >> > > > > > method
> > >> > > > > > > > > for the
> > >> > > > > > > > > > > >> > Java
> > >> > > > > > > > > > > >> > > > > >> client
> > >> > > > > > > > > > > >> > > > > >> > > still
> > >> > > > > > > > > > > >> > > > > >> > > > > take
> > >> > > > > > > > > > > >> > > > > >> > > > > > > a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >> `java.util.regex.Pattern` of
> > >> > > > do
> > >> > > > > > we
> > >> > > > > > > > > have to
> > >> > > > > > > > > > > >> > introduce
> > >> > > > > > > > > > > >> > > > > an
> > >> > > > > > > > > > > >> > > > > >> > > overload?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> > >> question. I
> > >> > > > > > forgot
> > >> > > > > > > > > about
> > >> > > > > > > > > > > >> that
> > >> > > > > > > > > > > >> > one.
> > >> > > > > > > > > > > >> > > > > As
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> `java.util.regex.Pattern` is
> > >> > > > not
> > >> > > > > > fully
> > >> > > > > > > > > > > >> compatible
> > >> > > > > > > > > > > >> > with
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > engine
> > >> > > > > > > > > > > >> > > > > >> > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> > >> might be
> > >> > > > better
> > >> > > > > > to
> > >> > > > > > > > > deprecate
> > >> > > > > > > > > > > >> > it and
> > >> > > > > > > > > > > >> > > > > >> use an
> > >> > > > > > > > > > > >> > > > > >> > > > > overload
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a
> string.
> > >> We would
> > >> > > > > > rely on
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> server
> > >> > > > > > > > > > > >> > side
> > >> > > > > > > > > > > >> > > > > >> > > validation.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> migration, I
> > >> think
> > >> > > > that
> > >> > > > > > we
> > >> > > > > > > > > could
> > >> > > > > > > > > > > >> still
> > >> > > > > > > > > > > >> > try to
> > >> > > > > > > > > > > >> > > > > >> > > toString
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it.
> That
> > >> should
> > >> > > > > > work, I
> > >> > > > > > > > > think, in
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> majority
> > >> > > > > > > > > > > >> > > > > >> > > of the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> > >> Justine
> > >> > > > that we
> > >> > > > > > > > > should be
> > >> > > > > > > > > > > >> > clearer
> > >> > > > > > > > > > > >> > > > > about
> > >> > > > > > > > > > > >> > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> > > > > reason
> > >> > > > > > > > > > > >> > > > > >> > > > > > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > >> > > > IBP/metadata.version
> > >> > > > > > from the
> > >> > > > > > > > > > > >> feature
> > >> > > > > > > > > > > >> > flag.
> > >> > > > > > > > > > > >> > > > > >> Maybe
> > >> > > > > > > > > > > >> > > > > >> > > we
> > >> > > > > > > > > > > >> > > > > >> > > > > mean
> > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > >> default for
> > >> > > > the
> > >> > > > > > feature
> > >> > > > > > > > > flag
> > >> > > > > > > > > > > >> to
> > >> > > > > > > > > > > >> > true
> > >> > > > > > > > > > > >> > > > > >> based
> > >> > > > > > > > > > > >> > > > > >> > on
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> once
> > >> we
> > >> > > > want to
> > >> > > > > > make
> > >> > > > > > > > > it the
> > >> > > > > > > > > > > >> > default.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> that
> > >> feature
> > >> > > > > > flag
> > >> > > > > > > > > mainly
> > >> > > > > > > > > > > >> > during the
> > >> > > > > > > > > > > >> > > > > >> > > development
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not
> have
> > >> > > > mentioned
> > >> > > > > > it, I
> > >> > > > > > > > > think,
> > >> > > > > > > > > > > >> > because
> > >> > > > > > > > > > > >> > > > > we
> > >> > > > > > > > > > > >> > > > > >> > > could
> > >> > > > > > > > > > > >> > > > > >> > > > > use
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config
> for
> > >> it.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> > >> protocol APIs
> > >> > > > > > don't
> > >> > > > > > > > > mention
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > required
> > >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > >> > > > > > > > > > > >> > > > > >> > > > > > > would
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that
> for
> > >> > > > consistency.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit
> odd
> > >> that
> > >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > >> > > > > > > > > > > >> > > > > requires
> > >> > > > > > > > > > > >> > > > > >> > > "Read
> > >> > > > > > > > > > > >> > > > > >> > > > > Group"
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems
> to do
> > >> more
> > >> > > > than
> > >> > > > > > > > > reading.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is
> how the
> > >> > > > current
> > >> > > > > > > > > protocol works
> > >> > > > > > > > > > > >> > though.
> > >> > > > > > > > > > > >> > > > > We
> > >> > > > > > > > > > > >> > > > > >> > only
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read
> Group" to
> > >> join a
> > >> > > > > > group. We
> > >> > > > > > > > > could
> > >> > > > > > > > > > > >> > consider
> > >> > > > > > > > > > > >> > > > > >> > > changing
> > >> > > > > > > > > > > >> > > > > >> > > > > this
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure
> that
> > >> it is
> > >> > > > worth
> > >> > > > > > it.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> > >> recreation
> > >> > > > > > handled by
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > consumer
> > >> > > > > > > > > > > >> > > > > with
> > >> > > > > > > > > > > >> > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It
> would be
> > >> good to
> > >> > > > > > have a
> > >> > > > > > > > > section
> > >> > > > > > > > > > > >> on
> > >> > > > > > > > > > > >> > this.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> protocol
> > >> > > > > > perspective, the
> > >> > > > > > > > > new
> > >> > > > > > > > > > > >> topic
> > >> > > > > > > > > > > >> > will
> > >> > > > > > > > > > > >> > > > > >> have a
> > >> > > > > > > > > > > >> > > > > >> > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> > >> treat it
> > >> > > > like a
> > >> > > > > > > > > topic with
> > >> > > > > > > > > > > >> a
> > >> > > > > > > > > > > >> > > > > different
> > >> > > > > > > > > > > >> > > > > >> > > name.
> > >> > > > > > > > > > > >> > > > > >> > > > > The
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that
> the
> > >> > > > fetch/commit
> > >> > > > > > > > > offsets
> > >> > > > > > > > > > > >> APIs
> > >> > > > > > > > > > > >> > do not
> > >> > > > > > > > > > > >> > > > > >> > support
> > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
> > >> would
> > >> > > > reuse the
> > >> > > > > > > > > offsets
> > >> > > > > > > > > > > >> based
> > >> > > > > > > > > > > >> > on the
> > >> > > > > > > > > > > >> > > > > >> > same.
> > >> > > > > > > > > > > >> > > > > >> > > I
> > >> > > > > > > > > > > >> > > > > >> > > > > think
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should
> update
> > >> those
> > >> > > > APIs
> > >> > > > > > as well
> > >> > > > > > > > > in
> > >> > > > > > > > > > > >> order
> > >> > > > > > > > > > > >> > to be
> > >> > > > > > > > > > > >> > > > > >> > > consistent
> > >> > > > > > > > > > > >> > > > > >> > > > > end
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> > >> strengthen
> > >> > > > the
> > >> > > > > > > > > semantics of
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> consumer.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> mentions
> > >> we will
> > >> > > > > > write the
> > >> > > > > > > > > new
> > >> > > > > > > > > > > >> > coordinator
> > >> > > > > > > > > > > >> > > > > >> in
> > >> > > > > > > > > > > >> > > > > >> > > Java.
> > >> > > > > > > > > > > >> > > > > >> > > > > Even
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > >> implementation
> > >> > > > > > detail, do
> > >> > > > > > > > > we plan
> > >> > > > > > > > > > > >> to
> > >> > > > > > > > > > > >> > have a
> > >> > > > > > > > > > > >> > > > > >> new
> > >> > > > > > > > > > > >> > > > > >> > > gradle
> > >> > > > > > > > > > > >> > > > > >> > > > > > > module
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> > >> scalability
> > >> > > > > > goal when
> > >> > > > > > > > > it
> > >> > > > > > > > > > > >> comes
> > >> > > > > > > > > > > >> > to how
> > >> > > > > > > > > > > >> > > > > >> many
> > >> > > > > > > > > > > >> > > > > >> > > > > members
> > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> > >> support?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have
> numbers
> > >> at the
> > >> > > > > > moment. The
> > >> > > > > > > > > > > >> protocol
> > >> > > > > > > > > > > >> > should
> > >> > > > > > > > > > > >> > > > > >> > > support
> > >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per
> group.
> > >> We will
> > >> > > > > > measure
> > >> > > > > > > > > this when
> > >> > > > > > > > > > > >> > we have
> > >> > > > > > > > > > > >> > > > > a
> > >> > > > > > > > > > > >> > > > > >> > first
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
> > >> that we
> > >> > > > might
> > >> > > > > > have
> > >> > > > > > > > > other
> > >> > > > > > > > > > > >> > > > > bottlenecks
> > >> > > > > > > > > > > >> > > > > >> > down
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> > >> commits).
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> consider
> > >> having
> > >> > > > > > > > > SubscribedTopidIds
> > >> > > > > > > > > > > >> > instead
> > >> > > > > > > > > > > >> > > > > of
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> SubscribedTopicNames
> > >> in
> > >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > >> > > > > > > > > > > >> > > > > >> Is
> > >> > > > > > > > > > > >> > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> > > > > idea
> > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> > >> resolve the
> > >> > > > > > regex on
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > server, we
> > >> > > > > > > > > > > >> > > > > >> can do
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > same
> > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name?
> The
> > >> > > > difference
> > >> > > > > > is that
> > >> > > > > > > > > > > >> sending
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> regex is
> > >> > > > > > > > > > > >> > > > > >> > > more
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending
> the
> > >> topic
> > >> > > > names
> > >> > > > > > is less
> > >> > > > > > > > > > > >> > efficient.
> > >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> easier
> > >> to
> > >> > > > handle
> > >> > > > > > if we
> > >> > > > > > > > > have
> > >> > > > > > > > > > > >> topic
> > >> > > > > > > > > > > >> > ids.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > >> consolidate the
> > >> > > > > > metadata
> > >> > > > > > > > > lookup
> > >> > > > > > > > > > > >> > on the
> > >> > > > > > > > > > > >> > > > > >> server
> > >> > > > > > > > > > > >> > > > > >> > > for
> > >> > > > > > > > > > > >> > > > > >> > > > > both
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
> > >> with your
> > >> > > > > > point. As
> > >> > > > > > > > > a
> > >> > > > > > > > > > > >> second
> > >> > > > > > > > > > > >> > > > > though,
> > >> > > > > > > > > > > >> > > > > >> > using
> > >> > > > > > > > > > > >> > > > > >> > > > > topic
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better
> here
> > >> for the
> > >> > > > > > delete and
> > >> > > > > > > > > > > >> > recreation
> > >> > > > > > > > > > > >> > > > > case.
> > >> > > > > > > > > > > >> > > > > >> > > Also, I
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
> > >> allow
> > >> > > > users to
> > >> > > > > > > > > subscribe
> > >> > > > > > > > > > > >> with
> > >> > > > > > > > > > > >> > topic
> > >> > > > > > > > > > > >> > > > > >> ids
> > >> > > > > > > > > > > >> > > > > >> > in
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that
> is
> > >> the only
> > >> > > > > > way to be
> > >> > > > > > > > > > > >> really
> > >> > > > > > > > > > > >> > robust
> > >> > > > > > > > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> > > topic
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12,
> 2022 at
> > >> 1:38 PM
> > >> > > > > > David
> > >> > > > > > > > > Jacot <
> > >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> > >> comments.
> > >> > > > Please
> > >> > > > > > find
> > >> > > > > > > > > my
> > >> > > > > > > > > > > >> answers
> > >> > > > > > > > > > > >> > > > > below.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > >> protocol
> > >> > > > relies on
> > >> > > > > > > > > topic IDs
> > >> > > > > > > > > > > >> > with the
> > >> > > > > > > > > > > >> > > > > >> > > exception
> > >> > > > > > > > > > > >> > > > > >> > > > > of the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based
> in
> > >> the
> > >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > >> > > > > > > > > > > >> > > > > >> I
> > >> > > > > > > > > > > >> > > > > >> > am
> > >> > > > > > > > > > > >> > > > > >> > > not
> > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> names
> > >> is the
> > >> > > > > > right call
> > >> > > > > > > > > here.
> > >> > > > > > > > > > > >> I
> > >> > > > > > > > > > > >> > need to
> > >> > > > > > > > > > > >> > > > > >> > think
> > >> > > > > > > > > > > >> > > > > >> > > > > about it
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > >> Obviously, the
> > >> > > > > > KIP does
> > >> > > > > > > > > not
> > >> > > > > > > > > > > >> > change the
> > >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to
> use
> > >> topic
> > >> > > > IDs.
> > >> > > > > > This
> > >> > > > > > > > > may be
> > >> > > > > > > > > > > >> > something
> > >> > > > > > > > > > > >> > > > > >> that
> > >> > > > > > > > > > > >> > > > > >> > we
> > >> > > > > > > > > > > >> > > > > >> > > > > should
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as
> it
> > >> would
> > >> > > > give
> > >> > > > > > better
> > >> > > > > > > > > > > >> overall
> > >> > > > > > > > > > > >> > > > > >> guarantee in
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> > >> think that
> > >> > > > I
> > >> > > > > > should
> > >> > > > > > > > > not
> > >> > > > > > > > > > > >> have
> > >> > > > > > > > > > > >> > > > > mentioned
> > >> > > > > > > > > > > >> > > > > >> > this
> > >> > > > > > > > > > > >> > > > > >> > > > > flag at
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove
> > >> it. We can
> > >> > > > > > use an
> > >> > > > > > > > > internal
> > >> > > > > > > > > > > >> > > > > >> configuration
> > >> > > > > > > > > > > >> > > > > >> > > while
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > >> feature.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> types
> > >> will be
> > >> > > > > > > > > supported. The
> > >> > > > > > > > > > > >> > change is
> > >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement
> is
> > >> that the
> > >> > > > > > cluster
> > >> > > > > > > > > uses
> > >> > > > > > > > > > > >> topic
> > >> > > > > > > > > > > >> > IDs.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11,
> 2022
> > >> at 9:53
> > >> > > > PM
> > >> > > > > > > > > Guozhang
> > >> > > > > > > > > > > >> Wang <
> > >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> > >> feedback.
> > >> > > > Here
> > >> > > > > > are
> > >> > > > > > > > > some
> > >> > > > > > > > > > > >> replies
> > >> > > > > > > > > > > >> > > > > inlined
> > >> > > > > > > > > > > >> > > > > >> > > below:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> 2022
> > >> at
> > >> > > > 2:53 AM
> > >> > > > > > > > > Ismael Juma
> > >> > > > > > > > > > > >> <
> > >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
> > >> KIP. This
> > >> > > > has
> > >> > > > > > the
> > >> > > > > > > > > > > >> potential
> > >> > > > > > > > > > > >> > to be a
> > >> > > > > > > > > > > >> > > > > >> > great
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > >> > > > questions/comments:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think
> it's
> > >> > > > premature to
> > >> > > > > > talk
> > >> > > > > > > > > about
> > >> > > > > > > > > > > >> > target
> > >> > > > > > > > > > > >> > > > > >> versions
> > >> > > > > > > > > > > >> > > > > >> > > for
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> > >> existing
> > >> > > > group
> > >> > > > > > > > > protocol.
> > >> > > > > > > > > > > >> > Unlike
> > >> > > > > > > > > > > >> > > > > >> KRaft,
> > >> > > > > > > > > > > >> > > > > >> > > this
> > >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> protocol
> > >> and hence
> > >> > > > > > > > > > > >> > deprecation/removal will
> > >> > > > > > > > > > > >> > > > > >> be
> > >> > > > > > > > > > > >> > > > > >> > > heavily
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> > >> applications
> > >> > > > > > migrate
> > >> > > > > > > > > to the
> > >> > > > > > > > > > > >> new
> > >> > > > > > > > > > > >> > > > > >> protocol.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree
> with
> > >> you. I
> > >> > > > think
> > >> > > > > > we can
> > >> > > > > > > > > > > >> remove
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> proposed
> > >> > > > > > > > > > > >> > > > > >> > > > > timeline
> > >> > > > > > > > > > > >> > > > > >> > > > > > > in
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > >> > > > Deprecation,
> > >> > > > > > and
> > >> > > > > > > > > Migration
> > >> > > > > > > > > > > >> > Plan` and
> > >> > > > > > > > > > > >> > > > > >> > > instead
> > >> > > > > > > > > > > >> > > > > >> > > > > just
> > >> > > > > > > > > > > >> > > > > >> > > > > > > state
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> decide
> > >> in the
> > >> > > > > > future
> > >> > > > > > > > > about
> > >> > > > > > > > > > > >> when
> > >> > > > > > > > > > > >> > we
> > >> > > > > > > > > > > >> > > > > would
> > >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > >> behaviors.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP
> says we
> > >> > > > intend to
> > >> > > > > > > > > release this
> > >> > > > > > > > > > > >> > in 4.x,
> > >> > > > > > > > > > > >> > > > > >> but
> > >> > > > > > > > > > > >> > > > > >> > it
> > >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > >> > > > > > > > > > > >> > > > > >> > > > > > > made
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> added
> > >> that as
> > >> > > > a
> > >> > > > > > way to
> > >> > > > > > > > > > > >> estimate
> > >> > > > > > > > > > > >> > when
> > >> > > > > > > > > > > >> > > > > we'd
> > >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > >> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > >> protocol, I
> > >> > > > also
> > >> > > > > > suggest
> > >> > > > > > > > > > > >> removing
> > >> > > > > > > > > > > >> > this
> > >> > > > > > > > > > > >> > > > > >> part.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's
> not
> > >> > > > specifically
> > >> > > > > > > > > related to
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> more
> > >> for
> > >> > > > client
> > >> > > > > > > > > upgrades.
> > >> > > > > > > > > > > >> I.e.
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> > > broker-side
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation
> may
> > >> be done
> > >> > > > > > first,
> > >> > > > > > > > > and then
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > client
> > >> > > > > > > > > > > >> > > > > >> > side,
> > >> > > > > > > > > > > >> > > > > >> > > > > and we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> > >> "released"
> > >> > > > by
> > >> > > > > > the
> > >> > > > > > > > > time
> > >> > > > > > > > > > > >> clients
> > >> > > > > > > > > > > >> > > > > >> > > implementations
> > >> > > > > > > > > > > >> > > > > >> > > > > are
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> > >> enable the
> > >> > > > > > feature the
> > >> > > > > > > > > > > >> clients
> > >> > > > > > > > > > > >> > need to
> > >> > > > > > > > > > > >> > > > > >> > first
> > >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> > >> rolling
> > >> > > > bounce
> > >> > > > > > and
> > >> > > > > > > > > then set
> > >> > > > > > > > > > > >> > the flag
> > >> > > > > > > > > > > >> > > > > >> > with a
> > >> > > > > > > > > > > >> > > > > >> > > > > second
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and
> hence
> > >> we feel
> > >> > > > it's
> > >> > > > > > > > > better to be
> > >> > > > > > > > > > > >> > released
> > >> > > > > > > > > > > >> > > > > >> in a
> > >> > > > > > > > > > > >> > > > > >> > > major
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> > >> flesh out
> > >> > > > the
> > >> > > > > > > > > details of
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > migration
> > >> > > > > > > > > > > >> > > > > >> > > story.
> > >> > > > > > > > > > > >> > > > > >> > > > > It
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying
> we
> > >> will
> > >> > > > support
> > >> > > > > > online
> > >> > > > > > > > > > > >> > migrations. Is
> > >> > > > > > > > > > > >> > > > > >> that
> > >> > > > > > > > > > > >> > > > > >> > > > > correct?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > We
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this
> in
> > >> detail.
> > >> > > > It
> > >> > > > > > could
> > >> > > > > > > > > also be
> > >> > > > > > > > > > > >> > done as a
> > >> > > > > > > > > > > >> > > > > >> > > separate
> > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> that's
> > >> the
> > >> > > > part we
> > >> > > > > > can
> > >> > > > > > > > > be more
> > >> > > > > > > > > > > >> > concrete
> > >> > > > > > > > > > > >> > > > > >> about
> > >> > > > > > > > > > > >> > > > > >> > > for
> > >> > > > > > > > > > > >> > > > > >> > > > > sure
> > >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related
> to
> > >> your
> > >> > > > > > question 2)
> > >> > > > > > > > > above).
> > >> > > > > > > > > > > >> > We will
> > >> > > > > > > > > > > >> > > > > >> work
> > >> > > > > > > > > > > >> > > > > >> > on
> > >> > > > > > > > > > > >> > > > > >> > > > > making
> > >> > > > > > > > > > > >> > > > > >> > > > > > > it
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > >> parallel as we
> > >> > > > > > solicit
> > >> > > > > > > > > more
> > >> > > > > > > > > > > >> > feedback.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
> > >> that we are
> > >> > > > > > pushing
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > pattern
> > >> > > > > > > > > > > >> > > > > >> > > subscriptions
> > >> > > > > > > > > > > >> > > > > >> > > > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> like
> > >> there
> > >> > > > > > could be
> > >> > > > > > > > > some
> > >> > > > > > > > > > > >> tricky
> > >> > > > > > > > > > > >> > > > > >> > > compatibility
> > >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> mechanism
> > >> for
> > >> > > > users to
> > >> > > > > > > > > detect that
> > >> > > > > > > > > > > >> > they
> > >> > > > > > > > > > > >> > > > > need
> > >> > > > > > > > > > > >> > > > > >> to
> > >> > > > > > > > > > > >> > > > > >> > > update
> > >> > > > > > > > > > > >> > > > > >> > > > > > > their
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> switching
> > >> to the
> > >> > > > new
> > >> > > > > > > > > protocol?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> need
> > >> some
> > >> > > > > > tooling for
> > >> > > > > > > > > > > >> non-java
> > >> > > > > > > > > > > >> > client
> > >> > > > > > > > > > > >> > > > > >> > users
> > >> > > > > > > > > > > >> > > > > >> > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > sort
> > >> > > > > > > > > > > >> > > > > >> > > > > > > of
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> > >> client before
> > >> > > > > > > > > switching to
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > new
> > >> > > > > > > > > > > >> > > > > >> > protocol.
> > >> > > > > > > > > > > >> > > > > >> > > I
> > >> > > > > > > > > > > >> > > > > >> > > > > do not
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on
> > >> top of my
> > >> > > > head
> > >> > > > > > > > > though,
> > >> > > > > > > > > > > >> maybe
> > >> > > > > > > > > > > >> > others
> > >> > > > > > > > > > > >> > > > > >> like
> > >> > > > > > > > > > > >> > > > > >> > > @Matt
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to
> the
> > >> last
> > >> > > > > > question,
> > >> > > > > > > > > will the
> > >> > > > > > > > > > > >> > Java
> > >> > > > > > > > > > > >> > > > > >> client
> > >> > > > > > > > > > > >> > > > > >> > > allow
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> > >> current
> > >> > > > regex
> > >> > > > > > > > > engine for
> > >> > > > > > > > > > > >> > > > > >> compatibility
> > >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > For
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be
> handy
> > >> to keep
> > >> > > > using
> > >> > > > > > > > > client
> > >> > > > > > > > > > > >> based
> > >> > > > > > > > > > > >> > regex
> > >> > > > > > > > > > > >> > > > > at
> > >> > > > > > > > > > > >> > > > > >> > > first to
> > >> > > > > > > > > > > >> > > > > >> > > > > keep
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> simple
> > >> and
> > >> > > > then
> > >> > > > > > migrate
> > >> > > > > > > > > to
> > >> > > > > > > > > > > >> > server based
> > >> > > > > > > > > > > >> > > > > >> > > regexes
> > >> > > > > > > > > > > >> > > > > >> > > > > as a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have
> not
> > >> thought
> > >> > > > > > about
> > >> > > > > > > > > that for
> > >> > > > > > > > > > > >> > java
> > >> > > > > > > > > > > >> > > > > >> clients,
> > >> > > > > > > > > > > >> > > > > >> > > and
> > >> > > > > > > > > > > >> > > > > >> > > > > we can
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind
> of
> > >> > > > > > compatibility
> > >> > > > > > > > > issues do
> > >> > > > > > > > > > > >> > you have
> > >> > > > > > > > > > > >> > > > > in
> > >> > > > > > > > > > > >> > > > > >> > > mind?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
> > >> that the
> > >> > > > group
> > >> > > > > > > > > coordinator
> > >> > > > > > > > > > > >> > will be
> > >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > >> configurations and
> > >> > > > that
> > >> > > > > > the
> > >> > > > > > > > > > > >> > configurations
> > >> > > > > > > > > > > >> > > > > will
> > >> > > > > > > > > > > >> > > > > >> be
> > >> > > > > > > > > > > >> > > > > >> > > > > deleted
> > >> > > > > > > > > > > >> > > > > >> > > > > > > when
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> deleted.
> > >> Will a
> > >> > > > > > > > > transition to
> > >> > > > > > > > > > > >> DEAD
> > >> > > > > > > > > > > >> > > > > trigger
> > >> > > > > > > > > > > >> > > > > >> > > deletion
> > >> > > > > > > > > > > >> > > > > >> > > > > of
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> configurations?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> DEAD
> > >> state
> > >> > > > is an
> > >> > > > > > > > > ending
> > >> > > > > > > > > > > >> state
> > >> > > > > > > > > > > >> > (we
> > >> > > > > > > > > > > >> > > > > would
> > >> > > > > > > > > > > >> > > > > >> > only
> > >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> > >> group is
> > >> > > > EMPTY
> > >> > > > > > and
> > >> > > > > > > > > also all
> > >> > > > > > > > > > > >> of
> > >> > > > > > > > > > > >> > its
> > >> > > > > > > > > > > >> > > > > >> > metadata
> > >> > > > > > > > > > > >> > > > > >> > > are
> > >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited
> to
> > >> DEAD this
> > >> > > > > > group
> > >> > > > > > > > > would
> > >> > > > > > > > > > > >> never
> > >> > > > > > > > > > > >> > be
> > >> > > > > > > > > > > >> > > > > >> revived.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> > >> choice to
> > >> > > > store
> > >> > > > > > the
> > >> > > > > > > > > configs
> > >> > > > > > > > > > > >> in
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > group
> > >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to
> list all
> > >> > > > cluster
> > >> > > > > > configs
> > >> > > > > > > > > and
> > >> > > > > > > > > > > >> their
> > >> > > > > > > > > > > >> > > > > values?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> > >> question,
> > >> > > > and our
> > >> > > > > > > > > thoughts
> > >> > > > > > > > > > > >> are
> > >> > > > > > > > > > > >> > that
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > so-called
> > >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations"
> are
> > >> > > > overrides
> > >> > > > > > of the
> > >> > > > > > > > > > > >> > cluster-level
> > >> > > > > > > > > > > >> > > > > >> > > > > configurations
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> > >> group so
> > >> > > > when an
> > >> > > > > > > > > admin list
> > >> > > > > > > > > > > >> > cluster
> > >> > > > > > > > > > > >> > > > > >> > configs
> > >> > > > > > > > > > > >> > > > > >> > > it's
> > >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> > >> cluster-level
> > >> > > > > > > > > "defaults", not
> > >> > > > > > > > > > > >> > showing
> > >> > > > > > > > > > > >> > > > > >> any
> > >> > > > > > > > > > > >> > > > > >> > > > > per-group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> > >> someone
> > >> > > > > > configure a
> > >> > > > > > > > > group
> > >> > > > > > > > > > > >> > before
> > >> > > > > > > > > > > >> > > > > >> starting
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> > >> allowing
> > >> > > > the
> > >> > > > > > explicit
> > >> > > > > > > > > > > >> > creation of
> > >> > > > > > > > > > > >> > > > > >> groups?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could
> be
> > >> > > > decoupled
> > >> > > > > > from
> > >> > > > > > > > > the group
> > >> > > > > > > > > > > >> > > > > lifecycle.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can
> be
> > >> created
> > >> > > > > > before
> > >> > > > > > > > > the group
> > >> > > > > > > > > > > >> > itself
> > >> > > > > > > > > > > >> > > > > as
> > >> > > > > > > > > > > >> > > > > >> an
> > >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course,
> this
> > >> > > > requires
> > >> > > > > > the
> > >> > > > > > > > > > > >> corresponding
> > >> > > > > > > > > > > >> > > > > >> request to
> > >> > > > > > > > > > > >> > > > > >> > > be
> > >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> coordinator
> > >> based on
> > >> > > > the
> > >> > > > > > group
> > >> > > > > > > > > id ---
> > >> > > > > > > > > > > >> > the only
> > >> > > > > > > > > > > >> > > > > >> > thing
> > >> > > > > > > > > > > >> > > > > >> > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> > >> itself is
> > >> > > > gone
> > >> > > > > > we
> > >> > > > > > > > > also check
> > >> > > > > > > > > > > >> > if there
> > >> > > > > > > > > > > >> > > > > >> are
> > >> > > > > > > > > > > >> > > > > >> > > any
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> related to
> > >> that
> > >> > > > group
> > >> > > > > > and
> > >> > > > > > > > > delete
> > >> > > > > > > > > > > >> as
> > >> > > > > > > > > > > >> > well.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> > >> indeed
> > >> > > > > > introduces an
> > >> > > > > > > > > > > >> > asymmetry on
> > >> > > > > > > > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > creation
> > >> > > > > > > > > > > >> > > > > >> > > > > /
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of
> the
> > >> config
> > >> > > > > > entities,
> > >> > > > > > > > > and we
> > >> > > > > > > > > > > >> > would like
> > >> > > > > > > > > > > >> > > > > >> to
> > >> > > > > > > > > > > >> > > > > >> > > hear
> > >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> whether we
> > >> should
> > >> > > > aim
> > >> > > > > > for
> > >> > > > > > > > > symmetry
> > >> > > > > > > > > > > >> > i.e.
> > >> > > > > > > > > > > >> > > > > >> totally
> > >> > > > > > > > > > > >> > > > > >> > > > > decouple
> > >> > > > > > > > > > > >> > > > > >> > > > > > > group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and
> hence
> > >> not
> > >> > > > delete
> > >> > > > > > them at
> > >> > > > > > > > > all
> > >> > > > > > > > > > > >> when
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> group
> > >> > > > > > > > > > > >> > > > > >> > is
> > >> > > > > > > > > > > >> > > > > >> > > > > gone,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > but
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> > >> deletion
> > >> > > > > > operations
> > >> > > > > > > > > by
> > >> > > > > > > > > > > >> > themselves.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > >> > > > > > Consumer.subscribe
> > >> > > > > > > > > method for
> > >> > > > > > > > > > > >> > the Java
> > >> > > > > > > > > > > >> > > > > >> > client
> > >> > > > > > > > > > > >> > > > > >> > > > > still
> > >> > > > > > > > > > > >> > > > > >> > > > > > > take
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > `java.util.regex.Pattern` of
> > >> > > > > > do we
> > >> > > > > > > > > have
> > >> > > > > > > > > > > >> to
> > >> > > > > > > > > > > >> > > > > >> introduce an
> > >> > > > > > > > > > > >> > > > > >> > > > > overload?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do
> not
> > >> need to
> > >> > > > > > introduce
> > >> > > > > > > > > an
> > >> > > > > > > > > > > >> > overload, but
> > >> > > > > > > > > > > >> > > > > >> I'm
> > >> > > > > > > > > > > >> > > > > >> > > all
> > >> > > > > > > > > > > >> > > > > >> > > > > ears
> > >> > > > > > > > > > > >> > > > > >> > > > > > > if
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > >> compatibility
> > >> > > > > > issues
> > >> > > > > > > > > that we
> > >> > > > > > > > > > > >> may
> > >> > > > > > > > > > > >> > > > > overlook.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree
> with
> > >> Justine
> > >> > > > > > that we
> > >> > > > > > > > > should
> > >> > > > > > > > > > > >> be
> > >> > > > > > > > > > > >> > clearer
> > >> > > > > > > > > > > >> > > > > >> > about
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > >> > > > > > IBP/metadata.version
> > >> > > > > > > > > from the
> > >> > > > > > > > > > > >> > feature
> > >> > > > > > > > > > > >> > > > > >> flag.
> > >> > > > > > > > > > > >> > > > > >> > > Maybe
> > >> > > > > > > > > > > >> > > > > >> > > > > we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> > >> default
> > >> > > > for
> > >> > > > > > the
> > >> > > > > > > > > feature
> > >> > > > > > > > > > > >> flag
> > >> > > > > > > > > > > >> > to
> > >> > > > > > > > > > > >> > > > > true
> > >> > > > > > > > > > > >> > > > > >> > > based on
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> metadata.version
> > >> once we
> > >> > > > > > want to
> > >> > > > > > > > > make it
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> default.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> > >> protocol
> > >> > > > APIs
> > >> > > > > > don't
> > >> > > > > > > > > mention
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > >> required
> > >> > > > > > > > > > > >> > > > > >> > > ACLs,
> > >> > > > > > > > > > > >> > > > > >> > > > > it
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add
> that
> > >> for
> > >> > > > > > consistency.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > >> certainly do
> > >> > > > that.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a
> bit
> > >> odd that
> > >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > >> > > > > > > > > > > >> > > > > >> > requires
> > >> > > > > > > > > > > >> > > > > >> > > > > "Read
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it
> seems
> > >> to do
> > >> > > > more
> > >> > > > > > than
> > >> > > > > > > > > reading.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> thought
> > >> myself
> > >> > > > as
> > >> > > > > > well,
> > >> > > > > > > > > but in
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > end we
> > >> > > > > > > > > > > >> > > > > >> > could
> > >> > > > > > > > > > > >> > > > > >> > > not
> > >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> alternative:
> > >> adding
> > >> > > > > > Write
> > >> > > > > > > > > Group
> > >> > > > > > > > > > > >> seems
> > >> > > > > > > > > > > >> > an
> > >> > > > > > > > > > > >> > > > > >> overkill
> > >> > > > > > > > > > > >> > > > > >> > > here
> > >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > >> elsewhere (we
> > >> > > > only
> > >> > > > > > have
> > >> > > > > > > > > Read /
> > >> > > > > > > > > > > >> > Delete
> > >> > > > > > > > > > > >> > > > > and
> > >> > > > > > > > > > > >> > > > > >> > > Describe
> > >> > > > > > > > > > > >> > > > > >> > > > > on
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would
> like to
> > >> hear
> > >> > > > others
> > >> > > > > > > > > thoughts.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> topic
> > >> > > > recreation
> > >> > > > > > > > > handled by
> > >> > > > > > > > > > > >> the
> > >> > > > > > > > > > > >> > > > > consumer
> > >> > > > > > > > > > > >> > > > > >> > with
> > >> > > > > > > > > > > >> > > > > >> > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> > >> would be
> > >> > > > good
> > >> > > > > > to have
> > >> > > > > > > > > a
> > >> > > > > > > > > > > >> > section on
> > >> > > > > > > > > > > >> > > > > >> this.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with
> regex
> > >> > > > > > subscription
> > >> > > > > > > > > right? Yes
> > >> > > > > > > > > > > >> > we can
> > >> > > > > > > > > > > >> > > > > >> add a
> > >> > > > > > > > > > > >> > > > > >> > > > > section
> > >> > > > > > > > > > > >> > > > > >> > > > > > > about
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> basically
> > >> the
> > >> > > > idea
> > >> > > > > > is that
> > >> > > > > > > > > > > >> consumer
> > >> > > > > > > > > > > >> > would
> > >> > > > > > > > > > > >> > > > > be
> > >> > > > > > > > > > > >> > > > > >> > > totally
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> protocol as
> > >> it's
> > >> > > > > > handled all
> > >> > > > > > > > > by the
> > >> > > > > > > > > > > >> > brokers.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> > >> mentions we
> > >> > > > will
> > >> > > > > > write
> > >> > > > > > > > > the
> > >> > > > > > > > > > > >> new
> > >> > > > > > > > > > > >> > > > > >> coordinator
> > >> > > > > > > > > > > >> > > > > >> > in
> > >> > > > > > > > > > > >> > > > > >> > > > > Java.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > >> implementation
> > >> > > > > > detail,
> > >> > > > > > > > > do we
> > >> > > > > > > > > > > >> > plan to
> > >> > > > > > > > > > > >> > > > > >> have a
> > >> > > > > > > > > > > >> > > > > >> > > new
> > >> > > > > > > > > > > >> > > > > >> > > > > gradle
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> thought
> > >> about
> > >> > > > > > that. But
> > >> > > > > > > > > I think
> > >> > > > > > > > > > > >> > the
> > >> > > > > > > > > > > >> > > > > answer
> > >> > > > > > > > > > > >> > > > > >> > > should
> > >> > > > > > > > > > > >> > > > > >> > > > > be
> > >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we
> have a
> > >> > > > scalability
> > >> > > > > > goal
> > >> > > > > > > > > when it
> > >> > > > > > > > > > > >> > comes to
> > >> > > > > > > > > > > >> > > > > >> how
> > >> > > > > > > > > > > >> > > > > >> > > many
> > >> > > > > > > > > > > >> > > > > >> > > > > > > members
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> protocol can
> > >> > > > support?
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group,
> I
> > >> think we
> > >> > > > > > should
> > >> > > > > > > > > shoot for
> > >> > > > > > > > > > > >> > 1000s of
> > >> > > > > > > > > > > >> > > > > >> > > members.
> > >> > > > > > > > > > > >> > > > > >> > > > > But
> > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability
> goals
> > >> also
> > >> > > > depend
> > >> > > > > > on the
> > >> > > > > > > > > offset
> > >> > > > > > > > > > > >> > > > > management
> > >> > > > > > > > > > > >> > > > > >> > > (commit,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of
> the
> > >> > > > coordinator
> > >> > > > > > > > > which we
> > >> > > > > > > > > > > >> did
> > >> > > > > > > > > > > >> > not
> > >> > > > > > > > > > > >> > > > > >> cover in
> > >> > > > > > > > > > > >> > > > > >> > > this
> > >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> > >> number that
> > >> > > > > > applies
> > >> > > > > > > > > > > >> > universally.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > >> consider
> > >> > > > having
> > >> > > > > > > > > > > >> > SubscribedTopidIds
> > >> > > > > > > > > > > >> > > > > >> instead
> > >> > > > > > > > > > > >> > > > > >> > of
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> SubscribedTopicNames in
> > >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > >> > > > > > > > > > > >> > > > > >> > > Is the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have
> to
> > >> resolve
> > >> > > > the
> > >> > > > > > regex
> > >> > > > > > > > > on the
> > >> > > > > > > > > > > >> > server,
> > >> > > > > > > > > > > >> > > > > we
> > >> > > > > > > > > > > >> > > > > >> > can
> > >> > > > > > > > > > > >> > > > > >> > > do
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > same
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic
> name?
> > >> The
> > >> > > > > > difference is
> > >> > > > > > > > > that
> > >> > > > > > > > > > > >> > sending the
> > >> > > > > > > > > > > >> > > > > >> > regex
> > >> > > > > > > > > > > >> > > > > >> > > is
> > >> > > > > > > > > > > >> > > > > >> > > > > more
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas
> sending
> > >> the topic
> > >> > > > > > names is
> > >> > > > > > > > > less
> > >> > > > > > > > > > > >> > efficient.
> > >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> > >> easier to
> > >> > > > > > handle if
> > >> > > > > > > > > we have
> > >> > > > > > > > > > > >> > topic
> > >> > > > > > > > > > > >> > > > > ids.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason
> to
> > >> still
> > >> > > > let
> > >> > > > > > the
> > >> > > > > > > > > clients
> > >> > > > > > > > > > > >> send
> > >> > > > > > > > > > > >> > names
> > >> > > > > > > > > > > >> > > > > >> is to
> > >> > > > > > > > > > > >> > > > > >> > > keep
> > >> > > > > > > > > > > >> > > > > >> > > > > the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> names
> > >> -> ids
> > >> > > > on
> > >> > > > > > the
> > >> > > > > > > > > broker /
> > >> > > > > > > > > > > >> > admin
> > >> > > > > > > > > > > >> > > > > client
> > >> > > > > > > > > > > >> > > > > >> > > only.
> > >> > > > > > > > > > > >> > > > > >> > > > > Note
> > >> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we
> added
> > >> topic id
> > >> > > > in
> > >> > > > > > > > > KIP-516, we
> > >> > > > > > > > > > > >> > never
> > >> > > > > > > > > > > >> > > > > >> > > implemented the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> consumer/producers
> > >> > > > leveraging
> > >> > > > > > the
> > >> > > > > > > > > related
> > >> > > > > > > > > > > >> > newer
> > >> > > > > > > > > > > >> > > > > >> versioned
> > >> > > > > > > > > > > >> > > > > >> > > RPCs,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> topic
> > >> id as
> > >> > > > empty
> > >> > > > > > UUID.
> > >> > > > > > > > > We
> > >> > > > > > > > > > > >> want
> > >> > > > > > > > > > > >> > to keep
> > >> > > > > > > > > > > >> > > > > >> the
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and
> only
> > >> > > > delegate
> > >> > > > > > the
> > >> > > > > > > > > reasoning
> > >> > > > > > > > > > > >> on
> > >> > > > > > > > > > > >> > broker
> > >> > > > > > > > > > > >> > > > > >> and
> > >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
> > >> 2022 at
> > >> > > > 10:45
> > >> > > > > > AM
> > >> > > > > > > > > David
> > >> > > > > > > > > > > >> Jacot
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> <dj...@confluent.io.invalid>
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would
> like to
> > >> start a
> > >> > > > > > > > > discussion
> > >> > > > > > > > > > > >> thread
> > >> > > > > > > > > > > >> > on
> > >> > > > > > > > > > > >> > > > > >> KIP-848:
> > >> > > > > > > > > > > >> > > > > >> > > The
> > >> > > > > > > > > > > >> > > > > >> > > > > Next
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation
> of
> > >> the
> > >> > > > Consumer
> > >> > > > > > > > > Rebalance
> > >> > > > > > > > > > > >> > Protocol.
> > >> > > > > > > > > > > >> > > > > >> With
> > >> > > > > > > > > > > >> > > > > >> > > this
> > >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > we
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> > >> rebalance
> > >> > > > > > protocol
> > >> > > > > > > > > (for
> > >> > > > > > > > > > > >> > consumers)
> > >> > > > > > > > > > > >> > > > > >> more
> > >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > more
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> > >> easier to
> > >> > > > > > implement for
> > >> > > > > > > > > > > >> > clients, and
> > >> > > > > > > > > > > >> > > > > >> easier
> > >> > > > > > > > > > > >> > > > > >> > > to
> > >> > > > > > > > > > > >> > > > > >> > > > > debug
> > >> > > > > > > > > > > >> > > > > >> > > > > > > for
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> here:
> > >> > > > > > > > > > > >> > > > > >> > > > >
> > >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take
> a
> > >> look and
> > >> > > > let
> > >> > > > > > me
> > >> > > > > > > > > know what
> > >> > > > > > > > > > > >> > you
> > >> > > > > > > > > > > >> > > > > think.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will
> be
> > >> away from
> > >> > > > > > July
> > >> > > > > > > > > 18th to
> > >> > > > > > > > > > > >> > August 8th.
> > >> > > > > > > > > > > >> > > > > >> That
> > >> > > > > > > > > > > >> > > > > >> > > gives
> > >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> read
> > >> and
> > >> > > > digest
> > >> > > > > > this
> > >> > > > > > > > > long
> > >> > > > > > > > > > > >> KIP.
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > >> > > > > > > > > > > >> > > > > >> > > > > > >
> > >> > > > > > > > > > > >> > > > > >> > > > >
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> > >
> > >> > > > > > > > > > > >> > > > > >> >
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > > >> --
> > >> > > > > > > > > > > >> > > > > >> -- Guozhang
> > >> > > > > > > > > > > >> > > > > >>
> > >> > > > > > > > > > > >> > > > > >
> > >> > > > > > > > > > > >> > > > >
> > >> > > > > > > > > > > >> >
> > >> > > > > > > > > > > >>
> > >> > > > > > > > > > > >
> > >> > > > > > > > >
> > >> > > > > >
> > >> > > > >
> > >> > > > >
> > >> > > > > --
> > >> > > > > -- Guozhang
> > >> > > >
> > >> > >
> > >> > >
> > >> > > --
> > >> > > -- Guozhang
> > >>
> > >
> > >
> > > --
> > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Guozhang,

> 1) The client-side "PartitionAssignor#Assignment" has an Error byte field,
> while the broker-side "PartitionAssignor#Assignment" does not. And the
> rationale seems to be assuming that we should always be able to do the
> assignment at the broker-side assignor without errors. Personally I think
> it's still potentially beneficial to add the Error field even for
> broker-side assignors, e.g. for some edge cases where some subscribed
> topics are not recognized with the current broker's metadata. What do you
> think?

Yeah, that seems reasonable. However, I wonder if it would be better
to use an exception on the server side. This is what we usually do for
server side plugins. On the client side, we use a field because the
errors are not defined in advance.

Your comment also makes me think about what we should do when the
server side assignor fails. I suppose that we have to keep the current
assignment until a new event occurs. For instance, in your example,
the coordinator would have to trigger a rebalance when unrecognized
topics are available. This would be part of the metadata monitoring.

> 2) The client-side "GroupMember" has three additional fields
> reason/version/metadata compared with the broker-side GroupMember. I agree
> that broker-side assignor would not need reason/metadata since they are
> blackbox strings/bytes to the assignor, but what about version? E.g. is it
> possible that we evolve our broker-side built-in assignor but the old
> versioned clients would not be able to work with the new version, in which
> case we need to let the broker being aware of this and upgrade its behavior
> to cooperate with the clients?

I don't really see the benefits here because server side assignors
don't have metadata at all. They only assign topic-partitions. They
are not supposed to generate metadata nor to receive metadata from the
members.

> 3) Also related to 2) above, for the client-side "GroupMember", instead of
> including these three fields, what about just adding the "Metadata" field
> class which has these three fields? Also, there are two "Metadata"
> currently in the APIs, the first is a class that encodes
> reason/version/metadata, and the second is just the encoded metadata bytes.
> I'm wondering what about just naming the first as memberMetadata, which
> then has a bytebuffer field Metadata, or instead naming the second
> bytebuffer field as metadataBytes?

That's a good point. Let me try to rationalize this interface based on
your suggestions.

Best,
David

On Tue, Sep 13, 2022 at 9:21 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hello David,
>
> Just had a few more nit questions about the public APIs:
>
> 1) The client-side "PartitionAssignor#Assignment" has an Error byte field,
> while the broker-side "PartitionAssignor#Assignment" does not. And the
> rationale seems to be assuming that we should always be able to do the
> assignment at the broker-side assignor without errors. Personally I think
> it's still potentially beneficial to add the Error field even for
> broker-side assignors, e.g. for some edge cases where some subscribed
> topics are not recognized with the current broker's metadata. What do you
> think?
>
> 2) The client-side "GroupMember" has three additional fields
> reason/version/metadata compared with the broker-side GroupMember. I agree
> that broker-side assignor would not need reason/metadata since they are
> blackbox strings/bytes to the assignor, but what about version? E.g. is it
> possible that we evolve our broker-side built-in assignor but the old
> versioned clients would not be able to work with the new version, in which
> case we need to let the broker being aware of this and upgrade its behavior
> to cooperate with the clients?
>
> 3) Also related to 2) above, for the client-side "GroupMember", instead of
> including these three fields, what about just adding the "Metadata" field
> class which has these three fields? Also, there are two "Metadata"
> currently in the APIs, the first is a class that encodes
> reason/version/metadata, and the second is just the encoded metadata bytes.
> I'm wondering what about just naming the first as memberMetadata, which
> then has a bytebuffer field Metadata, or instead naming the second
> bytebuffer field as metadataBytes?
>
> Guozhang
>
>
>
>
>
>
>
> On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> > Hello David,
> >
> > Thanks for bringing this question up. I think the main benefits as you
> > listed is 2) above if it stays; just to clarify, we would only be able to
> > save one round trip if the rebalance is still triggered by the broker; if
> > the rebalance is triggered by the client then the num.round.trips are the
> > same:
> >
> > 1) With GroupPrepareAssignment:
> >
> > T0: client decides to do a new assignment, suppose it has already sent a
> > HB and hence has to wait for it to return first since only one request /
> > response can be inflight with the coordinator's socket.
> > T1: client receives the HB response, and then sends the
> > GroupPrepareAssignment request.
> > T2: the GroupPrepareAssignment response is returned.
> > T3: it calculates the new assignment, and sends a GroupInstallAssignment
> > request.
> >
> > In total, two round trips.
> >
> > 2) Without GroupPrepareAssignment:
> >
> > T0: client decides to do a new assignment, suppose it has already sent a
> > HB and hence has to wait for it to return first since only one request /
> > response can be inflight with the coordinator's socket.
> > T1: client receives the HB response, and then sends the new HB request
> > with the flag indicating a new rebalance needed..
> > T2: the HB response with the optional member metadata map is returned.
> > T3: it calculates the new assignment, and sends a GroupInstallAssignment
> > request.
> >
> > In total, two round trips as well.
> >
> > -----------------------------
> >
> > So to complete the full picture here, we'd need to modify both HB request
> > and response so that the client can also indicate a new rebalance via the
> > HB request as well, right?
> >
> > Assuming all above is true, I think it's okay to merge the
> > GroupPrepareAssignment into HB given that we can make the additional fields
> > encoding the full member (subscription) metadata and topic metadata as
> > optional fields.
> >
> >
> > Guozhang
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > On Mon, Sep 12, 2022 at 5:22 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> >> Hi all,
> >>
> >> During an offline conversation, someone asked why we need the
> >> ConsumerGroupPrepareAssignment API and suggested that we could instead
> >> provide the group state in the heartbeat response. This has a few
> >> advantages: 1) it does not require using a special error code to
> >> signal that a new assignment is required as the signal would be the
> >> provided group state; 2) it removes one round trip when a client side
> >> assignor is used. The downside is that it makes the heartbeat
> >> response's definition quite large. I recall that I went with the
> >> current approach due to this.
> >>
> >> Providing the group state in the heartbeat response is appealing. What
> >> do you guys think?
> >>
> >> Best,
> >> David
> >>
> >> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io> wrote:
> >> >
> >> > Hi Guozhang,
> >> >
> >> > 1. I have added a reference to the relevant chapter instead of
> >> > repeating the whole thing. Does that work for you?
> >> >
> >> > 2. The "Rebalance Triggers" section you are referring to is about when
> >> > a rebalance should be triggered for the non-upgraded members using the
> >> > old protocol. The section mentions that a rebalance must be triggered
> >> > when a new assignment is installed. This implies that the group epoch
> >> > was updated either by a native member or a non-upgraded member. For
> >> > the latter, the JoinGroup request would be the trigger. I have added a
> >> > reference to the relevant chapter in the "JoinGroup Handling" section
> >> > as well. Does that make sense?
> >> >
> >> > Thanks,
> >> > David
> >> >
> >> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> >> wrote:
> >> > >
> >> > > Hello David,
> >> > >
> >> > > Alright I think that's sufficient. Just to make that clear in the doc,
> >> > > could we update:
> >> > >
> >> > > 1) the heartbeat request handling section, stating when coordinator
> >> will
> >> > > trigger rebalance based on the HB's member metadata / reason?
> >> > > 2) the "Rebalance Triggers" section to include what we described in
> >> "Group
> >> > > Epoch - Trigger a rebalance" section as well?
> >> > >
> >> > >
> >> > > Guozhang
> >> > >
> >> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> >> <dj...@confluent.io.invalid>
> >> > > wrote:
> >> > >
> >> > > > Hi Guozhang,
> >> > > >
> >> > > > I thought that the assignor will always be consulted when the next
> >> > > > heartbeat request is constructed. In other words,
> >> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
> >> This
> >> > > > gives the opportunity for the assignor to enforce a rebalance by
> >> > > > setting the reason to a non-zero value or by changing the bytes. Do
> >> > > > you think that this is not sufficient? Are you concerned by the
> >> delay?
> >> > > >
> >> > > > Best,
> >> > > > David
> >> > > >
> >> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
> >> wrote:
> >> > > > >
> >> > > > > Hello David,
> >> > > > >
> >> > > > > One of Jun's comments make me thinking:
> >> > > > >
> >> > > > > ```
> >> > > > > In this case, a new assignment is triggered by the client side
> >> > > > > assignor. When constructing the HB, the consumer will always
> >> consult
> >> > > > > the client side assignor and propagate the information to the
> >> group
> >> > > > > coordinator. In other words, we don't expect users to call
> >> > > > > Consumer#enforceRebalance anymore.
> >> > > > > ```
> >> > > > >
> >> > > > > As I looked at the current PartitionAssignor's interface, we
> >> actually do
> >> > > > > not have a way yet to instruct how to construct the next HB
> >> request, e.g.
> >> > > > > when the assignor wants to enforce a new rebalance with a new
> >> assignment,
> >> > > > > we'd need some customizable APIs inside the PartitionAssignor to
> >> indicate
> >> > > > > the next HB telling broker about so. WDYT about adding such an
> >> API on the
> >> > > > > PartitionAssignor?
> >> > > > >
> >> > > > >
> >> > > > > Guozhang
> >> > > > >
> >> > > > >
> >> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> >> <dj...@confluent.io.invalid>
> >> > > > > wrote:
> >> > > > >
> >> > > > > > Hi Jun,
> >> > > > > >
> >> > > > > > I have updated the KIP to include your feedback. I have also
> >> tried to
> >> > > > > > clarify the parts which were not cleared.
> >> > > > > >
> >> > > > > > Best,
> >> > > > > > David
> >> > > > > >
> >> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <djacot@confluent.io
> >> >
> >> > > > wrote:
> >> > > > > > >
> >> > > > > > > Hi Jun,
> >> > > > > > >
> >> > > > > > > Thanks for your feedback. Let me start by answering your
> >> questions
> >> > > > > > > inline and I will update the KIP next week.
> >> > > > > > >
> >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> >> seem to
> >> > > > be
> >> > > > > > fewer
> >> > > > > > > > RPCs during rebalance and more efficient support of
> >> wildcard. A few
> >> > > > > > > > comments below.
> >> > > > > > >
> >> > > > > > > I would also add that the KIP removes the global sync barrier
> >> in the
> >> > > > > > > protocol which is essential to improve group stability and
> >> > > > > > > scalability, and the KIP also simplifies the client by moving
> >> most of
> >> > > > > > > the logic to the server side.
> >> > > > > > >
> >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> >> rolling
> >> > > > > > changing
> >> > > > > > > > of the partition assignor in the consumers?
> >> > > > > > >
> >> > > > > > > Definitely. The group coordinator will use the assignor used
> >> by a
> >> > > > > > > majority of the members. This allows the group to move from
> >> one
> >> > > > > > > assignor to another by a roll. This is explained in the
> >> Assignor
> >> > > > > > > Selection chapter.
> >> > > > > > >
> >> > > > > > > > 30.2 For each field, could you explain whether it's
> >> required in
> >> > > > every
> >> > > > > > > > request or the scenarios when it needs to be filled? For
> >> example,
> >> > > > it's
> >> > > > > > not
> >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> >> > > > > > >
> >> > > > > > > The client is expected to set those fields in case of a
> >> connection
> >> > > > > > > issue (e.g. timeout) or when the fields have changed since
> >> the last
> >> > > > > > > HB. The server populates those fields as long as the member
> >> is not
> >> > > > > > > fully reconciled - the member should acknowledge that it has
> >> the
> >> > > > > > > expected epoch and assignment. I will clarify this in the KIP.
> >> > > > > > >
> >> > > > > > > > 31. In the current consumer protocol, the rack affinity
> >> between the
> >> > > > > > client
> >> > > > > > > > and the broker is only considered during fetching, but not
> >> during
> >> > > > > > assigning
> >> > > > > > > > partitions to consumers. Sometimes, once the assignment is
> >> made,
> >> > > > there
> >> > > > > > is
> >> > > > > > > > no opportunity for read affinity because no replicas of
> >> assigned
> >> > > > > > partitions
> >> > > > > > > > are close to the member. I am wondering if we should use
> >> this
> >> > > > > > opportunity
> >> > > > > > > > to address this by including rack in GroupMember.
> >> > > > > > >
> >> > > > > > > That's an interesting idea. I don't see any issue with adding
> >> the
> >> > > > rack
> >> > > > > > > to the members. I will do so.
> >> > > > > > >
> >> > > > > > > > 32. On the metric side, often, it's useful to know how busy
> >> a group
> >> > > > > > > > coordinator is. By moving the event loop model, it seems
> >> that we
> >> > > > could
> >> > > > > > add
> >> > > > > > > > a metric that tracks the fraction of the time the event
> >> loop is
> >> > > > doing
> >> > > > > > the
> >> > > > > > > > actual work.
> >> > > > > > >
> >> > > > > > > That's a great idea. I will add it. Thanks.
> >> > > > > > >
> >> > > > > > > > 33. Could we add a section on coordinator failover
> >> handling? For
> >> > > > > > example,
> >> > > > > > > > does it need to trigger the check if any group with the
> >> wildcard
> >> > > > > > > > subscription now has a new matching topic?
> >> > > > > > >
> >> > > > > > > Sure. When the new group coordinator takes over, it has to:
> >> > > > > > > * Setup the session timeouts.
> >> > > > > > > * Trigger a new assignment if a client side assignor is used.
> >> We
> >> > > > don't
> >> > > > > > > store the information about the member selected to run the
> >> assignment
> >> > > > > > > so we have to start a new one.
> >> > > > > > > * Update the topics metadata, verify the wildcard
> >> subscriptions, and
> >> > > > > > > trigger a rebalance if needed.
> >> > > > > > >
> >> > > > > > > > 34. ConsumerGroupMetadataValue,
> >> > > > ConsumerGroupPartitionMetadataValue,
> >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> >> the epoch
> >> > > > > > field
> >> > > > > > > > reflects? For example, does the epoch in
> >> ConsumerGroupMetadataValue
> >> > > > > > reflect
> >> > > > > > > > the latest group epoch? What about the one in
> >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> >> > > > > > ConsumerGroupMemberMetadataValue?
> >> > > > > > >
> >> > > > > > > Sure. I will clarify that but it is always the latest group
> >> epoch.
> >> > > > > > > When the group state is updated, the group epoch is bumped so
> >> we use
> >> > > > > > > that one for all the change records related to the update.
> >> > > > > > >
> >> > > > > > > > 35. "the group coordinator will ensure that the following
> >> > > > invariants
> >> > > > > > are
> >> > > > > > > > met: ... All members exists." It's possible for a member
> >> not to
> >> > > > get any
> >> > > > > > > > assigned partitions, right?
> >> > > > > > >
> >> > > > > > > That's right. Here I meant that the members provided by the
> >> assignor
> >> > > > > > > in the assignment must exist in the group. The assignor can
> >> not make
> >> > > > > > > up new member ids.
> >> > > > > > >
> >> > > > > > > > 36. "He can rejoins the group with a member epoch equals to
> >> 0":
> >> > > > When
> >> > > > > > would
> >> > > > > > > > a consumer rejoin and what member id would be used?
> >> > > > > > >
> >> > > > > > > A member is expected to abandon all its partitions and
> >> rejoins when
> >> > > > it
> >> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> >> group
> >> > > > > > > coordinator will have removed the member from the group. The
> >> member
> >> > > > > > > can rejoin the group with the same member id but with 0 as
> >> epoch. Let
> >> > > > > > > me see if I can clarify this in the KIP.
> >> > > > > > >
> >> > > > > > > > 37. "Instead, power users will have the ability to trigger a
> >> > > > > > reassignment
> >> > > > > > > > by either providing a non-zero reason or by updating the
> >> assignor
> >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> >> deprecation
> >> > > > of
> >> > > > > > > > Consumer#enforeRebalance.
> >> > > > > > >
> >> > > > > > > In this case, a new assignment is triggered by the client side
> >> > > > > > > assignor. When constructing the HB, the consumer will always
> >> consult
> >> > > > > > > the client side assignor and propagate the information to the
> >> group
> >> > > > > > > coordinator. In other words, we don't expect users to call
> >> > > > > > > Consumer#enforceRebalance anymore.
> >> > > > > > >
> >> > > > > > > > 38. The reassignment examples are nice. But the section
> >> seems to
> >> > > > have
> >> > > > > > > > multiple typos.
> >> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> >> gets into
> >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> >> into
> >> > > > > > epoch=3,
> >> > > > > > > > partitions=[foo-1] too early.
> >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> >> epoch=2,
> >> > > > > > > > partitions=[foo-0].
> >> > > > > > >
> >> > > > > > > Sorry for that! I will revise them.
> >> > > > > > >
> >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> >> from
> >> > > > any
> >> > > > > > old
> >> > > > > > > > version to new one?
> >> > > > > > >
> >> > > > > > > We will support upgrading from the consumer protocol version
> >> 3,
> >> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
> >> earliest
> >> > > > > > > version is unknown at the moment. This is explained in the
> >> migration
> >> > > > > > > plan chapter.
> >> > > > > > >
> >> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> >> based on
> >> > > > it
> >> > > > > > > next week.
> >> > > > > > >
> >> > > > > > > Best,
> >> > > > > > > David
> >> > > > > > >
> >> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> >> <ju...@confluent.io.invalid>
> >> > > > wrote:
> >> > > > > > > >
> >> > > > > > > > Hi, David,
> >> > > > > > > >
> >> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> >> seem to
> >> > > > be
> >> > > > > > fewer
> >> > > > > > > > RPCs during rebalance and more efficient support of
> >> wildcard. A few
> >> > > > > > > > comments below.
> >> > > > > > > >
> >> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> >> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> >> rolling
> >> > > > > > changing
> >> > > > > > > > of the partition assignor in the consumers?
> >> > > > > > > > 30.2 For each field, could you explain whether it's
> >> required in
> >> > > > every
> >> > > > > > > > request or the scenarios when it needs to be filled? For
> >> example,
> >> > > > it's
> >> > > > > > not
> >> > > > > > > > clear to me when TopicPartitions needs to be filled.
> >> > > > > > > >
> >> > > > > > > > 31. In the current consumer protocol, the rack affinity
> >> between the
> >> > > > > > client
> >> > > > > > > > and the broker is only considered during fetching, but not
> >> during
> >> > > > > > assigning
> >> > > > > > > > partitions to consumers. Sometimes, once the assignment is
> >> made,
> >> > > > there
> >> > > > > > is
> >> > > > > > > > no opportunity for read affinity because no replicas of
> >> assigned
> >> > > > > > partitions
> >> > > > > > > > are close to the member. I am wondering if we should use
> >> this
> >> > > > > > opportunity
> >> > > > > > > > to address this by including rack in GroupMember.
> >> > > > > > > >
> >> > > > > > > > 32. On the metric side, often, it's useful to know how busy
> >> a group
> >> > > > > > > > coordinator is. By moving the event loop model, it seems
> >> that we
> >> > > > could
> >> > > > > > add
> >> > > > > > > > a metric that tracks the fraction of the time the event
> >> loop is
> >> > > > doing
> >> > > > > > the
> >> > > > > > > > actual work.
> >> > > > > > > >
> >> > > > > > > > 33. Could we add a section on coordinator failover
> >> handling? For
> >> > > > > > example,
> >> > > > > > > > does it need to trigger the check if any group with the
> >> wildcard
> >> > > > > > > > subscription now has a new matching topic?
> >> > > > > > > >
> >> > > > > > > > 34. ConsumerGroupMetadataValue,
> >> > > > ConsumerGroupPartitionMetadataValue,
> >> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> >> the epoch
> >> > > > > > field
> >> > > > > > > > reflects? For example, does the epoch in
> >> ConsumerGroupMetadataValue
> >> > > > > > reflect
> >> > > > > > > > the latest group epoch? What about the one in
> >> > > > > > > > ConsumerGroupPartitionMetadataValue and
> >> > > > > > ConsumerGroupMemberMetadataValue?
> >> > > > > > > >
> >> > > > > > > > 35. "the group coordinator will ensure that the following
> >> > > > invariants
> >> > > > > > are
> >> > > > > > > > met: ... All members exists." It's possible for a member
> >> not to
> >> > > > get any
> >> > > > > > > > assigned partitions, right?
> >> > > > > > > >
> >> > > > > > > > 36. "He can rejoins the group with a member epoch equals to
> >> 0":
> >> > > > When
> >> > > > > > would
> >> > > > > > > > a consumer rejoin and what member id would be used?
> >> > > > > > > >
> >> > > > > > > > 37. "Instead, power users will have the ability to trigger a
> >> > > > > > reassignment
> >> > > > > > > > by either providing a non-zero reason or by updating the
> >> assignor
> >> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> >> deprecation
> >> > > > of
> >> > > > > > > > Consumer#enforeRebalance.
> >> > > > > > > >
> >> > > > > > > > 38. The reassignment examples are nice. But the section
> >> seems to
> >> > > > have
> >> > > > > > > > multiple typos.
> >> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> >> gets into
> >> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> >> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> >> into
> >> > > > > > epoch=3,
> >> > > > > > > > partitions=[foo-1] too early.
> >> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> >> epoch=2,
> >> > > > > > > > partitions=[foo-0].
> >> > > > > > > >
> >> > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> >> from
> >> > > > any
> >> > > > > > old
> >> > > > > > > > version to new one?
> >> > > > > > > >
> >> > > > > > > > Thanks,
> >> > > > > > > >
> >> > > > > > > > Jun
> >> > > > > > > >
> >> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> >> > > > > > <dj...@confluent.io.invalid>
> >> > > > > > > > wrote:
> >> > > > > > > >
> >> > > > > > > > > Hi all,
> >> > > > > > > > >
> >> > > > > > > > > The KIP states that we will re-implement the coordinator
> >> in
> >> > > > Java. I
> >> > > > > > > > > discussed this offline with a few folks and folks are
> >> concerned
> >> > > > that
> >> > > > > > > > > we could introduce many regressions in the old protocol
> >> if we do
> >> > > > so.
> >> > > > > > > > > Therefore, I am going to remove this statement from the
> >> KIP. It
> >> > > > is an
> >> > > > > > > > > implementation detail after all so it does not have to be
> >> > > > decided at
> >> > > > > > > > > this stage. We will likely start by trying to refactor the
> >> > > > current
> >> > > > > > > > > implementation as a first step.
> >> > > > > > > > >
> >> > > > > > > > > Cheers,
> >> > > > > > > > > David
> >> > > > > > > > >
> >> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> >> djacot@confluent.io
> >> > > > >
> >> > > > > > wrote:
> >> > > > > > > > > >
> >> > > > > > > > > > Hi Luke,
> >> > > > > > > > > >
> >> > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> >> > > > > > reconciling,
> >> > > > > > > > > stable,
> >> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> >> right?
> >> > > > > > > > > >
> >> > > > > > > > > > This sentence does not refer to those group states but
> >> rather
> >> > > > to a
> >> > > > > > > > > > state machine replication (SMR). This refers to the
> >> entire
> >> > > > state of
> >> > > > > > > > > > group coordinator which is replicated via the log
> >> layer. I will
> >> > > > > > > > > > clarify this in the KIP.
> >> > > > > > > > > >
> >> > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
> >> as an
> >> > > > event
> >> > > > > > > > > loop"?
> >> > > > > > > > > >
> >> > > > > > > > > > The idea is to follow a model similar to the new quorum
> >> > > > > > controller. We
> >> > > > > > > > > > will have N threads to process events. Each
> >> __consumer_offsets
> >> > > > > > > > > > partition is assigned to a unique thread and all the
> >> events
> >> > > > (e.g.
> >> > > > > > > > > > requests, callbacks, etc.) are processed by this
> >> thread. This
> >> > > > > > simplify
> >> > > > > > > > > > concurrency and will enable us to do simulation testing
> >> for the
> >> > > > > > group
> >> > > > > > > > > > coordinator.
> >> > > > > > > > > >
> >> > > > > > > > > > > 1.3. Why do we need a state machine per
> >> *__consumer_offsets*
> >> > > > > > > > > partitions?
> >> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> >> group
> >> > > > > > coordinator?
> >> > > > > > > > > For
> >> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> >> groups, and
> >> > > > > > both
> >> > > > > > > > > exist in
> >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> >> for it,
> >> > > > or
> >> > > > > > 2?
> >> > > > > > > > > >
> >> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> >> > > > > > > > > >
> >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> >> number
> >> > > > of
> >> > > > > > threads
> >> > > > > > > > > used
> >> > > > > > > > > > > to run the state machines. But I'm wondering if the
> >> purpose
> >> > > > of
> >> > > > > > the
> >> > > > > > > > > threads
> >> > > > > > > > > > > is only to keep the state of each consumer group (or
> >> > > > > > > > > *__consumer_offsets*
> >> > > > > > > > > > > partitions?), and no heavy computation, why should we
> >> need
> >> > > > > > > > > multi-threads
> >> > > > > > > > > > > here?
> >> > > > > > > > > >
> >> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> >> > > > processing as
> >> > > > > > the
> >> > > > > > > > > > computation could be heavy.
> >> > > > > > > > > >
> >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> >> default
> >> > > > session
> >> > > > > > > > > timeout not
> >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> >> min/max
> >> > > > > > session
> >> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> >> > > > > > > > > > >
> >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> >> timeout to
> >> > > > detect
> >> > > > > > client
> >> > > > > > > > > > > failures when using the consumer group protocol.
> >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> >> minimum
> >> > > > > > session
> >> > > > > > > > > timeout.
> >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> >> maximum
> >> > > > > > session
> >> > > > > > > > > timeout.
> >> > > > > > > > > >
> >> > > > > > > > > > This is indeed a mistake. The default session timeout
> >> should
> >> > > > be 45s
> >> > > > > > > > > > (the current default).
> >> > > > > > > > > >
> >> > > > > > > > > > > 2.2. The default server side assignor are [range,
> >> uniform],
> >> > > > > > which means
> >> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> >> why not
> >> > > > > > uniform
> >> > > > > > > > > one? I
> >> > > > > > > > > > > thought usually users will choose uniform assignor
> >> (former
> >> > > > sticky
> >> > > > > > > > > assinor)
> >> > > > > > > > > > > for better evenly distribution. Any other reason we
> >> choose
> >> > > > range
> >> > > > > > > > > assignor
> >> > > > > > > > > > > as default?
> >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> >> server side
> >> > > > > > assignors.
> >> > > > > > > > > >
> >> > > > > > > > > > The order on the server side has no influence because
> >> the
> >> > > > client
> >> > > > > > must
> >> > > > > > > > > > chose the selector that he wants to use. There is no
> >> default
> >> > > > in the
> >> > > > > > > > > > current proposal. If the assignor is not specified by
> >> the
> >> > > > client,
> >> > > > > > the
> >> > > > > > > > > > request is rejected. The default client value for
> >> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> >> > > > > > > > > >
> >> > > > > > > > > > Thanks for your very good comments, Luke. I hope that my
> >> > > > answers
> >> > > > > > help
> >> > > > > > > > > > to clarify things. I will update the KIP as well based
> >> on your
> >> > > > > > > > > > feedback.
> >> > > > > > > > > >
> >> > > > > > > > > > Cheers,
> >> > > > > > > > > > David
> >> > > > > > > > > >
> >> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> >> showuon@gmail.com>
> >> > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > Hi David,
> >> > > > > > > > > > >
> >> > > > > > > > > > > Thanks for the update.
> >> > > > > > > > > > >
> >> > > > > > > > > > > Some more questions:
> >> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> >> > > > > > > > > > > > The new group coordinator will have a state machine
> >> per
> >> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> >> machine is
> >> > > > > > modelled
> >> > > > > > > > > as an
> >> > > > > > > > > > > event loop. Those state machines will be executed in
> >> > > > > > > > > > > *group.coordinator.threads* threads.
> >> > > > > > > > > > >
> >> > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> >> > > > > > reconciling,
> >> > > > > > > > > stable,
> >> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> >> right?
> >> > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
> >> as an
> >> > > > event
> >> > > > > > > > > loop"?
> >> > > > > > > > > > > 1.3. Why do we need a state machine per
> >> *__consumer_offsets*
> >> > > > > > > > > partitions?
> >> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> >> group
> >> > > > > > coordinator?
> >> > > > > > > > > For
> >> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> >> groups, and
> >> > > > > > both
> >> > > > > > > > > exist in
> >> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> >> for it,
> >> > > > or
> >> > > > > > 2?
> >> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> >> number
> >> > > > of
> >> > > > > > threads
> >> > > > > > > > > used
> >> > > > > > > > > > > to run the state machines. But I'm wondering if the
> >> purpose
> >> > > > of
> >> > > > > > the
> >> > > > > > > > > threads
> >> > > > > > > > > > > is only to keep the state of each consumer group (or
> >> > > > > > > > > *__consumer_offsets*
> >> > > > > > > > > > > partitions?), and no heavy computation, why should we
> >> need
> >> > > > > > > > > multi-threads
> >> > > > > > > > > > > here?
> >> > > > > > > > > > >
> >> > > > > > > > > > > 2. For the default value in the new configs:
> >> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> >> default
> >> > > > session
> >> > > > > > > > > timeout not
> >> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> >> min/max
> >> > > > > > session
> >> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> >> > > > > > > > > > >
> >> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> >> timeout to
> >> > > > detect
> >> > > > > > client
> >> > > > > > > > > > > failures when using the consumer group protocol.
> >> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> >> minimum
> >> > > > > > session
> >> > > > > > > > > timeout.
> >> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> >> maximum
> >> > > > > > session
> >> > > > > > > > > timeout.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > 2.2. The default server side assignor are [range,
> >> uniform],
> >> > > > > > which means
> >> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> >> why not
> >> > > > > > uniform
> >> > > > > > > > > one? I
> >> > > > > > > > > > > thought usually users will choose uniform assignor
> >> (former
> >> > > > sticky
> >> > > > > > > > > assinor)
> >> > > > > > > > > > > for better evenly distribution. Any other reason we
> >> choose
> >> > > > range
> >> > > > > > > > > assignor
> >> > > > > > > > > > > as default?
> >> > > > > > > > > > > group.consumer.assignors List range, uniform The
> >> server side
> >> > > > > > assignors.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Thank you.
> >> > > > > > > > > > > Luke
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> >> showuon@gmail.com
> >> > > > >
> >> > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > > Hi Sagar,
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > I have some thoughts about Kafka Connect
> >> integrating with
> >> > > > > > KIP-848,
> >> > > > > > > > > but I
> >> > > > > > > > > > > > think we should have a separate discussion thread
> >> for the
> >> > > > Kafka
> >> > > > > > > > > Connect
> >> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> >> Rebalance
> >> > > > > > Protocol
> >> > > > > > > > > [1],
> >> > > > > > > > > > > > and let this discussion thread focus on consumer
> >> rebalance
> >> > > > > > protocol,
> >> > > > > > > > > WDYT?
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > [1]
> >> > > > > > > > > > > >
> >> > > > > > > > >
> >> > > > > >
> >> > > >
> >> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Thank you.
> >> > > > > > > > > > > > Luke
> >> > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> >> > > > > > sagarmeansocean@gmail.com>
> >> > > > > > > > > wrote:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
> >> like
> >> > > > there's
> >> > > > > > > > > agreement on
> >> > > > > > > > > > > >> using separate APIs for Connect. I would revisit
> >> the doc
> >> > > > and
> >> > > > > > see
> >> > > > > > > > > what
> >> > > > > > > > > > > >> changes are to be made.
> >> > > > > > > > > > > >>
> >> > > > > > > > > > > >> Thanks!
> >> > > > > > > > > > > >> Sagar.
> >> > > > > > > > > > > >>
> >> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> >> > > > > > > > > <dj...@confluent.io.invalid>
> >> > > > > > > > > > > >> wrote:
> >> > > > > > > > > > > >>
> >> > > > > > > > > > > >> > Hi Sagar,
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >> > Thanks for the feedback and the document. That's
> >> really
> >> > > > > > helpful. I
> >> > > > > > > > > > > >> > will take a look at it.
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >> > Overall, it seems to me that both Connect and the
> >> > > > Consumer
> >> > > > > > could
> >> > > > > > > > > share
> >> > > > > > > > > > > >> > the same underlying "engine". The main
> >> difference is
> >> > > > that
> >> > > > > > the
> >> > > > > > > > > Consumer
> >> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> >> Connect
> >> > > > assigns
> >> > > > > > tasks
> >> > > > > > > > > to
> >> > > > > > > > > > > >> > workers. I see two ways to move forward:
> >> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> >> different
> >> > > > > > resource
> >> > > > > > > > > types
> >> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> >> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> >> dedicated
> >> > > > APIs
> >> > > > > > > > > would be
> >> > > > > > > > > > > >> > similar to the new ones but different on the
> >> > > > > > content/resources and
> >> > > > > > > > > > > >> > they would rely on the same engine on the
> >> coordinator
> >> > > > side.
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >> > I personally lean towards 2) because I am not a
> >> fan of
> >> > > > > > > > > overcharging
> >> > > > > > > > > > > >> > APIs to serve different purposes. That being
> >> said, I am
> >> > > > not
> >> > > > > > > > > opposed to
> >> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >> > I think that we can continue to discuss it here
> >> for now
> >> > > > in
> >> > > > > > order
> >> > > > > > > > > to
> >> > > > > > > > > > > >> > ensure that this KIP is compatible with what we
> >> will do
> >> > > > for
> >> > > > > > > > > Connect in
> >> > > > > > > > > > > >> > the future.
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >> > Best,
> >> > > > > > > > > > > >> > David
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> >> > > > > > djacot@confluent.io>
> >> > > > > > > > > wrote:
> >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > >> > > Hi all,
> >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > >> > > I am back from vacation. I will go through and
> >> address
> >> > > > > > your
> >> > > > > > > > > comments
> >> > > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > >> > > Cheers,
> >> > > > > > > > > > > >> > > David
> >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris
> >> <
> >> > > > > > > > > gharris1727@gmail.com
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >> > wrote:
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > Hey All!
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> >> > > > cooperative
> >> > > > > > > > > rebalancing
> >> > > > > > > > > > > >> > making it
> >> > > > > > > > > > > >> > > > down the stack!
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > I had a few questions:
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> >> describes how
> >> > > > > > member
> >> > > > > > > > > epoch
> >> > > > > > > > > > > >> > should
> >> > > > > > > > > > > >> > > > advance in step with the group epoch and
> >> assignment
> >> > > > > > epoch
> >> > > > > > > > > values. I
> >> > > > > > > > > > > >> > think
> >> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> >> described
> >> > > > in
> >> > > > > > the
> >> > > > > > > > > KIP. When
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > protocol is incrementally assigning
> >> partitions to a
> >> > > > > > worker,
> >> > > > > > > > > what
> >> > > > > > > > > > > >> member
> >> > > > > > > > > > > >> > > > epoch does each incremental assignment use?
> >> Are
> >> > > > member
> >> > > > > > epochs
> >> > > > > > > > > > > >> re-used,
> >> > > > > > > > > > > >> > and
> >> > > > > > > > > > > >> > > > a single member epoch can correspond to
> >> multiple
> >> > > > > > different
> >> > > > > > > > > > > >> > (monotonically
> >> > > > > > > > > > > >> > > > larger) assignments?
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque
> >> to the
> >> > > > group
> >> > > > > > > > > > > >> coordinator? If
> >> > > > > > > > > > > >> > > > not, should custom client-side assignor
> >> > > > implementations
> >> > > > > > > > > interact
> >> > > > > > > > > > > >> with
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > Reason field, and how is its common meaning
> >> agreed
> >> > > > > > upon? If
> >> > > > > > > > > so, what
> >> > > > > > > > > > > >> > is the
> >> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> >> including
> >> > > > such
> >> > > > > > > > > functionality
> >> > > > > > > > > > > >> > in the
> >> > > > > > > > > > > >> > > > opaque metadata?
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> >> "Thanks to
> >> > > > > > this, the
> >> > > > > > > > > input
> >> > > > > > > > > > > >> of
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > client side assignor is entirely driven by
> >> the group
> >> > > > > > > > > coordinator.
> >> > > > > > > > > > > >> The
> >> > > > > > > > > > > >> > > > consumer is no longer responsible for
> >> maintaining
> >> > > > any
> >> > > > > > state
> >> > > > > > > > > besides
> >> > > > > > > > > > > >> its
> >> > > > > > > > > > > >> > > > assigned partitions." Does this mean that the
> >> > > > > > client-side
> >> > > > > > > > > assignor
> >> > > > > > > > > > > >> MAY
> >> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> >> (such as
> >> > > > > > partition
> >> > > > > > > > > > > >> > throughput,
> >> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or
> >> that
> >> > > > > > additional
> >> > > > > > > > > > > >> > non-Metadata
> >> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> >> > > > > > > > > > > >> > > > for
> >> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> >> > > > > > > > > > > >> > > > and
> >> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> >> > > > > > that
> >> > > > > > > > > seem to
> >> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> >> these two
> >> > > > > > > > > implementations
> >> > > > > > > > > > > >> to
> >> > > > > > > > > > > >> > be
> >> > > > > > > > > > > >> > > > unified? This would serve to promote feature
> >> parity
> >> > > > of
> >> > > > > > > > > server-side
> >> > > > > > > > > > > >> and
> >> > > > > > > > > > > >> > > > client-side assignors, and would also
> >> facilitate
> >> > > > > > operational
> >> > > > > > > > > > > >> > flexibility in
> >> > > > > > > > > > > >> > > > certain situations. For example, if a
> >> server-side
> >> > > > > > assignor
> >> > > > > > > > > has some
> >> > > > > > > > > > > >> > poor
> >> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> >> patched
> >> > > > > > assignor to
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > client
> >> > > > > > > > > > > >> > > > and switching one consumer group to a
> >> client-side
> >> > > > > > assignor
> >> > > > > > > > > may be
> >> > > > > > > > > > > >> > faster
> >> > > > > > > > > > > >> > > > and less risky than patching all of the
> >> brokers.
> >> > > > With
> >> > > > > > the
> >> > > > > > > > > currently
> >> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> >> > > > reimplementation
> >> > > > > > would
> >> > > > > > > > > have
> >> > > > > > > > > > > >> to be
> >> > > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> >> > > > > > significantly,
> >> > > > > > > > > then it
> >> > > > > > > > > > > >> is
> >> > > > > > > > > > > >> > > > possible that a reimplementation would not be
> >> > > > possible.
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > --
> >> > > > > > > > > > > >> > > > Greg Harris
> >> > > > > > > > > > > >> > > > gharris1727@gmail.com
> >> > > > > > > > > > > >> > > > github.com/gharris1727
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> >> > > > > > > > > sagarmeansocean@gmail.com>
> >> > > > > > > > > > > >> > wrote:
> >> > > > > > > > > > > >> > > >
> >> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > I created a confluence page to discuss how
> >> Connect
> >> > > > > > would
> >> > > > > > > > > need to
> >> > > > > > > > > > > >> > change
> >> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> >> Here's the
> >> > > > page:
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >>
> >> > > > > > > > >
> >> > > > > >
> >> > > >
> >> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > It's also pretty longish and I have tried
> >> to keep
> >> > > > a
> >> > > > > > format
> >> > > > > > > > > > > >> similar to
> >> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also,
> >> do you
> >> > > > > > think this
> >> > > > > > > > > > > >> should
> >> > > > > > > > > > > >> > be
> >> > > > > > > > > > > >> > > > > moved to a separate discussion thread or
> >> is this
> >> > > > one
> >> > > > > > fine?
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > Thanks!
> >> > > > > > > > > > > >> > > > > Sagar.
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> >> > > > > > > > > sagarmeansocean@gmail.com>
> >> > > > > > > > > > > >> > wrote:
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > > Hello Guozhang,
> >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> >> Streams.
> >> > > > > > Sure, I
> >> > > > > > > > > would do
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > > analysis and come up with such a
> >> document.
> >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > > Thanks!
> >> > > > > > > > > > > >> > > > > > Sagar.
> >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang
> >> Wang <
> >> > > > > > > > > > > >> wangguoz@gmail.com>
> >> > > > > > > > > > > >> > > > > wrote:
> >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> Hello Sagar,
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > > >> It would be great if you could come
> >> back with
> >> > > > some
> >> > > > > > > > > analysis on
> >> > > > > > > > > > > >> > how to
> >> > > > > > > > > > > >> > > > > >> implement the Connect side integration
> >> with
> >> > > > the new
> >> > > > > > > > > protocol;
> >> > > > > > > > > > > >> so
> >> > > > > > > > > > > >> > far
> >> > > > > > > > > > > >> > > > > >> besides leveraging on the new "protocol
> >> type"
> >> > > > we
> >> > > > > > did not
> >> > > > > > > > > yet
> >> > > > > > > > > > > >> think
> >> > > > > > > > > > > >> > > > > through
> >> > > > > > > > > > > >> > > > > >> the Connect side implementations. For
> >> Streams
> >> > > > > > here's a
> >> > > > > > > > > draft of
> >> > > > > > > > > > > >> > > > > >> integration
> >> > > > > > > > > > > >> > > > > >> plan:
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >>
> >> > > > > > > > >
> >> > > > > >
> >> > > >
> >> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >> > > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> >> > > > > > > > > > > >> sagarmeansocean@gmail.com
> >> > > > > > > > > > > >> > >
> >> > > > > > > > > > > >> > > > > wrote:
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > > >> > Hi David,
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> >> reason I
> >> > > > thought
> >> > > > > > > > > connect can
> >> > > > > > > > > > > >> > also fit
> >> > > > > > > > > > > >> > > > > >> into
> >> > > > > > > > > > > >> > > > > >> > this new scheme is that even today the
> >> > > > connect
> >> > > > > > uses a
> >> > > > > > > > > > > >> > > > > WorkerCoordinator
> >> > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to
> >> empower
> >> > > > > > > > > rebalances of
> >> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> >> WorkerCoordinator sets
> >> > > > the
> >> > > > > > > > > > > >> protocolType()
> >> > > > > > > > > > > >> > to
> >> > > > > > > > > > > >> > > > > >> connect
> >> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> >> plumbing
> >> > > > into
> >> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> > I think the changes to support
> >> connect would
> >> > > > be
> >> > > > > > > > > similar at a
> >> > > > > > > > > > > >> > high
> >> > > > > > > > > > > >> > > > > level
> >> > > > > > > > > > > >> > > > > >> to
> >> > > > > > > > > > > >> > > > > >> > the changes in streams mainly because
> >> of the
> >> > > > > > Client
> >> > > > > > > > > side
> >> > > > > > > > > > > >> > assignors
> >> > > > > > > > > > > >> > > > > being
> >> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
> >> level, we
> >> > > > > > might
> >> > > > > > > > > need to
> >> > > > > > > > > > > >> make
> >> > > > > > > > > > > >> > a lot
> >> > > > > > > > > > > >> > > > > of
> >> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> >> assignment
> >> > > > protocol
> >> > > > > > like
> >> > > > > > > > > > > >> enhancing
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> >> SyncGroup and
> >> > > > > > using
> >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> >> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> >> streams (or
> >> > > > > > there
> >> > > > > > > > > might be
> >> > > > > > > > > > > >> > > > > >> deviations). I
> >> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> >> analysis of
> >> > > > the
> >> > > > > > same
> >> > > > > > > > > and we
> >> > > > > > > > > > > >> > can have
> >> > > > > > > > > > > >> > > > > a
> >> > > > > > > > > > > >> > > > > >> > separate discussion thread for that
> >> as that
> >> > > > would
> >> > > > > > > > > derail this
> >> > > > > > > > > > > >> > > > > discussion
> >> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
> >> good to
> >> > > > you.
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> > Thanks!
> >> > > > > > > > > > > >> > > > > >> > Sagar.
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David
> >> Jacot
> >> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> > wrote:
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> >> `Assignment#error`.
> >> > > > > > Sure, I
> >> > > > > > > > > can
> >> > > > > > > > > > > >> > mention it.
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from
> >> his
> >> > > > current
> >> > > > > > > > > assignment
> >> > > > > > > > > > > >> to
> >> > > > > > > > > > > >> > his
> >> > > > > > > > > > > >> > > > > >> > > target assignment when he can move
> >> to
> >> > > > epoch 3.
> >> > > > > > When
> >> > > > > > > > > that
> >> > > > > > > > > > > >> > happens,
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
> >> persisted
> >> > > > > > with all
> >> > > > > > > > > its
> >> > > > > > > > > > > >> > assigned
> >> > > > > > > > > > > >> > > > > >> > > partitions even if they are not all
> >> revoked
> >> > > > > > yet. In
> >> > > > > > > > > other
> >> > > > > > > > > > > >> > words, the
> >> > > > > > > > > > > >> > > > > >> > > member assignment becomes the target
> >> > > > > > assignment.
> >> > > > > > > > > This is
> >> > > > > > > > > > > >> > basically
> >> > > > > > > > > > > >> > > > > an
> >> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> >> write all
> >> > > > the
> >> > > > > > > > > changes to
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > log.
> >> > > > > > > > > > > >> > > > > The
> >> > > > > > > > > > > >> > > > > >> > > examples are based on the persisted
> >> state
> >> > > > so I
> >> > > > > > > > > understand
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> >> improve
> >> > > > this in
> >> > > > > > the
> >> > > > > > > > > > > >> > description.
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> >> reuse the
> >> > > > > > protocol
> >> > > > > > > > > with a
> >> > > > > > > > > > > >> > client side
> >> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> >> protocol. The
> >> > > > > > assignment
> >> > > > > > > > > is
> >> > > > > > > > > > > >> about
> >> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> >> > > > Connect
> >> > > > > > fit
> >> > > > > > > > > into this?
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> > > Best,
> >> > > > > > > > > > > >> > > > > >> > > David
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> >> Sagar <
> >> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> >> > > > > > > > > > > >> > > > > >> wrote:
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > Hi David,
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
> >> minor
> >> > > > > > observations:
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> >> section in
> >> > > > Client
> >> > > > > > Side
> >> > > > > > > > > mode
> >> > > > > > > > > > > >> > Assignment
> >> > > > > > > > > > > >> > > > > >> > > process,
> >> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case,
> >> the
> >> > > > client
> >> > > > > > side
> >> > > > > > > > > assignor
> >> > > > > > > > > > > >> can
> >> > > > > > > > > > > >> > > > > return
> >> > > > > > > > > > > >> > > > > >> an
> >> > > > > > > > > > > >> > > > > >> > > > error to the group coordinator`.
> >> In this
> >> > > > > > case are
> >> > > > > > > > > you
> >> > > > > > > > > > > >> > referring to
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> >> AssignmentError
> >> > > > that's
> >> > > > > > > > > listed down
> >> > > > > > > > > > > >> > towards
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > end?
> >> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
> >> make sense
> >> > > > to
> >> > > > > > > > > mention this
> >> > > > > > > > > > > >> > > > > explicitly
> >> > > > > > > > > > > >> > > > > >> > here?
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I
> >> have a
> >> > > > > > slight
> >> > > > > > > > > > > >> confusion,
> >> > > > > > > > > > > >> > not
> >> > > > > > > > > > > >> > > > > sure
> >> > > > > > > > > > > >> > > > > >> if
> >> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
> >> this step:
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> >> coordinator
> >> > > > > > > > > transitions him
> >> > > > > > > > > > > >> to
> >> > > > > > > > > > > >> > epoch
> >> > > > > > > > > > > >> > > > > 3
> >> > > > > > > > > > > >> > > > > >> > > because
> >> > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> >> > > > persists
> >> > > > > > the
> >> > > > > > > > > change and
> >> > > > > > > > > > > >> > reply.
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> >> > > > > > > > > > > >> > > > > >> > > >       - A
> >> > > > > > > > > > > >> > > > > >> > > >       - B
> >> > > > > > > > > > > >> > > > > >> > > >       - C
> >> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> >> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> >> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> >> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> >> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> >> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> >> partitions=[foo-0,
> >> > > > > > foo-1]
> >> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> >> partitions=[foo-2]
> >> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> >> partitions=[foo-1]
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions
> >> to
> >> > > > epoch 3
> >> > > > > > but
> >> > > > > > > > > cannot
> >> > > > > > > > > > > >> get
> >> > > > > > > > > > > >> > foo-1
> >> > > > > > > > > > > >> > > > > >> yet.
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C
> >> can't
> >> > > > get
> >> > > > > > the
> >> > > > > > > > > foo-1
> >> > > > > > > > > > > >> > partition
> >> > > > > > > > > > > >> > > > > yet,
> >> > > > > > > > > > > >> > > > > >> > but
> >> > > > > > > > > > > >> > > > > >> > > > based on the description above,
> >> it seems
> >> > > > it
> >> > > > > > > > > already has
> >> > > > > > > > > > > >> it.
> >> > > > > > > > > > > >> > Do you
> >> > > > > > > > > > > >> > > > > >> > think
> >> > > > > > > > > > > >> > > > > >> > > it
> >> > > > > > > > > > > >> > > > > >> > > > would be better to remove it and
> >> > > > populate it
> >> > > > > > only
> >> > > > > > > > > when it
> >> > > > > > > > > > > >> > actually
> >> > > > > > > > > > > >> > > > > >> gets
> >> > > > > > > > > > > >> > > > > >> > > it?
> >> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> >> places, so
> >> > > > have
> >> > > > > > I
> >> > > > > > > > > > > >> understood it
> >> > > > > > > > > > > >> > > > > >> > incorrectly
> >> > > > > > > > > > > >> > > > > >> > > ?
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
> >> out of
> >> > > > scope
> >> > > > > > of
> >> > > > > > > > > this
> >> > > > > > > > > > > >> > discussion,
> >> > > > > > > > > > > >> > > > > but
> >> > > > > > > > > > > >> > > > > >> > from
> >> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> >> probably be
> >> > > > > > running in
> >> > > > > > > > > client
> >> > > > > > > > > > > >> > side
> >> > > > > > > > > > > >> > > > > >> assignor
> >> > > > > > > > > > > >> > > > > >> > > mode
> >> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> >> protocol as it
> >> > > > has
> >> > > > > > its
> >> > > > > > > > > own
> >> > > > > > > > > > > >> Custom
> >> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> >> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > Thanks!
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > Sagar.
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
> >> David
> >> > > > Jacot
> >> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
> >> move
> >> > > > forward
> >> > > > > > with
> >> > > > > > > > > > > >> > specialized API
> >> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> >> generic API.
> >> > > > For
> >> > > > > > > > > Connect, we
> >> > > > > > > > > > > >> > can apply
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> >> reuse/share the
> >> > > > core
> >> > > > > > > > > > > >> > implementation on
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> >> registry, I
> >> > > > > > think
> >> > > > > > > > > that we
> >> > > > > > > > > > > >> > should
> >> > > > > > > > > > > >> > > > > >> consider
> >> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> >> simple
> >> > > > > > > > > membership/leader
> >> > > > > > > > > > > >> > election.
> >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > Best,
> >> > > > > > > > > > > >> > > > > >> > > > > David
> >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22
> >> AM Ismael
> >> > > > > > Juma <
> >> > > > > > > > > > > >> > ismael@juma.me.uk
> >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> >> java.util.regex.Pattern
> >> > > > vs
> >> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> >> > > > > > > > > > > >> > > > > >> > > > > should
> >> > > > > > > > > > > >> > > > > >> > > > > > document the differences in
> >> more
> >> > > > detail
> >> > > > > > before
> >> > > > > > > > > > > >> deciding
> >> > > > > > > > > > > >> > one
> >> > > > > > > > > > > >> > > > > way
> >> > > > > > > > > > > >> > > > > >> or
> >> > > > > > > > > > > >> > > > > >> > > > > another.
> >> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> >> > > > > > > > > java.util.regex.Pattern,
> >> > > > > > > > > > > >> they
> >> > > > > > > > > > > >> > expect
> >> > > > > > > > > > > >> > > > > >> > their
> >> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If
> >> we are
> >> > > > doing
> >> > > > > > > > > something
> >> > > > > > > > > > > >> > different,
> >> > > > > > > > > > > >> > > > > >> then
> >> > > > > > > > > > > >> > > > > >> > we
> >> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> >> overload
> >> > > > with
> >> > > > > > our own
> >> > > > > > > > > > > >> Pattern
> >> > > > > > > > > > > >> > class
> >> > > > > > > > > > > >> > > > > (I
> >> > > > > > > > > > > >> > > > > >> > > don't
> >> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> >> re2j's at
> >> > > > this
> >> > > > > > > > > point).
> >> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> >> major new
> >> > > > > > protocol
> >> > > > > > > > > should
> >> > > > > > > > > > > >> > > > > integrate
> >> > > > > > > > > > > >> > > > > >> > fully
> >> > > > > > > > > > > >> > > > > >> > > > > with
> >> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> >> > > > > > recreation case
> >> > > > > > > > > > > >> > correctly.
> >> > > > > > > > > > > >> > > > > That's
> >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> > > main
> >> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> >> agree with
> >> > > > > > David
> >> > > > > > > > > that we'd
> >> > > > > > > > > > > >> > want to
> >> > > > > > > > > > > >> > > > > add
> >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > >> > > > > >> > > > > ids
> >> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
> >> that don't
> >> > > > > > have it
> >> > > > > > > > > yet and
> >> > > > > > > > > > > >> > that we
> >> > > > > > > > > > > >> > > > > can
> >> > > > > > > > > > > >> > > > > >> > > probably
> >> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals versus
> >> adding
> >> > > > new
> >> > > > > > APIs
> >> > > > > > > > > to the
> >> > > > > > > > > > > >> > Java
> >> > > > > > > > > > > >> > > > > >> Consumer
> >> > > > > > > > > > > >> > > > > >> > > > > (unless
> >> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs
> >> is
> >> > > > required
> >> > > > > > for
> >> > > > > > > > > > > >> reasonable
> >> > > > > > > > > > > >> > > > > >> semantics).
> >> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about
> >> the
> >> > > > > > coordinator
> >> > > > > > > > > storing
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > >> configs.
> >> > > > > > > > > > > >> > > > > >> > > It's
> >> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> >> > > > centralized
> >> > > > > > in the
> >> > > > > > > > > > > >> metadata
> >> > > > > > > > > > > >> > log for
> >> > > > > > > > > > > >> > > > > >> > > various
> >> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> >> visibility,
> >> > > > > > consistency,
> >> > > > > > > > > > > >> etc.).
> >> > > > > > > > > > > >> > > > > >> Similarly, I
> >> > > > > > > > > > > >> > > > > >> > > am
> >> > > > > > > > > > > >> > > > > >> > > > > not
> >> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> >> deleting
> >> > > > > > configs in a
> >> > > > > > > > > way
> >> > > > > > > > > > > >> that
> >> > > > > > > > > > > >> > they
> >> > > > > > > > > > > >> > > > > >> cannot
> >> > > > > > > > > > > >> > > > > >> > > be
> >> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property
> >> for modern
> >> > > > > > systems
> >> > > > > > > > > is to
> >> > > > > > > > > > > >> > minimize
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > number
> >> > > > > > > > > > > >> > > > > >> > > > > of
> >> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> >> scenarios.
> >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47
> >> PM David
> >> > > > > > Jacot
> >> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers
> >> are
> >> > > > below:
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> >> especially
> >> > > > > > the last
> >> > > > > > > > > step
> >> > > > > > > > > > > >> when
> >> > > > > > > > > > > >> > > > > clients
> >> > > > > > > > > > > >> > > > > >> > > flip the
> >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> >> protocol, in
> >> > > > which
> >> > > > > > we
> >> > > > > > > > > would
> >> > > > > > > > > > > >> have a
> >> > > > > > > > > > > >> > > > > window
> >> > > > > > > > > > > >> > > > > >> > where
> >> > > > > > > > > > > >> > > > > >> > > > > both
> >> > > > > > > > > > > >> > > > > >> > > > > > > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> >> > > > protocols /
> >> > > > > > rpcs
> >> > > > > > > > > are
> >> > > > > > > > > > > >> used
> >> > > > > > > > > > > >> > by
> >> > > > > > > > > > > >> > > > > >> members
> >> > > > > > > > > > > >> > > > > >> > of
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > same
> >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> >> coordinator could
> >> > > > > > "mimic"
> >> > > > > > > > > the old
> >> > > > > > > > > > > >> > behavior
> >> > > > > > > > > > > >> > > > > >> while
> >> > > > > > > > > > > >> > > > > >> > > > > using the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
> >> we
> >> > > > need to
> >> > > > > > > > > present
> >> > > > > > > > > > > >> about.
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a
> >> new
> >> > > > version
> >> > > > > > of KIP
> >> > > > > > > > > which
> >> > > > > > > > > > > >> > includes
> >> > > > > > > > > > > >> > > > > >> more
> >> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> >> > > > "Supporting
> >> > > > > > > > > Online
> >> > > > > > > > > > > >> > Consumer
> >> > > > > > > > > > > >> > > > > Group
> >> > > > > > > > > > > >> > > > > >> > > Upgrade"
> >> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> >> > > > Deprecation,
> >> > > > > > and
> >> > > > > > > > > Migration
> >> > > > > > > > > > > >> > Plan". I
> >> > > > > > > > > > > >> > > > > >> think
> >> > > > > > > > > > > >> > > > > >> > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
> >> few cases
> >> > > > > > now but
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > overall idea
> >> > > > > > > > > > > >> > > > > >> and
> >> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> >> understandable.
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> >> ids. So
> >> > > > far as
> >> > > > > > > > > KIP-516 the
> >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > >> > > > > >> are
> >> > > > > > > > > > > >> > > > > >> > > only
> >> > > > > > > > > > > >> > > > > >> > > > > used
> >> > > > > > > > > > > >> > > > > >> > > > > > > as
> >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> >> client,
> >> > > > but
> >> > > > > > they
> >> > > > > > > > > are not
> >> > > > > > > > > > > >> > exposed
> >> > > > > > > > > > > >> > > > > via
> >> > > > > > > > > > > >> > > > > >> any
> >> > > > > > > > > > > >> > > > > >> > > > > public
> >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
> >> the
> >> > > > > > question is,
> >> > > > > > > > > first
> >> > > > > > > > > > > >> > should we
> >> > > > > > > > > > > >> > > > > >> let
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> >> the
> >> > > > names
> >> > > > > > -> ids
> >> > > > > > > > > mapping
> >> > > > > > > > > > > >> > itself
> >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> > fully
> >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> >> existing
> >> > > > RPCs
> >> > > > > > and the
> >> > > > > > > > > new
> >> > > > > > > > > > > >> RPCs
> >> > > > > > > > > > > >> > with
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > >> > > > > >> > > > > ids;
> >> > > > > > > > > > > >> > > > > >> > > > > > > and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> >> consider
> >> > > > > > exposing
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > >> > > > > in
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > consumer
> >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> >> > > > > > > > > subscribe/assign, as
> >> > > > > > > > > > > >> well
> >> > > > > > > > > > > >> > as in
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> >> topic
> >> > > > > > > > > > > >> > deletion-and-recreation).
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> >> include
> >> > > > > > converting
> >> > > > > > > > > all
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > offsets
> >> > > > > > > > > > > >> > > > > >> > > related
> >> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in
> >> this
> >> > > > KIP,
> >> > > > > > the
> >> > > > > > > > > consumer
> >> > > > > > > > > > > >> > would be
> >> > > > > > > > > > > >> > > > > >> able
> >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
> >> ids. That
> >> > > > > > being
> >> > > > > > > > > said, it
> >> > > > > > > > > > > >> > still has
> >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> > > provide
> >> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in various
> >> APIs so
> >> > > > > > having a
> >> > > > > > > > > > > >> mapping
> >> > > > > > > > > > > >> > in the
> >> > > > > > > > > > > >> > > > > >> > > consumer
> >> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> >> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> >> opinion on
> >> > > > > > this.
> >> > > > > > > > > Here I
> >> > > > > > > > > > > >> > wonder if
> >> > > > > > > > > > > >> > > > > >> this
> >> > > > > > > > > > > >> > > > > >> > > goes
> >> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
> >> KIP. I
> >> > > > would
> >> > > > > > rather
> >> > > > > > > > > focus
> >> > > > > > > > > > > >> on
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> > internals
> >> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider
> >> this
> >> > > > > > separately if
> >> > > > > > > > > we see
> >> > > > > > > > > > > >> > value in
> >> > > > > > > > > > > >> > > > > >> doing
> >> > > > > > > > > > > >> > > > > >> > > it.
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> >> point about
> >> > > > > > using
> >> > > > > > > > > topic ids
> >> > > > > > > > > > > >> > in the
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> ConsumerGroupHeartbeatRequest, I
> >> > > > think
> >> > > > > > that
> >> > > > > > > > > there
> >> > > > > > > > > > > >> is
> >> > > > > > > > > > > >> > one
> >> > > > > > > > > > > >> > > > > >> > advantage
> >> > > > > > > > > > > >> > > > > >> > > in
> >> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer
> >> will
> >> > > > have
> >> > > > > > the
> >> > > > > > > > > > > >> opportunity
> >> > > > > > > > > > > >> > to
> >> > > > > > > > > > > >> > > > > >> validate
> >> > > > > > > > > > > >> > > > > >> > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> >> passing
> >> > > > them
> >> > > > > > into
> >> > > > > > > > > the
> >> > > > > > > > > > > >> group
> >> > > > > > > > > > > >> > > > > rebalance
> >> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> >> > > > coordinator
> >> > > > > > will
> >> > > > > > > > > also
> >> > > > > > > > > > > >> notice
> >> > > > > > > > > > > >> > it but
> >> > > > > > > > > > > >> > > > > >> it
> >> > > > > > > > > > > >> > > > > >> > > does
> >> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> >> reject an
> >> > > > > > invalid
> >> > > > > > > > > topic in
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> response.
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> >> on all
> >> > > > other
> >> > > > > > minor
> >> > > > > > > > > > > >> questions
> >> > > > > > > > > > > >> > > > > except
> >> > > > > > > > > > > >> > > > > >> for
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> >> question:
> >> > > > > > personally
> >> > > > > > > > > I think
> >> > > > > > > > > > > >> > it's not
> >> > > > > > > > > > > >> > > > > >> > > necessary
> >> > > > > > > > > > > >> > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> >> API with
> >> > > > > > Pattern,
> >> > > > > > > > > but
> >> > > > > > > > > > > >> > instead we
> >> > > > > > > > > > > >> > > > > >> still
> >> > > > > > > > > > > >> > > > > >> > > use
> >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> >> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> >> that our
> >> > > > > > > > > subscription may
> >> > > > > > > > > > > >> be
> >> > > > > > > > > > > >> > > > > >> rejected by
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> >> case is a
> >> > > > > > very rare
> >> > > > > > > > > > > >> scenario
> >> > > > > > > > > > > >> > I felt
> >> > > > > > > > > > > >> > > > > >> > using
> >> > > > > > > > > > > >> > > > > >> > > an
> >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> >> > > > > > subscription may
> >> > > > > > > > > be
> >> > > > > > > > > > > >> more
> >> > > > > > > > > > > >> > > > > >> vulnerable
> >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > >> > > > > >> > > > > various
> >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
> >> look at
> >> > > > the
> >> > > > > > > > > differences
> >> > > > > > > > > > > >> > between
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > two
> >> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> >> understand the
> >> > > > > > potential
> >> > > > > > > > > issues.
> >> > > > > > > > > > > >> My
> >> > > > > > > > > > > >> > > > > >> > > understanding is
> >> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all the
> >> basic
> >> > > > > > regular
> >> > > > > > > > > > > >> > expressions. The
> >> > > > > > > > > > > >> > > > > >> > > differences
> >> > > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
> >> about
> >> > > > the
> >> > > > > > various
> >> > > > > > > > > > > >> character
> >> > > > > > > > > > > >> > > > > >> classes. I
> >> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
> >> think
> >> > > > about
> >> > > > > > this.
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> >> > > > > > > > > > > >> > > > > >> > > > > > > David
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> >> 11:28 PM
> >> > > > > > Guozhang
> >> > > > > > > > > Wang <
> >> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> >> > > > > > > > > > > >> > > > > >> > > >
> >> > > > > > > > > > > >> > > > > >> > > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on
> >> the high
> >> > > > > > level
> >> > > > > > > > > there are
> >> > > > > > > > > > > >> > two meta
> >> > > > > > > > > > > >> > > > > >> > > points we
> >> > > > > > > > > > > >> > > > > >> > > > > need
> >> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> >> especially
> >> > > > > > the last
> >> > > > > > > > > step
> >> > > > > > > > > > > >> when
> >> > > > > > > > > > > >> > > > > clients
> >> > > > > > > > > > > >> > > > > >> > > flip the
> >> > > > > > > > > > > >> > > > > >> > > > > > > flag
> >> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> >> protocol, in
> >> > > > which
> >> > > > > > we
> >> > > > > > > > > would
> >> > > > > > > > > > > >> have a
> >> > > > > > > > > > > >> > > > > window
> >> > > > > > > > > > > >> > > > > >> > where
> >> > > > > > > > > > > >> > > > > >> > > > > both
> >> > > > > > > > > > > >> > > > > >> > > > > > > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> >> > > > protocols /
> >> > > > > > rpcs
> >> > > > > > > > > are
> >> > > > > > > > > > > >> used
> >> > > > > > > > > > > >> > by
> >> > > > > > > > > > > >> > > > > >> members
> >> > > > > > > > > > > >> > > > > >> > of
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > same
> >> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> >> coordinator could
> >> > > > > > "mimic"
> >> > > > > > > > > the old
> >> > > > > > > > > > > >> > behavior
> >> > > > > > > > > > > >> > > > > >> while
> >> > > > > > > > > > > >> > > > > >> > > > > using the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
> >> we
> >> > > > need to
> >> > > > > > > > > present
> >> > > > > > > > > > > >> about.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> >> ids. So
> >> > > > far as
> >> > > > > > > > > KIP-516 the
> >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > >> > > > > >> are
> >> > > > > > > > > > > >> > > > > >> > > only
> >> > > > > > > > > > > >> > > > > >> > > > > used
> >> > > > > > > > > > > >> > > > > >> > > > > > > as
> >> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> >> client,
> >> > > > but
> >> > > > > > they
> >> > > > > > > > > are not
> >> > > > > > > > > > > >> > exposed
> >> > > > > > > > > > > >> > > > > via
> >> > > > > > > > > > > >> > > > > >> any
> >> > > > > > > > > > > >> > > > > >> > > > > public
> >> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> >> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
> >> the
> >> > > > > > question is,
> >> > > > > > > > > first
> >> > > > > > > > > > > >> > should we
> >> > > > > > > > > > > >> > > > > >> let
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> >> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> >> the
> >> > > > names
> >> > > > > > -> ids
> >> > > > > > > > > mapping
> >> > > > > > > > > > > >> > itself
> >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> > fully
> >> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> >> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> >> existing
> >> > > > RPCs
> >> > > > > > and the
> >> > > > > > > > > new
> >> > > > > > > > > > > >> RPCs
> >> > > > > > > > > > > >> > with
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > >> > > > > >> > > > > ids;
> >> > > > > > > > > > > >> > > > > >> > > > > > > and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> >> consider
> >> > > > > > exposing
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > topic ids
> >> > > > > > > > > > > >> > > > > in
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > consumer
> >> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> >> > > > > > > > > subscribe/assign, as
> >> > > > > > > > > > > >> well
> >> > > > > > > > > > > >> > as in
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > rebalance
> >> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> >> topic
> >> > > > > > > > > > > >> > deletion-and-recreation).
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> >> on all
> >> > > > other
> >> > > > > > minor
> >> > > > > > > > > > > >> questions
> >> > > > > > > > > > > >> > > > > except
> >> > > > > > > > > > > >> > > > > >> for
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> >> question:
> >> > > > > > personally
> >> > > > > > > > > I think
> >> > > > > > > > > > > >> > it's not
> >> > > > > > > > > > > >> > > > > >> > > necessary
> >> > > > > > > > > > > >> > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> >> API with
> >> > > > > > Pattern,
> >> > > > > > > > > but
> >> > > > > > > > > > > >> > instead we
> >> > > > > > > > > > > >> > > > > >> still
> >> > > > > > > > > > > >> > > > > >> > > use
> >> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> >> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> >> that our
> >> > > > > > > > > subscription may
> >> > > > > > > > > > > >> be
> >> > > > > > > > > > > >> > > > > >> rejected by
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > server.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> >> case is a
> >> > > > > > very rare
> >> > > > > > > > > > > >> scenario
> >> > > > > > > > > > > >> > I felt
> >> > > > > > > > > > > >> > > > > >> > using
> >> > > > > > > > > > > >> > > > > >> > > an
> >> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> >> > > > > > subscription may
> >> > > > > > > > > be
> >> > > > > > > > > > > >> more
> >> > > > > > > > > > > >> > > > > >> vulnerable
> >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > >> > > > > >> > > > > various
> >> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
> >> 5:23 AM
> >> > > > > > David Jacot
> >> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> >> feedback. Let
> >> > > > me
> >> > > > > > answer
> >> > > > > > > > > your
> >> > > > > > > > > > > >> > questions
> >> > > > > > > > > > > >> > > > > >> > inline.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> >> premature to
> >> > > > > > talk about
> >> > > > > > > > > > > >> target
> >> > > > > > > > > > > >> > > > > versions
> >> > > > > > > > > > > >> > > > > >> for
> >> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> >> existing group
> >> > > > > > protocol.
> >> > > > > > > > > > > >> Unlike
> >> > > > > > > > > > > >> > KRaft,
> >> > > > > > > > > > > >> > > > > >> this
> >> > > > > > > > > > > >> > > > > >> > > > > affects a
> >> > > > > > > > > > > >> > > > > >> > > > > > > core
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
> >> hence
> >> > > > > > > > > deprecation/removal
> >> > > > > > > > > > > >> > will be
> >> > > > > > > > > > > >> > > > > >> > heavily
> >> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> >> applications
> >> > > > migrate
> >> > > > > > to
> >> > > > > > > > > the new
> >> > > > > > > > > > > >> > protocol.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> >> > > > remove it.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> >> intend to
> >> > > > > > release
> >> > > > > > > > > this in
> >> > > > > > > > > > > >> > 4.x, but
> >> > > > > > > > > > > >> > > > > it
> >> > > > > > > > > > > >> > > > > >> > > wasn't
> >> > > > > > > > > > > >> > > > > >> > > > > made
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that
> >> as a
> >> > > > way to
> >> > > > > > > > > estimate
> >> > > > > > > > > > > >> when
> >> > > > > > > > > > > >> > we'd
> >> > > > > > > > > > > >> > > > > >> > > deprecate
> >> > > > > > > > > > > >> > > > > >> > > > > and
> >> > > > > > > > > > > >> > > > > >> > > > > > > remove
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I
> >> also
> >> > > > > > suggest
> >> > > > > > > > > removing
> >> > > > > > > > > > > >> > this part.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> >> reasoning. As
> >> > > > > > > > > explained, I
> >> > > > > > > > > > > >> plan
> >> > > > > > > > > > > >> > to
> >> > > > > > > > > > > >> > > > > >> rewrite
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> >> while we
> >> > > > > > implement
> >> > > > > > > > > the new
> >> > > > > > > > > > > >> > protocol.
> >> > > > > > > > > > > >> > > > > >> This
> >> > > > > > > > > > > >> > > > > >> > > means
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals will
> >> be
> >> > > > slightly
> >> > > > > > > > > different
> >> > > > > > > > > > > >> > (e.g.
> >> > > > > > > > > > > >> > > > > >> threading
> >> > > > > > > > > > > >> > > > > >> > > > > model).
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
> >> tighten
> >> > > > the
> >> > > > > > > > > switch from
> >> > > > > > > > > > > >> > the old
> >> > > > > > > > > > > >> > > > > >> group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
> >> group
> >> > > > > > coordinator
> >> > > > > > > > > to a
> >> > > > > > > > > > > >> > major
> >> > > > > > > > > > > >> > > > > >> release.
> >> > > > > > > > > > > >> > > > > >> > > The
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to
> >> use a
> >> > > > flag
> >> > > > > > to do
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > switch
> >> > > > > > > > > > > >> > > > > >> instead
> >> > > > > > > > > > > >> > > > > >> > of
> >> > > > > > > > > > > >> > > > > >> > > > > relying
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
> >> out the
> >> > > > > > details of
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > migration
> >> > > > > > > > > > > >> > > > > >> story.
> >> > > > > > > > > > > >> > > > > >> > > It
> >> > > > > > > > > > > >> > > > > >> > > > > sounds
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
> >> support
> >> > > > > > online
> >> > > > > > > > > > > >> migrations.
> >> > > > > > > > > > > >> > Is
> >> > > > > > > > > > > >> > > > > that
> >> > > > > > > > > > > >> > > > > >> > > correct?
> >> > > > > > > > > > > >> > > > > >> > > > > We
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> >> detail. It
> >> > > > could
> >> > > > > > also
> >> > > > > > > > > be done
> >> > > > > > > > > > > >> > as a
> >> > > > > > > > > > > >> > > > > >> separate
> >> > > > > > > > > > > >> > > > > >> > > KIP,
> >> > > > > > > > > > > >> > > > > >> > > > > if
> >> > > > > > > > > > > >> > > > > >> > > > > > > it's
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
> >> online
> >> > > > > > migrations
> >> > > > > > > > > for the
> >> > > > > > > > > > > >> > group.
> >> > > > > > > > > > > >> > > > > That
> >> > > > > > > > > > > >> > > > > >> > > means
> >> > > > > > > > > > > >> > > > > >> > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> >> protocol
> >> > > > > > will be
> >> > > > > > > > > able to
> >> > > > > > > > > > > >> > switch to
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain
> >> how that
> >> > > > > > will work
> >> > > > > > > > > > > >> though.
> >> > > > > > > > > > > >> > It is
> >> > > > > > > > > > > >> > > > > >> > > basically a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> >> > > > upgraded or
> >> > > > > > > > > rolled to a
> >> > > > > > > > > > > >> > software
> >> > > > > > > > > > > >> > > > > >> > > supporting
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
> >> Both the
> >> > > > > > old and
> >> > > > > > > > > the new
> >> > > > > > > > > > > >> > > > > >> coordinator
> >> > > > > > > > > > > >> > > > > >> > > will
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> >> protocol and
> >> > > > rely
> >> > > > > > on the
> >> > > > > > > > > same
> >> > > > > > > > > > > >> > persisted
> >> > > > > > > > > > > >> > > > > >> > > metadata so
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work together.
> >> This
> >> > > > point
> >> > > > > > is an
> >> > > > > > > > > > > >> offline
> >> > > > > > > > > > > >> > > > > >> migration.
> >> > > > > > > > > > > >> > > > > >> > We
> >> > > > > > > > > > > >> > > > > >> > > > > cannot
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> >> because it
> >> > > > would
> >> > > > > > require
> >> > > > > > > > > > > >> > shutting down
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > current
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> >> starting up the
> >> > > > > > new one
> >> > > > > > > > > and
> >> > > > > > > > > > > >> that
> >> > > > > > > > > > > >> > would
> >> > > > > > > > > > > >> > > > > >> cause
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> >> metadata
> >> > > > > > version/IBP
> >> > > > > > > > > must be
> >> > > > > > > > > > > >> > upgraded
> >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> X
> >> > > > > > > > > > > >> > > > > >> > in
> >> > > > > > > > > > > >> > > > > >> > > > > order
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> >> protocol.
> >> > > > This
> >> > > > > > cannot
> >> > > > > > > > > be done
> >> > > > > > > > > > > >> > before
> >> > > > > > > > > > > >> > > > > 1)
> >> > > > > > > > > > > >> > > > > >> is
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the
> >> old
> >> > > > > > coordinator
> >> > > > > > > > > doesn't
> >> > > > > > > > > > > >> > support
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> >> > > > upgraded
> >> > > > > > to a
> >> > > > > > > > > version
> >> > > > > > > > > > > >> > > > > supporting
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > online
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> >> KIP-792).
> >> > > > If
> >> > > > > > the
> >> > > > > > > > > consumer
> >> > > > > > > > > > > >> is
> >> > > > > > > > > > > >> > > > > already
> >> > > > > > > > > > > >> > > > > >> > > there.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at
> >> this
> >> > > > point.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must
> >> be rolled
> >> > > > > > with the
> >> > > > > > > > > > > >> feature
> >> > > > > > > > > > > >> > flag
> >> > > > > > > > > > > >> > > > > >> turned
> >> > > > > > > > > > > >> > > > > >> > > on.
> >> > > > > > > > > > > >> > > > > >> > > > > The
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> >> automatically
> >> > > > > > > > > converted when
> >> > > > > > > > > > > >> > the first
> >> > > > > > > > > > > >> > > > > >> > > consumer
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
> >> joins
> >> > > > the
> >> > > > > > group.
> >> > > > > > > > > While
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > members
> >> > > > > > > > > > > >> > > > > >> > > using the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> >> > > > upgraded,
> >> > > > > > the old
> >> > > > > > > > > > > >> protocol
> >> > > > > > > > > > > >> > is
> >> > > > > > > > > > > >> > > > > >> proxied
> >> > > > > > > > > > > >> > > > > >> > > into
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
> >> this in
> >> > > > the
> >> > > > > > KIP.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we
> >> are
> >> > > > > > pushing the
> >> > > > > > > > > pattern
> >> > > > > > > > > > > >> > > > > >> subscriptions
> >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
> >> there
> >> > > > could be
> >> > > > > > some
> >> > > > > > > > > tricky
> >> > > > > > > > > > > >> > > > > >> compatibility
> >> > > > > > > > > > > >> > > > > >> > > > > issues.
> >> > > > > > > > > > > >> > > > > >> > > > > > > Will
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
> >> users to
> >> > > > > > detect
> >> > > > > > > > > that
> >> > > > > > > > > > > >> they
> >> > > > > > > > > > > >> > need to
> >> > > > > > > > > > > >> > > > > >> > update
> >> > > > > > > > > > > >> > > > > >> > > > > their
> >> > > > > > > > > > > >> > > > > >> > > > > > > regex
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
> >> the new
> >> > > > > > protocol?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit
> >> more
> >> > > > > > optimistic
> >> > > > > > > > > than
> >> > > > > > > > > > > >> you
> >> > > > > > > > > > > >> > on this
> >> > > > > > > > > > > >> > > > > >> > > point. I
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> >> majority of
> >> > > > the
> >> > > > > > cases
> >> > > > > > > > > are
> >> > > > > > > > > > > >> simple
> >> > > > > > > > > > > >> > > > > regexes
> >> > > > > > > > > > > >> > > > > >> > which
> >> > > > > > > > > > > >> > > > > >> > > > > should
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> >> engine. The
> >> > > > > > > > > coordinator will
> >> > > > > > > > > > > >> > verify
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > regex
> >> > > > > > > > > > > >> > > > > >> > > > > anyway
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer
> >> if the
> >> > > > > > regex is
> >> > > > > > > > > not
> >> > > > > > > > > > > >> valid.
> >> > > > > > > > > > > >> > > > > Coming
> >> > > > > > > > > > > >> > > > > >> > back
> >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
> >> worst
> >> > > > case,
> >> > > > > > the
> >> > > > > > > > > first
> >> > > > > > > > > > > >> > upgraded
> >> > > > > > > > > > > >> > > > > >> > consumer
> >> > > > > > > > > > > >> > > > > >> > > > > joining
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> >> rejected.
> >> > > > This
> >> > > > > > should
> >> > > > > > > > > be used
> >> > > > > > > > > > > >> > as the
> >> > > > > > > > > > > >> > > > > >> last
> >> > > > > > > > > > > >> > > > > >> > > > > defence, I
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> >> > > > validate
> >> > > > > > their
> >> > > > > > > > > regex
> >> > > > > > > > > > > >> > before
> >> > > > > > > > > > > >> > > > > >> > upgrading
> >> > > > > > > > > > > >> > > > > >> > > > > their
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
> >> them with
> >> > > > > > another
> >> > > > > > > > > group.
> >> > > > > > > > > > > >> For
> >> > > > > > > > > > > >> > > > > >> instance,
> >> > > > > > > > > > > >> > > > > >> > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> >> pre-prod
> >> > > > > > environment.
> >> > > > > > > > > > > >> Another
> >> > > > > > > > > > > >> > way
> >> > > > > > > > > > > >> > > > > >> would be
> >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> >> consumer-group tool
> >> > > > to
> >> > > > > > > > > provide a
> >> > > > > > > > > > > >> regex
> >> > > > > > > > > > > >> > > > > >> validation
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> >> > > > enough in
> >> > > > > > your
> >> > > > > > > > > > > >> opinion?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> >> > > > question,
> >> > > > > > will
> >> > > > > > > > > the
> >> > > > > > > > > > > >> Java
> >> > > > > > > > > > > >> > client
> >> > > > > > > > > > > >> > > > > >> allow
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > users to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> >> current regex
> >> > > > > > engine for
> >> > > > > > > > > > > >> > compatibility
> >> > > > > > > > > > > >> > > > > >> > > reasons?
> >> > > > > > > > > > > >> > > > > >> > > > > For
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
> >> keep using
> >> > > > > > client
> >> > > > > > > > > based
> >> > > > > > > > > > > >> > regex at
> >> > > > > > > > > > > >> > > > > >> first
> >> > > > > > > > > > > >> > > > > >> > to
> >> > > > > > > > > > > >> > > > > >> > > keep
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and
> >> then
> >> > > > > > migrate to
> >> > > > > > > > > server
> >> > > > > > > > > > > >> > based
> >> > > > > > > > > > > >> > > > > >> regexes
> >> > > > > > > > > > > >> > > > > >> > > as a
> >> > > > > > > > > > > >> > > > > >> > > > > > > second
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your point
> >> but I
> >> > > > am
> >> > > > > > > > > concerned that
> >> > > > > > > > > > > >> > this
> >> > > > > > > > > > > >> > > > > would
> >> > > > > > > > > > > >> > > > > >> > > allow
> >> > > > > > > > > > > >> > > > > >> > > > > users
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
> >> this mode.
> >> > > > That
> >> > > > > > > > > would go
> >> > > > > > > > > > > >> > against our
> >> > > > > > > > > > > >> > > > > >> goal
> >> > > > > > > > > > > >> > > > > >> > > of
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
> >> because
> >> > > > we
> >> > > > > > would
> >> > > > > > > > > have to
> >> > > > > > > > > > > >> > continue
> >> > > > > > > > > > > >> > > > > >> > > monitoring
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> >> client
> >> > > > side. I
> >> > > > > > would
> >> > > > > > > > > rather
> >> > > > > > > > > > > >> > not do
> >> > > > > > > > > > > >> > > > > >> this.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
> >> the group
> >> > > > > > > > > coordinator
> >> > > > > > > > > > > >> will
> >> > > > > > > > > > > >> > be
> >> > > > > > > > > > > >> > > > > >> > > responsible for
> >> > > > > > > > > > > >> > > > > >> > > > > > > storing
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations
> >> and that
> >> > > > the
> >> > > > > > > > > > > >> configurations
> >> > > > > > > > > > > >> > will be
> >> > > > > > > > > > > >> > > > > >> > > deleted
> >> > > > > > > > > > > >> > > > > >> > > > > when
> >> > > > > > > > > > > >> > > > > >> > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
> >> Will a
> >> > > > > > transition to
> >> > > > > > > > > DEAD
> >> > > > > > > > > > > >> > trigger
> >> > > > > > > > > > > >> > > > > >> > deletion
> >> > > > > > > > > > > >> > > > > >> > > of
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> >> > > > configurations
> >> > > > > > will be
> >> > > > > > > > > > > >> deleted
> >> > > > > > > > > > > >> > when
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > group is
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> >> together.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to
> >> store
> >> > > > the
> >> > > > > > > > > configs in
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > group
> >> > > > > > > > > > > >> > > > > >> > > coordinator
> >> > > > > > > > > > > >> > > > > >> > > > > > > make it
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> >> cluster
> >> > > > > > configs and
> >> > > > > > > > > their
> >> > > > > > > > > > > >> > values?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
> >> group
> >> > > > > > > > > configurations are
> >> > > > > > > > > > > >> > overrides
> >> > > > > > > > > > > >> > > > > >> of
> >> > > > > > > > > > > >> > > > > >> > > cluster
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to
> >> know
> >> > > > all
> >> > > > > > the
> >> > > > > > > > > overrides
> >> > > > > > > > > > > >> > though,
> >> > > > > > > > > > > >> > > > > you
> >> > > > > > > > > > > >> > > > > >> > > would
> >> > > > > > > > > > > >> > > > > >> > > > > have
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> >> > > > coordinators.
> >> > > > > > You
> >> > > > > > > > > cannot
> >> > > > > > > > > > > >> > rely on
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > metadata
> >> > > > > > > > > > > >> > > > > >> > > > > log
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> >> > > > configure a
> >> > > > > > group
> >> > > > > > > > > before
> >> > > > > > > > > > > >> > starting
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > consumers?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> >> allowing the
> >> > > > > > explicit
> >> > > > > > > > > creation
> >> > > > > > > > > > > >> of
> >> > > > > > > > > > > >> > > > > groups?
> >> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> >> decoupled
> >> > > > from
> >> > > > > > the
> >> > > > > > > > > group
> >> > > > > > > > > > > >> > lifecycle.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> >> > > > > > automatically
> >> > > > > > > > > created in
> >> > > > > > > > > > > >> > this
> >> > > > > > > > > > > >> > > > > case.
> >> > > > > > > > > > > >> > > > > >> > > However,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
> >> lost after
> >> > > > the
> >> > > > > > > > > retention
> >> > > > > > > > > > > >> > period of
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> >> > > > Consumer.subscribe
> >> > > > > > method
> >> > > > > > > > > for the
> >> > > > > > > > > > > >> > Java
> >> > > > > > > > > > > >> > > > > >> client
> >> > > > > > > > > > > >> > > > > >> > > still
> >> > > > > > > > > > > >> > > > > >> > > > > take
> >> > > > > > > > > > > >> > > > > >> > > > > > > a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >> `java.util.regex.Pattern` of
> >> > > > do
> >> > > > > > we
> >> > > > > > > > > have to
> >> > > > > > > > > > > >> > introduce
> >> > > > > > > > > > > >> > > > > an
> >> > > > > > > > > > > >> > > > > >> > > overload?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> >> question. I
> >> > > > > > forgot
> >> > > > > > > > > about
> >> > > > > > > > > > > >> that
> >> > > > > > > > > > > >> > one.
> >> > > > > > > > > > > >> > > > > As
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> `java.util.regex.Pattern` is
> >> > > > not
> >> > > > > > fully
> >> > > > > > > > > > > >> compatible
> >> > > > > > > > > > > >> > with
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > engine
> >> > > > > > > > > > > >> > > > > >> > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> >> might be
> >> > > > better
> >> > > > > > to
> >> > > > > > > > > deprecate
> >> > > > > > > > > > > >> > it and
> >> > > > > > > > > > > >> > > > > >> use an
> >> > > > > > > > > > > >> > > > > >> > > > > overload
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string.
> >> We would
> >> > > > > > rely on
> >> > > > > > > > > the
> >> > > > > > > > > > > >> server
> >> > > > > > > > > > > >> > side
> >> > > > > > > > > > > >> > > > > >> > > validation.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I
> >> think
> >> > > > that
> >> > > > > > we
> >> > > > > > > > > could
> >> > > > > > > > > > > >> still
> >> > > > > > > > > > > >> > try to
> >> > > > > > > > > > > >> > > > > >> > > toString
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
> >> should
> >> > > > > > work, I
> >> > > > > > > > > think, in
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> majority
> >> > > > > > > > > > > >> > > > > >> > > of the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> >> Justine
> >> > > > that we
> >> > > > > > > > > should be
> >> > > > > > > > > > > >> > clearer
> >> > > > > > > > > > > >> > > > > about
> >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> > > > > reason
> >> > > > > > > > > > > >> > > > > >> > > > > > > to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> >> > > > IBP/metadata.version
> >> > > > > > from the
> >> > > > > > > > > > > >> feature
> >> > > > > > > > > > > >> > flag.
> >> > > > > > > > > > > >> > > > > >> Maybe
> >> > > > > > > > > > > >> > > > > >> > > we
> >> > > > > > > > > > > >> > > > > >> > > > > mean
> >> > > > > > > > > > > >> > > > > >> > > > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> >> default for
> >> > > > the
> >> > > > > > feature
> >> > > > > > > > > flag
> >> > > > > > > > > > > >> to
> >> > > > > > > > > > > >> > true
> >> > > > > > > > > > > >> > > > > >> based
> >> > > > > > > > > > > >> > > > > >> > on
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once
> >> we
> >> > > > want to
> >> > > > > > make
> >> > > > > > > > > it the
> >> > > > > > > > > > > >> > default.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that
> >> feature
> >> > > > > > flag
> >> > > > > > > > > mainly
> >> > > > > > > > > > > >> > during the
> >> > > > > > > > > > > >> > > > > >> > > development
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> >> > > > mentioned
> >> > > > > > it, I
> >> > > > > > > > > think,
> >> > > > > > > > > > > >> > because
> >> > > > > > > > > > > >> > > > > we
> >> > > > > > > > > > > >> > > > > >> > > could
> >> > > > > > > > > > > >> > > > > >> > > > > use
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for
> >> it.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> >> protocol APIs
> >> > > > > > don't
> >> > > > > > > > > mention
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > required
> >> > > > > > > > > > > >> > > > > >> > > ACLs, it
> >> > > > > > > > > > > >> > > > > >> > > > > > > would
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> >> > > > consistency.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
> >> that
> >> > > > > > > > > > > >> ConsumerGroupHeartbeat
> >> > > > > > > > > > > >> > > > > requires
> >> > > > > > > > > > > >> > > > > >> > > "Read
> >> > > > > > > > > > > >> > > > > >> > > > > Group"
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do
> >> more
> >> > > > than
> >> > > > > > > > > reading.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> >> > > > current
> >> > > > > > > > > protocol works
> >> > > > > > > > > > > >> > though.
> >> > > > > > > > > > > >> > > > > We
> >> > > > > > > > > > > >> > > > > >> > only
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to
> >> join a
> >> > > > > > group. We
> >> > > > > > > > > could
> >> > > > > > > > > > > >> > consider
> >> > > > > > > > > > > >> > > > > >> > > changing
> >> > > > > > > > > > > >> > > > > >> > > > > this
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that
> >> it is
> >> > > > worth
> >> > > > > > it.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> >> recreation
> >> > > > > > handled by
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > consumer
> >> > > > > > > > > > > >> > > > > with
> >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> > > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be
> >> good to
> >> > > > > > have a
> >> > > > > > > > > section
> >> > > > > > > > > > > >> on
> >> > > > > > > > > > > >> > this.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> >> > > > > > perspective, the
> >> > > > > > > > > new
> >> > > > > > > > > > > >> topic
> >> > > > > > > > > > > >> > will
> >> > > > > > > > > > > >> > > > > >> have a
> >> > > > > > > > > > > >> > > > > >> > > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> >> treat it
> >> > > > like a
> >> > > > > > > > > topic with
> >> > > > > > > > > > > >> a
> >> > > > > > > > > > > >> > > > > different
> >> > > > > > > > > > > >> > > > > >> > > name.
> >> > > > > > > > > > > >> > > > > >> > > > > The
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> >> > > > fetch/commit
> >> > > > > > > > > offsets
> >> > > > > > > > > > > >> APIs
> >> > > > > > > > > > > >> > do not
> >> > > > > > > > > > > >> > > > > >> > support
> >> > > > > > > > > > > >> > > > > >> > > > > topic
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
> >> would
> >> > > > reuse the
> >> > > > > > > > > offsets
> >> > > > > > > > > > > >> based
> >> > > > > > > > > > > >> > on the
> >> > > > > > > > > > > >> > > > > >> > same.
> >> > > > > > > > > > > >> > > > > >> > > I
> >> > > > > > > > > > > >> > > > > >> > > > > think
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
> >> those
> >> > > > APIs
> >> > > > > > as well
> >> > > > > > > > > in
> >> > > > > > > > > > > >> order
> >> > > > > > > > > > > >> > to be
> >> > > > > > > > > > > >> > > > > >> > > consistent
> >> > > > > > > > > > > >> > > > > >> > > > > end
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> >> strengthen
> >> > > > the
> >> > > > > > > > > semantics of
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > >> consumer.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions
> >> we will
> >> > > > > > write the
> >> > > > > > > > > new
> >> > > > > > > > > > > >> > coordinator
> >> > > > > > > > > > > >> > > > > >> in
> >> > > > > > > > > > > >> > > > > >> > > Java.
> >> > > > > > > > > > > >> > > > > >> > > > > Even
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> >> implementation
> >> > > > > > detail, do
> >> > > > > > > > > we plan
> >> > > > > > > > > > > >> to
> >> > > > > > > > > > > >> > have a
> >> > > > > > > > > > > >> > > > > >> new
> >> > > > > > > > > > > >> > > > > >> > > gradle
> >> > > > > > > > > > > >> > > > > >> > > > > > > module
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> >> scalability
> >> > > > > > goal when
> >> > > > > > > > > it
> >> > > > > > > > > > > >> comes
> >> > > > > > > > > > > >> > to how
> >> > > > > > > > > > > >> > > > > >> many
> >> > > > > > > > > > > >> > > > > >> > > > > members
> >> > > > > > > > > > > >> > > > > >> > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> >> support?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers
> >> at the
> >> > > > > > moment. The
> >> > > > > > > > > > > >> protocol
> >> > > > > > > > > > > >> > should
> >> > > > > > > > > > > >> > > > > >> > > support
> >> > > > > > > > > > > >> > > > > >> > > > > 1000s
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group.
> >> We will
> >> > > > > > measure
> >> > > > > > > > > this when
> >> > > > > > > > > > > >> > we have
> >> > > > > > > > > > > >> > > > > a
> >> > > > > > > > > > > >> > > > > >> > first
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
> >> that we
> >> > > > might
> >> > > > > > have
> >> > > > > > > > > other
> >> > > > > > > > > > > >> > > > > bottlenecks
> >> > > > > > > > > > > >> > > > > >> > down
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> >> commits).
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> >> having
> >> > > > > > > > > SubscribedTopidIds
> >> > > > > > > > > > > >> > instead
> >> > > > > > > > > > > >> > > > > of
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames
> >> in
> >> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> >> > > > > > > > > > > >> > > > > >> Is
> >> > > > > > > > > > > >> > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> > > > > idea
> >> > > > > > > > > > > >> > > > > >> > > > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> >> resolve the
> >> > > > > > regex on
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > server, we
> >> > > > > > > > > > > >> > > > > >> can do
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > same
> >> > > > > > > > > > > >> > > > > >> > > > > > > for
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> >> > > > difference
> >> > > > > > is that
> >> > > > > > > > > > > >> sending
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> regex is
> >> > > > > > > > > > > >> > > > > >> > > more
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
> >> topic
> >> > > > names
> >> > > > > > is less
> >> > > > > > > > > > > >> > efficient.
> >> > > > > > > > > > > >> > > > > >> > > Furthermore,
> >> > > > > > > > > > > >> > > > > >> > > > > > > delete
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier
> >> to
> >> > > > handle
> >> > > > > > if we
> >> > > > > > > > > have
> >> > > > > > > > > > > >> topic
> >> > > > > > > > > > > >> > ids.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> >> consolidate the
> >> > > > > > metadata
> >> > > > > > > > > lookup
> >> > > > > > > > > > > >> > on the
> >> > > > > > > > > > > >> > > > > >> server
> >> > > > > > > > > > > >> > > > > >> > > for
> >> > > > > > > > > > > >> > > > > >> > > > > both
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
> >> with your
> >> > > > > > point. As
> >> > > > > > > > > a
> >> > > > > > > > > > > >> second
> >> > > > > > > > > > > >> > > > > though,
> >> > > > > > > > > > > >> > > > > >> > using
> >> > > > > > > > > > > >> > > > > >> > > > > topic
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here
> >> for the
> >> > > > > > delete and
> >> > > > > > > > > > > >> > recreation
> >> > > > > > > > > > > >> > > > > case.
> >> > > > > > > > > > > >> > > > > >> > > Also, I
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
> >> allow
> >> > > > users to
> >> > > > > > > > > subscribe
> >> > > > > > > > > > > >> with
> >> > > > > > > > > > > >> > topic
> >> > > > > > > > > > > >> > > > > >> ids
> >> > > > > > > > > > > >> > > > > >> > in
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that is
> >> the only
> >> > > > > > way to be
> >> > > > > > > > > > > >> really
> >> > > > > > > > > > > >> > robust
> >> > > > > > > > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> > > topic
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at
> >> 1:38 PM
> >> > > > > > David
> >> > > > > > > > > Jacot <
> >> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> >> comments.
> >> > > > Please
> >> > > > > > find
> >> > > > > > > > > my
> >> > > > > > > > > > > >> answers
> >> > > > > > > > > > > >> > > > > below.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> >> protocol
> >> > > > relies on
> >> > > > > > > > > topic IDs
> >> > > > > > > > > > > >> > with the
> >> > > > > > > > > > > >> > > > > >> > > exception
> >> > > > > > > > > > > >> > > > > >> > > > > of the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in
> >> the
> >> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> >> > > > > > > > > > > >> > > > > >> I
> >> > > > > > > > > > > >> > > > > >> > am
> >> > > > > > > > > > > >> > > > > >> > > not
> >> > > > > > > > > > > >> > > > > >> > > > > sure
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names
> >> is the
> >> > > > > > right call
> >> > > > > > > > > here.
> >> > > > > > > > > > > >> I
> >> > > > > > > > > > > >> > need to
> >> > > > > > > > > > > >> > > > > >> > think
> >> > > > > > > > > > > >> > > > > >> > > > > about it
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> >> Obviously, the
> >> > > > > > KIP does
> >> > > > > > > > > not
> >> > > > > > > > > > > >> > change the
> >> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
> >> topic
> >> > > > IDs.
> >> > > > > > This
> >> > > > > > > > > may be
> >> > > > > > > > > > > >> > something
> >> > > > > > > > > > > >> > > > > >> that
> >> > > > > > > > > > > >> > > > > >> > we
> >> > > > > > > > > > > >> > > > > >> > > > > should
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
> >> would
> >> > > > give
> >> > > > > > better
> >> > > > > > > > > > > >> overall
> >> > > > > > > > > > > >> > > > > >> guarantee in
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> >> think that
> >> > > > I
> >> > > > > > should
> >> > > > > > > > > not
> >> > > > > > > > > > > >> have
> >> > > > > > > > > > > >> > > > > mentioned
> >> > > > > > > > > > > >> > > > > >> > this
> >> > > > > > > > > > > >> > > > > >> > > > > flag at
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove
> >> it. We can
> >> > > > > > use an
> >> > > > > > > > > internal
> >> > > > > > > > > > > >> > > > > >> configuration
> >> > > > > > > > > > > >> > > > > >> > > while
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> >> feature.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
> >> will be
> >> > > > > > > > > supported. The
> >> > > > > > > > > > > >> > change is
> >> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
> >> that the
> >> > > > > > cluster
> >> > > > > > > > > uses
> >> > > > > > > > > > > >> topic
> >> > > > > > > > > > > >> > IDs.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022
> >> at 9:53
> >> > > > PM
> >> > > > > > > > > Guozhang
> >> > > > > > > > > > > >> Wang <
> >> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> >> feedback.
> >> > > > Here
> >> > > > > > are
> >> > > > > > > > > some
> >> > > > > > > > > > > >> replies
> >> > > > > > > > > > > >> > > > > inlined
> >> > > > > > > > > > > >> > > > > >> > > below:
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022
> >> at
> >> > > > 2:53 AM
> >> > > > > > > > > Ismael Juma
> >> > > > > > > > > > > >> <
> >> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> >> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
> >> KIP. This
> >> > > > has
> >> > > > > > the
> >> > > > > > > > > > > >> potential
> >> > > > > > > > > > > >> > to be a
> >> > > > > > > > > > > >> > > > > >> > great
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> >> > > > questions/comments:
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> >> > > > premature to
> >> > > > > > talk
> >> > > > > > > > > about
> >> > > > > > > > > > > >> > target
> >> > > > > > > > > > > >> > > > > >> versions
> >> > > > > > > > > > > >> > > > > >> > > for
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> >> existing
> >> > > > group
> >> > > > > > > > > protocol.
> >> > > > > > > > > > > >> > Unlike
> >> > > > > > > > > > > >> > > > > >> KRaft,
> >> > > > > > > > > > > >> > > > > >> > > this
> >> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
> >> and hence
> >> > > > > > > > > > > >> > deprecation/removal will
> >> > > > > > > > > > > >> > > > > >> be
> >> > > > > > > > > > > >> > > > > >> > > heavily
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> >> applications
> >> > > > > > migrate
> >> > > > > > > > > to the
> >> > > > > > > > > > > >> new
> >> > > > > > > > > > > >> > > > > >> protocol.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
> >> you. I
> >> > > > think
> >> > > > > > we can
> >> > > > > > > > > > > >> remove
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> proposed
> >> > > > > > > > > > > >> > > > > >> > > > > timeline
> >> > > > > > > > > > > >> > > > > >> > > > > > > in
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> >> > > > Deprecation,
> >> > > > > > and
> >> > > > > > > > > Migration
> >> > > > > > > > > > > >> > Plan` and
> >> > > > > > > > > > > >> > > > > >> > > instead
> >> > > > > > > > > > > >> > > > > >> > > > > just
> >> > > > > > > > > > > >> > > > > >> > > > > > > state
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide
> >> in the
> >> > > > > > future
> >> > > > > > > > > about
> >> > > > > > > > > > > >> when
> >> > > > > > > > > > > >> > we
> >> > > > > > > > > > > >> > > > > would
> >> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> >> behaviors.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> >> > > > intend to
> >> > > > > > > > > release this
> >> > > > > > > > > > > >> > in 4.x,
> >> > > > > > > > > > > >> > > > > >> but
> >> > > > > > > > > > > >> > > > > >> > it
> >> > > > > > > > > > > >> > > > > >> > > > > wasn't
> >> > > > > > > > > > > >> > > > > >> > > > > > > made
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
> >> that as
> >> > > > a
> >> > > > > > way to
> >> > > > > > > > > > > >> estimate
> >> > > > > > > > > > > >> > when
> >> > > > > > > > > > > >> > > > > we'd
> >> > > > > > > > > > > >> > > > > >> > > > > deprecate
> >> > > > > > > > > > > >> > > > > >> > > > > > > and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> >> protocol, I
> >> > > > also
> >> > > > > > suggest
> >> > > > > > > > > > > >> removing
> >> > > > > > > > > > > >> > this
> >> > > > > > > > > > > >> > > > > >> part.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> >> > > > specifically
> >> > > > > > > > > related to
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more
> >> for
> >> > > > client
> >> > > > > > > > > upgrades.
> >> > > > > > > > > > > >> I.e.
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> > > broker-side
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may
> >> be done
> >> > > > > > first,
> >> > > > > > > > > and then
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > client
> >> > > > > > > > > > > >> > > > > >> > side,
> >> > > > > > > > > > > >> > > > > >> > > > > and we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> >> "released"
> >> > > > by
> >> > > > > > the
> >> > > > > > > > > time
> >> > > > > > > > > > > >> clients
> >> > > > > > > > > > > >> > > > > >> > > implementations
> >> > > > > > > > > > > >> > > > > >> > > > > are
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> >> enable the
> >> > > > > > feature the
> >> > > > > > > > > > > >> clients
> >> > > > > > > > > > > >> > need to
> >> > > > > > > > > > > >> > > > > >> > first
> >> > > > > > > > > > > >> > > > > >> > > > > swap-in
> >> > > > > > > > > > > >> > > > > >> > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> >> rolling
> >> > > > bounce
> >> > > > > > and
> >> > > > > > > > > then set
> >> > > > > > > > > > > >> > the flag
> >> > > > > > > > > > > >> > > > > >> > with a
> >> > > > > > > > > > > >> > > > > >> > > > > second
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence
> >> we feel
> >> > > > it's
> >> > > > > > > > > better to be
> >> > > > > > > > > > > >> > released
> >> > > > > > > > > > > >> > > > > >> in a
> >> > > > > > > > > > > >> > > > > >> > > major
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> >> flesh out
> >> > > > the
> >> > > > > > > > > details of
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > migration
> >> > > > > > > > > > > >> > > > > >> > > story.
> >> > > > > > > > > > > >> > > > > >> > > > > It
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
> >> will
> >> > > > support
> >> > > > > > online
> >> > > > > > > > > > > >> > migrations. Is
> >> > > > > > > > > > > >> > > > > >> that
> >> > > > > > > > > > > >> > > > > >> > > > > correct?
> >> > > > > > > > > > > >> > > > > >> > > > > > > We
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> >> detail.
> >> > > > It
> >> > > > > > could
> >> > > > > > > > > also be
> >> > > > > > > > > > > >> > done as a
> >> > > > > > > > > > > >> > > > > >> > > separate
> >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> >> > > > > > > > > > > >> > > > > >> > > > > > > if
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's
> >> the
> >> > > > part we
> >> > > > > > can
> >> > > > > > > > > be more
> >> > > > > > > > > > > >> > concrete
> >> > > > > > > > > > > >> > > > > >> about
> >> > > > > > > > > > > >> > > > > >> > > for
> >> > > > > > > > > > > >> > > > > >> > > > > sure
> >> > > > > > > > > > > >> > > > > >> > > > > > > (and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to
> >> your
> >> > > > > > question 2)
> >> > > > > > > > > above).
> >> > > > > > > > > > > >> > We will
> >> > > > > > > > > > > >> > > > > >> work
> >> > > > > > > > > > > >> > > > > >> > on
> >> > > > > > > > > > > >> > > > > >> > > > > making
> >> > > > > > > > > > > >> > > > > >> > > > > > > it
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> >> parallel as we
> >> > > > > > solicit
> >> > > > > > > > > more
> >> > > > > > > > > > > >> > feedback.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
> >> that we are
> >> > > > > > pushing
> >> > > > > > > > > the
> >> > > > > > > > > > > >> > pattern
> >> > > > > > > > > > > >> > > > > >> > > subscriptions
> >> > > > > > > > > > > >> > > > > >> > > > > to
> >> > > > > > > > > > > >> > > > > >> > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like
> >> there
> >> > > > > > could be
> >> > > > > > > > > some
> >> > > > > > > > > > > >> tricky
> >> > > > > > > > > > > >> > > > > >> > > compatibility
> >> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism
> >> for
> >> > > > users to
> >> > > > > > > > > detect that
> >> > > > > > > > > > > >> > they
> >> > > > > > > > > > > >> > > > > need
> >> > > > > > > > > > > >> > > > > >> to
> >> > > > > > > > > > > >> > > > > >> > > update
> >> > > > > > > > > > > >> > > > > >> > > > > > > their
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching
> >> to the
> >> > > > new
> >> > > > > > > > > protocol?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need
> >> some
> >> > > > > > tooling for
> >> > > > > > > > > > > >> non-java
> >> > > > > > > > > > > >> > client
> >> > > > > > > > > > > >> > > > > >> > users
> >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > >> > > > > >> > > > > sort
> >> > > > > > > > > > > >> > > > > >> > > > > > > of
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> >> client before
> >> > > > > > > > > switching to
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > new
> >> > > > > > > > > > > >> > > > > >> > protocol.
> >> > > > > > > > > > > >> > > > > >> > > I
> >> > > > > > > > > > > >> > > > > >> > > > > do not
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on
> >> top of my
> >> > > > head
> >> > > > > > > > > though,
> >> > > > > > > > > > > >> maybe
> >> > > > > > > > > > > >> > others
> >> > > > > > > > > > > >> > > > > >> like
> >> > > > > > > > > > > >> > > > > >> > > @Matt
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the
> >> last
> >> > > > > > question,
> >> > > > > > > > > will the
> >> > > > > > > > > > > >> > Java
> >> > > > > > > > > > > >> > > > > >> client
> >> > > > > > > > > > > >> > > > > >> > > allow
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> >> current
> >> > > > regex
> >> > > > > > > > > engine for
> >> > > > > > > > > > > >> > > > > >> compatibility
> >> > > > > > > > > > > >> > > > > >> > > > > reasons?
> >> > > > > > > > > > > >> > > > > >> > > > > > > For
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy
> >> to keep
> >> > > > using
> >> > > > > > > > > client
> >> > > > > > > > > > > >> based
> >> > > > > > > > > > > >> > regex
> >> > > > > > > > > > > >> > > > > at
> >> > > > > > > > > > > >> > > > > >> > > first to
> >> > > > > > > > > > > >> > > > > >> > > > > keep
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple
> >> and
> >> > > > then
> >> > > > > > migrate
> >> > > > > > > > > to
> >> > > > > > > > > > > >> > server based
> >> > > > > > > > > > > >> > > > > >> > > regexes
> >> > > > > > > > > > > >> > > > > >> > > > > as a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not
> >> thought
> >> > > > > > about
> >> > > > > > > > > that for
> >> > > > > > > > > > > >> > java
> >> > > > > > > > > > > >> > > > > >> clients,
> >> > > > > > > > > > > >> > > > > >> > > and
> >> > > > > > > > > > > >> > > > > >> > > > > we can
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> >> > > > > > compatibility
> >> > > > > > > > > issues do
> >> > > > > > > > > > > >> > you have
> >> > > > > > > > > > > >> > > > > in
> >> > > > > > > > > > > >> > > > > >> > > mind?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
> >> that the
> >> > > > group
> >> > > > > > > > > coordinator
> >> > > > > > > > > > > >> > will be
> >> > > > > > > > > > > >> > > > > >> > > > > responsible for
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> >> configurations and
> >> > > > that
> >> > > > > > the
> >> > > > > > > > > > > >> > configurations
> >> > > > > > > > > > > >> > > > > will
> >> > > > > > > > > > > >> > > > > >> be
> >> > > > > > > > > > > >> > > > > >> > > > > deleted
> >> > > > > > > > > > > >> > > > > >> > > > > > > when
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted.
> >> Will a
> >> > > > > > > > > transition to
> >> > > > > > > > > > > >> DEAD
> >> > > > > > > > > > > >> > > > > trigger
> >> > > > > > > > > > > >> > > > > >> > > deletion
> >> > > > > > > > > > > >> > > > > >> > > > > of
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD
> >> state
> >> > > > is an
> >> > > > > > > > > ending
> >> > > > > > > > > > > >> state
> >> > > > > > > > > > > >> > (we
> >> > > > > > > > > > > >> > > > > would
> >> > > > > > > > > > > >> > > > > >> > only
> >> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> >> group is
> >> > > > EMPTY
> >> > > > > > and
> >> > > > > > > > > also all
> >> > > > > > > > > > > >> of
> >> > > > > > > > > > > >> > its
> >> > > > > > > > > > > >> > > > > >> > metadata
> >> > > > > > > > > > > >> > > > > >> > > are
> >> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
> >> DEAD this
> >> > > > > > group
> >> > > > > > > > > would
> >> > > > > > > > > > > >> never
> >> > > > > > > > > > > >> > be
> >> > > > > > > > > > > >> > > > > >> revived.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> >> choice to
> >> > > > store
> >> > > > > > the
> >> > > > > > > > > configs
> >> > > > > > > > > > > >> in
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > group
> >> > > > > > > > > > > >> > > > > >> > > > > coordinator
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> >> > > > cluster
> >> > > > > > configs
> >> > > > > > > > > and
> >> > > > > > > > > > > >> their
> >> > > > > > > > > > > >> > > > > values?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> >> question,
> >> > > > and our
> >> > > > > > > > > thoughts
> >> > > > > > > > > > > >> are
> >> > > > > > > > > > > >> > that
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > so-called
> >> > > > > > > > > > > >> > > > > >> > > > > > > "group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> >> > > > overrides
> >> > > > > > of the
> >> > > > > > > > > > > >> > cluster-level
> >> > > > > > > > > > > >> > > > > >> > > > > configurations
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> >> group so
> >> > > > when an
> >> > > > > > > > > admin list
> >> > > > > > > > > > > >> > cluster
> >> > > > > > > > > > > >> > > > > >> > configs
> >> > > > > > > > > > > >> > > > > >> > > it's
> >> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> >> cluster-level
> >> > > > > > > > > "defaults", not
> >> > > > > > > > > > > >> > showing
> >> > > > > > > > > > > >> > > > > >> any
> >> > > > > > > > > > > >> > > > > >> > > > > per-group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> >> someone
> >> > > > > > configure a
> >> > > > > > > > > group
> >> > > > > > > > > > > >> > before
> >> > > > > > > > > > > >> > > > > >> starting
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> >> allowing
> >> > > > the
> >> > > > > > explicit
> >> > > > > > > > > > > >> > creation of
> >> > > > > > > > > > > >> > > > > >> groups?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> >> > > > decoupled
> >> > > > > > from
> >> > > > > > > > > the group
> >> > > > > > > > > > > >> > > > > lifecycle.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
> >> created
> >> > > > > > before
> >> > > > > > > > > the group
> >> > > > > > > > > > > >> > itself
> >> > > > > > > > > > > >> > > > > as
> >> > > > > > > > > > > >> > > > > >> an
> >> > > > > > > > > > > >> > > > > >> > > > > > > independent
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> >> > > > requires
> >> > > > > > the
> >> > > > > > > > > > > >> corresponding
> >> > > > > > > > > > > >> > > > > >> request to
> >> > > > > > > > > > > >> > > > > >> > > be
> >> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> >> based on
> >> > > > the
> >> > > > > > group
> >> > > > > > > > > id ---
> >> > > > > > > > > > > >> > the only
> >> > > > > > > > > > > >> > > > > >> > thing
> >> > > > > > > > > > > >> > > > > >> > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> >> itself is
> >> > > > gone
> >> > > > > > we
> >> > > > > > > > > also check
> >> > > > > > > > > > > >> > if there
> >> > > > > > > > > > > >> > > > > >> are
> >> > > > > > > > > > > >> > > > > >> > > any
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to
> >> that
> >> > > > group
> >> > > > > > and
> >> > > > > > > > > delete
> >> > > > > > > > > > > >> as
> >> > > > > > > > > > > >> > well.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> >> indeed
> >> > > > > > introduces an
> >> > > > > > > > > > > >> > asymmetry on
> >> > > > > > > > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > creation
> >> > > > > > > > > > > >> > > > > >> > > > > /
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> >> config
> >> > > > > > entities,
> >> > > > > > > > > and we
> >> > > > > > > > > > > >> > would like
> >> > > > > > > > > > > >> > > > > >> to
> >> > > > > > > > > > > >> > > > > >> > > hear
> >> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we
> >> should
> >> > > > aim
> >> > > > > > for
> >> > > > > > > > > symmetry
> >> > > > > > > > > > > >> > i.e.
> >> > > > > > > > > > > >> > > > > >> totally
> >> > > > > > > > > > > >> > > > > >> > > > > decouple
> >> > > > > > > > > > > >> > > > > >> > > > > > > group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence
> >> not
> >> > > > delete
> >> > > > > > them at
> >> > > > > > > > > all
> >> > > > > > > > > > > >> when
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> group
> >> > > > > > > > > > > >> > > > > >> > is
> >> > > > > > > > > > > >> > > > > >> > > > > gone,
> >> > > > > > > > > > > >> > > > > >> > > > > > > but
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> >> deletion
> >> > > > > > operations
> >> > > > > > > > > by
> >> > > > > > > > > > > >> > themselves.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> >> > > > > > Consumer.subscribe
> >> > > > > > > > > method for
> >> > > > > > > > > > > >> > the Java
> >> > > > > > > > > > > >> > > > > >> > client
> >> > > > > > > > > > > >> > > > > >> > > > > still
> >> > > > > > > > > > > >> > > > > >> > > > > > > take
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > `java.util.regex.Pattern` of
> >> > > > > > do we
> >> > > > > > > > > have
> >> > > > > > > > > > > >> to
> >> > > > > > > > > > > >> > > > > >> introduce an
> >> > > > > > > > > > > >> > > > > >> > > > > overload?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
> >> need to
> >> > > > > > introduce
> >> > > > > > > > > an
> >> > > > > > > > > > > >> > overload, but
> >> > > > > > > > > > > >> > > > > >> I'm
> >> > > > > > > > > > > >> > > > > >> > > all
> >> > > > > > > > > > > >> > > > > >> > > > > ears
> >> > > > > > > > > > > >> > > > > >> > > > > > > if
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> >> compatibility
> >> > > > > > issues
> >> > > > > > > > > that we
> >> > > > > > > > > > > >> may
> >> > > > > > > > > > > >> > > > > overlook.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
> >> Justine
> >> > > > > > that we
> >> > > > > > > > > should
> >> > > > > > > > > > > >> be
> >> > > > > > > > > > > >> > clearer
> >> > > > > > > > > > > >> > > > > >> > about
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > reason
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> >> > > > > > IBP/metadata.version
> >> > > > > > > > > from the
> >> > > > > > > > > > > >> > feature
> >> > > > > > > > > > > >> > > > > >> flag.
> >> > > > > > > > > > > >> > > > > >> > > Maybe
> >> > > > > > > > > > > >> > > > > >> > > > > we
> >> > > > > > > > > > > >> > > > > >> > > > > > > mean
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> >> default
> >> > > > for
> >> > > > > > the
> >> > > > > > > > > feature
> >> > > > > > > > > > > >> flag
> >> > > > > > > > > > > >> > to
> >> > > > > > > > > > > >> > > > > true
> >> > > > > > > > > > > >> > > > > >> > > based on
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
> >> once we
> >> > > > > > want to
> >> > > > > > > > > make it
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > >> default.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> >> protocol
> >> > > > APIs
> >> > > > > > don't
> >> > > > > > > > > mention
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > >> required
> >> > > > > > > > > > > >> > > > > >> > > ACLs,
> >> > > > > > > > > > > >> > > > > >> > > > > it
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that
> >> for
> >> > > > > > consistency.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> >> certainly do
> >> > > > that.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
> >> odd that
> >> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> >> > > > > > > > > > > >> > > > > >> > requires
> >> > > > > > > > > > > >> > > > > >> > > > > "Read
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems
> >> to do
> >> > > > more
> >> > > > > > than
> >> > > > > > > > > reading.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
> >> myself
> >> > > > as
> >> > > > > > well,
> >> > > > > > > > > but in
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > end we
> >> > > > > > > > > > > >> > > > > >> > could
> >> > > > > > > > > > > >> > > > > >> > > not
> >> > > > > > > > > > > >> > > > > >> > > > > > > find a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative:
> >> adding
> >> > > > > > Write
> >> > > > > > > > > Group
> >> > > > > > > > > > > >> seems
> >> > > > > > > > > > > >> > an
> >> > > > > > > > > > > >> > > > > >> overkill
> >> > > > > > > > > > > >> > > > > >> > > here
> >> > > > > > > > > > > >> > > > > >> > > > > > > since we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> >> elsewhere (we
> >> > > > only
> >> > > > > > have
> >> > > > > > > > > Read /
> >> > > > > > > > > > > >> > Delete
> >> > > > > > > > > > > >> > > > > and
> >> > > > > > > > > > > >> > > > > >> > > Describe
> >> > > > > > > > > > > >> > > > > >> > > > > on
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to
> >> hear
> >> > > > others
> >> > > > > > > > > thoughts.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> >> > > > recreation
> >> > > > > > > > > handled by
> >> > > > > > > > > > > >> the
> >> > > > > > > > > > > >> > > > > consumer
> >> > > > > > > > > > > >> > > > > >> > with
> >> > > > > > > > > > > >> > > > > >> > > the
> >> > > > > > > > > > > >> > > > > >> > > > > new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> >> would be
> >> > > > good
> >> > > > > > to have
> >> > > > > > > > > a
> >> > > > > > > > > > > >> > section on
> >> > > > > > > > > > > >> > > > > >> this.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> >> > > > > > subscription
> >> > > > > > > > > right? Yes
> >> > > > > > > > > > > >> > we can
> >> > > > > > > > > > > >> > > > > >> add a
> >> > > > > > > > > > > >> > > > > >> > > > > section
> >> > > > > > > > > > > >> > > > > >> > > > > > > about
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically
> >> the
> >> > > > idea
> >> > > > > > is that
> >> > > > > > > > > > > >> consumer
> >> > > > > > > > > > > >> > would
> >> > > > > > > > > > > >> > > > > be
> >> > > > > > > > > > > >> > > > > >> > > totally
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as
> >> it's
> >> > > > > > handled all
> >> > > > > > > > > by the
> >> > > > > > > > > > > >> > brokers.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> >> mentions we
> >> > > > will
> >> > > > > > write
> >> > > > > > > > > the
> >> > > > > > > > > > > >> new
> >> > > > > > > > > > > >> > > > > >> coordinator
> >> > > > > > > > > > > >> > > > > >> > in
> >> > > > > > > > > > > >> > > > > >> > > > > Java.
> >> > > > > > > > > > > >> > > > > >> > > > > > > Even
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> >> implementation
> >> > > > > > detail,
> >> > > > > > > > > do we
> >> > > > > > > > > > > >> > plan to
> >> > > > > > > > > > > >> > > > > >> have a
> >> > > > > > > > > > > >> > > > > >> > > new
> >> > > > > > > > > > > >> > > > > >> > > > > gradle
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought
> >> about
> >> > > > > > that. But
> >> > > > > > > > > I think
> >> > > > > > > > > > > >> > the
> >> > > > > > > > > > > >> > > > > answer
> >> > > > > > > > > > > >> > > > > >> > > should
> >> > > > > > > > > > > >> > > > > >> > > > > be
> >> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> >> > > > scalability
> >> > > > > > goal
> >> > > > > > > > > when it
> >> > > > > > > > > > > >> > comes to
> >> > > > > > > > > > > >> > > > > >> how
> >> > > > > > > > > > > >> > > > > >> > > many
> >> > > > > > > > > > > >> > > > > >> > > > > > > members
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> >> > > > support?
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> >> think we
> >> > > > > > should
> >> > > > > > > > > shoot for
> >> > > > > > > > > > > >> > 1000s of
> >> > > > > > > > > > > >> > > > > >> > > members.
> >> > > > > > > > > > > >> > > > > >> > > > > But
> >> > > > > > > > > > > >> > > > > >> > > > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
> >> also
> >> > > > depend
> >> > > > > > on the
> >> > > > > > > > > offset
> >> > > > > > > > > > > >> > > > > management
> >> > > > > > > > > > > >> > > > > >> > > (commit,
> >> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> >> > > > coordinator
> >> > > > > > > > > which we
> >> > > > > > > > > > > >> did
> >> > > > > > > > > > > >> > not
> >> > > > > > > > > > > >> > > > > >> cover in
> >> > > > > > > > > > > >> > > > > >> > > this
> >> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> >> number that
> >> > > > > > applies
> >> > > > > > > > > > > >> > universally.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> >> consider
> >> > > > having
> >> > > > > > > > > > > >> > SubscribedTopidIds
> >> > > > > > > > > > > >> > > > > >> instead
> >> > > > > > > > > > > >> > > > > >> > of
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> SubscribedTopicNames in
> >> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> >> > > > > > > > > > > >> > > > > >> > > Is the
> >> > > > > > > > > > > >> > > > > >> > > > > > > idea
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
> >> resolve
> >> > > > the
> >> > > > > > regex
> >> > > > > > > > > on the
> >> > > > > > > > > > > >> > server,
> >> > > > > > > > > > > >> > > > > we
> >> > > > > > > > > > > >> > > > > >> > can
> >> > > > > > > > > > > >> > > > > >> > > do
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > same
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name?
> >> The
> >> > > > > > difference is
> >> > > > > > > > > that
> >> > > > > > > > > > > >> > sending the
> >> > > > > > > > > > > >> > > > > >> > regex
> >> > > > > > > > > > > >> > > > > >> > > is
> >> > > > > > > > > > > >> > > > > >> > > > > more
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
> >> the topic
> >> > > > > > names is
> >> > > > > > > > > less
> >> > > > > > > > > > > >> > efficient.
> >> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> >> easier to
> >> > > > > > handle if
> >> > > > > > > > > we have
> >> > > > > > > > > > > >> > topic
> >> > > > > > > > > > > >> > > > > ids.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
> >> still
> >> > > > let
> >> > > > > > the
> >> > > > > > > > > clients
> >> > > > > > > > > > > >> send
> >> > > > > > > > > > > >> > names
> >> > > > > > > > > > > >> > > > > >> is to
> >> > > > > > > > > > > >> > > > > >> > > keep
> >> > > > > > > > > > > >> > > > > >> > > > > the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names
> >> -> ids
> >> > > > on
> >> > > > > > the
> >> > > > > > > > > broker /
> >> > > > > > > > > > > >> > admin
> >> > > > > > > > > > > >> > > > > client
> >> > > > > > > > > > > >> > > > > >> > > only.
> >> > > > > > > > > > > >> > > > > >> > > > > Note
> >> > > > > > > > > > > >> > > > > >> > > > > > > that
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> >> topic id
> >> > > > in
> >> > > > > > > > > KIP-516, we
> >> > > > > > > > > > > >> > never
> >> > > > > > > > > > > >> > > > > >> > > implemented the
> >> > > > > > > > > > > >> > > > > >> > > > > > > logic
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> >> > > > leveraging
> >> > > > > > the
> >> > > > > > > > > related
> >> > > > > > > > > > > >> > newer
> >> > > > > > > > > > > >> > > > > >> versioned
> >> > > > > > > > > > > >> > > > > >> > > RPCs,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic
> >> id as
> >> > > > empty
> >> > > > > > UUID.
> >> > > > > > > > > We
> >> > > > > > > > > > > >> want
> >> > > > > > > > > > > >> > to keep
> >> > > > > > > > > > > >> > > > > >> the
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> >> > > > delegate
> >> > > > > > the
> >> > > > > > > > > reasoning
> >> > > > > > > > > > > >> on
> >> > > > > > > > > > > >> > broker
> >> > > > > > > > > > > >> > > > > >> and
> >> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
> >> 2022 at
> >> > > > 10:45
> >> > > > > > AM
> >> > > > > > > > > David
> >> > > > > > > > > > > >> Jacot
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> <dj...@confluent.io.invalid>
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to
> >> start a
> >> > > > > > > > > discussion
> >> > > > > > > > > > > >> thread
> >> > > > > > > > > > > >> > on
> >> > > > > > > > > > > >> > > > > >> KIP-848:
> >> > > > > > > > > > > >> > > > > >> > > The
> >> > > > > > > > > > > >> > > > > >> > > > > Next
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of
> >> the
> >> > > > Consumer
> >> > > > > > > > > Rebalance
> >> > > > > > > > > > > >> > Protocol.
> >> > > > > > > > > > > >> > > > > >> With
> >> > > > > > > > > > > >> > > > > >> > > this
> >> > > > > > > > > > > >> > > > > >> > > > > KIP,
> >> > > > > > > > > > > >> > > > > >> > > > > > > we
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> >> rebalance
> >> > > > > > protocol
> >> > > > > > > > > (for
> >> > > > > > > > > > > >> > consumers)
> >> > > > > > > > > > > >> > > > > >> more
> >> > > > > > > > > > > >> > > > > >> > > > > reliable,
> >> > > > > > > > > > > >> > > > > >> > > > > > > more
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> >> easier to
> >> > > > > > implement for
> >> > > > > > > > > > > >> > clients, and
> >> > > > > > > > > > > >> > > > > >> easier
> >> > > > > > > > > > > >> > > > > >> > > to
> >> > > > > > > > > > > >> > > > > >> > > > > debug
> >> > > > > > > > > > > >> > > > > >> > > > > > > for
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
> >> look and
> >> > > > let
> >> > > > > > me
> >> > > > > > > > > know what
> >> > > > > > > > > > > >> > you
> >> > > > > > > > > > > >> > > > > think.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
> >> away from
> >> > > > > > July
> >> > > > > > > > > 18th to
> >> > > > > > > > > > > >> > August 8th.
> >> > > > > > > > > > > >> > > > > >> That
> >> > > > > > > > > > > >> > > > > >> > > gives
> >> > > > > > > > > > > >> > > > > >> > > > > > > you a
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read
> >> and
> >> > > > digest
> >> > > > > > this
> >> > > > > > > > > long
> >> > > > > > > > > > > >> KIP.
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> >> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> >> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > > > > > --
> >> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> >> > > > > > > > > > > >> > > > > >> > > > > > >
> >> > > > > > > > > > > >> > > > > >> > > > >
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> > >
> >> > > > > > > > > > > >> > > > > >> >
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > > >> --
> >> > > > > > > > > > > >> > > > > >> -- Guozhang
> >> > > > > > > > > > > >> > > > > >>
> >> > > > > > > > > > > >> > > > > >
> >> > > > > > > > > > > >> > > > >
> >> > > > > > > > > > > >> >
> >> > > > > > > > > > > >>
> >> > > > > > > > > > > >
> >> > > > > > > > >
> >> > > > > >
> >> > > > >
> >> > > > >
> >> > > > > --
> >> > > > > -- Guozhang
> >> > > >
> >> > >
> >> > >
> >> > > --
> >> > > -- Guozhang
> >>
> >
> >
> > --
> > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

 and  and ` Hi Guozhang (again ;) ),

> Another quick thing I just realized about "PartitionAssignor#onAssignment":
> in the new proposal, this function could be triggered multiple time within
> a group generation as it will be triggered whenever the broker gives the
> client new assignments, whereas today it is only triggered once per
> rebalance. I think it's worth emphasizing this a bit in the doc, especially
> around:
>
> 1) how many times `onAssignment` could be triggered with a single rebalance
> (especially if the new assignment is decided by the client-assignor).
> 2) when the Assignment metadata could change across multiple `onAssignment`
> calls, and how users should handle it.
> 3) the order of calling `onAssignment` and `onPartitionsRevoked/Assigned`.

Yeah, that's right. Within a rebalance, `onAssignment` is called once
when the member transitions to a new epoch. This one contains the full
metadata provided by the client side assignor. Then, `onAssignment`
can be called max N times where N is the number of partitions pending
revocation by other members. Let me try to clarify this in the KIP.

> For example, today the Streams client relies on the following assumptions:
> 1) within a rebalance the onPartitionsRevoked would be called before the
> onAssignment, and onAssignment is called before onPartitionsAssigned (today
> we would trigger all three only once within a generation), 2) onAssignment
> contains the source-of-truth of the final assignment, and hence we can
> simply ignore onPartitionsAssigned and just rely on onAssignment to reason
> about which tasks are removed / assigned. And each time it calls, the
> corresponding metadata change are "final" for this new generation.

That makes sense. My goal was to keep the same sequence for those. The
only difference is that they can be called multiple times. I will
clarify this in the doc.

> We can change however Streams handle the new protocol, but I think there
> are other clients externally which may also make such assumptions about the
> ordering / call times about those callbacks, and hence worth clarifying
> those in the new proposal.

That makes sense. Thanks!

Best,
David

On Tue, Sep 13, 2022 at 10:58 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hi David,
>
> Another quick thing I just realized about "PartitionAssignor#onAssignment":
> in the new proposal, this function could be triggered multiple time within
> a group generation as it will be triggered whenever the broker gives the
> client new assignments, whereas today it is only triggered once per
> rebalance. I think it's worth emphasizing this a bit in the doc, especially
> around:
>
> 1) how many times `onAssignment` could be triggered with a single rebalance
> (especially if the new assignment is decided by the client-assignor).
> 2) when the Assignment metadata could change across multiple `onAssignment`
> calls, and how users should handle it.
> 3) the order of calling `onAssignment` and `onPartitionsRevoked/Assigned`.
>
> For example, today the Streams client relies on the following assumptions:
> 1) within a rebalance the onPartitionsRevoked would be called before the
> onAssignment, and onAssignment is called before onPartitionsAssigned (today
> we would trigger all three only once within a generation), 2) onAssignment
> contains the source-of-truth of the final assignment, and hence we can
> simply ignore onPartitionsAssigned and just rely on onAssignment to reason
> about which tasks are removed / assigned. And each time it calls, the
> corresponding metadata change are "final" for this new generation.
>
> We can change however Streams handle the new protocol, but I think there
> are other clients externally which may also make such assumptions about the
> ordering / call times about those callbacks, and hence worth clarifying
> those in the new proposal.
>
>
>
> Guozhang
>
>
>
>
> On Tue, Sep 13, 2022 at 12:20 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> > Hello David,
> >
> > Just had a few more nit questions about the public APIs:
> >
> > 1) The client-side "PartitionAssignor#Assignment" has an Error byte field,
> > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > rationale seems to be assuming that we should always be able to do the
> > assignment at the broker-side assignor without errors. Personally I think
> > it's still potentially beneficial to add the Error field even for
> > broker-side assignors, e.g. for some edge cases where some subscribed
> > topics are not recognized with the current broker's metadata. What do you
> > think?
> >
> > 2) The client-side "GroupMember" has three additional fields
> > reason/version/metadata compared with the broker-side GroupMember. I agree
> > that broker-side assignor would not need reason/metadata since they are
> > blackbox strings/bytes to the assignor, but what about version? E.g. is it
> > possible that we evolve our broker-side built-in assignor but the old
> > versioned clients would not be able to work with the new version, in which
> > case we need to let the broker being aware of this and upgrade its behavior
> > to cooperate with the clients?
> >
> > 3) Also related to 2) above, for the client-side "GroupMember", instead of
> > including these three fields, what about just adding the "Metadata" field
> > class which has these three fields? Also, there are two "Metadata"
> > currently in the APIs, the first is a class that encodes
> > reason/version/metadata, and the second is just the encoded metadata bytes.
> > I'm wondering what about just naming the first as memberMetadata, which
> > then has a bytebuffer field Metadata, or instead naming the second
> > bytebuffer field as metadataBytes?
> >
> > Guozhang
> >
> >
> >
> >
> >
> >
> >
> > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> >> Hello David,
> >>
> >> Thanks for bringing this question up. I think the main benefits as you
> >> listed is 2) above if it stays; just to clarify, we would only be able to
> >> save one round trip if the rebalance is still triggered by the broker; if
> >> the rebalance is triggered by the client then the num.round.trips are the
> >> same:
> >>
> >> 1) With GroupPrepareAssignment:
> >>
> >> T0: client decides to do a new assignment, suppose it has already sent a
> >> HB and hence has to wait for it to return first since only one request /
> >> response can be inflight with the coordinator's socket.
> >> T1: client receives the HB response, and then sends the
> >> GroupPrepareAssignment request.
> >> T2: the GroupPrepareAssignment response is returned.
> >> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> >> request.
> >>
> >> In total, two round trips.
> >>
> >> 2) Without GroupPrepareAssignment:
> >>
> >> T0: client decides to do a new assignment, suppose it has already sent a
> >> HB and hence has to wait for it to return first since only one request /
> >> response can be inflight with the coordinator's socket.
> >> T1: client receives the HB response, and then sends the new HB request
> >> with the flag indicating a new rebalance needed..
> >> T2: the HB response with the optional member metadata map is returned.
> >> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> >> request.
> >>
> >> In total, two round trips as well.
> >>
> >> -----------------------------
> >>
> >> So to complete the full picture here, we'd need to modify both HB request
> >> and response so that the client can also indicate a new rebalance via the
> >> HB request as well, right?
> >>
> >> Assuming all above is true, I think it's okay to merge the
> >> GroupPrepareAssignment into HB given that we can make the additional fields
> >> encoding the full member (subscription) metadata and topic metadata as
> >> optional fields.
> >>
> >>
> >> Guozhang
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >> On Mon, Sep 12, 2022 at 5:22 AM David Jacot <dj...@confluent.io.invalid>
> >> wrote:
> >>
> >>> Hi all,
> >>>
> >>> During an offline conversation, someone asked why we need the
> >>> ConsumerGroupPrepareAssignment API and suggested that we could instead
> >>> provide the group state in the heartbeat response. This has a few
> >>> advantages: 1) it does not require using a special error code to
> >>> signal that a new assignment is required as the signal would be the
> >>> provided group state; 2) it removes one round trip when a client side
> >>> assignor is used. The downside is that it makes the heartbeat
> >>> response's definition quite large. I recall that I went with the
> >>> current approach due to this.
> >>>
> >>> Providing the group state in the heartbeat response is appealing. What
> >>> do you guys think?
> >>>
> >>> Best,
> >>> David
> >>>
> >>> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io> wrote:
> >>> >
> >>> > Hi Guozhang,
> >>> >
> >>> > 1. I have added a reference to the relevant chapter instead of
> >>> > repeating the whole thing. Does that work for you?
> >>> >
> >>> > 2. The "Rebalance Triggers" section you are referring to is about when
> >>> > a rebalance should be triggered for the non-upgraded members using the
> >>> > old protocol. The section mentions that a rebalance must be triggered
> >>> > when a new assignment is installed. This implies that the group epoch
> >>> > was updated either by a native member or a non-upgraded member. For
> >>> > the latter, the JoinGroup request would be the trigger. I have added a
> >>> > reference to the relevant chapter in the "JoinGroup Handling" section
> >>> > as well. Does that make sense?
> >>> >
> >>> > Thanks,
> >>> > David
> >>> >
> >>> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> >>> wrote:
> >>> > >
> >>> > > Hello David,
> >>> > >
> >>> > > Alright I think that's sufficient. Just to make that clear in the
> >>> doc,
> >>> > > could we update:
> >>> > >
> >>> > > 1) the heartbeat request handling section, stating when coordinator
> >>> will
> >>> > > trigger rebalance based on the HB's member metadata / reason?
> >>> > > 2) the "Rebalance Triggers" section to include what we described in
> >>> "Group
> >>> > > Epoch - Trigger a rebalance" section as well?
> >>> > >
> >>> > >
> >>> > > Guozhang
> >>> > >
> >>> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> >>> <dj...@confluent.io.invalid>
> >>> > > wrote:
> >>> > >
> >>> > > > Hi Guozhang,
> >>> > > >
> >>> > > > I thought that the assignor will always be consulted when the next
> >>> > > > heartbeat request is constructed. In other words,
> >>> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
> >>> This
> >>> > > > gives the opportunity for the assignor to enforce a rebalance by
> >>> > > > setting the reason to a non-zero value or by changing the bytes. Do
> >>> > > > you think that this is not sufficient? Are you concerned by the
> >>> delay?
> >>> > > >
> >>> > > > Best,
> >>> > > > David
> >>> > > >
> >>> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
> >>> wrote:
> >>> > > > >
> >>> > > > > Hello David,
> >>> > > > >
> >>> > > > > One of Jun's comments make me thinking:
> >>> > > > >
> >>> > > > > ```
> >>> > > > > In this case, a new assignment is triggered by the client side
> >>> > > > > assignor. When constructing the HB, the consumer will always
> >>> consult
> >>> > > > > the client side assignor and propagate the information to the
> >>> group
> >>> > > > > coordinator. In other words, we don't expect users to call
> >>> > > > > Consumer#enforceRebalance anymore.
> >>> > > > > ```
> >>> > > > >
> >>> > > > > As I looked at the current PartitionAssignor's interface, we
> >>> actually do
> >>> > > > > not have a way yet to instruct how to construct the next HB
> >>> request, e.g.
> >>> > > > > when the assignor wants to enforce a new rebalance with a new
> >>> assignment,
> >>> > > > > we'd need some customizable APIs inside the PartitionAssignor to
> >>> indicate
> >>> > > > > the next HB telling broker about so. WDYT about adding such an
> >>> API on the
> >>> > > > > PartitionAssignor?
> >>> > > > >
> >>> > > > >
> >>> > > > > Guozhang
> >>> > > > >
> >>> > > > >
> >>> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> >>> <dj...@confluent.io.invalid>
> >>> > > > > wrote:
> >>> > > > >
> >>> > > > > > Hi Jun,
> >>> > > > > >
> >>> > > > > > I have updated the KIP to include your feedback. I have also
> >>> tried to
> >>> > > > > > clarify the parts which were not cleared.
> >>> > > > > >
> >>> > > > > > Best,
> >>> > > > > > David
> >>> > > > > >
> >>> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> >>> djacot@confluent.io>
> >>> > > > wrote:
> >>> > > > > > >
> >>> > > > > > > Hi Jun,
> >>> > > > > > >
> >>> > > > > > > Thanks for your feedback. Let me start by answering your
> >>> questions
> >>> > > > > > > inline and I will update the KIP next week.
> >>> > > > > > >
> >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> >>> seem to
> >>> > > > be
> >>> > > > > > fewer
> >>> > > > > > > > RPCs during rebalance and more efficient support of
> >>> wildcard. A few
> >>> > > > > > > > comments below.
> >>> > > > > > >
> >>> > > > > > > I would also add that the KIP removes the global sync
> >>> barrier in the
> >>> > > > > > > protocol which is essential to improve group stability and
> >>> > > > > > > scalability, and the KIP also simplifies the client by
> >>> moving most of
> >>> > > > > > > the logic to the server side.
> >>> > > > > > >
> >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> >>> rolling
> >>> > > > > > changing
> >>> > > > > > > > of the partition assignor in the consumers?
> >>> > > > > > >
> >>> > > > > > > Definitely. The group coordinator will use the assignor used
> >>> by a
> >>> > > > > > > majority of the members. This allows the group to move from
> >>> one
> >>> > > > > > > assignor to another by a roll. This is explained in the
> >>> Assignor
> >>> > > > > > > Selection chapter.
> >>> > > > > > >
> >>> > > > > > > > 30.2 For each field, could you explain whether it's
> >>> required in
> >>> > > > every
> >>> > > > > > > > request or the scenarios when it needs to be filled? For
> >>> example,
> >>> > > > it's
> >>> > > > > > not
> >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> >>> > > > > > >
> >>> > > > > > > The client is expected to set those fields in case of a
> >>> connection
> >>> > > > > > > issue (e.g. timeout) or when the fields have changed since
> >>> the last
> >>> > > > > > > HB. The server populates those fields as long as the member
> >>> is not
> >>> > > > > > > fully reconciled - the member should acknowledge that it has
> >>> the
> >>> > > > > > > expected epoch and assignment. I will clarify this in the
> >>> KIP.
> >>> > > > > > >
> >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> >>> between the
> >>> > > > > > client
> >>> > > > > > > > and the broker is only considered during fetching, but not
> >>> during
> >>> > > > > > assigning
> >>> > > > > > > > partitions to consumers. Sometimes, once the assignment is
> >>> made,
> >>> > > > there
> >>> > > > > > is
> >>> > > > > > > > no opportunity for read affinity because no replicas of
> >>> assigned
> >>> > > > > > partitions
> >>> > > > > > > > are close to the member. I am wondering if we should use
> >>> this
> >>> > > > > > opportunity
> >>> > > > > > > > to address this by including rack in GroupMember.
> >>> > > > > > >
> >>> > > > > > > That's an interesting idea. I don't see any issue with
> >>> adding the
> >>> > > > rack
> >>> > > > > > > to the members. I will do so.
> >>> > > > > > >
> >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> >>> busy a group
> >>> > > > > > > > coordinator is. By moving the event loop model, it seems
> >>> that we
> >>> > > > could
> >>> > > > > > add
> >>> > > > > > > > a metric that tracks the fraction of the time the event
> >>> loop is
> >>> > > > doing
> >>> > > > > > the
> >>> > > > > > > > actual work.
> >>> > > > > > >
> >>> > > > > > > That's a great idea. I will add it. Thanks.
> >>> > > > > > >
> >>> > > > > > > > 33. Could we add a section on coordinator failover
> >>> handling? For
> >>> > > > > > example,
> >>> > > > > > > > does it need to trigger the check if any group with the
> >>> wildcard
> >>> > > > > > > > subscription now has a new matching topic?
> >>> > > > > > >
> >>> > > > > > > Sure. When the new group coordinator takes over, it has to:
> >>> > > > > > > * Setup the session timeouts.
> >>> > > > > > > * Trigger a new assignment if a client side assignor is
> >>> used. We
> >>> > > > don't
> >>> > > > > > > store the information about the member selected to run the
> >>> assignment
> >>> > > > > > > so we have to start a new one.
> >>> > > > > > > * Update the topics metadata, verify the wildcard
> >>> subscriptions, and
> >>> > > > > > > trigger a rebalance if needed.
> >>> > > > > > >
> >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> >>> > > > ConsumerGroupPartitionMetadataValue,
> >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> >>> the epoch
> >>> > > > > > field
> >>> > > > > > > > reflects? For example, does the epoch in
> >>> ConsumerGroupMetadataValue
> >>> > > > > > reflect
> >>> > > > > > > > the latest group epoch? What about the one in
> >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> >>> > > > > > ConsumerGroupMemberMetadataValue?
> >>> > > > > > >
> >>> > > > > > > Sure. I will clarify that but it is always the latest group
> >>> epoch.
> >>> > > > > > > When the group state is updated, the group epoch is bumped
> >>> so we use
> >>> > > > > > > that one for all the change records related to the update.
> >>> > > > > > >
> >>> > > > > > > > 35. "the group coordinator will ensure that the following
> >>> > > > invariants
> >>> > > > > > are
> >>> > > > > > > > met: ... All members exists." It's possible for a member
> >>> not to
> >>> > > > get any
> >>> > > > > > > > assigned partitions, right?
> >>> > > > > > >
> >>> > > > > > > That's right. Here I meant that the members provided by the
> >>> assignor
> >>> > > > > > > in the assignment must exist in the group. The assignor can
> >>> not make
> >>> > > > > > > up new member ids.
> >>> > > > > > >
> >>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
> >>> to 0":
> >>> > > > When
> >>> > > > > > would
> >>> > > > > > > > a consumer rejoin and what member id would be used?
> >>> > > > > > >
> >>> > > > > > > A member is expected to abandon all its partitions and
> >>> rejoins when
> >>> > > > it
> >>> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> >>> group
> >>> > > > > > > coordinator will have removed the member from the group. The
> >>> member
> >>> > > > > > > can rejoin the group with the same member id but with 0 as
> >>> epoch. Let
> >>> > > > > > > me see if I can clarify this in the KIP.
> >>> > > > > > >
> >>> > > > > > > > 37. "Instead, power users will have the ability to trigger
> >>> a
> >>> > > > > > reassignment
> >>> > > > > > > > by either providing a non-zero reason or by updating the
> >>> assignor
> >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> >>> deprecation
> >>> > > > of
> >>> > > > > > > > Consumer#enforeRebalance.
> >>> > > > > > >
> >>> > > > > > > In this case, a new assignment is triggered by the client
> >>> side
> >>> > > > > > > assignor. When constructing the HB, the consumer will always
> >>> consult
> >>> > > > > > > the client side assignor and propagate the information to
> >>> the group
> >>> > > > > > > coordinator. In other words, we don't expect users to call
> >>> > > > > > > Consumer#enforceRebalance anymore.
> >>> > > > > > >
> >>> > > > > > > > 38. The reassignment examples are nice. But the section
> >>> seems to
> >>> > > > have
> >>> > > > > > > > multiple typos.
> >>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> >>> gets into
> >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> >>> into
> >>> > > > > > epoch=3,
> >>> > > > > > > > partitions=[foo-1] too early.
> >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> >>> epoch=2,
> >>> > > > > > > > partitions=[foo-0].
> >>> > > > > > >
> >>> > > > > > > Sorry for that! I will revise them.
> >>> > > > > > >
> >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> >>> upgrade from
> >>> > > > any
> >>> > > > > > old
> >>> > > > > > > > version to new one?
> >>> > > > > > >
> >>> > > > > > > We will support upgrading from the consumer protocol version
> >>> 3,
> >>> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
> >>> earliest
> >>> > > > > > > version is unknown at the moment. This is explained in the
> >>> migration
> >>> > > > > > > plan chapter.
> >>> > > > > > >
> >>> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> >>> based on
> >>> > > > it
> >>> > > > > > > next week.
> >>> > > > > > >
> >>> > > > > > > Best,
> >>> > > > > > > David
> >>> > > > > > >
> >>> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> >>> <ju...@confluent.io.invalid>
> >>> > > > wrote:
> >>> > > > > > > >
> >>> > > > > > > > Hi, David,
> >>> > > > > > > >
> >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> >>> seem to
> >>> > > > be
> >>> > > > > > fewer
> >>> > > > > > > > RPCs during rebalance and more efficient support of
> >>> wildcard. A few
> >>> > > > > > > > comments below.
> >>> > > > > > > >
> >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> >>> rolling
> >>> > > > > > changing
> >>> > > > > > > > of the partition assignor in the consumers?
> >>> > > > > > > > 30.2 For each field, could you explain whether it's
> >>> required in
> >>> > > > every
> >>> > > > > > > > request or the scenarios when it needs to be filled? For
> >>> example,
> >>> > > > it's
> >>> > > > > > not
> >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> >>> > > > > > > >
> >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> >>> between the
> >>> > > > > > client
> >>> > > > > > > > and the broker is only considered during fetching, but not
> >>> during
> >>> > > > > > assigning
> >>> > > > > > > > partitions to consumers. Sometimes, once the assignment is
> >>> made,
> >>> > > > there
> >>> > > > > > is
> >>> > > > > > > > no opportunity for read affinity because no replicas of
> >>> assigned
> >>> > > > > > partitions
> >>> > > > > > > > are close to the member. I am wondering if we should use
> >>> this
> >>> > > > > > opportunity
> >>> > > > > > > > to address this by including rack in GroupMember.
> >>> > > > > > > >
> >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> >>> busy a group
> >>> > > > > > > > coordinator is. By moving the event loop model, it seems
> >>> that we
> >>> > > > could
> >>> > > > > > add
> >>> > > > > > > > a metric that tracks the fraction of the time the event
> >>> loop is
> >>> > > > doing
> >>> > > > > > the
> >>> > > > > > > > actual work.
> >>> > > > > > > >
> >>> > > > > > > > 33. Could we add a section on coordinator failover
> >>> handling? For
> >>> > > > > > example,
> >>> > > > > > > > does it need to trigger the check if any group with the
> >>> wildcard
> >>> > > > > > > > subscription now has a new matching topic?
> >>> > > > > > > >
> >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> >>> > > > ConsumerGroupPartitionMetadataValue,
> >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> >>> the epoch
> >>> > > > > > field
> >>> > > > > > > > reflects? For example, does the epoch in
> >>> ConsumerGroupMetadataValue
> >>> > > > > > reflect
> >>> > > > > > > > the latest group epoch? What about the one in
> >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> >>> > > > > > ConsumerGroupMemberMetadataValue?
> >>> > > > > > > >
> >>> > > > > > > > 35. "the group coordinator will ensure that the following
> >>> > > > invariants
> >>> > > > > > are
> >>> > > > > > > > met: ... All members exists." It's possible for a member
> >>> not to
> >>> > > > get any
> >>> > > > > > > > assigned partitions, right?
> >>> > > > > > > >
> >>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
> >>> to 0":
> >>> > > > When
> >>> > > > > > would
> >>> > > > > > > > a consumer rejoin and what member id would be used?
> >>> > > > > > > >
> >>> > > > > > > > 37. "Instead, power users will have the ability to trigger
> >>> a
> >>> > > > > > reassignment
> >>> > > > > > > > by either providing a non-zero reason or by updating the
> >>> assignor
> >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> >>> deprecation
> >>> > > > of
> >>> > > > > > > > Consumer#enforeRebalance.
> >>> > > > > > > >
> >>> > > > > > > > 38. The reassignment examples are nice. But the section
> >>> seems to
> >>> > > > have
> >>> > > > > > > > multiple typos.
> >>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> >>> gets into
> >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> >>> into
> >>> > > > > > epoch=3,
> >>> > > > > > > > partitions=[foo-1] too early.
> >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> >>> epoch=2,
> >>> > > > > > > > partitions=[foo-0].
> >>> > > > > > > >
> >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> >>> upgrade from
> >>> > > > any
> >>> > > > > > old
> >>> > > > > > > > version to new one?
> >>> > > > > > > >
> >>> > > > > > > > Thanks,
> >>> > > > > > > >
> >>> > > > > > > > Jun
> >>> > > > > > > >
> >>> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> >>> > > > > > <dj...@confluent.io.invalid>
> >>> > > > > > > > wrote:
> >>> > > > > > > >
> >>> > > > > > > > > Hi all,
> >>> > > > > > > > >
> >>> > > > > > > > > The KIP states that we will re-implement the coordinator
> >>> in
> >>> > > > Java. I
> >>> > > > > > > > > discussed this offline with a few folks and folks are
> >>> concerned
> >>> > > > that
> >>> > > > > > > > > we could introduce many regressions in the old protocol
> >>> if we do
> >>> > > > so.
> >>> > > > > > > > > Therefore, I am going to remove this statement from the
> >>> KIP. It
> >>> > > > is an
> >>> > > > > > > > > implementation detail after all so it does not have to be
> >>> > > > decided at
> >>> > > > > > > > > this stage. We will likely start by trying to refactor
> >>> the
> >>> > > > current
> >>> > > > > > > > > implementation as a first step.
> >>> > > > > > > > >
> >>> > > > > > > > > Cheers,
> >>> > > > > > > > > David
> >>> > > > > > > > >
> >>> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> >>> djacot@confluent.io
> >>> > > > >
> >>> > > > > > wrote:
> >>> > > > > > > > > >
> >>> > > > > > > > > > Hi Luke,
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> >>> assigning,
> >>> > > > > > reconciling,
> >>> > > > > > > > > stable,
> >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> >>> right?
> >>> > > > > > > > > >
> >>> > > > > > > > > > This sentence does not refer to those group states but
> >>> rather
> >>> > > > to a
> >>> > > > > > > > > > state machine replication (SMR). This refers to the
> >>> entire
> >>> > > > state of
> >>> > > > > > > > > > group coordinator which is replicated via the log
> >>> layer. I will
> >>> > > > > > > > > > clarify this in the KIP.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> >>> modelled as an
> >>> > > > event
> >>> > > > > > > > > loop"?
> >>> > > > > > > > > >
> >>> > > > > > > > > > The idea is to follow a model similar to the new quorum
> >>> > > > > > controller. We
> >>> > > > > > > > > > will have N threads to process events. Each
> >>> __consumer_offsets
> >>> > > > > > > > > > partition is assigned to a unique thread and all the
> >>> events
> >>> > > > (e.g.
> >>> > > > > > > > > > requests, callbacks, etc.) are processed by this
> >>> thread. This
> >>> > > > > > simplify
> >>> > > > > > > > > > concurrency and will enable us to do simulation
> >>> testing for the
> >>> > > > > > group
> >>> > > > > > > > > > coordinator.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> >>> *__consumer_offsets*
> >>> > > > > > > > > partitions?
> >>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> >>> group
> >>> > > > > > coordinator?
> >>> > > > > > > > > For
> >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> >>> groups, and
> >>> > > > > > both
> >>> > > > > > > > > exist in
> >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> >>> for it,
> >>> > > > or
> >>> > > > > > 2?
> >>> > > > > > > > > >
> >>> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> >>> number
> >>> > > > of
> >>> > > > > > threads
> >>> > > > > > > > > used
> >>> > > > > > > > > > > to run the state machines. But I'm wondering if the
> >>> purpose
> >>> > > > of
> >>> > > > > > the
> >>> > > > > > > > > threads
> >>> > > > > > > > > > > is only to keep the state of each consumer group (or
> >>> > > > > > > > > *__consumer_offsets*
> >>> > > > > > > > > > > partitions?), and no heavy computation, why should
> >>> we need
> >>> > > > > > > > > multi-threads
> >>> > > > > > > > > > > here?
> >>> > > > > > > > > >
> >>> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> >>> > > > processing as
> >>> > > > > > the
> >>> > > > > > > > > > computation could be heavy.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> >>> default
> >>> > > > session
> >>> > > > > > > > > timeout not
> >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> >>> min/max
> >>> > > > > > session
> >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> >>> timeout to
> >>> > > > detect
> >>> > > > > > client
> >>> > > > > > > > > > > failures when using the consumer group protocol.
> >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> >>> minimum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> >>> maximum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > >
> >>> > > > > > > > > > This is indeed a mistake. The default session timeout
> >>> should
> >>> > > > be 45s
> >>> > > > > > > > > > (the current default).
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> >>> uniform],
> >>> > > > > > which means
> >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> >>> why not
> >>> > > > > > uniform
> >>> > > > > > > > > one? I
> >>> > > > > > > > > > > thought usually users will choose uniform assignor
> >>> (former
> >>> > > > sticky
> >>> > > > > > > > > assinor)
> >>> > > > > > > > > > > for better evenly distribution. Any other reason we
> >>> choose
> >>> > > > range
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > as default?
> >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> >>> server side
> >>> > > > > > assignors.
> >>> > > > > > > > > >
> >>> > > > > > > > > > The order on the server side has no influence because
> >>> the
> >>> > > > client
> >>> > > > > > must
> >>> > > > > > > > > > chose the selector that he wants to use. There is no
> >>> default
> >>> > > > in the
> >>> > > > > > > > > > current proposal. If the assignor is not specified by
> >>> the
> >>> > > > client,
> >>> > > > > > the
> >>> > > > > > > > > > request is rejected. The default client value for
> >>> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> >>> > > > > > > > > >
> >>> > > > > > > > > > Thanks for your very good comments, Luke. I hope that
> >>> my
> >>> > > > answers
> >>> > > > > > help
> >>> > > > > > > > > > to clarify things. I will update the KIP as well based
> >>> on your
> >>> > > > > > > > > > feedback.
> >>> > > > > > > > > >
> >>> > > > > > > > > > Cheers,
> >>> > > > > > > > > > David
> >>> > > > > > > > > >
> >>> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> >>> showuon@gmail.com>
> >>> > > > > > wrote:
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Hi David,
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Thanks for the update.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Some more questions:
> >>> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> >>> > > > > > > > > > > > The new group coordinator will have a state
> >>> machine per
> >>> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> >>> machine is
> >>> > > > > > modelled
> >>> > > > > > > > > as an
> >>> > > > > > > > > > > event loop. Those state machines will be executed in
> >>> > > > > > > > > > > *group.coordinator.threads* threads.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> >>> assigning,
> >>> > > > > > reconciling,
> >>> > > > > > > > > stable,
> >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> >>> right?
> >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> >>> modelled as an
> >>> > > > event
> >>> > > > > > > > > loop"?
> >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> >>> *__consumer_offsets*
> >>> > > > > > > > > partitions?
> >>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> >>> group
> >>> > > > > > coordinator?
> >>> > > > > > > > > For
> >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> >>> groups, and
> >>> > > > > > both
> >>> > > > > > > > > exist in
> >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> >>> for it,
> >>> > > > or
> >>> > > > > > 2?
> >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> >>> number
> >>> > > > of
> >>> > > > > > threads
> >>> > > > > > > > > used
> >>> > > > > > > > > > > to run the state machines. But I'm wondering if the
> >>> purpose
> >>> > > > of
> >>> > > > > > the
> >>> > > > > > > > > threads
> >>> > > > > > > > > > > is only to keep the state of each consumer group (or
> >>> > > > > > > > > *__consumer_offsets*
> >>> > > > > > > > > > > partitions?), and no heavy computation, why should
> >>> we need
> >>> > > > > > > > > multi-threads
> >>> > > > > > > > > > > here?
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > 2. For the default value in the new configs:
> >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> >>> default
> >>> > > > session
> >>> > > > > > > > > timeout not
> >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> >>> min/max
> >>> > > > > > session
> >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> >>> timeout to
> >>> > > > detect
> >>> > > > > > client
> >>> > > > > > > > > > > failures when using the consumer group protocol.
> >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> >>> minimum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> >>> maximum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> >>> uniform],
> >>> > > > > > which means
> >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> >>> why not
> >>> > > > > > uniform
> >>> > > > > > > > > one? I
> >>> > > > > > > > > > > thought usually users will choose uniform assignor
> >>> (former
> >>> > > > sticky
> >>> > > > > > > > > assinor)
> >>> > > > > > > > > > > for better evenly distribution. Any other reason we
> >>> choose
> >>> > > > range
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > as default?
> >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> >>> server side
> >>> > > > > > assignors.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Thank you.
> >>> > > > > > > > > > > Luke
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> >>> showuon@gmail.com
> >>> > > > >
> >>> > > > > > wrote:
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > > Hi Sagar,
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > I have some thoughts about Kafka Connect
> >>> integrating with
> >>> > > > > > KIP-848,
> >>> > > > > > > > > but I
> >>> > > > > > > > > > > > think we should have a separate discussion thread
> >>> for the
> >>> > > > Kafka
> >>> > > > > > > > > Connect
> >>> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> >>> Rebalance
> >>> > > > > > Protocol
> >>> > > > > > > > > [1],
> >>> > > > > > > > > > > > and let this discussion thread focus on consumer
> >>> rebalance
> >>> > > > > > protocol,
> >>> > > > > > > > > WDYT?
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > [1]
> >>> > > > > > > > > > > >
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > >
> >>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > Thank you.
> >>> > > > > > > > > > > > Luke
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> >>> > > > > > sagarmeansocean@gmail.com>
> >>> > > > > > > > > wrote:
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
> >>> like
> >>> > > > there's
> >>> > > > > > > > > agreement on
> >>> > > > > > > > > > > >> using separate APIs for Connect. I would revisit
> >>> the doc
> >>> > > > and
> >>> > > > > > see
> >>> > > > > > > > > what
> >>> > > > > > > > > > > >> changes are to be made.
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >> Thanks!
> >>> > > > > > > > > > > >> Sagar.
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> >>> > > > > > > > > <dj...@confluent.io.invalid>
> >>> > > > > > > > > > > >> wrote:
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >> > Hi Sagar,
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > Thanks for the feedback and the document.
> >>> That's really
> >>> > > > > > helpful. I
> >>> > > > > > > > > > > >> > will take a look at it.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > Overall, it seems to me that both Connect and
> >>> the
> >>> > > > Consumer
> >>> > > > > > could
> >>> > > > > > > > > share
> >>> > > > > > > > > > > >> > the same underlying "engine". The main
> >>> difference is
> >>> > > > that
> >>> > > > > > the
> >>> > > > > > > > > Consumer
> >>> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> >>> Connect
> >>> > > > assigns
> >>> > > > > > tasks
> >>> > > > > > > > > to
> >>> > > > > > > > > > > >> > workers. I see two ways to move forward:
> >>> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> >>> different
> >>> > > > > > resource
> >>> > > > > > > > > types
> >>> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> >>> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> >>> dedicated
> >>> > > > APIs
> >>> > > > > > > > > would be
> >>> > > > > > > > > > > >> > similar to the new ones but different on the
> >>> > > > > > content/resources and
> >>> > > > > > > > > > > >> > they would rely on the same engine on the
> >>> coordinator
> >>> > > > side.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > I personally lean towards 2) because I am not a
> >>> fan of
> >>> > > > > > > > > overcharging
> >>> > > > > > > > > > > >> > APIs to serve different purposes. That being
> >>> said, I am
> >>> > > > not
> >>> > > > > > > > > opposed to
> >>> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > I think that we can continue to discuss it here
> >>> for now
> >>> > > > in
> >>> > > > > > order
> >>> > > > > > > > > to
> >>> > > > > > > > > > > >> > ensure that this KIP is compatible with what we
> >>> will do
> >>> > > > for
> >>> > > > > > > > > Connect in
> >>> > > > > > > > > > > >> > the future.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > Best,
> >>> > > > > > > > > > > >> > David
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> >>> > > > > > djacot@confluent.io>
> >>> > > > > > > > > wrote:
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > Hi all,
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > I am back from vacation. I will go through
> >>> and address
> >>> > > > > > your
> >>> > > > > > > > > comments
> >>> > > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > Cheers,
> >>> > > > > > > > > > > >> > > David
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> >>> Harris <
> >>> > > > > > > > > gharris1727@gmail.com
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > Hey All!
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> >>> > > > cooperative
> >>> > > > > > > > > rebalancing
> >>> > > > > > > > > > > >> > making it
> >>> > > > > > > > > > > >> > > > down the stack!
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > I had a few questions:
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> >>> describes how
> >>> > > > > > member
> >>> > > > > > > > > epoch
> >>> > > > > > > > > > > >> > should
> >>> > > > > > > > > > > >> > > > advance in step with the group epoch and
> >>> assignment
> >>> > > > > > epoch
> >>> > > > > > > > > values. I
> >>> > > > > > > > > > > >> > think
> >>> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> >>> described
> >>> > > > in
> >>> > > > > > the
> >>> > > > > > > > > KIP. When
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > protocol is incrementally assigning
> >>> partitions to a
> >>> > > > > > worker,
> >>> > > > > > > > > what
> >>> > > > > > > > > > > >> member
> >>> > > > > > > > > > > >> > > > epoch does each incremental assignment use?
> >>> Are
> >>> > > > member
> >>> > > > > > epochs
> >>> > > > > > > > > > > >> re-used,
> >>> > > > > > > > > > > >> > and
> >>> > > > > > > > > > > >> > > > a single member epoch can correspond to
> >>> multiple
> >>> > > > > > different
> >>> > > > > > > > > > > >> > (monotonically
> >>> > > > > > > > > > > >> > > > larger) assignments?
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque
> >>> to the
> >>> > > > group
> >>> > > > > > > > > > > >> coordinator? If
> >>> > > > > > > > > > > >> > > > not, should custom client-side assignor
> >>> > > > implementations
> >>> > > > > > > > > interact
> >>> > > > > > > > > > > >> with
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > Reason field, and how is its common meaning
> >>> agreed
> >>> > > > > > upon? If
> >>> > > > > > > > > so, what
> >>> > > > > > > > > > > >> > is the
> >>> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> >>> including
> >>> > > > such
> >>> > > > > > > > > functionality
> >>> > > > > > > > > > > >> > in the
> >>> > > > > > > > > > > >> > > > opaque metadata?
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> >>> "Thanks to
> >>> > > > > > this, the
> >>> > > > > > > > > input
> >>> > > > > > > > > > > >> of
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > client side assignor is entirely driven by
> >>> the group
> >>> > > > > > > > > coordinator.
> >>> > > > > > > > > > > >> The
> >>> > > > > > > > > > > >> > > > consumer is no longer responsible for
> >>> maintaining
> >>> > > > any
> >>> > > > > > state
> >>> > > > > > > > > besides
> >>> > > > > > > > > > > >> its
> >>> > > > > > > > > > > >> > > > assigned partitions." Does this mean that
> >>> the
> >>> > > > > > client-side
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > >> MAY
> >>> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> >>> (such as
> >>> > > > > > partition
> >>> > > > > > > > > > > >> > throughput,
> >>> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or
> >>> that
> >>> > > > > > additional
> >>> > > > > > > > > > > >> > non-Metadata
> >>> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> >>> > > > > > > > > > > >> > > > for
> >>> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> >>> > > > > > > > > > > >> > > > and
> >>> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> >>> > > > > > that
> >>> > > > > > > > > seem to
> >>> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> >>> these two
> >>> > > > > > > > > implementations
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > unified? This would serve to promote
> >>> feature parity
> >>> > > > of
> >>> > > > > > > > > server-side
> >>> > > > > > > > > > > >> and
> >>> > > > > > > > > > > >> > > > client-side assignors, and would also
> >>> facilitate
> >>> > > > > > operational
> >>> > > > > > > > > > > >> > flexibility in
> >>> > > > > > > > > > > >> > > > certain situations. For example, if a
> >>> server-side
> >>> > > > > > assignor
> >>> > > > > > > > > has some
> >>> > > > > > > > > > > >> > poor
> >>> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> >>> patched
> >>> > > > > > assignor to
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > client
> >>> > > > > > > > > > > >> > > > and switching one consumer group to a
> >>> client-side
> >>> > > > > > assignor
> >>> > > > > > > > > may be
> >>> > > > > > > > > > > >> > faster
> >>> > > > > > > > > > > >> > > > and less risky than patching all of the
> >>> brokers.
> >>> > > > With
> >>> > > > > > the
> >>> > > > > > > > > currently
> >>> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> >>> > > > reimplementation
> >>> > > > > > would
> >>> > > > > > > > > have
> >>> > > > > > > > > > > >> to be
> >>> > > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> >>> > > > > > significantly,
> >>> > > > > > > > > then it
> >>> > > > > > > > > > > >> is
> >>> > > > > > > > > > > >> > > > possible that a reimplementation would not
> >>> be
> >>> > > > possible.
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > --
> >>> > > > > > > > > > > >> > > > Greg Harris
> >>> > > > > > > > > > > >> > > > gharris1727@gmail.com
> >>> > > > > > > > > > > >> > > > github.com/gharris1727
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> >>> > > > > > > > > sagarmeansocean@gmail.com>
> >>> > > > > > > > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > I created a confluence page to discuss
> >>> how Connect
> >>> > > > > > would
> >>> > > > > > > > > need to
> >>> > > > > > > > > > > >> > change
> >>> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> >>> Here's the
> >>> > > > page:
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >>
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > >
> >>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > It's also pretty longish and I have tried
> >>> to keep
> >>> > > > a
> >>> > > > > > format
> >>> > > > > > > > > > > >> similar to
> >>> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> >>> Also, do you
> >>> > > > > > think this
> >>> > > > > > > > > > > >> should
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > > moved to a separate discussion thread or
> >>> is this
> >>> > > > one
> >>> > > > > > fine?
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > Thanks!
> >>> > > > > > > > > > > >> > > > > Sagar.
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> >>> > > > > > > > > sagarmeansocean@gmail.com>
> >>> > > > > > > > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > > Hello Guozhang,
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> >>> Streams.
> >>> > > > > > Sure, I
> >>> > > > > > > > > would do
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > > analysis and come up with such a
> >>> document.
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > > Thanks!
> >>> > > > > > > > > > > >> > > > > > Sagar.
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> >>> Guozhang Wang <
> >>> > > > > > > > > > > >> wangguoz@gmail.com>
> >>> > > > > > > > > > > >> > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> Hello Sagar,
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> It would be great if you could come
> >>> back with
> >>> > > > some
> >>> > > > > > > > > analysis on
> >>> > > > > > > > > > > >> > how to
> >>> > > > > > > > > > > >> > > > > >> implement the Connect side integration
> >>> with
> >>> > > > the new
> >>> > > > > > > > > protocol;
> >>> > > > > > > > > > > >> so
> >>> > > > > > > > > > > >> > far
> >>> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> >>> "protocol type"
> >>> > > > we
> >>> > > > > > did not
> >>> > > > > > > > > yet
> >>> > > > > > > > > > > >> think
> >>> > > > > > > > > > > >> > > > > through
> >>> > > > > > > > > > > >> > > > > >> the Connect side implementations. For
> >>> Streams
> >>> > > > > > here's a
> >>> > > > > > > > > draft of
> >>> > > > > > > > > > > >> > > > > >> integration
> >>> > > > > > > > > > > >> > > > > >> plan:
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >>
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > >
> >>> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >>> > > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar
> >>> <
> >>> > > > > > > > > > > >> sagarmeansocean@gmail.com
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> > Hi David,
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> >>> reason I
> >>> > > > thought
> >>> > > > > > > > > connect can
> >>> > > > > > > > > > > >> > also fit
> >>> > > > > > > > > > > >> > > > > >> into
> >>> > > > > > > > > > > >> > > > > >> > this new scheme is that even today
> >>> the
> >>> > > > connect
> >>> > > > > > uses a
> >>> > > > > > > > > > > >> > > > > WorkerCoordinator
> >>> > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator
> >>> to empower
> >>> > > > > > > > > rebalances of
> >>> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> >>> WorkerCoordinator sets
> >>> > > > the
> >>> > > > > > > > > > > >> protocolType()
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> connect
> >>> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> >>> plumbing
> >>> > > > into
> >>> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > I think the changes to support
> >>> connect would
> >>> > > > be
> >>> > > > > > > > > similar at a
> >>> > > > > > > > > > > >> > high
> >>> > > > > > > > > > > >> > > > > level
> >>> > > > > > > > > > > >> > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> >>> because of the
> >>> > > > > > Client
> >>> > > > > > > > > side
> >>> > > > > > > > > > > >> > assignors
> >>> > > > > > > > > > > >> > > > > being
> >>> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
> >>> level, we
> >>> > > > > > might
> >>> > > > > > > > > need to
> >>> > > > > > > > > > > >> make
> >>> > > > > > > > > > > >> > a lot
> >>> > > > > > > > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> >>> assignment
> >>> > > > protocol
> >>> > > > > > like
> >>> > > > > > > > > > > >> enhancing
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> >>> SyncGroup and
> >>> > > > > > using
> >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> >>> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> >>> streams (or
> >>> > > > > > there
> >>> > > > > > > > > might be
> >>> > > > > > > > > > > >> > > > > >> deviations). I
> >>> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> >>> analysis of
> >>> > > > the
> >>> > > > > > same
> >>> > > > > > > > > and we
> >>> > > > > > > > > > > >> > can have
> >>> > > > > > > > > > > >> > > > > a
> >>> > > > > > > > > > > >> > > > > >> > separate discussion thread for that
> >>> as that
> >>> > > > would
> >>> > > > > > > > > derail this
> >>> > > > > > > > > > > >> > > > > discussion
> >>> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
> >>> good to
> >>> > > > you.
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > Thanks!
> >>> > > > > > > > > > > >> > > > > >> > Sagar.
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> >>> David Jacot
> >>> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> >>> `Assignment#error`.
> >>> > > > > > Sure, I
> >>> > > > > > > > > can
> >>> > > > > > > > > > > >> > mention it.
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> >>> from his
> >>> > > > current
> >>> > > > > > > > > assignment
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > his
> >>> > > > > > > > > > > >> > > > > >> > > target assignment when he can move
> >>> to
> >>> > > > epoch 3.
> >>> > > > > > When
> >>> > > > > > > > > that
> >>> > > > > > > > > > > >> > happens,
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
> >>> persisted
> >>> > > > > > with all
> >>> > > > > > > > > its
> >>> > > > > > > > > > > >> > assigned
> >>> > > > > > > > > > > >> > > > > >> > > partitions even if they are not
> >>> all revoked
> >>> > > > > > yet. In
> >>> > > > > > > > > other
> >>> > > > > > > > > > > >> > words, the
> >>> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> >>> target
> >>> > > > > > assignment.
> >>> > > > > > > > > This is
> >>> > > > > > > > > > > >> > basically
> >>> > > > > > > > > > > >> > > > > an
> >>> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> >>> write all
> >>> > > > the
> >>> > > > > > > > > changes to
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > log.
> >>> > > > > > > > > > > >> > > > > The
> >>> > > > > > > > > > > >> > > > > >> > > examples are based on the
> >>> persisted state
> >>> > > > so I
> >>> > > > > > > > > understand
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> >>> improve
> >>> > > > this in
> >>> > > > > > the
> >>> > > > > > > > > > > >> > description.
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> >>> reuse the
> >>> > > > > > protocol
> >>> > > > > > > > > with a
> >>> > > > > > > > > > > >> > client side
> >>> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> >>> protocol. The
> >>> > > > > > assignment
> >>> > > > > > > > > is
> >>> > > > > > > > > > > >> about
> >>> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> >>> could
> >>> > > > Connect
> >>> > > > > > fit
> >>> > > > > > > > > into this?
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > David
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> >>> Sagar <
> >>> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> >>> > > > > > > > > > > >> > > > > >> wrote:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Hi David,
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
> >>> minor
> >>> > > > > > observations:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> >>> section in
> >>> > > > Client
> >>> > > > > > Side
> >>> > > > > > > > > mode
> >>> > > > > > > > > > > >> > Assignment
> >>> > > > > > > > > > > >> > > > > >> > > process,
> >>> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case,
> >>> the
> >>> > > > client
> >>> > > > > > side
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > >> can
> >>> > > > > > > > > > > >> > > > > return
> >>> > > > > > > > > > > >> > > > > >> an
> >>> > > > > > > > > > > >> > > > > >> > > > error to the group coordinator`.
> >>> In this
> >>> > > > > > case are
> >>> > > > > > > > > you
> >>> > > > > > > > > > > >> > referring to
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> >>> AssignmentError
> >>> > > > that's
> >>> > > > > > > > > listed down
> >>> > > > > > > > > > > >> > towards
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > end?
> >>> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
> >>> make sense
> >>> > > > to
> >>> > > > > > > > > mention this
> >>> > > > > > > > > > > >> > > > > explicitly
> >>> > > > > > > > > > > >> > > > > >> > here?
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section,
> >>> I have a
> >>> > > > > > slight
> >>> > > > > > > > > > > >> confusion,
> >>> > > > > > > > > > > >> > not
> >>> > > > > > > > > > > >> > > > > sure
> >>> > > > > > > > > > > >> > > > > >> if
> >>> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
> >>> this step:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> >>> coordinator
> >>> > > > > > > > > transitions him
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > epoch
> >>> > > > > > > > > > > >> > > > > 3
> >>> > > > > > > > > > > >> > > > > >> > > because
> >>> > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> >>> > > > persists
> >>> > > > > > the
> >>> > > > > > > > > change and
> >>> > > > > > > > > > > >> > reply.
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> >>> > > > > > > > > > > >> > > > > >> > > >       - A
> >>> > > > > > > > > > > >> > > > > >> > > >       - B
> >>> > > > > > > > > > > >> > > > > >> > > >       - C
> >>> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> >>> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> >>> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> >>> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> >>> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> >>> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> >>> partitions=[foo-0,
> >>> > > > > > foo-1]
> >>> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> >>> partitions=[foo-2]
> >>> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> >>> partitions=[foo-1]
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> >>> transitions to
> >>> > > > epoch 3
> >>> > > > > > but
> >>> > > > > > > > > cannot
> >>> > > > > > > > > > > >> get
> >>> > > > > > > > > > > >> > foo-1
> >>> > > > > > > > > > > >> > > > > >> yet.
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member
> >>> C can't
> >>> > > > get
> >>> > > > > > the
> >>> > > > > > > > > foo-1
> >>> > > > > > > > > > > >> > partition
> >>> > > > > > > > > > > >> > > > > yet,
> >>> > > > > > > > > > > >> > > > > >> > but
> >>> > > > > > > > > > > >> > > > > >> > > > based on the description above,
> >>> it seems
> >>> > > > it
> >>> > > > > > > > > already has
> >>> > > > > > > > > > > >> it.
> >>> > > > > > > > > > > >> > Do you
> >>> > > > > > > > > > > >> > > > > >> > think
> >>> > > > > > > > > > > >> > > > > >> > > it
> >>> > > > > > > > > > > >> > > > > >> > > > would be better to remove it and
> >>> > > > populate it
> >>> > > > > > only
> >>> > > > > > > > > when it
> >>> > > > > > > > > > > >> > actually
> >>> > > > > > > > > > > >> > > > > >> gets
> >>> > > > > > > > > > > >> > > > > >> > > it?
> >>> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> >>> places, so
> >>> > > > have
> >>> > > > > > I
> >>> > > > > > > > > > > >> understood it
> >>> > > > > > > > > > > >> > > > > >> > incorrectly
> >>> > > > > > > > > > > >> > > > > >> > > ?
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
> >>> out of
> >>> > > > scope
> >>> > > > > > of
> >>> > > > > > > > > this
> >>> > > > > > > > > > > >> > discussion,
> >>> > > > > > > > > > > >> > > > > but
> >>> > > > > > > > > > > >> > > > > >> > from
> >>> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> >>> probably be
> >>> > > > > > running in
> >>> > > > > > > > > client
> >>> > > > > > > > > > > >> > side
> >>> > > > > > > > > > > >> > > > > >> assignor
> >>> > > > > > > > > > > >> > > > > >> > > mode
> >>> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> >>> protocol as it
> >>> > > > has
> >>> > > > > > its
> >>> > > > > > > > > own
> >>> > > > > > > > > > > >> Custom
> >>> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> >>> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Thanks!
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Sagar.
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
> >>> David
> >>> > > > Jacot
> >>> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
> >>> move
> >>> > > > forward
> >>> > > > > > with
> >>> > > > > > > > > > > >> > specialized API
> >>> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> >>> generic API.
> >>> > > > For
> >>> > > > > > > > > Connect, we
> >>> > > > > > > > > > > >> > can apply
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> >>> reuse/share the
> >>> > > > core
> >>> > > > > > > > > > > >> > implementation on
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> >>> registry, I
> >>> > > > > > think
> >>> > > > > > > > > that we
> >>> > > > > > > > > > > >> > should
> >>> > > > > > > > > > > >> > > > > >> consider
> >>> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> >>> simple
> >>> > > > > > > > > membership/leader
> >>> > > > > > > > > > > >> > election.
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22
> >>> AM Ismael
> >>> > > > > > Juma <
> >>> > > > > > > > > > > >> > ismael@juma.me.uk
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> >>> java.util.regex.Pattern
> >>> > > > vs
> >>> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> >>> > > > > > > > > > > >> > > > > >> > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > document the differences in
> >>> more
> >>> > > > detail
> >>> > > > > > before
> >>> > > > > > > > > > > >> deciding
> >>> > > > > > > > > > > >> > one
> >>> > > > > > > > > > > >> > > > > way
> >>> > > > > > > > > > > >> > > > > >> or
> >>> > > > > > > > > > > >> > > > > >> > > > > another.
> >>> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> >>> > > > > > > > > java.util.regex.Pattern,
> >>> > > > > > > > > > > >> they
> >>> > > > > > > > > > > >> > expect
> >>> > > > > > > > > > > >> > > > > >> > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If
> >>> we are
> >>> > > > doing
> >>> > > > > > > > > something
> >>> > > > > > > > > > > >> > different,
> >>> > > > > > > > > > > >> > > > > >> then
> >>> > > > > > > > > > > >> > > > > >> > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> >>> overload
> >>> > > > with
> >>> > > > > > our own
> >>> > > > > > > > > > > >> Pattern
> >>> > > > > > > > > > > >> > class
> >>> > > > > > > > > > > >> > > > > (I
> >>> > > > > > > > > > > >> > > > > >> > > don't
> >>> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> >>> re2j's at
> >>> > > > this
> >>> > > > > > > > > point).
> >>> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> >>> major new
> >>> > > > > > protocol
> >>> > > > > > > > > should
> >>> > > > > > > > > > > >> > > > > integrate
> >>> > > > > > > > > > > >> > > > > >> > fully
> >>> > > > > > > > > > > >> > > > > >> > > > > with
> >>> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> >>> topic
> >>> > > > > > recreation case
> >>> > > > > > > > > > > >> > correctly.
> >>> > > > > > > > > > > >> > > > > That's
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > main
> >>> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> >>> agree with
> >>> > > > > > David
> >>> > > > > > > > > that we'd
> >>> > > > > > > > > > > >> > want to
> >>> > > > > > > > > > > >> > > > > add
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > ids
> >>> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
> >>> that don't
> >>> > > > > > have it
> >>> > > > > > > > > yet and
> >>> > > > > > > > > > > >> > that we
> >>> > > > > > > > > > > >> > > > > can
> >>> > > > > > > > > > > >> > > > > >> > > probably
> >>> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> >>> versus adding
> >>> > > > new
> >>> > > > > > APIs
> >>> > > > > > > > > to the
> >>> > > > > > > > > > > >> > Java
> >>> > > > > > > > > > > >> > > > > >> Consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > (unless
> >>> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs
> >>> is
> >>> > > > required
> >>> > > > > > for
> >>> > > > > > > > > > > >> reasonable
> >>> > > > > > > > > > > >> > > > > >> semantics).
> >>> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about
> >>> the
> >>> > > > > > coordinator
> >>> > > > > > > > > storing
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> configs.
> >>> > > > > > > > > > > >> > > > > >> > > It's
> >>> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> >>> > > > centralized
> >>> > > > > > in the
> >>> > > > > > > > > > > >> metadata
> >>> > > > > > > > > > > >> > log for
> >>> > > > > > > > > > > >> > > > > >> > > various
> >>> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> >>> visibility,
> >>> > > > > > consistency,
> >>> > > > > > > > > > > >> etc.).
> >>> > > > > > > > > > > >> > > > > >> Similarly, I
> >>> > > > > > > > > > > >> > > > > >> > > am
> >>> > > > > > > > > > > >> > > > > >> > > > > not
> >>> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> >>> deleting
> >>> > > > > > configs in a
> >>> > > > > > > > > way
> >>> > > > > > > > > > > >> that
> >>> > > > > > > > > > > >> > they
> >>> > > > > > > > > > > >> > > > > >> cannot
> >>> > > > > > > > > > > >> > > > > >> > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property
> >>> for modern
> >>> > > > > > systems
> >>> > > > > > > > > is to
> >>> > > > > > > > > > > >> > minimize
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > number
> >>> > > > > > > > > > > >> > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> >>> scenarios.
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47
> >>> PM David
> >>> > > > > > Jacot
> >>> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> >>> answers are
> >>> > > > below:
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> >>> especially
> >>> > > > > > the last
> >>> > > > > > > > > step
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > > > > clients
> >>> > > > > > > > > > > >> > > > > >> > > flip the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> >>> protocol, in
> >>> > > > which
> >>> > > > > > we
> >>> > > > > > > > > would
> >>> > > > > > > > > > > >> have a
> >>> > > > > > > > > > > >> > > > > window
> >>> > > > > > > > > > > >> > > > > >> > where
> >>> > > > > > > > > > > >> > > > > >> > > > > both
> >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> >>> > > > protocols /
> >>> > > > > > rpcs
> >>> > > > > > > > > are
> >>> > > > > > > > > > > >> used
> >>> > > > > > > > > > > >> > by
> >>> > > > > > > > > > > >> > > > > >> members
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> >>> coordinator could
> >>> > > > > > "mimic"
> >>> > > > > > > > > the old
> >>> > > > > > > > > > > >> > behavior
> >>> > > > > > > > > > > >> > > > > >> while
> >>> > > > > > > > > > > >> > > > > >> > > > > using the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> >>> something we
> >>> > > > need to
> >>> > > > > > > > > present
> >>> > > > > > > > > > > >> about.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a
> >>> new
> >>> > > > version
> >>> > > > > > of KIP
> >>> > > > > > > > > which
> >>> > > > > > > > > > > >> > includes
> >>> > > > > > > > > > > >> > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> >>> > > > "Supporting
> >>> > > > > > > > > Online
> >>> > > > > > > > > > > >> > Consumer
> >>> > > > > > > > > > > >> > > > > Group
> >>> > > > > > > > > > > >> > > > > >> > > Upgrade"
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> >>> > > > Deprecation,
> >>> > > > > > and
> >>> > > > > > > > > Migration
> >>> > > > > > > > > > > >> > Plan". I
> >>> > > > > > > > > > > >> > > > > >> think
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
> >>> few cases
> >>> > > > > > now but
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > overall idea
> >>> > > > > > > > > > > >> > > > > >> and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> >>> understandable.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> >>> ids. So
> >>> > > > far as
> >>> > > > > > > > > KIP-516 the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > >> are
> >>> > > > > > > > > > > >> > > > > >> > > only
> >>> > > > > > > > > > > >> > > > > >> > > > > used
> >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> >>> client,
> >>> > > > but
> >>> > > > > > they
> >>> > > > > > > > > are not
> >>> > > > > > > > > > > >> > exposed
> >>> > > > > > > > > > > >> > > > > via
> >>> > > > > > > > > > > >> > > > > >> any
> >>> > > > > > > > > > > >> > > > > >> > > > > public
> >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> >>> think the
> >>> > > > > > question is,
> >>> > > > > > > > > first
> >>> > > > > > > > > > > >> > should we
> >>> > > > > > > > > > > >> > > > > >> let
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> >>> the
> >>> > > > names
> >>> > > > > > -> ids
> >>> > > > > > > > > mapping
> >>> > > > > > > > > > > >> > itself
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > fully
> >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> >>> existing
> >>> > > > RPCs
> >>> > > > > > and the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> RPCs
> >>> > > > > > > > > > > >> > with
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> >>> consider
> >>> > > > > > exposing
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> >>> > > > > > > > > subscribe/assign, as
> >>> > > > > > > > > > > >> well
> >>> > > > > > > > > > > >> > as in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> >>> topic
> >>> > > > > > > > > > > >> > deletion-and-recreation).
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> >>> include
> >>> > > > > > converting
> >>> > > > > > > > > all
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > offsets
> >>> > > > > > > > > > > >> > > > > >> > > related
> >>> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in
> >>> this
> >>> > > > KIP,
> >>> > > > > > the
> >>> > > > > > > > > consumer
> >>> > > > > > > > > > > >> > would be
> >>> > > > > > > > > > > >> > > > > >> able
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
> >>> ids. That
> >>> > > > > > being
> >>> > > > > > > > > said, it
> >>> > > > > > > > > > > >> > still has
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > provide
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> >>> various APIs so
> >>> > > > > > having a
> >>> > > > > > > > > > > >> mapping
> >>> > > > > > > > > > > >> > in the
> >>> > > > > > > > > > > >> > > > > >> > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> >>> opinion on
> >>> > > > > > this.
> >>> > > > > > > > > Here I
> >>> > > > > > > > > > > >> > wonder if
> >>> > > > > > > > > > > >> > > > > >> this
> >>> > > > > > > > > > > >> > > > > >> > > goes
> >>> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
> >>> KIP. I
> >>> > > > would
> >>> > > > > > rather
> >>> > > > > > > > > focus
> >>> > > > > > > > > > > >> on
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > internals
> >>> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider
> >>> this
> >>> > > > > > separately if
> >>> > > > > > > > > we see
> >>> > > > > > > > > > > >> > value in
> >>> > > > > > > > > > > >> > > > > >> doing
> >>> > > > > > > > > > > >> > > > > >> > > it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> >>> point about
> >>> > > > > > using
> >>> > > > > > > > > topic ids
> >>> > > > > > > > > > > >> > in the
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> ConsumerGroupHeartbeatRequest, I
> >>> > > > think
> >>> > > > > > that
> >>> > > > > > > > > there
> >>> > > > > > > > > > > >> is
> >>> > > > > > > > > > > >> > one
> >>> > > > > > > > > > > >> > > > > >> > advantage
> >>> > > > > > > > > > > >> > > > > >> > > in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer
> >>> will
> >>> > > > have
> >>> > > > > > the
> >>> > > > > > > > > > > >> opportunity
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> validate
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> >>> passing
> >>> > > > them
> >>> > > > > > into
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> group
> >>> > > > > > > > > > > >> > > > > rebalance
> >>> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> >>> > > > coordinator
> >>> > > > > > will
> >>> > > > > > > > > also
> >>> > > > > > > > > > > >> notice
> >>> > > > > > > > > > > >> > it but
> >>> > > > > > > > > > > >> > > > > >> it
> >>> > > > > > > > > > > >> > > > > >> > > does
> >>> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> >>> reject an
> >>> > > > > > invalid
> >>> > > > > > > > > topic in
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> response.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> >>> on all
> >>> > > > other
> >>> > > > > > minor
> >>> > > > > > > > > > > >> questions
> >>> > > > > > > > > > > >> > > > > except
> >>> > > > > > > > > > > >> > > > > >> for
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> >>> question:
> >>> > > > > > personally
> >>> > > > > > > > > I think
> >>> > > > > > > > > > > >> > it's not
> >>> > > > > > > > > > > >> > > > > >> > > necessary
> >>> > > > > > > > > > > >> > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> >>> API with
> >>> > > > > > Pattern,
> >>> > > > > > > > > but
> >>> > > > > > > > > > > >> > instead we
> >>> > > > > > > > > > > >> > > > > >> still
> >>> > > > > > > > > > > >> > > > > >> > > use
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> >>> that our
> >>> > > > > > > > > subscription may
> >>> > > > > > > > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> rejected by
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> >>> case is a
> >>> > > > > > very rare
> >>> > > > > > > > > > > >> scenario
> >>> > > > > > > > > > > >> > I felt
> >>> > > > > > > > > > > >> > > > > >> > using
> >>> > > > > > > > > > > >> > > > > >> > > an
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> >>> > > > > > subscription may
> >>> > > > > > > > > be
> >>> > > > > > > > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> vulnerable
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > > > various
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
> >>> look at
> >>> > > > the
> >>> > > > > > > > > differences
> >>> > > > > > > > > > > >> > between
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > two
> >>> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> >>> understand the
> >>> > > > > > potential
> >>> > > > > > > > > issues.
> >>> > > > > > > > > > > >> My
> >>> > > > > > > > > > > >> > > > > >> > > understanding is
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all
> >>> the basic
> >>> > > > > > regular
> >>> > > > > > > > > > > >> > expressions. The
> >>> > > > > > > > > > > >> > > > > >> > > differences
> >>> > > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
> >>> about
> >>> > > > the
> >>> > > > > > various
> >>> > > > > > > > > > > >> character
> >>> > > > > > > > > > > >> > > > > >> classes. I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
> >>> think
> >>> > > > about
> >>> > > > > > this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> >>> 11:28 PM
> >>> > > > > > Guozhang
> >>> > > > > > > > > Wang <
> >>> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on
> >>> the high
> >>> > > > > > level
> >>> > > > > > > > > there are
> >>> > > > > > > > > > > >> > two meta
> >>> > > > > > > > > > > >> > > > > >> > > points we
> >>> > > > > > > > > > > >> > > > > >> > > > > need
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> >>> especially
> >>> > > > > > the last
> >>> > > > > > > > > step
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > > > > clients
> >>> > > > > > > > > > > >> > > > > >> > > flip the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> >>> protocol, in
> >>> > > > which
> >>> > > > > > we
> >>> > > > > > > > > would
> >>> > > > > > > > > > > >> have a
> >>> > > > > > > > > > > >> > > > > window
> >>> > > > > > > > > > > >> > > > > >> > where
> >>> > > > > > > > > > > >> > > > > >> > > > > both
> >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> >>> > > > protocols /
> >>> > > > > > rpcs
> >>> > > > > > > > > are
> >>> > > > > > > > > > > >> used
> >>> > > > > > > > > > > >> > by
> >>> > > > > > > > > > > >> > > > > >> members
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> >>> coordinator could
> >>> > > > > > "mimic"
> >>> > > > > > > > > the old
> >>> > > > > > > > > > > >> > behavior
> >>> > > > > > > > > > > >> > > > > >> while
> >>> > > > > > > > > > > >> > > > > >> > > > > using the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> >>> something we
> >>> > > > need to
> >>> > > > > > > > > present
> >>> > > > > > > > > > > >> about.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> >>> ids. So
> >>> > > > far as
> >>> > > > > > > > > KIP-516 the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > >> are
> >>> > > > > > > > > > > >> > > > > >> > > only
> >>> > > > > > > > > > > >> > > > > >> > > > > used
> >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> >>> client,
> >>> > > > but
> >>> > > > > > they
> >>> > > > > > > > > are not
> >>> > > > > > > > > > > >> > exposed
> >>> > > > > > > > > > > >> > > > > via
> >>> > > > > > > > > > > >> > > > > >> any
> >>> > > > > > > > > > > >> > > > > >> > > > > public
> >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> >>> think the
> >>> > > > > > question is,
> >>> > > > > > > > > first
> >>> > > > > > > > > > > >> > should we
> >>> > > > > > > > > > > >> > > > > >> let
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> >>> the
> >>> > > > names
> >>> > > > > > -> ids
> >>> > > > > > > > > mapping
> >>> > > > > > > > > > > >> > itself
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > fully
> >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> >>> existing
> >>> > > > RPCs
> >>> > > > > > and the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> RPCs
> >>> > > > > > > > > > > >> > with
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> >>> consider
> >>> > > > > > exposing
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> >>> > > > > > > > > subscribe/assign, as
> >>> > > > > > > > > > > >> well
> >>> > > > > > > > > > > >> > as in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> >>> topic
> >>> > > > > > > > > > > >> > deletion-and-recreation).
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> >>> on all
> >>> > > > other
> >>> > > > > > minor
> >>> > > > > > > > > > > >> questions
> >>> > > > > > > > > > > >> > > > > except
> >>> > > > > > > > > > > >> > > > > >> for
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> >>> question:
> >>> > > > > > personally
> >>> > > > > > > > > I think
> >>> > > > > > > > > > > >> > it's not
> >>> > > > > > > > > > > >> > > > > >> > > necessary
> >>> > > > > > > > > > > >> > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> >>> API with
> >>> > > > > > Pattern,
> >>> > > > > > > > > but
> >>> > > > > > > > > > > >> > instead we
> >>> > > > > > > > > > > >> > > > > >> still
> >>> > > > > > > > > > > >> > > > > >> > > use
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> >>> that our
> >>> > > > > > > > > subscription may
> >>> > > > > > > > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> rejected by
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> >>> case is a
> >>> > > > > > very rare
> >>> > > > > > > > > > > >> scenario
> >>> > > > > > > > > > > >> > I felt
> >>> > > > > > > > > > > >> > > > > >> > using
> >>> > > > > > > > > > > >> > > > > >> > > an
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> >>> > > > > > subscription may
> >>> > > > > > > > > be
> >>> > > > > > > > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> vulnerable
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > > > various
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
> >>> 5:23 AM
> >>> > > > > > David Jacot
> >>> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> >>> feedback. Let
> >>> > > > me
> >>> > > > > > answer
> >>> > > > > > > > > your
> >>> > > > > > > > > > > >> > questions
> >>> > > > > > > > > > > >> > > > > >> > inline.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> >>> premature to
> >>> > > > > > talk about
> >>> > > > > > > > > > > >> target
> >>> > > > > > > > > > > >> > > > > versions
> >>> > > > > > > > > > > >> > > > > >> for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> >>> existing group
> >>> > > > > > protocol.
> >>> > > > > > > > > > > >> Unlike
> >>> > > > > > > > > > > >> > KRaft,
> >>> > > > > > > > > > > >> > > > > >> this
> >>> > > > > > > > > > > >> > > > > >> > > > > affects a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > core
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
> >>> hence
> >>> > > > > > > > > deprecation/removal
> >>> > > > > > > > > > > >> > will be
> >>> > > > > > > > > > > >> > > > > >> > heavily
> >>> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> >>> applications
> >>> > > > migrate
> >>> > > > > > to
> >>> > > > > > > > > the new
> >>> > > > > > > > > > > >> > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I
> >>> will
> >>> > > > remove it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> >>> intend to
> >>> > > > > > release
> >>> > > > > > > > > this in
> >>> > > > > > > > > > > >> > 4.x, but
> >>> > > > > > > > > > > >> > > > > it
> >>> > > > > > > > > > > >> > > > > >> > > wasn't
> >>> > > > > > > > > > > >> > > > > >> > > > > made
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> >>> that as a
> >>> > > > way to
> >>> > > > > > > > > estimate
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > we'd
> >>> > > > > > > > > > > >> > > > > >> > > deprecate
> >>> > > > > > > > > > > >> > > > > >> > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > remove
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol,
> >>> I also
> >>> > > > > > suggest
> >>> > > > > > > > > removing
> >>> > > > > > > > > > > >> > this part.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> >>> reasoning. As
> >>> > > > > > > > > explained, I
> >>> > > > > > > > > > > >> plan
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> rewrite
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> >>> while we
> >>> > > > > > implement
> >>> > > > > > > > > the new
> >>> > > > > > > > > > > >> > protocol.
> >>> > > > > > > > > > > >> > > > > >> This
> >>> > > > > > > > > > > >> > > > > >> > > means
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> >>> will be
> >>> > > > slightly
> >>> > > > > > > > > different
> >>> > > > > > > > > > > >> > (e.g.
> >>> > > > > > > > > > > >> > > > > >> threading
> >>> > > > > > > > > > > >> > > > > >> > > > > model).
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
> >>> tighten
> >>> > > > the
> >>> > > > > > > > > switch from
> >>> > > > > > > > > > > >> > the old
> >>> > > > > > > > > > > >> > > > > >> group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
> >>> group
> >>> > > > > > coordinator
> >>> > > > > > > > > to a
> >>> > > > > > > > > > > >> > major
> >>> > > > > > > > > > > >> > > > > >> release.
> >>> > > > > > > > > > > >> > > > > >> > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be
> >>> to use a
> >>> > > > flag
> >>> > > > > > to do
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > switch
> >>> > > > > > > > > > > >> > > > > >> instead
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > > > relying
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> >>> upgrade.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
> >>> out the
> >>> > > > > > details of
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > migration
> >>> > > > > > > > > > > >> > > > > >> story.
> >>> > > > > > > > > > > >> > > > > >> > > It
> >>> > > > > > > > > > > >> > > > > >> > > > > sounds
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
> >>> support
> >>> > > > > > online
> >>> > > > > > > > > > > >> migrations.
> >>> > > > > > > > > > > >> > Is
> >>> > > > > > > > > > > >> > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > correct?
> >>> > > > > > > > > > > >> > > > > >> > > > > We
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> >>> detail. It
> >>> > > > could
> >>> > > > > > also
> >>> > > > > > > > > be done
> >>> > > > > > > > > > > >> > as a
> >>> > > > > > > > > > > >> > > > > >> separate
> >>> > > > > > > > > > > >> > > > > >> > > KIP,
> >>> > > > > > > > > > > >> > > > > >> > > > > if
> >>> > > > > > > > > > > >> > > > > >> > > > > > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
> >>> online
> >>> > > > > > migrations
> >>> > > > > > > > > for the
> >>> > > > > > > > > > > >> > group.
> >>> > > > > > > > > > > >> > > > > That
> >>> > > > > > > > > > > >> > > > > >> > > means
> >>> > > > > > > > > > > >> > > > > >> > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> >>> protocol
> >>> > > > > > will be
> >>> > > > > > > > > able to
> >>> > > > > > > > > > > >> > switch to
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain
> >>> how that
> >>> > > > > > will work
> >>> > > > > > > > > > > >> though.
> >>> > > > > > > > > > > >> > It is
> >>> > > > > > > > > > > >> > > > > >> > > basically a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> >>> > > > upgraded or
> >>> > > > > > > > > rolled to a
> >>> > > > > > > > > > > >> > software
> >>> > > > > > > > > > > >> > > > > >> > > supporting
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
> >>> Both the
> >>> > > > > > old and
> >>> > > > > > > > > the new
> >>> > > > > > > > > > > >> > > > > >> coordinator
> >>> > > > > > > > > > > >> > > > > >> > > will
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> >>> protocol and
> >>> > > > rely
> >>> > > > > > on the
> >>> > > > > > > > > same
> >>> > > > > > > > > > > >> > persisted
> >>> > > > > > > > > > > >> > > > > >> > > metadata so
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> >>> together. This
> >>> > > > point
> >>> > > > > > is an
> >>> > > > > > > > > > > >> offline
> >>> > > > > > > > > > > >> > > > > >> migration.
> >>> > > > > > > > > > > >> > > > > >> > We
> >>> > > > > > > > > > > >> > > > > >> > > > > cannot
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> >>> because it
> >>> > > > would
> >>> > > > > > require
> >>> > > > > > > > > > > >> > shutting down
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > current
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> >>> starting up the
> >>> > > > > > new one
> >>> > > > > > > > > and
> >>> > > > > > > > > > > >> that
> >>> > > > > > > > > > > >> > would
> >>> > > > > > > > > > > >> > > > > >> cause
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> >>> metadata
> >>> > > > > > version/IBP
> >>> > > > > > > > > must be
> >>> > > > > > > > > > > >> > upgraded
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> X
> >>> > > > > > > > > > > >> > > > > >> > in
> >>> > > > > > > > > > > >> > > > > >> > > > > order
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> >>> protocol.
> >>> > > > This
> >>> > > > > > cannot
> >>> > > > > > > > > be done
> >>> > > > > > > > > > > >> > before
> >>> > > > > > > > > > > >> > > > > 1)
> >>> > > > > > > > > > > >> > > > > >> is
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the
> >>> old
> >>> > > > > > coordinator
> >>> > > > > > > > > doesn't
> >>> > > > > > > > > > > >> > support
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must
> >>> be
> >>> > > > upgraded
> >>> > > > > > to a
> >>> > > > > > > > > version
> >>> > > > > > > > > > > >> > > > > supporting
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > online
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> >>> KIP-792).
> >>> > > > If
> >>> > > > > > the
> >>> > > > > > > > > consumer
> >>> > > > > > > > > > > >> is
> >>> > > > > > > > > > > >> > > > > already
> >>> > > > > > > > > > > >> > > > > >> > > there.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done
> >>> at this
> >>> > > > point.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must
> >>> be rolled
> >>> > > > > > with the
> >>> > > > > > > > > > > >> feature
> >>> > > > > > > > > > > >> > flag
> >>> > > > > > > > > > > >> > > > > >> turned
> >>> > > > > > > > > > > >> > > > > >> > > on.
> >>> > > > > > > > > > > >> > > > > >> > > > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> >>> automatically
> >>> > > > > > > > > converted when
> >>> > > > > > > > > > > >> > the first
> >>> > > > > > > > > > > >> > > > > >> > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
> >>> joins
> >>> > > > the
> >>> > > > > > group.
> >>> > > > > > > > > While
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > members
> >>> > > > > > > > > > > >> > > > > >> > > using the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> >>> > > > upgraded,
> >>> > > > > > the old
> >>> > > > > > > > > > > >> protocol
> >>> > > > > > > > > > > >> > is
> >>> > > > > > > > > > > >> > > > > >> proxied
> >>> > > > > > > > > > > >> > > > > >> > > into
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
> >>> this in
> >>> > > > the
> >>> > > > > > KIP.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that
> >>> we are
> >>> > > > > > pushing the
> >>> > > > > > > > > pattern
> >>> > > > > > > > > > > >> > > > > >> subscriptions
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
> >>> there
> >>> > > > could be
> >>> > > > > > some
> >>> > > > > > > > > tricky
> >>> > > > > > > > > > > >> > > > > >> compatibility
> >>> > > > > > > > > > > >> > > > > >> > > > > issues.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Will
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
> >>> users to
> >>> > > > > > detect
> >>> > > > > > > > > that
> >>> > > > > > > > > > > >> they
> >>> > > > > > > > > > > >> > need to
> >>> > > > > > > > > > > >> > > > > >> > update
> >>> > > > > > > > > > > >> > > > > >> > > > > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > > regex
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
> >>> the new
> >>> > > > > > protocol?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a
> >>> bit more
> >>> > > > > > optimistic
> >>> > > > > > > > > than
> >>> > > > > > > > > > > >> you
> >>> > > > > > > > > > > >> > on this
> >>> > > > > > > > > > > >> > > > > >> > > point. I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> >>> majority of
> >>> > > > the
> >>> > > > > > cases
> >>> > > > > > > > > are
> >>> > > > > > > > > > > >> simple
> >>> > > > > > > > > > > >> > > > > regexes
> >>> > > > > > > > > > > >> > > > > >> > which
> >>> > > > > > > > > > > >> > > > > >> > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> >>> engine. The
> >>> > > > > > > > > coordinator will
> >>> > > > > > > > > > > >> > verify
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > regex
> >>> > > > > > > > > > > >> > > > > >> > > > > anyway
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> >>> consumer if the
> >>> > > > > > regex is
> >>> > > > > > > > > not
> >>> > > > > > > > > > > >> valid.
> >>> > > > > > > > > > > >> > > > > Coming
> >>> > > > > > > > > > > >> > > > > >> > back
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
> >>> worst
> >>> > > > case,
> >>> > > > > > the
> >>> > > > > > > > > first
> >>> > > > > > > > > > > >> > upgraded
> >>> > > > > > > > > > > >> > > > > >> > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > joining
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> >>> rejected.
> >>> > > > This
> >>> > > > > > should
> >>> > > > > > > > > be used
> >>> > > > > > > > > > > >> > as the
> >>> > > > > > > > > > > >> > > > > >> last
> >>> > > > > > > > > > > >> > > > > >> > > > > defence, I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers
> >>> to
> >>> > > > validate
> >>> > > > > > their
> >>> > > > > > > > > regex
> >>> > > > > > > > > > > >> > before
> >>> > > > > > > > > > > >> > > > > >> > upgrading
> >>> > > > > > > > > > > >> > > > > >> > > > > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
> >>> them with
> >>> > > > > > another
> >>> > > > > > > > > group.
> >>> > > > > > > > > > > >> For
> >>> > > > > > > > > > > >> > > > > >> instance,
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> >>> pre-prod
> >>> > > > > > environment.
> >>> > > > > > > > > > > >> Another
> >>> > > > > > > > > > > >> > way
> >>> > > > > > > > > > > >> > > > > >> would be
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> >>> consumer-group tool
> >>> > > > to
> >>> > > > > > > > > provide a
> >>> > > > > > > > > > > >> regex
> >>> > > > > > > > > > > >> > > > > >> validation
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this
> >>> be
> >>> > > > enough in
> >>> > > > > > your
> >>> > > > > > > > > > > >> opinion?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the
> >>> last
> >>> > > > question,
> >>> > > > > > will
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> Java
> >>> > > > > > > > > > > >> > client
> >>> > > > > > > > > > > >> > > > > >> allow
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > users to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> >>> current regex
> >>> > > > > > engine for
> >>> > > > > > > > > > > >> > compatibility
> >>> > > > > > > > > > > >> > > > > >> > > reasons?
> >>> > > > > > > > > > > >> > > > > >> > > > > For
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
> >>> keep using
> >>> > > > > > client
> >>> > > > > > > > > based
> >>> > > > > > > > > > > >> > regex at
> >>> > > > > > > > > > > >> > > > > >> first
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > keep
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple
> >>> and then
> >>> > > > > > migrate to
> >>> > > > > > > > > server
> >>> > > > > > > > > > > >> > based
> >>> > > > > > > > > > > >> > > > > >> regexes
> >>> > > > > > > > > > > >> > > > > >> > > as a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > second
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> >>> point but I
> >>> > > > am
> >>> > > > > > > > > concerned that
> >>> > > > > > > > > > > >> > this
> >>> > > > > > > > > > > >> > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > allow
> >>> > > > > > > > > > > >> > > > > >> > > > > users
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
> >>> this mode.
> >>> > > > That
> >>> > > > > > > > > would go
> >>> > > > > > > > > > > >> > against our
> >>> > > > > > > > > > > >> > > > > >> goal
> >>> > > > > > > > > > > >> > > > > >> > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
> >>> because
> >>> > > > we
> >>> > > > > > would
> >>> > > > > > > > > have to
> >>> > > > > > > > > > > >> > continue
> >>> > > > > > > > > > > >> > > > > >> > > monitoring
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> >>> client
> >>> > > > side. I
> >>> > > > > > would
> >>> > > > > > > > > rather
> >>> > > > > > > > > > > >> > not do
> >>> > > > > > > > > > > >> > > > > >> this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
> >>> the group
> >>> > > > > > > > > coordinator
> >>> > > > > > > > > > > >> will
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > > >> > > responsible for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > storing
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations
> >>> and that
> >>> > > > the
> >>> > > > > > > > > > > >> configurations
> >>> > > > > > > > > > > >> > will be
> >>> > > > > > > > > > > >> > > > > >> > > deleted
> >>> > > > > > > > > > > >> > > > > >> > > > > when
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
> >>> Will a
> >>> > > > > > transition to
> >>> > > > > > > > > DEAD
> >>> > > > > > > > > > > >> > trigger
> >>> > > > > > > > > > > >> > > > > >> > deletion
> >>> > > > > > > > > > > >> > > > > >> > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> >>> > > > configurations
> >>> > > > > > will be
> >>> > > > > > > > > > > >> deleted
> >>> > > > > > > > > > > >> > when
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > group is
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> >>> together.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice
> >>> to store
> >>> > > > the
> >>> > > > > > > > > configs in
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > group
> >>> > > > > > > > > > > >> > > > > >> > > coordinator
> >>> > > > > > > > > > > >> > > > > >> > > > > > > make it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> >>> cluster
> >>> > > > > > configs and
> >>> > > > > > > > > their
> >>> > > > > > > > > > > >> > values?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
> >>> group
> >>> > > > > > > > > configurations are
> >>> > > > > > > > > > > >> > overrides
> >>> > > > > > > > > > > >> > > > > >> of
> >>> > > > > > > > > > > >> > > > > >> > > cluster
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want
> >>> to know
> >>> > > > all
> >>> > > > > > the
> >>> > > > > > > > > overrides
> >>> > > > > > > > > > > >> > though,
> >>> > > > > > > > > > > >> > > > > you
> >>> > > > > > > > > > > >> > > > > >> > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > have
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> >>> > > > coordinators.
> >>> > > > > > You
> >>> > > > > > > > > cannot
> >>> > > > > > > > > > > >> > rely on
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > metadata
> >>> > > > > > > > > > > >> > > > > >> > > > > log
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> >>> > > > configure a
> >>> > > > > > group
> >>> > > > > > > > > before
> >>> > > > > > > > > > > >> > starting
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > consumers?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> >>> allowing the
> >>> > > > > > explicit
> >>> > > > > > > > > creation
> >>> > > > > > > > > > > >> of
> >>> > > > > > > > > > > >> > > > > groups?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> >>> decoupled
> >>> > > > from
> >>> > > > > > the
> >>> > > > > > > > > group
> >>> > > > > > > > > > > >> > lifecycle.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> >>> > > > > > automatically
> >>> > > > > > > > > created in
> >>> > > > > > > > > > > >> > this
> >>> > > > > > > > > > > >> > > > > case.
> >>> > > > > > > > > > > >> > > > > >> > > However,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
> >>> lost after
> >>> > > > the
> >>> > > > > > > > > retention
> >>> > > > > > > > > > > >> > period of
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> >>> > > > Consumer.subscribe
> >>> > > > > > method
> >>> > > > > > > > > for the
> >>> > > > > > > > > > > >> > Java
> >>> > > > > > > > > > > >> > > > > >> client
> >>> > > > > > > > > > > >> > > > > >> > > still
> >>> > > > > > > > > > > >> > > > > >> > > > > take
> >>> > > > > > > > > > > >> > > > > >> > > > > > > a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> `java.util.regex.Pattern` of
> >>> > > > do
> >>> > > > > > we
> >>> > > > > > > > > have to
> >>> > > > > > > > > > > >> > introduce
> >>> > > > > > > > > > > >> > > > > an
> >>> > > > > > > > > > > >> > > > > >> > > overload?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> >>> question. I
> >>> > > > > > forgot
> >>> > > > > > > > > about
> >>> > > > > > > > > > > >> that
> >>> > > > > > > > > > > >> > one.
> >>> > > > > > > > > > > >> > > > > As
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> `java.util.regex.Pattern` is
> >>> > > > not
> >>> > > > > > fully
> >>> > > > > > > > > > > >> compatible
> >>> > > > > > > > > > > >> > with
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > engine
> >>> > > > > > > > > > > >> > > > > >> > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> >>> might be
> >>> > > > better
> >>> > > > > > to
> >>> > > > > > > > > deprecate
> >>> > > > > > > > > > > >> > it and
> >>> > > > > > > > > > > >> > > > > >> use an
> >>> > > > > > > > > > > >> > > > > >> > > > > overload
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string.
> >>> We would
> >>> > > > > > rely on
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> server
> >>> > > > > > > > > > > >> > side
> >>> > > > > > > > > > > >> > > > > >> > > validation.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration,
> >>> I think
> >>> > > > that
> >>> > > > > > we
> >>> > > > > > > > > could
> >>> > > > > > > > > > > >> still
> >>> > > > > > > > > > > >> > try to
> >>> > > > > > > > > > > >> > > > > >> > > toString
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
> >>> should
> >>> > > > > > work, I
> >>> > > > > > > > > think, in
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> majority
> >>> > > > > > > > > > > >> > > > > >> > > of the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> >>> Justine
> >>> > > > that we
> >>> > > > > > > > > should be
> >>> > > > > > > > > > > >> > clearer
> >>> > > > > > > > > > > >> > > > > about
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > > > reason
> >>> > > > > > > > > > > >> > > > > >> > > > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> >>> > > > IBP/metadata.version
> >>> > > > > > from the
> >>> > > > > > > > > > > >> feature
> >>> > > > > > > > > > > >> > flag.
> >>> > > > > > > > > > > >> > > > > >> Maybe
> >>> > > > > > > > > > > >> > > > > >> > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > mean
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> >>> default for
> >>> > > > the
> >>> > > > > > feature
> >>> > > > > > > > > flag
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > true
> >>> > > > > > > > > > > >> > > > > >> based
> >>> > > > > > > > > > > >> > > > > >> > on
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> >>> once we
> >>> > > > want to
> >>> > > > > > make
> >>> > > > > > > > > it the
> >>> > > > > > > > > > > >> > default.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> >>> that feature
> >>> > > > > > flag
> >>> > > > > > > > > mainly
> >>> > > > > > > > > > > >> > during the
> >>> > > > > > > > > > > >> > > > > >> > > development
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not
> >>> have
> >>> > > > mentioned
> >>> > > > > > it, I
> >>> > > > > > > > > think,
> >>> > > > > > > > > > > >> > because
> >>> > > > > > > > > > > >> > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > could
> >>> > > > > > > > > > > >> > > > > >> > > > > use
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for
> >>> it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> >>> protocol APIs
> >>> > > > > > don't
> >>> > > > > > > > > mention
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > required
> >>> > > > > > > > > > > >> > > > > >> > > ACLs, it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> >>> > > > consistency.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
> >>> that
> >>> > > > > > > > > > > >> ConsumerGroupHeartbeat
> >>> > > > > > > > > > > >> > > > > requires
> >>> > > > > > > > > > > >> > > > > >> > > "Read
> >>> > > > > > > > > > > >> > > > > >> > > > > Group"
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to
> >>> do more
> >>> > > > than
> >>> > > > > > > > > reading.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how
> >>> the
> >>> > > > current
> >>> > > > > > > > > protocol works
> >>> > > > > > > > > > > >> > though.
> >>> > > > > > > > > > > >> > > > > We
> >>> > > > > > > > > > > >> > > > > >> > only
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group"
> >>> to join a
> >>> > > > > > group. We
> >>> > > > > > > > > could
> >>> > > > > > > > > > > >> > consider
> >>> > > > > > > > > > > >> > > > > >> > > changing
> >>> > > > > > > > > > > >> > > > > >> > > > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that
> >>> it is
> >>> > > > worth
> >>> > > > > > it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> >>> recreation
> >>> > > > > > handled by
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > consumer
> >>> > > > > > > > > > > >> > > > > with
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would
> >>> be good to
> >>> > > > > > have a
> >>> > > > > > > > > section
> >>> > > > > > > > > > > >> on
> >>> > > > > > > > > > > >> > this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> >>> > > > > > perspective, the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> topic
> >>> > > > > > > > > > > >> > will
> >>> > > > > > > > > > > >> > > > > >> have a
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> >>> treat it
> >>> > > > like a
> >>> > > > > > > > > topic with
> >>> > > > > > > > > > > >> a
> >>> > > > > > > > > > > >> > > > > different
> >>> > > > > > > > > > > >> > > > > >> > > name.
> >>> > > > > > > > > > > >> > > > > >> > > > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> >>> > > > fetch/commit
> >>> > > > > > > > > offsets
> >>> > > > > > > > > > > >> APIs
> >>> > > > > > > > > > > >> > do not
> >>> > > > > > > > > > > >> > > > > >> > support
> >>> > > > > > > > > > > >> > > > > >> > > > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
> >>> would
> >>> > > > reuse the
> >>> > > > > > > > > offsets
> >>> > > > > > > > > > > >> based
> >>> > > > > > > > > > > >> > on the
> >>> > > > > > > > > > > >> > > > > >> > same.
> >>> > > > > > > > > > > >> > > > > >> > > I
> >>> > > > > > > > > > > >> > > > > >> > > > > think
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
> >>> those
> >>> > > > APIs
> >>> > > > > > as well
> >>> > > > > > > > > in
> >>> > > > > > > > > > > >> order
> >>> > > > > > > > > > > >> > to be
> >>> > > > > > > > > > > >> > > > > >> > > consistent
> >>> > > > > > > > > > > >> > > > > >> > > > > end
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> >>> strengthen
> >>> > > > the
> >>> > > > > > > > > semantics of
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> consumer.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions
> >>> we will
> >>> > > > > > write the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> > coordinator
> >>> > > > > > > > > > > >> > > > > >> in
> >>> > > > > > > > > > > >> > > > > >> > > Java.
> >>> > > > > > > > > > > >> > > > > >> > > > > Even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> >>> implementation
> >>> > > > > > detail, do
> >>> > > > > > > > > we plan
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > have a
> >>> > > > > > > > > > > >> > > > > >> new
> >>> > > > > > > > > > > >> > > > > >> > > gradle
> >>> > > > > > > > > > > >> > > > > >> > > > > > > module
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> >>> scalability
> >>> > > > > > goal when
> >>> > > > > > > > > it
> >>> > > > > > > > > > > >> comes
> >>> > > > > > > > > > > >> > to how
> >>> > > > > > > > > > > >> > > > > >> many
> >>> > > > > > > > > > > >> > > > > >> > > > > members
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> >>> support?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers
> >>> at the
> >>> > > > > > moment. The
> >>> > > > > > > > > > > >> protocol
> >>> > > > > > > > > > > >> > should
> >>> > > > > > > > > > > >> > > > > >> > > support
> >>> > > > > > > > > > > >> > > > > >> > > > > 1000s
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group.
> >>> We will
> >>> > > > > > measure
> >>> > > > > > > > > this when
> >>> > > > > > > > > > > >> > we have
> >>> > > > > > > > > > > >> > > > > a
> >>> > > > > > > > > > > >> > > > > >> > first
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
> >>> that we
> >>> > > > might
> >>> > > > > > have
> >>> > > > > > > > > other
> >>> > > > > > > > > > > >> > > > > bottlenecks
> >>> > > > > > > > > > > >> > > > > >> > down
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> >>> commits).
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> >>> having
> >>> > > > > > > > > SubscribedTopidIds
> >>> > > > > > > > > > > >> > instead
> >>> > > > > > > > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames
> >>> in
> >>> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> >>> > > > > > > > > > > >> > > > > >> Is
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > > > idea
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> >>> resolve the
> >>> > > > > > regex on
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > server, we
> >>> > > > > > > > > > > >> > > > > >> can do
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> >>> > > > difference
> >>> > > > > > is that
> >>> > > > > > > > > > > >> sending
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> regex is
> >>> > > > > > > > > > > >> > > > > >> > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
> >>> topic
> >>> > > > names
> >>> > > > > > is less
> >>> > > > > > > > > > > >> > efficient.
> >>> > > > > > > > > > > >> > > > > >> > > Furthermore,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > delete
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier
> >>> to
> >>> > > > handle
> >>> > > > > > if we
> >>> > > > > > > > > have
> >>> > > > > > > > > > > >> topic
> >>> > > > > > > > > > > >> > ids.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> >>> consolidate the
> >>> > > > > > metadata
> >>> > > > > > > > > lookup
> >>> > > > > > > > > > > >> > on the
> >>> > > > > > > > > > > >> > > > > >> server
> >>> > > > > > > > > > > >> > > > > >> > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > both
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
> >>> with your
> >>> > > > > > point. As
> >>> > > > > > > > > a
> >>> > > > > > > > > > > >> second
> >>> > > > > > > > > > > >> > > > > though,
> >>> > > > > > > > > > > >> > > > > >> > using
> >>> > > > > > > > > > > >> > > > > >> > > > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here
> >>> for the
> >>> > > > > > delete and
> >>> > > > > > > > > > > >> > recreation
> >>> > > > > > > > > > > >> > > > > case.
> >>> > > > > > > > > > > >> > > > > >> > > Also, I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
> >>> allow
> >>> > > > users to
> >>> > > > > > > > > subscribe
> >>> > > > > > > > > > > >> with
> >>> > > > > > > > > > > >> > topic
> >>> > > > > > > > > > > >> > > > > >> ids
> >>> > > > > > > > > > > >> > > > > >> > in
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that is
> >>> the only
> >>> > > > > > way to be
> >>> > > > > > > > > > > >> really
> >>> > > > > > > > > > > >> > robust
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022
> >>> at 1:38 PM
> >>> > > > > > David
> >>> > > > > > > > > Jacot <
> >>> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> >>> comments.
> >>> > > > Please
> >>> > > > > > find
> >>> > > > > > > > > my
> >>> > > > > > > > > > > >> answers
> >>> > > > > > > > > > > >> > > > > below.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> >>> protocol
> >>> > > > relies on
> >>> > > > > > > > > topic IDs
> >>> > > > > > > > > > > >> > with the
> >>> > > > > > > > > > > >> > > > > >> > > exception
> >>> > > > > > > > > > > >> > > > > >> > > > > of the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in
> >>> the
> >>> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> >>> > > > > > > > > > > >> > > > > >> I
> >>> > > > > > > > > > > >> > > > > >> > am
> >>> > > > > > > > > > > >> > > > > >> > > not
> >>> > > > > > > > > > > >> > > > > >> > > > > sure
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names
> >>> is the
> >>> > > > > > right call
> >>> > > > > > > > > here.
> >>> > > > > > > > > > > >> I
> >>> > > > > > > > > > > >> > need to
> >>> > > > > > > > > > > >> > > > > >> > think
> >>> > > > > > > > > > > >> > > > > >> > > > > about it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> >>> Obviously, the
> >>> > > > > > KIP does
> >>> > > > > > > > > not
> >>> > > > > > > > > > > >> > change the
> >>> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
> >>> topic
> >>> > > > IDs.
> >>> > > > > > This
> >>> > > > > > > > > may be
> >>> > > > > > > > > > > >> > something
> >>> > > > > > > > > > > >> > > > > >> that
> >>> > > > > > > > > > > >> > > > > >> > we
> >>> > > > > > > > > > > >> > > > > >> > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
> >>> would
> >>> > > > give
> >>> > > > > > better
> >>> > > > > > > > > > > >> overall
> >>> > > > > > > > > > > >> > > > > >> guarantee in
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> >>> think that
> >>> > > > I
> >>> > > > > > should
> >>> > > > > > > > > not
> >>> > > > > > > > > > > >> have
> >>> > > > > > > > > > > >> > > > > mentioned
> >>> > > > > > > > > > > >> > > > > >> > this
> >>> > > > > > > > > > > >> > > > > >> > > > > flag at
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove
> >>> it. We can
> >>> > > > > > use an
> >>> > > > > > > > > internal
> >>> > > > > > > > > > > >> > > > > >> configuration
> >>> > > > > > > > > > > >> > > > > >> > > while
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> >>> feature.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
> >>> will be
> >>> > > > > > > > > supported. The
> >>> > > > > > > > > > > >> > change is
> >>> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
> >>> that the
> >>> > > > > > cluster
> >>> > > > > > > > > uses
> >>> > > > > > > > > > > >> topic
> >>> > > > > > > > > > > >> > IDs.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022
> >>> at 9:53
> >>> > > > PM
> >>> > > > > > > > > Guozhang
> >>> > > > > > > > > > > >> Wang <
> >>> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> >>> feedback.
> >>> > > > Here
> >>> > > > > > are
> >>> > > > > > > > > some
> >>> > > > > > > > > > > >> replies
> >>> > > > > > > > > > > >> > > > > inlined
> >>> > > > > > > > > > > >> > > > > >> > > below:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> >>> 2022 at
> >>> > > > 2:53 AM
> >>> > > > > > > > > Ismael Juma
> >>> > > > > > > > > > > >> <
> >>> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
> >>> KIP. This
> >>> > > > has
> >>> > > > > > the
> >>> > > > > > > > > > > >> potential
> >>> > > > > > > > > > > >> > to be a
> >>> > > > > > > > > > > >> > > > > >> > great
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> >>> > > > questions/comments:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> >>> > > > premature to
> >>> > > > > > talk
> >>> > > > > > > > > about
> >>> > > > > > > > > > > >> > target
> >>> > > > > > > > > > > >> > > > > >> versions
> >>> > > > > > > > > > > >> > > > > >> > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> >>> existing
> >>> > > > group
> >>> > > > > > > > > protocol.
> >>> > > > > > > > > > > >> > Unlike
> >>> > > > > > > > > > > >> > > > > >> KRaft,
> >>> > > > > > > > > > > >> > > > > >> > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
> >>> and hence
> >>> > > > > > > > > > > >> > deprecation/removal will
> >>> > > > > > > > > > > >> > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> > > heavily
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> >>> applications
> >>> > > > > > migrate
> >>> > > > > > > > > to the
> >>> > > > > > > > > > > >> new
> >>> > > > > > > > > > > >> > > > > >> protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
> >>> you. I
> >>> > > > think
> >>> > > > > > we can
> >>> > > > > > > > > > > >> remove
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> proposed
> >>> > > > > > > > > > > >> > > > > >> > > > > timeline
> >>> > > > > > > > > > > >> > > > > >> > > > > > > in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> >>> > > > Deprecation,
> >>> > > > > > and
> >>> > > > > > > > > Migration
> >>> > > > > > > > > > > >> > Plan` and
> >>> > > > > > > > > > > >> > > > > >> > > instead
> >>> > > > > > > > > > > >> > > > > >> > > > > just
> >>> > > > > > > > > > > >> > > > > >> > > > > > > state
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> >>> decide in the
> >>> > > > > > future
> >>> > > > > > > > > about
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > we
> >>> > > > > > > > > > > >> > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> >>> behaviors.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says
> >>> we
> >>> > > > intend to
> >>> > > > > > > > > release this
> >>> > > > > > > > > > > >> > in 4.x,
> >>> > > > > > > > > > > >> > > > > >> but
> >>> > > > > > > > > > > >> > > > > >> > it
> >>> > > > > > > > > > > >> > > > > >> > > > > wasn't
> >>> > > > > > > > > > > >> > > > > >> > > > > > > made
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
> >>> that as
> >>> > > > a
> >>> > > > > > way to
> >>> > > > > > > > > > > >> estimate
> >>> > > > > > > > > > > >> > when
> >>> > > > > > > > > > > >> > > > > we'd
> >>> > > > > > > > > > > >> > > > > >> > > > > deprecate
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> >>> protocol, I
> >>> > > > also
> >>> > > > > > suggest
> >>> > > > > > > > > > > >> removing
> >>> > > > > > > > > > > >> > this
> >>> > > > > > > > > > > >> > > > > >> part.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> >>> > > > specifically
> >>> > > > > > > > > related to
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> >>> more for
> >>> > > > client
> >>> > > > > > > > > upgrades.
> >>> > > > > > > > > > > >> I.e.
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > broker-side
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may
> >>> be done
> >>> > > > > > first,
> >>> > > > > > > > > and then
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > client
> >>> > > > > > > > > > > >> > > > > >> > side,
> >>> > > > > > > > > > > >> > > > > >> > > > > and we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> >>> "released"
> >>> > > > by
> >>> > > > > > the
> >>> > > > > > > > > time
> >>> > > > > > > > > > > >> clients
> >>> > > > > > > > > > > >> > > > > >> > > implementations
> >>> > > > > > > > > > > >> > > > > >> > > > > are
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> >>> enable the
> >>> > > > > > feature the
> >>> > > > > > > > > > > >> clients
> >>> > > > > > > > > > > >> > need to
> >>> > > > > > > > > > > >> > > > > >> > first
> >>> > > > > > > > > > > >> > > > > >> > > > > swap-in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> >>> rolling
> >>> > > > bounce
> >>> > > > > > and
> >>> > > > > > > > > then set
> >>> > > > > > > > > > > >> > the flag
> >>> > > > > > > > > > > >> > > > > >> > with a
> >>> > > > > > > > > > > >> > > > > >> > > > > second
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence
> >>> we feel
> >>> > > > it's
> >>> > > > > > > > > better to be
> >>> > > > > > > > > > > >> > released
> >>> > > > > > > > > > > >> > > > > >> in a
> >>> > > > > > > > > > > >> > > > > >> > > major
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> >>> flesh out
> >>> > > > the
> >>> > > > > > > > > details of
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > migration
> >>> > > > > > > > > > > >> > > > > >> > > story.
> >>> > > > > > > > > > > >> > > > > >> > > > > It
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
> >>> will
> >>> > > > support
> >>> > > > > > online
> >>> > > > > > > > > > > >> > migrations. Is
> >>> > > > > > > > > > > >> > > > > >> that
> >>> > > > > > > > > > > >> > > > > >> > > > > correct?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > We
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> >>> detail.
> >>> > > > It
> >>> > > > > > could
> >>> > > > > > > > > also be
> >>> > > > > > > > > > > >> > done as a
> >>> > > > > > > > > > > >> > > > > >> > > separate
> >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's
> >>> the
> >>> > > > part we
> >>> > > > > > can
> >>> > > > > > > > > be more
> >>> > > > > > > > > > > >> > concrete
> >>> > > > > > > > > > > >> > > > > >> about
> >>> > > > > > > > > > > >> > > > > >> > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > sure
> >>> > > > > > > > > > > >> > > > > >> > > > > > > (and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to
> >>> your
> >>> > > > > > question 2)
> >>> > > > > > > > > above).
> >>> > > > > > > > > > > >> > We will
> >>> > > > > > > > > > > >> > > > > >> work
> >>> > > > > > > > > > > >> > > > > >> > on
> >>> > > > > > > > > > > >> > > > > >> > > > > making
> >>> > > > > > > > > > > >> > > > > >> > > > > > > it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> >>> parallel as we
> >>> > > > > > solicit
> >>> > > > > > > > > more
> >>> > > > > > > > > > > >> > feedback.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
> >>> that we are
> >>> > > > > > pushing
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > pattern
> >>> > > > > > > > > > > >> > > > > >> > > subscriptions
> >>> > > > > > > > > > > >> > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> >>> like there
> >>> > > > > > could be
> >>> > > > > > > > > some
> >>> > > > > > > > > > > >> tricky
> >>> > > > > > > > > > > >> > > > > >> > > compatibility
> >>> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism
> >>> for
> >>> > > > users to
> >>> > > > > > > > > detect that
> >>> > > > > > > > > > > >> > they
> >>> > > > > > > > > > > >> > > > > need
> >>> > > > > > > > > > > >> > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> > > update
> >>> > > > > > > > > > > >> > > > > >> > > > > > > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching
> >>> to the
> >>> > > > new
> >>> > > > > > > > > protocol?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> >>> need some
> >>> > > > > > tooling for
> >>> > > > > > > > > > > >> non-java
> >>> > > > > > > > > > > >> > client
> >>> > > > > > > > > > > >> > > > > >> > users
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > sort
> >>> > > > > > > > > > > >> > > > > >> > > > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> >>> client before
> >>> > > > > > > > > switching to
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > new
> >>> > > > > > > > > > > >> > > > > >> > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > I
> >>> > > > > > > > > > > >> > > > > >> > > > > do not
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on
> >>> top of my
> >>> > > > head
> >>> > > > > > > > > though,
> >>> > > > > > > > > > > >> maybe
> >>> > > > > > > > > > > >> > others
> >>> > > > > > > > > > > >> > > > > >> like
> >>> > > > > > > > > > > >> > > > > >> > > @Matt
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to
> >>> the last
> >>> > > > > > question,
> >>> > > > > > > > > will the
> >>> > > > > > > > > > > >> > Java
> >>> > > > > > > > > > > >> > > > > >> client
> >>> > > > > > > > > > > >> > > > > >> > > allow
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> >>> current
> >>> > > > regex
> >>> > > > > > > > > engine for
> >>> > > > > > > > > > > >> > > > > >> compatibility
> >>> > > > > > > > > > > >> > > > > >> > > > > reasons?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > For
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy
> >>> to keep
> >>> > > > using
> >>> > > > > > > > > client
> >>> > > > > > > > > > > >> based
> >>> > > > > > > > > > > >> > regex
> >>> > > > > > > > > > > >> > > > > at
> >>> > > > > > > > > > > >> > > > > >> > > first to
> >>> > > > > > > > > > > >> > > > > >> > > > > keep
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> >>> simple and
> >>> > > > then
> >>> > > > > > migrate
> >>> > > > > > > > > to
> >>> > > > > > > > > > > >> > server based
> >>> > > > > > > > > > > >> > > > > >> > > regexes
> >>> > > > > > > > > > > >> > > > > >> > > > > as a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have
> >>> not thought
> >>> > > > > > about
> >>> > > > > > > > > that for
> >>> > > > > > > > > > > >> > java
> >>> > > > > > > > > > > >> > > > > >> clients,
> >>> > > > > > > > > > > >> > > > > >> > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > we can
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> >>> > > > > > compatibility
> >>> > > > > > > > > issues do
> >>> > > > > > > > > > > >> > you have
> >>> > > > > > > > > > > >> > > > > in
> >>> > > > > > > > > > > >> > > > > >> > > mind?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
> >>> that the
> >>> > > > group
> >>> > > > > > > > > coordinator
> >>> > > > > > > > > > > >> > will be
> >>> > > > > > > > > > > >> > > > > >> > > > > responsible for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> >>> configurations and
> >>> > > > that
> >>> > > > > > the
> >>> > > > > > > > > > > >> > configurations
> >>> > > > > > > > > > > >> > > > > will
> >>> > > > > > > > > > > >> > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> > > > > deleted
> >>> > > > > > > > > > > >> > > > > >> > > > > > > when
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> >>> deleted. Will a
> >>> > > > > > > > > transition to
> >>> > > > > > > > > > > >> DEAD
> >>> > > > > > > > > > > >> > > > > trigger
> >>> > > > > > > > > > > >> > > > > >> > > deletion
> >>> > > > > > > > > > > >> > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> >>> DEAD state
> >>> > > > is an
> >>> > > > > > > > > ending
> >>> > > > > > > > > > > >> state
> >>> > > > > > > > > > > >> > (we
> >>> > > > > > > > > > > >> > > > > would
> >>> > > > > > > > > > > >> > > > > >> > only
> >>> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> >>> group is
> >>> > > > EMPTY
> >>> > > > > > and
> >>> > > > > > > > > also all
> >>> > > > > > > > > > > >> of
> >>> > > > > > > > > > > >> > its
> >>> > > > > > > > > > > >> > > > > >> > metadata
> >>> > > > > > > > > > > >> > > > > >> > > are
> >>> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
> >>> DEAD this
> >>> > > > > > group
> >>> > > > > > > > > would
> >>> > > > > > > > > > > >> never
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > > >> revived.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> >>> choice to
> >>> > > > store
> >>> > > > > > the
> >>> > > > > > > > > configs
> >>> > > > > > > > > > > >> in
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > coordinator
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list
> >>> all
> >>> > > > cluster
> >>> > > > > > configs
> >>> > > > > > > > > and
> >>> > > > > > > > > > > >> their
> >>> > > > > > > > > > > >> > > > > values?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> >>> question,
> >>> > > > and our
> >>> > > > > > > > > thoughts
> >>> > > > > > > > > > > >> are
> >>> > > > > > > > > > > >> > that
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > so-called
> >>> > > > > > > > > > > >> > > > > >> > > > > > > "group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> >>> > > > overrides
> >>> > > > > > of the
> >>> > > > > > > > > > > >> > cluster-level
> >>> > > > > > > > > > > >> > > > > >> > > > > configurations
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> >>> group so
> >>> > > > when an
> >>> > > > > > > > > admin list
> >>> > > > > > > > > > > >> > cluster
> >>> > > > > > > > > > > >> > > > > >> > configs
> >>> > > > > > > > > > > >> > > > > >> > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> >>> cluster-level
> >>> > > > > > > > > "defaults", not
> >>> > > > > > > > > > > >> > showing
> >>> > > > > > > > > > > >> > > > > >> any
> >>> > > > > > > > > > > >> > > > > >> > > > > per-group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> >>> someone
> >>> > > > > > configure a
> >>> > > > > > > > > group
> >>> > > > > > > > > > > >> > before
> >>> > > > > > > > > > > >> > > > > >> starting
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> >>> allowing
> >>> > > > the
> >>> > > > > > explicit
> >>> > > > > > > > > > > >> > creation of
> >>> > > > > > > > > > > >> > > > > >> groups?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> >>> > > > decoupled
> >>> > > > > > from
> >>> > > > > > > > > the group
> >>> > > > > > > > > > > >> > > > > lifecycle.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
> >>> created
> >>> > > > > > before
> >>> > > > > > > > > the group
> >>> > > > > > > > > > > >> > itself
> >>> > > > > > > > > > > >> > > > > as
> >>> > > > > > > > > > > >> > > > > >> an
> >>> > > > > > > > > > > >> > > > > >> > > > > > > independent
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> >>> > > > requires
> >>> > > > > > the
> >>> > > > > > > > > > > >> corresponding
> >>> > > > > > > > > > > >> > > > > >> request to
> >>> > > > > > > > > > > >> > > > > >> > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> >>> based on
> >>> > > > the
> >>> > > > > > group
> >>> > > > > > > > > id ---
> >>> > > > > > > > > > > >> > the only
> >>> > > > > > > > > > > >> > > > > >> > thing
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> >>> itself is
> >>> > > > gone
> >>> > > > > > we
> >>> > > > > > > > > also check
> >>> > > > > > > > > > > >> > if there
> >>> > > > > > > > > > > >> > > > > >> are
> >>> > > > > > > > > > > >> > > > > >> > > any
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related
> >>> to that
> >>> > > > group
> >>> > > > > > and
> >>> > > > > > > > > delete
> >>> > > > > > > > > > > >> as
> >>> > > > > > > > > > > >> > well.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> >>> indeed
> >>> > > > > > introduces an
> >>> > > > > > > > > > > >> > asymmetry on
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > creation
> >>> > > > > > > > > > > >> > > > > >> > > > > /
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> >>> config
> >>> > > > > > entities,
> >>> > > > > > > > > and we
> >>> > > > > > > > > > > >> > would like
> >>> > > > > > > > > > > >> > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> > > hear
> >>> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether
> >>> we should
> >>> > > > aim
> >>> > > > > > for
> >>> > > > > > > > > symmetry
> >>> > > > > > > > > > > >> > i.e.
> >>> > > > > > > > > > > >> > > > > >> totally
> >>> > > > > > > > > > > >> > > > > >> > > > > decouple
> >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence
> >>> not
> >>> > > > delete
> >>> > > > > > them at
> >>> > > > > > > > > all
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> group
> >>> > > > > > > > > > > >> > > > > >> > is
> >>> > > > > > > > > > > >> > > > > >> > > > > gone,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > but
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> >>> deletion
> >>> > > > > > operations
> >>> > > > > > > > > by
> >>> > > > > > > > > > > >> > themselves.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> >>> > > > > > Consumer.subscribe
> >>> > > > > > > > > method for
> >>> > > > > > > > > > > >> > the Java
> >>> > > > > > > > > > > >> > > > > >> > client
> >>> > > > > > > > > > > >> > > > > >> > > > > still
> >>> > > > > > > > > > > >> > > > > >> > > > > > > take
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > `java.util.regex.Pattern` of
> >>> > > > > > do we
> >>> > > > > > > > > have
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> introduce an
> >>> > > > > > > > > > > >> > > > > >> > > > > overload?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
> >>> need to
> >>> > > > > > introduce
> >>> > > > > > > > > an
> >>> > > > > > > > > > > >> > overload, but
> >>> > > > > > > > > > > >> > > > > >> I'm
> >>> > > > > > > > > > > >> > > > > >> > > all
> >>> > > > > > > > > > > >> > > > > >> > > > > ears
> >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> >>> compatibility
> >>> > > > > > issues
> >>> > > > > > > > > that we
> >>> > > > > > > > > > > >> may
> >>> > > > > > > > > > > >> > > > > overlook.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
> >>> Justine
> >>> > > > > > that we
> >>> > > > > > > > > should
> >>> > > > > > > > > > > >> be
> >>> > > > > > > > > > > >> > clearer
> >>> > > > > > > > > > > >> > > > > >> > about
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > reason
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> >>> > > > > > IBP/metadata.version
> >>> > > > > > > > > from the
> >>> > > > > > > > > > > >> > feature
> >>> > > > > > > > > > > >> > > > > >> flag.
> >>> > > > > > > > > > > >> > > > > >> > > Maybe
> >>> > > > > > > > > > > >> > > > > >> > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > mean
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> >>> default
> >>> > > > for
> >>> > > > > > the
> >>> > > > > > > > > feature
> >>> > > > > > > > > > > >> flag
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > true
> >>> > > > > > > > > > > >> > > > > >> > > based on
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
> >>> once we
> >>> > > > > > want to
> >>> > > > > > > > > make it
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> default.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> >>> protocol
> >>> > > > APIs
> >>> > > > > > don't
> >>> > > > > > > > > mention
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> required
> >>> > > > > > > > > > > >> > > > > >> > > ACLs,
> >>> > > > > > > > > > > >> > > > > >> > > > > it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that
> >>> for
> >>> > > > > > consistency.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> >>> certainly do
> >>> > > > that.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
> >>> odd that
> >>> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> >>> > > > > > > > > > > >> > > > > >> > requires
> >>> > > > > > > > > > > >> > > > > >> > > > > "Read
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems
> >>> to do
> >>> > > > more
> >>> > > > > > than
> >>> > > > > > > > > reading.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
> >>> myself
> >>> > > > as
> >>> > > > > > well,
> >>> > > > > > > > > but in
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > end we
> >>> > > > > > > > > > > >> > > > > >> > could
> >>> > > > > > > > > > > >> > > > > >> > > not
> >>> > > > > > > > > > > >> > > > > >> > > > > > > find a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> >>> alternative: adding
> >>> > > > > > Write
> >>> > > > > > > > > Group
> >>> > > > > > > > > > > >> seems
> >>> > > > > > > > > > > >> > an
> >>> > > > > > > > > > > >> > > > > >> overkill
> >>> > > > > > > > > > > >> > > > > >> > > here
> >>> > > > > > > > > > > >> > > > > >> > > > > > > since we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> >>> elsewhere (we
> >>> > > > only
> >>> > > > > > have
> >>> > > > > > > > > Read /
> >>> > > > > > > > > > > >> > Delete
> >>> > > > > > > > > > > >> > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > Describe
> >>> > > > > > > > > > > >> > > > > >> > > > > on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like
> >>> to hear
> >>> > > > others
> >>> > > > > > > > > thoughts.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> >>> > > > recreation
> >>> > > > > > > > > handled by
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > with
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> >>> would be
> >>> > > > good
> >>> > > > > > to have
> >>> > > > > > > > > a
> >>> > > > > > > > > > > >> > section on
> >>> > > > > > > > > > > >> > > > > >> this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> >>> > > > > > subscription
> >>> > > > > > > > > right? Yes
> >>> > > > > > > > > > > >> > we can
> >>> > > > > > > > > > > >> > > > > >> add a
> >>> > > > > > > > > > > >> > > > > >> > > > > section
> >>> > > > > > > > > > > >> > > > > >> > > > > > > about
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> >>> basically the
> >>> > > > idea
> >>> > > > > > is that
> >>> > > > > > > > > > > >> consumer
> >>> > > > > > > > > > > >> > would
> >>> > > > > > > > > > > >> > > > > be
> >>> > > > > > > > > > > >> > > > > >> > > totally
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol
> >>> as it's
> >>> > > > > > handled all
> >>> > > > > > > > > by the
> >>> > > > > > > > > > > >> > brokers.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> >>> mentions we
> >>> > > > will
> >>> > > > > > write
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> new
> >>> > > > > > > > > > > >> > > > > >> coordinator
> >>> > > > > > > > > > > >> > > > > >> > in
> >>> > > > > > > > > > > >> > > > > >> > > > > Java.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> >>> implementation
> >>> > > > > > detail,
> >>> > > > > > > > > do we
> >>> > > > > > > > > > > >> > plan to
> >>> > > > > > > > > > > >> > > > > >> have a
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > gradle
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> >>> thought about
> >>> > > > > > that. But
> >>> > > > > > > > > I think
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > answer
> >>> > > > > > > > > > > >> > > > > >> > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> >>> > > > scalability
> >>> > > > > > goal
> >>> > > > > > > > > when it
> >>> > > > > > > > > > > >> > comes to
> >>> > > > > > > > > > > >> > > > > >> how
> >>> > > > > > > > > > > >> > > > > >> > > many
> >>> > > > > > > > > > > >> > > > > >> > > > > > > members
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol
> >>> can
> >>> > > > support?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> >>> think we
> >>> > > > > > should
> >>> > > > > > > > > shoot for
> >>> > > > > > > > > > > >> > 1000s of
> >>> > > > > > > > > > > >> > > > > >> > > members.
> >>> > > > > > > > > > > >> > > > > >> > > > > But
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
> >>> also
> >>> > > > depend
> >>> > > > > > on the
> >>> > > > > > > > > offset
> >>> > > > > > > > > > > >> > > > > management
> >>> > > > > > > > > > > >> > > > > >> > > (commit,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> >>> > > > coordinator
> >>> > > > > > > > > which we
> >>> > > > > > > > > > > >> did
> >>> > > > > > > > > > > >> > not
> >>> > > > > > > > > > > >> > > > > >> cover in
> >>> > > > > > > > > > > >> > > > > >> > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> >>> number that
> >>> > > > > > applies
> >>> > > > > > > > > > > >> > universally.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> >>> consider
> >>> > > > having
> >>> > > > > > > > > > > >> > SubscribedTopidIds
> >>> > > > > > > > > > > >> > > > > >> instead
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> SubscribedTopicNames in
> >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> >>> > > > > > > > > > > >> > > > > >> > > Is the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > idea
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
> >>> resolve
> >>> > > > the
> >>> > > > > > regex
> >>> > > > > > > > > on the
> >>> > > > > > > > > > > >> > server,
> >>> > > > > > > > > > > >> > > > > we
> >>> > > > > > > > > > > >> > > > > >> > can
> >>> > > > > > > > > > > >> > > > > >> > > do
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name?
> >>> The
> >>> > > > > > difference is
> >>> > > > > > > > > that
> >>> > > > > > > > > > > >> > sending the
> >>> > > > > > > > > > > >> > > > > >> > regex
> >>> > > > > > > > > > > >> > > > > >> > > is
> >>> > > > > > > > > > > >> > > > > >> > > > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
> >>> the topic
> >>> > > > > > names is
> >>> > > > > > > > > less
> >>> > > > > > > > > > > >> > efficient.
> >>> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> >>> easier to
> >>> > > > > > handle if
> >>> > > > > > > > > we have
> >>> > > > > > > > > > > >> > topic
> >>> > > > > > > > > > > >> > > > > ids.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
> >>> still
> >>> > > > let
> >>> > > > > > the
> >>> > > > > > > > > clients
> >>> > > > > > > > > > > >> send
> >>> > > > > > > > > > > >> > names
> >>> > > > > > > > > > > >> > > > > >> is to
> >>> > > > > > > > > > > >> > > > > >> > > keep
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names
> >>> -> ids
> >>> > > > on
> >>> > > > > > the
> >>> > > > > > > > > broker /
> >>> > > > > > > > > > > >> > admin
> >>> > > > > > > > > > > >> > > > > client
> >>> > > > > > > > > > > >> > > > > >> > > only.
> >>> > > > > > > > > > > >> > > > > >> > > > > Note
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> >>> topic id
> >>> > > > in
> >>> > > > > > > > > KIP-516, we
> >>> > > > > > > > > > > >> > never
> >>> > > > > > > > > > > >> > > > > >> > > implemented the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > logic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> >>> > > > leveraging
> >>> > > > > > the
> >>> > > > > > > > > related
> >>> > > > > > > > > > > >> > newer
> >>> > > > > > > > > > > >> > > > > >> versioned
> >>> > > > > > > > > > > >> > > > > >> > > RPCs,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic
> >>> id as
> >>> > > > empty
> >>> > > > > > UUID.
> >>> > > > > > > > > We
> >>> > > > > > > > > > > >> want
> >>> > > > > > > > > > > >> > to keep
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> >>> > > > delegate
> >>> > > > > > the
> >>> > > > > > > > > reasoning
> >>> > > > > > > > > > > >> on
> >>> > > > > > > > > > > >> > broker
> >>> > > > > > > > > > > >> > > > > >> and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
> >>> 2022 at
> >>> > > > 10:45
> >>> > > > > > AM
> >>> > > > > > > > > David
> >>> > > > > > > > > > > >> Jacot
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> <dj...@confluent.io.invalid>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like
> >>> to start a
> >>> > > > > > > > > discussion
> >>> > > > > > > > > > > >> thread
> >>> > > > > > > > > > > >> > on
> >>> > > > > > > > > > > >> > > > > >> KIP-848:
> >>> > > > > > > > > > > >> > > > > >> > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > Next
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of
> >>> the
> >>> > > > Consumer
> >>> > > > > > > > > Rebalance
> >>> > > > > > > > > > > >> > Protocol.
> >>> > > > > > > > > > > >> > > > > >> With
> >>> > > > > > > > > > > >> > > > > >> > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> >>> rebalance
> >>> > > > > > protocol
> >>> > > > > > > > > (for
> >>> > > > > > > > > > > >> > consumers)
> >>> > > > > > > > > > > >> > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> > > > > reliable,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> >>> easier to
> >>> > > > > > implement for
> >>> > > > > > > > > > > >> > clients, and
> >>> > > > > > > > > > > >> > > > > >> easier
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > debug
> >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> >>> here:
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
> >>> look and
> >>> > > > let
> >>> > > > > > me
> >>> > > > > > > > > know what
> >>> > > > > > > > > > > >> > you
> >>> > > > > > > > > > > >> > > > > think.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
> >>> away from
> >>> > > > > > July
> >>> > > > > > > > > 18th to
> >>> > > > > > > > > > > >> > August 8th.
> >>> > > > > > > > > > > >> > > > > >> That
> >>> > > > > > > > > > > >> > > > > >> > > gives
> >>> > > > > > > > > > > >> > > > > >> > > > > > > you a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> >>> read and
> >>> > > > digest
> >>> > > > > > this
> >>> > > > > > > > > long
> >>> > > > > > > > > > > >> KIP.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > --
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> --
> >>> > > > > > > > > > > >> > > > > >> -- Guozhang
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > > >
> >>> > > > >
> >>> > > > > --
> >>> > > > > -- Guozhang
> >>> > > >
> >>> > >
> >>> > >
> >>> > > --
> >>> > > -- Guozhang
> >>>
> >>
> >>
> >> --
> >> -- Guozhang
> >>
> >
> >
> > --
> > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Jason,

Thanks for your comments.

1. That's an interesting point of view. In my mind, I had the
canonical case where a consumer leaves and then rejoins the group with
the same instance id. In this case, it makes sense to replace the
former member with the new one. This means that the former member is
fenced from this point on. Do you think that we could have a case
where two members with the same static id are active? That seems a bit
counter intuitive to me. Assuming that we would not fence the former
member immediately, one issue would be that we would not know when the
handoff must take place. I guess that we would have to wait until the
former member reaches its session timeout or something like that.

2. Yeah, those configurations are not used by the new protocol. I will
deprecate them in the KIP. Note that the rebalance timeout (max poll
interval) remains on the client though.

3. That's an interesting idea. We could have a regex type for instance
but that would fundamentally be a string, right? One benefit of this
approach is that we could validate the regex at the protocol level.
The downside is that it would require the RE2 library as a dependency
of the common module to achieve this. My plan was to simply pass
through the regex to the server and to let the server do all the
validation. With this, we don't need the client to depend on the RE2
library and that is a good thing in my opinion.

4. That's a good point. A no-op in this case also makes sense. We
could also add a warning to notify the user. I will update the KIP.

5. That would also work. I think that we need a configuration for
Streams anyway because we cannot infer it in that case. This is why I
went with the same configuration for both consumer and streams. I do
agree that we could remove it on the consumer side though. I don't
feel strong either way. What do you think?

Thanks,
David

On Wed, Sep 21, 2022 at 2:28 AM Jason Gustafson
<ja...@confluent.io.invalid> wrote:
>
> Hi David,
>
> Thanks for all the hard work on this.  I'm steadily making my way through
> the doc and had a few initial questions:
>
> 1. The KIP mentions the following in regard to static membership:
>
> > Static membership, introduced in KIP-345, is still supported by this new
> rebalance protocol. When a member rejoins with the same Instance ID, the
> group coordinator replaces the old member with the new member. The new
> member can continue from where it left off.
>
> Can you clarify this a little bit? Suppose that the current member is still
> active and a new one arrives with the same instance ID. Do we fence the
> current member immediately or do we attempt to transition gracefully?
>
> 2. I wasn't very clear about the plan for existing client configurations
> which overlap with the new broker configurations. For example,
> heartbeat.interval.ms. Would the setting on the server override this? Do we
> plan to deprecate the configuration?
>
> 3. The KIP adds support for regex subscription through a separate field in
> the heartbeat request. Perhaps it would make sense to make this a field
> type in the RPC schema definitions. I can imagine reusing this in the
> Metadata request API for example. By the way, does this introduce a
> dependence on the RE2 library? Or will the client simply pass through the
> string that is provided by the user?
>
> 4. In regard to the deprecation of `enforceRebalance`, I wonder if it would
> be better to make it a no-op instead of letting it raise an exception. Just
> a little concerned that might make upgrading more difficult.
>
> 5. I'm trying to decide if `group.protocol` is pulling enough weight to
> exist. Could the behavior be inferred from other configurations? For
> example, perhaps by default, group.remote.assignor should be empty, which
> implies the "generic" protocol. If a user provides a non-empty value, then
> we change to "consumer." That would give the opt-in behavior that it looks
> like you want.
>
> Thanks,
> Jason
>
>
>
>
>
> On Mon, Sep 19, 2022 at 5:33 PM Jun Rao <ju...@confluent.io.invalid> wrote:
>
> > Hi, David,
> >
> > Thanks for the reply. A few more comments.
> >
> > 40. Is there an API to change PartitionAssignor#metadata? I assume the
> > member fields are just for reads?
> >
> > 43.2 Sounds good.
> >
> > 44.1 Thanks. Could we document that the coordinator selects the assignor
> > when multiple assignors are specified in the client?
> >
> > 44.2 Encoding the group state in ConsumerGroupPrepareAssignmentResponse
> > sounds simpler and clearer to me.
> >
> > 48. Yes, that's the confusing part. I assume member assignment is what the
> > member client has. Could we include member assignment from both the
> > coordinator and the client's point of view?
> >
> > 50. We only include Uuid in PartitionAssignor.Group.TopicMetadata. How does
> > KStream do co-partitioning across topics?
> >
> > 51. PartitionAssignor.metadata: Is this really an assignor level thing? The
> > use case mentioned is to use metadata for encoding lags in standby tasks.
> > However, the lag is independent of the assignor. For example, if a member
> > sets two assignors, does it need to set the same metadata in both
> > assignors?
> >
> > 52. "power users will have the ability to trigger a reassignment by either
> > providing a non-zero reason or by updating the assignor metadata." : Do we
> > need both reason and metadata to trigger an assignment? Could we just use
> > metadata?
> >
> > 53. MemberAssignment.metadata: How is that used? Does that overwrite the
> > member metadata on the coordinator?
> >
> > Thanks,
> >
> > Jun
> >
> >
> > On Tue, Sep 13, 2022 at 1:59 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > > Hi David,
> > >
> > > Another quick thing I just realized about
> > "PartitionAssignor#onAssignment":
> > > in the new proposal, this function could be triggered multiple time
> > within
> > > a group generation as it will be triggered whenever the broker gives the
> > > client new assignments, whereas today it is only triggered once per
> > > rebalance. I think it's worth emphasizing this a bit in the doc,
> > especially
> > > around:
> > >
> > > 1) how many times `onAssignment` could be triggered with a single
> > rebalance
> > > (especially if the new assignment is decided by the client-assignor).
> > > 2) when the Assignment metadata could change across multiple
> > `onAssignment`
> > > calls, and how users should handle it.
> > > 3) the order of calling `onAssignment` and
> > `onPartitionsRevoked/Assigned`.
> > >
> > > For example, today the Streams client relies on the following
> > assumptions:
> > > 1) within a rebalance the onPartitionsRevoked would be called before the
> > > onAssignment, and onAssignment is called before onPartitionsAssigned
> > (today
> > > we would trigger all three only once within a generation), 2)
> > onAssignment
> > > contains the source-of-truth of the final assignment, and hence we can
> > > simply ignore onPartitionsAssigned and just rely on onAssignment to
> > reason
> > > about which tasks are removed / assigned. And each time it calls, the
> > > corresponding metadata change are "final" for this new generation.
> > >
> > > We can change however Streams handle the new protocol, but I think there
> > > are other clients externally which may also make such assumptions about
> > the
> > > ordering / call times about those callbacks, and hence worth clarifying
> > > those in the new proposal.
> > >
> > >
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > > On Tue, Sep 13, 2022 at 12:20 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > > > Hello David,
> > > >
> > > > Just had a few more nit questions about the public APIs:
> > > >
> > > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > > field,
> > > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > > rationale seems to be assuming that we should always be able to do the
> > > > assignment at the broker-side assignor without errors. Personally I
> > think
> > > > it's still potentially beneficial to add the Error field even for
> > > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > > topics are not recognized with the current broker's metadata. What do
> > you
> > > > think?
> > > >
> > > > 2) The client-side "GroupMember" has three additional fields
> > > > reason/version/metadata compared with the broker-side GroupMember. I
> > > agree
> > > > that broker-side assignor would not need reason/metadata since they are
> > > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > > it
> > > > possible that we evolve our broker-side built-in assignor but the old
> > > > versioned clients would not be able to work with the new version, in
> > > which
> > > > case we need to let the broker being aware of this and upgrade its
> > > behavior
> > > > to cooperate with the clients?
> > > >
> > > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > > of
> > > > including these three fields, what about just adding the "Metadata"
> > field
> > > > class which has these three fields? Also, there are two "Metadata"
> > > > currently in the APIs, the first is a class that encodes
> > > > reason/version/metadata, and the second is just the encoded metadata
> > > bytes.
> > > > I'm wondering what about just naming the first as memberMetadata, which
> > > > then has a bytebuffer field Metadata, or instead naming the second
> > > > bytebuffer field as metadataBytes?
> > > >
> > > > Guozhang
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > >
> > > >> Hello David,
> > > >>
> > > >> Thanks for bringing this question up. I think the main benefits as you
> > > >> listed is 2) above if it stays; just to clarify, we would only be able
> > > to
> > > >> save one round trip if the rebalance is still triggered by the broker;
> > > if
> > > >> the rebalance is triggered by the client then the num.round.trips are
> > > the
> > > >> same:
> > > >>
> > > >> 1) With GroupPrepareAssignment:
> > > >>
> > > >> T0: client decides to do a new assignment, suppose it has already
> > sent a
> > > >> HB and hence has to wait for it to return first since only one
> > request /
> > > >> response can be inflight with the coordinator's socket.
> > > >> T1: client receives the HB response, and then sends the
> > > >> GroupPrepareAssignment request.
> > > >> T2: the GroupPrepareAssignment response is returned.
> > > >> T3: it calculates the new assignment, and sends a
> > GroupInstallAssignment
> > > >> request.
> > > >>
> > > >> In total, two round trips.
> > > >>
> > > >> 2) Without GroupPrepareAssignment:
> > > >>
> > > >> T0: client decides to do a new assignment, suppose it has already
> > sent a
> > > >> HB and hence has to wait for it to return first since only one
> > request /
> > > >> response can be inflight with the coordinator's socket.
> > > >> T1: client receives the HB response, and then sends the new HB request
> > > >> with the flag indicating a new rebalance needed..
> > > >> T2: the HB response with the optional member metadata map is returned.
> > > >> T3: it calculates the new assignment, and sends a
> > GroupInstallAssignment
> > > >> request.
> > > >>
> > > >> In total, two round trips as well.
> > > >>
> > > >> -----------------------------
> > > >>
> > > >> So to complete the full picture here, we'd need to modify both HB
> > > request
> > > >> and response so that the client can also indicate a new rebalance via
> > > the
> > > >> HB request as well, right?
> > > >>
> > > >> Assuming all above is true, I think it's okay to merge the
> > > >> GroupPrepareAssignment into HB given that we can make the additional
> > > fields
> > > >> encoding the full member (subscription) metadata and topic metadata as
> > > >> optional fields.
> > > >>
> > > >>
> > > >> Guozhang
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >> On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> > <djacot@confluent.io.invalid
> > > >
> > > >> wrote:
> > > >>
> > > >>> Hi all,
> > > >>>
> > > >>> During an offline conversation, someone asked why we need the
> > > >>> ConsumerGroupPrepareAssignment API and suggested that we could
> > instead
> > > >>> provide the group state in the heartbeat response. This has a few
> > > >>> advantages: 1) it does not require using a special error code to
> > > >>> signal that a new assignment is required as the signal would be the
> > > >>> provided group state; 2) it removes one round trip when a client side
> > > >>> assignor is used. The downside is that it makes the heartbeat
> > > >>> response's definition quite large. I recall that I went with the
> > > >>> current approach due to this.
> > > >>>
> > > >>> Providing the group state in the heartbeat response is appealing.
> > What
> > > >>> do you guys think?
> > > >>>
> > > >>> Best,
> > > >>> David
> > > >>>
> > > >>> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > >>> >
> > > >>> > Hi Guozhang,
> > > >>> >
> > > >>> > 1. I have added a reference to the relevant chapter instead of
> > > >>> > repeating the whole thing. Does that work for you?
> > > >>> >
> > > >>> > 2. The "Rebalance Triggers" section you are referring to is about
> > > when
> > > >>> > a rebalance should be triggered for the non-upgraded members using
> > > the
> > > >>> > old protocol. The section mentions that a rebalance must be
> > triggered
> > > >>> > when a new assignment is installed. This implies that the group
> > epoch
> > > >>> > was updated either by a native member or a non-upgraded member. For
> > > >>> > the latter, the JoinGroup request would be the trigger. I have
> > added
> > > a
> > > >>> > reference to the relevant chapter in the "JoinGroup Handling"
> > section
> > > >>> > as well. Does that make sense?
> > > >>> >
> > > >>> > Thanks,
> > > >>> > David
> > > >>> >
> > > >>> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> > > >>> wrote:
> > > >>> > >
> > > >>> > > Hello David,
> > > >>> > >
> > > >>> > > Alright I think that's sufficient. Just to make that clear in the
> > > >>> doc,
> > > >>> > > could we update:
> > > >>> > >
> > > >>> > > 1) the heartbeat request handling section, stating when
> > coordinator
> > > >>> will
> > > >>> > > trigger rebalance based on the HB's member metadata / reason?
> > > >>> > > 2) the "Rebalance Triggers" section to include what we described
> > in
> > > >>> "Group
> > > >>> > > Epoch - Trigger a rebalance" section as well?
> > > >>> > >
> > > >>> > >
> > > >>> > > Guozhang
> > > >>> > >
> > > >>> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > > >>> <dj...@confluent.io.invalid>
> > > >>> > > wrote:
> > > >>> > >
> > > >>> > > > Hi Guozhang,
> > > >>> > > >
> > > >>> > > > I thought that the assignor will always be consulted when the
> > > next
> > > >>> > > > heartbeat request is constructed. In other words,
> > > >>> > > > `PartitionAssignor#metadata` will be called for every
> > heartbeat.
> > > >>> This
> > > >>> > > > gives the opportunity for the assignor to enforce a rebalance
> > by
> > > >>> > > > setting the reason to a non-zero value or by changing the
> > bytes.
> > > Do
> > > >>> > > > you think that this is not sufficient? Are you concerned by the
> > > >>> delay?
> > > >>> > > >
> > > >>> > > > Best,
> > > >>> > > > David
> > > >>> > > >
> > > >>> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> > wangguoz@gmail.com
> > > >
> > > >>> wrote:
> > > >>> > > > >
> > > >>> > > > > Hello David,
> > > >>> > > > >
> > > >>> > > > > One of Jun's comments make me thinking:
> > > >>> > > > >
> > > >>> > > > > ```
> > > >>> > > > > In this case, a new assignment is triggered by the client
> > side
> > > >>> > > > > assignor. When constructing the HB, the consumer will always
> > > >>> consult
> > > >>> > > > > the client side assignor and propagate the information to the
> > > >>> group
> > > >>> > > > > coordinator. In other words, we don't expect users to call
> > > >>> > > > > Consumer#enforceRebalance anymore.
> > > >>> > > > > ```
> > > >>> > > > >
> > > >>> > > > > As I looked at the current PartitionAssignor's interface, we
> > > >>> actually do
> > > >>> > > > > not have a way yet to instruct how to construct the next HB
> > > >>> request, e.g.
> > > >>> > > > > when the assignor wants to enforce a new rebalance with a new
> > > >>> assignment,
> > > >>> > > > > we'd need some customizable APIs inside the PartitionAssignor
> > > to
> > > >>> indicate
> > > >>> > > > > the next HB telling broker about so. WDYT about adding such
> > an
> > > >>> API on the
> > > >>> > > > > PartitionAssignor?
> > > >>> > > > >
> > > >>> > > > >
> > > >>> > > > > Guozhang
> > > >>> > > > >
> > > >>> > > > >
> > > >>> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > > >>> <dj...@confluent.io.invalid>
> > > >>> > > > > wrote:
> > > >>> > > > >
> > > >>> > > > > > Hi Jun,
> > > >>> > > > > >
> > > >>> > > > > > I have updated the KIP to include your feedback. I have
> > also
> > > >>> tried to
> > > >>> > > > > > clarify the parts which were not cleared.
> > > >>> > > > > >
> > > >>> > > > > > Best,
> > > >>> > > > > > David
> > > >>> > > > > >
> > > >>> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > > >>> djacot@confluent.io>
> > > >>> > > > wrote:
> > > >>> > > > > > >
> > > >>> > > > > > > Hi Jun,
> > > >>> > > > > > >
> > > >>> > > > > > > Thanks for your feedback. Let me start by answering your
> > > >>> questions
> > > >>> > > > > > > inline and I will update the KIP next week.
> > > >>> > > > > > >
> > > >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > KIP
> > > >>> seem to
> > > >>> > > > be
> > > >>> > > > > > fewer
> > > >>> > > > > > > > RPCs during rebalance and more efficient support of
> > > >>> wildcard. A few
> > > >>> > > > > > > > comments below.
> > > >>> > > > > > >
> > > >>> > > > > > > I would also add that the KIP removes the global sync
> > > >>> barrier in the
> > > >>> > > > > > > protocol which is essential to improve group stability
> > and
> > > >>> > > > > > > scalability, and the KIP also simplifies the client by
> > > >>> moving most of
> > > >>> > > > > > > the logic to the server side.
> > > >>> > > > > > >
> > > >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > support
> > > >>> rolling
> > > >>> > > > > > changing
> > > >>> > > > > > > > of the partition assignor in the consumers?
> > > >>> > > > > > >
> > > >>> > > > > > > Definitely. The group coordinator will use the assignor
> > > used
> > > >>> by a
> > > >>> > > > > > > majority of the members. This allows the group to move
> > from
> > > >>> one
> > > >>> > > > > > > assignor to another by a roll. This is explained in the
> > > >>> Assignor
> > > >>> > > > > > > Selection chapter.
> > > >>> > > > > > >
> > > >>> > > > > > > > 30.2 For each field, could you explain whether it's
> > > >>> required in
> > > >>> > > > every
> > > >>> > > > > > > > request or the scenarios when it needs to be filled?
> > For
> > > >>> example,
> > > >>> > > > it's
> > > >>> > > > > > not
> > > >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > >>> > > > > > >
> > > >>> > > > > > > The client is expected to set those fields in case of a
> > > >>> connection
> > > >>> > > > > > > issue (e.g. timeout) or when the fields have changed
> > since
> > > >>> the last
> > > >>> > > > > > > HB. The server populates those fields as long as the
> > member
> > > >>> is not
> > > >>> > > > > > > fully reconciled - the member should acknowledge that it
> > > has
> > > >>> the
> > > >>> > > > > > > expected epoch and assignment. I will clarify this in the
> > > >>> KIP.
> > > >>> > > > > > >
> > > >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > > >>> between the
> > > >>> > > > > > client
> > > >>> > > > > > > > and the broker is only considered during fetching, but
> > > not
> > > >>> during
> > > >>> > > > > > assigning
> > > >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > > is
> > > >>> made,
> > > >>> > > > there
> > > >>> > > > > > is
> > > >>> > > > > > > > no opportunity for read affinity because no replicas of
> > > >>> assigned
> > > >>> > > > > > partitions
> > > >>> > > > > > > > are close to the member. I am wondering if we should
> > use
> > > >>> this
> > > >>> > > > > > opportunity
> > > >>> > > > > > > > to address this by including rack in GroupMember.
> > > >>> > > > > > >
> > > >>> > > > > > > That's an interesting idea. I don't see any issue with
> > > >>> adding the
> > > >>> > > > rack
> > > >>> > > > > > > to the members. I will do so.
> > > >>> > > > > > >
> > > >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> > > >>> busy a group
> > > >>> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > >>> that we
> > > >>> > > > could
> > > >>> > > > > > add
> > > >>> > > > > > > > a metric that tracks the fraction of the time the event
> > > >>> loop is
> > > >>> > > > doing
> > > >>> > > > > > the
> > > >>> > > > > > > > actual work.
> > > >>> > > > > > >
> > > >>> > > > > > > That's a great idea. I will add it. Thanks.
> > > >>> > > > > > >
> > > >>> > > > > > > > 33. Could we add a section on coordinator failover
> > > >>> handling? For
> > > >>> > > > > > example,
> > > >>> > > > > > > > does it need to trigger the check if any group with the
> > > >>> wildcard
> > > >>> > > > > > > > subscription now has a new matching topic?
> > > >>> > > > > > >
> > > >>> > > > > > > Sure. When the new group coordinator takes over, it has
> > to:
> > > >>> > > > > > > * Setup the session timeouts.
> > > >>> > > > > > > * Trigger a new assignment if a client side assignor is
> > > >>> used. We
> > > >>> > > > don't
> > > >>> > > > > > > store the information about the member selected to run
> > the
> > > >>> assignment
> > > >>> > > > > > > so we have to start a new one.
> > > >>> > > > > > > * Update the topics metadata, verify the wildcard
> > > >>> subscriptions, and
> > > >>> > > > > > > trigger a rebalance if needed.
> > > >>> > > > > > >
> > > >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > >>> > > > ConsumerGroupPartitionMetadataValue,
> > > >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > >>> the epoch
> > > >>> > > > > > field
> > > >>> > > > > > > > reflects? For example, does the epoch in
> > > >>> ConsumerGroupMetadataValue
> > > >>> > > > > > reflect
> > > >>> > > > > > > > the latest group epoch? What about the one in
> > > >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > >>> > > > > > ConsumerGroupMemberMetadataValue?
> > > >>> > > > > > >
> > > >>> > > > > > > Sure. I will clarify that but it is always the latest
> > group
> > > >>> epoch.
> > > >>> > > > > > > When the group state is updated, the group epoch is
> > bumped
> > > >>> so we use
> > > >>> > > > > > > that one for all the change records related to the
> > update.
> > > >>> > > > > > >
> > > >>> > > > > > > > 35. "the group coordinator will ensure that the
> > following
> > > >>> > > > invariants
> > > >>> > > > > > are
> > > >>> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > >>> not to
> > > >>> > > > get any
> > > >>> > > > > > > > assigned partitions, right?
> > > >>> > > > > > >
> > > >>> > > > > > > That's right. Here I meant that the members provided by
> > the
> > > >>> assignor
> > > >>> > > > > > > in the assignment must exist in the group. The assignor
> > can
> > > >>> not make
> > > >>> > > > > > > up new member ids.
> > > >>> > > > > > >
> > > >>> > > > > > > > 36. "He can rejoins the group with a member epoch
> > equals
> > > >>> to 0":
> > > >>> > > > When
> > > >>> > > > > > would
> > > >>> > > > > > > > a consumer rejoin and what member id would be used?
> > > >>> > > > > > >
> > > >>> > > > > > > A member is expected to abandon all its partitions and
> > > >>> rejoins when
> > > >>> > > > it
> > > >>> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> > > >>> group
> > > >>> > > > > > > coordinator will have removed the member from the group.
> > > The
> > > >>> member
> > > >>> > > > > > > can rejoin the group with the same member id but with 0
> > as
> > > >>> epoch. Let
> > > >>> > > > > > > me see if I can clarify this in the KIP.
> > > >>> > > > > > >
> > > >>> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger
> > > >>> a
> > > >>> > > > > > reassignment
> > > >>> > > > > > > > by either providing a non-zero reason or by updating
> > the
> > > >>> assignor
> > > >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > >>> deprecation
> > > >>> > > > of
> > > >>> > > > > > > > Consumer#enforeRebalance.
> > > >>> > > > > > >
> > > >>> > > > > > > In this case, a new assignment is triggered by the client
> > > >>> side
> > > >>> > > > > > > assignor. When constructing the HB, the consumer will
> > > always
> > > >>> consult
> > > >>> > > > > > > the client side assignor and propagate the information to
> > > >>> the group
> > > >>> > > > > > > coordinator. In other words, we don't expect users to
> > call
> > > >>> > > > > > > Consumer#enforceRebalance anymore.
> > > >>> > > > > > >
> > > >>> > > > > > > > 38. The reassignment examples are nice. But the section
> > > >>> seems to
> > > >>> > > > have
> > > >>> > > > > > > > multiple typos.
> > > >>> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > immediately
> > > >>> gets into
> > > >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > > get
> > > >>> into
> > > >>> > > > > > epoch=3,
> > > >>> > > > > > > > partitions=[foo-1] too early.
> > > >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > >>> epoch=2,
> > > >>> > > > > > > > partitions=[foo-0].
> > > >>> > > > > > >
> > > >>> > > > > > > Sorry for that! I will revise them.
> > > >>> > > > > > >
> > > >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > >>> upgrade from
> > > >>> > > > any
> > > >>> > > > > > old
> > > >>> > > > > > > > version to new one?
> > > >>> > > > > > >
> > > >>> > > > > > > We will support upgrading from the consumer protocol
> > > version
> > > >>> 3,
> > > >>> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> > > the
> > > >>> earliest
> > > >>> > > > > > > version is unknown at the moment. This is explained in
> > the
> > > >>> migration
> > > >>> > > > > > > plan chapter.
> > > >>> > > > > > >
> > > >>> > > > > > > Thanks again for your feedback, Jun. I will update the
> > KIP
> > > >>> based on
> > > >>> > > > it
> > > >>> > > > > > > next week.
> > > >>> > > > > > >
> > > >>> > > > > > > Best,
> > > >>> > > > > > > David
> > > >>> > > > > > >
> > > >>> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > > >>> <ju...@confluent.io.invalid>
> > > >>> > > > wrote:
> > > >>> > > > > > > >
> > > >>> > > > > > > > Hi, David,
> > > >>> > > > > > > >
> > > >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> > KIP
> > > >>> seem to
> > > >>> > > > be
> > > >>> > > > > > fewer
> > > >>> > > > > > > > RPCs during rebalance and more efficient support of
> > > >>> wildcard. A few
> > > >>> > > > > > > > comments below.
> > > >>> > > > > > > >
> > > >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> > support
> > > >>> rolling
> > > >>> > > > > > changing
> > > >>> > > > > > > > of the partition assignor in the consumers?
> > > >>> > > > > > > > 30.2 For each field, could you explain whether it's
> > > >>> required in
> > > >>> > > > every
> > > >>> > > > > > > > request or the scenarios when it needs to be filled?
> > For
> > > >>> example,
> > > >>> > > > it's
> > > >>> > > > > > not
> > > >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > >>> > > > > > > >
> > > >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > > >>> between the
> > > >>> > > > > > client
> > > >>> > > > > > > > and the broker is only considered during fetching, but
> > > not
> > > >>> during
> > > >>> > > > > > assigning
> > > >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > > is
> > > >>> made,
> > > >>> > > > there
> > > >>> > > > > > is
> > > >>> > > > > > > > no opportunity for read affinity because no replicas of
> > > >>> assigned
> > > >>> > > > > > partitions
> > > >>> > > > > > > > are close to the member. I am wondering if we should
> > use
> > > >>> this
> > > >>> > > > > > opportunity
> > > >>> > > > > > > > to address this by including rack in GroupMember.
> > > >>> > > > > > > >
> > > >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> > > >>> busy a group
> > > >>> > > > > > > > coordinator is. By moving the event loop model, it
> > seems
> > > >>> that we
> > > >>> > > > could
> > > >>> > > > > > add
> > > >>> > > > > > > > a metric that tracks the fraction of the time the event
> > > >>> loop is
> > > >>> > > > doing
> > > >>> > > > > > the
> > > >>> > > > > > > > actual work.
> > > >>> > > > > > > >
> > > >>> > > > > > > > 33. Could we add a section on coordinator failover
> > > >>> handling? For
> > > >>> > > > > > example,
> > > >>> > > > > > > > does it need to trigger the check if any group with the
> > > >>> wildcard
> > > >>> > > > > > > > subscription now has a new matching topic?
> > > >>> > > > > > > >
> > > >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > >>> > > > ConsumerGroupPartitionMetadataValue,
> > > >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> > what
> > > >>> the epoch
> > > >>> > > > > > field
> > > >>> > > > > > > > reflects? For example, does the epoch in
> > > >>> ConsumerGroupMetadataValue
> > > >>> > > > > > reflect
> > > >>> > > > > > > > the latest group epoch? What about the one in
> > > >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > >>> > > > > > ConsumerGroupMemberMetadataValue?
> > > >>> > > > > > > >
> > > >>> > > > > > > > 35. "the group coordinator will ensure that the
> > following
> > > >>> > > > invariants
> > > >>> > > > > > are
> > > >>> > > > > > > > met: ... All members exists." It's possible for a
> > member
> > > >>> not to
> > > >>> > > > get any
> > > >>> > > > > > > > assigned partitions, right?
> > > >>> > > > > > > >
> > > >>> > > > > > > > 36. "He can rejoins the group with a member epoch
> > equals
> > > >>> to 0":
> > > >>> > > > When
> > > >>> > > > > > would
> > > >>> > > > > > > > a consumer rejoin and what member id would be used?
> > > >>> > > > > > > >
> > > >>> > > > > > > > 37. "Instead, power users will have the ability to
> > > trigger
> > > >>> a
> > > >>> > > > > > reassignment
> > > >>> > > > > > > > by either providing a non-zero reason or by updating
> > the
> > > >>> assignor
> > > >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > > >>> deprecation
> > > >>> > > > of
> > > >>> > > > > > > > Consumer#enforeRebalance.
> > > >>> > > > > > > >
> > > >>> > > > > > > > 38. The reassignment examples are nice. But the section
> > > >>> seems to
> > > >>> > > > have
> > > >>> > > > > > > > multiple typos.
> > > >>> > > > > > > > 38.1 When the group transitions to epoch 2, B
> > immediately
> > > >>> gets into
> > > >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > > get
> > > >>> into
> > > >>> > > > > > epoch=3,
> > > >>> > > > > > > > partitions=[foo-1] too early.
> > > >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > > >>> epoch=2,
> > > >>> > > > > > > > partitions=[foo-0].
> > > >>> > > > > > > >
> > > >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > > >>> upgrade from
> > > >>> > > > any
> > > >>> > > > > > old
> > > >>> > > > > > > > version to new one?
> > > >>> > > > > > > >
> > > >>> > > > > > > > Thanks,
> > > >>> > > > > > > >
> > > >>> > > > > > > > Jun
> > > >>> > > > > > > >
> > > >>> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > >>> > > > > > <dj...@confluent.io.invalid>
> > > >>> > > > > > > > wrote:
> > > >>> > > > > > > >
> > > >>> > > > > > > > > Hi all,
> > > >>> > > > > > > > >
> > > >>> > > > > > > > > The KIP states that we will re-implement the
> > > coordinator
> > > >>> in
> > > >>> > > > Java. I
> > > >>> > > > > > > > > discussed this offline with a few folks and folks are
> > > >>> concerned
> > > >>> > > > that
> > > >>> > > > > > > > > we could introduce many regressions in the old
> > protocol
> > > >>> if we do
> > > >>> > > > so.
> > > >>> > > > > > > > > Therefore, I am going to remove this statement from
> > the
> > > >>> KIP. It
> > > >>> > > > is an
> > > >>> > > > > > > > > implementation detail after all so it does not have
> > to
> > > be
> > > >>> > > > decided at
> > > >>> > > > > > > > > this stage. We will likely start by trying to
> > refactor
> > > >>> the
> > > >>> > > > current
> > > >>> > > > > > > > > implementation as a first step.
> > > >>> > > > > > > > >
> > > >>> > > > > > > > > Cheers,
> > > >>> > > > > > > > > David
> > > >>> > > > > > > > >
> > > >>> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > > >>> djacot@confluent.io
> > > >>> > > > >
> > > >>> > > > > > wrote:
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > Hi Luke,
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > >>> assigning,
> > > >>> > > > > > reconciling,
> > > >>> > > > > > > > > stable,
> > > >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > > >>> right?
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > This sentence does not refer to those group states
> > > but
> > > >>> rather
> > > >>> > > > to a
> > > >>> > > > > > > > > > state machine replication (SMR). This refers to the
> > > >>> entire
> > > >>> > > > state of
> > > >>> > > > > > > > > > group coordinator which is replicated via the log
> > > >>> layer. I will
> > > >>> > > > > > > > > > clarify this in the KIP.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > >>> modelled as an
> > > >>> > > > event
> > > >>> > > > > > > > > loop"?
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > The idea is to follow a model similar to the new
> > > quorum
> > > >>> > > > > > controller. We
> > > >>> > > > > > > > > > will have N threads to process events. Each
> > > >>> __consumer_offsets
> > > >>> > > > > > > > > > partition is assigned to a unique thread and all
> > the
> > > >>> events
> > > >>> > > > (e.g.
> > > >>> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > > >>> thread. This
> > > >>> > > > > > simplify
> > > >>> > > > > > > > > > concurrency and will enable us to do simulation
> > > >>> testing for the
> > > >>> > > > > > group
> > > >>> > > > > > > > > > coordinator.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > >>> *__consumer_offsets*
> > > >>> > > > > > > > > partitions?
> > > >>> > > > > > > > > > > Not a state machine "per consumer group" owned
> > by a
> > > >>> group
> > > >>> > > > > > coordinator?
> > > >>> > > > > > > > > For
> > > >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > > >>> groups, and
> > > >>> > > > > > both
> > > >>> > > > > > > > > exist in
> > > >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > >>> for it,
> > > >>> > > > or
> > > >>> > > > > > 2?
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > > the
> > > >>> number
> > > >>> > > > of
> > > >>> > > > > > threads
> > > >>> > > > > > > > > used
> > > >>> > > > > > > > > > > to run the state machines. But I'm wondering if
> > the
> > > >>> purpose
> > > >>> > > > of
> > > >>> > > > > > the
> > > >>> > > > > > > > > threads
> > > >>> > > > > > > > > > > is only to keep the state of each consumer group
> > > (or
> > > >>> > > > > > > > > *__consumer_offsets*
> > > >>> > > > > > > > > > > partitions?), and no heavy computation, why
> > should
> > > >>> we need
> > > >>> > > > > > > > > multi-threads
> > > >>> > > > > > > > > > > here?
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > See 1.2. The idea is to have an ability to shard
> > the
> > > >>> > > > processing as
> > > >>> > > > > > the
> > > >>> > > > > > > > > > computation could be heavy.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > >>> default
> > > >>> > > > session
> > > >>> > > > > > > > > timeout not
> > > >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > > the
> > > >>> min/max
> > > >>> > > > > > session
> > > >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > >>> timeout to
> > > >>> > > > detect
> > > >>> > > > > > client
> > > >>> > > > > > > > > > > failures when using the consumer group protocol.
> > > >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > The
> > > >>> minimum
> > > >>> > > > > > session
> > > >>> > > > > > > > > timeout.
> > > >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > The
> > > >>> maximum
> > > >>> > > > > > session
> > > >>> > > > > > > > > timeout.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > This is indeed a mistake. The default session
> > timeout
> > > >>> should
> > > >>> > > > be 45s
> > > >>> > > > > > > > > > (the current default).
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > > >>> uniform],
> > > >>> > > > > > which means
> > > >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > know
> > > >>> why not
> > > >>> > > > > > uniform
> > > >>> > > > > > > > > one? I
> > > >>> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > >>> (former
> > > >>> > > > sticky
> > > >>> > > > > > > > > assinor)
> > > >>> > > > > > > > > > > for better evenly distribution. Any other reason
> > we
> > > >>> choose
> > > >>> > > > range
> > > >>> > > > > > > > > assignor
> > > >>> > > > > > > > > > > as default?
> > > >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > >>> server side
> > > >>> > > > > > assignors.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > The order on the server side has no influence
> > because
> > > >>> the
> > > >>> > > > client
> > > >>> > > > > > must
> > > >>> > > > > > > > > > chose the selector that he wants to use. There is
> > no
> > > >>> default
> > > >>> > > > in the
> > > >>> > > > > > > > > > current proposal. If the assignor is not specified
> > by
> > > >>> the
> > > >>> > > > client,
> > > >>> > > > > > the
> > > >>> > > > > > > > > > request is rejected. The default client value for
> > > >>> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> > that
> > > >>> my
> > > >>> > > > answers
> > > >>> > > > > > help
> > > >>> > > > > > > > > > to clarify things. I will update the KIP as well
> > > based
> > > >>> on your
> > > >>> > > > > > > > > > feedback.
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > Cheers,
> > > >>> > > > > > > > > > David
> > > >>> > > > > > > > > >
> > > >>> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > > >>> showuon@gmail.com>
> > > >>> > > > > > wrote:
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > Hi David,
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > Thanks for the update.
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > Some more questions:
> > > >>> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > >>> > > > > > > > > > > > The new group coordinator will have a state
> > > >>> machine per
> > > >>> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> > > >>> machine is
> > > >>> > > > > > modelled
> > > >>> > > > > > > > > as an
> > > >>> > > > > > > > > > > event loop. Those state machines will be executed
> > > in
> > > >>> > > > > > > > > > > *group.coordinator.threads* threads.
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > > >>> assigning,
> > > >>> > > > > > reconciling,
> > > >>> > > > > > > > > stable,
> > > >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > > >>> right?
> > > >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > > >>> modelled as an
> > > >>> > > > event
> > > >>> > > > > > > > > loop"?
> > > >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> > > >>> *__consumer_offsets*
> > > >>> > > > > > > > > partitions?
> > > >>> > > > > > > > > > > Not a state machine "per consumer group" owned
> > by a
> > > >>> group
> > > >>> > > > > > coordinator?
> > > >>> > > > > > > > > For
> > > >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > > >>> groups, and
> > > >>> > > > > > both
> > > >>> > > > > > > > > exist in
> > > >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > > machine
> > > >>> for it,
> > > >>> > > > or
> > > >>> > > > > > 2?
> > > >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > > the
> > > >>> number
> > > >>> > > > of
> > > >>> > > > > > threads
> > > >>> > > > > > > > > used
> > > >>> > > > > > > > > > > to run the state machines. But I'm wondering if
> > the
> > > >>> purpose
> > > >>> > > > of
> > > >>> > > > > > the
> > > >>> > > > > > > > > threads
> > > >>> > > > > > > > > > > is only to keep the state of each consumer group
> > > (or
> > > >>> > > > > > > > > *__consumer_offsets*
> > > >>> > > > > > > > > > > partitions?), and no heavy computation, why
> > should
> > > >>> we need
> > > >>> > > > > > > > > multi-threads
> > > >>> > > > > > > > > > > here?
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > 2. For the default value in the new configs:
> > > >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > > >>> default
> > > >>> > > > session
> > > >>> > > > > > > > > timeout not
> > > >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > > the
> > > >>> min/max
> > > >>> > > > > > session
> > > >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > > >>> timeout to
> > > >>> > > > detect
> > > >>> > > > > > client
> > > >>> > > > > > > > > > > failures when using the consumer group protocol.
> > > >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> > The
> > > >>> minimum
> > > >>> > > > > > session
> > > >>> > > > > > > > > timeout.
> > > >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> > The
> > > >>> maximum
> > > >>> > > > > > session
> > > >>> > > > > > > > > timeout.
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > > >>> uniform],
> > > >>> > > > > > which means
> > > >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> > know
> > > >>> why not
> > > >>> > > > > > uniform
> > > >>> > > > > > > > > one? I
> > > >>> > > > > > > > > > > thought usually users will choose uniform
> > assignor
> > > >>> (former
> > > >>> > > > sticky
> > > >>> > > > > > > > > assinor)
> > > >>> > > > > > > > > > > for better evenly distribution. Any other reason
> > we
> > > >>> choose
> > > >>> > > > range
> > > >>> > > > > > > > > assignor
> > > >>> > > > > > > > > > > as default?
> > > >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > > >>> server side
> > > >>> > > > > > assignors.
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > Thank you.
> > > >>> > > > > > > > > > > Luke
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > > >>> showuon@gmail.com
> > > >>> > > > >
> > > >>> > > > > > wrote:
> > > >>> > > > > > > > > > >
> > > >>> > > > > > > > > > > > Hi Sagar,
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > > >>> integrating with
> > > >>> > > > > > KIP-848,
> > > >>> > > > > > > > > but I
> > > >>> > > > > > > > > > > > think we should have a separate discussion
> > thread
> > > >>> for the
> > > >>> > > > Kafka
> > > >>> > > > > > > > > Connect
> > > >>> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> > Consumer
> > > >>> Rebalance
> > > >>> > > > > > Protocol
> > > >>> > > > > > > > > [1],
> > > >>> > > > > > > > > > > > and let this discussion thread focus on
> > consumer
> > > >>> rebalance
> > > >>> > > > > > protocol,
> > > >>> > > > > > > > > WDYT?
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > > > > > [1]
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > >
> > > >>> > > > > >
> > > >>> > > >
> > > >>>
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > > > > > Thank you.
> > > >>> > > > > > > > > > > > Luke
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > >>> > > > > > sagarmeansocean@gmail.com>
> > > >>> > > > > > > > > wrote:
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> > Looks
> > > >>> like
> > > >>> > > > there's
> > > >>> > > > > > > > > agreement on
> > > >>> > > > > > > > > > > >> using separate APIs for Connect. I would
> > revisit
> > > >>> the doc
> > > >>> > > > and
> > > >>> > > > > > see
> > > >>> > > > > > > > > what
> > > >>> > > > > > > > > > > >> changes are to be made.
> > > >>> > > > > > > > > > > >>
> > > >>> > > > > > > > > > > >> Thanks!
> > > >>> > > > > > > > > > > >> Sagar.
> > > >>> > > > > > > > > > > >>
> > > >>> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > >>> > > > > > > > > <dj...@confluent.io.invalid>
> > > >>> > > > > > > > > > > >> wrote:
> > > >>> > > > > > > > > > > >>
> > > >>> > > > > > > > > > > >> > Hi Sagar,
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > > >>> That's really
> > > >>> > > > > > helpful. I
> > > >>> > > > > > > > > > > >> > will take a look at it.
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> > and
> > > >>> the
> > > >>> > > > Consumer
> > > >>> > > > > > could
> > > >>> > > > > > > > > share
> > > >>> > > > > > > > > > > >> > the same underlying "engine". The main
> > > >>> difference is
> > > >>> > > > that
> > > >>> > > > > > the
> > > >>> > > > > > > > > Consumer
> > > >>> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > > >>> Connect
> > > >>> > > > assigns
> > > >>> > > > > > tasks
> > > >>> > > > > > > > > to
> > > >>> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > >>> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> > support
> > > >>> different
> > > >>> > > > > > resource
> > > >>> > > > > > > > > types
> > > >>> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > >>> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect.
> > The
> > > >>> dedicated
> > > >>> > > > APIs
> > > >>> > > > > > > > > would be
> > > >>> > > > > > > > > > > >> > similar to the new ones but different on the
> > > >>> > > > > > content/resources and
> > > >>> > > > > > > > > > > >> > they would rely on the same engine on the
> > > >>> coordinator
> > > >>> > > > side.
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >> > I personally lean towards 2) because I am
> > not
> > > a
> > > >>> fan of
> > > >>> > > > > > > > > overcharging
> > > >>> > > > > > > > > > > >> > APIs to serve different purposes. That being
> > > >>> said, I am
> > > >>> > > > not
> > > >>> > > > > > > > > opposed to
> > > >>> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >> > I think that we can continue to discuss it
> > > here
> > > >>> for now
> > > >>> > > > in
> > > >>> > > > > > order
> > > >>> > > > > > > > > to
> > > >>> > > > > > > > > > > >> > ensure that this KIP is compatible with what
> > > we
> > > >>> will do
> > > >>> > > > for
> > > >>> > > > > > > > > Connect in
> > > >>> > > > > > > > > > > >> > the future.
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >> > Best,
> > > >>> > > > > > > > > > > >> > David
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > >>> > > > > > djacot@confluent.io>
> > > >>> > > > > > > > > wrote:
> > > >>> > > > > > > > > > > >> > >
> > > >>> > > > > > > > > > > >> > > Hi all,
> > > >>> > > > > > > > > > > >> > >
> > > >>> > > > > > > > > > > >> > > I am back from vacation. I will go through
> > > >>> and address
> > > >>> > > > > > your
> > > >>> > > > > > > > > comments
> > > >>> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > > feedback.
> > > >>> > > > > > > > > > > >> > >
> > > >>> > > > > > > > > > > >> > > Cheers,
> > > >>> > > > > > > > > > > >> > > David
> > > >>> > > > > > > > > > > >> > >
> > > >>> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > > >>> Harris <
> > > >>> > > > > > > > > gharris1727@gmail.com
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >> > wrote:
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > Hey All!
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to
> > see
> > > >>> > > > cooperative
> > > >>> > > > > > > > > rebalancing
> > > >>> > > > > > > > > > > >> > making it
> > > >>> > > > > > > > > > > >> > > > down the stack!
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > I had a few questions:
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > > >>> describes how
> > > >>> > > > > > member
> > > >>> > > > > > > > > epoch
> > > >>> > > > > > > > > > > >> > should
> > > >>> > > > > > > > > > > >> > > > advance in step with the group epoch and
> > > >>> assignment
> > > >>> > > > > > epoch
> > > >>> > > > > > > > > values. I
> > > >>> > > > > > > > > > > >> > think
> > > >>> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> > > >>> described
> > > >>> > > > in
> > > >>> > > > > > the
> > > >>> > > > > > > > > KIP. When
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > > >>> partitions to a
> > > >>> > > > > > worker,
> > > >>> > > > > > > > > what
> > > >>> > > > > > > > > > > >> member
> > > >>> > > > > > > > > > > >> > > > epoch does each incremental assignment
> > > use?
> > > >>> Are
> > > >>> > > > member
> > > >>> > > > > > epochs
> > > >>> > > > > > > > > > > >> re-used,
> > > >>> > > > > > > > > > > >> > and
> > > >>> > > > > > > > > > > >> > > > a single member epoch can correspond to
> > > >>> multiple
> > > >>> > > > > > different
> > > >>> > > > > > > > > > > >> > (monotonically
> > > >>> > > > > > > > > > > >> > > > larger) assignments?
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> > opaque
> > > >>> to the
> > > >>> > > > group
> > > >>> > > > > > > > > > > >> coordinator? If
> > > >>> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > > >>> > > > implementations
> > > >>> > > > > > > > > interact
> > > >>> > > > > > > > > > > >> with
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > Reason field, and how is its common
> > > meaning
> > > >>> agreed
> > > >>> > > > > > upon? If
> > > >>> > > > > > > > > so, what
> > > >>> > > > > > > > > > > >> > is the
> > > >>> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > > >>> including
> > > >>> > > > such
> > > >>> > > > > > > > > functionality
> > > >>> > > > > > > > > > > >> > in the
> > > >>> > > > > > > > > > > >> > > > opaque metadata?
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> > > >>> "Thanks to
> > > >>> > > > > > this, the
> > > >>> > > > > > > > > input
> > > >>> > > > > > > > > > > >> of
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > client side assignor is entirely driven
> > by
> > > >>> the group
> > > >>> > > > > > > > > coordinator.
> > > >>> > > > > > > > > > > >> The
> > > >>> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > > >>> maintaining
> > > >>> > > > any
> > > >>> > > > > > state
> > > >>> > > > > > > > > besides
> > > >>> > > > > > > > > > > >> its
> > > >>> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> > that
> > > >>> the
> > > >>> > > > > > client-side
> > > >>> > > > > > > > > assignor
> > > >>> > > > > > > > > > > >> MAY
> > > >>> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> > state
> > > >>> (such as
> > > >>> > > > > > partition
> > > >>> > > > > > > > > > > >> > throughput,
> > > >>> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc),
> > > or
> > > >>> that
> > > >>> > > > > > additional
> > > >>> > > > > > > > > > > >> > non-Metadata
> > > >>> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> > > >>> > > > > > > > > > > >> > > > for
> > > >>> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > >>> > > > > > > > > > > >> > > > and
> > > >>> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > >>> > > > > > that
> > > >>> > > > > > > > > seem to
> > > >>> > > > > > > > > > > >> > > > overlap significantly. Is it possible
> > for
> > > >>> these two
> > > >>> > > > > > > > > implementations
> > > >>> > > > > > > > > > > >> to
> > > >>> > > > > > > > > > > >> > be
> > > >>> > > > > > > > > > > >> > > > unified? This would serve to promote
> > > >>> feature parity
> > > >>> > > > of
> > > >>> > > > > > > > > server-side
> > > >>> > > > > > > > > > > >> and
> > > >>> > > > > > > > > > > >> > > > client-side assignors, and would also
> > > >>> facilitate
> > > >>> > > > > > operational
> > > >>> > > > > > > > > > > >> > flexibility in
> > > >>> > > > > > > > > > > >> > > > certain situations. For example, if a
> > > >>> server-side
> > > >>> > > > > > assignor
> > > >>> > > > > > > > > has some
> > > >>> > > > > > > > > > > >> > poor
> > > >>> > > > > > > > > > > >> > > > behavior and needs a patch, deploying
> > the
> > > >>> patched
> > > >>> > > > > > assignor to
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > client
> > > >>> > > > > > > > > > > >> > > > and switching one consumer group to a
> > > >>> client-side
> > > >>> > > > > > assignor
> > > >>> > > > > > > > > may be
> > > >>> > > > > > > > > > > >> > faster
> > > >>> > > > > > > > > > > >> > > > and less risky than patching all of the
> > > >>> brokers.
> > > >>> > > > With
> > > >>> > > > > > the
> > > >>> > > > > > > > > currently
> > > >>> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > >>> > > > reimplementation
> > > >>> > > > > > would
> > > >>> > > > > > > > > have
> > > >>> > > > > > > > > > > >> to be
> > > >>> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > > diverged
> > > >>> > > > > > significantly,
> > > >>> > > > > > > > > then it
> > > >>> > > > > > > > > > > >> is
> > > >>> > > > > > > > > > > >> > > > possible that a reimplementation would
> > not
> > > >>> be
> > > >>> > > > possible.
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > --
> > > >>> > > > > > > > > > > >> > > > Greg Harris
> > > >>> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > >>> > > > > > > > > > > >> > > > github.com/gharris1727
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > >>> > > > > > > > > sagarmeansocean@gmail.com>
> > > >>> > > > > > > > > > > >> > wrote:
> > > >>> > > > > > > > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > I created a confluence page to discuss
> > > >>> how Connect
> > > >>> > > > > > would
> > > >>> > > > > > > > > need to
> > > >>> > > > > > > > > > > >> > change
> > > >>> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> > > >>> Here's the
> > > >>> > > > page:
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >>
> > > >>> > > > > > > > >
> > > >>> > > > > >
> > > >>> > > >
> > > >>>
> > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > > tried
> > > >>> to keep
> > > >>> > > > a
> > > >>> > > > > > format
> > > >>> > > > > > > > > > > >> similar to
> > > >>> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> > > >>> Also, do you
> > > >>> > > > > > think this
> > > >>> > > > > > > > > > > >> should
> > > >>> > > > > > > > > > > >> > be
> > > >>> > > > > > > > > > > >> > > > > moved to a separate discussion thread
> > or
> > > >>> is this
> > > >>> > > > one
> > > >>> > > > > > fine?
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > Thanks!
> > > >>> > > > > > > > > > > >> > > > > Sagar.
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar
> > <
> > > >>> > > > > > > > > sagarmeansocean@gmail.com>
> > > >>> > > > > > > > > > > >> > wrote:
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > >>> > > > > > > > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> > Kafka
> > > >>> Streams.
> > > >>> > > > > > Sure, I
> > > >>> > > > > > > > > would do
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > > >>> document.
> > > >>> > > > > > > > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > > Thanks!
> > > >>> > > > > > > > > > > >> > > > > > Sagar.
> > > >>> > > > > > > > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > > >>> Guozhang Wang <
> > > >>> > > > > > > > > > > >> wangguoz@gmail.com>
> > > >>> > > > > > > > > > > >> > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > > >> It would be great if you could come
> > > >>> back with
> > > >>> > > > some
> > > >>> > > > > > > > > analysis on
> > > >>> > > > > > > > > > > >> > how to
> > > >>> > > > > > > > > > > >> > > > > >> implement the Connect side
> > > integration
> > > >>> with
> > > >>> > > > the new
> > > >>> > > > > > > > > protocol;
> > > >>> > > > > > > > > > > >> so
> > > >>> > > > > > > > > > > >> > far
> > > >>> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > > >>> "protocol type"
> > > >>> > > > we
> > > >>> > > > > > did not
> > > >>> > > > > > > > > yet
> > > >>> > > > > > > > > > > >> think
> > > >>> > > > > > > > > > > >> > > > > through
> > > >>> > > > > > > > > > > >> > > > > >> the Connect side implementations.
> > For
> > > >>> Streams
> > > >>> > > > > > here's a
> > > >>> > > > > > > > > draft of
> > > >>> > > > > > > > > > > >> > > > > >> integration
> > > >>> > > > > > > > > > > >> > > > > >> plan:
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >>
> > > >>> > > > > > > > >
> > > >>> > > > > >
> > > >>> > > >
> > > >>>
> > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > >>> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> > > Connect.
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> > > Sagar
> > > >>> <
> > > >>> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > >>> > > > > > > > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > > >> > Hi David,
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> > > >>> reason I
> > > >>> > > > thought
> > > >>> > > > > > > > > connect can
> > > >>> > > > > > > > > > > >> > also fit
> > > >>> > > > > > > > > > > >> > > > > >> into
> > > >>> > > > > > > > > > > >> > > > > >> > this new scheme is that even
> > today
> > > >>> the
> > > >>> > > > connect
> > > >>> > > > > > uses a
> > > >>> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > >>> > > > > > > > > > > >> > > > > >> > extending from
> > AbstractCoordinator
> > > >>> to empower
> > > >>> > > > > > > > > rebalances of
> > > >>> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > > >>> WorkerCoordinator sets
> > > >>> > > > the
> > > >>> > > > > > > > > > > >> protocolType()
> > > >>> > > > > > > > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > >> connect
> > > >>> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> > > >>> plumbing
> > > >>> > > > into
> > > >>> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > > >>> connect would
> > > >>> > > > be
> > > >>> > > > > > > > > similar at a
> > > >>> > > > > > > > > > > >> > high
> > > >>> > > > > > > > > > > >> > > > > level
> > > >>> > > > > > > > > > > >> > > > > >> to
> > > >>> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> > > >>> because of the
> > > >>> > > > > > Client
> > > >>> > > > > > > > > side
> > > >>> > > > > > > > > > > >> > assignors
> > > >>> > > > > > > > > > > >> > > > > being
> > > >>> > > > > > > > > > > >> > > > > >> > used in both. At an
> > implementation
> > > >>> level, we
> > > >>> > > > > > might
> > > >>> > > > > > > > > need to
> > > >>> > > > > > > > > > > >> make
> > > >>> > > > > > > > > > > >> > a lot
> > > >>> > > > > > > > > > > >> > > > > of
> > > >>> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > > >>> assignment
> > > >>> > > > protocol
> > > >>> > > > > > like
> > > >>> > > > > > > > > > > >> enhancing
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > > >>> SyncGroup and
> > > >>> > > > > > using
> > > >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > >>> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> > > >>> streams (or
> > > >>> > > > > > there
> > > >>> > > > > > > > > might be
> > > >>> > > > > > > > > > > >> > > > > >> deviations). I
> > > >>> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> > > >>> analysis of
> > > >>> > > > the
> > > >>> > > > > > same
> > > >>> > > > > > > > > and we
> > > >>> > > > > > > > > > > >> > can have
> > > >>> > > > > > > > > > > >> > > > > a
> > > >>> > > > > > > > > > > >> > > > > >> > separate discussion thread for
> > that
> > > >>> as that
> > > >>> > > > would
> > > >>> > > > > > > > > derail this
> > > >>> > > > > > > > > > > >> > > > > discussion
> > > >>> > > > > > > > > > > >> > > > > >> > thread. Let me know if that
> > sounds
> > > >>> good to
> > > >>> > > > you.
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> > Thanks!
> > > >>> > > > > > > > > > > >> > > > > >> > Sagar.
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> > > >>> David Jacot
> > > >>> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > > >>> `Assignment#error`.
> > > >>> > > > > > Sure, I
> > > >>> > > > > > > > > can
> > > >>> > > > > > > > > > > >> > mention it.
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> > > >>> from his
> > > >>> > > > current
> > > >>> > > > > > > > > assignment
> > > >>> > > > > > > > > > > >> to
> > > >>> > > > > > > > > > > >> > his
> > > >>> > > > > > > > > > > >> > > > > >> > > target assignment when he can
> > > move
> > > >>> to
> > > >>> > > > epoch 3.
> > > >>> > > > > > When
> > > >>> > > > > > > > > that
> > > >>> > > > > > > > > > > >> > happens,
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > member assignment is updated
> > and
> > > >>> persisted
> > > >>> > > > > > with all
> > > >>> > > > > > > > > its
> > > >>> > > > > > > > > > > >> > assigned
> > > >>> > > > > > > > > > > >> > > > > >> > > partitions even if they are not
> > > >>> all revoked
> > > >>> > > > > > yet. In
> > > >>> > > > > > > > > other
> > > >>> > > > > > > > > > > >> > words, the
> > > >>> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> > > >>> target
> > > >>> > > > > > assignment.
> > > >>> > > > > > > > > This is
> > > >>> > > > > > > > > > > >> > basically
> > > >>> > > > > > > > > > > >> > > > > an
> > > >>> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> > > >>> write all
> > > >>> > > > the
> > > >>> > > > > > > > > changes to
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > log.
> > > >>> > > > > > > > > > > >> > > > > The
> > > >>> > > > > > > > > > > >> > > > > >> > > examples are based on the
> > > >>> persisted state
> > > >>> > > > so I
> > > >>> > > > > > > > > understand
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> > > >>> improve
> > > >>> > > > this in
> > > >>> > > > > > the
> > > >>> > > > > > > > > > > >> > description.
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> > > >>> reuse the
> > > >>> > > > > > protocol
> > > >>> > > > > > > > > with a
> > > >>> > > > > > > > > > > >> > client side
> > > >>> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > > >>> protocol. The
> > > >>> > > > > > assignment
> > > >>> > > > > > > > > is
> > > >>> > > > > > > > > > > >> about
> > > >>> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> > > >>> could
> > > >>> > > > Connect
> > > >>> > > > > > fit
> > > >>> > > > > > > > > into this?
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> > > Best,
> > > >>> > > > > > > > > > > >> > > > > >> > > David
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> > > >>> Sagar <
> > > >>> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > >>> > > > > > > > > > > >> > > > > >> wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just
> > had
> > > >>> minor
> > > >>> > > > > > observations:
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > > >>> section in
> > > >>> > > > Client
> > > >>> > > > > > Side
> > > >>> > > > > > > > > mode
> > > >>> > > > > > > > > > > >> > Assignment
> > > >>> > > > > > > > > > > >> > > > > >> > > process,
> > > >>> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this
> > case,
> > > >>> the
> > > >>> > > > client
> > > >>> > > > > > side
> > > >>> > > > > > > > > assignor
> > > >>> > > > > > > > > > > >> can
> > > >>> > > > > > > > > > > >> > > > > return
> > > >>> > > > > > > > > > > >> > > > > >> an
> > > >>> > > > > > > > > > > >> > > > > >> > > > error to the group
> > > coordinator`.
> > > >>> In this
> > > >>> > > > > > case are
> > > >>> > > > > > > > > you
> > > >>> > > > > > > > > > > >> > referring to
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > > >>> AssignmentError
> > > >>> > > > that's
> > > >>> > > > > > > > > listed down
> > > >>> > > > > > > > > > > >> > towards
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > end?
> > > >>> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
> > > >>> make sense
> > > >>> > > > to
> > > >>> > > > > > > > > mention this
> > > >>> > > > > > > > > > > >> > > > > explicitly
> > > >>> > > > > > > > > > > >> > > > > >> > here?
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> > section,
> > > >>> I have a
> > > >>> > > > > > slight
> > > >>> > > > > > > > > > > >> confusion,
> > > >>> > > > > > > > > > > >> > not
> > > >>> > > > > > > > > > > >> > > > > sure
> > > >>> > > > > > > > > > > >> > > > > >> if
> > > >>> > > > > > > > > > > >> > > > > >> > > > others have the same.
> > Consider
> > > >>> this step:
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> > > >>> coordinator
> > > >>> > > > > > > > > transitions him
> > > >>> > > > > > > > > > > >> to
> > > >>> > > > > > > > > > > >> > epoch
> > > >>> > > > > > > > > > > >> > > > > 3
> > > >>> > > > > > > > > > > >> > > > > >> > > because
> > > >>> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> > revoke.
> > > It
> > > >>> > > > persists
> > > >>> > > > > > the
> > > >>> > > > > > > > > change and
> > > >>> > > > > > > > > > > >> > reply.
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - A
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - B
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - C
> > > >>> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > > (epoch=3)
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - A -
> > partitions=[foo-0]
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - B -
> > partitions=[foo-2]
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - C -
> > partitions=[foo-1]
> > > >>> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > > >>> partitions=[foo-0,
> > > >>> > > > > > foo-1]
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > > >>> partitions=[foo-2]
> > > >>> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > > >>> partitions=[foo-1]
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > > >>> transitions to
> > > >>> > > > epoch 3
> > > >>> > > > > > but
> > > >>> > > > > > > > > cannot
> > > >>> > > > > > > > > > > >> get
> > > >>> > > > > > > > > > > >> > foo-1
> > > >>> > > > > > > > > > > >> > > > > >> yet.
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that
> > member
> > > >>> C can't
> > > >>> > > > get
> > > >>> > > > > > the
> > > >>> > > > > > > > > foo-1
> > > >>> > > > > > > > > > > >> > partition
> > > >>> > > > > > > > > > > >> > > > > yet,
> > > >>> > > > > > > > > > > >> > > > > >> > but
> > > >>> > > > > > > > > > > >> > > > > >> > > > based on the description
> > above,
> > > >>> it seems
> > > >>> > > > it
> > > >>> > > > > > > > > already has
> > > >>> > > > > > > > > > > >> it.
> > > >>> > > > > > > > > > > >> > Do you
> > > >>> > > > > > > > > > > >> > > > > >> > think
> > > >>> > > > > > > > > > > >> > > > > >> > > it
> > > >>> > > > > > > > > > > >> > > > > >> > > > would be better to remove it
> > > and
> > > >>> > > > populate it
> > > >>> > > > > > only
> > > >>> > > > > > > > > when it
> > > >>> > > > > > > > > > > >> > actually
> > > >>> > > > > > > > > > > >> > > > > >> gets
> > > >>> > > > > > > > > > > >> > > > > >> > > it?
> > > >>> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> > > >>> places, so
> > > >>> > > > have
> > > >>> > > > > > I
> > > >>> > > > > > > > > > > >> understood it
> > > >>> > > > > > > > > > > >> > > > > >> > incorrectly
> > > >>> > > > > > > > > > > >> > > > > >> > > ?
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might
> > be
> > > >>> out of
> > > >>> > > > scope
> > > >>> > > > > > of
> > > >>> > > > > > > > > this
> > > >>> > > > > > > > > > > >> > discussion,
> > > >>> > > > > > > > > > > >> > > > > but
> > > >>> > > > > > > > > > > >> > > > > >> > from
> > > >>> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > > >>> probably be
> > > >>> > > > > > running in
> > > >>> > > > > > > > > client
> > > >>> > > > > > > > > > > >> > side
> > > >>> > > > > > > > > > > >> > > > > >> assignor
> > > >>> > > > > > > > > > > >> > > > > >> > > mode
> > > >>> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > > >>> protocol as it
> > > >>> > > > has
> > > >>> > > > > > its
> > > >>> > > > > > > > > own
> > > >>> > > > > > > > > > > >> Custom
> > > >>> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > >>> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00
> > PM
> > > >>> David
> > > >>> > > > Jacot
> > > >>> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is
> > to
> > > >>> move
> > > >>> > > > forward
> > > >>> > > > > > with
> > > >>> > > > > > > > > > > >> > specialized API
> > > >>> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> > > >>> generic API.
> > > >>> > > > For
> > > >>> > > > > > > > > Connect, we
> > > >>> > > > > > > > > > > >> > can apply
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > > >>> reuse/share the
> > > >>> > > > core
> > > >>> > > > > > > > > > > >> > implementation on
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> > > >>> registry, I
> > > >>> > > > > > think
> > > >>> > > > > > > > > that we
> > > >>> > > > > > > > > > > >> > should
> > > >>> > > > > > > > > > > >> > > > > >> consider
> > > >>> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> > > >>> simple
> > > >>> > > > > > > > > membership/leader
> > > >>> > > > > > > > > > > >> > election.
> > > >>> > > > > > > > > > > >> > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > David
> > > >>> > > > > > > > > > > >> > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at
> > 10:22
> > > >>> AM Ismael
> > > >>> > > > > > Juma <
> > > >>> > > > > > > > > > > >> > ismael@juma.me.uk
> > > >>> > > > > > > > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > > >>> java.util.regex.Pattern
> > > >>> > > > vs
> > > >>> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > document the differences
> > in
> > > >>> more
> > > >>> > > > detail
> > > >>> > > > > > before
> > > >>> > > > > > > > > > > >> deciding
> > > >>> > > > > > > > > > > >> > one
> > > >>> > > > > > > > > > > >> > > > > way
> > > >>> > > > > > > > > > > >> > > > > >> or
> > > >>> > > > > > > > > > > >> > > > > >> > > > > another.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > >>> > > > > > > > > java.util.regex.Pattern,
> > > >>> > > > > > > > > > > >> they
> > > >>> > > > > > > > > > > >> > expect
> > > >>> > > > > > > > > > > >> > > > > >> > their
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored.
> > If
> > > >>> we are
> > > >>> > > > doing
> > > >>> > > > > > > > > something
> > > >>> > > > > > > > > > > >> > different,
> > > >>> > > > > > > > > > > >> > > > > >> then
> > > >>> > > > > > > > > > > >> > > > > >> > we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> > > >>> overload
> > > >>> > > > with
> > > >>> > > > > > our own
> > > >>> > > > > > > > > > > >> Pattern
> > > >>> > > > > > > > > > > >> > class
> > > >>> > > > > > > > > > > >> > > > > (I
> > > >>> > > > > > > > > > > >> > > > > >> > > don't
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> > > >>> re2j's at
> > > >>> > > > this
> > > >>> > > > > > > > > point).
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids,
> > any
> > > >>> major new
> > > >>> > > > > > protocol
> > > >>> > > > > > > > > should
> > > >>> > > > > > > > > > > >> > > > > integrate
> > > >>> > > > > > > > > > > >> > > > > >> > fully
> > > >>> > > > > > > > > > > >> > > > > >> > > > > with
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> > > >>> topic
> > > >>> > > > > > recreation case
> > > >>> > > > > > > > > > > >> > correctly.
> > > >>> > > > > > > > > > > >> > > > > That's
> > > >>> > > > > > > > > > > >> > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> > > main
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> > > >>> agree with
> > > >>> > > > > > David
> > > >>> > > > > > > > > that we'd
> > > >>> > > > > > > > > > > >> > want to
> > > >>> > > > > > > > > > > >> > > > > add
> > > >>> > > > > > > > > > > >> > > > > >> > > topic
> > > >>> > > > > > > > > > > >> > > > > >> > > > > ids
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
> > > >>> that don't
> > > >>> > > > > > have it
> > > >>> > > > > > > > > yet and
> > > >>> > > > > > > > > > > >> > that we
> > > >>> > > > > > > > > > > >> > > > > can
> > > >>> > > > > > > > > > > >> > > > > >> > > probably
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> > > >>> versus adding
> > > >>> > > > new
> > > >>> > > > > > APIs
> > > >>> > > > > > > > > to the
> > > >>> > > > > > > > > > > >> > Java
> > > >>> > > > > > > > > > > >> > > > > >> Consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new
> > > APIs
> > > >>> is
> > > >>> > > > required
> > > >>> > > > > > for
> > > >>> > > > > > > > > > > >> reasonable
> > > >>> > > > > > > > > > > >> > > > > >> semantics).
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> > > about
> > > >>> the
> > > >>> > > > > > coordinator
> > > >>> > > > > > > > > storing
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> configs.
> > > >>> > > > > > > > > > > >> > > > > >> > > It's
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to
> > be
> > > >>> > > > centralized
> > > >>> > > > > > in the
> > > >>> > > > > > > > > > > >> metadata
> > > >>> > > > > > > > > > > >> > log for
> > > >>> > > > > > > > > > > >> > > > > >> > > various
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > > >>> visibility,
> > > >>> > > > > > consistency,
> > > >>> > > > > > > > > > > >> etc.).
> > > >>> > > > > > > > > > > >> > > > > >> Similarly, I
> > > >>> > > > > > > > > > > >> > > > > >> > > am
> > > >>> > > > > > > > > > > >> > > > > >> > > > > not
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> > > >>> deleting
> > > >>> > > > > > configs in a
> > > >>> > > > > > > > > way
> > > >>> > > > > > > > > > > >> that
> > > >>> > > > > > > > > > > >> > they
> > > >>> > > > > > > > > > > >> > > > > >> cannot
> > > >>> > > > > > > > > > > >> > > > > >> > > be
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> > property
> > > >>> for modern
> > > >>> > > > > > systems
> > > >>> > > > > > > > > is to
> > > >>> > > > > > > > > > > >> > minimize
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > number
> > > >>> > > > > > > > > > > >> > > > > >> > > > > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> > > >>> scenarios.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at
> > > 3:47
> > > >>> PM David
> > > >>> > > > > > Jacot
> > > >>> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> > > >>> answers are
> > > >>> > > > below:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > path,
> > > >>> especially
> > > >>> > > > > > the last
> > > >>> > > > > > > > > step
> > > >>> > > > > > > > > > > >> when
> > > >>> > > > > > > > > > > >> > > > > clients
> > > >>> > > > > > > > > > > >> > > > > >> > > flip the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > >>> protocol, in
> > > >>> > > > which
> > > >>> > > > > > we
> > > >>> > > > > > > > > would
> > > >>> > > > > > > > > > > >> have a
> > > >>> > > > > > > > > > > >> > > > > window
> > > >>> > > > > > > > > > > >> > > > > >> > where
> > > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > > old
> > > >>> > > > protocols /
> > > >>> > > > > > rpcs
> > > >>> > > > > > > > > are
> > > >>> > > > > > > > > > > >> used
> > > >>> > > > > > > > > > > >> > by
> > > >>> > > > > > > > > > > >> > > > > >> members
> > > >>> > > > > > > > > > > >> > > > > >> > of
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > >>> coordinator could
> > > >>> > > > > > "mimic"
> > > >>> > > > > > > > > the old
> > > >>> > > > > > > > > > > >> > behavior
> > > >>> > > > > > > > > > > >> > > > > >> while
> > > >>> > > > > > > > > > > >> > > > > >> > > > > using the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > >>> something we
> > > >>> > > > need to
> > > >>> > > > > > > > > present
> > > >>> > > > > > > > > > > >> about.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> > published a
> > > >>> new
> > > >>> > > > version
> > > >>> > > > > > of KIP
> > > >>> > > > > > > > > which
> > > >>> > > > > > > > > > > >> > includes
> > > >>> > > > > > > > > > > >> > > > > >> more
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See
> > > the
> > > >>> > > > "Supporting
> > > >>> > > > > > > > > Online
> > > >>> > > > > > > > > > > >> > Consumer
> > > >>> > > > > > > > > > > >> > > > > Group
> > > >>> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > > >>> > > > Deprecation,
> > > >>> > > > > > and
> > > >>> > > > > > > > > Migration
> > > >>> > > > > > > > > > > >> > Plan". I
> > > >>> > > > > > > > > > > >> > > > > >> think
> > > >>> > > > > > > > > > > >> > > > > >> > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> > through a
> > > >>> few cases
> > > >>> > > > > > now but
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > overall idea
> > > >>> > > > > > > > > > > >> > > > > >> and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > > >>> understandable.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > > >>> ids. So
> > > >>> > > > far as
> > > >>> > > > > > > > > KIP-516 the
> > > >>> > > > > > > > > > > >> > topic ids
> > > >>> > > > > > > > > > > >> > > > > >> are
> > > >>> > > > > > > > > > > >> > > > > >> > > only
> > > >>> > > > > > > > > > > >> > > > > >> > > > > used
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > admin
> > > >>> client,
> > > >>> > > > but
> > > >>> > > > > > they
> > > >>> > > > > > > > > are not
> > > >>> > > > > > > > > > > >> > exposed
> > > >>> > > > > > > > > > > >> > > > > via
> > > >>> > > > > > > > > > > >> > > > > >> any
> > > >>> > > > > > > > > > > >> > > > > >> > > > > public
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > > >>> think the
> > > >>> > > > > > question is,
> > > >>> > > > > > > > > first
> > > >>> > > > > > > > > > > >> > should we
> > > >>> > > > > > > > > > > >> > > > > >> let
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > maintaining
> > > >>> the
> > > >>> > > > names
> > > >>> > > > > > -> ids
> > > >>> > > > > > > > > mapping
> > > >>> > > > > > > > > > > >> > itself
> > > >>> > > > > > > > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > fully
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > > >>> existing
> > > >>> > > > RPCs
> > > >>> > > > > > and the
> > > >>> > > > > > > > > new
> > > >>> > > > > > > > > > > >> RPCs
> > > >>> > > > > > > > > > > >> > with
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > topic
> > > >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > > ever
> > > >>> consider
> > > >>> > > > > > exposing
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > topic ids
> > > >>> > > > > > > > > > > >> > > > > in
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > > (both
> > > >>> > > > > > > > > subscribe/assign, as
> > > >>> > > > > > > > > > > >> well
> > > >>> > > > > > > > > > > >> > as in
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > like
> > > >>> topic
> > > >>> > > > > > > > > > > >> > deletion-and-recreation).
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we
> > would
> > > >>> include
> > > >>> > > > > > converting
> > > >>> > > > > > > > > all
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > offsets
> > > >>> > > > > > > > > > > >> > > > > >> > > related
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids
> > > in
> > > >>> this
> > > >>> > > > KIP,
> > > >>> > > > > > the
> > > >>> > > > > > > > > consumer
> > > >>> > > > > > > > > > > >> > would be
> > > >>> > > > > > > > > > > >> > > > > >> able
> > > >>> > > > > > > > > > > >> > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with
> > topic
> > > >>> ids. That
> > > >>> > > > > > being
> > > >>> > > > > > > > > said, it
> > > >>> > > > > > > > > > > >> > still has
> > > >>> > > > > > > > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > provide
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> > > >>> various APIs so
> > > >>> > > > > > having a
> > > >>> > > > > > > > > > > >> mapping
> > > >>> > > > > > > > > > > >> > in the
> > > >>> > > > > > > > > > > >> > > > > >> > > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a
> > strong
> > > >>> opinion on
> > > >>> > > > > > this.
> > > >>> > > > > > > > > Here I
> > > >>> > > > > > > > > > > >> > wonder if
> > > >>> > > > > > > > > > > >> > > > > >> this
> > > >>> > > > > > > > > > > >> > > > > >> > > goes
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of
> > this
> > > >>> KIP. I
> > > >>> > > > would
> > > >>> > > > > > rather
> > > >>> > > > > > > > > focus
> > > >>> > > > > > > > > > > >> on
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> > internals
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> > consider
> > > >>> this
> > > >>> > > > > > separately if
> > > >>> > > > > > > > > we see
> > > >>> > > > > > > > > > > >> > value in
> > > >>> > > > > > > > > > > >> > > > > >> doing
> > > >>> > > > > > > > > > > >> > > > > >> > > it.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> > > >>> point about
> > > >>> > > > > > using
> > > >>> > > > > > > > > topic ids
> > > >>> > > > > > > > > > > >> > in the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> ConsumerGroupHeartbeatRequest, I
> > > >>> > > > think
> > > >>> > > > > > that
> > > >>> > > > > > > > > there
> > > >>> > > > > > > > > > > >> is
> > > >>> > > > > > > > > > > >> > one
> > > >>> > > > > > > > > > > >> > > > > >> > advantage
> > > >>> > > > > > > > > > > >> > > > > >> > > in
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> > > consumer
> > > >>> will
> > > >>> > > > have
> > > >>> > > > > > the
> > > >>> > > > > > > > > > > >> opportunity
> > > >>> > > > > > > > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > >> validate
> > > >>> > > > > > > > > > > >> > > > > >> > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists
> > before
> > > >>> passing
> > > >>> > > > them
> > > >>> > > > > > into
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> group
> > > >>> > > > > > > > > > > >> > > > > rebalance
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously,
> > the
> > > >>> > > > coordinator
> > > >>> > > > > > will
> > > >>> > > > > > > > > also
> > > >>> > > > > > > > > > > >> notice
> > > >>> > > > > > > > > > > >> > it but
> > > >>> > > > > > > > > > > >> > > > > >> it
> > > >>> > > > > > > > > > > >> > > > > >> > > does
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way
> > to
> > > >>> reject an
> > > >>> > > > > > invalid
> > > >>> > > > > > > > > topic in
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> response.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > David
> > > >>> on all
> > > >>> > > > other
> > > >>> > > > > > minor
> > > >>> > > > > > > > > > > >> questions
> > > >>> > > > > > > > > > > >> > > > > except
> > > >>> > > > > > > > > > > >> > > > > >> for
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > > >>> question:
> > > >>> > > > > > personally
> > > >>> > > > > > > > > I think
> > > >>> > > > > > > > > > > >> > it's not
> > > >>> > > > > > > > > > > >> > > > > >> > > necessary
> > > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > subscribe
> > > >>> API with
> > > >>> > > > > > Pattern,
> > > >>> > > > > > > > > but
> > > >>> > > > > > > > > > > >> > instead we
> > > >>> > > > > > > > > > > >> > > > > >> still
> > > >>> > > > > > > > > > > >> > > > > >> > > use
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > documenting
> > > >>> that our
> > > >>> > > > > > > > > subscription may
> > > >>> > > > > > > > > > > >> be
> > > >>> > > > > > > > > > > >> > > > > >> rejected by
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > incompatible
> > > >>> case is a
> > > >>> > > > > > very rare
> > > >>> > > > > > > > > > > >> scenario
> > > >>> > > > > > > > > > > >> > I felt
> > > >>> > > > > > > > > > > >> > > > > >> > using
> > > >>> > > > > > > > > > > >> > > > > >> > > an
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > > based
> > > >>> > > > > > subscription may
> > > >>> > > > > > > > > be
> > > >>> > > > > > > > > > > >> more
> > > >>> > > > > > > > > > > >> > > > > >> vulnerable
> > > >>> > > > > > > > > > > >> > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > various
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have
> > > to
> > > >>> look at
> > > >>> > > > the
> > > >>> > > > > > > > > differences
> > > >>> > > > > > > > > > > >> > between
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > two
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > > >>> understand the
> > > >>> > > > > > potential
> > > >>> > > > > > > > > issues.
> > > >>> > > > > > > > > > > >> My
> > > >>> > > > > > > > > > > >> > > > > >> > > understanding is
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all
> > > >>> the basic
> > > >>> > > > > > regular
> > > >>> > > > > > > > > > > >> > expressions. The
> > > >>> > > > > > > > > > > >> > > > > >> > > differences
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> > > mainly
> > > >>> about
> > > >>> > > > the
> > > >>> > > > > > various
> > > >>> > > > > > > > > > > >> character
> > > >>> > > > > > > > > > > >> > > > > >> classes. I
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other
> > people
> > > >>> think
> > > >>> > > > about
> > > >>> > > > > > this.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> > > >>> 11:28 PM
> > > >>> > > > > > Guozhang
> > > >>> > > > > > > > > Wang <
> > > >>> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > >>> > > > > > > > > > > >> > > > > >> > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think
> > > on
> > > >>> the high
> > > >>> > > > > > level
> > > >>> > > > > > > > > there are
> > > >>> > > > > > > > > > > >> > two meta
> > > >>> > > > > > > > > > > >> > > > > >> > > points we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > need
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit
> > > more:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> > path,
> > > >>> especially
> > > >>> > > > > > the last
> > > >>> > > > > > > > > step
> > > >>> > > > > > > > > > > >> when
> > > >>> > > > > > > > > > > >> > > > > clients
> > > >>> > > > > > > > > > > >> > > > > >> > > flip the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > > >>> protocol, in
> > > >>> > > > which
> > > >>> > > > > > we
> > > >>> > > > > > > > > would
> > > >>> > > > > > > > > > > >> have a
> > > >>> > > > > > > > > > > >> > > > > window
> > > >>> > > > > > > > > > > >> > > > > >> > where
> > > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > > old
> > > >>> > > > protocols /
> > > >>> > > > > > rpcs
> > > >>> > > > > > > > > are
> > > >>> > > > > > > > > > > >> used
> > > >>> > > > > > > > > > > >> > by
> > > >>> > > > > > > > > > > >> > > > > >> members
> > > >>> > > > > > > > > > > >> > > > > >> > of
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > > >>> coordinator could
> > > >>> > > > > > "mimic"
> > > >>> > > > > > > > > the old
> > > >>> > > > > > > > > > > >> > behavior
> > > >>> > > > > > > > > > > >> > > > > >> while
> > > >>> > > > > > > > > > > >> > > > > >> > > > > using the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > > >>> something we
> > > >>> > > > need to
> > > >>> > > > > > > > > present
> > > >>> > > > > > > > > > > >> about.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > > >>> ids. So
> > > >>> > > > far as
> > > >>> > > > > > > > > KIP-516 the
> > > >>> > > > > > > > > > > >> > topic ids
> > > >>> > > > > > > > > > > >> > > > > >> are
> > > >>> > > > > > > > > > > >> > > > > >> > > only
> > > >>> > > > > > > > > > > >> > > > > >> > > > > used
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> > admin
> > > >>> client,
> > > >>> > > > but
> > > >>> > > > > > they
> > > >>> > > > > > > > > are not
> > > >>> > > > > > > > > > > >> > exposed
> > > >>> > > > > > > > > > > >> > > > > via
> > > >>> > > > > > > > > > > >> > > > > >> any
> > > >>> > > > > > > > > > > >> > > > > >> > > > > public
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > > >>> think the
> > > >>> > > > > > question is,
> > > >>> > > > > > > > > first
> > > >>> > > > > > > > > > > >> > should we
> > > >>> > > > > > > > > > > >> > > > > >> let
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > > maintaining
> > > >>> the
> > > >>> > > > names
> > > >>> > > > > > -> ids
> > > >>> > > > > > > > > mapping
> > > >>> > > > > > > > > > > >> > itself
> > > >>> > > > > > > > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > fully
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > > >>> existing
> > > >>> > > > RPCs
> > > >>> > > > > > and the
> > > >>> > > > > > > > > new
> > > >>> > > > > > > > > > > >> RPCs
> > > >>> > > > > > > > > > > >> > with
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > topic
> > > >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > > ever
> > > >>> consider
> > > >>> > > > > > exposing
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > topic ids
> > > >>> > > > > > > > > > > >> > > > > in
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > > (both
> > > >>> > > > > > > > > subscribe/assign, as
> > > >>> > > > > > > > > > > >> well
> > > >>> > > > > > > > > > > >> > as in
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> > like
> > > >>> topic
> > > >>> > > > > > > > > > > >> > deletion-and-recreation).
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> > David
> > > >>> on all
> > > >>> > > > other
> > > >>> > > > > > minor
> > > >>> > > > > > > > > > > >> questions
> > > >>> > > > > > > > > > > >> > > > > except
> > > >>> > > > > > > > > > > >> > > > > >> for
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > > >>> question:
> > > >>> > > > > > personally
> > > >>> > > > > > > > > I think
> > > >>> > > > > > > > > > > >> > it's not
> > > >>> > > > > > > > > > > >> > > > > >> > > necessary
> > > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> > subscribe
> > > >>> API with
> > > >>> > > > > > Pattern,
> > > >>> > > > > > > > > but
> > > >>> > > > > > > > > > > >> > instead we
> > > >>> > > > > > > > > > > >> > > > > >> still
> > > >>> > > > > > > > > > > >> > > > > >> > > use
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> > documenting
> > > >>> that our
> > > >>> > > > > > > > > subscription may
> > > >>> > > > > > > > > > > >> be
> > > >>> > > > > > > > > > > >> > > > > >> rejected by
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> > incompatible
> > > >>> case is a
> > > >>> > > > > > very rare
> > > >>> > > > > > > > > > > >> scenario
> > > >>> > > > > > > > > > > >> > I felt
> > > >>> > > > > > > > > > > >> > > > > >> > using
> > > >>> > > > > > > > > > > >> > > > > >> > > an
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > > based
> > > >>> > > > > > subscription may
> > > >>> > > > > > > > > be
> > > >>> > > > > > > > > > > >> more
> > > >>> > > > > > > > > > > >> > > > > >> vulnerable
> > > >>> > > > > > > > > > > >> > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > various
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022
> > at
> > > >>> 5:23 AM
> > > >>> > > > > > David Jacot
> > > >>> > > > > > > > > > > >> > > > > >> > > > >
> > <djacot@confluent.io.invalid
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > > >>> feedback. Let
> > > >>> > > > me
> > > >>> > > > > > answer
> > > >>> > > > > > > > > your
> > > >>> > > > > > > > > > > >> > questions
> > > >>> > > > > > > > > > > >> > > > > >> > inline.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> > > >>> premature to
> > > >>> > > > > > talk about
> > > >>> > > > > > > > > > > >> target
> > > >>> > > > > > > > > > > >> > > > > versions
> > > >>> > > > > > > > > > > >> > > > > >> for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > > >>> existing group
> > > >>> > > > > > protocol.
> > > >>> > > > > > > > > > > >> Unlike
> > > >>> > > > > > > > > > > >> > KRaft,
> > > >>> > > > > > > > > > > >> > > > > >> this
> > > >>> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol
> > and
> > > >>> hence
> > > >>> > > > > > > > > deprecation/removal
> > > >>> > > > > > > > > > > >> > will be
> > > >>> > > > > > > > > > > >> > > > > >> > heavily
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > > >>> applications
> > > >>> > > > migrate
> > > >>> > > > > > to
> > > >>> > > > > > > > > the new
> > > >>> > > > > > > > > > > >> > protocol.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I
> > > >>> will
> > > >>> > > > remove it.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says
> > we
> > > >>> intend to
> > > >>> > > > > > release
> > > >>> > > > > > > > > this in
> > > >>> > > > > > > > > > > >> > 4.x, but
> > > >>> > > > > > > > > > > >> > > > > it
> > > >>> > > > > > > > > > > >> > > > > >> > > wasn't
> > > >>> > > > > > > > > > > >> > > > > >> > > > > made
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> > > >>> that as a
> > > >>> > > > way to
> > > >>> > > > > > > > > estimate
> > > >>> > > > > > > > > > > >> when
> > > >>> > > > > > > > > > > >> > we'd
> > > >>> > > > > > > > > > > >> > > > > >> > > deprecate
> > > >>> > > > > > > > > > > >> > > > > >> > > > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> > protocol,
> > > >>> I also
> > > >>> > > > > > suggest
> > > >>> > > > > > > > > removing
> > > >>> > > > > > > > > > > >> > this part.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> > > >>> reasoning. As
> > > >>> > > > > > > > > explained, I
> > > >>> > > > > > > > > > > >> plan
> > > >>> > > > > > > > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > >> rewrite
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> > > >>> while we
> > > >>> > > > > > implement
> > > >>> > > > > > > > > the new
> > > >>> > > > > > > > > > > >> > protocol.
> > > >>> > > > > > > > > > > >> > > > > >> This
> > > >>> > > > > > > > > > > >> > > > > >> > > means
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> > > >>> will be
> > > >>> > > > slightly
> > > >>> > > > > > > > > different
> > > >>> > > > > > > > > > > >> > (e.g.
> > > >>> > > > > > > > > > > >> > > > > >> threading
> > > >>> > > > > > > > > > > >> > > > > >> > > > > model).
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted
> > > to
> > > >>> tighten
> > > >>> > > > the
> > > >>> > > > > > > > > switch from
> > > >>> > > > > > > > > > > >> > the old
> > > >>> > > > > > > > > > > >> > > > > >> group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the
> > > new
> > > >>> group
> > > >>> > > > > > coordinator
> > > >>> > > > > > > > > to a
> > > >>> > > > > > > > > > > >> > major
> > > >>> > > > > > > > > > > >> > > > > >> release.
> > > >>> > > > > > > > > > > >> > > > > >> > > The
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would
> > be
> > > >>> to use a
> > > >>> > > > flag
> > > >>> > > > > > to do
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > switch
> > > >>> > > > > > > > > > > >> > > > > >> instead
> > > >>> > > > > > > > > > > >> > > > > >> > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > relying
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> > > >>> upgrade.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to
> > flesh
> > > >>> out the
> > > >>> > > > > > details of
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > migration
> > > >>> > > > > > > > > > > >> > > > > >> story.
> > > >>> > > > > > > > > > > >> > > > > >> > > It
> > > >>> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we
> > > will
> > > >>> support
> > > >>> > > > > > online
> > > >>> > > > > > > > > > > >> migrations.
> > > >>> > > > > > > > > > > >> > Is
> > > >>> > > > > > > > > > > >> > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > correct?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > We
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> > > >>> detail. It
> > > >>> > > > could
> > > >>> > > > > > also
> > > >>> > > > > > > > > be done
> > > >>> > > > > > > > > > > >> > as a
> > > >>> > > > > > > > > > > >> > > > > >> separate
> > > >>> > > > > > > > > > > >> > > > > >> > > KIP,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > if
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> > support
> > > >>> online
> > > >>> > > > > > migrations
> > > >>> > > > > > > > > for the
> > > >>> > > > > > > > > > > >> > group.
> > > >>> > > > > > > > > > > >> > > > > That
> > > >>> > > > > > > > > > > >> > > > > >> > > means
> > > >>> > > > > > > > > > > >> > > > > >> > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the
> > old
> > > >>> protocol
> > > >>> > > > > > will be
> > > >>> > > > > > > > > able to
> > > >>> > > > > > > > > > > >> > switch to
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> > > explain
> > > >>> how that
> > > >>> > > > > > will work
> > > >>> > > > > > > > > > > >> though.
> > > >>> > > > > > > > > > > >> > It is
> > > >>> > > > > > > > > > > >> > > > > >> > > basically a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must
> > > be
> > > >>> > > > upgraded or
> > > >>> > > > > > > > > rolled to a
> > > >>> > > > > > > > > > > >> > software
> > > >>> > > > > > > > > > > >> > > > > >> > > supporting
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > > coordinator.
> > > >>> Both the
> > > >>> > > > > > old and
> > > >>> > > > > > > > > the new
> > > >>> > > > > > > > > > > >> > > > > >> coordinator
> > > >>> > > > > > > > > > > >> > > > > >> > > will
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > > >>> protocol and
> > > >>> > > > rely
> > > >>> > > > > > on the
> > > >>> > > > > > > > > same
> > > >>> > > > > > > > > > > >> > persisted
> > > >>> > > > > > > > > > > >> > > > > >> > > metadata so
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> > > >>> together. This
> > > >>> > > > point
> > > >>> > > > > > is an
> > > >>> > > > > > > > > > > >> offline
> > > >>> > > > > > > > > > > >> > > > > >> migration.
> > > >>> > > > > > > > > > > >> > > > > >> > We
> > > >>> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > > >>> because it
> > > >>> > > > would
> > > >>> > > > > > require
> > > >>> > > > > > > > > > > >> > shutting down
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > current
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > > >>> starting up the
> > > >>> > > > > > new one
> > > >>> > > > > > > > > and
> > > >>> > > > > > > > > > > >> that
> > > >>> > > > > > > > > > > >> > would
> > > >>> > > > > > > > > > > >> > > > > >> cause
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > > >>> metadata
> > > >>> > > > > > version/IBP
> > > >>> > > > > > > > > must be
> > > >>> > > > > > > > > > > >> > upgraded
> > > >>> > > > > > > > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> X
> > > >>> > > > > > > > > > > >> > > > > >> > in
> > > >>> > > > > > > > > > > >> > > > > >> > > > > order
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> > > >>> protocol.
> > > >>> > > > This
> > > >>> > > > > > cannot
> > > >>> > > > > > > > > be done
> > > >>> > > > > > > > > > > >> > before
> > > >>> > > > > > > > > > > >> > > > > 1)
> > > >>> > > > > > > > > > > >> > > > > >> is
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because
> > > the
> > > >>> old
> > > >>> > > > > > coordinator
> > > >>> > > > > > > > > doesn't
> > > >>> > > > > > > > > > > >> > support
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers
> > must
> > > >>> be
> > > >>> > > > upgraded
> > > >>> > > > > > to a
> > > >>> > > > > > > > > version
> > > >>> > > > > > > > > > > >> > > > > supporting
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > online
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must
> > have
> > > >>> KIP-792).
> > > >>> > > > If
> > > >>> > > > > > the
> > > >>> > > > > > > > > consumer
> > > >>> > > > > > > > > > > >> is
> > > >>> > > > > > > > > > > >> > > > > already
> > > >>> > > > > > > > > > > >> > > > > >> > > there.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be
> > done
> > > >>> at this
> > > >>> > > > point.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers
> > must
> > > >>> be rolled
> > > >>> > > > > > with the
> > > >>> > > > > > > > > > > >> feature
> > > >>> > > > > > > > > > > >> > flag
> > > >>> > > > > > > > > > > >> > > > > >> turned
> > > >>> > > > > > > > > > > >> > > > > >> > > on.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > The
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> > > >>> automatically
> > > >>> > > > > > > > > converted when
> > > >>> > > > > > > > > > > >> > the first
> > > >>> > > > > > > > > > > >> > > > > >> > > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> > > protocol
> > > >>> joins
> > > >>> > > > the
> > > >>> > > > > > group.
> > > >>> > > > > > > > > While
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > members
> > > >>> > > > > > > > > > > >> > > > > >> > > using the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> > > being
> > > >>> > > > upgraded,
> > > >>> > > > > > the old
> > > >>> > > > > > > > > > > >> protocol
> > > >>> > > > > > > > > > > >> > is
> > > >>> > > > > > > > > > > >> > > > > >> proxied
> > > >>> > > > > > > > > > > >> > > > > >> > > into
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all
> > of
> > > >>> this in
> > > >>> > > > the
> > > >>> > > > > > KIP.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy
> > that
> > > >>> we are
> > > >>> > > > > > pushing the
> > > >>> > > > > > > > > pattern
> > > >>> > > > > > > > > > > >> > > > > >> subscriptions
> > > >>> > > > > > > > > > > >> > > > > >> > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
> > > >>> there
> > > >>> > > > could be
> > > >>> > > > > > some
> > > >>> > > > > > > > > tricky
> > > >>> > > > > > > > > > > >> > > > > >> compatibility
> > > >>> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism
> > > for
> > > >>> users to
> > > >>> > > > > > detect
> > > >>> > > > > > > > > that
> > > >>> > > > > > > > > > > >> they
> > > >>> > > > > > > > > > > >> > need to
> > > >>> > > > > > > > > > > >> > > > > >> > update
> > > >>> > > > > > > > > > > >> > > > > >> > > > > their
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching
> > to
> > > >>> the new
> > > >>> > > > > > protocol?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a
> > > >>> bit more
> > > >>> > > > > > optimistic
> > > >>> > > > > > > > > than
> > > >>> > > > > > > > > > > >> you
> > > >>> > > > > > > > > > > >> > on this
> > > >>> > > > > > > > > > > >> > > > > >> > > point. I
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > > >>> majority of
> > > >>> > > > the
> > > >>> > > > > > cases
> > > >>> > > > > > > > > are
> > > >>> > > > > > > > > > > >> simple
> > > >>> > > > > > > > > > > >> > > > > regexes
> > > >>> > > > > > > > > > > >> > > > > >> > which
> > > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> > > >>> engine. The
> > > >>> > > > > > > > > coordinator will
> > > >>> > > > > > > > > > > >> > verify
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > regex
> > > >>> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> > > >>> consumer if the
> > > >>> > > > > > regex is
> > > >>> > > > > > > > > not
> > > >>> > > > > > > > > > > >> valid.
> > > >>> > > > > > > > > > > >> > > > > Coming
> > > >>> > > > > > > > > > > >> > > > > >> > back
> > > >>> > > > > > > > > > > >> > > > > >> > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in
> > > the
> > > >>> worst
> > > >>> > > > case,
> > > >>> > > > > > the
> > > >>> > > > > > > > > first
> > > >>> > > > > > > > > > > >> > upgraded
> > > >>> > > > > > > > > > > >> > > > > >> > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > joining
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> > > >>> rejected.
> > > >>> > > > This
> > > >>> > > > > > should
> > > >>> > > > > > > > > be used
> > > >>> > > > > > > > > > > >> > as the
> > > >>> > > > > > > > > > > >> > > > > >> last
> > > >>> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> > customers
> > > >>> to
> > > >>> > > > validate
> > > >>> > > > > > their
> > > >>> > > > > > > > > regex
> > > >>> > > > > > > > > > > >> > before
> > > >>> > > > > > > > > > > >> > > > > >> > upgrading
> > > >>> > > > > > > > > > > >> > > > > >> > > > > their
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to
> > test
> > > >>> them with
> > > >>> > > > > > another
> > > >>> > > > > > > > > group.
> > > >>> > > > > > > > > > > >> For
> > > >>> > > > > > > > > > > >> > > > > >> instance,
> > > >>> > > > > > > > > > > >> > > > > >> > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> > > >>> pre-prod
> > > >>> > > > > > environment.
> > > >>> > > > > > > > > > > >> Another
> > > >>> > > > > > > > > > > >> > way
> > > >>> > > > > > > > > > > >> > > > > >> would be
> > > >>> > > > > > > > > > > >> > > > > >> > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > > >>> consumer-group tool
> > > >>> > > > to
> > > >>> > > > > > > > > provide a
> > > >>> > > > > > > > > > > >> regex
> > > >>> > > > > > > > > > > >> > > > > >> validation
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would
> > this
> > > >>> be
> > > >>> > > > enough in
> > > >>> > > > > > your
> > > >>> > > > > > > > > > > >> opinion?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the
> > > >>> last
> > > >>> > > > question,
> > > >>> > > > > > will
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> Java
> > > >>> > > > > > > > > > > >> > client
> > > >>> > > > > > > > > > > >> > > > > >> allow
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > > >>> current regex
> > > >>> > > > > > engine for
> > > >>> > > > > > > > > > > >> > compatibility
> > > >>> > > > > > > > > > > >> > > > > >> > > reasons?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > For
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy
> > to
> > > >>> keep using
> > > >>> > > > > > client
> > > >>> > > > > > > > > based
> > > >>> > > > > > > > > > > >> > regex at
> > > >>> > > > > > > > > > > >> > > > > >> first
> > > >>> > > > > > > > > > > >> > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > >> > > keep
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple
> > > >>> and then
> > > >>> > > > > > migrate to
> > > >>> > > > > > > > > server
> > > >>> > > > > > > > > > > >> > based
> > > >>> > > > > > > > > > > >> > > > > >> regexes
> > > >>> > > > > > > > > > > >> > > > > >> > > as a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> > > >>> point but I
> > > >>> > > > am
> > > >>> > > > > > > > > concerned that
> > > >>> > > > > > > > > > > >> > this
> > > >>> > > > > > > > > > > >> > > > > would
> > > >>> > > > > > > > > > > >> > > > > >> > > allow
> > > >>> > > > > > > > > > > >> > > > > >> > > > > users
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
> > > >>> this mode.
> > > >>> > > > That
> > > >>> > > > > > > > > would go
> > > >>> > > > > > > > > > > >> > against our
> > > >>> > > > > > > > > > > >> > > > > >> goal
> > > >>> > > > > > > > > > > >> > > > > >> > > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> > > client
> > > >>> because
> > > >>> > > > we
> > > >>> > > > > > would
> > > >>> > > > > > > > > have to
> > > >>> > > > > > > > > > > >> > continue
> > > >>> > > > > > > > > > > >> > > > > >> > > monitoring
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> > > >>> client
> > > >>> > > > side. I
> > > >>> > > > > > would
> > > >>> > > > > > > > > rather
> > > >>> > > > > > > > > > > >> > not do
> > > >>> > > > > > > > > > > >> > > > > >> this.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say
> > that
> > > >>> the group
> > > >>> > > > > > > > > coordinator
> > > >>> > > > > > > > > > > >> will
> > > >>> > > > > > > > > > > >> > be
> > > >>> > > > > > > > > > > >> > > > > >> > > responsible for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> > configurations
> > > >>> and that
> > > >>> > > > the
> > > >>> > > > > > > > > > > >> configurations
> > > >>> > > > > > > > > > > >> > will be
> > > >>> > > > > > > > > > > >> > > > > >> > > deleted
> > > >>> > > > > > > > > > > >> > > > > >> > > > > when
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
> > > >>> Will a
> > > >>> > > > > > transition to
> > > >>> > > > > > > > > DEAD
> > > >>> > > > > > > > > > > >> > trigger
> > > >>> > > > > > > > > > > >> > > > > >> > deletion
> > > >>> > > > > > > > > > > >> > > > > >> > > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > >>> > > > configurations
> > > >>> > > > > > will be
> > > >>> > > > > > > > > > > >> deleted
> > > >>> > > > > > > > > > > >> > when
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > group is
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > > >>> together.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the
> > choice
> > > >>> to store
> > > >>> > > > the
> > > >>> > > > > > > > > configs in
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > group
> > > >>> > > > > > > > > > > >> > > > > >> > > coordinator
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list
> > all
> > > >>> cluster
> > > >>> > > > > > configs and
> > > >>> > > > > > > > > their
> > > >>> > > > > > > > > > > >> > values?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so.
> > The
> > > >>> group
> > > >>> > > > > > > > > configurations are
> > > >>> > > > > > > > > > > >> > overrides
> > > >>> > > > > > > > > > > >> > > > > >> of
> > > >>> > > > > > > > > > > >> > > > > >> > > cluster
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you
> > want
> > > >>> to know
> > > >>> > > > all
> > > >>> > > > > > the
> > > >>> > > > > > > > > overrides
> > > >>> > > > > > > > > > > >> > though,
> > > >>> > > > > > > > > > > >> > > > > you
> > > >>> > > > > > > > > > > >> > > > > >> > > would
> > > >>> > > > > > > > > > > >> > > > > >> > > > > have
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the
> > group
> > > >>> > > > coordinators.
> > > >>> > > > > > You
> > > >>> > > > > > > > > cannot
> > > >>> > > > > > > > > > > >> > rely on
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > metadata
> > > >>> > > > > > > > > > > >> > > > > >> > > > > log
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> > > someone
> > > >>> > > > configure a
> > > >>> > > > > > group
> > > >>> > > > > > > > > before
> > > >>> > > > > > > > > > > >> > starting
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > > >>> allowing the
> > > >>> > > > > > explicit
> > > >>> > > > > > > > > creation
> > > >>> > > > > > > > > > > >> of
> > > >>> > > > > > > > > > > >> > > > > groups?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> > > >>> decoupled
> > > >>> > > > from
> > > >>> > > > > > the
> > > >>> > > > > > > > > group
> > > >>> > > > > > > > > > > >> > lifecycle.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will
> > > be
> > > >>> > > > > > automatically
> > > >>> > > > > > > > > created in
> > > >>> > > > > > > > > > > >> > this
> > > >>> > > > > > > > > > > >> > > > > case.
> > > >>> > > > > > > > > > > >> > > > > >> > > However,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
> > > >>> lost after
> > > >>> > > > the
> > > >>> > > > > > > > > retention
> > > >>> > > > > > > > > > > >> > period of
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > >>> > > > Consumer.subscribe
> > > >>> > > > > > method
> > > >>> > > > > > > > > for the
> > > >>> > > > > > > > > > > >> > Java
> > > >>> > > > > > > > > > > >> > > > > >> client
> > > >>> > > > > > > > > > > >> > > > > >> > > still
> > > >>> > > > > > > > > > > >> > > > > >> > > > > take
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >>> `java.util.regex.Pattern` of
> > > >>> > > > do
> > > >>> > > > > > we
> > > >>> > > > > > > > > have to
> > > >>> > > > > > > > > > > >> > introduce
> > > >>> > > > > > > > > > > >> > > > > an
> > > >>> > > > > > > > > > > >> > > > > >> > > overload?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> > > >>> question. I
> > > >>> > > > > > forgot
> > > >>> > > > > > > > > about
> > > >>> > > > > > > > > > > >> that
> > > >>> > > > > > > > > > > >> > one.
> > > >>> > > > > > > > > > > >> > > > > As
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> `java.util.regex.Pattern` is
> > > >>> > > > not
> > > >>> > > > > > fully
> > > >>> > > > > > > > > > > >> compatible
> > > >>> > > > > > > > > > > >> > with
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > engine
> > > >>> > > > > > > > > > > >> > > > > >> > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> > > >>> might be
> > > >>> > > > better
> > > >>> > > > > > to
> > > >>> > > > > > > > > deprecate
> > > >>> > > > > > > > > > > >> > it and
> > > >>> > > > > > > > > > > >> > > > > >> use an
> > > >>> > > > > > > > > > > >> > > > > >> > > > > overload
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a
> > string.
> > > >>> We would
> > > >>> > > > > > rely on
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> server
> > > >>> > > > > > > > > > > >> > side
> > > >>> > > > > > > > > > > >> > > > > >> > > validation.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> > migration,
> > > >>> I think
> > > >>> > > > that
> > > >>> > > > > > we
> > > >>> > > > > > > > > could
> > > >>> > > > > > > > > > > >> still
> > > >>> > > > > > > > > > > >> > try to
> > > >>> > > > > > > > > > > >> > > > > >> > > toString
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it.
> > > That
> > > >>> should
> > > >>> > > > > > work, I
> > > >>> > > > > > > > > think, in
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> majority
> > > >>> > > > > > > > > > > >> > > > > >> > > of the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> > > >>> Justine
> > > >>> > > > that we
> > > >>> > > > > > > > > should be
> > > >>> > > > > > > > > > > >> > clearer
> > > >>> > > > > > > > > > > >> > > > > about
> > > >>> > > > > > > > > > > >> > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > reason
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > >>> > > > IBP/metadata.version
> > > >>> > > > > > from the
> > > >>> > > > > > > > > > > >> feature
> > > >>> > > > > > > > > > > >> > flag.
> > > >>> > > > > > > > > > > >> > > > > >> Maybe
> > > >>> > > > > > > > > > > >> > > > > >> > > we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > mean
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > > >>> default for
> > > >>> > > > the
> > > >>> > > > > > feature
> > > >>> > > > > > > > > flag
> > > >>> > > > > > > > > > > >> to
> > > >>> > > > > > > > > > > >> > true
> > > >>> > > > > > > > > > > >> > > > > >> based
> > > >>> > > > > > > > > > > >> > > > > >> > on
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> > > >>> once we
> > > >>> > > > want to
> > > >>> > > > > > make
> > > >>> > > > > > > > > it the
> > > >>> > > > > > > > > > > >> > default.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> > > >>> that feature
> > > >>> > > > > > flag
> > > >>> > > > > > > > > mainly
> > > >>> > > > > > > > > > > >> > during the
> > > >>> > > > > > > > > > > >> > > > > >> > > development
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not
> > > >>> have
> > > >>> > > > mentioned
> > > >>> > > > > > it, I
> > > >>> > > > > > > > > think,
> > > >>> > > > > > > > > > > >> > because
> > > >>> > > > > > > > > > > >> > > > > we
> > > >>> > > > > > > > > > > >> > > > > >> > > could
> > > >>> > > > > > > > > > > >> > > > > >> > > > > use
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config
> > > for
> > > >>> it.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> > > >>> protocol APIs
> > > >>> > > > > > don't
> > > >>> > > > > > > > > mention
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > required
> > > >>> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that
> > > for
> > > >>> > > > consistency.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit
> > odd
> > > >>> that
> > > >>> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > >>> > > > > > > > > > > >> > > > > requires
> > > >>> > > > > > > > > > > >> > > > > >> > > "Read
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems
> > to
> > > >>> do more
> > > >>> > > > than
> > > >>> > > > > > > > > reading.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is
> > how
> > > >>> the
> > > >>> > > > current
> > > >>> > > > > > > > > protocol works
> > > >>> > > > > > > > > > > >> > though.
> > > >>> > > > > > > > > > > >> > > > > We
> > > >>> > > > > > > > > > > >> > > > > >> > only
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read
> > Group"
> > > >>> to join a
> > > >>> > > > > > group. We
> > > >>> > > > > > > > > could
> > > >>> > > > > > > > > > > >> > consider
> > > >>> > > > > > > > > > > >> > > > > >> > > changing
> > > >>> > > > > > > > > > > >> > > > > >> > > > > this
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure
> > > that
> > > >>> it is
> > > >>> > > > worth
> > > >>> > > > > > it.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> > > >>> recreation
> > > >>> > > > > > handled by
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > consumer
> > > >>> > > > > > > > > > > >> > > > > with
> > > >>> > > > > > > > > > > >> > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It
> > would
> > > >>> be good to
> > > >>> > > > > > have a
> > > >>> > > > > > > > > section
> > > >>> > > > > > > > > > > >> on
> > > >>> > > > > > > > > > > >> > this.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> > > protocol
> > > >>> > > > > > perspective, the
> > > >>> > > > > > > > > new
> > > >>> > > > > > > > > > > >> topic
> > > >>> > > > > > > > > > > >> > will
> > > >>> > > > > > > > > > > >> > > > > >> have a
> > > >>> > > > > > > > > > > >> > > > > >> > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> > > >>> treat it
> > > >>> > > > like a
> > > >>> > > > > > > > > topic with
> > > >>> > > > > > > > > > > >> a
> > > >>> > > > > > > > > > > >> > > > > different
> > > >>> > > > > > > > > > > >> > > > > >> > > name.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > The
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that
> > > the
> > > >>> > > > fetch/commit
> > > >>> > > > > > > > > offsets
> > > >>> > > > > > > > > > > >> APIs
> > > >>> > > > > > > > > > > >> > do not
> > > >>> > > > > > > > > > > >> > > > > >> > support
> > > >>> > > > > > > > > > > >> > > > > >> > > > > topic
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
> > > >>> would
> > > >>> > > > reuse the
> > > >>> > > > > > > > > offsets
> > > >>> > > > > > > > > > > >> based
> > > >>> > > > > > > > > > > >> > on the
> > > >>> > > > > > > > > > > >> > > > > >> > same.
> > > >>> > > > > > > > > > > >> > > > > >> > > I
> > > >>> > > > > > > > > > > >> > > > > >> > > > > think
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should
> > update
> > > >>> those
> > > >>> > > > APIs
> > > >>> > > > > > as well
> > > >>> > > > > > > > > in
> > > >>> > > > > > > > > > > >> order
> > > >>> > > > > > > > > > > >> > to be
> > > >>> > > > > > > > > > > >> > > > > >> > > consistent
> > > >>> > > > > > > > > > > >> > > > > >> > > > > end
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> > > >>> strengthen
> > > >>> > > > the
> > > >>> > > > > > > > > semantics of
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> consumer.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> > > mentions
> > > >>> we will
> > > >>> > > > > > write the
> > > >>> > > > > > > > > new
> > > >>> > > > > > > > > > > >> > coordinator
> > > >>> > > > > > > > > > > >> > > > > >> in
> > > >>> > > > > > > > > > > >> > > > > >> > > Java.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Even
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > > >>> implementation
> > > >>> > > > > > detail, do
> > > >>> > > > > > > > > we plan
> > > >>> > > > > > > > > > > >> to
> > > >>> > > > > > > > > > > >> > have a
> > > >>> > > > > > > > > > > >> > > > > >> new
> > > >>> > > > > > > > > > > >> > > > > >> > > gradle
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> > > >>> scalability
> > > >>> > > > > > goal when
> > > >>> > > > > > > > > it
> > > >>> > > > > > > > > > > >> comes
> > > >>> > > > > > > > > > > >> > to how
> > > >>> > > > > > > > > > > >> > > > > >> many
> > > >>> > > > > > > > > > > >> > > > > >> > > > > members
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol
> > can
> > > >>> support?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have
> > numbers
> > > >>> at the
> > > >>> > > > > > moment. The
> > > >>> > > > > > > > > > > >> protocol
> > > >>> > > > > > > > > > > >> > should
> > > >>> > > > > > > > > > > >> > > > > >> > > support
> > > >>> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per
> > group.
> > > >>> We will
> > > >>> > > > > > measure
> > > >>> > > > > > > > > this when
> > > >>> > > > > > > > > > > >> > we have
> > > >>> > > > > > > > > > > >> > > > > a
> > > >>> > > > > > > > > > > >> > > > > >> > first
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation.
> > Note
> > > >>> that we
> > > >>> > > > might
> > > >>> > > > > > have
> > > >>> > > > > > > > > other
> > > >>> > > > > > > > > > > >> > > > > bottlenecks
> > > >>> > > > > > > > > > > >> > > > > >> > down
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> > > >>> commits).
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> > consider
> > > >>> having
> > > >>> > > > > > > > > SubscribedTopidIds
> > > >>> > > > > > > > > > > >> > instead
> > > >>> > > > > > > > > > > >> > > > > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > SubscribedTopicNames
> > > >>> in
> > > >>> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > >>> > > > > > > > > > > >> > > > > >> Is
> > > >>> > > > > > > > > > > >> > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > idea
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> > > >>> resolve the
> > > >>> > > > > > regex on
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > server, we
> > > >>> > > > > > > > > > > >> > > > > >> can do
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name?
> > The
> > > >>> > > > difference
> > > >>> > > > > > is that
> > > >>> > > > > > > > > > > >> sending
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> regex is
> > > >>> > > > > > > > > > > >> > > > > >> > > more
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending
> > the
> > > >>> topic
> > > >>> > > > names
> > > >>> > > > > > is less
> > > >>> > > > > > > > > > > >> > efficient.
> > > >>> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> > > easier
> > > >>> to
> > > >>> > > > handle
> > > >>> > > > > > if we
> > > >>> > > > > > > > > have
> > > >>> > > > > > > > > > > >> topic
> > > >>> > > > > > > > > > > >> > ids.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > > >>> consolidate the
> > > >>> > > > > > metadata
> > > >>> > > > > > > > > lookup
> > > >>> > > > > > > > > > > >> > on the
> > > >>> > > > > > > > > > > >> > > > > >> server
> > > >>> > > > > > > > > > > >> > > > > >> > > for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do
> > agree
> > > >>> with your
> > > >>> > > > > > point. As
> > > >>> > > > > > > > > a
> > > >>> > > > > > > > > > > >> second
> > > >>> > > > > > > > > > > >> > > > > though,
> > > >>> > > > > > > > > > > >> > > > > >> > using
> > > >>> > > > > > > > > > > >> > > > > >> > > > > topic
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better
> > > here
> > > >>> for the
> > > >>> > > > > > delete and
> > > >>> > > > > > > > > > > >> > recreation
> > > >>> > > > > > > > > > > >> > > > > case.
> > > >>> > > > > > > > > > > >> > > > > >> > > Also, I
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
> > > >>> allow
> > > >>> > > > users to
> > > >>> > > > > > > > > subscribe
> > > >>> > > > > > > > > > > >> with
> > > >>> > > > > > > > > > > >> > topic
> > > >>> > > > > > > > > > > >> > > > > >> ids
> > > >>> > > > > > > > > > > >> > > > > >> > in
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that
> > > is
> > > >>> the only
> > > >>> > > > > > way to be
> > > >>> > > > > > > > > > > >> really
> > > >>> > > > > > > > > > > >> > robust
> > > >>> > > > > > > > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > topic
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12,
> > 2022
> > > >>> at 1:38 PM
> > > >>> > > > > > David
> > > >>> > > > > > > > > Jacot <
> > > >>> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> > > >>> comments.
> > > >>> > > > Please
> > > >>> > > > > > find
> > > >>> > > > > > > > > my
> > > >>> > > > > > > > > > > >> answers
> > > >>> > > > > > > > > > > >> > > > > below.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > > >>> protocol
> > > >>> > > > relies on
> > > >>> > > > > > > > > topic IDs
> > > >>> > > > > > > > > > > >> > with the
> > > >>> > > > > > > > > > > >> > > > > >> > > exception
> > > >>> > > > > > > > > > > >> > > > > >> > > > > of the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based
> > > in
> > > >>> the
> > > >>> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > >>> > > > > > > > > > > >> > > > > >> I
> > > >>> > > > > > > > > > > >> > > > > >> > am
> > > >>> > > > > > > > > > > >> > > > > >> > > not
> > > >>> > > > > > > > > > > >> > > > > >> > > > > sure
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> > > names
> > > >>> is the
> > > >>> > > > > > right call
> > > >>> > > > > > > > > here.
> > > >>> > > > > > > > > > > >> I
> > > >>> > > > > > > > > > > >> > need to
> > > >>> > > > > > > > > > > >> > > > > >> > think
> > > >>> > > > > > > > > > > >> > > > > >> > > > > about it
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > > >>> Obviously, the
> > > >>> > > > > > KIP does
> > > >>> > > > > > > > > not
> > > >>> > > > > > > > > > > >> > change the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to
> > use
> > > >>> topic
> > > >>> > > > IDs.
> > > >>> > > > > > This
> > > >>> > > > > > > > > may be
> > > >>> > > > > > > > > > > >> > something
> > > >>> > > > > > > > > > > >> > > > > >> that
> > > >>> > > > > > > > > > > >> > > > > >> > we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as
> > > it
> > > >>> would
> > > >>> > > > give
> > > >>> > > > > > better
> > > >>> > > > > > > > > > > >> overall
> > > >>> > > > > > > > > > > >> > > > > >> guarantee in
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> > > >>> think that
> > > >>> > > > I
> > > >>> > > > > > should
> > > >>> > > > > > > > > not
> > > >>> > > > > > > > > > > >> have
> > > >>> > > > > > > > > > > >> > > > > mentioned
> > > >>> > > > > > > > > > > >> > > > > >> > this
> > > >>> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will
> > remove
> > > >>> it. We can
> > > >>> > > > > > use an
> > > >>> > > > > > > > > internal
> > > >>> > > > > > > > > > > >> > > > > >> configuration
> > > >>> > > > > > > > > > > >> > > > > >> > > while
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > > >>> feature.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> > > types
> > > >>> will be
> > > >>> > > > > > > > > supported. The
> > > >>> > > > > > > > > > > >> > change is
> > > >>> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement
> > is
> > > >>> that the
> > > >>> > > > > > cluster
> > > >>> > > > > > > > > uses
> > > >>> > > > > > > > > > > >> topic
> > > >>> > > > > > > > > > > >> > IDs.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11,
> > > 2022
> > > >>> at 9:53
> > > >>> > > > PM
> > > >>> > > > > > > > > Guozhang
> > > >>> > > > > > > > > > > >> Wang <
> > > >>> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> > > >>> feedback.
> > > >>> > > > Here
> > > >>> > > > > > are
> > > >>> > > > > > > > > some
> > > >>> > > > > > > > > > > >> replies
> > > >>> > > > > > > > > > > >> > > > > inlined
> > > >>> > > > > > > > > > > >> > > > > >> > > below:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> > > >>> 2022 at
> > > >>> > > > 2:53 AM
> > > >>> > > > > > > > > Ismael Juma
> > > >>> > > > > > > > > > > >> <
> > > >>> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for
> > the
> > > >>> KIP. This
> > > >>> > > > has
> > > >>> > > > > > the
> > > >>> > > > > > > > > > > >> potential
> > > >>> > > > > > > > > > > >> > to be a
> > > >>> > > > > > > > > > > >> > > > > >> > great
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > >>> > > > questions/comments:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think
> > it's
> > > >>> > > > premature to
> > > >>> > > > > > talk
> > > >>> > > > > > > > > about
> > > >>> > > > > > > > > > > >> > target
> > > >>> > > > > > > > > > > >> > > > > >> versions
> > > >>> > > > > > > > > > > >> > > > > >> > > for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of
> > the
> > > >>> existing
> > > >>> > > > group
> > > >>> > > > > > > > > protocol.
> > > >>> > > > > > > > > > > >> > Unlike
> > > >>> > > > > > > > > > > >> > > > > >> KRaft,
> > > >>> > > > > > > > > > > >> > > > > >> > > this
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> > protocol
> > > >>> and hence
> > > >>> > > > > > > > > > > >> > deprecation/removal will
> > > >>> > > > > > > > > > > >> > > > > >> be
> > > >>> > > > > > > > > > > >> > > > > >> > > heavily
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> > > >>> applications
> > > >>> > > > > > migrate
> > > >>> > > > > > > > > to the
> > > >>> > > > > > > > > > > >> new
> > > >>> > > > > > > > > > > >> > > > > >> protocol.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree
> > with
> > > >>> you. I
> > > >>> > > > think
> > > >>> > > > > > we can
> > > >>> > > > > > > > > > > >> remove
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> proposed
> > > >>> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > > >>> > > > Deprecation,
> > > >>> > > > > > and
> > > >>> > > > > > > > > Migration
> > > >>> > > > > > > > > > > >> > Plan` and
> > > >>> > > > > > > > > > > >> > > > > >> > > instead
> > > >>> > > > > > > > > > > >> > > > > >> > > > > just
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> > > >>> decide in the
> > > >>> > > > > > future
> > > >>> > > > > > > > > about
> > > >>> > > > > > > > > > > >> when
> > > >>> > > > > > > > > > > >> > we
> > > >>> > > > > > > > > > > >> > > > > would
> > > >>> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > > >>> behaviors.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP
> > says
> > > >>> we
> > > >>> > > > intend to
> > > >>> > > > > > > > > release this
> > > >>> > > > > > > > > > > >> > in 4.x,
> > > >>> > > > > > > > > > > >> > > > > >> but
> > > >>> > > > > > > > > > > >> > > > > >> > it
> > > >>> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> > > added
> > > >>> that as
> > > >>> > > > a
> > > >>> > > > > > way to
> > > >>> > > > > > > > > > > >> estimate
> > > >>> > > > > > > > > > > >> > when
> > > >>> > > > > > > > > > > >> > > > > we'd
> > > >>> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > > >>> protocol, I
> > > >>> > > > also
> > > >>> > > > > > suggest
> > > >>> > > > > > > > > > > >> removing
> > > >>> > > > > > > > > > > >> > this
> > > >>> > > > > > > > > > > >> > > > > >> part.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's
> > > not
> > > >>> > > > specifically
> > > >>> > > > > > > > > related to
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> > > >>> more for
> > > >>> > > > client
> > > >>> > > > > > > > > upgrades.
> > > >>> > > > > > > > > > > >> I.e.
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> > > broker-side
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation
> > > may
> > > >>> be done
> > > >>> > > > > > first,
> > > >>> > > > > > > > > and then
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > client
> > > >>> > > > > > > > > > > >> > > > > >> > side,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > and we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> > > >>> "released"
> > > >>> > > > by
> > > >>> > > > > > the
> > > >>> > > > > > > > > time
> > > >>> > > > > > > > > > > >> clients
> > > >>> > > > > > > > > > > >> > > > > >> > > implementations
> > > >>> > > > > > > > > > > >> > > > > >> > > > > are
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> > > >>> enable the
> > > >>> > > > > > feature the
> > > >>> > > > > > > > > > > >> clients
> > > >>> > > > > > > > > > > >> > need to
> > > >>> > > > > > > > > > > >> > > > > >> > first
> > > >>> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> > > >>> rolling
> > > >>> > > > bounce
> > > >>> > > > > > and
> > > >>> > > > > > > > > then set
> > > >>> > > > > > > > > > > >> > the flag
> > > >>> > > > > > > > > > > >> > > > > >> > with a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > second
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and
> > hence
> > > >>> we feel
> > > >>> > > > it's
> > > >>> > > > > > > > > better to be
> > > >>> > > > > > > > > > > >> > released
> > > >>> > > > > > > > > > > >> > > > > >> in a
> > > >>> > > > > > > > > > > >> > > > > >> > > major
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> > > >>> flesh out
> > > >>> > > > the
> > > >>> > > > > > > > > details of
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > migration
> > > >>> > > > > > > > > > > >> > > > > >> > > story.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > It
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying
> > we
> > > >>> will
> > > >>> > > > support
> > > >>> > > > > > online
> > > >>> > > > > > > > > > > >> > migrations. Is
> > > >>> > > > > > > > > > > >> > > > > >> that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this
> > in
> > > >>> detail.
> > > >>> > > > It
> > > >>> > > > > > could
> > > >>> > > > > > > > > also be
> > > >>> > > > > > > > > > > >> > done as a
> > > >>> > > > > > > > > > > >> > > > > >> > > separate
> > > >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > > that's
> > > >>> the
> > > >>> > > > part we
> > > >>> > > > > > can
> > > >>> > > > > > > > > be more
> > > >>> > > > > > > > > > > >> > concrete
> > > >>> > > > > > > > > > > >> > > > > >> about
> > > >>> > > > > > > > > > > >> > > > > >> > > for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > sure
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related
> > > to
> > > >>> your
> > > >>> > > > > > question 2)
> > > >>> > > > > > > > > above).
> > > >>> > > > > > > > > > > >> > We will
> > > >>> > > > > > > > > > > >> > > > > >> work
> > > >>> > > > > > > > > > > >> > > > > >> > on
> > > >>> > > > > > > > > > > >> > > > > >> > > > > making
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > > >>> parallel as we
> > > >>> > > > > > solicit
> > > >>> > > > > > > > > more
> > > >>> > > > > > > > > > > >> > feedback.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
> > > >>> that we are
> > > >>> > > > > > pushing
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > pattern
> > > >>> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> > > >>> like there
> > > >>> > > > > > could be
> > > >>> > > > > > > > > some
> > > >>> > > > > > > > > > > >> tricky
> > > >>> > > > > > > > > > > >> > > > > >> > > compatibility
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > > mechanism
> > > >>> for
> > > >>> > > > users to
> > > >>> > > > > > > > > detect that
> > > >>> > > > > > > > > > > >> > they
> > > >>> > > > > > > > > > > >> > > > > need
> > > >>> > > > > > > > > > > >> > > > > >> to
> > > >>> > > > > > > > > > > >> > > > > >> > > update
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > > switching
> > > >>> to the
> > > >>> > > > new
> > > >>> > > > > > > > > protocol?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> > > >>> need some
> > > >>> > > > > > tooling for
> > > >>> > > > > > > > > > > >> non-java
> > > >>> > > > > > > > > > > >> > client
> > > >>> > > > > > > > > > > >> > > > > >> > users
> > > >>> > > > > > > > > > > >> > > > > >> > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > sort
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> > > >>> client before
> > > >>> > > > > > > > > switching to
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > new
> > > >>> > > > > > > > > > > >> > > > > >> > protocol.
> > > >>> > > > > > > > > > > >> > > > > >> > > I
> > > >>> > > > > > > > > > > >> > > > > >> > > > > do not
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea
> > on
> > > >>> top of my
> > > >>> > > > head
> > > >>> > > > > > > > > though,
> > > >>> > > > > > > > > > > >> maybe
> > > >>> > > > > > > > > > > >> > others
> > > >>> > > > > > > > > > > >> > > > > >> like
> > > >>> > > > > > > > > > > >> > > > > >> > > @Matt
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to
> > > >>> the last
> > > >>> > > > > > question,
> > > >>> > > > > > > > > will the
> > > >>> > > > > > > > > > > >> > Java
> > > >>> > > > > > > > > > > >> > > > > >> client
> > > >>> > > > > > > > > > > >> > > > > >> > > allow
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with
> > the
> > > >>> current
> > > >>> > > > regex
> > > >>> > > > > > > > > engine for
> > > >>> > > > > > > > > > > >> > > > > >> compatibility
> > > >>> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be
> > handy
> > > >>> to keep
> > > >>> > > > using
> > > >>> > > > > > > > > client
> > > >>> > > > > > > > > > > >> based
> > > >>> > > > > > > > > > > >> > regex
> > > >>> > > > > > > > > > > >> > > > > at
> > > >>> > > > > > > > > > > >> > > > > >> > > first to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > keep
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> > > >>> simple and
> > > >>> > > > then
> > > >>> > > > > > migrate
> > > >>> > > > > > > > > to
> > > >>> > > > > > > > > > > >> > server based
> > > >>> > > > > > > > > > > >> > > > > >> > > regexes
> > > >>> > > > > > > > > > > >> > > > > >> > > > > as a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have
> > > >>> not thought
> > > >>> > > > > > about
> > > >>> > > > > > > > > that for
> > > >>> > > > > > > > > > > >> > java
> > > >>> > > > > > > > > > > >> > > > > >> clients,
> > > >>> > > > > > > > > > > >> > > > > >> > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > we can
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind
> > > of
> > > >>> > > > > > compatibility
> > > >>> > > > > > > > > issues do
> > > >>> > > > > > > > > > > >> > you have
> > > >>> > > > > > > > > > > >> > > > > in
> > > >>> > > > > > > > > > > >> > > > > >> > > mind?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we
> > say
> > > >>> that the
> > > >>> > > > group
> > > >>> > > > > > > > > coordinator
> > > >>> > > > > > > > > > > >> > will be
> > > >>> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > > >>> configurations and
> > > >>> > > > that
> > > >>> > > > > > the
> > > >>> > > > > > > > > > > >> > configurations
> > > >>> > > > > > > > > > > >> > > > > will
> > > >>> > > > > > > > > > > >> > > > > >> be
> > > >>> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> > > >>> deleted. Will a
> > > >>> > > > > > > > > transition to
> > > >>> > > > > > > > > > > >> DEAD
> > > >>> > > > > > > > > > > >> > > > > trigger
> > > >>> > > > > > > > > > > >> > > > > >> > > deletion
> > > >>> > > > > > > > > > > >> > > > > >> > > > > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > configurations?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> > > >>> DEAD state
> > > >>> > > > is an
> > > >>> > > > > > > > > ending
> > > >>> > > > > > > > > > > >> state
> > > >>> > > > > > > > > > > >> > (we
> > > >>> > > > > > > > > > > >> > > > > would
> > > >>> > > > > > > > > > > >> > > > > >> > only
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> > > >>> group is
> > > >>> > > > EMPTY
> > > >>> > > > > > and
> > > >>> > > > > > > > > also all
> > > >>> > > > > > > > > > > >> of
> > > >>> > > > > > > > > > > >> > its
> > > >>> > > > > > > > > > > >> > > > > >> > metadata
> > > >>> > > > > > > > > > > >> > > > > >> > > are
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited
> > to
> > > >>> DEAD this
> > > >>> > > > > > group
> > > >>> > > > > > > > > would
> > > >>> > > > > > > > > > > >> never
> > > >>> > > > > > > > > > > >> > be
> > > >>> > > > > > > > > > > >> > > > > >> revived.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> > > >>> choice to
> > > >>> > > > store
> > > >>> > > > > > the
> > > >>> > > > > > > > > configs
> > > >>> > > > > > > > > > > >> in
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to
> > list
> > > >>> all
> > > >>> > > > cluster
> > > >>> > > > > > configs
> > > >>> > > > > > > > > and
> > > >>> > > > > > > > > > > >> their
> > > >>> > > > > > > > > > > >> > > > > values?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> > > >>> question,
> > > >>> > > > and our
> > > >>> > > > > > > > > thoughts
> > > >>> > > > > > > > > > > >> are
> > > >>> > > > > > > > > > > >> > that
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > so-called
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations"
> > > are
> > > >>> > > > overrides
> > > >>> > > > > > of the
> > > >>> > > > > > > > > > > >> > cluster-level
> > > >>> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> > > >>> group so
> > > >>> > > > when an
> > > >>> > > > > > > > > admin list
> > > >>> > > > > > > > > > > >> > cluster
> > > >>> > > > > > > > > > > >> > > > > >> > configs
> > > >>> > > > > > > > > > > >> > > > > >> > > it's
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> > > >>> cluster-level
> > > >>> > > > > > > > > "defaults", not
> > > >>> > > > > > > > > > > >> > showing
> > > >>> > > > > > > > > > > >> > > > > >> any
> > > >>> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> > > >>> someone
> > > >>> > > > > > configure a
> > > >>> > > > > > > > > group
> > > >>> > > > > > > > > > > >> > before
> > > >>> > > > > > > > > > > >> > > > > >> starting
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> > > >>> allowing
> > > >>> > > > the
> > > >>> > > > > > explicit
> > > >>> > > > > > > > > > > >> > creation of
> > > >>> > > > > > > > > > > >> > > > > >> groups?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could
> > > be
> > > >>> > > > decoupled
> > > >>> > > > > > from
> > > >>> > > > > > > > > the group
> > > >>> > > > > > > > > > > >> > > > > lifecycle.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can
> > > be
> > > >>> created
> > > >>> > > > > > before
> > > >>> > > > > > > > > the group
> > > >>> > > > > > > > > > > >> > itself
> > > >>> > > > > > > > > > > >> > > > > as
> > > >>> > > > > > > > > > > >> > > > > >> an
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course,
> > > this
> > > >>> > > > requires
> > > >>> > > > > > the
> > > >>> > > > > > > > > > > >> corresponding
> > > >>> > > > > > > > > > > >> > > > > >> request to
> > > >>> > > > > > > > > > > >> > > > > >> > > be
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> > coordinator
> > > >>> based on
> > > >>> > > > the
> > > >>> > > > > > group
> > > >>> > > > > > > > > id ---
> > > >>> > > > > > > > > > > >> > the only
> > > >>> > > > > > > > > > > >> > > > > >> > thing
> > > >>> > > > > > > > > > > >> > > > > >> > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> > > >>> itself is
> > > >>> > > > gone
> > > >>> > > > > > we
> > > >>> > > > > > > > > also check
> > > >>> > > > > > > > > > > >> > if there
> > > >>> > > > > > > > > > > >> > > > > >> are
> > > >>> > > > > > > > > > > >> > > > > >> > > any
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> > related
> > > >>> to that
> > > >>> > > > group
> > > >>> > > > > > and
> > > >>> > > > > > > > > delete
> > > >>> > > > > > > > > > > >> as
> > > >>> > > > > > > > > > > >> > well.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> > > >>> indeed
> > > >>> > > > > > introduces an
> > > >>> > > > > > > > > > > >> > asymmetry on
> > > >>> > > > > > > > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > creation
> > > >>> > > > > > > > > > > >> > > > > >> > > > > /
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of
> > the
> > > >>> config
> > > >>> > > > > > entities,
> > > >>> > > > > > > > > and we
> > > >>> > > > > > > > > > > >> > would like
> > > >>> > > > > > > > > > > >> > > > > >> to
> > > >>> > > > > > > > > > > >> > > > > >> > > hear
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> > whether
> > > >>> we should
> > > >>> > > > aim
> > > >>> > > > > > for
> > > >>> > > > > > > > > symmetry
> > > >>> > > > > > > > > > > >> > i.e.
> > > >>> > > > > > > > > > > >> > > > > >> totally
> > > >>> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and
> > hence
> > > >>> not
> > > >>> > > > delete
> > > >>> > > > > > them at
> > > >>> > > > > > > > > all
> > > >>> > > > > > > > > > > >> when
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> group
> > > >>> > > > > > > > > > > >> > > > > >> > is
> > > >>> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> > explicit
> > > >>> deletion
> > > >>> > > > > > operations
> > > >>> > > > > > > > > by
> > > >>> > > > > > > > > > > >> > themselves.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > >>> > > > > > Consumer.subscribe
> > > >>> > > > > > > > > method for
> > > >>> > > > > > > > > > > >> > the Java
> > > >>> > > > > > > > > > > >> > > > > >> > client
> > > >>> > > > > > > > > > > >> > > > > >> > > > > still
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > `java.util.regex.Pattern` of
> > > >>> > > > > > do we
> > > >>> > > > > > > > > have
> > > >>> > > > > > > > > > > >> to
> > > >>> > > > > > > > > > > >> > > > > >> introduce an
> > > >>> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do
> > not
> > > >>> need to
> > > >>> > > > > > introduce
> > > >>> > > > > > > > > an
> > > >>> > > > > > > > > > > >> > overload, but
> > > >>> > > > > > > > > > > >> > > > > >> I'm
> > > >>> > > > > > > > > > > >> > > > > >> > > all
> > > >>> > > > > > > > > > > >> > > > > >> > > > > ears
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > > >>> compatibility
> > > >>> > > > > > issues
> > > >>> > > > > > > > > that we
> > > >>> > > > > > > > > > > >> may
> > > >>> > > > > > > > > > > >> > > > > overlook.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree
> > > with
> > > >>> Justine
> > > >>> > > > > > that we
> > > >>> > > > > > > > > should
> > > >>> > > > > > > > > > > >> be
> > > >>> > > > > > > > > > > >> > clearer
> > > >>> > > > > > > > > > > >> > > > > >> > about
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > >>> > > > > > IBP/metadata.version
> > > >>> > > > > > > > > from the
> > > >>> > > > > > > > > > > >> > feature
> > > >>> > > > > > > > > > > >> > > > > >> flag.
> > > >>> > > > > > > > > > > >> > > > > >> > > Maybe
> > > >>> > > > > > > > > > > >> > > > > >> > > > > we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch
> > the
> > > >>> default
> > > >>> > > > for
> > > >>> > > > > > the
> > > >>> > > > > > > > > feature
> > > >>> > > > > > > > > > > >> flag
> > > >>> > > > > > > > > > > >> > to
> > > >>> > > > > > > > > > > >> > > > > true
> > > >>> > > > > > > > > > > >> > > > > >> > > based on
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > metadata.version
> > > >>> once we
> > > >>> > > > > > want to
> > > >>> > > > > > > > > make it
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> default.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> > > >>> protocol
> > > >>> > > > APIs
> > > >>> > > > > > don't
> > > >>> > > > > > > > > mention
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > >> required
> > > >>> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > it
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add
> > > that
> > > >>> for
> > > >>> > > > > > consistency.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > > >>> certainly do
> > > >>> > > > that.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a
> > bit
> > > >>> odd that
> > > >>> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > >>> > > > > > > > > > > >> > > > > >> > requires
> > > >>> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it
> > seems
> > > >>> to do
> > > >>> > > > more
> > > >>> > > > > > than
> > > >>> > > > > > > > > reading.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> > > thought
> > > >>> myself
> > > >>> > > > as
> > > >>> > > > > > well,
> > > >>> > > > > > > > > but in
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > end we
> > > >>> > > > > > > > > > > >> > > > > >> > could
> > > >>> > > > > > > > > > > >> > > > > >> > > not
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > > >>> alternative: adding
> > > >>> > > > > > Write
> > > >>> > > > > > > > > Group
> > > >>> > > > > > > > > > > >> seems
> > > >>> > > > > > > > > > > >> > an
> > > >>> > > > > > > > > > > >> > > > > >> overkill
> > > >>> > > > > > > > > > > >> > > > > >> > > here
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > > >>> elsewhere (we
> > > >>> > > > only
> > > >>> > > > > > have
> > > >>> > > > > > > > > Read /
> > > >>> > > > > > > > > > > >> > Delete
> > > >>> > > > > > > > > > > >> > > > > and
> > > >>> > > > > > > > > > > >> > > > > >> > > Describe
> > > >>> > > > > > > > > > > >> > > > > >> > > > > on
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would
> > like
> > > >>> to hear
> > > >>> > > > others
> > > >>> > > > > > > > > thoughts.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> > > topic
> > > >>> > > > recreation
> > > >>> > > > > > > > > handled by
> > > >>> > > > > > > > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > consumer
> > > >>> > > > > > > > > > > >> > > > > >> > with
> > > >>> > > > > > > > > > > >> > > > > >> > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> > > >>> would be
> > > >>> > > > good
> > > >>> > > > > > to have
> > > >>> > > > > > > > > a
> > > >>> > > > > > > > > > > >> > section on
> > > >>> > > > > > > > > > > >> > > > > >> this.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with
> > > regex
> > > >>> > > > > > subscription
> > > >>> > > > > > > > > right? Yes
> > > >>> > > > > > > > > > > >> > we can
> > > >>> > > > > > > > > > > >> > > > > >> add a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > section
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> > > >>> basically the
> > > >>> > > > idea
> > > >>> > > > > > is that
> > > >>> > > > > > > > > > > >> consumer
> > > >>> > > > > > > > > > > >> > would
> > > >>> > > > > > > > > > > >> > > > > be
> > > >>> > > > > > > > > > > >> > > > > >> > > totally
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> > protocol
> > > >>> as it's
> > > >>> > > > > > handled all
> > > >>> > > > > > > > > by the
> > > >>> > > > > > > > > > > >> > brokers.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> > > >>> mentions we
> > > >>> > > > will
> > > >>> > > > > > write
> > > >>> > > > > > > > > the
> > > >>> > > > > > > > > > > >> new
> > > >>> > > > > > > > > > > >> > > > > >> coordinator
> > > >>> > > > > > > > > > > >> > > > > >> > in
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > > >>> implementation
> > > >>> > > > > > detail,
> > > >>> > > > > > > > > do we
> > > >>> > > > > > > > > > > >> > plan to
> > > >>> > > > > > > > > > > >> > > > > >> have a
> > > >>> > > > > > > > > > > >> > > > > >> > > new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> > > >>> thought about
> > > >>> > > > > > that. But
> > > >>> > > > > > > > > I think
> > > >>> > > > > > > > > > > >> > the
> > > >>> > > > > > > > > > > >> > > > > answer
> > > >>> > > > > > > > > > > >> > > > > >> > > should
> > > >>> > > > > > > > > > > >> > > > > >> > > > > be
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we
> > have
> > > a
> > > >>> > > > scalability
> > > >>> > > > > > goal
> > > >>> > > > > > > > > when it
> > > >>> > > > > > > > > > > >> > comes to
> > > >>> > > > > > > > > > > >> > > > > >> how
> > > >>> > > > > > > > > > > >> > > > > >> > > many
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> > protocol
> > > >>> can
> > > >>> > > > support?
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> > group, I
> > > >>> think we
> > > >>> > > > > > should
> > > >>> > > > > > > > > shoot for
> > > >>> > > > > > > > > > > >> > 1000s of
> > > >>> > > > > > > > > > > >> > > > > >> > > members.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > But
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability
> > goals
> > > >>> also
> > > >>> > > > depend
> > > >>> > > > > > on the
> > > >>> > > > > > > > > offset
> > > >>> > > > > > > > > > > >> > > > > management
> > > >>> > > > > > > > > > > >> > > > > >> > > (commit,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of
> > > the
> > > >>> > > > coordinator
> > > >>> > > > > > > > > which we
> > > >>> > > > > > > > > > > >> did
> > > >>> > > > > > > > > > > >> > not
> > > >>> > > > > > > > > > > >> > > > > >> cover in
> > > >>> > > > > > > > > > > >> > > > > >> > > this
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> > > >>> number that
> > > >>> > > > > > applies
> > > >>> > > > > > > > > > > >> > universally.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > > >>> consider
> > > >>> > > > having
> > > >>> > > > > > > > > > > >> > SubscribedTopidIds
> > > >>> > > > > > > > > > > >> > > > > >> instead
> > > >>> > > > > > > > > > > >> > > > > >> > of
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> SubscribedTopicNames in
> > > >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > >>> > > > > > > > > > > >> > > > > >> > > Is the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have
> > > to
> > > >>> resolve
> > > >>> > > > the
> > > >>> > > > > > regex
> > > >>> > > > > > > > > on the
> > > >>> > > > > > > > > > > >> > server,
> > > >>> > > > > > > > > > > >> > > > > we
> > > >>> > > > > > > > > > > >> > > > > >> > can
> > > >>> > > > > > > > > > > >> > > > > >> > > do
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic
> > name?
> > > >>> The
> > > >>> > > > > > difference is
> > > >>> > > > > > > > > that
> > > >>> > > > > > > > > > > >> > sending the
> > > >>> > > > > > > > > > > >> > > > > >> > regex
> > > >>> > > > > > > > > > > >> > > > > >> > > is
> > > >>> > > > > > > > > > > >> > > > > >> > > > > more
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas
> > sending
> > > >>> the topic
> > > >>> > > > > > names is
> > > >>> > > > > > > > > less
> > > >>> > > > > > > > > > > >> > efficient.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> > > >>> easier to
> > > >>> > > > > > handle if
> > > >>> > > > > > > > > we have
> > > >>> > > > > > > > > > > >> > topic
> > > >>> > > > > > > > > > > >> > > > > ids.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason
> > > to
> > > >>> still
> > > >>> > > > let
> > > >>> > > > > > the
> > > >>> > > > > > > > > clients
> > > >>> > > > > > > > > > > >> send
> > > >>> > > > > > > > > > > >> > names
> > > >>> > > > > > > > > > > >> > > > > >> is to
> > > >>> > > > > > > > > > > >> > > > > >> > > keep
> > > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> > > names
> > > >>> -> ids
> > > >>> > > > on
> > > >>> > > > > > the
> > > >>> > > > > > > > > broker /
> > > >>> > > > > > > > > > > >> > admin
> > > >>> > > > > > > > > > > >> > > > > client
> > > >>> > > > > > > > > > > >> > > > > >> > > only.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Note
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we
> > added
> > > >>> topic id
> > > >>> > > > in
> > > >>> > > > > > > > > KIP-516, we
> > > >>> > > > > > > > > > > >> > never
> > > >>> > > > > > > > > > > >> > > > > >> > > implemented the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > consumer/producers
> > > >>> > > > leveraging
> > > >>> > > > > > the
> > > >>> > > > > > > > > related
> > > >>> > > > > > > > > > > >> > newer
> > > >>> > > > > > > > > > > >> > > > > >> versioned
> > > >>> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> > > topic
> > > >>> id as
> > > >>> > > > empty
> > > >>> > > > > > UUID.
> > > >>> > > > > > > > > We
> > > >>> > > > > > > > > > > >> want
> > > >>> > > > > > > > > > > >> > to keep
> > > >>> > > > > > > > > > > >> > > > > >> the
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and
> > > only
> > > >>> > > > delegate
> > > >>> > > > > > the
> > > >>> > > > > > > > > reasoning
> > > >>> > > > > > > > > > > >> on
> > > >>> > > > > > > > > > > >> > broker
> > > >>> > > > > > > > > > > >> > > > > >> and
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul
> > 6,
> > > >>> 2022 at
> > > >>> > > > 10:45
> > > >>> > > > > > AM
> > > >>> > > > > > > > > David
> > > >>> > > > > > > > > > > >> Jacot
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> <dj...@confluent.io.invalid>
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would
> > like
> > > >>> to start a
> > > >>> > > > > > > > > discussion
> > > >>> > > > > > > > > > > >> thread
> > > >>> > > > > > > > > > > >> > on
> > > >>> > > > > > > > > > > >> > > > > >> KIP-848:
> > > >>> > > > > > > > > > > >> > > > > >> > > The
> > > >>> > > > > > > > > > > >> > > > > >> > > > > Next
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation
> > of
> > > >>> the
> > > >>> > > > Consumer
> > > >>> > > > > > > > > Rebalance
> > > >>> > > > > > > > > > > >> > Protocol.
> > > >>> > > > > > > > > > > >> > > > > >> With
> > > >>> > > > > > > > > > > >> > > > > >> > > this
> > > >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> > > >>> rebalance
> > > >>> > > > > > protocol
> > > >>> > > > > > > > > (for
> > > >>> > > > > > > > > > > >> > consumers)
> > > >>> > > > > > > > > > > >> > > > > >> more
> > > >>> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> > > >>> easier to
> > > >>> > > > > > implement for
> > > >>> > > > > > > > > > > >> > clients, and
> > > >>> > > > > > > > > > > >> > > > > >> easier
> > > >>> > > > > > > > > > > >> > > > > >> > > to
> > > >>> > > > > > > > > > > >> > > > > >> > > > > debug
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> > > >>> here:
> > > >>> > > > > > > > > > > >> > > > > >> > > > >
> > > >>> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please
> > take a
> > > >>> look and
> > > >>> > > > let
> > > >>> > > > > > me
> > > >>> > > > > > > > > know what
> > > >>> > > > > > > > > > > >> > you
> > > >>> > > > > > > > > > > >> > > > > think.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will
> > be
> > > >>> away from
> > > >>> > > > > > July
> > > >>> > > > > > > > > 18th to
> > > >>> > > > > > > > > > > >> > August 8th.
> > > >>> > > > > > > > > > > >> > > > > >> That
> > > >>> > > > > > > > > > > >> > > > > >> > > gives
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> > > >>> read and
> > > >>> > > > digest
> > > >>> > > > > > this
> > > >>> > > > > > > > > long
> > > >>> > > > > > > > > > > >> KIP.
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > > > >
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> > >
> > > >>> > > > > > > > > > > >> > > > > >> >
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > > >> --
> > > >>> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > >>> > > > > > > > > > > >> > > > > >>
> > > >>> > > > > > > > > > > >> > > > > >
> > > >>> > > > > > > > > > > >> > > > >
> > > >>> > > > > > > > > > > >> >
> > > >>> > > > > > > > > > > >>
> > > >>> > > > > > > > > > > >
> > > >>> > > > > > > > >
> > > >>> > > > > >
> > > >>> > > > >
> > > >>> > > > >
> > > >>> > > > > --
> > > >>> > > > > -- Guozhang
> > > >>> > > >
> > > >>> > >
> > > >>> > >
> > > >>> > > --
> > > >>> > > -- Guozhang
> > > >>>
> > > >>
> > > >>
> > > >> --
> > > >> -- Guozhang
> > > >>
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
> > >
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Jason Gustafson <ja...@confluent.io.INVALID>]

Hi David,

Thanks for all the hard work on this.  I'm steadily making my way through
the doc and had a few initial questions:

1. The KIP mentions the following in regard to static membership:

> Static membership, introduced in KIP-345, is still supported by this new
rebalance protocol. When a member rejoins with the same Instance ID, the
group coordinator replaces the old member with the new member. The new
member can continue from where it left off.

Can you clarify this a little bit? Suppose that the current member is still
active and a new one arrives with the same instance ID. Do we fence the
current member immediately or do we attempt to transition gracefully?

2. I wasn't very clear about the plan for existing client configurations
which overlap with the new broker configurations. For example,
heartbeat.interval.ms. Would the setting on the server override this? Do we
plan to deprecate the configuration?

3. The KIP adds support for regex subscription through a separate field in
the heartbeat request. Perhaps it would make sense to make this a field
type in the RPC schema definitions. I can imagine reusing this in the
Metadata request API for example. By the way, does this introduce a
dependence on the RE2 library? Or will the client simply pass through the
string that is provided by the user?

4. In regard to the deprecation of `enforceRebalance`, I wonder if it would
be better to make it a no-op instead of letting it raise an exception. Just
a little concerned that might make upgrading more difficult.

5. I'm trying to decide if `group.protocol` is pulling enough weight to
exist. Could the behavior be inferred from other configurations? For
example, perhaps by default, group.remote.assignor should be empty, which
implies the "generic" protocol. If a user provides a non-empty value, then
we change to "consumer." That would give the opt-in behavior that it looks
like you want.

Thanks,
Jason





On Mon, Sep 19, 2022 at 5:33 PM Jun Rao <ju...@confluent.io.invalid> wrote:

> Hi, David,
>
> Thanks for the reply. A few more comments.
>
> 40. Is there an API to change PartitionAssignor#metadata? I assume the
> member fields are just for reads?
>
> 43.2 Sounds good.
>
> 44.1 Thanks. Could we document that the coordinator selects the assignor
> when multiple assignors are specified in the client?
>
> 44.2 Encoding the group state in ConsumerGroupPrepareAssignmentResponse
> sounds simpler and clearer to me.
>
> 48. Yes, that's the confusing part. I assume member assignment is what the
> member client has. Could we include member assignment from both the
> coordinator and the client's point of view?
>
> 50. We only include Uuid in PartitionAssignor.Group.TopicMetadata. How does
> KStream do co-partitioning across topics?
>
> 51. PartitionAssignor.metadata: Is this really an assignor level thing? The
> use case mentioned is to use metadata for encoding lags in standby tasks.
> However, the lag is independent of the assignor. For example, if a member
> sets two assignors, does it need to set the same metadata in both
> assignors?
>
> 52. "power users will have the ability to trigger a reassignment by either
> providing a non-zero reason or by updating the assignor metadata." : Do we
> need both reason and metadata to trigger an assignment? Could we just use
> metadata?
>
> 53. MemberAssignment.metadata: How is that used? Does that overwrite the
> member metadata on the coordinator?
>
> Thanks,
>
> Jun
>
>
> On Tue, Sep 13, 2022 at 1:59 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> > Hi David,
> >
> > Another quick thing I just realized about
> "PartitionAssignor#onAssignment":
> > in the new proposal, this function could be triggered multiple time
> within
> > a group generation as it will be triggered whenever the broker gives the
> > client new assignments, whereas today it is only triggered once per
> > rebalance. I think it's worth emphasizing this a bit in the doc,
> especially
> > around:
> >
> > 1) how many times `onAssignment` could be triggered with a single
> rebalance
> > (especially if the new assignment is decided by the client-assignor).
> > 2) when the Assignment metadata could change across multiple
> `onAssignment`
> > calls, and how users should handle it.
> > 3) the order of calling `onAssignment` and
> `onPartitionsRevoked/Assigned`.
> >
> > For example, today the Streams client relies on the following
> assumptions:
> > 1) within a rebalance the onPartitionsRevoked would be called before the
> > onAssignment, and onAssignment is called before onPartitionsAssigned
> (today
> > we would trigger all three only once within a generation), 2)
> onAssignment
> > contains the source-of-truth of the final assignment, and hence we can
> > simply ignore onPartitionsAssigned and just rely on onAssignment to
> reason
> > about which tasks are removed / assigned. And each time it calls, the
> > corresponding metadata change are "final" for this new generation.
> >
> > We can change however Streams handle the new protocol, but I think there
> > are other clients externally which may also make such assumptions about
> the
> > ordering / call times about those callbacks, and hence worth clarifying
> > those in the new proposal.
> >
> >
> >
> > Guozhang
> >
> >
> >
> >
> > On Tue, Sep 13, 2022 at 12:20 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> >
> > > Hello David,
> > >
> > > Just had a few more nit questions about the public APIs:
> > >
> > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > field,
> > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > rationale seems to be assuming that we should always be able to do the
> > > assignment at the broker-side assignor without errors. Personally I
> think
> > > it's still potentially beneficial to add the Error field even for
> > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > topics are not recognized with the current broker's metadata. What do
> you
> > > think?
> > >
> > > 2) The client-side "GroupMember" has three additional fields
> > > reason/version/metadata compared with the broker-side GroupMember. I
> > agree
> > > that broker-side assignor would not need reason/metadata since they are
> > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > it
> > > possible that we evolve our broker-side built-in assignor but the old
> > > versioned clients would not be able to work with the new version, in
> > which
> > > case we need to let the broker being aware of this and upgrade its
> > behavior
> > > to cooperate with the clients?
> > >
> > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > of
> > > including these three fields, what about just adding the "Metadata"
> field
> > > class which has these three fields? Also, there are two "Metadata"
> > > currently in the APIs, the first is a class that encodes
> > > reason/version/metadata, and the second is just the encoded metadata
> > bytes.
> > > I'm wondering what about just naming the first as memberMetadata, which
> > > then has a bytebuffer field Metadata, or instead naming the second
> > > bytebuffer field as metadataBytes?
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > >> Hello David,
> > >>
> > >> Thanks for bringing this question up. I think the main benefits as you
> > >> listed is 2) above if it stays; just to clarify, we would only be able
> > to
> > >> save one round trip if the rebalance is still triggered by the broker;
> > if
> > >> the rebalance is triggered by the client then the num.round.trips are
> > the
> > >> same:
> > >>
> > >> 1) With GroupPrepareAssignment:
> > >>
> > >> T0: client decides to do a new assignment, suppose it has already
> sent a
> > >> HB and hence has to wait for it to return first since only one
> request /
> > >> response can be inflight with the coordinator's socket.
> > >> T1: client receives the HB response, and then sends the
> > >> GroupPrepareAssignment request.
> > >> T2: the GroupPrepareAssignment response is returned.
> > >> T3: it calculates the new assignment, and sends a
> GroupInstallAssignment
> > >> request.
> > >>
> > >> In total, two round trips.
> > >>
> > >> 2) Without GroupPrepareAssignment:
> > >>
> > >> T0: client decides to do a new assignment, suppose it has already
> sent a
> > >> HB and hence has to wait for it to return first since only one
> request /
> > >> response can be inflight with the coordinator's socket.
> > >> T1: client receives the HB response, and then sends the new HB request
> > >> with the flag indicating a new rebalance needed..
> > >> T2: the HB response with the optional member metadata map is returned.
> > >> T3: it calculates the new assignment, and sends a
> GroupInstallAssignment
> > >> request.
> > >>
> > >> In total, two round trips as well.
> > >>
> > >> -----------------------------
> > >>
> > >> So to complete the full picture here, we'd need to modify both HB
> > request
> > >> and response so that the client can also indicate a new rebalance via
> > the
> > >> HB request as well, right?
> > >>
> > >> Assuming all above is true, I think it's okay to merge the
> > >> GroupPrepareAssignment into HB given that we can make the additional
> > fields
> > >> encoding the full member (subscription) metadata and topic metadata as
> > >> optional fields.
> > >>
> > >>
> > >> Guozhang
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >> On Mon, Sep 12, 2022 at 5:22 AM David Jacot
> <djacot@confluent.io.invalid
> > >
> > >> wrote:
> > >>
> > >>> Hi all,
> > >>>
> > >>> During an offline conversation, someone asked why we need the
> > >>> ConsumerGroupPrepareAssignment API and suggested that we could
> instead
> > >>> provide the group state in the heartbeat response. This has a few
> > >>> advantages: 1) it does not require using a special error code to
> > >>> signal that a new assignment is required as the signal would be the
> > >>> provided group state; 2) it removes one round trip when a client side
> > >>> assignor is used. The downside is that it makes the heartbeat
> > >>> response's definition quite large. I recall that I went with the
> > >>> current approach due to this.
> > >>>
> > >>> Providing the group state in the heartbeat response is appealing.
> What
> > >>> do you guys think?
> > >>>
> > >>> Best,
> > >>> David
> > >>>
> > >>> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > wrote:
> > >>> >
> > >>> > Hi Guozhang,
> > >>> >
> > >>> > 1. I have added a reference to the relevant chapter instead of
> > >>> > repeating the whole thing. Does that work for you?
> > >>> >
> > >>> > 2. The "Rebalance Triggers" section you are referring to is about
> > when
> > >>> > a rebalance should be triggered for the non-upgraded members using
> > the
> > >>> > old protocol. The section mentions that a rebalance must be
> triggered
> > >>> > when a new assignment is installed. This implies that the group
> epoch
> > >>> > was updated either by a native member or a non-upgraded member. For
> > >>> > the latter, the JoinGroup request would be the trigger. I have
> added
> > a
> > >>> > reference to the relevant chapter in the "JoinGroup Handling"
> section
> > >>> > as well. Does that make sense?
> > >>> >
> > >>> > Thanks,
> > >>> > David
> > >>> >
> > >>> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> > >>> wrote:
> > >>> > >
> > >>> > > Hello David,
> > >>> > >
> > >>> > > Alright I think that's sufficient. Just to make that clear in the
> > >>> doc,
> > >>> > > could we update:
> > >>> > >
> > >>> > > 1) the heartbeat request handling section, stating when
> coordinator
> > >>> will
> > >>> > > trigger rebalance based on the HB's member metadata / reason?
> > >>> > > 2) the "Rebalance Triggers" section to include what we described
> in
> > >>> "Group
> > >>> > > Epoch - Trigger a rebalance" section as well?
> > >>> > >
> > >>> > >
> > >>> > > Guozhang
> > >>> > >
> > >>> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > >>> <dj...@confluent.io.invalid>
> > >>> > > wrote:
> > >>> > >
> > >>> > > > Hi Guozhang,
> > >>> > > >
> > >>> > > > I thought that the assignor will always be consulted when the
> > next
> > >>> > > > heartbeat request is constructed. In other words,
> > >>> > > > `PartitionAssignor#metadata` will be called for every
> heartbeat.
> > >>> This
> > >>> > > > gives the opportunity for the assignor to enforce a rebalance
> by
> > >>> > > > setting the reason to a non-zero value or by changing the
> bytes.
> > Do
> > >>> > > > you think that this is not sufficient? Are you concerned by the
> > >>> delay?
> > >>> > > >
> > >>> > > > Best,
> > >>> > > > David
> > >>> > > >
> > >>> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <
> wangguoz@gmail.com
> > >
> > >>> wrote:
> > >>> > > > >
> > >>> > > > > Hello David,
> > >>> > > > >
> > >>> > > > > One of Jun's comments make me thinking:
> > >>> > > > >
> > >>> > > > > ```
> > >>> > > > > In this case, a new assignment is triggered by the client
> side
> > >>> > > > > assignor. When constructing the HB, the consumer will always
> > >>> consult
> > >>> > > > > the client side assignor and propagate the information to the
> > >>> group
> > >>> > > > > coordinator. In other words, we don't expect users to call
> > >>> > > > > Consumer#enforceRebalance anymore.
> > >>> > > > > ```
> > >>> > > > >
> > >>> > > > > As I looked at the current PartitionAssignor's interface, we
> > >>> actually do
> > >>> > > > > not have a way yet to instruct how to construct the next HB
> > >>> request, e.g.
> > >>> > > > > when the assignor wants to enforce a new rebalance with a new
> > >>> assignment,
> > >>> > > > > we'd need some customizable APIs inside the PartitionAssignor
> > to
> > >>> indicate
> > >>> > > > > the next HB telling broker about so. WDYT about adding such
> an
> > >>> API on the
> > >>> > > > > PartitionAssignor?
> > >>> > > > >
> > >>> > > > >
> > >>> > > > > Guozhang
> > >>> > > > >
> > >>> > > > >
> > >>> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > >>> <dj...@confluent.io.invalid>
> > >>> > > > > wrote:
> > >>> > > > >
> > >>> > > > > > Hi Jun,
> > >>> > > > > >
> > >>> > > > > > I have updated the KIP to include your feedback. I have
> also
> > >>> tried to
> > >>> > > > > > clarify the parts which were not cleared.
> > >>> > > > > >
> > >>> > > > > > Best,
> > >>> > > > > > David
> > >>> > > > > >
> > >>> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > >>> djacot@confluent.io>
> > >>> > > > wrote:
> > >>> > > > > > >
> > >>> > > > > > > Hi Jun,
> > >>> > > > > > >
> > >>> > > > > > > Thanks for your feedback. Let me start by answering your
> > >>> questions
> > >>> > > > > > > inline and I will update the KIP next week.
> > >>> > > > > > >
> > >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> KIP
> > >>> seem to
> > >>> > > > be
> > >>> > > > > > fewer
> > >>> > > > > > > > RPCs during rebalance and more efficient support of
> > >>> wildcard. A few
> > >>> > > > > > > > comments below.
> > >>> > > > > > >
> > >>> > > > > > > I would also add that the KIP removes the global sync
> > >>> barrier in the
> > >>> > > > > > > protocol which is essential to improve group stability
> and
> > >>> > > > > > > scalability, and the KIP also simplifies the client by
> > >>> moving most of
> > >>> > > > > > > the logic to the server side.
> > >>> > > > > > >
> > >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> support
> > >>> rolling
> > >>> > > > > > changing
> > >>> > > > > > > > of the partition assignor in the consumers?
> > >>> > > > > > >
> > >>> > > > > > > Definitely. The group coordinator will use the assignor
> > used
> > >>> by a
> > >>> > > > > > > majority of the members. This allows the group to move
> from
> > >>> one
> > >>> > > > > > > assignor to another by a roll. This is explained in the
> > >>> Assignor
> > >>> > > > > > > Selection chapter.
> > >>> > > > > > >
> > >>> > > > > > > > 30.2 For each field, could you explain whether it's
> > >>> required in
> > >>> > > > every
> > >>> > > > > > > > request or the scenarios when it needs to be filled?
> For
> > >>> example,
> > >>> > > > it's
> > >>> > > > > > not
> > >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > >>> > > > > > >
> > >>> > > > > > > The client is expected to set those fields in case of a
> > >>> connection
> > >>> > > > > > > issue (e.g. timeout) or when the fields have changed
> since
> > >>> the last
> > >>> > > > > > > HB. The server populates those fields as long as the
> member
> > >>> is not
> > >>> > > > > > > fully reconciled - the member should acknowledge that it
> > has
> > >>> the
> > >>> > > > > > > expected epoch and assignment. I will clarify this in the
> > >>> KIP.
> > >>> > > > > > >
> > >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > >>> between the
> > >>> > > > > > client
> > >>> > > > > > > > and the broker is only considered during fetching, but
> > not
> > >>> during
> > >>> > > > > > assigning
> > >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > is
> > >>> made,
> > >>> > > > there
> > >>> > > > > > is
> > >>> > > > > > > > no opportunity for read affinity because no replicas of
> > >>> assigned
> > >>> > > > > > partitions
> > >>> > > > > > > > are close to the member. I am wondering if we should
> use
> > >>> this
> > >>> > > > > > opportunity
> > >>> > > > > > > > to address this by including rack in GroupMember.
> > >>> > > > > > >
> > >>> > > > > > > That's an interesting idea. I don't see any issue with
> > >>> adding the
> > >>> > > > rack
> > >>> > > > > > > to the members. I will do so.
> > >>> > > > > > >
> > >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> > >>> busy a group
> > >>> > > > > > > > coordinator is. By moving the event loop model, it
> seems
> > >>> that we
> > >>> > > > could
> > >>> > > > > > add
> > >>> > > > > > > > a metric that tracks the fraction of the time the event
> > >>> loop is
> > >>> > > > doing
> > >>> > > > > > the
> > >>> > > > > > > > actual work.
> > >>> > > > > > >
> > >>> > > > > > > That's a great idea. I will add it. Thanks.
> > >>> > > > > > >
> > >>> > > > > > > > 33. Could we add a section on coordinator failover
> > >>> handling? For
> > >>> > > > > > example,
> > >>> > > > > > > > does it need to trigger the check if any group with the
> > >>> wildcard
> > >>> > > > > > > > subscription now has a new matching topic?
> > >>> > > > > > >
> > >>> > > > > > > Sure. When the new group coordinator takes over, it has
> to:
> > >>> > > > > > > * Setup the session timeouts.
> > >>> > > > > > > * Trigger a new assignment if a client side assignor is
> > >>> used. We
> > >>> > > > don't
> > >>> > > > > > > store the information about the member selected to run
> the
> > >>> assignment
> > >>> > > > > > > so we have to start a new one.
> > >>> > > > > > > * Update the topics metadata, verify the wildcard
> > >>> subscriptions, and
> > >>> > > > > > > trigger a rebalance if needed.
> > >>> > > > > > >
> > >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> > >>> > > > ConsumerGroupPartitionMetadataValue,
> > >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> what
> > >>> the epoch
> > >>> > > > > > field
> > >>> > > > > > > > reflects? For example, does the epoch in
> > >>> ConsumerGroupMetadataValue
> > >>> > > > > > reflect
> > >>> > > > > > > > the latest group epoch? What about the one in
> > >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > >>> > > > > > ConsumerGroupMemberMetadataValue?
> > >>> > > > > > >
> > >>> > > > > > > Sure. I will clarify that but it is always the latest
> group
> > >>> epoch.
> > >>> > > > > > > When the group state is updated, the group epoch is
> bumped
> > >>> so we use
> > >>> > > > > > > that one for all the change records related to the
> update.
> > >>> > > > > > >
> > >>> > > > > > > > 35. "the group coordinator will ensure that the
> following
> > >>> > > > invariants
> > >>> > > > > > are
> > >>> > > > > > > > met: ... All members exists." It's possible for a
> member
> > >>> not to
> > >>> > > > get any
> > >>> > > > > > > > assigned partitions, right?
> > >>> > > > > > >
> > >>> > > > > > > That's right. Here I meant that the members provided by
> the
> > >>> assignor
> > >>> > > > > > > in the assignment must exist in the group. The assignor
> can
> > >>> not make
> > >>> > > > > > > up new member ids.
> > >>> > > > > > >
> > >>> > > > > > > > 36. "He can rejoins the group with a member epoch
> equals
> > >>> to 0":
> > >>> > > > When
> > >>> > > > > > would
> > >>> > > > > > > > a consumer rejoin and what member id would be used?
> > >>> > > > > > >
> > >>> > > > > > > A member is expected to abandon all its partitions and
> > >>> rejoins when
> > >>> > > > it
> > >>> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> > >>> group
> > >>> > > > > > > coordinator will have removed the member from the group.
> > The
> > >>> member
> > >>> > > > > > > can rejoin the group with the same member id but with 0
> as
> > >>> epoch. Let
> > >>> > > > > > > me see if I can clarify this in the KIP.
> > >>> > > > > > >
> > >>> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger
> > >>> a
> > >>> > > > > > reassignment
> > >>> > > > > > > > by either providing a non-zero reason or by updating
> the
> > >>> assignor
> > >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > >>> deprecation
> > >>> > > > of
> > >>> > > > > > > > Consumer#enforeRebalance.
> > >>> > > > > > >
> > >>> > > > > > > In this case, a new assignment is triggered by the client
> > >>> side
> > >>> > > > > > > assignor. When constructing the HB, the consumer will
> > always
> > >>> consult
> > >>> > > > > > > the client side assignor and propagate the information to
> > >>> the group
> > >>> > > > > > > coordinator. In other words, we don't expect users to
> call
> > >>> > > > > > > Consumer#enforceRebalance anymore.
> > >>> > > > > > >
> > >>> > > > > > > > 38. The reassignment examples are nice. But the section
> > >>> seems to
> > >>> > > > have
> > >>> > > > > > > > multiple typos.
> > >>> > > > > > > > 38.1 When the group transitions to epoch 2, B
> immediately
> > >>> gets into
> > >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > >>> into
> > >>> > > > > > epoch=3,
> > >>> > > > > > > > partitions=[foo-1] too early.
> > >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > >>> epoch=2,
> > >>> > > > > > > > partitions=[foo-0].
> > >>> > > > > > >
> > >>> > > > > > > Sorry for that! I will revise them.
> > >>> > > > > > >
> > >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > >>> upgrade from
> > >>> > > > any
> > >>> > > > > > old
> > >>> > > > > > > > version to new one?
> > >>> > > > > > >
> > >>> > > > > > > We will support upgrading from the consumer protocol
> > version
> > >>> 3,
> > >>> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> > the
> > >>> earliest
> > >>> > > > > > > version is unknown at the moment. This is explained in
> the
> > >>> migration
> > >>> > > > > > > plan chapter.
> > >>> > > > > > >
> > >>> > > > > > > Thanks again for your feedback, Jun. I will update the
> KIP
> > >>> based on
> > >>> > > > it
> > >>> > > > > > > next week.
> > >>> > > > > > >
> > >>> > > > > > > Best,
> > >>> > > > > > > David
> > >>> > > > > > >
> > >>> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > >>> <ju...@confluent.io.invalid>
> > >>> > > > wrote:
> > >>> > > > > > > >
> > >>> > > > > > > > Hi, David,
> > >>> > > > > > > >
> > >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the
> KIP
> > >>> seem to
> > >>> > > > be
> > >>> > > > > > fewer
> > >>> > > > > > > > RPCs during rebalance and more efficient support of
> > >>> wildcard. A few
> > >>> > > > > > > > comments below.
> > >>> > > > > > > >
> > >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to
> support
> > >>> rolling
> > >>> > > > > > changing
> > >>> > > > > > > > of the partition assignor in the consumers?
> > >>> > > > > > > > 30.2 For each field, could you explain whether it's
> > >>> required in
> > >>> > > > every
> > >>> > > > > > > > request or the scenarios when it needs to be filled?
> For
> > >>> example,
> > >>> > > > it's
> > >>> > > > > > not
> > >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > >>> > > > > > > >
> > >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > >>> between the
> > >>> > > > > > client
> > >>> > > > > > > > and the broker is only considered during fetching, but
> > not
> > >>> during
> > >>> > > > > > assigning
> > >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > is
> > >>> made,
> > >>> > > > there
> > >>> > > > > > is
> > >>> > > > > > > > no opportunity for read affinity because no replicas of
> > >>> assigned
> > >>> > > > > > partitions
> > >>> > > > > > > > are close to the member. I am wondering if we should
> use
> > >>> this
> > >>> > > > > > opportunity
> > >>> > > > > > > > to address this by including rack in GroupMember.
> > >>> > > > > > > >
> > >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> > >>> busy a group
> > >>> > > > > > > > coordinator is. By moving the event loop model, it
> seems
> > >>> that we
> > >>> > > > could
> > >>> > > > > > add
> > >>> > > > > > > > a metric that tracks the fraction of the time the event
> > >>> loop is
> > >>> > > > doing
> > >>> > > > > > the
> > >>> > > > > > > > actual work.
> > >>> > > > > > > >
> > >>> > > > > > > > 33. Could we add a section on coordinator failover
> > >>> handling? For
> > >>> > > > > > example,
> > >>> > > > > > > > does it need to trigger the check if any group with the
> > >>> wildcard
> > >>> > > > > > > > subscription now has a new matching topic?
> > >>> > > > > > > >
> > >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> > >>> > > > ConsumerGroupPartitionMetadataValue,
> > >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document
> what
> > >>> the epoch
> > >>> > > > > > field
> > >>> > > > > > > > reflects? For example, does the epoch in
> > >>> ConsumerGroupMetadataValue
> > >>> > > > > > reflect
> > >>> > > > > > > > the latest group epoch? What about the one in
> > >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > >>> > > > > > ConsumerGroupMemberMetadataValue?
> > >>> > > > > > > >
> > >>> > > > > > > > 35. "the group coordinator will ensure that the
> following
> > >>> > > > invariants
> > >>> > > > > > are
> > >>> > > > > > > > met: ... All members exists." It's possible for a
> member
> > >>> not to
> > >>> > > > get any
> > >>> > > > > > > > assigned partitions, right?
> > >>> > > > > > > >
> > >>> > > > > > > > 36. "He can rejoins the group with a member epoch
> equals
> > >>> to 0":
> > >>> > > > When
> > >>> > > > > > would
> > >>> > > > > > > > a consumer rejoin and what member id would be used?
> > >>> > > > > > > >
> > >>> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger
> > >>> a
> > >>> > > > > > reassignment
> > >>> > > > > > > > by either providing a non-zero reason or by updating
> the
> > >>> assignor
> > >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > >>> deprecation
> > >>> > > > of
> > >>> > > > > > > > Consumer#enforeRebalance.
> > >>> > > > > > > >
> > >>> > > > > > > > 38. The reassignment examples are nice. But the section
> > >>> seems to
> > >>> > > > have
> > >>> > > > > > > > multiple typos.
> > >>> > > > > > > > 38.1 When the group transitions to epoch 2, B
> immediately
> > >>> gets into
> > >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > >>> into
> > >>> > > > > > epoch=3,
> > >>> > > > > > > > partitions=[foo-1] too early.
> > >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > >>> epoch=2,
> > >>> > > > > > > > partitions=[foo-0].
> > >>> > > > > > > >
> > >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > >>> upgrade from
> > >>> > > > any
> > >>> > > > > > old
> > >>> > > > > > > > version to new one?
> > >>> > > > > > > >
> > >>> > > > > > > > Thanks,
> > >>> > > > > > > >
> > >>> > > > > > > > Jun
> > >>> > > > > > > >
> > >>> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > >>> > > > > > <dj...@confluent.io.invalid>
> > >>> > > > > > > > wrote:
> > >>> > > > > > > >
> > >>> > > > > > > > > Hi all,
> > >>> > > > > > > > >
> > >>> > > > > > > > > The KIP states that we will re-implement the
> > coordinator
> > >>> in
> > >>> > > > Java. I
> > >>> > > > > > > > > discussed this offline with a few folks and folks are
> > >>> concerned
> > >>> > > > that
> > >>> > > > > > > > > we could introduce many regressions in the old
> protocol
> > >>> if we do
> > >>> > > > so.
> > >>> > > > > > > > > Therefore, I am going to remove this statement from
> the
> > >>> KIP. It
> > >>> > > > is an
> > >>> > > > > > > > > implementation detail after all so it does not have
> to
> > be
> > >>> > > > decided at
> > >>> > > > > > > > > this stage. We will likely start by trying to
> refactor
> > >>> the
> > >>> > > > current
> > >>> > > > > > > > > implementation as a first step.
> > >>> > > > > > > > >
> > >>> > > > > > > > > Cheers,
> > >>> > > > > > > > > David
> > >>> > > > > > > > >
> > >>> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > >>> djacot@confluent.io
> > >>> > > > >
> > >>> > > > > > wrote:
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > Hi Luke,
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > >>> assigning,
> > >>> > > > > > reconciling,
> > >>> > > > > > > > > stable,
> > >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > >>> right?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > This sentence does not refer to those group states
> > but
> > >>> rather
> > >>> > > > to a
> > >>> > > > > > > > > > state machine replication (SMR). This refers to the
> > >>> entire
> > >>> > > > state of
> > >>> > > > > > > > > > group coordinator which is replicated via the log
> > >>> layer. I will
> > >>> > > > > > > > > > clarify this in the KIP.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > >>> modelled as an
> > >>> > > > event
> > >>> > > > > > > > > loop"?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > The idea is to follow a model similar to the new
> > quorum
> > >>> > > > > > controller. We
> > >>> > > > > > > > > > will have N threads to process events. Each
> > >>> __consumer_offsets
> > >>> > > > > > > > > > partition is assigned to a unique thread and all
> the
> > >>> events
> > >>> > > > (e.g.
> > >>> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > >>> thread. This
> > >>> > > > > > simplify
> > >>> > > > > > > > > > concurrency and will enable us to do simulation
> > >>> testing for the
> > >>> > > > > > group
> > >>> > > > > > > > > > coordinator.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> > >>> *__consumer_offsets*
> > >>> > > > > > > > > partitions?
> > >>> > > > > > > > > > > Not a state machine "per consumer group" owned
> by a
> > >>> group
> > >>> > > > > > coordinator?
> > >>> > > > > > > > > For
> > >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > >>> groups, and
> > >>> > > > > > both
> > >>> > > > > > > > > exist in
> > >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > >>> for it,
> > >>> > > > or
> > >>> > > > > > 2?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > >>> number
> > >>> > > > of
> > >>> > > > > > threads
> > >>> > > > > > > > > used
> > >>> > > > > > > > > > > to run the state machines. But I'm wondering if
> the
> > >>> purpose
> > >>> > > > of
> > >>> > > > > > the
> > >>> > > > > > > > > threads
> > >>> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > >>> > > > > > > > > *__consumer_offsets*
> > >>> > > > > > > > > > > partitions?), and no heavy computation, why
> should
> > >>> we need
> > >>> > > > > > > > > multi-threads
> > >>> > > > > > > > > > > here?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > See 1.2. The idea is to have an ability to shard
> the
> > >>> > > > processing as
> > >>> > > > > > the
> > >>> > > > > > > > > > computation could be heavy.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > >>> default
> > >>> > > > session
> > >>> > > > > > > > > timeout not
> > >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > >>> min/max
> > >>> > > > > > session
> > >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > >>> timeout to
> > >>> > > > detect
> > >>> > > > > > client
> > >>> > > > > > > > > > > failures when using the consumer group protocol.
> > >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> The
> > >>> minimum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> The
> > >>> maximum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > This is indeed a mistake. The default session
> timeout
> > >>> should
> > >>> > > > be 45s
> > >>> > > > > > > > > > (the current default).
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > >>> uniform],
> > >>> > > > > > which means
> > >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> know
> > >>> why not
> > >>> > > > > > uniform
> > >>> > > > > > > > > one? I
> > >>> > > > > > > > > > > thought usually users will choose uniform
> assignor
> > >>> (former
> > >>> > > > sticky
> > >>> > > > > > > > > assinor)
> > >>> > > > > > > > > > > for better evenly distribution. Any other reason
> we
> > >>> choose
> > >>> > > > range
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > as default?
> > >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > >>> server side
> > >>> > > > > > assignors.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > The order on the server side has no influence
> because
> > >>> the
> > >>> > > > client
> > >>> > > > > > must
> > >>> > > > > > > > > > chose the selector that he wants to use. There is
> no
> > >>> default
> > >>> > > > in the
> > >>> > > > > > > > > > current proposal. If the assignor is not specified
> by
> > >>> the
> > >>> > > > client,
> > >>> > > > > > the
> > >>> > > > > > > > > > request is rejected. The default client value for
> > >>> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > Thanks for your very good comments, Luke. I hope
> that
> > >>> my
> > >>> > > > answers
> > >>> > > > > > help
> > >>> > > > > > > > > > to clarify things. I will update the KIP as well
> > based
> > >>> on your
> > >>> > > > > > > > > > feedback.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > Cheers,
> > >>> > > > > > > > > > David
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > >>> showuon@gmail.com>
> > >>> > > > > > wrote:
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Hi David,
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Thanks for the update.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Some more questions:
> > >>> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > >>> > > > > > > > > > > > The new group coordinator will have a state
> > >>> machine per
> > >>> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> > >>> machine is
> > >>> > > > > > modelled
> > >>> > > > > > > > > as an
> > >>> > > > > > > > > > > event loop. Those state machines will be executed
> > in
> > >>> > > > > > > > > > > *group.coordinator.threads* threads.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > >>> assigning,
> > >>> > > > > > reconciling,
> > >>> > > > > > > > > stable,
> > >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > >>> right?
> > >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > >>> modelled as an
> > >>> > > > event
> > >>> > > > > > > > > loop"?
> > >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> > >>> *__consumer_offsets*
> > >>> > > > > > > > > partitions?
> > >>> > > > > > > > > > > Not a state machine "per consumer group" owned
> by a
> > >>> group
> > >>> > > > > > coordinator?
> > >>> > > > > > > > > For
> > >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > >>> groups, and
> > >>> > > > > > both
> > >>> > > > > > > > > exist in
> > >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > >>> for it,
> > >>> > > > or
> > >>> > > > > > 2?
> > >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > >>> number
> > >>> > > > of
> > >>> > > > > > threads
> > >>> > > > > > > > > used
> > >>> > > > > > > > > > > to run the state machines. But I'm wondering if
> the
> > >>> purpose
> > >>> > > > of
> > >>> > > > > > the
> > >>> > > > > > > > > threads
> > >>> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > >>> > > > > > > > > *__consumer_offsets*
> > >>> > > > > > > > > > > partitions?), and no heavy computation, why
> should
> > >>> we need
> > >>> > > > > > > > > multi-threads
> > >>> > > > > > > > > > > here?
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > 2. For the default value in the new configs:
> > >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > >>> default
> > >>> > > > session
> > >>> > > > > > > > > timeout not
> > >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > >>> min/max
> > >>> > > > > > session
> > >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > >>> timeout to
> > >>> > > > detect
> > >>> > > > > > client
> > >>> > > > > > > > > > > failures when using the consumer group protocol.
> > >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s
> The
> > >>> minimum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s
> The
> > >>> maximum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > >>> uniform],
> > >>> > > > > > which means
> > >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to
> know
> > >>> why not
> > >>> > > > > > uniform
> > >>> > > > > > > > > one? I
> > >>> > > > > > > > > > > thought usually users will choose uniform
> assignor
> > >>> (former
> > >>> > > > sticky
> > >>> > > > > > > > > assinor)
> > >>> > > > > > > > > > > for better evenly distribution. Any other reason
> we
> > >>> choose
> > >>> > > > range
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > as default?
> > >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > >>> server side
> > >>> > > > > > assignors.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Thank you.
> > >>> > > > > > > > > > > Luke
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > >>> showuon@gmail.com
> > >>> > > > >
> > >>> > > > > > wrote:
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > > Hi Sagar,
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > >>> integrating with
> > >>> > > > > > KIP-848,
> > >>> > > > > > > > > but I
> > >>> > > > > > > > > > > > think we should have a separate discussion
> thread
> > >>> for the
> > >>> > > > Kafka
> > >>> > > > > > > > > Connect
> > >>> > > > > > > > > > > > KIP: Integrating Kafka Connect With New
> Consumer
> > >>> Rebalance
> > >>> > > > > > Protocol
> > >>> > > > > > > > > [1],
> > >>> > > > > > > > > > > > and let this discussion thread focus on
> consumer
> > >>> rebalance
> > >>> > > > > > protocol,
> > >>> > > > > > > > > WDYT?
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > [1]
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > >
> > >>>
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > Thank you.
> > >>> > > > > > > > > > > > Luke
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > >>> > > > > > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > wrote:
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> Thank you Guozhang/David for the feedback.
> Looks
> > >>> like
> > >>> > > > there's
> > >>> > > > > > > > > agreement on
> > >>> > > > > > > > > > > >> using separate APIs for Connect. I would
> revisit
> > >>> the doc
> > >>> > > > and
> > >>> > > > > > see
> > >>> > > > > > > > > what
> > >>> > > > > > > > > > > >> changes are to be made.
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >> Thanks!
> > >>> > > > > > > > > > > >> Sagar.
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > >>> > > > > > > > > <dj...@confluent.io.invalid>
> > >>> > > > > > > > > > > >> wrote:
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >> > Hi Sagar,
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > >>> That's really
> > >>> > > > > > helpful. I
> > >>> > > > > > > > > > > >> > will take a look at it.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > Overall, it seems to me that both Connect
> and
> > >>> the
> > >>> > > > Consumer
> > >>> > > > > > could
> > >>> > > > > > > > > share
> > >>> > > > > > > > > > > >> > the same underlying "engine". The main
> > >>> difference is
> > >>> > > > that
> > >>> > > > > > the
> > >>> > > > > > > > > Consumer
> > >>> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > >>> Connect
> > >>> > > > assigns
> > >>> > > > > > tasks
> > >>> > > > > > > > > to
> > >>> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > >>> > > > > > > > > > > >> > 1) We extend the new proposed APIs to
> support
> > >>> different
> > >>> > > > > > resource
> > >>> > > > > > > > > types
> > >>> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > >>> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect.
> The
> > >>> dedicated
> > >>> > > > APIs
> > >>> > > > > > > > > would be
> > >>> > > > > > > > > > > >> > similar to the new ones but different on the
> > >>> > > > > > content/resources and
> > >>> > > > > > > > > > > >> > they would rely on the same engine on the
> > >>> coordinator
> > >>> > > > side.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > I personally lean towards 2) because I am
> not
> > a
> > >>> fan of
> > >>> > > > > > > > > overcharging
> > >>> > > > > > > > > > > >> > APIs to serve different purposes. That being
> > >>> said, I am
> > >>> > > > not
> > >>> > > > > > > > > opposed to
> > >>> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > I think that we can continue to discuss it
> > here
> > >>> for now
> > >>> > > > in
> > >>> > > > > > order
> > >>> > > > > > > > > to
> > >>> > > > > > > > > > > >> > ensure that this KIP is compatible with what
> > we
> > >>> will do
> > >>> > > > for
> > >>> > > > > > > > > Connect in
> > >>> > > > > > > > > > > >> > the future.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > Best,
> > >>> > > > > > > > > > > >> > David
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > >>> > > > > > djacot@confluent.io>
> > >>> > > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > Hi all,
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > I am back from vacation. I will go through
> > >>> and address
> > >>> > > > > > your
> > >>> > > > > > > > > comments
> > >>> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > feedback.
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > Cheers,
> > >>> > > > > > > > > > > >> > > David
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > >>> Harris <
> > >>> > > > > > > > > gharris1727@gmail.com
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > Hey All!
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to
> see
> > >>> > > > cooperative
> > >>> > > > > > > > > rebalancing
> > >>> > > > > > > > > > > >> > making it
> > >>> > > > > > > > > > > >> > > > down the stack!
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > I had a few questions:
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > >>> describes how
> > >>> > > > > > member
> > >>> > > > > > > > > epoch
> > >>> > > > > > > > > > > >> > should
> > >>> > > > > > > > > > > >> > > > advance in step with the group epoch and
> > >>> assignment
> > >>> > > > > > epoch
> > >>> > > > > > > > > values. I
> > >>> > > > > > > > > > > >> > think
> > >>> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> > >>> described
> > >>> > > > in
> > >>> > > > > > the
> > >>> > > > > > > > > KIP. When
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > >>> partitions to a
> > >>> > > > > > worker,
> > >>> > > > > > > > > what
> > >>> > > > > > > > > > > >> member
> > >>> > > > > > > > > > > >> > > > epoch does each incremental assignment
> > use?
> > >>> Are
> > >>> > > > member
> > >>> > > > > > epochs
> > >>> > > > > > > > > > > >> re-used,
> > >>> > > > > > > > > > > >> > and
> > >>> > > > > > > > > > > >> > > > a single member epoch can correspond to
> > >>> multiple
> > >>> > > > > > different
> > >>> > > > > > > > > > > >> > (monotonically
> > >>> > > > > > > > > > > >> > > > larger) assignments?
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field
> opaque
> > >>> to the
> > >>> > > > group
> > >>> > > > > > > > > > > >> coordinator? If
> > >>> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > >>> > > > implementations
> > >>> > > > > > > > > interact
> > >>> > > > > > > > > > > >> with
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > Reason field, and how is its common
> > meaning
> > >>> agreed
> > >>> > > > > > upon? If
> > >>> > > > > > > > > so, what
> > >>> > > > > > > > > > > >> > is the
> > >>> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > >>> including
> > >>> > > > such
> > >>> > > > > > > > > functionality
> > >>> > > > > > > > > > > >> > in the
> > >>> > > > > > > > > > > >> > > > opaque metadata?
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> > >>> "Thanks to
> > >>> > > > > > this, the
> > >>> > > > > > > > > input
> > >>> > > > > > > > > > > >> of
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > client side assignor is entirely driven
> by
> > >>> the group
> > >>> > > > > > > > > coordinator.
> > >>> > > > > > > > > > > >> The
> > >>> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > >>> maintaining
> > >>> > > > any
> > >>> > > > > > state
> > >>> > > > > > > > > besides
> > >>> > > > > > > > > > > >> its
> > >>> > > > > > > > > > > >> > > > assigned partitions." Does this mean
> that
> > >>> the
> > >>> > > > > > client-side
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > >> MAY
> > >>> > > > > > > > > > > >> > > > incorporate additional non-Metadata
> state
> > >>> (such as
> > >>> > > > > > partition
> > >>> > > > > > > > > > > >> > throughput,
> > >>> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc),
> > or
> > >>> that
> > >>> > > > > > additional
> > >>> > > > > > > > > > > >> > non-Metadata
> > >>> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> > >>> > > > > > > > > > > >> > > > for
> > >>> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > >>> > > > > > > > > > > >> > > > and
> > >>> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > >>> > > > > > that
> > >>> > > > > > > > > seem to
> > >>> > > > > > > > > > > >> > > > overlap significantly. Is it possible
> for
> > >>> these two
> > >>> > > > > > > > > implementations
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > unified? This would serve to promote
> > >>> feature parity
> > >>> > > > of
> > >>> > > > > > > > > server-side
> > >>> > > > > > > > > > > >> and
> > >>> > > > > > > > > > > >> > > > client-side assignors, and would also
> > >>> facilitate
> > >>> > > > > > operational
> > >>> > > > > > > > > > > >> > flexibility in
> > >>> > > > > > > > > > > >> > > > certain situations. For example, if a
> > >>> server-side
> > >>> > > > > > assignor
> > >>> > > > > > > > > has some
> > >>> > > > > > > > > > > >> > poor
> > >>> > > > > > > > > > > >> > > > behavior and needs a patch, deploying
> the
> > >>> patched
> > >>> > > > > > assignor to
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > client
> > >>> > > > > > > > > > > >> > > > and switching one consumer group to a
> > >>> client-side
> > >>> > > > > > assignor
> > >>> > > > > > > > > may be
> > >>> > > > > > > > > > > >> > faster
> > >>> > > > > > > > > > > >> > > > and less risky than patching all of the
> > >>> brokers.
> > >>> > > > With
> > >>> > > > > > the
> > >>> > > > > > > > > currently
> > >>> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > >>> > > > reimplementation
> > >>> > > > > > would
> > >>> > > > > > > > > have
> > >>> > > > > > > > > > > >> to be
> > >>> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > diverged
> > >>> > > > > > significantly,
> > >>> > > > > > > > > then it
> > >>> > > > > > > > > > > >> is
> > >>> > > > > > > > > > > >> > > > possible that a reimplementation would
> not
> > >>> be
> > >>> > > > possible.
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > --
> > >>> > > > > > > > > > > >> > > > Greg Harris
> > >>> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > >>> > > > > > > > > > > >> > > > github.com/gharris1727
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > >>> > > > > > > > > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > I created a confluence page to discuss
> > >>> how Connect
> > >>> > > > > > would
> > >>> > > > > > > > > need to
> > >>> > > > > > > > > > > >> > change
> > >>> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> > >>> Here's the
> > >>> > > > page:
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > >
> > >>>
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > tried
> > >>> to keep
> > >>> > > > a
> > >>> > > > > > format
> > >>> > > > > > > > > > > >> similar to
> > >>> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> > >>> Also, do you
> > >>> > > > > > think this
> > >>> > > > > > > > > > > >> should
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > > moved to a separate discussion thread
> or
> > >>> is this
> > >>> > > > one
> > >>> > > > > > fine?
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > Thanks!
> > >>> > > > > > > > > > > >> > > > > Sagar.
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar
> <
> > >>> > > > > > > > > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > > Thank you so much for the doc on
> Kafka
> > >>> Streams.
> > >>> > > > > > Sure, I
> > >>> > > > > > > > > would do
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > >>> document.
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > > Thanks!
> > >>> > > > > > > > > > > >> > > > > > Sagar.
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > >>> Guozhang Wang <
> > >>> > > > > > > > > > > >> wangguoz@gmail.com>
> > >>> > > > > > > > > > > >> > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> It would be great if you could come
> > >>> back with
> > >>> > > > some
> > >>> > > > > > > > > analysis on
> > >>> > > > > > > > > > > >> > how to
> > >>> > > > > > > > > > > >> > > > > >> implement the Connect side
> > integration
> > >>> with
> > >>> > > > the new
> > >>> > > > > > > > > protocol;
> > >>> > > > > > > > > > > >> so
> > >>> > > > > > > > > > > >> > far
> > >>> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > >>> "protocol type"
> > >>> > > > we
> > >>> > > > > > did not
> > >>> > > > > > > > > yet
> > >>> > > > > > > > > > > >> think
> > >>> > > > > > > > > > > >> > > > > through
> > >>> > > > > > > > > > > >> > > > > >> the Connect side implementations.
> For
> > >>> Streams
> > >>> > > > > > here's a
> > >>> > > > > > > > > draft of
> > >>> > > > > > > > > > > >> > > > > >> integration
> > >>> > > > > > > > > > > >> > > > > >> plan:
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > >
> > >>>
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > >>> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> > Connect.
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> > Sagar
> > >>> <
> > >>> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> > Hi David,
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> > >>> reason I
> > >>> > > > thought
> > >>> > > > > > > > > connect can
> > >>> > > > > > > > > > > >> > also fit
> > >>> > > > > > > > > > > >> > > > > >> into
> > >>> > > > > > > > > > > >> > > > > >> > this new scheme is that even
> today
> > >>> the
> > >>> > > > connect
> > >>> > > > > > uses a
> > >>> > > > > > > > > > > >> > > > > WorkerCoordinator
> > >>> > > > > > > > > > > >> > > > > >> > extending from
> AbstractCoordinator
> > >>> to empower
> > >>> > > > > > > > > rebalances of
> > >>> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > >>> WorkerCoordinator sets
> > >>> > > > the
> > >>> > > > > > > > > > > >> protocolType()
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> connect
> > >>> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> > >>> plumbing
> > >>> > > > into
> > >>> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > >>> connect would
> > >>> > > > be
> > >>> > > > > > > > > similar at a
> > >>> > > > > > > > > > > >> > high
> > >>> > > > > > > > > > > >> > > > > level
> > >>> > > > > > > > > > > >> > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> > >>> because of the
> > >>> > > > > > Client
> > >>> > > > > > > > > side
> > >>> > > > > > > > > > > >> > assignors
> > >>> > > > > > > > > > > >> > > > > being
> > >>> > > > > > > > > > > >> > > > > >> > used in both. At an
> implementation
> > >>> level, we
> > >>> > > > > > might
> > >>> > > > > > > > > need to
> > >>> > > > > > > > > > > >> make
> > >>> > > > > > > > > > > >> > a lot
> > >>> > > > > > > > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > >>> assignment
> > >>> > > > protocol
> > >>> > > > > > like
> > >>> > > > > > > > > > > >> enhancing
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > >>> SyncGroup and
> > >>> > > > > > using
> > >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > >>> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> > >>> streams (or
> > >>> > > > > > there
> > >>> > > > > > > > > might be
> > >>> > > > > > > > > > > >> > > > > >> deviations). I
> > >>> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> > >>> analysis of
> > >>> > > > the
> > >>> > > > > > same
> > >>> > > > > > > > > and we
> > >>> > > > > > > > > > > >> > can have
> > >>> > > > > > > > > > > >> > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > separate discussion thread for
> that
> > >>> as that
> > >>> > > > would
> > >>> > > > > > > > > derail this
> > >>> > > > > > > > > > > >> > > > > discussion
> > >>> > > > > > > > > > > >> > > > > >> > thread. Let me know if that
> sounds
> > >>> good to
> > >>> > > > you.
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > Thanks!
> > >>> > > > > > > > > > > >> > > > > >> > Sagar.
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> > >>> David Jacot
> > >>> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > >>> `Assignment#error`.
> > >>> > > > > > Sure, I
> > >>> > > > > > > > > can
> > >>> > > > > > > > > > > >> > mention it.
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> > >>> from his
> > >>> > > > current
> > >>> > > > > > > > > assignment
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > his
> > >>> > > > > > > > > > > >> > > > > >> > > target assignment when he can
> > move
> > >>> to
> > >>> > > > epoch 3.
> > >>> > > > > > When
> > >>> > > > > > > > > that
> > >>> > > > > > > > > > > >> > happens,
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > member assignment is updated
> and
> > >>> persisted
> > >>> > > > > > with all
> > >>> > > > > > > > > its
> > >>> > > > > > > > > > > >> > assigned
> > >>> > > > > > > > > > > >> > > > > >> > > partitions even if they are not
> > >>> all revoked
> > >>> > > > > > yet. In
> > >>> > > > > > > > > other
> > >>> > > > > > > > > > > >> > words, the
> > >>> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> > >>> target
> > >>> > > > > > assignment.
> > >>> > > > > > > > > This is
> > >>> > > > > > > > > > > >> > basically
> > >>> > > > > > > > > > > >> > > > > an
> > >>> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> > >>> write all
> > >>> > > > the
> > >>> > > > > > > > > changes to
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > log.
> > >>> > > > > > > > > > > >> > > > > The
> > >>> > > > > > > > > > > >> > > > > >> > > examples are based on the
> > >>> persisted state
> > >>> > > > so I
> > >>> > > > > > > > > understand
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> > >>> improve
> > >>> > > > this in
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> > description.
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> > >>> reuse the
> > >>> > > > > > protocol
> > >>> > > > > > > > > with a
> > >>> > > > > > > > > > > >> > client side
> > >>> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > >>> protocol. The
> > >>> > > > > > assignment
> > >>> > > > > > > > > is
> > >>> > > > > > > > > > > >> about
> > >>> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> > >>> could
> > >>> > > > Connect
> > >>> > > > > > fit
> > >>> > > > > > > > > into this?
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > David
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> > >>> Sagar <
> > >>> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > > > >> > > > > >> wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just
> had
> > >>> minor
> > >>> > > > > > observations:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > >>> section in
> > >>> > > > Client
> > >>> > > > > > Side
> > >>> > > > > > > > > mode
> > >>> > > > > > > > > > > >> > Assignment
> > >>> > > > > > > > > > > >> > > > > >> > > process,
> > >>> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this
> case,
> > >>> the
> > >>> > > > client
> > >>> > > > > > side
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > >> can
> > >>> > > > > > > > > > > >> > > > > return
> > >>> > > > > > > > > > > >> > > > > >> an
> > >>> > > > > > > > > > > >> > > > > >> > > > error to the group
> > coordinator`.
> > >>> In this
> > >>> > > > > > case are
> > >>> > > > > > > > > you
> > >>> > > > > > > > > > > >> > referring to
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > >>> AssignmentError
> > >>> > > > that's
> > >>> > > > > > > > > listed down
> > >>> > > > > > > > > > > >> > towards
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > end?
> > >>> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
> > >>> make sense
> > >>> > > > to
> > >>> > > > > > > > > mention this
> > >>> > > > > > > > > > > >> > > > > explicitly
> > >>> > > > > > > > > > > >> > > > > >> > here?
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies
> section,
> > >>> I have a
> > >>> > > > > > slight
> > >>> > > > > > > > > > > >> confusion,
> > >>> > > > > > > > > > > >> > not
> > >>> > > > > > > > > > > >> > > > > sure
> > >>> > > > > > > > > > > >> > > > > >> if
> > >>> > > > > > > > > > > >> > > > > >> > > > others have the same.
> Consider
> > >>> this step:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> > >>> coordinator
> > >>> > > > > > > > > transitions him
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > epoch
> > >>> > > > > > > > > > > >> > > > > 3
> > >>> > > > > > > > > > > >> > > > > >> > > because
> > >>> > > > > > > > > > > >> > > > > >> > > > B has no partitions to
> revoke.
> > It
> > >>> > > > persists
> > >>> > > > > > the
> > >>> > > > > > > > > change and
> > >>> > > > > > > > > > > >> > reply.
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > >>> > > > > > > > > > > >> > > > > >> > > >       - A
> > >>> > > > > > > > > > > >> > > > > >> > > >       - B
> > >>> > > > > > > > > > > >> > > > > >> > > >       - C
> > >>> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > (epoch=3)
> > >>> > > > > > > > > > > >> > > > > >> > > >       - A -
> partitions=[foo-0]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - B -
> partitions=[foo-2]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - C -
> partitions=[foo-1]
> > >>> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > >>> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > >>> partitions=[foo-0,
> > >>> > > > > > foo-1]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > >>> partitions=[foo-2]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > >>> partitions=[foo-1]
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > >>> transitions to
> > >>> > > > epoch 3
> > >>> > > > > > but
> > >>> > > > > > > > > cannot
> > >>> > > > > > > > > > > >> get
> > >>> > > > > > > > > > > >> > foo-1
> > >>> > > > > > > > > > > >> > > > > >> yet.
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that
> member
> > >>> C can't
> > >>> > > > get
> > >>> > > > > > the
> > >>> > > > > > > > > foo-1
> > >>> > > > > > > > > > > >> > partition
> > >>> > > > > > > > > > > >> > > > > yet,
> > >>> > > > > > > > > > > >> > > > > >> > but
> > >>> > > > > > > > > > > >> > > > > >> > > > based on the description
> above,
> > >>> it seems
> > >>> > > > it
> > >>> > > > > > > > > already has
> > >>> > > > > > > > > > > >> it.
> > >>> > > > > > > > > > > >> > Do you
> > >>> > > > > > > > > > > >> > > > > >> > think
> > >>> > > > > > > > > > > >> > > > > >> > > it
> > >>> > > > > > > > > > > >> > > > > >> > > > would be better to remove it
> > and
> > >>> > > > populate it
> > >>> > > > > > only
> > >>> > > > > > > > > when it
> > >>> > > > > > > > > > > >> > actually
> > >>> > > > > > > > > > > >> > > > > >> gets
> > >>> > > > > > > > > > > >> > > > > >> > > it?
> > >>> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> > >>> places, so
> > >>> > > > have
> > >>> > > > > > I
> > >>> > > > > > > > > > > >> understood it
> > >>> > > > > > > > > > > >> > > > > >> > incorrectly
> > >>> > > > > > > > > > > >> > > > > >> > > ?
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might
> be
> > >>> out of
> > >>> > > > scope
> > >>> > > > > > of
> > >>> > > > > > > > > this
> > >>> > > > > > > > > > > >> > discussion,
> > >>> > > > > > > > > > > >> > > > > but
> > >>> > > > > > > > > > > >> > > > > >> > from
> > >>> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > >>> probably be
> > >>> > > > > > running in
> > >>> > > > > > > > > client
> > >>> > > > > > > > > > > >> > side
> > >>> > > > > > > > > > > >> > > > > >> assignor
> > >>> > > > > > > > > > > >> > > > > >> > > mode
> > >>> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > >>> protocol as it
> > >>> > > > has
> > >>> > > > > > its
> > >>> > > > > > > > > own
> > >>> > > > > > > > > > > >> Custom
> > >>> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > >>> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00
> PM
> > >>> David
> > >>> > > > Jacot
> > >>> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is
> to
> > >>> move
> > >>> > > > forward
> > >>> > > > > > with
> > >>> > > > > > > > > > > >> > specialized API
> > >>> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> > >>> generic API.
> > >>> > > > For
> > >>> > > > > > > > > Connect, we
> > >>> > > > > > > > > > > >> > can apply
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > >>> reuse/share the
> > >>> > > > core
> > >>> > > > > > > > > > > >> > implementation on
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> > >>> registry, I
> > >>> > > > > > think
> > >>> > > > > > > > > that we
> > >>> > > > > > > > > > > >> > should
> > >>> > > > > > > > > > > >> > > > > >> consider
> > >>> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> > >>> simple
> > >>> > > > > > > > > membership/leader
> > >>> > > > > > > > > > > >> > election.
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at
> 10:22
> > >>> AM Ismael
> > >>> > > > > > Juma <
> > >>> > > > > > > > > > > >> > ismael@juma.me.uk
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > >>> java.util.regex.Pattern
> > >>> > > > vs
> > >>> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > document the differences
> in
> > >>> more
> > >>> > > > detail
> > >>> > > > > > before
> > >>> > > > > > > > > > > >> deciding
> > >>> > > > > > > > > > > >> > one
> > >>> > > > > > > > > > > >> > > > > way
> > >>> > > > > > > > > > > >> > > > > >> or
> > >>> > > > > > > > > > > >> > > > > >> > > > > another.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > >>> > > > > > > > > java.util.regex.Pattern,
> > >>> > > > > > > > > > > >> they
> > >>> > > > > > > > > > > >> > expect
> > >>> > > > > > > > > > > >> > > > > >> > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored.
> If
> > >>> we are
> > >>> > > > doing
> > >>> > > > > > > > > something
> > >>> > > > > > > > > > > >> > different,
> > >>> > > > > > > > > > > >> > > > > >> then
> > >>> > > > > > > > > > > >> > > > > >> > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> > >>> overload
> > >>> > > > with
> > >>> > > > > > our own
> > >>> > > > > > > > > > > >> Pattern
> > >>> > > > > > > > > > > >> > class
> > >>> > > > > > > > > > > >> > > > > (I
> > >>> > > > > > > > > > > >> > > > > >> > > don't
> > >>> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> > >>> re2j's at
> > >>> > > > this
> > >>> > > > > > > > > point).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids,
> any
> > >>> major new
> > >>> > > > > > protocol
> > >>> > > > > > > > > should
> > >>> > > > > > > > > > > >> > > > > integrate
> > >>> > > > > > > > > > > >> > > > > >> > fully
> > >>> > > > > > > > > > > >> > > > > >> > > > > with
> > >>> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> > >>> topic
> > >>> > > > > > recreation case
> > >>> > > > > > > > > > > >> > correctly.
> > >>> > > > > > > > > > > >> > > > > That's
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > main
> > >>> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> > >>> agree with
> > >>> > > > > > David
> > >>> > > > > > > > > that we'd
> > >>> > > > > > > > > > > >> > want to
> > >>> > > > > > > > > > > >> > > > > add
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > ids
> > >>> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
> > >>> that don't
> > >>> > > > > > have it
> > >>> > > > > > > > > yet and
> > >>> > > > > > > > > > > >> > that we
> > >>> > > > > > > > > > > >> > > > > can
> > >>> > > > > > > > > > > >> > > > > >> > > probably
> > >>> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> > >>> versus adding
> > >>> > > > new
> > >>> > > > > > APIs
> > >>> > > > > > > > > to the
> > >>> > > > > > > > > > > >> > Java
> > >>> > > > > > > > > > > >> > > > > >> Consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > (unless
> > >>> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new
> > APIs
> > >>> is
> > >>> > > > required
> > >>> > > > > > for
> > >>> > > > > > > > > > > >> reasonable
> > >>> > > > > > > > > > > >> > > > > >> semantics).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> > about
> > >>> the
> > >>> > > > > > coordinator
> > >>> > > > > > > > > storing
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> configs.
> > >>> > > > > > > > > > > >> > > > > >> > > It's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to
> be
> > >>> > > > centralized
> > >>> > > > > > in the
> > >>> > > > > > > > > > > >> metadata
> > >>> > > > > > > > > > > >> > log for
> > >>> > > > > > > > > > > >> > > > > >> > > various
> > >>> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > >>> visibility,
> > >>> > > > > > consistency,
> > >>> > > > > > > > > > > >> etc.).
> > >>> > > > > > > > > > > >> > > > > >> Similarly, I
> > >>> > > > > > > > > > > >> > > > > >> > > am
> > >>> > > > > > > > > > > >> > > > > >> > > > > not
> > >>> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> > >>> deleting
> > >>> > > > > > configs in a
> > >>> > > > > > > > > way
> > >>> > > > > > > > > > > >> that
> > >>> > > > > > > > > > > >> > they
> > >>> > > > > > > > > > > >> > > > > >> cannot
> > >>> > > > > > > > > > > >> > > > > >> > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good
> property
> > >>> for modern
> > >>> > > > > > systems
> > >>> > > > > > > > > is to
> > >>> > > > > > > > > > > >> > minimize
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > number
> > >>> > > > > > > > > > > >> > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> > >>> scenarios.
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at
> > 3:47
> > >>> PM David
> > >>> > > > > > Jacot
> > >>> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> > >>> answers are
> > >>> > > > below:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> path,
> > >>> especially
> > >>> > > > > > the last
> > >>> > > > > > > > > step
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > > > > clients
> > >>> > > > > > > > > > > >> > > > > >> > > flip the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > >>> protocol, in
> > >>> > > > which
> > >>> > > > > > we
> > >>> > > > > > > > > would
> > >>> > > > > > > > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > window
> > >>> > > > > > > > > > > >> > > > > >> > where
> > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > old
> > >>> > > > protocols /
> > >>> > > > > > rpcs
> > >>> > > > > > > > > are
> > >>> > > > > > > > > > > >> used
> > >>> > > > > > > > > > > >> > by
> > >>> > > > > > > > > > > >> > > > > >> members
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > >>> coordinator could
> > >>> > > > > > "mimic"
> > >>> > > > > > > > > the old
> > >>> > > > > > > > > > > >> > behavior
> > >>> > > > > > > > > > > >> > > > > >> while
> > >>> > > > > > > > > > > >> > > > > >> > > > > using the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > >>> something we
> > >>> > > > need to
> > >>> > > > > > > > > present
> > >>> > > > > > > > > > > >> about.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just
> published a
> > >>> new
> > >>> > > > version
> > >>> > > > > > of KIP
> > >>> > > > > > > > > which
> > >>> > > > > > > > > > > >> > includes
> > >>> > > > > > > > > > > >> > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See
> > the
> > >>> > > > "Supporting
> > >>> > > > > > > > > Online
> > >>> > > > > > > > > > > >> > Consumer
> > >>> > > > > > > > > > > >> > > > > Group
> > >>> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > >>> > > > Deprecation,
> > >>> > > > > > and
> > >>> > > > > > > > > Migration
> > >>> > > > > > > > > > > >> > Plan". I
> > >>> > > > > > > > > > > >> > > > > >> think
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > I have to think
> through a
> > >>> few cases
> > >>> > > > > > now but
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > overall idea
> > >>> > > > > > > > > > > >> > > > > >> and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > >>> understandable.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > >>> ids. So
> > >>> > > > far as
> > >>> > > > > > > > > KIP-516 the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > >> are
> > >>> > > > > > > > > > > >> > > > > >> > > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > used
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> admin
> > >>> client,
> > >>> > > > but
> > >>> > > > > > they
> > >>> > > > > > > > > are not
> > >>> > > > > > > > > > > >> > exposed
> > >>> > > > > > > > > > > >> > > > > via
> > >>> > > > > > > > > > > >> > > > > >> any
> > >>> > > > > > > > > > > >> > > > > >> > > > > public
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > >>> think the
> > >>> > > > > > question is,
> > >>> > > > > > > > > first
> > >>> > > > > > > > > > > >> > should we
> > >>> > > > > > > > > > > >> > > > > >> let
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > maintaining
> > >>> the
> > >>> > > > names
> > >>> > > > > > -> ids
> > >>> > > > > > > > > mapping
> > >>> > > > > > > > > > > >> > itself
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > fully
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > >>> existing
> > >>> > > > RPCs
> > >>> > > > > > and the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> RPCs
> > >>> > > > > > > > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > ever
> > >>> consider
> > >>> > > > > > exposing
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > (both
> > >>> > > > > > > > > subscribe/assign, as
> > >>> > > > > > > > > > > >> well
> > >>> > > > > > > > > > > >> > as in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> like
> > >>> topic
> > >>> > > > > > > > > > > >> > deletion-and-recreation).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we
> would
> > >>> include
> > >>> > > > > > converting
> > >>> > > > > > > > > all
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > offsets
> > >>> > > > > > > > > > > >> > > > > >> > > related
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids
> > in
> > >>> this
> > >>> > > > KIP,
> > >>> > > > > > the
> > >>> > > > > > > > > consumer
> > >>> > > > > > > > > > > >> > would be
> > >>> > > > > > > > > > > >> > > > > >> able
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with
> topic
> > >>> ids. That
> > >>> > > > > > being
> > >>> > > > > > > > > said, it
> > >>> > > > > > > > > > > >> > still has
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > provide
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> > >>> various APIs so
> > >>> > > > > > having a
> > >>> > > > > > > > > > > >> mapping
> > >>> > > > > > > > > > > >> > in the
> > >>> > > > > > > > > > > >> > > > > >> > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a
> strong
> > >>> opinion on
> > >>> > > > > > this.
> > >>> > > > > > > > > Here I
> > >>> > > > > > > > > > > >> > wonder if
> > >>> > > > > > > > > > > >> > > > > >> this
> > >>> > > > > > > > > > > >> > > > > >> > > goes
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of
> this
> > >>> KIP. I
> > >>> > > > would
> > >>> > > > > > rather
> > >>> > > > > > > > > focus
> > >>> > > > > > > > > > > >> on
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > internals
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > here and we can
> consider
> > >>> this
> > >>> > > > > > separately if
> > >>> > > > > > > > > we see
> > >>> > > > > > > > > > > >> > value in
> > >>> > > > > > > > > > > >> > > > > >> doing
> > >>> > > > > > > > > > > >> > > > > >> > > it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> > >>> point about
> > >>> > > > > > using
> > >>> > > > > > > > > topic ids
> > >>> > > > > > > > > > > >> > in the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> ConsumerGroupHeartbeatRequest, I
> > >>> > > > think
> > >>> > > > > > that
> > >>> > > > > > > > > there
> > >>> > > > > > > > > > > >> is
> > >>> > > > > > > > > > > >> > one
> > >>> > > > > > > > > > > >> > > > > >> > advantage
> > >>> > > > > > > > > > > >> > > > > >> > > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> > consumer
> > >>> will
> > >>> > > > have
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> opportunity
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> validate
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists
> before
> > >>> passing
> > >>> > > > them
> > >>> > > > > > into
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> group
> > >>> > > > > > > > > > > >> > > > > rebalance
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously,
> the
> > >>> > > > coordinator
> > >>> > > > > > will
> > >>> > > > > > > > > also
> > >>> > > > > > > > > > > >> notice
> > >>> > > > > > > > > > > >> > it but
> > >>> > > > > > > > > > > >> > > > > >> it
> > >>> > > > > > > > > > > >> > > > > >> > > does
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way
> to
> > >>> reject an
> > >>> > > > > > invalid
> > >>> > > > > > > > > topic in
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> response.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> David
> > >>> on all
> > >>> > > > other
> > >>> > > > > > minor
> > >>> > > > > > > > > > > >> questions
> > >>> > > > > > > > > > > >> > > > > except
> > >>> > > > > > > > > > > >> > > > > >> for
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > >>> question:
> > >>> > > > > > personally
> > >>> > > > > > > > > I think
> > >>> > > > > > > > > > > >> > it's not
> > >>> > > > > > > > > > > >> > > > > >> > > necessary
> > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> subscribe
> > >>> API with
> > >>> > > > > > Pattern,
> > >>> > > > > > > > > but
> > >>> > > > > > > > > > > >> > instead we
> > >>> > > > > > > > > > > >> > > > > >> still
> > >>> > > > > > > > > > > >> > > > > >> > > use
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> documenting
> > >>> that our
> > >>> > > > > > > > > subscription may
> > >>> > > > > > > > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> rejected by
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> incompatible
> > >>> case is a
> > >>> > > > > > very rare
> > >>> > > > > > > > > > > >> scenario
> > >>> > > > > > > > > > > >> > I felt
> > >>> > > > > > > > > > > >> > > > > >> > using
> > >>> > > > > > > > > > > >> > > > > >> > > an
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > based
> > >>> > > > > > subscription may
> > >>> > > > > > > > > be
> > >>> > > > > > > > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> vulnerable
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > various
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have
> > to
> > >>> look at
> > >>> > > > the
> > >>> > > > > > > > > differences
> > >>> > > > > > > > > > > >> > between
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > two
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > >>> understand the
> > >>> > > > > > potential
> > >>> > > > > > > > > issues.
> > >>> > > > > > > > > > > >> My
> > >>> > > > > > > > > > > >> > > > > >> > > understanding is
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all
> > >>> the basic
> > >>> > > > > > regular
> > >>> > > > > > > > > > > >> > expressions. The
> > >>> > > > > > > > > > > >> > > > > >> > > differences
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> > mainly
> > >>> about
> > >>> > > > the
> > >>> > > > > > various
> > >>> > > > > > > > > > > >> character
> > >>> > > > > > > > > > > >> > > > > >> classes. I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other
> people
> > >>> think
> > >>> > > > about
> > >>> > > > > > this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> > >>> 11:28 PM
> > >>> > > > > > Guozhang
> > >>> > > > > > > > > Wang <
> > >>> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think
> > on
> > >>> the high
> > >>> > > > > > level
> > >>> > > > > > > > > there are
> > >>> > > > > > > > > > > >> > two meta
> > >>> > > > > > > > > > > >> > > > > >> > > points we
> > >>> > > > > > > > > > > >> > > > > >> > > > > need
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit
> > more:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration
> path,
> > >>> especially
> > >>> > > > > > the last
> > >>> > > > > > > > > step
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > > > > clients
> > >>> > > > > > > > > > > >> > > > > >> > > flip the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > >>> protocol, in
> > >>> > > > which
> > >>> > > > > > we
> > >>> > > > > > > > > would
> > >>> > > > > > > > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > window
> > >>> > > > > > > > > > > >> > > > > >> > where
> > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > old
> > >>> > > > protocols /
> > >>> > > > > > rpcs
> > >>> > > > > > > > > are
> > >>> > > > > > > > > > > >> used
> > >>> > > > > > > > > > > >> > by
> > >>> > > > > > > > > > > >> > > > > >> members
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > >>> coordinator could
> > >>> > > > > > "mimic"
> > >>> > > > > > > > > the old
> > >>> > > > > > > > > > > >> > behavior
> > >>> > > > > > > > > > > >> > > > > >> while
> > >>> > > > > > > > > > > >> > > > > >> > > > > using the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > >>> something we
> > >>> > > > need to
> > >>> > > > > > > > > present
> > >>> > > > > > > > > > > >> about.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > >>> ids. So
> > >>> > > > far as
> > >>> > > > > > > > > KIP-516 the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > >> are
> > >>> > > > > > > > > > > >> > > > > >> > > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > used
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and
> admin
> > >>> client,
> > >>> > > > but
> > >>> > > > > > they
> > >>> > > > > > > > > are not
> > >>> > > > > > > > > > > >> > exposed
> > >>> > > > > > > > > > > >> > > > > via
> > >>> > > > > > > > > > > >> > > > > >> any
> > >>> > > > > > > > > > > >> > > > > >> > > > > public
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > >>> think the
> > >>> > > > > > question is,
> > >>> > > > > > > > > first
> > >>> > > > > > > > > > > >> > should we
> > >>> > > > > > > > > > > >> > > > > >> let
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > maintaining
> > >>> the
> > >>> > > > names
> > >>> > > > > > -> ids
> > >>> > > > > > > > > mapping
> > >>> > > > > > > > > > > >> > itself
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > fully
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > >>> existing
> > >>> > > > RPCs
> > >>> > > > > > and the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> RPCs
> > >>> > > > > > > > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > ever
> > >>> consider
> > >>> > > > > > exposing
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > (both
> > >>> > > > > > > > > subscribe/assign, as
> > >>> > > > > > > > > > > >> well
> > >>> > > > > > > > > > > >> > as in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases
> like
> > >>> topic
> > >>> > > > > > > > > > > >> > deletion-and-recreation).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with
> David
> > >>> on all
> > >>> > > > other
> > >>> > > > > > minor
> > >>> > > > > > > > > > > >> questions
> > >>> > > > > > > > > > > >> > > > > except
> > >>> > > > > > > > > > > >> > > > > >> for
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > >>> question:
> > >>> > > > > > personally
> > >>> > > > > > > > > I think
> > >>> > > > > > > > > > > >> > it's not
> > >>> > > > > > > > > > > >> > > > > >> > > necessary
> > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the
> subscribe
> > >>> API with
> > >>> > > > > > Pattern,
> > >>> > > > > > > > > but
> > >>> > > > > > > > > > > >> > instead we
> > >>> > > > > > > > > > > >> > > > > >> still
> > >>> > > > > > > > > > > >> > > > > >> > > use
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just
> documenting
> > >>> that our
> > >>> > > > > > > > > subscription may
> > >>> > > > > > > > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> rejected by
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the
> incompatible
> > >>> case is a
> > >>> > > > > > very rare
> > >>> > > > > > > > > > > >> scenario
> > >>> > > > > > > > > > > >> > I felt
> > >>> > > > > > > > > > > >> > > > > >> > using
> > >>> > > > > > > > > > > >> > > > > >> > > an
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > based
> > >>> > > > > > subscription may
> > >>> > > > > > > > > be
> > >>> > > > > > > > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> vulnerable
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > various
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022
> at
> > >>> 5:23 AM
> > >>> > > > > > David Jacot
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > >>> feedback. Let
> > >>> > > > me
> > >>> > > > > > answer
> > >>> > > > > > > > > your
> > >>> > > > > > > > > > > >> > questions
> > >>> > > > > > > > > > > >> > > > > >> > inline.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> > >>> premature to
> > >>> > > > > > talk about
> > >>> > > > > > > > > > > >> target
> > >>> > > > > > > > > > > >> > > > > versions
> > >>> > > > > > > > > > > >> > > > > >> for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > >>> existing group
> > >>> > > > > > protocol.
> > >>> > > > > > > > > > > >> Unlike
> > >>> > > > > > > > > > > >> > KRaft,
> > >>> > > > > > > > > > > >> > > > > >> this
> > >>> > > > > > > > > > > >> > > > > >> > > > > affects a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > core
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol
> and
> > >>> hence
> > >>> > > > > > > > > deprecation/removal
> > >>> > > > > > > > > > > >> > will be
> > >>> > > > > > > > > > > >> > > > > >> > heavily
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > >>> applications
> > >>> > > > migrate
> > >>> > > > > > to
> > >>> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I
> > >>> will
> > >>> > > > remove it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says
> we
> > >>> intend to
> > >>> > > > > > release
> > >>> > > > > > > > > this in
> > >>> > > > > > > > > > > >> > 4.x, but
> > >>> > > > > > > > > > > >> > > > > it
> > >>> > > > > > > > > > > >> > > > > >> > > wasn't
> > >>> > > > > > > > > > > >> > > > > >> > > > > made
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> > >>> that as a
> > >>> > > > way to
> > >>> > > > > > > > > estimate
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > we'd
> > >>> > > > > > > > > > > >> > > > > >> > > deprecate
> > >>> > > > > > > > > > > >> > > > > >> > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group
> protocol,
> > >>> I also
> > >>> > > > > > suggest
> > >>> > > > > > > > > removing
> > >>> > > > > > > > > > > >> > this part.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> > >>> reasoning. As
> > >>> > > > > > > > > explained, I
> > >>> > > > > > > > > > > >> plan
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> rewrite
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> > >>> while we
> > >>> > > > > > implement
> > >>> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > protocol.
> > >>> > > > > > > > > > > >> > > > > >> This
> > >>> > > > > > > > > > > >> > > > > >> > > means
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> > >>> will be
> > >>> > > > slightly
> > >>> > > > > > > > > different
> > >>> > > > > > > > > > > >> > (e.g.
> > >>> > > > > > > > > > > >> > > > > >> threading
> > >>> > > > > > > > > > > >> > > > > >> > > > > model).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted
> > to
> > >>> tighten
> > >>> > > > the
> > >>> > > > > > > > > switch from
> > >>> > > > > > > > > > > >> > the old
> > >>> > > > > > > > > > > >> > > > > >> group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the
> > new
> > >>> group
> > >>> > > > > > coordinator
> > >>> > > > > > > > > to a
> > >>> > > > > > > > > > > >> > major
> > >>> > > > > > > > > > > >> > > > > >> release.
> > >>> > > > > > > > > > > >> > > > > >> > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would
> be
> > >>> to use a
> > >>> > > > flag
> > >>> > > > > > to do
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > switch
> > >>> > > > > > > > > > > >> > > > > >> instead
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > relying
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> > >>> upgrade.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to
> flesh
> > >>> out the
> > >>> > > > > > details of
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > migration
> > >>> > > > > > > > > > > >> > > > > >> story.
> > >>> > > > > > > > > > > >> > > > > >> > > It
> > >>> > > > > > > > > > > >> > > > > >> > > > > sounds
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we
> > will
> > >>> support
> > >>> > > > > > online
> > >>> > > > > > > > > > > >> migrations.
> > >>> > > > > > > > > > > >> > Is
> > >>> > > > > > > > > > > >> > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > correct?
> > >>> > > > > > > > > > > >> > > > > >> > > > > We
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> > >>> detail. It
> > >>> > > > could
> > >>> > > > > > also
> > >>> > > > > > > > > be done
> > >>> > > > > > > > > > > >> > as a
> > >>> > > > > > > > > > > >> > > > > >> separate
> > >>> > > > > > > > > > > >> > > > > >> > > KIP,
> > >>> > > > > > > > > > > >> > > > > >> > > > > if
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will
> support
> > >>> online
> > >>> > > > > > migrations
> > >>> > > > > > > > > for the
> > >>> > > > > > > > > > > >> > group.
> > >>> > > > > > > > > > > >> > > > > That
> > >>> > > > > > > > > > > >> > > > > >> > > means
> > >>> > > > > > > > > > > >> > > > > >> > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the
> old
> > >>> protocol
> > >>> > > > > > will be
> > >>> > > > > > > > > able to
> > >>> > > > > > > > > > > >> > switch to
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> > explain
> > >>> how that
> > >>> > > > > > will work
> > >>> > > > > > > > > > > >> though.
> > >>> > > > > > > > > > > >> > It is
> > >>> > > > > > > > > > > >> > > > > >> > > basically a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must
> > be
> > >>> > > > upgraded or
> > >>> > > > > > > > > rolled to a
> > >>> > > > > > > > > > > >> > software
> > >>> > > > > > > > > > > >> > > > > >> > > supporting
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > coordinator.
> > >>> Both the
> > >>> > > > > > old and
> > >>> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > > > > >> coordinator
> > >>> > > > > > > > > > > >> > > > > >> > > will
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > >>> protocol and
> > >>> > > > rely
> > >>> > > > > > on the
> > >>> > > > > > > > > same
> > >>> > > > > > > > > > > >> > persisted
> > >>> > > > > > > > > > > >> > > > > >> > > metadata so
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> > >>> together. This
> > >>> > > > point
> > >>> > > > > > is an
> > >>> > > > > > > > > > > >> offline
> > >>> > > > > > > > > > > >> > > > > >> migration.
> > >>> > > > > > > > > > > >> > > > > >> > We
> > >>> > > > > > > > > > > >> > > > > >> > > > > cannot
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > >>> because it
> > >>> > > > would
> > >>> > > > > > require
> > >>> > > > > > > > > > > >> > shutting down
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > current
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > >>> starting up the
> > >>> > > > > > new one
> > >>> > > > > > > > > and
> > >>> > > > > > > > > > > >> that
> > >>> > > > > > > > > > > >> > would
> > >>> > > > > > > > > > > >> > > > > >> cause
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > >>> metadata
> > >>> > > > > > version/IBP
> > >>> > > > > > > > > must be
> > >>> > > > > > > > > > > >> > upgraded
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> X
> > >>> > > > > > > > > > > >> > > > > >> > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > order
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> > >>> protocol.
> > >>> > > > This
> > >>> > > > > > cannot
> > >>> > > > > > > > > be done
> > >>> > > > > > > > > > > >> > before
> > >>> > > > > > > > > > > >> > > > > 1)
> > >>> > > > > > > > > > > >> > > > > >> is
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because
> > the
> > >>> old
> > >>> > > > > > coordinator
> > >>> > > > > > > > > doesn't
> > >>> > > > > > > > > > > >> > support
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers
> must
> > >>> be
> > >>> > > > upgraded
> > >>> > > > > > to a
> > >>> > > > > > > > > version
> > >>> > > > > > > > > > > >> > > > > supporting
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > online
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must
> have
> > >>> KIP-792).
> > >>> > > > If
> > >>> > > > > > the
> > >>> > > > > > > > > consumer
> > >>> > > > > > > > > > > >> is
> > >>> > > > > > > > > > > >> > > > > already
> > >>> > > > > > > > > > > >> > > > > >> > > there.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be
> done
> > >>> at this
> > >>> > > > point.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers
> must
> > >>> be rolled
> > >>> > > > > > with the
> > >>> > > > > > > > > > > >> feature
> > >>> > > > > > > > > > > >> > flag
> > >>> > > > > > > > > > > >> > > > > >> turned
> > >>> > > > > > > > > > > >> > > > > >> > > on.
> > >>> > > > > > > > > > > >> > > > > >> > > > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> > >>> automatically
> > >>> > > > > > > > > converted when
> > >>> > > > > > > > > > > >> > the first
> > >>> > > > > > > > > > > >> > > > > >> > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> > protocol
> > >>> joins
> > >>> > > > the
> > >>> > > > > > group.
> > >>> > > > > > > > > While
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > members
> > >>> > > > > > > > > > > >> > > > > >> > > using the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> > being
> > >>> > > > upgraded,
> > >>> > > > > > the old
> > >>> > > > > > > > > > > >> protocol
> > >>> > > > > > > > > > > >> > is
> > >>> > > > > > > > > > > >> > > > > >> proxied
> > >>> > > > > > > > > > > >> > > > > >> > > into
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all
> of
> > >>> this in
> > >>> > > > the
> > >>> > > > > > KIP.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy
> that
> > >>> we are
> > >>> > > > > > pushing the
> > >>> > > > > > > > > pattern
> > >>> > > > > > > > > > > >> > > > > >> subscriptions
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
> > >>> there
> > >>> > > > could be
> > >>> > > > > > some
> > >>> > > > > > > > > tricky
> > >>> > > > > > > > > > > >> > > > > >> compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > > > issues.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism
> > for
> > >>> users to
> > >>> > > > > > detect
> > >>> > > > > > > > > that
> > >>> > > > > > > > > > > >> they
> > >>> > > > > > > > > > > >> > need to
> > >>> > > > > > > > > > > >> > > > > >> > update
> > >>> > > > > > > > > > > >> > > > > >> > > > > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching
> to
> > >>> the new
> > >>> > > > > > protocol?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a
> > >>> bit more
> > >>> > > > > > optimistic
> > >>> > > > > > > > > than
> > >>> > > > > > > > > > > >> you
> > >>> > > > > > > > > > > >> > on this
> > >>> > > > > > > > > > > >> > > > > >> > > point. I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > >>> majority of
> > >>> > > > the
> > >>> > > > > > cases
> > >>> > > > > > > > > are
> > >>> > > > > > > > > > > >> simple
> > >>> > > > > > > > > > > >> > > > > regexes
> > >>> > > > > > > > > > > >> > > > > >> > which
> > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> > >>> engine. The
> > >>> > > > > > > > > coordinator will
> > >>> > > > > > > > > > > >> > verify
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > regex
> > >>> > > > > > > > > > > >> > > > > >> > > > > anyway
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> > >>> consumer if the
> > >>> > > > > > regex is
> > >>> > > > > > > > > not
> > >>> > > > > > > > > > > >> valid.
> > >>> > > > > > > > > > > >> > > > > Coming
> > >>> > > > > > > > > > > >> > > > > >> > back
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in
> > the
> > >>> worst
> > >>> > > > case,
> > >>> > > > > > the
> > >>> > > > > > > > > first
> > >>> > > > > > > > > > > >> > upgraded
> > >>> > > > > > > > > > > >> > > > > >> > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > joining
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> > >>> rejected.
> > >>> > > > This
> > >>> > > > > > should
> > >>> > > > > > > > > be used
> > >>> > > > > > > > > > > >> > as the
> > >>> > > > > > > > > > > >> > > > > >> last
> > >>> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for
> customers
> > >>> to
> > >>> > > > validate
> > >>> > > > > > their
> > >>> > > > > > > > > regex
> > >>> > > > > > > > > > > >> > before
> > >>> > > > > > > > > > > >> > > > > >> > upgrading
> > >>> > > > > > > > > > > >> > > > > >> > > > > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to
> test
> > >>> them with
> > >>> > > > > > another
> > >>> > > > > > > > > group.
> > >>> > > > > > > > > > > >> For
> > >>> > > > > > > > > > > >> > > > > >> instance,
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> > >>> pre-prod
> > >>> > > > > > environment.
> > >>> > > > > > > > > > > >> Another
> > >>> > > > > > > > > > > >> > way
> > >>> > > > > > > > > > > >> > > > > >> would be
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > >>> consumer-group tool
> > >>> > > > to
> > >>> > > > > > > > > provide a
> > >>> > > > > > > > > > > >> regex
> > >>> > > > > > > > > > > >> > > > > >> validation
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would
> this
> > >>> be
> > >>> > > > enough in
> > >>> > > > > > your
> > >>> > > > > > > > > > > >> opinion?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the
> > >>> last
> > >>> > > > question,
> > >>> > > > > > will
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> Java
> > >>> > > > > > > > > > > >> > client
> > >>> > > > > > > > > > > >> > > > > >> allow
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > >>> current regex
> > >>> > > > > > engine for
> > >>> > > > > > > > > > > >> > compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > reasons?
> > >>> > > > > > > > > > > >> > > > > >> > > > > For
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy
> to
> > >>> keep using
> > >>> > > > > > client
> > >>> > > > > > > > > based
> > >>> > > > > > > > > > > >> > regex at
> > >>> > > > > > > > > > > >> > > > > >> first
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > keep
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple
> > >>> and then
> > >>> > > > > > migrate to
> > >>> > > > > > > > > server
> > >>> > > > > > > > > > > >> > based
> > >>> > > > > > > > > > > >> > > > > >> regexes
> > >>> > > > > > > > > > > >> > > > > >> > > as a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > second
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> > >>> point but I
> > >>> > > > am
> > >>> > > > > > > > > concerned that
> > >>> > > > > > > > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > allow
> > >>> > > > > > > > > > > >> > > > > >> > > > > users
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
> > >>> this mode.
> > >>> > > > That
> > >>> > > > > > > > > would go
> > >>> > > > > > > > > > > >> > against our
> > >>> > > > > > > > > > > >> > > > > >> goal
> > >>> > > > > > > > > > > >> > > > > >> > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> > client
> > >>> because
> > >>> > > > we
> > >>> > > > > > would
> > >>> > > > > > > > > have to
> > >>> > > > > > > > > > > >> > continue
> > >>> > > > > > > > > > > >> > > > > >> > > monitoring
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> > >>> client
> > >>> > > > side. I
> > >>> > > > > > would
> > >>> > > > > > > > > rather
> > >>> > > > > > > > > > > >> > not do
> > >>> > > > > > > > > > > >> > > > > >> this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say
> that
> > >>> the group
> > >>> > > > > > > > > coordinator
> > >>> > > > > > > > > > > >> will
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > > >> > > responsible for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the
> configurations
> > >>> and that
> > >>> > > > the
> > >>> > > > > > > > > > > >> configurations
> > >>> > > > > > > > > > > >> > will be
> > >>> > > > > > > > > > > >> > > > > >> > > deleted
> > >>> > > > > > > > > > > >> > > > > >> > > > > when
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
> > >>> Will a
> > >>> > > > > > transition to
> > >>> > > > > > > > > DEAD
> > >>> > > > > > > > > > > >> > trigger
> > >>> > > > > > > > > > > >> > > > > >> > deletion
> > >>> > > > > > > > > > > >> > > > > >> > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > >>> > > > configurations
> > >>> > > > > > will be
> > >>> > > > > > > > > > > >> deleted
> > >>> > > > > > > > > > > >> > when
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > group is
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > >>> together.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the
> choice
> > >>> to store
> > >>> > > > the
> > >>> > > > > > > > > configs in
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > group
> > >>> > > > > > > > > > > >> > > > > >> > > coordinator
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list
> all
> > >>> cluster
> > >>> > > > > > configs and
> > >>> > > > > > > > > their
> > >>> > > > > > > > > > > >> > values?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so.
> The
> > >>> group
> > >>> > > > > > > > > configurations are
> > >>> > > > > > > > > > > >> > overrides
> > >>> > > > > > > > > > > >> > > > > >> of
> > >>> > > > > > > > > > > >> > > > > >> > > cluster
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you
> want
> > >>> to know
> > >>> > > > all
> > >>> > > > > > the
> > >>> > > > > > > > > overrides
> > >>> > > > > > > > > > > >> > though,
> > >>> > > > > > > > > > > >> > > > > you
> > >>> > > > > > > > > > > >> > > > > >> > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > have
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the
> group
> > >>> > > > coordinators.
> > >>> > > > > > You
> > >>> > > > > > > > > cannot
> > >>> > > > > > > > > > > >> > rely on
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > metadata
> > >>> > > > > > > > > > > >> > > > > >> > > > > log
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> > someone
> > >>> > > > configure a
> > >>> > > > > > group
> > >>> > > > > > > > > before
> > >>> > > > > > > > > > > >> > starting
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > >>> allowing the
> > >>> > > > > > explicit
> > >>> > > > > > > > > creation
> > >>> > > > > > > > > > > >> of
> > >>> > > > > > > > > > > >> > > > > groups?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> > >>> decoupled
> > >>> > > > from
> > >>> > > > > > the
> > >>> > > > > > > > > group
> > >>> > > > > > > > > > > >> > lifecycle.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will
> > be
> > >>> > > > > > automatically
> > >>> > > > > > > > > created in
> > >>> > > > > > > > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > case.
> > >>> > > > > > > > > > > >> > > > > >> > > However,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
> > >>> lost after
> > >>> > > > the
> > >>> > > > > > > > > retention
> > >>> > > > > > > > > > > >> > period of
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > >>> > > > Consumer.subscribe
> > >>> > > > > > method
> > >>> > > > > > > > > for the
> > >>> > > > > > > > > > > >> > Java
> > >>> > > > > > > > > > > >> > > > > >> client
> > >>> > > > > > > > > > > >> > > > > >> > > still
> > >>> > > > > > > > > > > >> > > > > >> > > > > take
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> `java.util.regex.Pattern` of
> > >>> > > > do
> > >>> > > > > > we
> > >>> > > > > > > > > have to
> > >>> > > > > > > > > > > >> > introduce
> > >>> > > > > > > > > > > >> > > > > an
> > >>> > > > > > > > > > > >> > > > > >> > > overload?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> > >>> question. I
> > >>> > > > > > forgot
> > >>> > > > > > > > > about
> > >>> > > > > > > > > > > >> that
> > >>> > > > > > > > > > > >> > one.
> > >>> > > > > > > > > > > >> > > > > As
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> `java.util.regex.Pattern` is
> > >>> > > > not
> > >>> > > > > > fully
> > >>> > > > > > > > > > > >> compatible
> > >>> > > > > > > > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > engine
> > >>> > > > > > > > > > > >> > > > > >> > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> > >>> might be
> > >>> > > > better
> > >>> > > > > > to
> > >>> > > > > > > > > deprecate
> > >>> > > > > > > > > > > >> > it and
> > >>> > > > > > > > > > > >> > > > > >> use an
> > >>> > > > > > > > > > > >> > > > > >> > > > > overload
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a
> string.
> > >>> We would
> > >>> > > > > > rely on
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> server
> > >>> > > > > > > > > > > >> > side
> > >>> > > > > > > > > > > >> > > > > >> > > validation.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > During the
> migration,
> > >>> I think
> > >>> > > > that
> > >>> > > > > > we
> > >>> > > > > > > > > could
> > >>> > > > > > > > > > > >> still
> > >>> > > > > > > > > > > >> > try to
> > >>> > > > > > > > > > > >> > > > > >> > > toString
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it.
> > That
> > >>> should
> > >>> > > > > > work, I
> > >>> > > > > > > > > think, in
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> majority
> > >>> > > > > > > > > > > >> > > > > >> > > of the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> > >>> Justine
> > >>> > > > that we
> > >>> > > > > > > > > should be
> > >>> > > > > > > > > > > >> > clearer
> > >>> > > > > > > > > > > >> > > > > about
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > reason
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > >>> > > > IBP/metadata.version
> > >>> > > > > > from the
> > >>> > > > > > > > > > > >> feature
> > >>> > > > > > > > > > > >> > flag.
> > >>> > > > > > > > > > > >> > > > > >> Maybe
> > >>> > > > > > > > > > > >> > > > > >> > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > mean
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > >>> default for
> > >>> > > > the
> > >>> > > > > > feature
> > >>> > > > > > > > > flag
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > true
> > >>> > > > > > > > > > > >> > > > > >> based
> > >>> > > > > > > > > > > >> > > > > >> > on
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> > >>> once we
> > >>> > > > want to
> > >>> > > > > > make
> > >>> > > > > > > > > it the
> > >>> > > > > > > > > > > >> > default.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> > >>> that feature
> > >>> > > > > > flag
> > >>> > > > > > > > > mainly
> > >>> > > > > > > > > > > >> > during the
> > >>> > > > > > > > > > > >> > > > > >> > > development
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not
> > >>> have
> > >>> > > > mentioned
> > >>> > > > > > it, I
> > >>> > > > > > > > > think,
> > >>> > > > > > > > > > > >> > because
> > >>> > > > > > > > > > > >> > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > could
> > >>> > > > > > > > > > > >> > > > > >> > > > > use
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config
> > for
> > >>> it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> > >>> protocol APIs
> > >>> > > > > > don't
> > >>> > > > > > > > > mention
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > required
> > >>> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that
> > for
> > >>> > > > consistency.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit
> odd
> > >>> that
> > >>> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > >>> > > > > > > > > > > >> > > > > requires
> > >>> > > > > > > > > > > >> > > > > >> > > "Read
> > >>> > > > > > > > > > > >> > > > > >> > > > > Group"
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems
> to
> > >>> do more
> > >>> > > > than
> > >>> > > > > > > > > reading.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is
> how
> > >>> the
> > >>> > > > current
> > >>> > > > > > > > > protocol works
> > >>> > > > > > > > > > > >> > though.
> > >>> > > > > > > > > > > >> > > > > We
> > >>> > > > > > > > > > > >> > > > > >> > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read
> Group"
> > >>> to join a
> > >>> > > > > > group. We
> > >>> > > > > > > > > could
> > >>> > > > > > > > > > > >> > consider
> > >>> > > > > > > > > > > >> > > > > >> > > changing
> > >>> > > > > > > > > > > >> > > > > >> > > > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure
> > that
> > >>> it is
> > >>> > > > worth
> > >>> > > > > > it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> > >>> recreation
> > >>> > > > > > handled by
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > consumer
> > >>> > > > > > > > > > > >> > > > > with
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It
> would
> > >>> be good to
> > >>> > > > > > have a
> > >>> > > > > > > > > section
> > >>> > > > > > > > > > > >> on
> > >>> > > > > > > > > > > >> > this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> > protocol
> > >>> > > > > > perspective, the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> topic
> > >>> > > > > > > > > > > >> > will
> > >>> > > > > > > > > > > >> > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> > >>> treat it
> > >>> > > > like a
> > >>> > > > > > > > > topic with
> > >>> > > > > > > > > > > >> a
> > >>> > > > > > > > > > > >> > > > > different
> > >>> > > > > > > > > > > >> > > > > >> > > name.
> > >>> > > > > > > > > > > >> > > > > >> > > > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that
> > the
> > >>> > > > fetch/commit
> > >>> > > > > > > > > offsets
> > >>> > > > > > > > > > > >> APIs
> > >>> > > > > > > > > > > >> > do not
> > >>> > > > > > > > > > > >> > > > > >> > support
> > >>> > > > > > > > > > > >> > > > > >> > > > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
> > >>> would
> > >>> > > > reuse the
> > >>> > > > > > > > > offsets
> > >>> > > > > > > > > > > >> based
> > >>> > > > > > > > > > > >> > on the
> > >>> > > > > > > > > > > >> > > > > >> > same.
> > >>> > > > > > > > > > > >> > > > > >> > > I
> > >>> > > > > > > > > > > >> > > > > >> > > > > think
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should
> update
> > >>> those
> > >>> > > > APIs
> > >>> > > > > > as well
> > >>> > > > > > > > > in
> > >>> > > > > > > > > > > >> order
> > >>> > > > > > > > > > > >> > to be
> > >>> > > > > > > > > > > >> > > > > >> > > consistent
> > >>> > > > > > > > > > > >> > > > > >> > > > > end
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> > >>> strengthen
> > >>> > > > the
> > >>> > > > > > > > > semantics of
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> consumer.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> > mentions
> > >>> we will
> > >>> > > > > > write the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> > coordinator
> > >>> > > > > > > > > > > >> > > > > >> in
> > >>> > > > > > > > > > > >> > > > > >> > > Java.
> > >>> > > > > > > > > > > >> > > > > >> > > > > Even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > >>> implementation
> > >>> > > > > > detail, do
> > >>> > > > > > > > > we plan
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > have a
> > >>> > > > > > > > > > > >> > > > > >> new
> > >>> > > > > > > > > > > >> > > > > >> > > gradle
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > module
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> > >>> scalability
> > >>> > > > > > goal when
> > >>> > > > > > > > > it
> > >>> > > > > > > > > > > >> comes
> > >>> > > > > > > > > > > >> > to how
> > >>> > > > > > > > > > > >> > > > > >> many
> > >>> > > > > > > > > > > >> > > > > >> > > > > members
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol
> can
> > >>> support?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have
> numbers
> > >>> at the
> > >>> > > > > > moment. The
> > >>> > > > > > > > > > > >> protocol
> > >>> > > > > > > > > > > >> > should
> > >>> > > > > > > > > > > >> > > > > >> > > support
> > >>> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per
> group.
> > >>> We will
> > >>> > > > > > measure
> > >>> > > > > > > > > this when
> > >>> > > > > > > > > > > >> > we have
> > >>> > > > > > > > > > > >> > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > first
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation.
> Note
> > >>> that we
> > >>> > > > might
> > >>> > > > > > have
> > >>> > > > > > > > > other
> > >>> > > > > > > > > > > >> > > > > bottlenecks
> > >>> > > > > > > > > > > >> > > > > >> > down
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> > >>> commits).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we
> consider
> > >>> having
> > >>> > > > > > > > > SubscribedTopidIds
> > >>> > > > > > > > > > > >> > instead
> > >>> > > > > > > > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > SubscribedTopicNames
> > >>> in
> > >>> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > >>> > > > > > > > > > > >> > > > > >> Is
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > idea
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> > >>> resolve the
> > >>> > > > > > regex on
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > server, we
> > >>> > > > > > > > > > > >> > > > > >> can do
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name?
> The
> > >>> > > > difference
> > >>> > > > > > is that
> > >>> > > > > > > > > > > >> sending
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> regex is
> > >>> > > > > > > > > > > >> > > > > >> > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending
> the
> > >>> topic
> > >>> > > > names
> > >>> > > > > > is less
> > >>> > > > > > > > > > > >> > efficient.
> > >>> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> > easier
> > >>> to
> > >>> > > > handle
> > >>> > > > > > if we
> > >>> > > > > > > > > have
> > >>> > > > > > > > > > > >> topic
> > >>> > > > > > > > > > > >> > ids.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > >>> consolidate the
> > >>> > > > > > metadata
> > >>> > > > > > > > > lookup
> > >>> > > > > > > > > > > >> > on the
> > >>> > > > > > > > > > > >> > > > > >> server
> > >>> > > > > > > > > > > >> > > > > >> > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do
> agree
> > >>> with your
> > >>> > > > > > point. As
> > >>> > > > > > > > > a
> > >>> > > > > > > > > > > >> second
> > >>> > > > > > > > > > > >> > > > > though,
> > >>> > > > > > > > > > > >> > > > > >> > using
> > >>> > > > > > > > > > > >> > > > > >> > > > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better
> > here
> > >>> for the
> > >>> > > > > > delete and
> > >>> > > > > > > > > > > >> > recreation
> > >>> > > > > > > > > > > >> > > > > case.
> > >>> > > > > > > > > > > >> > > > > >> > > Also, I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
> > >>> allow
> > >>> > > > users to
> > >>> > > > > > > > > subscribe
> > >>> > > > > > > > > > > >> with
> > >>> > > > > > > > > > > >> > topic
> > >>> > > > > > > > > > > >> > > > > >> ids
> > >>> > > > > > > > > > > >> > > > > >> > in
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that
> > is
> > >>> the only
> > >>> > > > > > way to be
> > >>> > > > > > > > > > > >> really
> > >>> > > > > > > > > > > >> > robust
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12,
> 2022
> > >>> at 1:38 PM
> > >>> > > > > > David
> > >>> > > > > > > > > Jacot <
> > >>> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> > >>> comments.
> > >>> > > > Please
> > >>> > > > > > find
> > >>> > > > > > > > > my
> > >>> > > > > > > > > > > >> answers
> > >>> > > > > > > > > > > >> > > > > below.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > >>> protocol
> > >>> > > > relies on
> > >>> > > > > > > > > topic IDs
> > >>> > > > > > > > > > > >> > with the
> > >>> > > > > > > > > > > >> > > > > >> > > exception
> > >>> > > > > > > > > > > >> > > > > >> > > > > of the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based
> > in
> > >>> the
> > >>> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > >>> > > > > > > > > > > >> > > > > >> I
> > >>> > > > > > > > > > > >> > > > > >> > am
> > >>> > > > > > > > > > > >> > > > > >> > > not
> > >>> > > > > > > > > > > >> > > > > >> > > > > sure
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> > names
> > >>> is the
> > >>> > > > > > right call
> > >>> > > > > > > > > here.
> > >>> > > > > > > > > > > >> I
> > >>> > > > > > > > > > > >> > need to
> > >>> > > > > > > > > > > >> > > > > >> > think
> > >>> > > > > > > > > > > >> > > > > >> > > > > about it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > >>> Obviously, the
> > >>> > > > > > KIP does
> > >>> > > > > > > > > not
> > >>> > > > > > > > > > > >> > change the
> > >>> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to
> use
> > >>> topic
> > >>> > > > IDs.
> > >>> > > > > > This
> > >>> > > > > > > > > may be
> > >>> > > > > > > > > > > >> > something
> > >>> > > > > > > > > > > >> > > > > >> that
> > >>> > > > > > > > > > > >> > > > > >> > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as
> > it
> > >>> would
> > >>> > > > give
> > >>> > > > > > better
> > >>> > > > > > > > > > > >> overall
> > >>> > > > > > > > > > > >> > > > > >> guarantee in
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> > >>> think that
> > >>> > > > I
> > >>> > > > > > should
> > >>> > > > > > > > > not
> > >>> > > > > > > > > > > >> have
> > >>> > > > > > > > > > > >> > > > > mentioned
> > >>> > > > > > > > > > > >> > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > flag at
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will
> remove
> > >>> it. We can
> > >>> > > > > > use an
> > >>> > > > > > > > > internal
> > >>> > > > > > > > > > > >> > > > > >> configuration
> > >>> > > > > > > > > > > >> > > > > >> > > while
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > >>> feature.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> > types
> > >>> will be
> > >>> > > > > > > > > supported. The
> > >>> > > > > > > > > > > >> > change is
> > >>> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement
> is
> > >>> that the
> > >>> > > > > > cluster
> > >>> > > > > > > > > uses
> > >>> > > > > > > > > > > >> topic
> > >>> > > > > > > > > > > >> > IDs.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11,
> > 2022
> > >>> at 9:53
> > >>> > > > PM
> > >>> > > > > > > > > Guozhang
> > >>> > > > > > > > > > > >> Wang <
> > >>> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> > >>> feedback.
> > >>> > > > Here
> > >>> > > > > > are
> > >>> > > > > > > > > some
> > >>> > > > > > > > > > > >> replies
> > >>> > > > > > > > > > > >> > > > > inlined
> > >>> > > > > > > > > > > >> > > > > >> > > below:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> > >>> 2022 at
> > >>> > > > 2:53 AM
> > >>> > > > > > > > > Ismael Juma
> > >>> > > > > > > > > > > >> <
> > >>> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for
> the
> > >>> KIP. This
> > >>> > > > has
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> potential
> > >>> > > > > > > > > > > >> > to be a
> > >>> > > > > > > > > > > >> > > > > >> > great
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > >>> > > > questions/comments:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think
> it's
> > >>> > > > premature to
> > >>> > > > > > talk
> > >>> > > > > > > > > about
> > >>> > > > > > > > > > > >> > target
> > >>> > > > > > > > > > > >> > > > > >> versions
> > >>> > > > > > > > > > > >> > > > > >> > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of
> the
> > >>> existing
> > >>> > > > group
> > >>> > > > > > > > > protocol.
> > >>> > > > > > > > > > > >> > Unlike
> > >>> > > > > > > > > > > >> > > > > >> KRaft,
> > >>> > > > > > > > > > > >> > > > > >> > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client
> protocol
> > >>> and hence
> > >>> > > > > > > > > > > >> > deprecation/removal will
> > >>> > > > > > > > > > > >> > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> > > heavily
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> > >>> applications
> > >>> > > > > > migrate
> > >>> > > > > > > > > to the
> > >>> > > > > > > > > > > >> new
> > >>> > > > > > > > > > > >> > > > > >> protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree
> with
> > >>> you. I
> > >>> > > > think
> > >>> > > > > > we can
> > >>> > > > > > > > > > > >> remove
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> proposed
> > >>> > > > > > > > > > > >> > > > > >> > > > > timeline
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > >>> > > > Deprecation,
> > >>> > > > > > and
> > >>> > > > > > > > > Migration
> > >>> > > > > > > > > > > >> > Plan` and
> > >>> > > > > > > > > > > >> > > > > >> > > instead
> > >>> > > > > > > > > > > >> > > > > >> > > > > just
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > state
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> > >>> decide in the
> > >>> > > > > > future
> > >>> > > > > > > > > about
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > we
> > >>> > > > > > > > > > > >> > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > >>> behaviors.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP
> says
> > >>> we
> > >>> > > > intend to
> > >>> > > > > > > > > release this
> > >>> > > > > > > > > > > >> > in 4.x,
> > >>> > > > > > > > > > > >> > > > > >> but
> > >>> > > > > > > > > > > >> > > > > >> > it
> > >>> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > made
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> > added
> > >>> that as
> > >>> > > > a
> > >>> > > > > > way to
> > >>> > > > > > > > > > > >> estimate
> > >>> > > > > > > > > > > >> > when
> > >>> > > > > > > > > > > >> > > > > we'd
> > >>> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > >>> protocol, I
> > >>> > > > also
> > >>> > > > > > suggest
> > >>> > > > > > > > > > > >> removing
> > >>> > > > > > > > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > >> part.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's
> > not
> > >>> > > > specifically
> > >>> > > > > > > > > related to
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> > >>> more for
> > >>> > > > client
> > >>> > > > > > > > > upgrades.
> > >>> > > > > > > > > > > >> I.e.
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > broker-side
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation
> > may
> > >>> be done
> > >>> > > > > > first,
> > >>> > > > > > > > > and then
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > client
> > >>> > > > > > > > > > > >> > > > > >> > side,
> > >>> > > > > > > > > > > >> > > > > >> > > > > and we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> > >>> "released"
> > >>> > > > by
> > >>> > > > > > the
> > >>> > > > > > > > > time
> > >>> > > > > > > > > > > >> clients
> > >>> > > > > > > > > > > >> > > > > >> > > implementations
> > >>> > > > > > > > > > > >> > > > > >> > > > > are
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> > >>> enable the
> > >>> > > > > > feature the
> > >>> > > > > > > > > > > >> clients
> > >>> > > > > > > > > > > >> > need to
> > >>> > > > > > > > > > > >> > > > > >> > first
> > >>> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> > >>> rolling
> > >>> > > > bounce
> > >>> > > > > > and
> > >>> > > > > > > > > then set
> > >>> > > > > > > > > > > >> > the flag
> > >>> > > > > > > > > > > >> > > > > >> > with a
> > >>> > > > > > > > > > > >> > > > > >> > > > > second
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and
> hence
> > >>> we feel
> > >>> > > > it's
> > >>> > > > > > > > > better to be
> > >>> > > > > > > > > > > >> > released
> > >>> > > > > > > > > > > >> > > > > >> in a
> > >>> > > > > > > > > > > >> > > > > >> > > major
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> > >>> flesh out
> > >>> > > > the
> > >>> > > > > > > > > details of
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > migration
> > >>> > > > > > > > > > > >> > > > > >> > > story.
> > >>> > > > > > > > > > > >> > > > > >> > > > > It
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying
> we
> > >>> will
> > >>> > > > support
> > >>> > > > > > online
> > >>> > > > > > > > > > > >> > migrations. Is
> > >>> > > > > > > > > > > >> > > > > >> that
> > >>> > > > > > > > > > > >> > > > > >> > > > > correct?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > We
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this
> in
> > >>> detail.
> > >>> > > > It
> > >>> > > > > > could
> > >>> > > > > > > > > also be
> > >>> > > > > > > > > > > >> > done as a
> > >>> > > > > > > > > > > >> > > > > >> > > separate
> > >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > that's
> > >>> the
> > >>> > > > part we
> > >>> > > > > > can
> > >>> > > > > > > > > be more
> > >>> > > > > > > > > > > >> > concrete
> > >>> > > > > > > > > > > >> > > > > >> about
> > >>> > > > > > > > > > > >> > > > > >> > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > sure
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related
> > to
> > >>> your
> > >>> > > > > > question 2)
> > >>> > > > > > > > > above).
> > >>> > > > > > > > > > > >> > We will
> > >>> > > > > > > > > > > >> > > > > >> work
> > >>> > > > > > > > > > > >> > > > > >> > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > making
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > >>> parallel as we
> > >>> > > > > > solicit
> > >>> > > > > > > > > more
> > >>> > > > > > > > > > > >> > feedback.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
> > >>> that we are
> > >>> > > > > > pushing
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > pattern
> > >>> > > > > > > > > > > >> > > > > >> > > subscriptions
> > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> > >>> like there
> > >>> > > > > > could be
> > >>> > > > > > > > > some
> > >>> > > > > > > > > > > >> tricky
> > >>> > > > > > > > > > > >> > > > > >> > > compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > mechanism
> > >>> for
> > >>> > > > users to
> > >>> > > > > > > > > detect that
> > >>> > > > > > > > > > > >> > they
> > >>> > > > > > > > > > > >> > > > > need
> > >>> > > > > > > > > > > >> > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> > > update
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > switching
> > >>> to the
> > >>> > > > new
> > >>> > > > > > > > > protocol?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> > >>> need some
> > >>> > > > > > tooling for
> > >>> > > > > > > > > > > >> non-java
> > >>> > > > > > > > > > > >> > client
> > >>> > > > > > > > > > > >> > > > > >> > users
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > sort
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> > >>> client before
> > >>> > > > > > > > > switching to
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > new
> > >>> > > > > > > > > > > >> > > > > >> > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > I
> > >>> > > > > > > > > > > >> > > > > >> > > > > do not
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea
> on
> > >>> top of my
> > >>> > > > head
> > >>> > > > > > > > > though,
> > >>> > > > > > > > > > > >> maybe
> > >>> > > > > > > > > > > >> > others
> > >>> > > > > > > > > > > >> > > > > >> like
> > >>> > > > > > > > > > > >> > > > > >> > > @Matt
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to
> > >>> the last
> > >>> > > > > > question,
> > >>> > > > > > > > > will the
> > >>> > > > > > > > > > > >> > Java
> > >>> > > > > > > > > > > >> > > > > >> client
> > >>> > > > > > > > > > > >> > > > > >> > > allow
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with
> the
> > >>> current
> > >>> > > > regex
> > >>> > > > > > > > > engine for
> > >>> > > > > > > > > > > >> > > > > >> compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > For
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be
> handy
> > >>> to keep
> > >>> > > > using
> > >>> > > > > > > > > client
> > >>> > > > > > > > > > > >> based
> > >>> > > > > > > > > > > >> > regex
> > >>> > > > > > > > > > > >> > > > > at
> > >>> > > > > > > > > > > >> > > > > >> > > first to
> > >>> > > > > > > > > > > >> > > > > >> > > > > keep
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> > >>> simple and
> > >>> > > > then
> > >>> > > > > > migrate
> > >>> > > > > > > > > to
> > >>> > > > > > > > > > > >> > server based
> > >>> > > > > > > > > > > >> > > > > >> > > regexes
> > >>> > > > > > > > > > > >> > > > > >> > > > > as a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have
> > >>> not thought
> > >>> > > > > > about
> > >>> > > > > > > > > that for
> > >>> > > > > > > > > > > >> > java
> > >>> > > > > > > > > > > >> > > > > >> clients,
> > >>> > > > > > > > > > > >> > > > > >> > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > we can
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind
> > of
> > >>> > > > > > compatibility
> > >>> > > > > > > > > issues do
> > >>> > > > > > > > > > > >> > you have
> > >>> > > > > > > > > > > >> > > > > in
> > >>> > > > > > > > > > > >> > > > > >> > > mind?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we
> say
> > >>> that the
> > >>> > > > group
> > >>> > > > > > > > > coordinator
> > >>> > > > > > > > > > > >> > will be
> > >>> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > >>> configurations and
> > >>> > > > that
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> > configurations
> > >>> > > > > > > > > > > >> > > > > will
> > >>> > > > > > > > > > > >> > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> > > > > deleted
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > when
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> > >>> deleted. Will a
> > >>> > > > > > > > > transition to
> > >>> > > > > > > > > > > >> DEAD
> > >>> > > > > > > > > > > >> > > > > trigger
> > >>> > > > > > > > > > > >> > > > > >> > > deletion
> > >>> > > > > > > > > > > >> > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> configurations?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> > >>> DEAD state
> > >>> > > > is an
> > >>> > > > > > > > > ending
> > >>> > > > > > > > > > > >> state
> > >>> > > > > > > > > > > >> > (we
> > >>> > > > > > > > > > > >> > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> > >>> group is
> > >>> > > > EMPTY
> > >>> > > > > > and
> > >>> > > > > > > > > also all
> > >>> > > > > > > > > > > >> of
> > >>> > > > > > > > > > > >> > its
> > >>> > > > > > > > > > > >> > > > > >> > metadata
> > >>> > > > > > > > > > > >> > > > > >> > > are
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited
> to
> > >>> DEAD this
> > >>> > > > > > group
> > >>> > > > > > > > > would
> > >>> > > > > > > > > > > >> never
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > > >> revived.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> > >>> choice to
> > >>> > > > store
> > >>> > > > > > the
> > >>> > > > > > > > > configs
> > >>> > > > > > > > > > > >> in
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to
> list
> > >>> all
> > >>> > > > cluster
> > >>> > > > > > configs
> > >>> > > > > > > > > and
> > >>> > > > > > > > > > > >> their
> > >>> > > > > > > > > > > >> > > > > values?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> > >>> question,
> > >>> > > > and our
> > >>> > > > > > > > > thoughts
> > >>> > > > > > > > > > > >> are
> > >>> > > > > > > > > > > >> > that
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > so-called
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations"
> > are
> > >>> > > > overrides
> > >>> > > > > > of the
> > >>> > > > > > > > > > > >> > cluster-level
> > >>> > > > > > > > > > > >> > > > > >> > > > > configurations
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> > >>> group so
> > >>> > > > when an
> > >>> > > > > > > > > admin list
> > >>> > > > > > > > > > > >> > cluster
> > >>> > > > > > > > > > > >> > > > > >> > configs
> > >>> > > > > > > > > > > >> > > > > >> > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> > >>> cluster-level
> > >>> > > > > > > > > "defaults", not
> > >>> > > > > > > > > > > >> > showing
> > >>> > > > > > > > > > > >> > > > > >> any
> > >>> > > > > > > > > > > >> > > > > >> > > > > per-group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> > >>> someone
> > >>> > > > > > configure a
> > >>> > > > > > > > > group
> > >>> > > > > > > > > > > >> > before
> > >>> > > > > > > > > > > >> > > > > >> starting
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> > >>> allowing
> > >>> > > > the
> > >>> > > > > > explicit
> > >>> > > > > > > > > > > >> > creation of
> > >>> > > > > > > > > > > >> > > > > >> groups?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could
> > be
> > >>> > > > decoupled
> > >>> > > > > > from
> > >>> > > > > > > > > the group
> > >>> > > > > > > > > > > >> > > > > lifecycle.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can
> > be
> > >>> created
> > >>> > > > > > before
> > >>> > > > > > > > > the group
> > >>> > > > > > > > > > > >> > itself
> > >>> > > > > > > > > > > >> > > > > as
> > >>> > > > > > > > > > > >> > > > > >> an
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course,
> > this
> > >>> > > > requires
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> corresponding
> > >>> > > > > > > > > > > >> > > > > >> request to
> > >>> > > > > > > > > > > >> > > > > >> > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right
> coordinator
> > >>> based on
> > >>> > > > the
> > >>> > > > > > group
> > >>> > > > > > > > > id ---
> > >>> > > > > > > > > > > >> > the only
> > >>> > > > > > > > > > > >> > > > > >> > thing
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> > >>> itself is
> > >>> > > > gone
> > >>> > > > > > we
> > >>> > > > > > > > > also check
> > >>> > > > > > > > > > > >> > if there
> > >>> > > > > > > > > > > >> > > > > >> are
> > >>> > > > > > > > > > > >> > > > > >> > > any
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities
> related
> > >>> to that
> > >>> > > > group
> > >>> > > > > > and
> > >>> > > > > > > > > delete
> > >>> > > > > > > > > > > >> as
> > >>> > > > > > > > > > > >> > well.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> > >>> indeed
> > >>> > > > > > introduces an
> > >>> > > > > > > > > > > >> > asymmetry on
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > creation
> > >>> > > > > > > > > > > >> > > > > >> > > > > /
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of
> the
> > >>> config
> > >>> > > > > > entities,
> > >>> > > > > > > > > and we
> > >>> > > > > > > > > > > >> > would like
> > >>> > > > > > > > > > > >> > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> > > hear
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings
> whether
> > >>> we should
> > >>> > > > aim
> > >>> > > > > > for
> > >>> > > > > > > > > symmetry
> > >>> > > > > > > > > > > >> > i.e.
> > >>> > > > > > > > > > > >> > > > > >> totally
> > >>> > > > > > > > > > > >> > > > > >> > > > > decouple
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and
> hence
> > >>> not
> > >>> > > > delete
> > >>> > > > > > them at
> > >>> > > > > > > > > all
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> group
> > >>> > > > > > > > > > > >> > > > > >> > is
> > >>> > > > > > > > > > > >> > > > > >> > > > > gone,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > but
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require
> explicit
> > >>> deletion
> > >>> > > > > > operations
> > >>> > > > > > > > > by
> > >>> > > > > > > > > > > >> > themselves.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > >>> > > > > > Consumer.subscribe
> > >>> > > > > > > > > method for
> > >>> > > > > > > > > > > >> > the Java
> > >>> > > > > > > > > > > >> > > > > >> > client
> > >>> > > > > > > > > > > >> > > > > >> > > > > still
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > take
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > `java.util.regex.Pattern` of
> > >>> > > > > > do we
> > >>> > > > > > > > > have
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> introduce an
> > >>> > > > > > > > > > > >> > > > > >> > > > > overload?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do
> not
> > >>> need to
> > >>> > > > > > introduce
> > >>> > > > > > > > > an
> > >>> > > > > > > > > > > >> > overload, but
> > >>> > > > > > > > > > > >> > > > > >> I'm
> > >>> > > > > > > > > > > >> > > > > >> > > all
> > >>> > > > > > > > > > > >> > > > > >> > > > > ears
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > >>> compatibility
> > >>> > > > > > issues
> > >>> > > > > > > > > that we
> > >>> > > > > > > > > > > >> may
> > >>> > > > > > > > > > > >> > > > > overlook.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree
> > with
> > >>> Justine
> > >>> > > > > > that we
> > >>> > > > > > > > > should
> > >>> > > > > > > > > > > >> be
> > >>> > > > > > > > > > > >> > clearer
> > >>> > > > > > > > > > > >> > > > > >> > about
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > >>> > > > > > IBP/metadata.version
> > >>> > > > > > > > > from the
> > >>> > > > > > > > > > > >> > feature
> > >>> > > > > > > > > > > >> > > > > >> flag.
> > >>> > > > > > > > > > > >> > > > > >> > > Maybe
> > >>> > > > > > > > > > > >> > > > > >> > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch
> the
> > >>> default
> > >>> > > > for
> > >>> > > > > > the
> > >>> > > > > > > > > feature
> > >>> > > > > > > > > > > >> flag
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > true
> > >>> > > > > > > > > > > >> > > > > >> > > based on
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > metadata.version
> > >>> once we
> > >>> > > > > > want to
> > >>> > > > > > > > > make it
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> default.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> > >>> protocol
> > >>> > > > APIs
> > >>> > > > > > don't
> > >>> > > > > > > > > mention
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> required
> > >>> > > > > > > > > > > >> > > > > >> > > ACLs,
> > >>> > > > > > > > > > > >> > > > > >> > > > > it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add
> > that
> > >>> for
> > >>> > > > > > consistency.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > >>> certainly do
> > >>> > > > that.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a
> bit
> > >>> odd that
> > >>> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > >>> > > > > > > > > > > >> > > > > >> > requires
> > >>> > > > > > > > > > > >> > > > > >> > > > > "Read
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it
> seems
> > >>> to do
> > >>> > > > more
> > >>> > > > > > than
> > >>> > > > > > > > > reading.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> > thought
> > >>> myself
> > >>> > > > as
> > >>> > > > > > well,
> > >>> > > > > > > > > but in
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > end we
> > >>> > > > > > > > > > > >> > > > > >> > could
> > >>> > > > > > > > > > > >> > > > > >> > > not
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > >>> alternative: adding
> > >>> > > > > > Write
> > >>> > > > > > > > > Group
> > >>> > > > > > > > > > > >> seems
> > >>> > > > > > > > > > > >> > an
> > >>> > > > > > > > > > > >> > > > > >> overkill
> > >>> > > > > > > > > > > >> > > > > >> > > here
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > >>> elsewhere (we
> > >>> > > > only
> > >>> > > > > > have
> > >>> > > > > > > > > Read /
> > >>> > > > > > > > > > > >> > Delete
> > >>> > > > > > > > > > > >> > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > Describe
> > >>> > > > > > > > > > > >> > > > > >> > > > > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would
> like
> > >>> to hear
> > >>> > > > others
> > >>> > > > > > > > > thoughts.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> > topic
> > >>> > > > recreation
> > >>> > > > > > > > > handled by
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> > >>> would be
> > >>> > > > good
> > >>> > > > > > to have
> > >>> > > > > > > > > a
> > >>> > > > > > > > > > > >> > section on
> > >>> > > > > > > > > > > >> > > > > >> this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with
> > regex
> > >>> > > > > > subscription
> > >>> > > > > > > > > right? Yes
> > >>> > > > > > > > > > > >> > we can
> > >>> > > > > > > > > > > >> > > > > >> add a
> > >>> > > > > > > > > > > >> > > > > >> > > > > section
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > about
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> > >>> basically the
> > >>> > > > idea
> > >>> > > > > > is that
> > >>> > > > > > > > > > > >> consumer
> > >>> > > > > > > > > > > >> > would
> > >>> > > > > > > > > > > >> > > > > be
> > >>> > > > > > > > > > > >> > > > > >> > > totally
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new
> protocol
> > >>> as it's
> > >>> > > > > > handled all
> > >>> > > > > > > > > by the
> > >>> > > > > > > > > > > >> > brokers.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> > >>> mentions we
> > >>> > > > will
> > >>> > > > > > write
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> new
> > >>> > > > > > > > > > > >> > > > > >> coordinator
> > >>> > > > > > > > > > > >> > > > > >> > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > Java.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > >>> implementation
> > >>> > > > > > detail,
> > >>> > > > > > > > > do we
> > >>> > > > > > > > > > > >> > plan to
> > >>> > > > > > > > > > > >> > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > gradle
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> > >>> thought about
> > >>> > > > > > that. But
> > >>> > > > > > > > > I think
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > answer
> > >>> > > > > > > > > > > >> > > > > >> > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we
> have
> > a
> > >>> > > > scalability
> > >>> > > > > > goal
> > >>> > > > > > > > > when it
> > >>> > > > > > > > > > > >> > comes to
> > >>> > > > > > > > > > > >> > > > > >> how
> > >>> > > > > > > > > > > >> > > > > >> > > many
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > members
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group
> protocol
> > >>> can
> > >>> > > > support?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a
> group, I
> > >>> think we
> > >>> > > > > > should
> > >>> > > > > > > > > shoot for
> > >>> > > > > > > > > > > >> > 1000s of
> > >>> > > > > > > > > > > >> > > > > >> > > members.
> > >>> > > > > > > > > > > >> > > > > >> > > > > But
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability
> goals
> > >>> also
> > >>> > > > depend
> > >>> > > > > > on the
> > >>> > > > > > > > > offset
> > >>> > > > > > > > > > > >> > > > > management
> > >>> > > > > > > > > > > >> > > > > >> > > (commit,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of
> > the
> > >>> > > > coordinator
> > >>> > > > > > > > > which we
> > >>> > > > > > > > > > > >> did
> > >>> > > > > > > > > > > >> > not
> > >>> > > > > > > > > > > >> > > > > >> cover in
> > >>> > > > > > > > > > > >> > > > > >> > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> > >>> number that
> > >>> > > > > > applies
> > >>> > > > > > > > > > > >> > universally.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > >>> consider
> > >>> > > > having
> > >>> > > > > > > > > > > >> > SubscribedTopidIds
> > >>> > > > > > > > > > > >> > > > > >> instead
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> SubscribedTopicNames in
> > >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > >>> > > > > > > > > > > >> > > > > >> > > Is the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have
> > to
> > >>> resolve
> > >>> > > > the
> > >>> > > > > > regex
> > >>> > > > > > > > > on the
> > >>> > > > > > > > > > > >> > server,
> > >>> > > > > > > > > > > >> > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > can
> > >>> > > > > > > > > > > >> > > > > >> > > do
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic
> name?
> > >>> The
> > >>> > > > > > difference is
> > >>> > > > > > > > > that
> > >>> > > > > > > > > > > >> > sending the
> > >>> > > > > > > > > > > >> > > > > >> > regex
> > >>> > > > > > > > > > > >> > > > > >> > > is
> > >>> > > > > > > > > > > >> > > > > >> > > > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas
> sending
> > >>> the topic
> > >>> > > > > > names is
> > >>> > > > > > > > > less
> > >>> > > > > > > > > > > >> > efficient.
> > >>> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> > >>> easier to
> > >>> > > > > > handle if
> > >>> > > > > > > > > we have
> > >>> > > > > > > > > > > >> > topic
> > >>> > > > > > > > > > > >> > > > > ids.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason
> > to
> > >>> still
> > >>> > > > let
> > >>> > > > > > the
> > >>> > > > > > > > > clients
> > >>> > > > > > > > > > > >> send
> > >>> > > > > > > > > > > >> > names
> > >>> > > > > > > > > > > >> > > > > >> is to
> > >>> > > > > > > > > > > >> > > > > >> > > keep
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> > names
> > >>> -> ids
> > >>> > > > on
> > >>> > > > > > the
> > >>> > > > > > > > > broker /
> > >>> > > > > > > > > > > >> > admin
> > >>> > > > > > > > > > > >> > > > > client
> > >>> > > > > > > > > > > >> > > > > >> > > only.
> > >>> > > > > > > > > > > >> > > > > >> > > > > Note
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we
> added
> > >>> topic id
> > >>> > > > in
> > >>> > > > > > > > > KIP-516, we
> > >>> > > > > > > > > > > >> > never
> > >>> > > > > > > > > > > >> > > > > >> > > implemented the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > consumer/producers
> > >>> > > > leveraging
> > >>> > > > > > the
> > >>> > > > > > > > > related
> > >>> > > > > > > > > > > >> > newer
> > >>> > > > > > > > > > > >> > > > > >> versioned
> > >>> > > > > > > > > > > >> > > > > >> > > RPCs,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> > topic
> > >>> id as
> > >>> > > > empty
> > >>> > > > > > UUID.
> > >>> > > > > > > > > We
> > >>> > > > > > > > > > > >> want
> > >>> > > > > > > > > > > >> > to keep
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and
> > only
> > >>> > > > delegate
> > >>> > > > > > the
> > >>> > > > > > > > > reasoning
> > >>> > > > > > > > > > > >> on
> > >>> > > > > > > > > > > >> > broker
> > >>> > > > > > > > > > > >> > > > > >> and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul
> 6,
> > >>> 2022 at
> > >>> > > > 10:45
> > >>> > > > > > AM
> > >>> > > > > > > > > David
> > >>> > > > > > > > > > > >> Jacot
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> <dj...@confluent.io.invalid>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would
> like
> > >>> to start a
> > >>> > > > > > > > > discussion
> > >>> > > > > > > > > > > >> thread
> > >>> > > > > > > > > > > >> > on
> > >>> > > > > > > > > > > >> > > > > >> KIP-848:
> > >>> > > > > > > > > > > >> > > > > >> > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > Next
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation
> of
> > >>> the
> > >>> > > > Consumer
> > >>> > > > > > > > > Rebalance
> > >>> > > > > > > > > > > >> > Protocol.
> > >>> > > > > > > > > > > >> > > > > >> With
> > >>> > > > > > > > > > > >> > > > > >> > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> > >>> rebalance
> > >>> > > > > > protocol
> > >>> > > > > > > > > (for
> > >>> > > > > > > > > > > >> > consumers)
> > >>> > > > > > > > > > > >> > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> > >>> easier to
> > >>> > > > > > implement for
> > >>> > > > > > > > > > > >> > clients, and
> > >>> > > > > > > > > > > >> > > > > >> easier
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > debug
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> > >>> here:
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please
> take a
> > >>> look and
> > >>> > > > let
> > >>> > > > > > me
> > >>> > > > > > > > > know what
> > >>> > > > > > > > > > > >> > you
> > >>> > > > > > > > > > > >> > > > > think.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will
> be
> > >>> away from
> > >>> > > > > > July
> > >>> > > > > > > > > 18th to
> > >>> > > > > > > > > > > >> > August 8th.
> > >>> > > > > > > > > > > >> > > > > >> That
> > >>> > > > > > > > > > > >> > > > > >> > > gives
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> > >>> read and
> > >>> > > > digest
> > >>> > > > > > this
> > >>> > > > > > > > > long
> > >>> > > > > > > > > > > >> KIP.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> --
> > >>> > > > > > > > > > > >> > > > > >> -- Guozhang
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > > >
> > >>> > > > >
> > >>> > > > > --
> > >>> > > > > -- Guozhang
> > >>> > > >
> > >>> > >
> > >>> > >
> > >>> > > --
> > >>> > > -- Guozhang
> > >>>
> > >>
> > >>
> > >> --
> > >> -- Guozhang
> > >>
> > >
> > >
> > > --
> > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
> >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Jun,

Thanks for your comments.

40. No, there is not. This is left to the implementer of the
interface. For instance, Streams would consult its internal state in
order to compute the metadata. The metadata is internal to the
assignor so providing a Consumer API to update it does not seem useful
to me. However, it may be useful to have a way to pass an instance of
the assignor vs passing a class to be instantiated.

44.1 This is explained in the `Assignor Selection` chapter. Do you
think that we need more details?

44.2 OK.

48. Right. The member assignment is what the client has whereas the
target is what the member will eventually have. "Eventually" is
important here because it means that the client side assignor takes a
decision on possibly stale metadata if we don't use the target
assignment. We could provide both but don't you think that providing
both will make things even more confusing? I think that users won't
know which one to use. I still believe that providing the current
target is the right one here.

50. TopicMetadata contains a Uuid and the number of partitions.
Knowing that partitions are always indexed from 0, it is possible to
co-partition across topics, no? Do you think that we should explicitly
pass all the partition indexes? That would make the protocol more
robust, I suppose.

51. Yeah, that is debatable. If they are multiple assignors, their
metadata do not have to be the same. For instance, one version could
encore the lag in the metadata whereas a newer version might use a
different approach. This is why, I think, that having it at the
assignor level makes sense.

52. We could just use metadata. With the reason, we are trying to
bring more visibility into the protocol for Streams. That would help
us to better see why Streams did something without requiring to parse
the bytes on the server side.

53. MemberAssignment.metadata is created by the assignor as part of
the new assignment. The new assignment overrides the target assignment
on the server side. The metadata in the group state is not impacted by
this. This one is provided by PartitionAssignor#metadata().

Best,
David

On Tue, Sep 20, 2022 at 2:32 AM Jun Rao <ju...@confluent.io.invalid> wrote:
>
> Hi, David,
>
> Thanks for the reply. A few more comments.
>
> 40. Is there an API to change PartitionAssignor#metadata? I assume the
> member fields are just for reads?
>
> 43.2 Sounds good.
>
> 44.1 Thanks. Could we document that the coordinator selects the assignor
> when multiple assignors are specified in the client?
>
> 44.2 Encoding the group state in ConsumerGroupPrepareAssignmentResponse
> sounds simpler and clearer to me.
>
> 48. Yes, that's the confusing part. I assume member assignment is what the
> member client has. Could we include member assignment from both the
> coordinator and the client's point of view?
>
> 50. We only include Uuid in PartitionAssignor.Group.TopicMetadata. How does
> KStream do co-partitioning across topics?
>
> 51. PartitionAssignor.metadata: Is this really an assignor level thing? The
> use case mentioned is to use metadata for encoding lags in standby tasks.
> However, the lag is independent of the assignor. For example, if a member
> sets two assignors, does it need to set the same metadata in both assignors?
>
> 52. "power users will have the ability to trigger a reassignment by either
> providing a non-zero reason or by updating the assignor metadata." : Do we
> need both reason and metadata to trigger an assignment? Could we just use
> metadata?
>
> 53. MemberAssignment.metadata: How is that used? Does that overwrite the
> member metadata on the coordinator?
>
> Thanks,
>
> Jun
>
>
> On Tue, Sep 13, 2022 at 1:59 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> > Hi David,
> >
> > Another quick thing I just realized about "PartitionAssignor#onAssignment":
> > in the new proposal, this function could be triggered multiple time within
> > a group generation as it will be triggered whenever the broker gives the
> > client new assignments, whereas today it is only triggered once per
> > rebalance. I think it's worth emphasizing this a bit in the doc, especially
> > around:
> >
> > 1) how many times `onAssignment` could be triggered with a single rebalance
> > (especially if the new assignment is decided by the client-assignor).
> > 2) when the Assignment metadata could change across multiple `onAssignment`
> > calls, and how users should handle it.
> > 3) the order of calling `onAssignment` and `onPartitionsRevoked/Assigned`.
> >
> > For example, today the Streams client relies on the following assumptions:
> > 1) within a rebalance the onPartitionsRevoked would be called before the
> > onAssignment, and onAssignment is called before onPartitionsAssigned (today
> > we would trigger all three only once within a generation), 2) onAssignment
> > contains the source-of-truth of the final assignment, and hence we can
> > simply ignore onPartitionsAssigned and just rely on onAssignment to reason
> > about which tasks are removed / assigned. And each time it calls, the
> > corresponding metadata change are "final" for this new generation.
> >
> > We can change however Streams handle the new protocol, but I think there
> > are other clients externally which may also make such assumptions about the
> > ordering / call times about those callbacks, and hence worth clarifying
> > those in the new proposal.
> >
> >
> >
> > Guozhang
> >
> >
> >
> >
> > On Tue, Sep 13, 2022 at 12:20 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > > Hello David,
> > >
> > > Just had a few more nit questions about the public APIs:
> > >
> > > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> > field,
> > > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > > rationale seems to be assuming that we should always be able to do the
> > > assignment at the broker-side assignor without errors. Personally I think
> > > it's still potentially beneficial to add the Error field even for
> > > broker-side assignors, e.g. for some edge cases where some subscribed
> > > topics are not recognized with the current broker's metadata. What do you
> > > think?
> > >
> > > 2) The client-side "GroupMember" has three additional fields
> > > reason/version/metadata compared with the broker-side GroupMember. I
> > agree
> > > that broker-side assignor would not need reason/metadata since they are
> > > blackbox strings/bytes to the assignor, but what about version? E.g. is
> > it
> > > possible that we evolve our broker-side built-in assignor but the old
> > > versioned clients would not be able to work with the new version, in
> > which
> > > case we need to let the broker being aware of this and upgrade its
> > behavior
> > > to cooperate with the clients?
> > >
> > > 3) Also related to 2) above, for the client-side "GroupMember", instead
> > of
> > > including these three fields, what about just adding the "Metadata" field
> > > class which has these three fields? Also, there are two "Metadata"
> > > currently in the APIs, the first is a class that encodes
> > > reason/version/metadata, and the second is just the encoded metadata
> > bytes.
> > > I'm wondering what about just naming the first as memberMetadata, which
> > > then has a bytebuffer field Metadata, or instead naming the second
> > > bytebuffer field as metadataBytes?
> > >
> > > Guozhang
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > >> Hello David,
> > >>
> > >> Thanks for bringing this question up. I think the main benefits as you
> > >> listed is 2) above if it stays; just to clarify, we would only be able
> > to
> > >> save one round trip if the rebalance is still triggered by the broker;
> > if
> > >> the rebalance is triggered by the client then the num.round.trips are
> > the
> > >> same:
> > >>
> > >> 1) With GroupPrepareAssignment:
> > >>
> > >> T0: client decides to do a new assignment, suppose it has already sent a
> > >> HB and hence has to wait for it to return first since only one request /
> > >> response can be inflight with the coordinator's socket.
> > >> T1: client receives the HB response, and then sends the
> > >> GroupPrepareAssignment request.
> > >> T2: the GroupPrepareAssignment response is returned.
> > >> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> > >> request.
> > >>
> > >> In total, two round trips.
> > >>
> > >> 2) Without GroupPrepareAssignment:
> > >>
> > >> T0: client decides to do a new assignment, suppose it has already sent a
> > >> HB and hence has to wait for it to return first since only one request /
> > >> response can be inflight with the coordinator's socket.
> > >> T1: client receives the HB response, and then sends the new HB request
> > >> with the flag indicating a new rebalance needed..
> > >> T2: the HB response with the optional member metadata map is returned.
> > >> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> > >> request.
> > >>
> > >> In total, two round trips as well.
> > >>
> > >> -----------------------------
> > >>
> > >> So to complete the full picture here, we'd need to modify both HB
> > request
> > >> and response so that the client can also indicate a new rebalance via
> > the
> > >> HB request as well, right?
> > >>
> > >> Assuming all above is true, I think it's okay to merge the
> > >> GroupPrepareAssignment into HB given that we can make the additional
> > fields
> > >> encoding the full member (subscription) metadata and topic metadata as
> > >> optional fields.
> > >>
> > >>
> > >> Guozhang
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >> On Mon, Sep 12, 2022 at 5:22 AM David Jacot <djacot@confluent.io.invalid
> > >
> > >> wrote:
> > >>
> > >>> Hi all,
> > >>>
> > >>> During an offline conversation, someone asked why we need the
> > >>> ConsumerGroupPrepareAssignment API and suggested that we could instead
> > >>> provide the group state in the heartbeat response. This has a few
> > >>> advantages: 1) it does not require using a special error code to
> > >>> signal that a new assignment is required as the signal would be the
> > >>> provided group state; 2) it removes one round trip when a client side
> > >>> assignor is used. The downside is that it makes the heartbeat
> > >>> response's definition quite large. I recall that I went with the
> > >>> current approach due to this.
> > >>>
> > >>> Providing the group state in the heartbeat response is appealing. What
> > >>> do you guys think?
> > >>>
> > >>> Best,
> > >>> David
> > >>>
> > >>> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> > wrote:
> > >>> >
> > >>> > Hi Guozhang,
> > >>> >
> > >>> > 1. I have added a reference to the relevant chapter instead of
> > >>> > repeating the whole thing. Does that work for you?
> > >>> >
> > >>> > 2. The "Rebalance Triggers" section you are referring to is about
> > when
> > >>> > a rebalance should be triggered for the non-upgraded members using
> > the
> > >>> > old protocol. The section mentions that a rebalance must be triggered
> > >>> > when a new assignment is installed. This implies that the group epoch
> > >>> > was updated either by a native member or a non-upgraded member. For
> > >>> > the latter, the JoinGroup request would be the trigger. I have added
> > a
> > >>> > reference to the relevant chapter in the "JoinGroup Handling" section
> > >>> > as well. Does that make sense?
> > >>> >
> > >>> > Thanks,
> > >>> > David
> > >>> >
> > >>> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> > >>> wrote:
> > >>> > >
> > >>> > > Hello David,
> > >>> > >
> > >>> > > Alright I think that's sufficient. Just to make that clear in the
> > >>> doc,
> > >>> > > could we update:
> > >>> > >
> > >>> > > 1) the heartbeat request handling section, stating when coordinator
> > >>> will
> > >>> > > trigger rebalance based on the HB's member metadata / reason?
> > >>> > > 2) the "Rebalance Triggers" section to include what we described in
> > >>> "Group
> > >>> > > Epoch - Trigger a rebalance" section as well?
> > >>> > >
> > >>> > >
> > >>> > > Guozhang
> > >>> > >
> > >>> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> > >>> <dj...@confluent.io.invalid>
> > >>> > > wrote:
> > >>> > >
> > >>> > > > Hi Guozhang,
> > >>> > > >
> > >>> > > > I thought that the assignor will always be consulted when the
> > next
> > >>> > > > heartbeat request is constructed. In other words,
> > >>> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
> > >>> This
> > >>> > > > gives the opportunity for the assignor to enforce a rebalance by
> > >>> > > > setting the reason to a non-zero value or by changing the bytes.
> > Do
> > >>> > > > you think that this is not sufficient? Are you concerned by the
> > >>> delay?
> > >>> > > >
> > >>> > > > Best,
> > >>> > > > David
> > >>> > > >
> > >>> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wangguoz@gmail.com
> > >
> > >>> wrote:
> > >>> > > > >
> > >>> > > > > Hello David,
> > >>> > > > >
> > >>> > > > > One of Jun's comments make me thinking:
> > >>> > > > >
> > >>> > > > > ```
> > >>> > > > > In this case, a new assignment is triggered by the client side
> > >>> > > > > assignor. When constructing the HB, the consumer will always
> > >>> consult
> > >>> > > > > the client side assignor and propagate the information to the
> > >>> group
> > >>> > > > > coordinator. In other words, we don't expect users to call
> > >>> > > > > Consumer#enforceRebalance anymore.
> > >>> > > > > ```
> > >>> > > > >
> > >>> > > > > As I looked at the current PartitionAssignor's interface, we
> > >>> actually do
> > >>> > > > > not have a way yet to instruct how to construct the next HB
> > >>> request, e.g.
> > >>> > > > > when the assignor wants to enforce a new rebalance with a new
> > >>> assignment,
> > >>> > > > > we'd need some customizable APIs inside the PartitionAssignor
> > to
> > >>> indicate
> > >>> > > > > the next HB telling broker about so. WDYT about adding such an
> > >>> API on the
> > >>> > > > > PartitionAssignor?
> > >>> > > > >
> > >>> > > > >
> > >>> > > > > Guozhang
> > >>> > > > >
> > >>> > > > >
> > >>> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> > >>> <dj...@confluent.io.invalid>
> > >>> > > > > wrote:
> > >>> > > > >
> > >>> > > > > > Hi Jun,
> > >>> > > > > >
> > >>> > > > > > I have updated the KIP to include your feedback. I have also
> > >>> tried to
> > >>> > > > > > clarify the parts which were not cleared.
> > >>> > > > > >
> > >>> > > > > > Best,
> > >>> > > > > > David
> > >>> > > > > >
> > >>> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> > >>> djacot@confluent.io>
> > >>> > > > wrote:
> > >>> > > > > > >
> > >>> > > > > > > Hi Jun,
> > >>> > > > > > >
> > >>> > > > > > > Thanks for your feedback. Let me start by answering your
> > >>> questions
> > >>> > > > > > > inline and I will update the KIP next week.
> > >>> > > > > > >
> > >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> > >>> seem to
> > >>> > > > be
> > >>> > > > > > fewer
> > >>> > > > > > > > RPCs during rebalance and more efficient support of
> > >>> wildcard. A few
> > >>> > > > > > > > comments below.
> > >>> > > > > > >
> > >>> > > > > > > I would also add that the KIP removes the global sync
> > >>> barrier in the
> > >>> > > > > > > protocol which is essential to improve group stability and
> > >>> > > > > > > scalability, and the KIP also simplifies the client by
> > >>> moving most of
> > >>> > > > > > > the logic to the server side.
> > >>> > > > > > >
> > >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> > >>> rolling
> > >>> > > > > > changing
> > >>> > > > > > > > of the partition assignor in the consumers?
> > >>> > > > > > >
> > >>> > > > > > > Definitely. The group coordinator will use the assignor
> > used
> > >>> by a
> > >>> > > > > > > majority of the members. This allows the group to move from
> > >>> one
> > >>> > > > > > > assignor to another by a roll. This is explained in the
> > >>> Assignor
> > >>> > > > > > > Selection chapter.
> > >>> > > > > > >
> > >>> > > > > > > > 30.2 For each field, could you explain whether it's
> > >>> required in
> > >>> > > > every
> > >>> > > > > > > > request or the scenarios when it needs to be filled? For
> > >>> example,
> > >>> > > > it's
> > >>> > > > > > not
> > >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > >>> > > > > > >
> > >>> > > > > > > The client is expected to set those fields in case of a
> > >>> connection
> > >>> > > > > > > issue (e.g. timeout) or when the fields have changed since
> > >>> the last
> > >>> > > > > > > HB. The server populates those fields as long as the member
> > >>> is not
> > >>> > > > > > > fully reconciled - the member should acknowledge that it
> > has
> > >>> the
> > >>> > > > > > > expected epoch and assignment. I will clarify this in the
> > >>> KIP.
> > >>> > > > > > >
> > >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > >>> between the
> > >>> > > > > > client
> > >>> > > > > > > > and the broker is only considered during fetching, but
> > not
> > >>> during
> > >>> > > > > > assigning
> > >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > is
> > >>> made,
> > >>> > > > there
> > >>> > > > > > is
> > >>> > > > > > > > no opportunity for read affinity because no replicas of
> > >>> assigned
> > >>> > > > > > partitions
> > >>> > > > > > > > are close to the member. I am wondering if we should use
> > >>> this
> > >>> > > > > > opportunity
> > >>> > > > > > > > to address this by including rack in GroupMember.
> > >>> > > > > > >
> > >>> > > > > > > That's an interesting idea. I don't see any issue with
> > >>> adding the
> > >>> > > > rack
> > >>> > > > > > > to the members. I will do so.
> > >>> > > > > > >
> > >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> > >>> busy a group
> > >>> > > > > > > > coordinator is. By moving the event loop model, it seems
> > >>> that we
> > >>> > > > could
> > >>> > > > > > add
> > >>> > > > > > > > a metric that tracks the fraction of the time the event
> > >>> loop is
> > >>> > > > doing
> > >>> > > > > > the
> > >>> > > > > > > > actual work.
> > >>> > > > > > >
> > >>> > > > > > > That's a great idea. I will add it. Thanks.
> > >>> > > > > > >
> > >>> > > > > > > > 33. Could we add a section on coordinator failover
> > >>> handling? For
> > >>> > > > > > example,
> > >>> > > > > > > > does it need to trigger the check if any group with the
> > >>> wildcard
> > >>> > > > > > > > subscription now has a new matching topic?
> > >>> > > > > > >
> > >>> > > > > > > Sure. When the new group coordinator takes over, it has to:
> > >>> > > > > > > * Setup the session timeouts.
> > >>> > > > > > > * Trigger a new assignment if a client side assignor is
> > >>> used. We
> > >>> > > > don't
> > >>> > > > > > > store the information about the member selected to run the
> > >>> assignment
> > >>> > > > > > > so we have to start a new one.
> > >>> > > > > > > * Update the topics metadata, verify the wildcard
> > >>> subscriptions, and
> > >>> > > > > > > trigger a rebalance if needed.
> > >>> > > > > > >
> > >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> > >>> > > > ConsumerGroupPartitionMetadataValue,
> > >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> > >>> the epoch
> > >>> > > > > > field
> > >>> > > > > > > > reflects? For example, does the epoch in
> > >>> ConsumerGroupMetadataValue
> > >>> > > > > > reflect
> > >>> > > > > > > > the latest group epoch? What about the one in
> > >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > >>> > > > > > ConsumerGroupMemberMetadataValue?
> > >>> > > > > > >
> > >>> > > > > > > Sure. I will clarify that but it is always the latest group
> > >>> epoch.
> > >>> > > > > > > When the group state is updated, the group epoch is bumped
> > >>> so we use
> > >>> > > > > > > that one for all the change records related to the update.
> > >>> > > > > > >
> > >>> > > > > > > > 35. "the group coordinator will ensure that the following
> > >>> > > > invariants
> > >>> > > > > > are
> > >>> > > > > > > > met: ... All members exists." It's possible for a member
> > >>> not to
> > >>> > > > get any
> > >>> > > > > > > > assigned partitions, right?
> > >>> > > > > > >
> > >>> > > > > > > That's right. Here I meant that the members provided by the
> > >>> assignor
> > >>> > > > > > > in the assignment must exist in the group. The assignor can
> > >>> not make
> > >>> > > > > > > up new member ids.
> > >>> > > > > > >
> > >>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
> > >>> to 0":
> > >>> > > > When
> > >>> > > > > > would
> > >>> > > > > > > > a consumer rejoin and what member id would be used?
> > >>> > > > > > >
> > >>> > > > > > > A member is expected to abandon all its partitions and
> > >>> rejoins when
> > >>> > > > it
> > >>> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> > >>> group
> > >>> > > > > > > coordinator will have removed the member from the group.
> > The
> > >>> member
> > >>> > > > > > > can rejoin the group with the same member id but with 0 as
> > >>> epoch. Let
> > >>> > > > > > > me see if I can clarify this in the KIP.
> > >>> > > > > > >
> > >>> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger
> > >>> a
> > >>> > > > > > reassignment
> > >>> > > > > > > > by either providing a non-zero reason or by updating the
> > >>> assignor
> > >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > >>> deprecation
> > >>> > > > of
> > >>> > > > > > > > Consumer#enforeRebalance.
> > >>> > > > > > >
> > >>> > > > > > > In this case, a new assignment is triggered by the client
> > >>> side
> > >>> > > > > > > assignor. When constructing the HB, the consumer will
> > always
> > >>> consult
> > >>> > > > > > > the client side assignor and propagate the information to
> > >>> the group
> > >>> > > > > > > coordinator. In other words, we don't expect users to call
> > >>> > > > > > > Consumer#enforceRebalance anymore.
> > >>> > > > > > >
> > >>> > > > > > > > 38. The reassignment examples are nice. But the section
> > >>> seems to
> > >>> > > > have
> > >>> > > > > > > > multiple typos.
> > >>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> > >>> gets into
> > >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > >>> into
> > >>> > > > > > epoch=3,
> > >>> > > > > > > > partitions=[foo-1] too early.
> > >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > >>> epoch=2,
> > >>> > > > > > > > partitions=[foo-0].
> > >>> > > > > > >
> > >>> > > > > > > Sorry for that! I will revise them.
> > >>> > > > > > >
> > >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > >>> upgrade from
> > >>> > > > any
> > >>> > > > > > old
> > >>> > > > > > > > version to new one?
> > >>> > > > > > >
> > >>> > > > > > > We will support upgrading from the consumer protocol
> > version
> > >>> 3,
> > >>> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> > the
> > >>> earliest
> > >>> > > > > > > version is unknown at the moment. This is explained in the
> > >>> migration
> > >>> > > > > > > plan chapter.
> > >>> > > > > > >
> > >>> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> > >>> based on
> > >>> > > > it
> > >>> > > > > > > next week.
> > >>> > > > > > >
> > >>> > > > > > > Best,
> > >>> > > > > > > David
> > >>> > > > > > >
> > >>> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> > >>> <ju...@confluent.io.invalid>
> > >>> > > > wrote:
> > >>> > > > > > > >
> > >>> > > > > > > > Hi, David,
> > >>> > > > > > > >
> > >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> > >>> seem to
> > >>> > > > be
> > >>> > > > > > fewer
> > >>> > > > > > > > RPCs during rebalance and more efficient support of
> > >>> wildcard. A few
> > >>> > > > > > > > comments below.
> > >>> > > > > > > >
> > >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> > >>> rolling
> > >>> > > > > > changing
> > >>> > > > > > > > of the partition assignor in the consumers?
> > >>> > > > > > > > 30.2 For each field, could you explain whether it's
> > >>> required in
> > >>> > > > every
> > >>> > > > > > > > request or the scenarios when it needs to be filled? For
> > >>> example,
> > >>> > > > it's
> > >>> > > > > > not
> > >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > >>> > > > > > > >
> > >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> > >>> between the
> > >>> > > > > > client
> > >>> > > > > > > > and the broker is only considered during fetching, but
> > not
> > >>> during
> > >>> > > > > > assigning
> > >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> > is
> > >>> made,
> > >>> > > > there
> > >>> > > > > > is
> > >>> > > > > > > > no opportunity for read affinity because no replicas of
> > >>> assigned
> > >>> > > > > > partitions
> > >>> > > > > > > > are close to the member. I am wondering if we should use
> > >>> this
> > >>> > > > > > opportunity
> > >>> > > > > > > > to address this by including rack in GroupMember.
> > >>> > > > > > > >
> > >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> > >>> busy a group
> > >>> > > > > > > > coordinator is. By moving the event loop model, it seems
> > >>> that we
> > >>> > > > could
> > >>> > > > > > add
> > >>> > > > > > > > a metric that tracks the fraction of the time the event
> > >>> loop is
> > >>> > > > doing
> > >>> > > > > > the
> > >>> > > > > > > > actual work.
> > >>> > > > > > > >
> > >>> > > > > > > > 33. Could we add a section on coordinator failover
> > >>> handling? For
> > >>> > > > > > example,
> > >>> > > > > > > > does it need to trigger the check if any group with the
> > >>> wildcard
> > >>> > > > > > > > subscription now has a new matching topic?
> > >>> > > > > > > >
> > >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> > >>> > > > ConsumerGroupPartitionMetadataValue,
> > >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> > >>> the epoch
> > >>> > > > > > field
> > >>> > > > > > > > reflects? For example, does the epoch in
> > >>> ConsumerGroupMetadataValue
> > >>> > > > > > reflect
> > >>> > > > > > > > the latest group epoch? What about the one in
> > >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > >>> > > > > > ConsumerGroupMemberMetadataValue?
> > >>> > > > > > > >
> > >>> > > > > > > > 35. "the group coordinator will ensure that the following
> > >>> > > > invariants
> > >>> > > > > > are
> > >>> > > > > > > > met: ... All members exists." It's possible for a member
> > >>> not to
> > >>> > > > get any
> > >>> > > > > > > > assigned partitions, right?
> > >>> > > > > > > >
> > >>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
> > >>> to 0":
> > >>> > > > When
> > >>> > > > > > would
> > >>> > > > > > > > a consumer rejoin and what member id would be used?
> > >>> > > > > > > >
> > >>> > > > > > > > 37. "Instead, power users will have the ability to
> > trigger
> > >>> a
> > >>> > > > > > reassignment
> > >>> > > > > > > > by either providing a non-zero reason or by updating the
> > >>> assignor
> > >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> > >>> deprecation
> > >>> > > > of
> > >>> > > > > > > > Consumer#enforeRebalance.
> > >>> > > > > > > >
> > >>> > > > > > > > 38. The reassignment examples are nice. But the section
> > >>> seems to
> > >>> > > > have
> > >>> > > > > > > > multiple typos.
> > >>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> > >>> gets into
> > >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> > get
> > >>> into
> > >>> > > > > > epoch=3,
> > >>> > > > > > > > partitions=[foo-1] too early.
> > >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> > >>> epoch=2,
> > >>> > > > > > > > partitions=[foo-0].
> > >>> > > > > > > >
> > >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> > >>> upgrade from
> > >>> > > > any
> > >>> > > > > > old
> > >>> > > > > > > > version to new one?
> > >>> > > > > > > >
> > >>> > > > > > > > Thanks,
> > >>> > > > > > > >
> > >>> > > > > > > > Jun
> > >>> > > > > > > >
> > >>> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > >>> > > > > > <dj...@confluent.io.invalid>
> > >>> > > > > > > > wrote:
> > >>> > > > > > > >
> > >>> > > > > > > > > Hi all,
> > >>> > > > > > > > >
> > >>> > > > > > > > > The KIP states that we will re-implement the
> > coordinator
> > >>> in
> > >>> > > > Java. I
> > >>> > > > > > > > > discussed this offline with a few folks and folks are
> > >>> concerned
> > >>> > > > that
> > >>> > > > > > > > > we could introduce many regressions in the old protocol
> > >>> if we do
> > >>> > > > so.
> > >>> > > > > > > > > Therefore, I am going to remove this statement from the
> > >>> KIP. It
> > >>> > > > is an
> > >>> > > > > > > > > implementation detail after all so it does not have to
> > be
> > >>> > > > decided at
> > >>> > > > > > > > > this stage. We will likely start by trying to refactor
> > >>> the
> > >>> > > > current
> > >>> > > > > > > > > implementation as a first step.
> > >>> > > > > > > > >
> > >>> > > > > > > > > Cheers,
> > >>> > > > > > > > > David
> > >>> > > > > > > > >
> > >>> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> > >>> djacot@confluent.io
> > >>> > > > >
> > >>> > > > > > wrote:
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > Hi Luke,
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > >>> assigning,
> > >>> > > > > > reconciling,
> > >>> > > > > > > > > stable,
> > >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > >>> right?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > This sentence does not refer to those group states
> > but
> > >>> rather
> > >>> > > > to a
> > >>> > > > > > > > > > state machine replication (SMR). This refers to the
> > >>> entire
> > >>> > > > state of
> > >>> > > > > > > > > > group coordinator which is replicated via the log
> > >>> layer. I will
> > >>> > > > > > > > > > clarify this in the KIP.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > >>> modelled as an
> > >>> > > > event
> > >>> > > > > > > > > loop"?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > The idea is to follow a model similar to the new
> > quorum
> > >>> > > > > > controller. We
> > >>> > > > > > > > > > will have N threads to process events. Each
> > >>> __consumer_offsets
> > >>> > > > > > > > > > partition is assigned to a unique thread and all the
> > >>> events
> > >>> > > > (e.g.
> > >>> > > > > > > > > > requests, callbacks, etc.) are processed by this
> > >>> thread. This
> > >>> > > > > > simplify
> > >>> > > > > > > > > > concurrency and will enable us to do simulation
> > >>> testing for the
> > >>> > > > > > group
> > >>> > > > > > > > > > coordinator.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> > >>> *__consumer_offsets*
> > >>> > > > > > > > > partitions?
> > >>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> > >>> group
> > >>> > > > > > coordinator?
> > >>> > > > > > > > > For
> > >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > >>> groups, and
> > >>> > > > > > both
> > >>> > > > > > > > > exist in
> > >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > >>> for it,
> > >>> > > > or
> > >>> > > > > > 2?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > >>> number
> > >>> > > > of
> > >>> > > > > > threads
> > >>> > > > > > > > > used
> > >>> > > > > > > > > > > to run the state machines. But I'm wondering if the
> > >>> purpose
> > >>> > > > of
> > >>> > > > > > the
> > >>> > > > > > > > > threads
> > >>> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > >>> > > > > > > > > *__consumer_offsets*
> > >>> > > > > > > > > > > partitions?), and no heavy computation, why should
> > >>> we need
> > >>> > > > > > > > > multi-threads
> > >>> > > > > > > > > > > here?
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > >>> > > > processing as
> > >>> > > > > > the
> > >>> > > > > > > > > > computation could be heavy.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > >>> default
> > >>> > > > session
> > >>> > > > > > > > > timeout not
> > >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > >>> min/max
> > >>> > > > > > session
> > >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > >>> timeout to
> > >>> > > > detect
> > >>> > > > > > client
> > >>> > > > > > > > > > > failures when using the consumer group protocol.
> > >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > >>> minimum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > >>> maximum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > This is indeed a mistake. The default session timeout
> > >>> should
> > >>> > > > be 45s
> > >>> > > > > > > > > > (the current default).
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > >>> uniform],
> > >>> > > > > > which means
> > >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> > >>> why not
> > >>> > > > > > uniform
> > >>> > > > > > > > > one? I
> > >>> > > > > > > > > > > thought usually users will choose uniform assignor
> > >>> (former
> > >>> > > > sticky
> > >>> > > > > > > > > assinor)
> > >>> > > > > > > > > > > for better evenly distribution. Any other reason we
> > >>> choose
> > >>> > > > range
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > as default?
> > >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > >>> server side
> > >>> > > > > > assignors.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > The order on the server side has no influence because
> > >>> the
> > >>> > > > client
> > >>> > > > > > must
> > >>> > > > > > > > > > chose the selector that he wants to use. There is no
> > >>> default
> > >>> > > > in the
> > >>> > > > > > > > > > current proposal. If the assignor is not specified by
> > >>> the
> > >>> > > > client,
> > >>> > > > > > the
> > >>> > > > > > > > > > request is rejected. The default client value for
> > >>> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > Thanks for your very good comments, Luke. I hope that
> > >>> my
> > >>> > > > answers
> > >>> > > > > > help
> > >>> > > > > > > > > > to clarify things. I will update the KIP as well
> > based
> > >>> on your
> > >>> > > > > > > > > > feedback.
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > Cheers,
> > >>> > > > > > > > > > David
> > >>> > > > > > > > > >
> > >>> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> > >>> showuon@gmail.com>
> > >>> > > > > > wrote:
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Hi David,
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Thanks for the update.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Some more questions:
> > >>> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > >>> > > > > > > > > > > > The new group coordinator will have a state
> > >>> machine per
> > >>> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> > >>> machine is
> > >>> > > > > > modelled
> > >>> > > > > > > > > as an
> > >>> > > > > > > > > > > event loop. Those state machines will be executed
> > in
> > >>> > > > > > > > > > > *group.coordinator.threads* threads.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> > >>> assigning,
> > >>> > > > > > reconciling,
> > >>> > > > > > > > > stable,
> > >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> > >>> right?
> > >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> > >>> modelled as an
> > >>> > > > event
> > >>> > > > > > > > > loop"?
> > >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> > >>> *__consumer_offsets*
> > >>> > > > > > > > > partitions?
> > >>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> > >>> group
> > >>> > > > > > coordinator?
> > >>> > > > > > > > > For
> > >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> > >>> groups, and
> > >>> > > > > > both
> > >>> > > > > > > > > exist in
> > >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> > machine
> > >>> for it,
> > >>> > > > or
> > >>> > > > > > 2?
> > >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> > the
> > >>> number
> > >>> > > > of
> > >>> > > > > > threads
> > >>> > > > > > > > > used
> > >>> > > > > > > > > > > to run the state machines. But I'm wondering if the
> > >>> purpose
> > >>> > > > of
> > >>> > > > > > the
> > >>> > > > > > > > > threads
> > >>> > > > > > > > > > > is only to keep the state of each consumer group
> > (or
> > >>> > > > > > > > > *__consumer_offsets*
> > >>> > > > > > > > > > > partitions?), and no heavy computation, why should
> > >>> we need
> > >>> > > > > > > > > multi-threads
> > >>> > > > > > > > > > > here?
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > 2. For the default value in the new configs:
> > >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> > >>> default
> > >>> > > > session
> > >>> > > > > > > > > timeout not
> > >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> > the
> > >>> min/max
> > >>> > > > > > session
> > >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> > >>> timeout to
> > >>> > > > detect
> > >>> > > > > > client
> > >>> > > > > > > > > > > failures when using the consumer group protocol.
> > >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> > >>> minimum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> > >>> maximum
> > >>> > > > > > session
> > >>> > > > > > > > > timeout.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> > >>> uniform],
> > >>> > > > > > which means
> > >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> > >>> why not
> > >>> > > > > > uniform
> > >>> > > > > > > > > one? I
> > >>> > > > > > > > > > > thought usually users will choose uniform assignor
> > >>> (former
> > >>> > > > sticky
> > >>> > > > > > > > > assinor)
> > >>> > > > > > > > > > > for better evenly distribution. Any other reason we
> > >>> choose
> > >>> > > > range
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > as default?
> > >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> > >>> server side
> > >>> > > > > > assignors.
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > Thank you.
> > >>> > > > > > > > > > > Luke
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> > >>> showuon@gmail.com
> > >>> > > > >
> > >>> > > > > > wrote:
> > >>> > > > > > > > > > >
> > >>> > > > > > > > > > > > Hi Sagar,
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > I have some thoughts about Kafka Connect
> > >>> integrating with
> > >>> > > > > > KIP-848,
> > >>> > > > > > > > > but I
> > >>> > > > > > > > > > > > think we should have a separate discussion thread
> > >>> for the
> > >>> > > > Kafka
> > >>> > > > > > > > > Connect
> > >>> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> > >>> Rebalance
> > >>> > > > > > Protocol
> > >>> > > > > > > > > [1],
> > >>> > > > > > > > > > > > and let this discussion thread focus on consumer
> > >>> rebalance
> > >>> > > > > > protocol,
> > >>> > > > > > > > > WDYT?
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > [1]
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > >
> > >>>
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > Thank you.
> > >>> > > > > > > > > > > > Luke
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > >>> > > > > > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > wrote:
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
> > >>> like
> > >>> > > > there's
> > >>> > > > > > > > > agreement on
> > >>> > > > > > > > > > > >> using separate APIs for Connect. I would revisit
> > >>> the doc
> > >>> > > > and
> > >>> > > > > > see
> > >>> > > > > > > > > what
> > >>> > > > > > > > > > > >> changes are to be made.
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >> Thanks!
> > >>> > > > > > > > > > > >> Sagar.
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > >>> > > > > > > > > <dj...@confluent.io.invalid>
> > >>> > > > > > > > > > > >> wrote:
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >> > Hi Sagar,
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > Thanks for the feedback and the document.
> > >>> That's really
> > >>> > > > > > helpful. I
> > >>> > > > > > > > > > > >> > will take a look at it.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > Overall, it seems to me that both Connect and
> > >>> the
> > >>> > > > Consumer
> > >>> > > > > > could
> > >>> > > > > > > > > share
> > >>> > > > > > > > > > > >> > the same underlying "engine". The main
> > >>> difference is
> > >>> > > > that
> > >>> > > > > > the
> > >>> > > > > > > > > Consumer
> > >>> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> > >>> Connect
> > >>> > > > assigns
> > >>> > > > > > tasks
> > >>> > > > > > > > > to
> > >>> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > >>> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> > >>> different
> > >>> > > > > > resource
> > >>> > > > > > > > > types
> > >>> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > >>> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> > >>> dedicated
> > >>> > > > APIs
> > >>> > > > > > > > > would be
> > >>> > > > > > > > > > > >> > similar to the new ones but different on the
> > >>> > > > > > content/resources and
> > >>> > > > > > > > > > > >> > they would rely on the same engine on the
> > >>> coordinator
> > >>> > > > side.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > I personally lean towards 2) because I am not
> > a
> > >>> fan of
> > >>> > > > > > > > > overcharging
> > >>> > > > > > > > > > > >> > APIs to serve different purposes. That being
> > >>> said, I am
> > >>> > > > not
> > >>> > > > > > > > > opposed to
> > >>> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > I think that we can continue to discuss it
> > here
> > >>> for now
> > >>> > > > in
> > >>> > > > > > order
> > >>> > > > > > > > > to
> > >>> > > > > > > > > > > >> > ensure that this KIP is compatible with what
> > we
> > >>> will do
> > >>> > > > for
> > >>> > > > > > > > > Connect in
> > >>> > > > > > > > > > > >> > the future.
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > Best,
> > >>> > > > > > > > > > > >> > David
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > >>> > > > > > djacot@confluent.io>
> > >>> > > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > Hi all,
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > I am back from vacation. I will go through
> > >>> and address
> > >>> > > > > > your
> > >>> > > > > > > > > comments
> > >>> > > > > > > > > > > >> > > in the coming days. Thanks for your
> > feedback.
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > Cheers,
> > >>> > > > > > > > > > > >> > > David
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> > >>> Harris <
> > >>> > > > > > > > > gharris1727@gmail.com
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > Hey All!
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > >>> > > > cooperative
> > >>> > > > > > > > > rebalancing
> > >>> > > > > > > > > > > >> > making it
> > >>> > > > > > > > > > > >> > > > down the stack!
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > I had a few questions:
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> > >>> describes how
> > >>> > > > > > member
> > >>> > > > > > > > > epoch
> > >>> > > > > > > > > > > >> > should
> > >>> > > > > > > > > > > >> > > > advance in step with the group epoch and
> > >>> assignment
> > >>> > > > > > epoch
> > >>> > > > > > > > > values. I
> > >>> > > > > > > > > > > >> > think
> > >>> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> > >>> described
> > >>> > > > in
> > >>> > > > > > the
> > >>> > > > > > > > > KIP. When
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > protocol is incrementally assigning
> > >>> partitions to a
> > >>> > > > > > worker,
> > >>> > > > > > > > > what
> > >>> > > > > > > > > > > >> member
> > >>> > > > > > > > > > > >> > > > epoch does each incremental assignment
> > use?
> > >>> Are
> > >>> > > > member
> > >>> > > > > > epochs
> > >>> > > > > > > > > > > >> re-used,
> > >>> > > > > > > > > > > >> > and
> > >>> > > > > > > > > > > >> > > > a single member epoch can correspond to
> > >>> multiple
> > >>> > > > > > different
> > >>> > > > > > > > > > > >> > (monotonically
> > >>> > > > > > > > > > > >> > > > larger) assignments?
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque
> > >>> to the
> > >>> > > > group
> > >>> > > > > > > > > > > >> coordinator? If
> > >>> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > >>> > > > implementations
> > >>> > > > > > > > > interact
> > >>> > > > > > > > > > > >> with
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > Reason field, and how is its common
> > meaning
> > >>> agreed
> > >>> > > > > > upon? If
> > >>> > > > > > > > > so, what
> > >>> > > > > > > > > > > >> > is the
> > >>> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> > >>> including
> > >>> > > > such
> > >>> > > > > > > > > functionality
> > >>> > > > > > > > > > > >> > in the
> > >>> > > > > > > > > > > >> > > > opaque metadata?
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> > >>> "Thanks to
> > >>> > > > > > this, the
> > >>> > > > > > > > > input
> > >>> > > > > > > > > > > >> of
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > client side assignor is entirely driven by
> > >>> the group
> > >>> > > > > > > > > coordinator.
> > >>> > > > > > > > > > > >> The
> > >>> > > > > > > > > > > >> > > > consumer is no longer responsible for
> > >>> maintaining
> > >>> > > > any
> > >>> > > > > > state
> > >>> > > > > > > > > besides
> > >>> > > > > > > > > > > >> its
> > >>> > > > > > > > > > > >> > > > assigned partitions." Does this mean that
> > >>> the
> > >>> > > > > > client-side
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > >> MAY
> > >>> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> > >>> (such as
> > >>> > > > > > partition
> > >>> > > > > > > > > > > >> > throughput,
> > >>> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc),
> > or
> > >>> that
> > >>> > > > > > additional
> > >>> > > > > > > > > > > >> > non-Metadata
> > >>> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> > >>> > > > > > > > > > > >> > > > for
> > >>> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > >>> > > > > > > > > > > >> > > > and
> > >>> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > >>> > > > > > that
> > >>> > > > > > > > > seem to
> > >>> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> > >>> these two
> > >>> > > > > > > > > implementations
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > unified? This would serve to promote
> > >>> feature parity
> > >>> > > > of
> > >>> > > > > > > > > server-side
> > >>> > > > > > > > > > > >> and
> > >>> > > > > > > > > > > >> > > > client-side assignors, and would also
> > >>> facilitate
> > >>> > > > > > operational
> > >>> > > > > > > > > > > >> > flexibility in
> > >>> > > > > > > > > > > >> > > > certain situations. For example, if a
> > >>> server-side
> > >>> > > > > > assignor
> > >>> > > > > > > > > has some
> > >>> > > > > > > > > > > >> > poor
> > >>> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> > >>> patched
> > >>> > > > > > assignor to
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > client
> > >>> > > > > > > > > > > >> > > > and switching one consumer group to a
> > >>> client-side
> > >>> > > > > > assignor
> > >>> > > > > > > > > may be
> > >>> > > > > > > > > > > >> > faster
> > >>> > > > > > > > > > > >> > > > and less risky than patching all of the
> > >>> brokers.
> > >>> > > > With
> > >>> > > > > > the
> > >>> > > > > > > > > currently
> > >>> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > >>> > > > reimplementation
> > >>> > > > > > would
> > >>> > > > > > > > > have
> > >>> > > > > > > > > > > >> to be
> > >>> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> > diverged
> > >>> > > > > > significantly,
> > >>> > > > > > > > > then it
> > >>> > > > > > > > > > > >> is
> > >>> > > > > > > > > > > >> > > > possible that a reimplementation would not
> > >>> be
> > >>> > > > possible.
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > --
> > >>> > > > > > > > > > > >> > > > Greg Harris
> > >>> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > >>> > > > > > > > > > > >> > > > github.com/gharris1727
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > >>> > > > > > > > > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > I created a confluence page to discuss
> > >>> how Connect
> > >>> > > > > > would
> > >>> > > > > > > > > need to
> > >>> > > > > > > > > > > >> > change
> > >>> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> > >>> Here's the
> > >>> > > > page:
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > >
> > >>>
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> > tried
> > >>> to keep
> > >>> > > > a
> > >>> > > > > > format
> > >>> > > > > > > > > > > >> similar to
> > >>> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> > >>> Also, do you
> > >>> > > > > > think this
> > >>> > > > > > > > > > > >> should
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > > moved to a separate discussion thread or
> > >>> is this
> > >>> > > > one
> > >>> > > > > > fine?
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > Thanks!
> > >>> > > > > > > > > > > >> > > > > Sagar.
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > >>> > > > > > > > > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> > >>> Streams.
> > >>> > > > > > Sure, I
> > >>> > > > > > > > > would do
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > > analysis and come up with such a
> > >>> document.
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > > Thanks!
> > >>> > > > > > > > > > > >> > > > > > Sagar.
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> > >>> Guozhang Wang <
> > >>> > > > > > > > > > > >> wangguoz@gmail.com>
> > >>> > > > > > > > > > > >> > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> It would be great if you could come
> > >>> back with
> > >>> > > > some
> > >>> > > > > > > > > analysis on
> > >>> > > > > > > > > > > >> > how to
> > >>> > > > > > > > > > > >> > > > > >> implement the Connect side
> > integration
> > >>> with
> > >>> > > > the new
> > >>> > > > > > > > > protocol;
> > >>> > > > > > > > > > > >> so
> > >>> > > > > > > > > > > >> > far
> > >>> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> > >>> "protocol type"
> > >>> > > > we
> > >>> > > > > > did not
> > >>> > > > > > > > > yet
> > >>> > > > > > > > > > > >> think
> > >>> > > > > > > > > > > >> > > > > through
> > >>> > > > > > > > > > > >> > > > > >> the Connect side implementations. For
> > >>> Streams
> > >>> > > > > > here's a
> > >>> > > > > > > > > draft of
> > >>> > > > > > > > > > > >> > > > > >> integration
> > >>> > > > > > > > > > > >> > > > > >> plan:
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > >
> > >>>
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > >>> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> > Connect.
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> > Sagar
> > >>> <
> > >>> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > >>> > > > > > > > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> > Hi David,
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> > >>> reason I
> > >>> > > > thought
> > >>> > > > > > > > > connect can
> > >>> > > > > > > > > > > >> > also fit
> > >>> > > > > > > > > > > >> > > > > >> into
> > >>> > > > > > > > > > > >> > > > > >> > this new scheme is that even today
> > >>> the
> > >>> > > > connect
> > >>> > > > > > uses a
> > >>> > > > > > > > > > > >> > > > > WorkerCoordinator
> > >>> > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator
> > >>> to empower
> > >>> > > > > > > > > rebalances of
> > >>> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> > >>> WorkerCoordinator sets
> > >>> > > > the
> > >>> > > > > > > > > > > >> protocolType()
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> connect
> > >>> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> > >>> plumbing
> > >>> > > > into
> > >>> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > I think the changes to support
> > >>> connect would
> > >>> > > > be
> > >>> > > > > > > > > similar at a
> > >>> > > > > > > > > > > >> > high
> > >>> > > > > > > > > > > >> > > > > level
> > >>> > > > > > > > > > > >> > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> > >>> because of the
> > >>> > > > > > Client
> > >>> > > > > > > > > side
> > >>> > > > > > > > > > > >> > assignors
> > >>> > > > > > > > > > > >> > > > > being
> > >>> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
> > >>> level, we
> > >>> > > > > > might
> > >>> > > > > > > > > need to
> > >>> > > > > > > > > > > >> make
> > >>> > > > > > > > > > > >> > a lot
> > >>> > > > > > > > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> > >>> assignment
> > >>> > > > protocol
> > >>> > > > > > like
> > >>> > > > > > > > > > > >> enhancing
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> > >>> SyncGroup and
> > >>> > > > > > using
> > >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > >>> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> > >>> streams (or
> > >>> > > > > > there
> > >>> > > > > > > > > might be
> > >>> > > > > > > > > > > >> > > > > >> deviations). I
> > >>> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> > >>> analysis of
> > >>> > > > the
> > >>> > > > > > same
> > >>> > > > > > > > > and we
> > >>> > > > > > > > > > > >> > can have
> > >>> > > > > > > > > > > >> > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > separate discussion thread for that
> > >>> as that
> > >>> > > > would
> > >>> > > > > > > > > derail this
> > >>> > > > > > > > > > > >> > > > > discussion
> > >>> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
> > >>> good to
> > >>> > > > you.
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > Thanks!
> > >>> > > > > > > > > > > >> > > > > >> > Sagar.
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> > >>> David Jacot
> > >>> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > wrote:
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> > >>> `Assignment#error`.
> > >>> > > > > > Sure, I
> > >>> > > > > > > > > can
> > >>> > > > > > > > > > > >> > mention it.
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> > >>> from his
> > >>> > > > current
> > >>> > > > > > > > > assignment
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > his
> > >>> > > > > > > > > > > >> > > > > >> > > target assignment when he can
> > move
> > >>> to
> > >>> > > > epoch 3.
> > >>> > > > > > When
> > >>> > > > > > > > > that
> > >>> > > > > > > > > > > >> > happens,
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
> > >>> persisted
> > >>> > > > > > with all
> > >>> > > > > > > > > its
> > >>> > > > > > > > > > > >> > assigned
> > >>> > > > > > > > > > > >> > > > > >> > > partitions even if they are not
> > >>> all revoked
> > >>> > > > > > yet. In
> > >>> > > > > > > > > other
> > >>> > > > > > > > > > > >> > words, the
> > >>> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> > >>> target
> > >>> > > > > > assignment.
> > >>> > > > > > > > > This is
> > >>> > > > > > > > > > > >> > basically
> > >>> > > > > > > > > > > >> > > > > an
> > >>> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> > >>> write all
> > >>> > > > the
> > >>> > > > > > > > > changes to
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > log.
> > >>> > > > > > > > > > > >> > > > > The
> > >>> > > > > > > > > > > >> > > > > >> > > examples are based on the
> > >>> persisted state
> > >>> > > > so I
> > >>> > > > > > > > > understand
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> > >>> improve
> > >>> > > > this in
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> > description.
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> > >>> reuse the
> > >>> > > > > > protocol
> > >>> > > > > > > > > with a
> > >>> > > > > > > > > > > >> > client side
> > >>> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> > >>> protocol. The
> > >>> > > > > > assignment
> > >>> > > > > > > > > is
> > >>> > > > > > > > > > > >> about
> > >>> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> > >>> could
> > >>> > > > Connect
> > >>> > > > > > fit
> > >>> > > > > > > > > into this?
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > David
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> > >>> Sagar <
> > >>> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > >>> > > > > > > > > > > >> > > > > >> wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
> > >>> minor
> > >>> > > > > > observations:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> > >>> section in
> > >>> > > > Client
> > >>> > > > > > Side
> > >>> > > > > > > > > mode
> > >>> > > > > > > > > > > >> > Assignment
> > >>> > > > > > > > > > > >> > > > > >> > > process,
> > >>> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case,
> > >>> the
> > >>> > > > client
> > >>> > > > > > side
> > >>> > > > > > > > > assignor
> > >>> > > > > > > > > > > >> can
> > >>> > > > > > > > > > > >> > > > > return
> > >>> > > > > > > > > > > >> > > > > >> an
> > >>> > > > > > > > > > > >> > > > > >> > > > error to the group
> > coordinator`.
> > >>> In this
> > >>> > > > > > case are
> > >>> > > > > > > > > you
> > >>> > > > > > > > > > > >> > referring to
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> > >>> AssignmentError
> > >>> > > > that's
> > >>> > > > > > > > > listed down
> > >>> > > > > > > > > > > >> > towards
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > end?
> > >>> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
> > >>> make sense
> > >>> > > > to
> > >>> > > > > > > > > mention this
> > >>> > > > > > > > > > > >> > > > > explicitly
> > >>> > > > > > > > > > > >> > > > > >> > here?
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section,
> > >>> I have a
> > >>> > > > > > slight
> > >>> > > > > > > > > > > >> confusion,
> > >>> > > > > > > > > > > >> > not
> > >>> > > > > > > > > > > >> > > > > sure
> > >>> > > > > > > > > > > >> > > > > >> if
> > >>> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
> > >>> this step:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> > >>> coordinator
> > >>> > > > > > > > > transitions him
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > epoch
> > >>> > > > > > > > > > > >> > > > > 3
> > >>> > > > > > > > > > > >> > > > > >> > > because
> > >>> > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke.
> > It
> > >>> > > > persists
> > >>> > > > > > the
> > >>> > > > > > > > > change and
> > >>> > > > > > > > > > > >> > reply.
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > >>> > > > > > > > > > > >> > > > > >> > > >       - A
> > >>> > > > > > > > > > > >> > > > > >> > > >       - B
> > >>> > > > > > > > > > > >> > > > > >> > > >       - C
> > >>> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> > (epoch=3)
> > >>> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > >>> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > >>> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> > >>> partitions=[foo-0,
> > >>> > > > > > foo-1]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> > >>> partitions=[foo-2]
> > >>> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> > >>> partitions=[foo-1]
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> > >>> transitions to
> > >>> > > > epoch 3
> > >>> > > > > > but
> > >>> > > > > > > > > cannot
> > >>> > > > > > > > > > > >> get
> > >>> > > > > > > > > > > >> > foo-1
> > >>> > > > > > > > > > > >> > > > > >> yet.
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member
> > >>> C can't
> > >>> > > > get
> > >>> > > > > > the
> > >>> > > > > > > > > foo-1
> > >>> > > > > > > > > > > >> > partition
> > >>> > > > > > > > > > > >> > > > > yet,
> > >>> > > > > > > > > > > >> > > > > >> > but
> > >>> > > > > > > > > > > >> > > > > >> > > > based on the description above,
> > >>> it seems
> > >>> > > > it
> > >>> > > > > > > > > already has
> > >>> > > > > > > > > > > >> it.
> > >>> > > > > > > > > > > >> > Do you
> > >>> > > > > > > > > > > >> > > > > >> > think
> > >>> > > > > > > > > > > >> > > > > >> > > it
> > >>> > > > > > > > > > > >> > > > > >> > > > would be better to remove it
> > and
> > >>> > > > populate it
> > >>> > > > > > only
> > >>> > > > > > > > > when it
> > >>> > > > > > > > > > > >> > actually
> > >>> > > > > > > > > > > >> > > > > >> gets
> > >>> > > > > > > > > > > >> > > > > >> > > it?
> > >>> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> > >>> places, so
> > >>> > > > have
> > >>> > > > > > I
> > >>> > > > > > > > > > > >> understood it
> > >>> > > > > > > > > > > >> > > > > >> > incorrectly
> > >>> > > > > > > > > > > >> > > > > >> > > ?
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
> > >>> out of
> > >>> > > > scope
> > >>> > > > > > of
> > >>> > > > > > > > > this
> > >>> > > > > > > > > > > >> > discussion,
> > >>> > > > > > > > > > > >> > > > > but
> > >>> > > > > > > > > > > >> > > > > >> > from
> > >>> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> > >>> probably be
> > >>> > > > > > running in
> > >>> > > > > > > > > client
> > >>> > > > > > > > > > > >> > side
> > >>> > > > > > > > > > > >> > > > > >> assignor
> > >>> > > > > > > > > > > >> > > > > >> > > mode
> > >>> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> > >>> protocol as it
> > >>> > > > has
> > >>> > > > > > its
> > >>> > > > > > > > > own
> > >>> > > > > > > > > > > >> Custom
> > >>> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > >>> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
> > >>> David
> > >>> > > > Jacot
> > >>> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
> > >>> move
> > >>> > > > forward
> > >>> > > > > > with
> > >>> > > > > > > > > > > >> > specialized API
> > >>> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> > >>> generic API.
> > >>> > > > For
> > >>> > > > > > > > > Connect, we
> > >>> > > > > > > > > > > >> > can apply
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> > >>> reuse/share the
> > >>> > > > core
> > >>> > > > > > > > > > > >> > implementation on
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> > >>> registry, I
> > >>> > > > > > think
> > >>> > > > > > > > > that we
> > >>> > > > > > > > > > > >> > should
> > >>> > > > > > > > > > > >> > > > > >> consider
> > >>> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> > >>> simple
> > >>> > > > > > > > > membership/leader
> > >>> > > > > > > > > > > >> > election.
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22
> > >>> AM Ismael
> > >>> > > > > > Juma <
> > >>> > > > > > > > > > > >> > ismael@juma.me.uk
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> > >>> java.util.regex.Pattern
> > >>> > > > vs
> > >>> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > document the differences in
> > >>> more
> > >>> > > > detail
> > >>> > > > > > before
> > >>> > > > > > > > > > > >> deciding
> > >>> > > > > > > > > > > >> > one
> > >>> > > > > > > > > > > >> > > > > way
> > >>> > > > > > > > > > > >> > > > > >> or
> > >>> > > > > > > > > > > >> > > > > >> > > > > another.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > >>> > > > > > > > > java.util.regex.Pattern,
> > >>> > > > > > > > > > > >> they
> > >>> > > > > > > > > > > >> > expect
> > >>> > > > > > > > > > > >> > > > > >> > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If
> > >>> we are
> > >>> > > > doing
> > >>> > > > > > > > > something
> > >>> > > > > > > > > > > >> > different,
> > >>> > > > > > > > > > > >> > > > > >> then
> > >>> > > > > > > > > > > >> > > > > >> > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> > >>> overload
> > >>> > > > with
> > >>> > > > > > our own
> > >>> > > > > > > > > > > >> Pattern
> > >>> > > > > > > > > > > >> > class
> > >>> > > > > > > > > > > >> > > > > (I
> > >>> > > > > > > > > > > >> > > > > >> > > don't
> > >>> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> > >>> re2j's at
> > >>> > > > this
> > >>> > > > > > > > > point).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> > >>> major new
> > >>> > > > > > protocol
> > >>> > > > > > > > > should
> > >>> > > > > > > > > > > >> > > > > integrate
> > >>> > > > > > > > > > > >> > > > > >> > fully
> > >>> > > > > > > > > > > >> > > > > >> > > > > with
> > >>> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> > >>> topic
> > >>> > > > > > recreation case
> > >>> > > > > > > > > > > >> > correctly.
> > >>> > > > > > > > > > > >> > > > > That's
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > main
> > >>> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> > >>> agree with
> > >>> > > > > > David
> > >>> > > > > > > > > that we'd
> > >>> > > > > > > > > > > >> > want to
> > >>> > > > > > > > > > > >> > > > > add
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > ids
> > >>> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
> > >>> that don't
> > >>> > > > > > have it
> > >>> > > > > > > > > yet and
> > >>> > > > > > > > > > > >> > that we
> > >>> > > > > > > > > > > >> > > > > can
> > >>> > > > > > > > > > > >> > > > > >> > > probably
> > >>> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> > >>> versus adding
> > >>> > > > new
> > >>> > > > > > APIs
> > >>> > > > > > > > > to the
> > >>> > > > > > > > > > > >> > Java
> > >>> > > > > > > > > > > >> > > > > >> Consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > (unless
> > >>> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new
> > APIs
> > >>> is
> > >>> > > > required
> > >>> > > > > > for
> > >>> > > > > > > > > > > >> reasonable
> > >>> > > > > > > > > > > >> > > > > >> semantics).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> > about
> > >>> the
> > >>> > > > > > coordinator
> > >>> > > > > > > > > storing
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> configs.
> > >>> > > > > > > > > > > >> > > > > >> > > It's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > >>> > > > centralized
> > >>> > > > > > in the
> > >>> > > > > > > > > > > >> metadata
> > >>> > > > > > > > > > > >> > log for
> > >>> > > > > > > > > > > >> > > > > >> > > various
> > >>> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> > >>> visibility,
> > >>> > > > > > consistency,
> > >>> > > > > > > > > > > >> etc.).
> > >>> > > > > > > > > > > >> > > > > >> Similarly, I
> > >>> > > > > > > > > > > >> > > > > >> > > am
> > >>> > > > > > > > > > > >> > > > > >> > > > > not
> > >>> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> > >>> deleting
> > >>> > > > > > configs in a
> > >>> > > > > > > > > way
> > >>> > > > > > > > > > > >> that
> > >>> > > > > > > > > > > >> > they
> > >>> > > > > > > > > > > >> > > > > >> cannot
> > >>> > > > > > > > > > > >> > > > > >> > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property
> > >>> for modern
> > >>> > > > > > systems
> > >>> > > > > > > > > is to
> > >>> > > > > > > > > > > >> > minimize
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > number
> > >>> > > > > > > > > > > >> > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> > >>> scenarios.
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at
> > 3:47
> > >>> PM David
> > >>> > > > > > Jacot
> > >>> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> > >>> answers are
> > >>> > > > below:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> > >>> especially
> > >>> > > > > > the last
> > >>> > > > > > > > > step
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > > > > clients
> > >>> > > > > > > > > > > >> > > > > >> > > flip the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > >>> protocol, in
> > >>> > > > which
> > >>> > > > > > we
> > >>> > > > > > > > > would
> > >>> > > > > > > > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > window
> > >>> > > > > > > > > > > >> > > > > >> > where
> > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > old
> > >>> > > > protocols /
> > >>> > > > > > rpcs
> > >>> > > > > > > > > are
> > >>> > > > > > > > > > > >> used
> > >>> > > > > > > > > > > >> > by
> > >>> > > > > > > > > > > >> > > > > >> members
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > >>> coordinator could
> > >>> > > > > > "mimic"
> > >>> > > > > > > > > the old
> > >>> > > > > > > > > > > >> > behavior
> > >>> > > > > > > > > > > >> > > > > >> while
> > >>> > > > > > > > > > > >> > > > > >> > > > > using the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > >>> something we
> > >>> > > > need to
> > >>> > > > > > > > > present
> > >>> > > > > > > > > > > >> about.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a
> > >>> new
> > >>> > > > version
> > >>> > > > > > of KIP
> > >>> > > > > > > > > which
> > >>> > > > > > > > > > > >> > includes
> > >>> > > > > > > > > > > >> > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See
> > the
> > >>> > > > "Supporting
> > >>> > > > > > > > > Online
> > >>> > > > > > > > > > > >> > Consumer
> > >>> > > > > > > > > > > >> > > > > Group
> > >>> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > >>> > > > Deprecation,
> > >>> > > > > > and
> > >>> > > > > > > > > Migration
> > >>> > > > > > > > > > > >> > Plan". I
> > >>> > > > > > > > > > > >> > > > > >> think
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
> > >>> few cases
> > >>> > > > > > now but
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > overall idea
> > >>> > > > > > > > > > > >> > > > > >> and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> > >>> understandable.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > >>> ids. So
> > >>> > > > far as
> > >>> > > > > > > > > KIP-516 the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > >> are
> > >>> > > > > > > > > > > >> > > > > >> > > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > used
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> > >>> client,
> > >>> > > > but
> > >>> > > > > > they
> > >>> > > > > > > > > are not
> > >>> > > > > > > > > > > >> > exposed
> > >>> > > > > > > > > > > >> > > > > via
> > >>> > > > > > > > > > > >> > > > > >> any
> > >>> > > > > > > > > > > >> > > > > >> > > > > public
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > >>> think the
> > >>> > > > > > question is,
> > >>> > > > > > > > > first
> > >>> > > > > > > > > > > >> > should we
> > >>> > > > > > > > > > > >> > > > > >> let
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > maintaining
> > >>> the
> > >>> > > > names
> > >>> > > > > > -> ids
> > >>> > > > > > > > > mapping
> > >>> > > > > > > > > > > >> > itself
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > fully
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > >>> existing
> > >>> > > > RPCs
> > >>> > > > > > and the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> RPCs
> > >>> > > > > > > > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > ever
> > >>> consider
> > >>> > > > > > exposing
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > (both
> > >>> > > > > > > > > subscribe/assign, as
> > >>> > > > > > > > > > > >> well
> > >>> > > > > > > > > > > >> > as in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> > >>> topic
> > >>> > > > > > > > > > > >> > deletion-and-recreation).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> > >>> include
> > >>> > > > > > converting
> > >>> > > > > > > > > all
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > offsets
> > >>> > > > > > > > > > > >> > > > > >> > > related
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids
> > in
> > >>> this
> > >>> > > > KIP,
> > >>> > > > > > the
> > >>> > > > > > > > > consumer
> > >>> > > > > > > > > > > >> > would be
> > >>> > > > > > > > > > > >> > > > > >> able
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
> > >>> ids. That
> > >>> > > > > > being
> > >>> > > > > > > > > said, it
> > >>> > > > > > > > > > > >> > still has
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > provide
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> > >>> various APIs so
> > >>> > > > > > having a
> > >>> > > > > > > > > > > >> mapping
> > >>> > > > > > > > > > > >> > in the
> > >>> > > > > > > > > > > >> > > > > >> > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> > >>> opinion on
> > >>> > > > > > this.
> > >>> > > > > > > > > Here I
> > >>> > > > > > > > > > > >> > wonder if
> > >>> > > > > > > > > > > >> > > > > >> this
> > >>> > > > > > > > > > > >> > > > > >> > > goes
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
> > >>> KIP. I
> > >>> > > > would
> > >>> > > > > > rather
> > >>> > > > > > > > > focus
> > >>> > > > > > > > > > > >> on
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > internals
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider
> > >>> this
> > >>> > > > > > separately if
> > >>> > > > > > > > > we see
> > >>> > > > > > > > > > > >> > value in
> > >>> > > > > > > > > > > >> > > > > >> doing
> > >>> > > > > > > > > > > >> > > > > >> > > it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> > >>> point about
> > >>> > > > > > using
> > >>> > > > > > > > > topic ids
> > >>> > > > > > > > > > > >> > in the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> ConsumerGroupHeartbeatRequest, I
> > >>> > > > think
> > >>> > > > > > that
> > >>> > > > > > > > > there
> > >>> > > > > > > > > > > >> is
> > >>> > > > > > > > > > > >> > one
> > >>> > > > > > > > > > > >> > > > > >> > advantage
> > >>> > > > > > > > > > > >> > > > > >> > > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> > consumer
> > >>> will
> > >>> > > > have
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> opportunity
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> validate
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> > >>> passing
> > >>> > > > them
> > >>> > > > > > into
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> group
> > >>> > > > > > > > > > > >> > > > > rebalance
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > >>> > > > coordinator
> > >>> > > > > > will
> > >>> > > > > > > > > also
> > >>> > > > > > > > > > > >> notice
> > >>> > > > > > > > > > > >> > it but
> > >>> > > > > > > > > > > >> > > > > >> it
> > >>> > > > > > > > > > > >> > > > > >> > > does
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> > >>> reject an
> > >>> > > > > > invalid
> > >>> > > > > > > > > topic in
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> response.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> > >>> on all
> > >>> > > > other
> > >>> > > > > > minor
> > >>> > > > > > > > > > > >> questions
> > >>> > > > > > > > > > > >> > > > > except
> > >>> > > > > > > > > > > >> > > > > >> for
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > >>> question:
> > >>> > > > > > personally
> > >>> > > > > > > > > I think
> > >>> > > > > > > > > > > >> > it's not
> > >>> > > > > > > > > > > >> > > > > >> > > necessary
> > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> > >>> API with
> > >>> > > > > > Pattern,
> > >>> > > > > > > > > but
> > >>> > > > > > > > > > > >> > instead we
> > >>> > > > > > > > > > > >> > > > > >> still
> > >>> > > > > > > > > > > >> > > > > >> > > use
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> > >>> that our
> > >>> > > > > > > > > subscription may
> > >>> > > > > > > > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> rejected by
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> > >>> case is a
> > >>> > > > > > very rare
> > >>> > > > > > > > > > > >> scenario
> > >>> > > > > > > > > > > >> > I felt
> > >>> > > > > > > > > > > >> > > > > >> > using
> > >>> > > > > > > > > > > >> > > > > >> > > an
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > based
> > >>> > > > > > subscription may
> > >>> > > > > > > > > be
> > >>> > > > > > > > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> vulnerable
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > various
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have
> > to
> > >>> look at
> > >>> > > > the
> > >>> > > > > > > > > differences
> > >>> > > > > > > > > > > >> > between
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > two
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> > >>> understand the
> > >>> > > > > > potential
> > >>> > > > > > > > > issues.
> > >>> > > > > > > > > > > >> My
> > >>> > > > > > > > > > > >> > > > > >> > > understanding is
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all
> > >>> the basic
> > >>> > > > > > regular
> > >>> > > > > > > > > > > >> > expressions. The
> > >>> > > > > > > > > > > >> > > > > >> > > differences
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> > mainly
> > >>> about
> > >>> > > > the
> > >>> > > > > > various
> > >>> > > > > > > > > > > >> character
> > >>> > > > > > > > > > > >> > > > > >> classes. I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
> > >>> think
> > >>> > > > about
> > >>> > > > > > this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> > >>> 11:28 PM
> > >>> > > > > > Guozhang
> > >>> > > > > > > > > Wang <
> > >>> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > >>> > > > > > > > > > > >> > > > > >> > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think
> > on
> > >>> the high
> > >>> > > > > > level
> > >>> > > > > > > > > there are
> > >>> > > > > > > > > > > >> > two meta
> > >>> > > > > > > > > > > >> > > > > >> > > points we
> > >>> > > > > > > > > > > >> > > > > >> > > > > need
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit
> > more:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> > >>> especially
> > >>> > > > > > the last
> > >>> > > > > > > > > step
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > > > > clients
> > >>> > > > > > > > > > > >> > > > > >> > > flip the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> > >>> protocol, in
> > >>> > > > which
> > >>> > > > > > we
> > >>> > > > > > > > > would
> > >>> > > > > > > > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > window
> > >>> > > > > > > > > > > >> > > > > >> > where
> > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> > old
> > >>> > > > protocols /
> > >>> > > > > > rpcs
> > >>> > > > > > > > > are
> > >>> > > > > > > > > > > >> used
> > >>> > > > > > > > > > > >> > by
> > >>> > > > > > > > > > > >> > > > > >> members
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> > >>> coordinator could
> > >>> > > > > > "mimic"
> > >>> > > > > > > > > the old
> > >>> > > > > > > > > > > >> > behavior
> > >>> > > > > > > > > > > >> > > > > >> while
> > >>> > > > > > > > > > > >> > > > > >> > > > > using the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> > >>> something we
> > >>> > > > need to
> > >>> > > > > > > > > present
> > >>> > > > > > > > > > > >> about.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> > >>> ids. So
> > >>> > > > far as
> > >>> > > > > > > > > KIP-516 the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > >> are
> > >>> > > > > > > > > > > >> > > > > >> > > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > used
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> > >>> client,
> > >>> > > > but
> > >>> > > > > > they
> > >>> > > > > > > > > are not
> > >>> > > > > > > > > > > >> > exposed
> > >>> > > > > > > > > > > >> > > > > via
> > >>> > > > > > > > > > > >> > > > > >> any
> > >>> > > > > > > > > > > >> > > > > >> > > > > public
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> > >>> think the
> > >>> > > > > > question is,
> > >>> > > > > > > > > first
> > >>> > > > > > > > > > > >> > should we
> > >>> > > > > > > > > > > >> > > > > >> let
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> > maintaining
> > >>> the
> > >>> > > > names
> > >>> > > > > > -> ids
> > >>> > > > > > > > > mapping
> > >>> > > > > > > > > > > >> > itself
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > fully
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> > >>> existing
> > >>> > > > RPCs
> > >>> > > > > > and the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> RPCs
> > >>> > > > > > > > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> > ever
> > >>> consider
> > >>> > > > > > exposing
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > topic ids
> > >>> > > > > > > > > > > >> > > > > in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> > (both
> > >>> > > > > > > > > subscribe/assign, as
> > >>> > > > > > > > > > > >> well
> > >>> > > > > > > > > > > >> > as in
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> > >>> topic
> > >>> > > > > > > > > > > >> > deletion-and-recreation).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> > >>> on all
> > >>> > > > other
> > >>> > > > > > minor
> > >>> > > > > > > > > > > >> questions
> > >>> > > > > > > > > > > >> > > > > except
> > >>> > > > > > > > > > > >> > > > > >> for
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> > >>> question:
> > >>> > > > > > personally
> > >>> > > > > > > > > I think
> > >>> > > > > > > > > > > >> > it's not
> > >>> > > > > > > > > > > >> > > > > >> > > necessary
> > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> > >>> API with
> > >>> > > > > > Pattern,
> > >>> > > > > > > > > but
> > >>> > > > > > > > > > > >> > instead we
> > >>> > > > > > > > > > > >> > > > > >> still
> > >>> > > > > > > > > > > >> > > > > >> > > use
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> > >>> that our
> > >>> > > > > > > > > subscription may
> > >>> > > > > > > > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> rejected by
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> > >>> case is a
> > >>> > > > > > very rare
> > >>> > > > > > > > > > > >> scenario
> > >>> > > > > > > > > > > >> > I felt
> > >>> > > > > > > > > > > >> > > > > >> > using
> > >>> > > > > > > > > > > >> > > > > >> > > an
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> > based
> > >>> > > > > > subscription may
> > >>> > > > > > > > > be
> > >>> > > > > > > > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> vulnerable
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > various
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
> > >>> 5:23 AM
> > >>> > > > > > David Jacot
> > >>> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> > >>> feedback. Let
> > >>> > > > me
> > >>> > > > > > answer
> > >>> > > > > > > > > your
> > >>> > > > > > > > > > > >> > questions
> > >>> > > > > > > > > > > >> > > > > >> > inline.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> > >>> premature to
> > >>> > > > > > talk about
> > >>> > > > > > > > > > > >> target
> > >>> > > > > > > > > > > >> > > > > versions
> > >>> > > > > > > > > > > >> > > > > >> for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> > >>> existing group
> > >>> > > > > > protocol.
> > >>> > > > > > > > > > > >> Unlike
> > >>> > > > > > > > > > > >> > KRaft,
> > >>> > > > > > > > > > > >> > > > > >> this
> > >>> > > > > > > > > > > >> > > > > >> > > > > affects a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > core
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
> > >>> hence
> > >>> > > > > > > > > deprecation/removal
> > >>> > > > > > > > > > > >> > will be
> > >>> > > > > > > > > > > >> > > > > >> > heavily
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> > >>> applications
> > >>> > > > migrate
> > >>> > > > > > to
> > >>> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I
> > >>> will
> > >>> > > > remove it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> > >>> intend to
> > >>> > > > > > release
> > >>> > > > > > > > > this in
> > >>> > > > > > > > > > > >> > 4.x, but
> > >>> > > > > > > > > > > >> > > > > it
> > >>> > > > > > > > > > > >> > > > > >> > > wasn't
> > >>> > > > > > > > > > > >> > > > > >> > > > > made
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> > >>> that as a
> > >>> > > > way to
> > >>> > > > > > > > > estimate
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > we'd
> > >>> > > > > > > > > > > >> > > > > >> > > deprecate
> > >>> > > > > > > > > > > >> > > > > >> > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol,
> > >>> I also
> > >>> > > > > > suggest
> > >>> > > > > > > > > removing
> > >>> > > > > > > > > > > >> > this part.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> > >>> reasoning. As
> > >>> > > > > > > > > explained, I
> > >>> > > > > > > > > > > >> plan
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> rewrite
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> > >>> while we
> > >>> > > > > > implement
> > >>> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > protocol.
> > >>> > > > > > > > > > > >> > > > > >> This
> > >>> > > > > > > > > > > >> > > > > >> > > means
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> > >>> will be
> > >>> > > > slightly
> > >>> > > > > > > > > different
> > >>> > > > > > > > > > > >> > (e.g.
> > >>> > > > > > > > > > > >> > > > > >> threading
> > >>> > > > > > > > > > > >> > > > > >> > > > > model).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted
> > to
> > >>> tighten
> > >>> > > > the
> > >>> > > > > > > > > switch from
> > >>> > > > > > > > > > > >> > the old
> > >>> > > > > > > > > > > >> > > > > >> group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the
> > new
> > >>> group
> > >>> > > > > > coordinator
> > >>> > > > > > > > > to a
> > >>> > > > > > > > > > > >> > major
> > >>> > > > > > > > > > > >> > > > > >> release.
> > >>> > > > > > > > > > > >> > > > > >> > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be
> > >>> to use a
> > >>> > > > flag
> > >>> > > > > > to do
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > switch
> > >>> > > > > > > > > > > >> > > > > >> instead
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > relying
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> > >>> upgrade.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
> > >>> out the
> > >>> > > > > > details of
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > migration
> > >>> > > > > > > > > > > >> > > > > >> story.
> > >>> > > > > > > > > > > >> > > > > >> > > It
> > >>> > > > > > > > > > > >> > > > > >> > > > > sounds
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we
> > will
> > >>> support
> > >>> > > > > > online
> > >>> > > > > > > > > > > >> migrations.
> > >>> > > > > > > > > > > >> > Is
> > >>> > > > > > > > > > > >> > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > correct?
> > >>> > > > > > > > > > > >> > > > > >> > > > > We
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> > >>> detail. It
> > >>> > > > could
> > >>> > > > > > also
> > >>> > > > > > > > > be done
> > >>> > > > > > > > > > > >> > as a
> > >>> > > > > > > > > > > >> > > > > >> separate
> > >>> > > > > > > > > > > >> > > > > >> > > KIP,
> > >>> > > > > > > > > > > >> > > > > >> > > > > if
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
> > >>> online
> > >>> > > > > > migrations
> > >>> > > > > > > > > for the
> > >>> > > > > > > > > > > >> > group.
> > >>> > > > > > > > > > > >> > > > > That
> > >>> > > > > > > > > > > >> > > > > >> > > means
> > >>> > > > > > > > > > > >> > > > > >> > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> > >>> protocol
> > >>> > > > > > will be
> > >>> > > > > > > > > able to
> > >>> > > > > > > > > > > >> > switch to
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> > explain
> > >>> how that
> > >>> > > > > > will work
> > >>> > > > > > > > > > > >> though.
> > >>> > > > > > > > > > > >> > It is
> > >>> > > > > > > > > > > >> > > > > >> > > basically a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must
> > be
> > >>> > > > upgraded or
> > >>> > > > > > > > > rolled to a
> > >>> > > > > > > > > > > >> > software
> > >>> > > > > > > > > > > >> > > > > >> > > supporting
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> > coordinator.
> > >>> Both the
> > >>> > > > > > old and
> > >>> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > > > > >> coordinator
> > >>> > > > > > > > > > > >> > > > > >> > > will
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> > >>> protocol and
> > >>> > > > rely
> > >>> > > > > > on the
> > >>> > > > > > > > > same
> > >>> > > > > > > > > > > >> > persisted
> > >>> > > > > > > > > > > >> > > > > >> > > metadata so
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> > >>> together. This
> > >>> > > > point
> > >>> > > > > > is an
> > >>> > > > > > > > > > > >> offline
> > >>> > > > > > > > > > > >> > > > > >> migration.
> > >>> > > > > > > > > > > >> > > > > >> > We
> > >>> > > > > > > > > > > >> > > > > >> > > > > cannot
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> > >>> because it
> > >>> > > > would
> > >>> > > > > > require
> > >>> > > > > > > > > > > >> > shutting down
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > current
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> > >>> starting up the
> > >>> > > > > > new one
> > >>> > > > > > > > > and
> > >>> > > > > > > > > > > >> that
> > >>> > > > > > > > > > > >> > would
> > >>> > > > > > > > > > > >> > > > > >> cause
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> > >>> metadata
> > >>> > > > > > version/IBP
> > >>> > > > > > > > > must be
> > >>> > > > > > > > > > > >> > upgraded
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> X
> > >>> > > > > > > > > > > >> > > > > >> > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > order
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> > >>> protocol.
> > >>> > > > This
> > >>> > > > > > cannot
> > >>> > > > > > > > > be done
> > >>> > > > > > > > > > > >> > before
> > >>> > > > > > > > > > > >> > > > > 1)
> > >>> > > > > > > > > > > >> > > > > >> is
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because
> > the
> > >>> old
> > >>> > > > > > coordinator
> > >>> > > > > > > > > doesn't
> > >>> > > > > > > > > > > >> > support
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must
> > >>> be
> > >>> > > > upgraded
> > >>> > > > > > to a
> > >>> > > > > > > > > version
> > >>> > > > > > > > > > > >> > > > > supporting
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > online
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> > >>> KIP-792).
> > >>> > > > If
> > >>> > > > > > the
> > >>> > > > > > > > > consumer
> > >>> > > > > > > > > > > >> is
> > >>> > > > > > > > > > > >> > > > > already
> > >>> > > > > > > > > > > >> > > > > >> > > there.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done
> > >>> at this
> > >>> > > > point.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must
> > >>> be rolled
> > >>> > > > > > with the
> > >>> > > > > > > > > > > >> feature
> > >>> > > > > > > > > > > >> > flag
> > >>> > > > > > > > > > > >> > > > > >> turned
> > >>> > > > > > > > > > > >> > > > > >> > > on.
> > >>> > > > > > > > > > > >> > > > > >> > > > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> > >>> automatically
> > >>> > > > > > > > > converted when
> > >>> > > > > > > > > > > >> > the first
> > >>> > > > > > > > > > > >> > > > > >> > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> > protocol
> > >>> joins
> > >>> > > > the
> > >>> > > > > > group.
> > >>> > > > > > > > > While
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > members
> > >>> > > > > > > > > > > >> > > > > >> > > using the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> > being
> > >>> > > > upgraded,
> > >>> > > > > > the old
> > >>> > > > > > > > > > > >> protocol
> > >>> > > > > > > > > > > >> > is
> > >>> > > > > > > > > > > >> > > > > >> proxied
> > >>> > > > > > > > > > > >> > > > > >> > > into
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
> > >>> this in
> > >>> > > > the
> > >>> > > > > > KIP.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that
> > >>> we are
> > >>> > > > > > pushing the
> > >>> > > > > > > > > pattern
> > >>> > > > > > > > > > > >> > > > > >> subscriptions
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
> > >>> there
> > >>> > > > could be
> > >>> > > > > > some
> > >>> > > > > > > > > tricky
> > >>> > > > > > > > > > > >> > > > > >> compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > > > issues.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism
> > for
> > >>> users to
> > >>> > > > > > detect
> > >>> > > > > > > > > that
> > >>> > > > > > > > > > > >> they
> > >>> > > > > > > > > > > >> > need to
> > >>> > > > > > > > > > > >> > > > > >> > update
> > >>> > > > > > > > > > > >> > > > > >> > > > > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
> > >>> the new
> > >>> > > > > > protocol?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a
> > >>> bit more
> > >>> > > > > > optimistic
> > >>> > > > > > > > > than
> > >>> > > > > > > > > > > >> you
> > >>> > > > > > > > > > > >> > on this
> > >>> > > > > > > > > > > >> > > > > >> > > point. I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> > >>> majority of
> > >>> > > > the
> > >>> > > > > > cases
> > >>> > > > > > > > > are
> > >>> > > > > > > > > > > >> simple
> > >>> > > > > > > > > > > >> > > > > regexes
> > >>> > > > > > > > > > > >> > > > > >> > which
> > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> > >>> engine. The
> > >>> > > > > > > > > coordinator will
> > >>> > > > > > > > > > > >> > verify
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > regex
> > >>> > > > > > > > > > > >> > > > > >> > > > > anyway
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> > >>> consumer if the
> > >>> > > > > > regex is
> > >>> > > > > > > > > not
> > >>> > > > > > > > > > > >> valid.
> > >>> > > > > > > > > > > >> > > > > Coming
> > >>> > > > > > > > > > > >> > > > > >> > back
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in
> > the
> > >>> worst
> > >>> > > > case,
> > >>> > > > > > the
> > >>> > > > > > > > > first
> > >>> > > > > > > > > > > >> > upgraded
> > >>> > > > > > > > > > > >> > > > > >> > consumer
> > >>> > > > > > > > > > > >> > > > > >> > > > > joining
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> > >>> rejected.
> > >>> > > > This
> > >>> > > > > > should
> > >>> > > > > > > > > be used
> > >>> > > > > > > > > > > >> > as the
> > >>> > > > > > > > > > > >> > > > > >> last
> > >>> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers
> > >>> to
> > >>> > > > validate
> > >>> > > > > > their
> > >>> > > > > > > > > regex
> > >>> > > > > > > > > > > >> > before
> > >>> > > > > > > > > > > >> > > > > >> > upgrading
> > >>> > > > > > > > > > > >> > > > > >> > > > > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
> > >>> them with
> > >>> > > > > > another
> > >>> > > > > > > > > group.
> > >>> > > > > > > > > > > >> For
> > >>> > > > > > > > > > > >> > > > > >> instance,
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> > >>> pre-prod
> > >>> > > > > > environment.
> > >>> > > > > > > > > > > >> Another
> > >>> > > > > > > > > > > >> > way
> > >>> > > > > > > > > > > >> > > > > >> would be
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> > >>> consumer-group tool
> > >>> > > > to
> > >>> > > > > > > > > provide a
> > >>> > > > > > > > > > > >> regex
> > >>> > > > > > > > > > > >> > > > > >> validation
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this
> > >>> be
> > >>> > > > enough in
> > >>> > > > > > your
> > >>> > > > > > > > > > > >> opinion?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the
> > >>> last
> > >>> > > > question,
> > >>> > > > > > will
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> Java
> > >>> > > > > > > > > > > >> > client
> > >>> > > > > > > > > > > >> > > > > >> allow
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> > >>> current regex
> > >>> > > > > > engine for
> > >>> > > > > > > > > > > >> > compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > reasons?
> > >>> > > > > > > > > > > >> > > > > >> > > > > For
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
> > >>> keep using
> > >>> > > > > > client
> > >>> > > > > > > > > based
> > >>> > > > > > > > > > > >> > regex at
> > >>> > > > > > > > > > > >> > > > > >> first
> > >>> > > > > > > > > > > >> > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > >> > > keep
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple
> > >>> and then
> > >>> > > > > > migrate to
> > >>> > > > > > > > > server
> > >>> > > > > > > > > > > >> > based
> > >>> > > > > > > > > > > >> > > > > >> regexes
> > >>> > > > > > > > > > > >> > > > > >> > > as a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > second
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> > >>> point but I
> > >>> > > > am
> > >>> > > > > > > > > concerned that
> > >>> > > > > > > > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > allow
> > >>> > > > > > > > > > > >> > > > > >> > > > > users
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
> > >>> this mode.
> > >>> > > > That
> > >>> > > > > > > > > would go
> > >>> > > > > > > > > > > >> > against our
> > >>> > > > > > > > > > > >> > > > > >> goal
> > >>> > > > > > > > > > > >> > > > > >> > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> > client
> > >>> because
> > >>> > > > we
> > >>> > > > > > would
> > >>> > > > > > > > > have to
> > >>> > > > > > > > > > > >> > continue
> > >>> > > > > > > > > > > >> > > > > >> > > monitoring
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> > >>> client
> > >>> > > > side. I
> > >>> > > > > > would
> > >>> > > > > > > > > rather
> > >>> > > > > > > > > > > >> > not do
> > >>> > > > > > > > > > > >> > > > > >> this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
> > >>> the group
> > >>> > > > > > > > > coordinator
> > >>> > > > > > > > > > > >> will
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > > >> > > responsible for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations
> > >>> and that
> > >>> > > > the
> > >>> > > > > > > > > > > >> configurations
> > >>> > > > > > > > > > > >> > will be
> > >>> > > > > > > > > > > >> > > > > >> > > deleted
> > >>> > > > > > > > > > > >> > > > > >> > > > > when
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
> > >>> Will a
> > >>> > > > > > transition to
> > >>> > > > > > > > > DEAD
> > >>> > > > > > > > > > > >> > trigger
> > >>> > > > > > > > > > > >> > > > > >> > deletion
> > >>> > > > > > > > > > > >> > > > > >> > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > >>> > > > configurations
> > >>> > > > > > will be
> > >>> > > > > > > > > > > >> deleted
> > >>> > > > > > > > > > > >> > when
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > group is
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> > >>> together.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice
> > >>> to store
> > >>> > > > the
> > >>> > > > > > > > > configs in
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > group
> > >>> > > > > > > > > > > >> > > > > >> > > coordinator
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> > >>> cluster
> > >>> > > > > > configs and
> > >>> > > > > > > > > their
> > >>> > > > > > > > > > > >> > values?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
> > >>> group
> > >>> > > > > > > > > configurations are
> > >>> > > > > > > > > > > >> > overrides
> > >>> > > > > > > > > > > >> > > > > >> of
> > >>> > > > > > > > > > > >> > > > > >> > > cluster
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want
> > >>> to know
> > >>> > > > all
> > >>> > > > > > the
> > >>> > > > > > > > > overrides
> > >>> > > > > > > > > > > >> > though,
> > >>> > > > > > > > > > > >> > > > > you
> > >>> > > > > > > > > > > >> > > > > >> > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > have
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > >>> > > > coordinators.
> > >>> > > > > > You
> > >>> > > > > > > > > cannot
> > >>> > > > > > > > > > > >> > rely on
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > metadata
> > >>> > > > > > > > > > > >> > > > > >> > > > > log
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> > someone
> > >>> > > > configure a
> > >>> > > > > > group
> > >>> > > > > > > > > before
> > >>> > > > > > > > > > > >> > starting
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> > >>> allowing the
> > >>> > > > > > explicit
> > >>> > > > > > > > > creation
> > >>> > > > > > > > > > > >> of
> > >>> > > > > > > > > > > >> > > > > groups?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> > >>> decoupled
> > >>> > > > from
> > >>> > > > > > the
> > >>> > > > > > > > > group
> > >>> > > > > > > > > > > >> > lifecycle.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will
> > be
> > >>> > > > > > automatically
> > >>> > > > > > > > > created in
> > >>> > > > > > > > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > case.
> > >>> > > > > > > > > > > >> > > > > >> > > However,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
> > >>> lost after
> > >>> > > > the
> > >>> > > > > > > > > retention
> > >>> > > > > > > > > > > >> > period of
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > >>> > > > Consumer.subscribe
> > >>> > > > > > method
> > >>> > > > > > > > > for the
> > >>> > > > > > > > > > > >> > Java
> > >>> > > > > > > > > > > >> > > > > >> client
> > >>> > > > > > > > > > > >> > > > > >> > > still
> > >>> > > > > > > > > > > >> > > > > >> > > > > take
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> `java.util.regex.Pattern` of
> > >>> > > > do
> > >>> > > > > > we
> > >>> > > > > > > > > have to
> > >>> > > > > > > > > > > >> > introduce
> > >>> > > > > > > > > > > >> > > > > an
> > >>> > > > > > > > > > > >> > > > > >> > > overload?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> > >>> question. I
> > >>> > > > > > forgot
> > >>> > > > > > > > > about
> > >>> > > > > > > > > > > >> that
> > >>> > > > > > > > > > > >> > one.
> > >>> > > > > > > > > > > >> > > > > As
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> `java.util.regex.Pattern` is
> > >>> > > > not
> > >>> > > > > > fully
> > >>> > > > > > > > > > > >> compatible
> > >>> > > > > > > > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > engine
> > >>> > > > > > > > > > > >> > > > > >> > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> > >>> might be
> > >>> > > > better
> > >>> > > > > > to
> > >>> > > > > > > > > deprecate
> > >>> > > > > > > > > > > >> > it and
> > >>> > > > > > > > > > > >> > > > > >> use an
> > >>> > > > > > > > > > > >> > > > > >> > > > > overload
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string.
> > >>> We would
> > >>> > > > > > rely on
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> server
> > >>> > > > > > > > > > > >> > side
> > >>> > > > > > > > > > > >> > > > > >> > > validation.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration,
> > >>> I think
> > >>> > > > that
> > >>> > > > > > we
> > >>> > > > > > > > > could
> > >>> > > > > > > > > > > >> still
> > >>> > > > > > > > > > > >> > try to
> > >>> > > > > > > > > > > >> > > > > >> > > toString
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it.
> > That
> > >>> should
> > >>> > > > > > work, I
> > >>> > > > > > > > > think, in
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> majority
> > >>> > > > > > > > > > > >> > > > > >> > > of the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> > >>> Justine
> > >>> > > > that we
> > >>> > > > > > > > > should be
> > >>> > > > > > > > > > > >> > clearer
> > >>> > > > > > > > > > > >> > > > > about
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > reason
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > >>> > > > IBP/metadata.version
> > >>> > > > > > from the
> > >>> > > > > > > > > > > >> feature
> > >>> > > > > > > > > > > >> > flag.
> > >>> > > > > > > > > > > >> > > > > >> Maybe
> > >>> > > > > > > > > > > >> > > > > >> > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > mean
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> > >>> default for
> > >>> > > > the
> > >>> > > > > > feature
> > >>> > > > > > > > > flag
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > true
> > >>> > > > > > > > > > > >> > > > > >> based
> > >>> > > > > > > > > > > >> > > > > >> > on
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> > >>> once we
> > >>> > > > want to
> > >>> > > > > > make
> > >>> > > > > > > > > it the
> > >>> > > > > > > > > > > >> > default.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> > >>> that feature
> > >>> > > > > > flag
> > >>> > > > > > > > > mainly
> > >>> > > > > > > > > > > >> > during the
> > >>> > > > > > > > > > > >> > > > > >> > > development
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not
> > >>> have
> > >>> > > > mentioned
> > >>> > > > > > it, I
> > >>> > > > > > > > > think,
> > >>> > > > > > > > > > > >> > because
> > >>> > > > > > > > > > > >> > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > could
> > >>> > > > > > > > > > > >> > > > > >> > > > > use
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config
> > for
> > >>> it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> > >>> protocol APIs
> > >>> > > > > > don't
> > >>> > > > > > > > > mention
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > required
> > >>> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that
> > for
> > >>> > > > consistency.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
> > >>> that
> > >>> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > >>> > > > > > > > > > > >> > > > > requires
> > >>> > > > > > > > > > > >> > > > > >> > > "Read
> > >>> > > > > > > > > > > >> > > > > >> > > > > Group"
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to
> > >>> do more
> > >>> > > > than
> > >>> > > > > > > > > reading.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how
> > >>> the
> > >>> > > > current
> > >>> > > > > > > > > protocol works
> > >>> > > > > > > > > > > >> > though.
> > >>> > > > > > > > > > > >> > > > > We
> > >>> > > > > > > > > > > >> > > > > >> > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group"
> > >>> to join a
> > >>> > > > > > group. We
> > >>> > > > > > > > > could
> > >>> > > > > > > > > > > >> > consider
> > >>> > > > > > > > > > > >> > > > > >> > > changing
> > >>> > > > > > > > > > > >> > > > > >> > > > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure
> > that
> > >>> it is
> > >>> > > > worth
> > >>> > > > > > it.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> > >>> recreation
> > >>> > > > > > handled by
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > consumer
> > >>> > > > > > > > > > > >> > > > > with
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would
> > >>> be good to
> > >>> > > > > > have a
> > >>> > > > > > > > > section
> > >>> > > > > > > > > > > >> on
> > >>> > > > > > > > > > > >> > this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> > protocol
> > >>> > > > > > perspective, the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> topic
> > >>> > > > > > > > > > > >> > will
> > >>> > > > > > > > > > > >> > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> > >>> treat it
> > >>> > > > like a
> > >>> > > > > > > > > topic with
> > >>> > > > > > > > > > > >> a
> > >>> > > > > > > > > > > >> > > > > different
> > >>> > > > > > > > > > > >> > > > > >> > > name.
> > >>> > > > > > > > > > > >> > > > > >> > > > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that
> > the
> > >>> > > > fetch/commit
> > >>> > > > > > > > > offsets
> > >>> > > > > > > > > > > >> APIs
> > >>> > > > > > > > > > > >> > do not
> > >>> > > > > > > > > > > >> > > > > >> > support
> > >>> > > > > > > > > > > >> > > > > >> > > > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
> > >>> would
> > >>> > > > reuse the
> > >>> > > > > > > > > offsets
> > >>> > > > > > > > > > > >> based
> > >>> > > > > > > > > > > >> > on the
> > >>> > > > > > > > > > > >> > > > > >> > same.
> > >>> > > > > > > > > > > >> > > > > >> > > I
> > >>> > > > > > > > > > > >> > > > > >> > > > > think
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
> > >>> those
> > >>> > > > APIs
> > >>> > > > > > as well
> > >>> > > > > > > > > in
> > >>> > > > > > > > > > > >> order
> > >>> > > > > > > > > > > >> > to be
> > >>> > > > > > > > > > > >> > > > > >> > > consistent
> > >>> > > > > > > > > > > >> > > > > >> > > > > end
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> > >>> strengthen
> > >>> > > > the
> > >>> > > > > > > > > semantics of
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> consumer.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> > mentions
> > >>> we will
> > >>> > > > > > write the
> > >>> > > > > > > > > new
> > >>> > > > > > > > > > > >> > coordinator
> > >>> > > > > > > > > > > >> > > > > >> in
> > >>> > > > > > > > > > > >> > > > > >> > > Java.
> > >>> > > > > > > > > > > >> > > > > >> > > > > Even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> > >>> implementation
> > >>> > > > > > detail, do
> > >>> > > > > > > > > we plan
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > have a
> > >>> > > > > > > > > > > >> > > > > >> new
> > >>> > > > > > > > > > > >> > > > > >> > > gradle
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > module
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> > >>> scalability
> > >>> > > > > > goal when
> > >>> > > > > > > > > it
> > >>> > > > > > > > > > > >> comes
> > >>> > > > > > > > > > > >> > to how
> > >>> > > > > > > > > > > >> > > > > >> many
> > >>> > > > > > > > > > > >> > > > > >> > > > > members
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> > >>> support?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers
> > >>> at the
> > >>> > > > > > moment. The
> > >>> > > > > > > > > > > >> protocol
> > >>> > > > > > > > > > > >> > should
> > >>> > > > > > > > > > > >> > > > > >> > > support
> > >>> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group.
> > >>> We will
> > >>> > > > > > measure
> > >>> > > > > > > > > this when
> > >>> > > > > > > > > > > >> > we have
> > >>> > > > > > > > > > > >> > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > first
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
> > >>> that we
> > >>> > > > might
> > >>> > > > > > have
> > >>> > > > > > > > > other
> > >>> > > > > > > > > > > >> > > > > bottlenecks
> > >>> > > > > > > > > > > >> > > > > >> > down
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> > >>> commits).
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> > >>> having
> > >>> > > > > > > > > SubscribedTopidIds
> > >>> > > > > > > > > > > >> > instead
> > >>> > > > > > > > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > SubscribedTopicNames
> > >>> in
> > >>> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > >>> > > > > > > > > > > >> > > > > >> Is
> > >>> > > > > > > > > > > >> > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > idea
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> > >>> resolve the
> > >>> > > > > > regex on
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > server, we
> > >>> > > > > > > > > > > >> > > > > >> can do
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> > >>> > > > difference
> > >>> > > > > > is that
> > >>> > > > > > > > > > > >> sending
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> regex is
> > >>> > > > > > > > > > > >> > > > > >> > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
> > >>> topic
> > >>> > > > names
> > >>> > > > > > is less
> > >>> > > > > > > > > > > >> > efficient.
> > >>> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> > easier
> > >>> to
> > >>> > > > handle
> > >>> > > > > > if we
> > >>> > > > > > > > > have
> > >>> > > > > > > > > > > >> topic
> > >>> > > > > > > > > > > >> > ids.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> > >>> consolidate the
> > >>> > > > > > metadata
> > >>> > > > > > > > > lookup
> > >>> > > > > > > > > > > >> > on the
> > >>> > > > > > > > > > > >> > > > > >> server
> > >>> > > > > > > > > > > >> > > > > >> > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > both
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
> > >>> with your
> > >>> > > > > > point. As
> > >>> > > > > > > > > a
> > >>> > > > > > > > > > > >> second
> > >>> > > > > > > > > > > >> > > > > though,
> > >>> > > > > > > > > > > >> > > > > >> > using
> > >>> > > > > > > > > > > >> > > > > >> > > > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better
> > here
> > >>> for the
> > >>> > > > > > delete and
> > >>> > > > > > > > > > > >> > recreation
> > >>> > > > > > > > > > > >> > > > > case.
> > >>> > > > > > > > > > > >> > > > > >> > > Also, I
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
> > >>> allow
> > >>> > > > users to
> > >>> > > > > > > > > subscribe
> > >>> > > > > > > > > > > >> with
> > >>> > > > > > > > > > > >> > topic
> > >>> > > > > > > > > > > >> > > > > >> ids
> > >>> > > > > > > > > > > >> > > > > >> > in
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that
> > is
> > >>> the only
> > >>> > > > > > way to be
> > >>> > > > > > > > > > > >> really
> > >>> > > > > > > > > > > >> > robust
> > >>> > > > > > > > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > topic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022
> > >>> at 1:38 PM
> > >>> > > > > > David
> > >>> > > > > > > > > Jacot <
> > >>> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> > >>> comments.
> > >>> > > > Please
> > >>> > > > > > find
> > >>> > > > > > > > > my
> > >>> > > > > > > > > > > >> answers
> > >>> > > > > > > > > > > >> > > > > below.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> > >>> protocol
> > >>> > > > relies on
> > >>> > > > > > > > > topic IDs
> > >>> > > > > > > > > > > >> > with the
> > >>> > > > > > > > > > > >> > > > > >> > > exception
> > >>> > > > > > > > > > > >> > > > > >> > > > > of the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based
> > in
> > >>> the
> > >>> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > >>> > > > > > > > > > > >> > > > > >> I
> > >>> > > > > > > > > > > >> > > > > >> > am
> > >>> > > > > > > > > > > >> > > > > >> > > not
> > >>> > > > > > > > > > > >> > > > > >> > > > > sure
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> > names
> > >>> is the
> > >>> > > > > > right call
> > >>> > > > > > > > > here.
> > >>> > > > > > > > > > > >> I
> > >>> > > > > > > > > > > >> > need to
> > >>> > > > > > > > > > > >> > > > > >> > think
> > >>> > > > > > > > > > > >> > > > > >> > > > > about it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> > >>> Obviously, the
> > >>> > > > > > KIP does
> > >>> > > > > > > > > not
> > >>> > > > > > > > > > > >> > change the
> > >>> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
> > >>> topic
> > >>> > > > IDs.
> > >>> > > > > > This
> > >>> > > > > > > > > may be
> > >>> > > > > > > > > > > >> > something
> > >>> > > > > > > > > > > >> > > > > >> that
> > >>> > > > > > > > > > > >> > > > > >> > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as
> > it
> > >>> would
> > >>> > > > give
> > >>> > > > > > better
> > >>> > > > > > > > > > > >> overall
> > >>> > > > > > > > > > > >> > > > > >> guarantee in
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> > >>> think that
> > >>> > > > I
> > >>> > > > > > should
> > >>> > > > > > > > > not
> > >>> > > > > > > > > > > >> have
> > >>> > > > > > > > > > > >> > > > > mentioned
> > >>> > > > > > > > > > > >> > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > flag at
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove
> > >>> it. We can
> > >>> > > > > > use an
> > >>> > > > > > > > > internal
> > >>> > > > > > > > > > > >> > > > > >> configuration
> > >>> > > > > > > > > > > >> > > > > >> > > while
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> > >>> feature.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> > types
> > >>> will be
> > >>> > > > > > > > > supported. The
> > >>> > > > > > > > > > > >> > change is
> > >>> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
> > >>> that the
> > >>> > > > > > cluster
> > >>> > > > > > > > > uses
> > >>> > > > > > > > > > > >> topic
> > >>> > > > > > > > > > > >> > IDs.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11,
> > 2022
> > >>> at 9:53
> > >>> > > > PM
> > >>> > > > > > > > > Guozhang
> > >>> > > > > > > > > > > >> Wang <
> > >>> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> > >>> feedback.
> > >>> > > > Here
> > >>> > > > > > are
> > >>> > > > > > > > > some
> > >>> > > > > > > > > > > >> replies
> > >>> > > > > > > > > > > >> > > > > inlined
> > >>> > > > > > > > > > > >> > > > > >> > > below:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> > >>> 2022 at
> > >>> > > > 2:53 AM
> > >>> > > > > > > > > Ismael Juma
> > >>> > > > > > > > > > > >> <
> > >>> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
> > >>> KIP. This
> > >>> > > > has
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> potential
> > >>> > > > > > > > > > > >> > to be a
> > >>> > > > > > > > > > > >> > > > > >> > great
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > >>> > > > questions/comments:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> > >>> > > > premature to
> > >>> > > > > > talk
> > >>> > > > > > > > > about
> > >>> > > > > > > > > > > >> > target
> > >>> > > > > > > > > > > >> > > > > >> versions
> > >>> > > > > > > > > > > >> > > > > >> > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> > >>> existing
> > >>> > > > group
> > >>> > > > > > > > > protocol.
> > >>> > > > > > > > > > > >> > Unlike
> > >>> > > > > > > > > > > >> > > > > >> KRaft,
> > >>> > > > > > > > > > > >> > > > > >> > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
> > >>> and hence
> > >>> > > > > > > > > > > >> > deprecation/removal will
> > >>> > > > > > > > > > > >> > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> > > heavily
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> > >>> applications
> > >>> > > > > > migrate
> > >>> > > > > > > > > to the
> > >>> > > > > > > > > > > >> new
> > >>> > > > > > > > > > > >> > > > > >> protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
> > >>> you. I
> > >>> > > > think
> > >>> > > > > > we can
> > >>> > > > > > > > > > > >> remove
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> proposed
> > >>> > > > > > > > > > > >> > > > > >> > > > > timeline
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > >>> > > > Deprecation,
> > >>> > > > > > and
> > >>> > > > > > > > > Migration
> > >>> > > > > > > > > > > >> > Plan` and
> > >>> > > > > > > > > > > >> > > > > >> > > instead
> > >>> > > > > > > > > > > >> > > > > >> > > > > just
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > state
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> > >>> decide in the
> > >>> > > > > > future
> > >>> > > > > > > > > about
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > we
> > >>> > > > > > > > > > > >> > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> > >>> behaviors.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says
> > >>> we
> > >>> > > > intend to
> > >>> > > > > > > > > release this
> > >>> > > > > > > > > > > >> > in 4.x,
> > >>> > > > > > > > > > > >> > > > > >> but
> > >>> > > > > > > > > > > >> > > > > >> > it
> > >>> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > made
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> > added
> > >>> that as
> > >>> > > > a
> > >>> > > > > > way to
> > >>> > > > > > > > > > > >> estimate
> > >>> > > > > > > > > > > >> > when
> > >>> > > > > > > > > > > >> > > > > we'd
> > >>> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> > >>> protocol, I
> > >>> > > > also
> > >>> > > > > > suggest
> > >>> > > > > > > > > > > >> removing
> > >>> > > > > > > > > > > >> > this
> > >>> > > > > > > > > > > >> > > > > >> part.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's
> > not
> > >>> > > > specifically
> > >>> > > > > > > > > related to
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> > >>> more for
> > >>> > > > client
> > >>> > > > > > > > > upgrades.
> > >>> > > > > > > > > > > >> I.e.
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> > > broker-side
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation
> > may
> > >>> be done
> > >>> > > > > > first,
> > >>> > > > > > > > > and then
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > client
> > >>> > > > > > > > > > > >> > > > > >> > side,
> > >>> > > > > > > > > > > >> > > > > >> > > > > and we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> > >>> "released"
> > >>> > > > by
> > >>> > > > > > the
> > >>> > > > > > > > > time
> > >>> > > > > > > > > > > >> clients
> > >>> > > > > > > > > > > >> > > > > >> > > implementations
> > >>> > > > > > > > > > > >> > > > > >> > > > > are
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> > >>> enable the
> > >>> > > > > > feature the
> > >>> > > > > > > > > > > >> clients
> > >>> > > > > > > > > > > >> > need to
> > >>> > > > > > > > > > > >> > > > > >> > first
> > >>> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> > >>> rolling
> > >>> > > > bounce
> > >>> > > > > > and
> > >>> > > > > > > > > then set
> > >>> > > > > > > > > > > >> > the flag
> > >>> > > > > > > > > > > >> > > > > >> > with a
> > >>> > > > > > > > > > > >> > > > > >> > > > > second
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence
> > >>> we feel
> > >>> > > > it's
> > >>> > > > > > > > > better to be
> > >>> > > > > > > > > > > >> > released
> > >>> > > > > > > > > > > >> > > > > >> in a
> > >>> > > > > > > > > > > >> > > > > >> > > major
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> > >>> flesh out
> > >>> > > > the
> > >>> > > > > > > > > details of
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > migration
> > >>> > > > > > > > > > > >> > > > > >> > > story.
> > >>> > > > > > > > > > > >> > > > > >> > > > > It
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
> > >>> will
> > >>> > > > support
> > >>> > > > > > online
> > >>> > > > > > > > > > > >> > migrations. Is
> > >>> > > > > > > > > > > >> > > > > >> that
> > >>> > > > > > > > > > > >> > > > > >> > > > > correct?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > We
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> > >>> detail.
> > >>> > > > It
> > >>> > > > > > could
> > >>> > > > > > > > > also be
> > >>> > > > > > > > > > > >> > done as a
> > >>> > > > > > > > > > > >> > > > > >> > > separate
> > >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> > that's
> > >>> the
> > >>> > > > part we
> > >>> > > > > > can
> > >>> > > > > > > > > be more
> > >>> > > > > > > > > > > >> > concrete
> > >>> > > > > > > > > > > >> > > > > >> about
> > >>> > > > > > > > > > > >> > > > > >> > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > sure
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related
> > to
> > >>> your
> > >>> > > > > > question 2)
> > >>> > > > > > > > > above).
> > >>> > > > > > > > > > > >> > We will
> > >>> > > > > > > > > > > >> > > > > >> work
> > >>> > > > > > > > > > > >> > > > > >> > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > making
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> > >>> parallel as we
> > >>> > > > > > solicit
> > >>> > > > > > > > > more
> > >>> > > > > > > > > > > >> > feedback.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
> > >>> that we are
> > >>> > > > > > pushing
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> > pattern
> > >>> > > > > > > > > > > >> > > > > >> > > subscriptions
> > >>> > > > > > > > > > > >> > > > > >> > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> > >>> like there
> > >>> > > > > > could be
> > >>> > > > > > > > > some
> > >>> > > > > > > > > > > >> tricky
> > >>> > > > > > > > > > > >> > > > > >> > > compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> > mechanism
> > >>> for
> > >>> > > > users to
> > >>> > > > > > > > > detect that
> > >>> > > > > > > > > > > >> > they
> > >>> > > > > > > > > > > >> > > > > need
> > >>> > > > > > > > > > > >> > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> > > update
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > their
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> > switching
> > >>> to the
> > >>> > > > new
> > >>> > > > > > > > > protocol?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> > >>> need some
> > >>> > > > > > tooling for
> > >>> > > > > > > > > > > >> non-java
> > >>> > > > > > > > > > > >> > client
> > >>> > > > > > > > > > > >> > > > > >> > users
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > sort
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> > >>> client before
> > >>> > > > > > > > > switching to
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > new
> > >>> > > > > > > > > > > >> > > > > >> > protocol.
> > >>> > > > > > > > > > > >> > > > > >> > > I
> > >>> > > > > > > > > > > >> > > > > >> > > > > do not
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on
> > >>> top of my
> > >>> > > > head
> > >>> > > > > > > > > though,
> > >>> > > > > > > > > > > >> maybe
> > >>> > > > > > > > > > > >> > others
> > >>> > > > > > > > > > > >> > > > > >> like
> > >>> > > > > > > > > > > >> > > > > >> > > @Matt
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to
> > >>> the last
> > >>> > > > > > question,
> > >>> > > > > > > > > will the
> > >>> > > > > > > > > > > >> > Java
> > >>> > > > > > > > > > > >> > > > > >> client
> > >>> > > > > > > > > > > >> > > > > >> > > allow
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> > >>> current
> > >>> > > > regex
> > >>> > > > > > > > > engine for
> > >>> > > > > > > > > > > >> > > > > >> compatibility
> > >>> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > For
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy
> > >>> to keep
> > >>> > > > using
> > >>> > > > > > > > > client
> > >>> > > > > > > > > > > >> based
> > >>> > > > > > > > > > > >> > regex
> > >>> > > > > > > > > > > >> > > > > at
> > >>> > > > > > > > > > > >> > > > > >> > > first to
> > >>> > > > > > > > > > > >> > > > > >> > > > > keep
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> > >>> simple and
> > >>> > > > then
> > >>> > > > > > migrate
> > >>> > > > > > > > > to
> > >>> > > > > > > > > > > >> > server based
> > >>> > > > > > > > > > > >> > > > > >> > > regexes
> > >>> > > > > > > > > > > >> > > > > >> > > > > as a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have
> > >>> not thought
> > >>> > > > > > about
> > >>> > > > > > > > > that for
> > >>> > > > > > > > > > > >> > java
> > >>> > > > > > > > > > > >> > > > > >> clients,
> > >>> > > > > > > > > > > >> > > > > >> > > and
> > >>> > > > > > > > > > > >> > > > > >> > > > > we can
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind
> > of
> > >>> > > > > > compatibility
> > >>> > > > > > > > > issues do
> > >>> > > > > > > > > > > >> > you have
> > >>> > > > > > > > > > > >> > > > > in
> > >>> > > > > > > > > > > >> > > > > >> > > mind?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
> > >>> that the
> > >>> > > > group
> > >>> > > > > > > > > coordinator
> > >>> > > > > > > > > > > >> > will be
> > >>> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> > >>> configurations and
> > >>> > > > that
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> > configurations
> > >>> > > > > > > > > > > >> > > > > will
> > >>> > > > > > > > > > > >> > > > > >> be
> > >>> > > > > > > > > > > >> > > > > >> > > > > deleted
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > when
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> > >>> deleted. Will a
> > >>> > > > > > > > > transition to
> > >>> > > > > > > > > > > >> DEAD
> > >>> > > > > > > > > > > >> > > > > trigger
> > >>> > > > > > > > > > > >> > > > > >> > > deletion
> > >>> > > > > > > > > > > >> > > > > >> > > > > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> > >>> DEAD state
> > >>> > > > is an
> > >>> > > > > > > > > ending
> > >>> > > > > > > > > > > >> state
> > >>> > > > > > > > > > > >> > (we
> > >>> > > > > > > > > > > >> > > > > would
> > >>> > > > > > > > > > > >> > > > > >> > only
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> > >>> group is
> > >>> > > > EMPTY
> > >>> > > > > > and
> > >>> > > > > > > > > also all
> > >>> > > > > > > > > > > >> of
> > >>> > > > > > > > > > > >> > its
> > >>> > > > > > > > > > > >> > > > > >> > metadata
> > >>> > > > > > > > > > > >> > > > > >> > > are
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
> > >>> DEAD this
> > >>> > > > > > group
> > >>> > > > > > > > > would
> > >>> > > > > > > > > > > >> never
> > >>> > > > > > > > > > > >> > be
> > >>> > > > > > > > > > > >> > > > > >> revived.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> > >>> choice to
> > >>> > > > store
> > >>> > > > > > the
> > >>> > > > > > > > > configs
> > >>> > > > > > > > > > > >> in
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list
> > >>> all
> > >>> > > > cluster
> > >>> > > > > > configs
> > >>> > > > > > > > > and
> > >>> > > > > > > > > > > >> their
> > >>> > > > > > > > > > > >> > > > > values?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> > >>> question,
> > >>> > > > and our
> > >>> > > > > > > > > thoughts
> > >>> > > > > > > > > > > >> are
> > >>> > > > > > > > > > > >> > that
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > so-called
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations"
> > are
> > >>> > > > overrides
> > >>> > > > > > of the
> > >>> > > > > > > > > > > >> > cluster-level
> > >>> > > > > > > > > > > >> > > > > >> > > > > configurations
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> > >>> group so
> > >>> > > > when an
> > >>> > > > > > > > > admin list
> > >>> > > > > > > > > > > >> > cluster
> > >>> > > > > > > > > > > >> > > > > >> > configs
> > >>> > > > > > > > > > > >> > > > > >> > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> > >>> cluster-level
> > >>> > > > > > > > > "defaults", not
> > >>> > > > > > > > > > > >> > showing
> > >>> > > > > > > > > > > >> > > > > >> any
> > >>> > > > > > > > > > > >> > > > > >> > > > > per-group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> > >>> someone
> > >>> > > > > > configure a
> > >>> > > > > > > > > group
> > >>> > > > > > > > > > > >> > before
> > >>> > > > > > > > > > > >> > > > > >> starting
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> > >>> allowing
> > >>> > > > the
> > >>> > > > > > explicit
> > >>> > > > > > > > > > > >> > creation of
> > >>> > > > > > > > > > > >> > > > > >> groups?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could
> > be
> > >>> > > > decoupled
> > >>> > > > > > from
> > >>> > > > > > > > > the group
> > >>> > > > > > > > > > > >> > > > > lifecycle.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can
> > be
> > >>> created
> > >>> > > > > > before
> > >>> > > > > > > > > the group
> > >>> > > > > > > > > > > >> > itself
> > >>> > > > > > > > > > > >> > > > > as
> > >>> > > > > > > > > > > >> > > > > >> an
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course,
> > this
> > >>> > > > requires
> > >>> > > > > > the
> > >>> > > > > > > > > > > >> corresponding
> > >>> > > > > > > > > > > >> > > > > >> request to
> > >>> > > > > > > > > > > >> > > > > >> > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> > >>> based on
> > >>> > > > the
> > >>> > > > > > group
> > >>> > > > > > > > > id ---
> > >>> > > > > > > > > > > >> > the only
> > >>> > > > > > > > > > > >> > > > > >> > thing
> > >>> > > > > > > > > > > >> > > > > >> > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> > >>> itself is
> > >>> > > > gone
> > >>> > > > > > we
> > >>> > > > > > > > > also check
> > >>> > > > > > > > > > > >> > if there
> > >>> > > > > > > > > > > >> > > > > >> are
> > >>> > > > > > > > > > > >> > > > > >> > > any
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related
> > >>> to that
> > >>> > > > group
> > >>> > > > > > and
> > >>> > > > > > > > > delete
> > >>> > > > > > > > > > > >> as
> > >>> > > > > > > > > > > >> > well.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> > >>> indeed
> > >>> > > > > > introduces an
> > >>> > > > > > > > > > > >> > asymmetry on
> > >>> > > > > > > > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > creation
> > >>> > > > > > > > > > > >> > > > > >> > > > > /
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> > >>> config
> > >>> > > > > > entities,
> > >>> > > > > > > > > and we
> > >>> > > > > > > > > > > >> > would like
> > >>> > > > > > > > > > > >> > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> > > hear
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether
> > >>> we should
> > >>> > > > aim
> > >>> > > > > > for
> > >>> > > > > > > > > symmetry
> > >>> > > > > > > > > > > >> > i.e.
> > >>> > > > > > > > > > > >> > > > > >> totally
> > >>> > > > > > > > > > > >> > > > > >> > > > > decouple
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence
> > >>> not
> > >>> > > > delete
> > >>> > > > > > them at
> > >>> > > > > > > > > all
> > >>> > > > > > > > > > > >> when
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> group
> > >>> > > > > > > > > > > >> > > > > >> > is
> > >>> > > > > > > > > > > >> > > > > >> > > > > gone,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > but
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> > >>> deletion
> > >>> > > > > > operations
> > >>> > > > > > > > > by
> > >>> > > > > > > > > > > >> > themselves.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > >>> > > > > > Consumer.subscribe
> > >>> > > > > > > > > method for
> > >>> > > > > > > > > > > >> > the Java
> > >>> > > > > > > > > > > >> > > > > >> > client
> > >>> > > > > > > > > > > >> > > > > >> > > > > still
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > take
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > `java.util.regex.Pattern` of
> > >>> > > > > > do we
> > >>> > > > > > > > > have
> > >>> > > > > > > > > > > >> to
> > >>> > > > > > > > > > > >> > > > > >> introduce an
> > >>> > > > > > > > > > > >> > > > > >> > > > > overload?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
> > >>> need to
> > >>> > > > > > introduce
> > >>> > > > > > > > > an
> > >>> > > > > > > > > > > >> > overload, but
> > >>> > > > > > > > > > > >> > > > > >> I'm
> > >>> > > > > > > > > > > >> > > > > >> > > all
> > >>> > > > > > > > > > > >> > > > > >> > > > > ears
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> > >>> compatibility
> > >>> > > > > > issues
> > >>> > > > > > > > > that we
> > >>> > > > > > > > > > > >> may
> > >>> > > > > > > > > > > >> > > > > overlook.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree
> > with
> > >>> Justine
> > >>> > > > > > that we
> > >>> > > > > > > > > should
> > >>> > > > > > > > > > > >> be
> > >>> > > > > > > > > > > >> > clearer
> > >>> > > > > > > > > > > >> > > > > >> > about
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > >>> > > > > > IBP/metadata.version
> > >>> > > > > > > > > from the
> > >>> > > > > > > > > > > >> > feature
> > >>> > > > > > > > > > > >> > > > > >> flag.
> > >>> > > > > > > > > > > >> > > > > >> > > Maybe
> > >>> > > > > > > > > > > >> > > > > >> > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> > >>> default
> > >>> > > > for
> > >>> > > > > > the
> > >>> > > > > > > > > feature
> > >>> > > > > > > > > > > >> flag
> > >>> > > > > > > > > > > >> > to
> > >>> > > > > > > > > > > >> > > > > true
> > >>> > > > > > > > > > > >> > > > > >> > > based on
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > metadata.version
> > >>> once we
> > >>> > > > > > want to
> > >>> > > > > > > > > make it
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> default.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> > >>> protocol
> > >>> > > > APIs
> > >>> > > > > > don't
> > >>> > > > > > > > > mention
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > >> required
> > >>> > > > > > > > > > > >> > > > > >> > > ACLs,
> > >>> > > > > > > > > > > >> > > > > >> > > > > it
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add
> > that
> > >>> for
> > >>> > > > > > consistency.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> > >>> certainly do
> > >>> > > > that.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
> > >>> odd that
> > >>> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > >>> > > > > > > > > > > >> > > > > >> > requires
> > >>> > > > > > > > > > > >> > > > > >> > > > > "Read
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems
> > >>> to do
> > >>> > > > more
> > >>> > > > > > than
> > >>> > > > > > > > > reading.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> > thought
> > >>> myself
> > >>> > > > as
> > >>> > > > > > well,
> > >>> > > > > > > > > but in
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > end we
> > >>> > > > > > > > > > > >> > > > > >> > could
> > >>> > > > > > > > > > > >> > > > > >> > > not
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> > >>> alternative: adding
> > >>> > > > > > Write
> > >>> > > > > > > > > Group
> > >>> > > > > > > > > > > >> seems
> > >>> > > > > > > > > > > >> > an
> > >>> > > > > > > > > > > >> > > > > >> overkill
> > >>> > > > > > > > > > > >> > > > > >> > > here
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> > >>> elsewhere (we
> > >>> > > > only
> > >>> > > > > > have
> > >>> > > > > > > > > Read /
> > >>> > > > > > > > > > > >> > Delete
> > >>> > > > > > > > > > > >> > > > > and
> > >>> > > > > > > > > > > >> > > > > >> > > Describe
> > >>> > > > > > > > > > > >> > > > > >> > > > > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like
> > >>> to hear
> > >>> > > > others
> > >>> > > > > > > > > thoughts.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> > topic
> > >>> > > > recreation
> > >>> > > > > > > > > handled by
> > >>> > > > > > > > > > > >> the
> > >>> > > > > > > > > > > >> > > > > consumer
> > >>> > > > > > > > > > > >> > > > > >> > with
> > >>> > > > > > > > > > > >> > > > > >> > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> > >>> would be
> > >>> > > > good
> > >>> > > > > > to have
> > >>> > > > > > > > > a
> > >>> > > > > > > > > > > >> > section on
> > >>> > > > > > > > > > > >> > > > > >> this.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with
> > regex
> > >>> > > > > > subscription
> > >>> > > > > > > > > right? Yes
> > >>> > > > > > > > > > > >> > we can
> > >>> > > > > > > > > > > >> > > > > >> add a
> > >>> > > > > > > > > > > >> > > > > >> > > > > section
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > about
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> > >>> basically the
> > >>> > > > idea
> > >>> > > > > > is that
> > >>> > > > > > > > > > > >> consumer
> > >>> > > > > > > > > > > >> > would
> > >>> > > > > > > > > > > >> > > > > be
> > >>> > > > > > > > > > > >> > > > > >> > > totally
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol
> > >>> as it's
> > >>> > > > > > handled all
> > >>> > > > > > > > > by the
> > >>> > > > > > > > > > > >> > brokers.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> > >>> mentions we
> > >>> > > > will
> > >>> > > > > > write
> > >>> > > > > > > > > the
> > >>> > > > > > > > > > > >> new
> > >>> > > > > > > > > > > >> > > > > >> coordinator
> > >>> > > > > > > > > > > >> > > > > >> > in
> > >>> > > > > > > > > > > >> > > > > >> > > > > Java.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> > >>> implementation
> > >>> > > > > > detail,
> > >>> > > > > > > > > do we
> > >>> > > > > > > > > > > >> > plan to
> > >>> > > > > > > > > > > >> > > > > >> have a
> > >>> > > > > > > > > > > >> > > > > >> > > new
> > >>> > > > > > > > > > > >> > > > > >> > > > > gradle
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> > >>> thought about
> > >>> > > > > > that. But
> > >>> > > > > > > > > I think
> > >>> > > > > > > > > > > >> > the
> > >>> > > > > > > > > > > >> > > > > answer
> > >>> > > > > > > > > > > >> > > > > >> > > should
> > >>> > > > > > > > > > > >> > > > > >> > > > > be
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have
> > a
> > >>> > > > scalability
> > >>> > > > > > goal
> > >>> > > > > > > > > when it
> > >>> > > > > > > > > > > >> > comes to
> > >>> > > > > > > > > > > >> > > > > >> how
> > >>> > > > > > > > > > > >> > > > > >> > > many
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > members
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol
> > >>> can
> > >>> > > > support?
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> > >>> think we
> > >>> > > > > > should
> > >>> > > > > > > > > shoot for
> > >>> > > > > > > > > > > >> > 1000s of
> > >>> > > > > > > > > > > >> > > > > >> > > members.
> > >>> > > > > > > > > > > >> > > > > >> > > > > But
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
> > >>> also
> > >>> > > > depend
> > >>> > > > > > on the
> > >>> > > > > > > > > offset
> > >>> > > > > > > > > > > >> > > > > management
> > >>> > > > > > > > > > > >> > > > > >> > > (commit,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of
> > the
> > >>> > > > coordinator
> > >>> > > > > > > > > which we
> > >>> > > > > > > > > > > >> did
> > >>> > > > > > > > > > > >> > not
> > >>> > > > > > > > > > > >> > > > > >> cover in
> > >>> > > > > > > > > > > >> > > > > >> > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> > >>> number that
> > >>> > > > > > applies
> > >>> > > > > > > > > > > >> > universally.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> > >>> consider
> > >>> > > > having
> > >>> > > > > > > > > > > >> > SubscribedTopidIds
> > >>> > > > > > > > > > > >> > > > > >> instead
> > >>> > > > > > > > > > > >> > > > > >> > of
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> SubscribedTopicNames in
> > >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > >>> > > > > > > > > > > >> > > > > >> > > Is the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have
> > to
> > >>> resolve
> > >>> > > > the
> > >>> > > > > > regex
> > >>> > > > > > > > > on the
> > >>> > > > > > > > > > > >> > server,
> > >>> > > > > > > > > > > >> > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > can
> > >>> > > > > > > > > > > >> > > > > >> > > do
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > same
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name?
> > >>> The
> > >>> > > > > > difference is
> > >>> > > > > > > > > that
> > >>> > > > > > > > > > > >> > sending the
> > >>> > > > > > > > > > > >> > > > > >> > regex
> > >>> > > > > > > > > > > >> > > > > >> > > is
> > >>> > > > > > > > > > > >> > > > > >> > > > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
> > >>> the topic
> > >>> > > > > > names is
> > >>> > > > > > > > > less
> > >>> > > > > > > > > > > >> > efficient.
> > >>> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> > >>> easier to
> > >>> > > > > > handle if
> > >>> > > > > > > > > we have
> > >>> > > > > > > > > > > >> > topic
> > >>> > > > > > > > > > > >> > > > > ids.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason
> > to
> > >>> still
> > >>> > > > let
> > >>> > > > > > the
> > >>> > > > > > > > > clients
> > >>> > > > > > > > > > > >> send
> > >>> > > > > > > > > > > >> > names
> > >>> > > > > > > > > > > >> > > > > >> is to
> > >>> > > > > > > > > > > >> > > > > >> > > keep
> > >>> > > > > > > > > > > >> > > > > >> > > > > the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> > names
> > >>> -> ids
> > >>> > > > on
> > >>> > > > > > the
> > >>> > > > > > > > > broker /
> > >>> > > > > > > > > > > >> > admin
> > >>> > > > > > > > > > > >> > > > > client
> > >>> > > > > > > > > > > >> > > > > >> > > only.
> > >>> > > > > > > > > > > >> > > > > >> > > > > Note
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> > >>> topic id
> > >>> > > > in
> > >>> > > > > > > > > KIP-516, we
> > >>> > > > > > > > > > > >> > never
> > >>> > > > > > > > > > > >> > > > > >> > > implemented the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > consumer/producers
> > >>> > > > leveraging
> > >>> > > > > > the
> > >>> > > > > > > > > related
> > >>> > > > > > > > > > > >> > newer
> > >>> > > > > > > > > > > >> > > > > >> versioned
> > >>> > > > > > > > > > > >> > > > > >> > > RPCs,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> > topic
> > >>> id as
> > >>> > > > empty
> > >>> > > > > > UUID.
> > >>> > > > > > > > > We
> > >>> > > > > > > > > > > >> want
> > >>> > > > > > > > > > > >> > to keep
> > >>> > > > > > > > > > > >> > > > > >> the
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and
> > only
> > >>> > > > delegate
> > >>> > > > > > the
> > >>> > > > > > > > > reasoning
> > >>> > > > > > > > > > > >> on
> > >>> > > > > > > > > > > >> > broker
> > >>> > > > > > > > > > > >> > > > > >> and
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
> > >>> 2022 at
> > >>> > > > 10:45
> > >>> > > > > > AM
> > >>> > > > > > > > > David
> > >>> > > > > > > > > > > >> Jacot
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> <dj...@confluent.io.invalid>
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like
> > >>> to start a
> > >>> > > > > > > > > discussion
> > >>> > > > > > > > > > > >> thread
> > >>> > > > > > > > > > > >> > on
> > >>> > > > > > > > > > > >> > > > > >> KIP-848:
> > >>> > > > > > > > > > > >> > > > > >> > > The
> > >>> > > > > > > > > > > >> > > > > >> > > > > Next
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of
> > >>> the
> > >>> > > > Consumer
> > >>> > > > > > > > > Rebalance
> > >>> > > > > > > > > > > >> > Protocol.
> > >>> > > > > > > > > > > >> > > > > >> With
> > >>> > > > > > > > > > > >> > > > > >> > > this
> > >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > we
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> > >>> rebalance
> > >>> > > > > > protocol
> > >>> > > > > > > > > (for
> > >>> > > > > > > > > > > >> > consumers)
> > >>> > > > > > > > > > > >> > > > > >> more
> > >>> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > more
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> > >>> easier to
> > >>> > > > > > implement for
> > >>> > > > > > > > > > > >> > clients, and
> > >>> > > > > > > > > > > >> > > > > >> easier
> > >>> > > > > > > > > > > >> > > > > >> > > to
> > >>> > > > > > > > > > > >> > > > > >> > > > > debug
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> > >>> here:
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
> > >>> look and
> > >>> > > > let
> > >>> > > > > > me
> > >>> > > > > > > > > know what
> > >>> > > > > > > > > > > >> > you
> > >>> > > > > > > > > > > >> > > > > think.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
> > >>> away from
> > >>> > > > > > July
> > >>> > > > > > > > > 18th to
> > >>> > > > > > > > > > > >> > August 8th.
> > >>> > > > > > > > > > > >> > > > > >> That
> > >>> > > > > > > > > > > >> > > > > >> > > gives
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> > >>> read and
> > >>> > > > digest
> > >>> > > > > > this
> > >>> > > > > > > > > long
> > >>> > > > > > > > > > > >> KIP.
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > >>> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > >>> > > > > > > > > > > >> > > > > >> > > > > > >
> > >>> > > > > > > > > > > >> > > > > >> > > > >
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> > >
> > >>> > > > > > > > > > > >> > > > > >> >
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >> --
> > >>> > > > > > > > > > > >> > > > > >> -- Guozhang
> > >>> > > > > > > > > > > >> > > > > >>
> > >>> > > > > > > > > > > >> > > > > >
> > >>> > > > > > > > > > > >> > > > >
> > >>> > > > > > > > > > > >> >
> > >>> > > > > > > > > > > >>
> > >>> > > > > > > > > > > >
> > >>> > > > > > > > >
> > >>> > > > > >
> > >>> > > > >
> > >>> > > > >
> > >>> > > > > --
> > >>> > > > > -- Guozhang
> > >>> > > >
> > >>> > >
> > >>> > >
> > >>> > > --
> > >>> > > -- Guozhang
> > >>>
> > >>
> > >>
> > >> --
> > >> -- Guozhang
> > >>
> > >
> > >
> > > --
> > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Jun Rao <ju...@confluent.io.INVALID>]

Hi, David,

Thanks for the reply. A few more comments.

40. Is there an API to change PartitionAssignor#metadata? I assume the
member fields are just for reads?

43.2 Sounds good.

44.1 Thanks. Could we document that the coordinator selects the assignor
when multiple assignors are specified in the client?

44.2 Encoding the group state in ConsumerGroupPrepareAssignmentResponse
sounds simpler and clearer to me.

48. Yes, that's the confusing part. I assume member assignment is what the
member client has. Could we include member assignment from both the
coordinator and the client's point of view?

50. We only include Uuid in PartitionAssignor.Group.TopicMetadata. How does
KStream do co-partitioning across topics?

51. PartitionAssignor.metadata: Is this really an assignor level thing? The
use case mentioned is to use metadata for encoding lags in standby tasks.
However, the lag is independent of the assignor. For example, if a member
sets two assignors, does it need to set the same metadata in both assignors?

52. "power users will have the ability to trigger a reassignment by either
providing a non-zero reason or by updating the assignor metadata." : Do we
need both reason and metadata to trigger an assignment? Could we just use
metadata?

53. MemberAssignment.metadata: How is that used? Does that overwrite the
member metadata on the coordinator?

Thanks,

Jun


On Tue, Sep 13, 2022 at 1:59 PM Guozhang Wang <wa...@gmail.com> wrote:

> Hi David,
>
> Another quick thing I just realized about "PartitionAssignor#onAssignment":
> in the new proposal, this function could be triggered multiple time within
> a group generation as it will be triggered whenever the broker gives the
> client new assignments, whereas today it is only triggered once per
> rebalance. I think it's worth emphasizing this a bit in the doc, especially
> around:
>
> 1) how many times `onAssignment` could be triggered with a single rebalance
> (especially if the new assignment is decided by the client-assignor).
> 2) when the Assignment metadata could change across multiple `onAssignment`
> calls, and how users should handle it.
> 3) the order of calling `onAssignment` and `onPartitionsRevoked/Assigned`.
>
> For example, today the Streams client relies on the following assumptions:
> 1) within a rebalance the onPartitionsRevoked would be called before the
> onAssignment, and onAssignment is called before onPartitionsAssigned (today
> we would trigger all three only once within a generation), 2) onAssignment
> contains the source-of-truth of the final assignment, and hence we can
> simply ignore onPartitionsAssigned and just rely on onAssignment to reason
> about which tasks are removed / assigned. And each time it calls, the
> corresponding metadata change are "final" for this new generation.
>
> We can change however Streams handle the new protocol, but I think there
> are other clients externally which may also make such assumptions about the
> ordering / call times about those callbacks, and hence worth clarifying
> those in the new proposal.
>
>
>
> Guozhang
>
>
>
>
> On Tue, Sep 13, 2022 at 12:20 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> > Hello David,
> >
> > Just had a few more nit questions about the public APIs:
> >
> > 1) The client-side "PartitionAssignor#Assignment" has an Error byte
> field,
> > while the broker-side "PartitionAssignor#Assignment" does not. And the
> > rationale seems to be assuming that we should always be able to do the
> > assignment at the broker-side assignor without errors. Personally I think
> > it's still potentially beneficial to add the Error field even for
> > broker-side assignors, e.g. for some edge cases where some subscribed
> > topics are not recognized with the current broker's metadata. What do you
> > think?
> >
> > 2) The client-side "GroupMember" has three additional fields
> > reason/version/metadata compared with the broker-side GroupMember. I
> agree
> > that broker-side assignor would not need reason/metadata since they are
> > blackbox strings/bytes to the assignor, but what about version? E.g. is
> it
> > possible that we evolve our broker-side built-in assignor but the old
> > versioned clients would not be able to work with the new version, in
> which
> > case we need to let the broker being aware of this and upgrade its
> behavior
> > to cooperate with the clients?
> >
> > 3) Also related to 2) above, for the client-side "GroupMember", instead
> of
> > including these three fields, what about just adding the "Metadata" field
> > class which has these three fields? Also, there are two "Metadata"
> > currently in the APIs, the first is a class that encodes
> > reason/version/metadata, and the second is just the encoded metadata
> bytes.
> > I'm wondering what about just naming the first as memberMetadata, which
> > then has a bytebuffer field Metadata, or instead naming the second
> > bytebuffer field as metadataBytes?
> >
> > Guozhang
> >
> >
> >
> >
> >
> >
> >
> > On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> >
> >> Hello David,
> >>
> >> Thanks for bringing this question up. I think the main benefits as you
> >> listed is 2) above if it stays; just to clarify, we would only be able
> to
> >> save one round trip if the rebalance is still triggered by the broker;
> if
> >> the rebalance is triggered by the client then the num.round.trips are
> the
> >> same:
> >>
> >> 1) With GroupPrepareAssignment:
> >>
> >> T0: client decides to do a new assignment, suppose it has already sent a
> >> HB and hence has to wait for it to return first since only one request /
> >> response can be inflight with the coordinator's socket.
> >> T1: client receives the HB response, and then sends the
> >> GroupPrepareAssignment request.
> >> T2: the GroupPrepareAssignment response is returned.
> >> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> >> request.
> >>
> >> In total, two round trips.
> >>
> >> 2) Without GroupPrepareAssignment:
> >>
> >> T0: client decides to do a new assignment, suppose it has already sent a
> >> HB and hence has to wait for it to return first since only one request /
> >> response can be inflight with the coordinator's socket.
> >> T1: client receives the HB response, and then sends the new HB request
> >> with the flag indicating a new rebalance needed..
> >> T2: the HB response with the optional member metadata map is returned.
> >> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> >> request.
> >>
> >> In total, two round trips as well.
> >>
> >> -----------------------------
> >>
> >> So to complete the full picture here, we'd need to modify both HB
> request
> >> and response so that the client can also indicate a new rebalance via
> the
> >> HB request as well, right?
> >>
> >> Assuming all above is true, I think it's okay to merge the
> >> GroupPrepareAssignment into HB given that we can make the additional
> fields
> >> encoding the full member (subscription) metadata and topic metadata as
> >> optional fields.
> >>
> >>
> >> Guozhang
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >>
> >> On Mon, Sep 12, 2022 at 5:22 AM David Jacot <djacot@confluent.io.invalid
> >
> >> wrote:
> >>
> >>> Hi all,
> >>>
> >>> During an offline conversation, someone asked why we need the
> >>> ConsumerGroupPrepareAssignment API and suggested that we could instead
> >>> provide the group state in the heartbeat response. This has a few
> >>> advantages: 1) it does not require using a special error code to
> >>> signal that a new assignment is required as the signal would be the
> >>> provided group state; 2) it removes one round trip when a client side
> >>> assignor is used. The downside is that it makes the heartbeat
> >>> response's definition quite large. I recall that I went with the
> >>> current approach due to this.
> >>>
> >>> Providing the group state in the heartbeat response is appealing. What
> >>> do you guys think?
> >>>
> >>> Best,
> >>> David
> >>>
> >>> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io>
> wrote:
> >>> >
> >>> > Hi Guozhang,
> >>> >
> >>> > 1. I have added a reference to the relevant chapter instead of
> >>> > repeating the whole thing. Does that work for you?
> >>> >
> >>> > 2. The "Rebalance Triggers" section you are referring to is about
> when
> >>> > a rebalance should be triggered for the non-upgraded members using
> the
> >>> > old protocol. The section mentions that a rebalance must be triggered
> >>> > when a new assignment is installed. This implies that the group epoch
> >>> > was updated either by a native member or a non-upgraded member. For
> >>> > the latter, the JoinGroup request would be the trigger. I have added
> a
> >>> > reference to the relevant chapter in the "JoinGroup Handling" section
> >>> > as well. Does that make sense?
> >>> >
> >>> > Thanks,
> >>> > David
> >>> >
> >>> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> >>> wrote:
> >>> > >
> >>> > > Hello David,
> >>> > >
> >>> > > Alright I think that's sufficient. Just to make that clear in the
> >>> doc,
> >>> > > could we update:
> >>> > >
> >>> > > 1) the heartbeat request handling section, stating when coordinator
> >>> will
> >>> > > trigger rebalance based on the HB's member metadata / reason?
> >>> > > 2) the "Rebalance Triggers" section to include what we described in
> >>> "Group
> >>> > > Epoch - Trigger a rebalance" section as well?
> >>> > >
> >>> > >
> >>> > > Guozhang
> >>> > >
> >>> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
> >>> <dj...@confluent.io.invalid>
> >>> > > wrote:
> >>> > >
> >>> > > > Hi Guozhang,
> >>> > > >
> >>> > > > I thought that the assignor will always be consulted when the
> next
> >>> > > > heartbeat request is constructed. In other words,
> >>> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
> >>> This
> >>> > > > gives the opportunity for the assignor to enforce a rebalance by
> >>> > > > setting the reason to a non-zero value or by changing the bytes.
> Do
> >>> > > > you think that this is not sufficient? Are you concerned by the
> >>> delay?
> >>> > > >
> >>> > > > Best,
> >>> > > > David
> >>> > > >
> >>> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wangguoz@gmail.com
> >
> >>> wrote:
> >>> > > > >
> >>> > > > > Hello David,
> >>> > > > >
> >>> > > > > One of Jun's comments make me thinking:
> >>> > > > >
> >>> > > > > ```
> >>> > > > > In this case, a new assignment is triggered by the client side
> >>> > > > > assignor. When constructing the HB, the consumer will always
> >>> consult
> >>> > > > > the client side assignor and propagate the information to the
> >>> group
> >>> > > > > coordinator. In other words, we don't expect users to call
> >>> > > > > Consumer#enforceRebalance anymore.
> >>> > > > > ```
> >>> > > > >
> >>> > > > > As I looked at the current PartitionAssignor's interface, we
> >>> actually do
> >>> > > > > not have a way yet to instruct how to construct the next HB
> >>> request, e.g.
> >>> > > > > when the assignor wants to enforce a new rebalance with a new
> >>> assignment,
> >>> > > > > we'd need some customizable APIs inside the PartitionAssignor
> to
> >>> indicate
> >>> > > > > the next HB telling broker about so. WDYT about adding such an
> >>> API on the
> >>> > > > > PartitionAssignor?
> >>> > > > >
> >>> > > > >
> >>> > > > > Guozhang
> >>> > > > >
> >>> > > > >
> >>> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> >>> <dj...@confluent.io.invalid>
> >>> > > > > wrote:
> >>> > > > >
> >>> > > > > > Hi Jun,
> >>> > > > > >
> >>> > > > > > I have updated the KIP to include your feedback. I have also
> >>> tried to
> >>> > > > > > clarify the parts which were not cleared.
> >>> > > > > >
> >>> > > > > > Best,
> >>> > > > > > David
> >>> > > > > >
> >>> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
> >>> djacot@confluent.io>
> >>> > > > wrote:
> >>> > > > > > >
> >>> > > > > > > Hi Jun,
> >>> > > > > > >
> >>> > > > > > > Thanks for your feedback. Let me start by answering your
> >>> questions
> >>> > > > > > > inline and I will update the KIP next week.
> >>> > > > > > >
> >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> >>> seem to
> >>> > > > be
> >>> > > > > > fewer
> >>> > > > > > > > RPCs during rebalance and more efficient support of
> >>> wildcard. A few
> >>> > > > > > > > comments below.
> >>> > > > > > >
> >>> > > > > > > I would also add that the KIP removes the global sync
> >>> barrier in the
> >>> > > > > > > protocol which is essential to improve group stability and
> >>> > > > > > > scalability, and the KIP also simplifies the client by
> >>> moving most of
> >>> > > > > > > the logic to the server side.
> >>> > > > > > >
> >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> >>> rolling
> >>> > > > > > changing
> >>> > > > > > > > of the partition assignor in the consumers?
> >>> > > > > > >
> >>> > > > > > > Definitely. The group coordinator will use the assignor
> used
> >>> by a
> >>> > > > > > > majority of the members. This allows the group to move from
> >>> one
> >>> > > > > > > assignor to another by a roll. This is explained in the
> >>> Assignor
> >>> > > > > > > Selection chapter.
> >>> > > > > > >
> >>> > > > > > > > 30.2 For each field, could you explain whether it's
> >>> required in
> >>> > > > every
> >>> > > > > > > > request or the scenarios when it needs to be filled? For
> >>> example,
> >>> > > > it's
> >>> > > > > > not
> >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> >>> > > > > > >
> >>> > > > > > > The client is expected to set those fields in case of a
> >>> connection
> >>> > > > > > > issue (e.g. timeout) or when the fields have changed since
> >>> the last
> >>> > > > > > > HB. The server populates those fields as long as the member
> >>> is not
> >>> > > > > > > fully reconciled - the member should acknowledge that it
> has
> >>> the
> >>> > > > > > > expected epoch and assignment. I will clarify this in the
> >>> KIP.
> >>> > > > > > >
> >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> >>> between the
> >>> > > > > > client
> >>> > > > > > > > and the broker is only considered during fetching, but
> not
> >>> during
> >>> > > > > > assigning
> >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> is
> >>> made,
> >>> > > > there
> >>> > > > > > is
> >>> > > > > > > > no opportunity for read affinity because no replicas of
> >>> assigned
> >>> > > > > > partitions
> >>> > > > > > > > are close to the member. I am wondering if we should use
> >>> this
> >>> > > > > > opportunity
> >>> > > > > > > > to address this by including rack in GroupMember.
> >>> > > > > > >
> >>> > > > > > > That's an interesting idea. I don't see any issue with
> >>> adding the
> >>> > > > rack
> >>> > > > > > > to the members. I will do so.
> >>> > > > > > >
> >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> >>> busy a group
> >>> > > > > > > > coordinator is. By moving the event loop model, it seems
> >>> that we
> >>> > > > could
> >>> > > > > > add
> >>> > > > > > > > a metric that tracks the fraction of the time the event
> >>> loop is
> >>> > > > doing
> >>> > > > > > the
> >>> > > > > > > > actual work.
> >>> > > > > > >
> >>> > > > > > > That's a great idea. I will add it. Thanks.
> >>> > > > > > >
> >>> > > > > > > > 33. Could we add a section on coordinator failover
> >>> handling? For
> >>> > > > > > example,
> >>> > > > > > > > does it need to trigger the check if any group with the
> >>> wildcard
> >>> > > > > > > > subscription now has a new matching topic?
> >>> > > > > > >
> >>> > > > > > > Sure. When the new group coordinator takes over, it has to:
> >>> > > > > > > * Setup the session timeouts.
> >>> > > > > > > * Trigger a new assignment if a client side assignor is
> >>> used. We
> >>> > > > don't
> >>> > > > > > > store the information about the member selected to run the
> >>> assignment
> >>> > > > > > > so we have to start a new one.
> >>> > > > > > > * Update the topics metadata, verify the wildcard
> >>> subscriptions, and
> >>> > > > > > > trigger a rebalance if needed.
> >>> > > > > > >
> >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> >>> > > > ConsumerGroupPartitionMetadataValue,
> >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> >>> the epoch
> >>> > > > > > field
> >>> > > > > > > > reflects? For example, does the epoch in
> >>> ConsumerGroupMetadataValue
> >>> > > > > > reflect
> >>> > > > > > > > the latest group epoch? What about the one in
> >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> >>> > > > > > ConsumerGroupMemberMetadataValue?
> >>> > > > > > >
> >>> > > > > > > Sure. I will clarify that but it is always the latest group
> >>> epoch.
> >>> > > > > > > When the group state is updated, the group epoch is bumped
> >>> so we use
> >>> > > > > > > that one for all the change records related to the update.
> >>> > > > > > >
> >>> > > > > > > > 35. "the group coordinator will ensure that the following
> >>> > > > invariants
> >>> > > > > > are
> >>> > > > > > > > met: ... All members exists." It's possible for a member
> >>> not to
> >>> > > > get any
> >>> > > > > > > > assigned partitions, right?
> >>> > > > > > >
> >>> > > > > > > That's right. Here I meant that the members provided by the
> >>> assignor
> >>> > > > > > > in the assignment must exist in the group. The assignor can
> >>> not make
> >>> > > > > > > up new member ids.
> >>> > > > > > >
> >>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
> >>> to 0":
> >>> > > > When
> >>> > > > > > would
> >>> > > > > > > > a consumer rejoin and what member id would be used?
> >>> > > > > > >
> >>> > > > > > > A member is expected to abandon all its partitions and
> >>> rejoins when
> >>> > > > it
> >>> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
> >>> group
> >>> > > > > > > coordinator will have removed the member from the group.
> The
> >>> member
> >>> > > > > > > can rejoin the group with the same member id but with 0 as
> >>> epoch. Let
> >>> > > > > > > me see if I can clarify this in the KIP.
> >>> > > > > > >
> >>> > > > > > > > 37. "Instead, power users will have the ability to
> trigger
> >>> a
> >>> > > > > > reassignment
> >>> > > > > > > > by either providing a non-zero reason or by updating the
> >>> assignor
> >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> >>> deprecation
> >>> > > > of
> >>> > > > > > > > Consumer#enforeRebalance.
> >>> > > > > > >
> >>> > > > > > > In this case, a new assignment is triggered by the client
> >>> side
> >>> > > > > > > assignor. When constructing the HB, the consumer will
> always
> >>> consult
> >>> > > > > > > the client side assignor and propagate the information to
> >>> the group
> >>> > > > > > > coordinator. In other words, we don't expect users to call
> >>> > > > > > > Consumer#enforceRebalance anymore.
> >>> > > > > > >
> >>> > > > > > > > 38. The reassignment examples are nice. But the section
> >>> seems to
> >>> > > > have
> >>> > > > > > > > multiple typos.
> >>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> >>> gets into
> >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> get
> >>> into
> >>> > > > > > epoch=3,
> >>> > > > > > > > partitions=[foo-1] too early.
> >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> >>> epoch=2,
> >>> > > > > > > > partitions=[foo-0].
> >>> > > > > > >
> >>> > > > > > > Sorry for that! I will revise them.
> >>> > > > > > >
> >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> >>> upgrade from
> >>> > > > any
> >>> > > > > > old
> >>> > > > > > > > version to new one?
> >>> > > > > > >
> >>> > > > > > > We will support upgrading from the consumer protocol
> version
> >>> 3,
> >>> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so
> the
> >>> earliest
> >>> > > > > > > version is unknown at the moment. This is explained in the
> >>> migration
> >>> > > > > > > plan chapter.
> >>> > > > > > >
> >>> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> >>> based on
> >>> > > > it
> >>> > > > > > > next week.
> >>> > > > > > >
> >>> > > > > > > Best,
> >>> > > > > > > David
> >>> > > > > > >
> >>> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> >>> <ju...@confluent.io.invalid>
> >>> > > > wrote:
> >>> > > > > > > >
> >>> > > > > > > > Hi, David,
> >>> > > > > > > >
> >>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> >>> seem to
> >>> > > > be
> >>> > > > > > fewer
> >>> > > > > > > > RPCs during rebalance and more efficient support of
> >>> wildcard. A few
> >>> > > > > > > > comments below.
> >>> > > > > > > >
> >>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> >>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> >>> rolling
> >>> > > > > > changing
> >>> > > > > > > > of the partition assignor in the consumers?
> >>> > > > > > > > 30.2 For each field, could you explain whether it's
> >>> required in
> >>> > > > every
> >>> > > > > > > > request or the scenarios when it needs to be filled? For
> >>> example,
> >>> > > > it's
> >>> > > > > > not
> >>> > > > > > > > clear to me when TopicPartitions needs to be filled.
> >>> > > > > > > >
> >>> > > > > > > > 31. In the current consumer protocol, the rack affinity
> >>> between the
> >>> > > > > > client
> >>> > > > > > > > and the broker is only considered during fetching, but
> not
> >>> during
> >>> > > > > > assigning
> >>> > > > > > > > partitions to consumers. Sometimes, once the assignment
> is
> >>> made,
> >>> > > > there
> >>> > > > > > is
> >>> > > > > > > > no opportunity for read affinity because no replicas of
> >>> assigned
> >>> > > > > > partitions
> >>> > > > > > > > are close to the member. I am wondering if we should use
> >>> this
> >>> > > > > > opportunity
> >>> > > > > > > > to address this by including rack in GroupMember.
> >>> > > > > > > >
> >>> > > > > > > > 32. On the metric side, often, it's useful to know how
> >>> busy a group
> >>> > > > > > > > coordinator is. By moving the event loop model, it seems
> >>> that we
> >>> > > > could
> >>> > > > > > add
> >>> > > > > > > > a metric that tracks the fraction of the time the event
> >>> loop is
> >>> > > > doing
> >>> > > > > > the
> >>> > > > > > > > actual work.
> >>> > > > > > > >
> >>> > > > > > > > 33. Could we add a section on coordinator failover
> >>> handling? For
> >>> > > > > > example,
> >>> > > > > > > > does it need to trigger the check if any group with the
> >>> wildcard
> >>> > > > > > > > subscription now has a new matching topic?
> >>> > > > > > > >
> >>> > > > > > > > 34. ConsumerGroupMetadataValue,
> >>> > > > ConsumerGroupPartitionMetadataValue,
> >>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
> >>> the epoch
> >>> > > > > > field
> >>> > > > > > > > reflects? For example, does the epoch in
> >>> ConsumerGroupMetadataValue
> >>> > > > > > reflect
> >>> > > > > > > > the latest group epoch? What about the one in
> >>> > > > > > > > ConsumerGroupPartitionMetadataValue and
> >>> > > > > > ConsumerGroupMemberMetadataValue?
> >>> > > > > > > >
> >>> > > > > > > > 35. "the group coordinator will ensure that the following
> >>> > > > invariants
> >>> > > > > > are
> >>> > > > > > > > met: ... All members exists." It's possible for a member
> >>> not to
> >>> > > > get any
> >>> > > > > > > > assigned partitions, right?
> >>> > > > > > > >
> >>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
> >>> to 0":
> >>> > > > When
> >>> > > > > > would
> >>> > > > > > > > a consumer rejoin and what member id would be used?
> >>> > > > > > > >
> >>> > > > > > > > 37. "Instead, power users will have the ability to
> trigger
> >>> a
> >>> > > > > > reassignment
> >>> > > > > > > > by either providing a non-zero reason or by updating the
> >>> assignor
> >>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> >>> deprecation
> >>> > > > of
> >>> > > > > > > > Consumer#enforeRebalance.
> >>> > > > > > > >
> >>> > > > > > > > 38. The reassignment examples are nice. But the section
> >>> seems to
> >>> > > > have
> >>> > > > > > > > multiple typos.
> >>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> >>> gets into
> >>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> >>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to
> get
> >>> into
> >>> > > > > > epoch=3,
> >>> > > > > > > > partitions=[foo-1] too early.
> >>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
> >>> epoch=2,
> >>> > > > > > > > partitions=[foo-0].
> >>> > > > > > > >
> >>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
> >>> upgrade from
> >>> > > > any
> >>> > > > > > old
> >>> > > > > > > > version to new one?
> >>> > > > > > > >
> >>> > > > > > > > Thanks,
> >>> > > > > > > >
> >>> > > > > > > > Jun
> >>> > > > > > > >
> >>> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> >>> > > > > > <dj...@confluent.io.invalid>
> >>> > > > > > > > wrote:
> >>> > > > > > > >
> >>> > > > > > > > > Hi all,
> >>> > > > > > > > >
> >>> > > > > > > > > The KIP states that we will re-implement the
> coordinator
> >>> in
> >>> > > > Java. I
> >>> > > > > > > > > discussed this offline with a few folks and folks are
> >>> concerned
> >>> > > > that
> >>> > > > > > > > > we could introduce many regressions in the old protocol
> >>> if we do
> >>> > > > so.
> >>> > > > > > > > > Therefore, I am going to remove this statement from the
> >>> KIP. It
> >>> > > > is an
> >>> > > > > > > > > implementation detail after all so it does not have to
> be
> >>> > > > decided at
> >>> > > > > > > > > this stage. We will likely start by trying to refactor
> >>> the
> >>> > > > current
> >>> > > > > > > > > implementation as a first step.
> >>> > > > > > > > >
> >>> > > > > > > > > Cheers,
> >>> > > > > > > > > David
> >>> > > > > > > > >
> >>> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> >>> djacot@confluent.io
> >>> > > > >
> >>> > > > > > wrote:
> >>> > > > > > > > > >
> >>> > > > > > > > > > Hi Luke,
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> >>> assigning,
> >>> > > > > > reconciling,
> >>> > > > > > > > > stable,
> >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> >>> right?
> >>> > > > > > > > > >
> >>> > > > > > > > > > This sentence does not refer to those group states
> but
> >>> rather
> >>> > > > to a
> >>> > > > > > > > > > state machine replication (SMR). This refers to the
> >>> entire
> >>> > > > state of
> >>> > > > > > > > > > group coordinator which is replicated via the log
> >>> layer. I will
> >>> > > > > > > > > > clarify this in the KIP.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> >>> modelled as an
> >>> > > > event
> >>> > > > > > > > > loop"?
> >>> > > > > > > > > >
> >>> > > > > > > > > > The idea is to follow a model similar to the new
> quorum
> >>> > > > > > controller. We
> >>> > > > > > > > > > will have N threads to process events. Each
> >>> __consumer_offsets
> >>> > > > > > > > > > partition is assigned to a unique thread and all the
> >>> events
> >>> > > > (e.g.
> >>> > > > > > > > > > requests, callbacks, etc.) are processed by this
> >>> thread. This
> >>> > > > > > simplify
> >>> > > > > > > > > > concurrency and will enable us to do simulation
> >>> testing for the
> >>> > > > > > group
> >>> > > > > > > > > > coordinator.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> >>> *__consumer_offsets*
> >>> > > > > > > > > partitions?
> >>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> >>> group
> >>> > > > > > coordinator?
> >>> > > > > > > > > For
> >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> >>> groups, and
> >>> > > > > > both
> >>> > > > > > > > > exist in
> >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> machine
> >>> for it,
> >>> > > > or
> >>> > > > > > 2?
> >>> > > > > > > > > >
> >>> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> the
> >>> number
> >>> > > > of
> >>> > > > > > threads
> >>> > > > > > > > > used
> >>> > > > > > > > > > > to run the state machines. But I'm wondering if the
> >>> purpose
> >>> > > > of
> >>> > > > > > the
> >>> > > > > > > > > threads
> >>> > > > > > > > > > > is only to keep the state of each consumer group
> (or
> >>> > > > > > > > > *__consumer_offsets*
> >>> > > > > > > > > > > partitions?), and no heavy computation, why should
> >>> we need
> >>> > > > > > > > > multi-threads
> >>> > > > > > > > > > > here?
> >>> > > > > > > > > >
> >>> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> >>> > > > processing as
> >>> > > > > > the
> >>> > > > > > > > > > computation could be heavy.
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> >>> default
> >>> > > > session
> >>> > > > > > > > > timeout not
> >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> the
> >>> min/max
> >>> > > > > > session
> >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> >>> timeout to
> >>> > > > detect
> >>> > > > > > client
> >>> > > > > > > > > > > failures when using the consumer group protocol.
> >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> >>> minimum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> >>> maximum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > >
> >>> > > > > > > > > > This is indeed a mistake. The default session timeout
> >>> should
> >>> > > > be 45s
> >>> > > > > > > > > > (the current default).
> >>> > > > > > > > > >
> >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> >>> uniform],
> >>> > > > > > which means
> >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> >>> why not
> >>> > > > > > uniform
> >>> > > > > > > > > one? I
> >>> > > > > > > > > > > thought usually users will choose uniform assignor
> >>> (former
> >>> > > > sticky
> >>> > > > > > > > > assinor)
> >>> > > > > > > > > > > for better evenly distribution. Any other reason we
> >>> choose
> >>> > > > range
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > as default?
> >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> >>> server side
> >>> > > > > > assignors.
> >>> > > > > > > > > >
> >>> > > > > > > > > > The order on the server side has no influence because
> >>> the
> >>> > > > client
> >>> > > > > > must
> >>> > > > > > > > > > chose the selector that he wants to use. There is no
> >>> default
> >>> > > > in the
> >>> > > > > > > > > > current proposal. If the assignor is not specified by
> >>> the
> >>> > > > client,
> >>> > > > > > the
> >>> > > > > > > > > > request is rejected. The default client value for
> >>> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> >>> > > > > > > > > >
> >>> > > > > > > > > > Thanks for your very good comments, Luke. I hope that
> >>> my
> >>> > > > answers
> >>> > > > > > help
> >>> > > > > > > > > > to clarify things. I will update the KIP as well
> based
> >>> on your
> >>> > > > > > > > > > feedback.
> >>> > > > > > > > > >
> >>> > > > > > > > > > Cheers,
> >>> > > > > > > > > > David
> >>> > > > > > > > > >
> >>> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> >>> showuon@gmail.com>
> >>> > > > > > wrote:
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Hi David,
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Thanks for the update.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Some more questions:
> >>> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> >>> > > > > > > > > > > > The new group coordinator will have a state
> >>> machine per
> >>> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> >>> machine is
> >>> > > > > > modelled
> >>> > > > > > > > > as an
> >>> > > > > > > > > > > event loop. Those state machines will be executed
> in
> >>> > > > > > > > > > > *group.coordinator.threads* threads.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
> >>> assigning,
> >>> > > > > > reconciling,
> >>> > > > > > > > > stable,
> >>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> >>> right?
> >>> > > > > > > > > > > 1.2. What do you mean "each state machine is
> >>> modelled as an
> >>> > > > event
> >>> > > > > > > > > loop"?
> >>> > > > > > > > > > > 1.3. Why do we need a state machine per
> >>> *__consumer_offsets*
> >>> > > > > > > > > partitions?
> >>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> >>> group
> >>> > > > > > coordinator?
> >>> > > > > > > > > For
> >>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> >>> groups, and
> >>> > > > > > both
> >>> > > > > > > > > exist in
> >>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state
> machine
> >>> for it,
> >>> > > > or
> >>> > > > > > 2?
> >>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is
> the
> >>> number
> >>> > > > of
> >>> > > > > > threads
> >>> > > > > > > > > used
> >>> > > > > > > > > > > to run the state machines. But I'm wondering if the
> >>> purpose
> >>> > > > of
> >>> > > > > > the
> >>> > > > > > > > > threads
> >>> > > > > > > > > > > is only to keep the state of each consumer group
> (or
> >>> > > > > > > > > *__consumer_offsets*
> >>> > > > > > > > > > > partitions?), and no heavy computation, why should
> >>> we need
> >>> > > > > > > > > multi-threads
> >>> > > > > > > > > > > here?
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > 2. For the default value in the new configs:
> >>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
> >>> default
> >>> > > > session
> >>> > > > > > > > > timeout not
> >>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought
> the
> >>> min/max
> >>> > > > > > session
> >>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
> >>> timeout to
> >>> > > > detect
> >>> > > > > > client
> >>> > > > > > > > > > > failures when using the consumer group protocol.
> >>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> >>> minimum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> >>> maximum
> >>> > > > > > session
> >>> > > > > > > > > timeout.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > 2.2. The default server side assignor are [range,
> >>> uniform],
> >>> > > > > > which means
> >>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> >>> why not
> >>> > > > > > uniform
> >>> > > > > > > > > one? I
> >>> > > > > > > > > > > thought usually users will choose uniform assignor
> >>> (former
> >>> > > > sticky
> >>> > > > > > > > > assinor)
> >>> > > > > > > > > > > for better evenly distribution. Any other reason we
> >>> choose
> >>> > > > range
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > as default?
> >>> > > > > > > > > > > group.consumer.assignors List range, uniform The
> >>> server side
> >>> > > > > > assignors.
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > Thank you.
> >>> > > > > > > > > > > Luke
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> >>> showuon@gmail.com
> >>> > > > >
> >>> > > > > > wrote:
> >>> > > > > > > > > > >
> >>> > > > > > > > > > > > Hi Sagar,
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > I have some thoughts about Kafka Connect
> >>> integrating with
> >>> > > > > > KIP-848,
> >>> > > > > > > > > but I
> >>> > > > > > > > > > > > think we should have a separate discussion thread
> >>> for the
> >>> > > > Kafka
> >>> > > > > > > > > Connect
> >>> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> >>> Rebalance
> >>> > > > > > Protocol
> >>> > > > > > > > > [1],
> >>> > > > > > > > > > > > and let this discussion thread focus on consumer
> >>> rebalance
> >>> > > > > > protocol,
> >>> > > > > > > > > WDYT?
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > [1]
> >>> > > > > > > > > > > >
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > >
> >>>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > Thank you.
> >>> > > > > > > > > > > > Luke
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> >>> > > > > > sagarmeansocean@gmail.com>
> >>> > > > > > > > > wrote:
> >>> > > > > > > > > > > >
> >>> > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
> >>> like
> >>> > > > there's
> >>> > > > > > > > > agreement on
> >>> > > > > > > > > > > >> using separate APIs for Connect. I would revisit
> >>> the doc
> >>> > > > and
> >>> > > > > > see
> >>> > > > > > > > > what
> >>> > > > > > > > > > > >> changes are to be made.
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >> Thanks!
> >>> > > > > > > > > > > >> Sagar.
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> >>> > > > > > > > > <dj...@confluent.io.invalid>
> >>> > > > > > > > > > > >> wrote:
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >> > Hi Sagar,
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > Thanks for the feedback and the document.
> >>> That's really
> >>> > > > > > helpful. I
> >>> > > > > > > > > > > >> > will take a look at it.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > Overall, it seems to me that both Connect and
> >>> the
> >>> > > > Consumer
> >>> > > > > > could
> >>> > > > > > > > > share
> >>> > > > > > > > > > > >> > the same underlying "engine". The main
> >>> difference is
> >>> > > > that
> >>> > > > > > the
> >>> > > > > > > > > Consumer
> >>> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> >>> Connect
> >>> > > > assigns
> >>> > > > > > tasks
> >>> > > > > > > > > to
> >>> > > > > > > > > > > >> > workers. I see two ways to move forward:
> >>> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> >>> different
> >>> > > > > > resource
> >>> > > > > > > > > types
> >>> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> >>> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> >>> dedicated
> >>> > > > APIs
> >>> > > > > > > > > would be
> >>> > > > > > > > > > > >> > similar to the new ones but different on the
> >>> > > > > > content/resources and
> >>> > > > > > > > > > > >> > they would rely on the same engine on the
> >>> coordinator
> >>> > > > side.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > I personally lean towards 2) because I am not
> a
> >>> fan of
> >>> > > > > > > > > overcharging
> >>> > > > > > > > > > > >> > APIs to serve different purposes. That being
> >>> said, I am
> >>> > > > not
> >>> > > > > > > > > opposed to
> >>> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > I think that we can continue to discuss it
> here
> >>> for now
> >>> > > > in
> >>> > > > > > order
> >>> > > > > > > > > to
> >>> > > > > > > > > > > >> > ensure that this KIP is compatible with what
> we
> >>> will do
> >>> > > > for
> >>> > > > > > > > > Connect in
> >>> > > > > > > > > > > >> > the future.
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > Best,
> >>> > > > > > > > > > > >> > David
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> >>> > > > > > djacot@confluent.io>
> >>> > > > > > > > > wrote:
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > Hi all,
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > I am back from vacation. I will go through
> >>> and address
> >>> > > > > > your
> >>> > > > > > > > > comments
> >>> > > > > > > > > > > >> > > in the coming days. Thanks for your
> feedback.
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > Cheers,
> >>> > > > > > > > > > > >> > > David
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
> >>> Harris <
> >>> > > > > > > > > gharris1727@gmail.com
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > Hey All!
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> >>> > > > cooperative
> >>> > > > > > > > > rebalancing
> >>> > > > > > > > > > > >> > making it
> >>> > > > > > > > > > > >> > > > down the stack!
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > I had a few questions:
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> >>> describes how
> >>> > > > > > member
> >>> > > > > > > > > epoch
> >>> > > > > > > > > > > >> > should
> >>> > > > > > > > > > > >> > > > advance in step with the group epoch and
> >>> assignment
> >>> > > > > > epoch
> >>> > > > > > > > > values. I
> >>> > > > > > > > > > > >> > think
> >>> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> >>> described
> >>> > > > in
> >>> > > > > > the
> >>> > > > > > > > > KIP. When
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > protocol is incrementally assigning
> >>> partitions to a
> >>> > > > > > worker,
> >>> > > > > > > > > what
> >>> > > > > > > > > > > >> member
> >>> > > > > > > > > > > >> > > > epoch does each incremental assignment
> use?
> >>> Are
> >>> > > > member
> >>> > > > > > epochs
> >>> > > > > > > > > > > >> re-used,
> >>> > > > > > > > > > > >> > and
> >>> > > > > > > > > > > >> > > > a single member epoch can correspond to
> >>> multiple
> >>> > > > > > different
> >>> > > > > > > > > > > >> > (monotonically
> >>> > > > > > > > > > > >> > > > larger) assignments?
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque
> >>> to the
> >>> > > > group
> >>> > > > > > > > > > > >> coordinator? If
> >>> > > > > > > > > > > >> > > > not, should custom client-side assignor
> >>> > > > implementations
> >>> > > > > > > > > interact
> >>> > > > > > > > > > > >> with
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > Reason field, and how is its common
> meaning
> >>> agreed
> >>> > > > > > upon? If
> >>> > > > > > > > > so, what
> >>> > > > > > > > > > > >> > is the
> >>> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> >>> including
> >>> > > > such
> >>> > > > > > > > > functionality
> >>> > > > > > > > > > > >> > in the
> >>> > > > > > > > > > > >> > > > opaque metadata?
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> >>> "Thanks to
> >>> > > > > > this, the
> >>> > > > > > > > > input
> >>> > > > > > > > > > > >> of
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > client side assignor is entirely driven by
> >>> the group
> >>> > > > > > > > > coordinator.
> >>> > > > > > > > > > > >> The
> >>> > > > > > > > > > > >> > > > consumer is no longer responsible for
> >>> maintaining
> >>> > > > any
> >>> > > > > > state
> >>> > > > > > > > > besides
> >>> > > > > > > > > > > >> its
> >>> > > > > > > > > > > >> > > > assigned partitions." Does this mean that
> >>> the
> >>> > > > > > client-side
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > >> MAY
> >>> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> >>> (such as
> >>> > > > > > partition
> >>> > > > > > > > > > > >> > throughput,
> >>> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc),
> or
> >>> that
> >>> > > > > > additional
> >>> > > > > > > > > > > >> > non-Metadata
> >>> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> >>> > > > > > > > > > > >> > > > for
> >>> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> >>> > > > > > > > > > > >> > > > and
> >>> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> >>> > > > > > that
> >>> > > > > > > > > seem to
> >>> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> >>> these two
> >>> > > > > > > > > implementations
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > unified? This would serve to promote
> >>> feature parity
> >>> > > > of
> >>> > > > > > > > > server-side
> >>> > > > > > > > > > > >> and
> >>> > > > > > > > > > > >> > > > client-side assignors, and would also
> >>> facilitate
> >>> > > > > > operational
> >>> > > > > > > > > > > >> > flexibility in
> >>> > > > > > > > > > > >> > > > certain situations. For example, if a
> >>> server-side
> >>> > > > > > assignor
> >>> > > > > > > > > has some
> >>> > > > > > > > > > > >> > poor
> >>> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> >>> patched
> >>> > > > > > assignor to
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > client
> >>> > > > > > > > > > > >> > > > and switching one consumer group to a
> >>> client-side
> >>> > > > > > assignor
> >>> > > > > > > > > may be
> >>> > > > > > > > > > > >> > faster
> >>> > > > > > > > > > > >> > > > and less risky than patching all of the
> >>> brokers.
> >>> > > > With
> >>> > > > > > the
> >>> > > > > > > > > currently
> >>> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> >>> > > > reimplementation
> >>> > > > > > would
> >>> > > > > > > > > have
> >>> > > > > > > > > > > >> to be
> >>> > > > > > > > > > > >> > > > assembled, and if the two APIs have
> diverged
> >>> > > > > > significantly,
> >>> > > > > > > > > then it
> >>> > > > > > > > > > > >> is
> >>> > > > > > > > > > > >> > > > possible that a reimplementation would not
> >>> be
> >>> > > > possible.
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > --
> >>> > > > > > > > > > > >> > > > Greg Harris
> >>> > > > > > > > > > > >> > > > gharris1727@gmail.com
> >>> > > > > > > > > > > >> > > > github.com/gharris1727
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> >>> > > > > > > > > sagarmeansocean@gmail.com>
> >>> > > > > > > > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > I created a confluence page to discuss
> >>> how Connect
> >>> > > > > > would
> >>> > > > > > > > > need to
> >>> > > > > > > > > > > >> > change
> >>> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
> >>> Here's the
> >>> > > > page:
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >>
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > >
> >>>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > It's also pretty longish and I have
> tried
> >>> to keep
> >>> > > > a
> >>> > > > > > format
> >>> > > > > > > > > > > >> similar to
> >>> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
> >>> Also, do you
> >>> > > > > > think this
> >>> > > > > > > > > > > >> should
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > > moved to a separate discussion thread or
> >>> is this
> >>> > > > one
> >>> > > > > > fine?
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > Thanks!
> >>> > > > > > > > > > > >> > > > > Sagar.
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> >>> > > > > > > > > sagarmeansocean@gmail.com>
> >>> > > > > > > > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > > Hello Guozhang,
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> >>> Streams.
> >>> > > > > > Sure, I
> >>> > > > > > > > > would do
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > > analysis and come up with such a
> >>> document.
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > > Thanks!
> >>> > > > > > > > > > > >> > > > > > Sagar.
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
> >>> Guozhang Wang <
> >>> > > > > > > > > > > >> wangguoz@gmail.com>
> >>> > > > > > > > > > > >> > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> Hello Sagar,
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> It would be great if you could come
> >>> back with
> >>> > > > some
> >>> > > > > > > > > analysis on
> >>> > > > > > > > > > > >> > how to
> >>> > > > > > > > > > > >> > > > > >> implement the Connect side
> integration
> >>> with
> >>> > > > the new
> >>> > > > > > > > > protocol;
> >>> > > > > > > > > > > >> so
> >>> > > > > > > > > > > >> > far
> >>> > > > > > > > > > > >> > > > > >> besides leveraging on the new
> >>> "protocol type"
> >>> > > > we
> >>> > > > > > did not
> >>> > > > > > > > > yet
> >>> > > > > > > > > > > >> think
> >>> > > > > > > > > > > >> > > > > through
> >>> > > > > > > > > > > >> > > > > >> the Connect side implementations. For
> >>> Streams
> >>> > > > > > here's a
> >>> > > > > > > > > draft of
> >>> > > > > > > > > > > >> > > > > >> integration
> >>> > > > > > > > > > > >> > > > > >> plan:
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >>
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > >
> >>>
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >>> > > > > > > > > > > >> > > > > >> just FYI for your analysis on
> Connect.
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM
> Sagar
> >>> <
> >>> > > > > > > > > > > >> sagarmeansocean@gmail.com
> >>> > > > > > > > > > > >> > >
> >>> > > > > > > > > > > >> > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> > Hi David,
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> >>> reason I
> >>> > > > thought
> >>> > > > > > > > > connect can
> >>> > > > > > > > > > > >> > also fit
> >>> > > > > > > > > > > >> > > > > >> into
> >>> > > > > > > > > > > >> > > > > >> > this new scheme is that even today
> >>> the
> >>> > > > connect
> >>> > > > > > uses a
> >>> > > > > > > > > > > >> > > > > WorkerCoordinator
> >>> > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator
> >>> to empower
> >>> > > > > > > > > rebalances of
> >>> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> >>> WorkerCoordinator sets
> >>> > > > the
> >>> > > > > > > > > > > >> protocolType()
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> connect
> >>> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> >>> plumbing
> >>> > > > into
> >>> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > I think the changes to support
> >>> connect would
> >>> > > > be
> >>> > > > > > > > > similar at a
> >>> > > > > > > > > > > >> > high
> >>> > > > > > > > > > > >> > > > > level
> >>> > > > > > > > > > > >> > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
> >>> because of the
> >>> > > > > > Client
> >>> > > > > > > > > side
> >>> > > > > > > > > > > >> > assignors
> >>> > > > > > > > > > > >> > > > > being
> >>> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
> >>> level, we
> >>> > > > > > might
> >>> > > > > > > > > need to
> >>> > > > > > > > > > > >> make
> >>> > > > > > > > > > > >> > a lot
> >>> > > > > > > > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > changes to get onto this new
> >>> assignment
> >>> > > > protocol
> >>> > > > > > like
> >>> > > > > > > > > > > >> enhancing
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> >>> SyncGroup and
> >>> > > > > > using
> >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> >>> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> >>> streams (or
> >>> > > > > > there
> >>> > > > > > > > > might be
> >>> > > > > > > > > > > >> > > > > >> deviations). I
> >>> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> >>> analysis of
> >>> > > > the
> >>> > > > > > same
> >>> > > > > > > > > and we
> >>> > > > > > > > > > > >> > can have
> >>> > > > > > > > > > > >> > > > > a
> >>> > > > > > > > > > > >> > > > > >> > separate discussion thread for that
> >>> as that
> >>> > > > would
> >>> > > > > > > > > derail this
> >>> > > > > > > > > > > >> > > > > discussion
> >>> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
> >>> good to
> >>> > > > you.
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > Thanks!
> >>> > > > > > > > > > > >> > > > > >> > Sagar.
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
> >>> David Jacot
> >>> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > wrote:
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> >>> `Assignment#error`.
> >>> > > > > > Sure, I
> >>> > > > > > > > > can
> >>> > > > > > > > > > > >> > mention it.
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
> >>> from his
> >>> > > > current
> >>> > > > > > > > > assignment
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > his
> >>> > > > > > > > > > > >> > > > > >> > > target assignment when he can
> move
> >>> to
> >>> > > > epoch 3.
> >>> > > > > > When
> >>> > > > > > > > > that
> >>> > > > > > > > > > > >> > happens,
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
> >>> persisted
> >>> > > > > > with all
> >>> > > > > > > > > its
> >>> > > > > > > > > > > >> > assigned
> >>> > > > > > > > > > > >> > > > > >> > > partitions even if they are not
> >>> all revoked
> >>> > > > > > yet. In
> >>> > > > > > > > > other
> >>> > > > > > > > > > > >> > words, the
> >>> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
> >>> target
> >>> > > > > > assignment.
> >>> > > > > > > > > This is
> >>> > > > > > > > > > > >> > basically
> >>> > > > > > > > > > > >> > > > > an
> >>> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> >>> write all
> >>> > > > the
> >>> > > > > > > > > changes to
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > log.
> >>> > > > > > > > > > > >> > > > > The
> >>> > > > > > > > > > > >> > > > > >> > > examples are based on the
> >>> persisted state
> >>> > > > so I
> >>> > > > > > > > > understand
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> >>> improve
> >>> > > > this in
> >>> > > > > > the
> >>> > > > > > > > > > > >> > description.
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
> >>> reuse the
> >>> > > > > > protocol
> >>> > > > > > > > > with a
> >>> > > > > > > > > > > >> > client side
> >>> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
> >>> protocol. The
> >>> > > > > > assignment
> >>> > > > > > > > > is
> >>> > > > > > > > > > > >> about
> >>> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
> >>> could
> >>> > > > Connect
> >>> > > > > > fit
> >>> > > > > > > > > into this?
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > David
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> >>> Sagar <
> >>> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> >>> > > > > > > > > > > >> > > > > >> wrote:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Hi David,
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
> >>> minor
> >>> > > > > > observations:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
> >>> section in
> >>> > > > Client
> >>> > > > > > Side
> >>> > > > > > > > > mode
> >>> > > > > > > > > > > >> > Assignment
> >>> > > > > > > > > > > >> > > > > >> > > process,
> >>> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case,
> >>> the
> >>> > > > client
> >>> > > > > > side
> >>> > > > > > > > > assignor
> >>> > > > > > > > > > > >> can
> >>> > > > > > > > > > > >> > > > > return
> >>> > > > > > > > > > > >> > > > > >> an
> >>> > > > > > > > > > > >> > > > > >> > > > error to the group
> coordinator`.
> >>> In this
> >>> > > > > > case are
> >>> > > > > > > > > you
> >>> > > > > > > > > > > >> > referring to
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> >>> AssignmentError
> >>> > > > that's
> >>> > > > > > > > > listed down
> >>> > > > > > > > > > > >> > towards
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > end?
> >>> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
> >>> make sense
> >>> > > > to
> >>> > > > > > > > > mention this
> >>> > > > > > > > > > > >> > > > > explicitly
> >>> > > > > > > > > > > >> > > > > >> > here?
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section,
> >>> I have a
> >>> > > > > > slight
> >>> > > > > > > > > > > >> confusion,
> >>> > > > > > > > > > > >> > not
> >>> > > > > > > > > > > >> > > > > sure
> >>> > > > > > > > > > > >> > > > > >> if
> >>> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
> >>> this step:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> >>> coordinator
> >>> > > > > > > > > transitions him
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > epoch
> >>> > > > > > > > > > > >> > > > > 3
> >>> > > > > > > > > > > >> > > > > >> > > because
> >>> > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke.
> It
> >>> > > > persists
> >>> > > > > > the
> >>> > > > > > > > > change and
> >>> > > > > > > > > > > >> > reply.
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> >>> > > > > > > > > > > >> > > > > >> > > >       - A
> >>> > > > > > > > > > > >> > > > > >> > > >       - B
> >>> > > > > > > > > > > >> > > > > >> > > >       - C
> >>> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment
> (epoch=3)
> >>> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> >>> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> >>> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> >>> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> >>> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> >>> partitions=[foo-0,
> >>> > > > > > foo-1]
> >>> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> >>> partitions=[foo-2]
> >>> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> >>> partitions=[foo-1]
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
> >>> transitions to
> >>> > > > epoch 3
> >>> > > > > > but
> >>> > > > > > > > > cannot
> >>> > > > > > > > > > > >> get
> >>> > > > > > > > > > > >> > foo-1
> >>> > > > > > > > > > > >> > > > > >> yet.
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member
> >>> C can't
> >>> > > > get
> >>> > > > > > the
> >>> > > > > > > > > foo-1
> >>> > > > > > > > > > > >> > partition
> >>> > > > > > > > > > > >> > > > > yet,
> >>> > > > > > > > > > > >> > > > > >> > but
> >>> > > > > > > > > > > >> > > > > >> > > > based on the description above,
> >>> it seems
> >>> > > > it
> >>> > > > > > > > > already has
> >>> > > > > > > > > > > >> it.
> >>> > > > > > > > > > > >> > Do you
> >>> > > > > > > > > > > >> > > > > >> > think
> >>> > > > > > > > > > > >> > > > > >> > > it
> >>> > > > > > > > > > > >> > > > > >> > > > would be better to remove it
> and
> >>> > > > populate it
> >>> > > > > > only
> >>> > > > > > > > > when it
> >>> > > > > > > > > > > >> > actually
> >>> > > > > > > > > > > >> > > > > >> gets
> >>> > > > > > > > > > > >> > > > > >> > > it?
> >>> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> >>> places, so
> >>> > > > have
> >>> > > > > > I
> >>> > > > > > > > > > > >> understood it
> >>> > > > > > > > > > > >> > > > > >> > incorrectly
> >>> > > > > > > > > > > >> > > > > >> > > ?
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
> >>> out of
> >>> > > > scope
> >>> > > > > > of
> >>> > > > > > > > > this
> >>> > > > > > > > > > > >> > discussion,
> >>> > > > > > > > > > > >> > > > > but
> >>> > > > > > > > > > > >> > > > > >> > from
> >>> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> >>> probably be
> >>> > > > > > running in
> >>> > > > > > > > > client
> >>> > > > > > > > > > > >> > side
> >>> > > > > > > > > > > >> > > > > >> assignor
> >>> > > > > > > > > > > >> > > > > >> > > mode
> >>> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
> >>> protocol as it
> >>> > > > has
> >>> > > > > > its
> >>> > > > > > > > > own
> >>> > > > > > > > > > > >> Custom
> >>> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> >>> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Thanks!
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > Sagar.
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
> >>> David
> >>> > > > Jacot
> >>> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
> >>> move
> >>> > > > forward
> >>> > > > > > with
> >>> > > > > > > > > > > >> > specialized API
> >>> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> >>> generic API.
> >>> > > > For
> >>> > > > > > > > > Connect, we
> >>> > > > > > > > > > > >> > can apply
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> >>> reuse/share the
> >>> > > > core
> >>> > > > > > > > > > > >> > implementation on
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> >>> registry, I
> >>> > > > > > think
> >>> > > > > > > > > that we
> >>> > > > > > > > > > > >> > should
> >>> > > > > > > > > > > >> > > > > >> consider
> >>> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> >>> simple
> >>> > > > > > > > > membership/leader
> >>> > > > > > > > > > > >> > election.
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22
> >>> AM Ismael
> >>> > > > > > Juma <
> >>> > > > > > > > > > > >> > ismael@juma.me.uk
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> >>> java.util.regex.Pattern
> >>> > > > vs
> >>> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> >>> > > > > > > > > > > >> > > > > >> > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > document the differences in
> >>> more
> >>> > > > detail
> >>> > > > > > before
> >>> > > > > > > > > > > >> deciding
> >>> > > > > > > > > > > >> > one
> >>> > > > > > > > > > > >> > > > > way
> >>> > > > > > > > > > > >> > > > > >> or
> >>> > > > > > > > > > > >> > > > > >> > > > > another.
> >>> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> >>> > > > > > > > > java.util.regex.Pattern,
> >>> > > > > > > > > > > >> they
> >>> > > > > > > > > > > >> > expect
> >>> > > > > > > > > > > >> > > > > >> > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If
> >>> we are
> >>> > > > doing
> >>> > > > > > > > > something
> >>> > > > > > > > > > > >> > different,
> >>> > > > > > > > > > > >> > > > > >> then
> >>> > > > > > > > > > > >> > > > > >> > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> >>> overload
> >>> > > > with
> >>> > > > > > our own
> >>> > > > > > > > > > > >> Pattern
> >>> > > > > > > > > > > >> > class
> >>> > > > > > > > > > > >> > > > > (I
> >>> > > > > > > > > > > >> > > > > >> > > don't
> >>> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> >>> re2j's at
> >>> > > > this
> >>> > > > > > > > > point).
> >>> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> >>> major new
> >>> > > > > > protocol
> >>> > > > > > > > > should
> >>> > > > > > > > > > > >> > > > > integrate
> >>> > > > > > > > > > > >> > > > > >> > fully
> >>> > > > > > > > > > > >> > > > > >> > > > > with
> >>> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
> >>> topic
> >>> > > > > > recreation case
> >>> > > > > > > > > > > >> > correctly.
> >>> > > > > > > > > > > >> > > > > That's
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > main
> >>> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> >>> agree with
> >>> > > > > > David
> >>> > > > > > > > > that we'd
> >>> > > > > > > > > > > >> > want to
> >>> > > > > > > > > > > >> > > > > add
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > ids
> >>> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
> >>> that don't
> >>> > > > > > have it
> >>> > > > > > > > > yet and
> >>> > > > > > > > > > > >> > that we
> >>> > > > > > > > > > > >> > > > > can
> >>> > > > > > > > > > > >> > > > > >> > > probably
> >>> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
> >>> versus adding
> >>> > > > new
> >>> > > > > > APIs
> >>> > > > > > > > > to the
> >>> > > > > > > > > > > >> > Java
> >>> > > > > > > > > > > >> > > > > >> Consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > (unless
> >>> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new
> APIs
> >>> is
> >>> > > > required
> >>> > > > > > for
> >>> > > > > > > > > > > >> reasonable
> >>> > > > > > > > > > > >> > > > > >> semantics).
> >>> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure
> about
> >>> the
> >>> > > > > > coordinator
> >>> > > > > > > > > storing
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> configs.
> >>> > > > > > > > > > > >> > > > > >> > > It's
> >>> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> >>> > > > centralized
> >>> > > > > > in the
> >>> > > > > > > > > > > >> metadata
> >>> > > > > > > > > > > >> > log for
> >>> > > > > > > > > > > >> > > > > >> > > various
> >>> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> >>> visibility,
> >>> > > > > > consistency,
> >>> > > > > > > > > > > >> etc.).
> >>> > > > > > > > > > > >> > > > > >> Similarly, I
> >>> > > > > > > > > > > >> > > > > >> > > am
> >>> > > > > > > > > > > >> > > > > >> > > > > not
> >>> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> >>> deleting
> >>> > > > > > configs in a
> >>> > > > > > > > > way
> >>> > > > > > > > > > > >> that
> >>> > > > > > > > > > > >> > they
> >>> > > > > > > > > > > >> > > > > >> cannot
> >>> > > > > > > > > > > >> > > > > >> > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property
> >>> for modern
> >>> > > > > > systems
> >>> > > > > > > > > is to
> >>> > > > > > > > > > > >> > minimize
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > number
> >>> > > > > > > > > > > >> > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> >>> scenarios.
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at
> 3:47
> >>> PM David
> >>> > > > > > Jacot
> >>> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
> >>> answers are
> >>> > > > below:
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> >>> especially
> >>> > > > > > the last
> >>> > > > > > > > > step
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > > > > clients
> >>> > > > > > > > > > > >> > > > > >> > > flip the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> >>> protocol, in
> >>> > > > which
> >>> > > > > > we
> >>> > > > > > > > > would
> >>> > > > > > > > > > > >> have a
> >>> > > > > > > > > > > >> > > > > window
> >>> > > > > > > > > > > >> > > > > >> > where
> >>> > > > > > > > > > > >> > > > > >> > > > > both
> >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> old
> >>> > > > protocols /
> >>> > > > > > rpcs
> >>> > > > > > > > > are
> >>> > > > > > > > > > > >> used
> >>> > > > > > > > > > > >> > by
> >>> > > > > > > > > > > >> > > > > >> members
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> >>> coordinator could
> >>> > > > > > "mimic"
> >>> > > > > > > > > the old
> >>> > > > > > > > > > > >> > behavior
> >>> > > > > > > > > > > >> > > > > >> while
> >>> > > > > > > > > > > >> > > > > >> > > > > using the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> >>> something we
> >>> > > > need to
> >>> > > > > > > > > present
> >>> > > > > > > > > > > >> about.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a
> >>> new
> >>> > > > version
> >>> > > > > > of KIP
> >>> > > > > > > > > which
> >>> > > > > > > > > > > >> > includes
> >>> > > > > > > > > > > >> > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See
> the
> >>> > > > "Supporting
> >>> > > > > > > > > Online
> >>> > > > > > > > > > > >> > Consumer
> >>> > > > > > > > > > > >> > > > > Group
> >>> > > > > > > > > > > >> > > > > >> > > Upgrade"
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> >>> > > > Deprecation,
> >>> > > > > > and
> >>> > > > > > > > > Migration
> >>> > > > > > > > > > > >> > Plan". I
> >>> > > > > > > > > > > >> > > > > >> think
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
> >>> few cases
> >>> > > > > > now but
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > overall idea
> >>> > > > > > > > > > > >> > > > > >> and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> >>> understandable.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> >>> ids. So
> >>> > > > far as
> >>> > > > > > > > > KIP-516 the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > >> are
> >>> > > > > > > > > > > >> > > > > >> > > only
> >>> > > > > > > > > > > >> > > > > >> > > > > used
> >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> >>> client,
> >>> > > > but
> >>> > > > > > they
> >>> > > > > > > > > are not
> >>> > > > > > > > > > > >> > exposed
> >>> > > > > > > > > > > >> > > > > via
> >>> > > > > > > > > > > >> > > > > >> any
> >>> > > > > > > > > > > >> > > > > >> > > > > public
> >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> >>> think the
> >>> > > > > > question is,
> >>> > > > > > > > > first
> >>> > > > > > > > > > > >> > should we
> >>> > > > > > > > > > > >> > > > > >> let
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> maintaining
> >>> the
> >>> > > > names
> >>> > > > > > -> ids
> >>> > > > > > > > > mapping
> >>> > > > > > > > > > > >> > itself
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > fully
> >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> >>> existing
> >>> > > > RPCs
> >>> > > > > > and the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> RPCs
> >>> > > > > > > > > > > >> > with
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> ever
> >>> consider
> >>> > > > > > exposing
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> (both
> >>> > > > > > > > > subscribe/assign, as
> >>> > > > > > > > > > > >> well
> >>> > > > > > > > > > > >> > as in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> >>> topic
> >>> > > > > > > > > > > >> > deletion-and-recreation).
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> >>> include
> >>> > > > > > converting
> >>> > > > > > > > > all
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > offsets
> >>> > > > > > > > > > > >> > > > > >> > > related
> >>> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids
> in
> >>> this
> >>> > > > KIP,
> >>> > > > > > the
> >>> > > > > > > > > consumer
> >>> > > > > > > > > > > >> > would be
> >>> > > > > > > > > > > >> > > > > >> able
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
> >>> ids. That
> >>> > > > > > being
> >>> > > > > > > > > said, it
> >>> > > > > > > > > > > >> > still has
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > provide
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
> >>> various APIs so
> >>> > > > > > having a
> >>> > > > > > > > > > > >> mapping
> >>> > > > > > > > > > > >> > in the
> >>> > > > > > > > > > > >> > > > > >> > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> >>> opinion on
> >>> > > > > > this.
> >>> > > > > > > > > Here I
> >>> > > > > > > > > > > >> > wonder if
> >>> > > > > > > > > > > >> > > > > >> this
> >>> > > > > > > > > > > >> > > > > >> > > goes
> >>> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
> >>> KIP. I
> >>> > > > would
> >>> > > > > > rather
> >>> > > > > > > > > focus
> >>> > > > > > > > > > > >> on
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > internals
> >>> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider
> >>> this
> >>> > > > > > separately if
> >>> > > > > > > > > we see
> >>> > > > > > > > > > > >> > value in
> >>> > > > > > > > > > > >> > > > > >> doing
> >>> > > > > > > > > > > >> > > > > >> > > it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> >>> point about
> >>> > > > > > using
> >>> > > > > > > > > topic ids
> >>> > > > > > > > > > > >> > in the
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> ConsumerGroupHeartbeatRequest, I
> >>> > > > think
> >>> > > > > > that
> >>> > > > > > > > > there
> >>> > > > > > > > > > > >> is
> >>> > > > > > > > > > > >> > one
> >>> > > > > > > > > > > >> > > > > >> > advantage
> >>> > > > > > > > > > > >> > > > > >> > > in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The
> consumer
> >>> will
> >>> > > > have
> >>> > > > > > the
> >>> > > > > > > > > > > >> opportunity
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> validate
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> >>> passing
> >>> > > > them
> >>> > > > > > into
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> group
> >>> > > > > > > > > > > >> > > > > rebalance
> >>> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> >>> > > > coordinator
> >>> > > > > > will
> >>> > > > > > > > > also
> >>> > > > > > > > > > > >> notice
> >>> > > > > > > > > > > >> > it but
> >>> > > > > > > > > > > >> > > > > >> it
> >>> > > > > > > > > > > >> > > > > >> > > does
> >>> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> >>> reject an
> >>> > > > > > invalid
> >>> > > > > > > > > topic in
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> response.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> >>> on all
> >>> > > > other
> >>> > > > > > minor
> >>> > > > > > > > > > > >> questions
> >>> > > > > > > > > > > >> > > > > except
> >>> > > > > > > > > > > >> > > > > >> for
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> >>> question:
> >>> > > > > > personally
> >>> > > > > > > > > I think
> >>> > > > > > > > > > > >> > it's not
> >>> > > > > > > > > > > >> > > > > >> > > necessary
> >>> > > > > > > > > > > >> > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> >>> API with
> >>> > > > > > Pattern,
> >>> > > > > > > > > but
> >>> > > > > > > > > > > >> > instead we
> >>> > > > > > > > > > > >> > > > > >> still
> >>> > > > > > > > > > > >> > > > > >> > > use
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> >>> that our
> >>> > > > > > > > > subscription may
> >>> > > > > > > > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> rejected by
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> >>> case is a
> >>> > > > > > very rare
> >>> > > > > > > > > > > >> scenario
> >>> > > > > > > > > > > >> > I felt
> >>> > > > > > > > > > > >> > > > > >> > using
> >>> > > > > > > > > > > >> > > > > >> > > an
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> based
> >>> > > > > > subscription may
> >>> > > > > > > > > be
> >>> > > > > > > > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> vulnerable
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > > > various
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have
> to
> >>> look at
> >>> > > > the
> >>> > > > > > > > > differences
> >>> > > > > > > > > > > >> > between
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > two
> >>> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
> >>> understand the
> >>> > > > > > potential
> >>> > > > > > > > > issues.
> >>> > > > > > > > > > > >> My
> >>> > > > > > > > > > > >> > > > > >> > > understanding is
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all
> >>> the basic
> >>> > > > > > regular
> >>> > > > > > > > > > > >> > expressions. The
> >>> > > > > > > > > > > >> > > > > >> > > differences
> >>> > > > > > > > > > > >> > > > > >> > > > > > > between the two are
> mainly
> >>> about
> >>> > > > the
> >>> > > > > > various
> >>> > > > > > > > > > > >> character
> >>> > > > > > > > > > > >> > > > > >> classes. I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
> >>> think
> >>> > > > about
> >>> > > > > > this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> >>> 11:28 PM
> >>> > > > > > Guozhang
> >>> > > > > > > > > Wang <
> >>> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> >>> > > > > > > > > > > >> > > > > >> > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think
> on
> >>> the high
> >>> > > > > > level
> >>> > > > > > > > > there are
> >>> > > > > > > > > > > >> > two meta
> >>> > > > > > > > > > > >> > > > > >> > > points we
> >>> > > > > > > > > > > >> > > > > >> > > > > need
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit
> more:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> >>> especially
> >>> > > > > > the last
> >>> > > > > > > > > step
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > > > > clients
> >>> > > > > > > > > > > >> > > > > >> > > flip the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > flag
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> >>> protocol, in
> >>> > > > which
> >>> > > > > > we
> >>> > > > > > > > > would
> >>> > > > > > > > > > > >> have a
> >>> > > > > > > > > > > >> > > > > window
> >>> > > > > > > > > > > >> > > > > >> > where
> >>> > > > > > > > > > > >> > > > > >> > > > > both
> >>> > > > > > > > > > > >> > > > > >> > > > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and
> old
> >>> > > > protocols /
> >>> > > > > > rpcs
> >>> > > > > > > > > are
> >>> > > > > > > > > > > >> used
> >>> > > > > > > > > > > >> > by
> >>> > > > > > > > > > > >> > > > > >> members
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
> >>> coordinator could
> >>> > > > > > "mimic"
> >>> > > > > > > > > the old
> >>> > > > > > > > > > > >> > behavior
> >>> > > > > > > > > > > >> > > > > >> while
> >>> > > > > > > > > > > >> > > > > >> > > > > using the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
> >>> something we
> >>> > > > need to
> >>> > > > > > > > > present
> >>> > > > > > > > > > > >> about.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
> >>> ids. So
> >>> > > > far as
> >>> > > > > > > > > KIP-516 the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > >> are
> >>> > > > > > > > > > > >> > > > > >> > > only
> >>> > > > > > > > > > > >> > > > > >> > > > > used
> >>> > > > > > > > > > > >> > > > > >> > > > > > > as
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> >>> client,
> >>> > > > but
> >>> > > > > > they
> >>> > > > > > > > > are not
> >>> > > > > > > > > > > >> > exposed
> >>> > > > > > > > > > > >> > > > > via
> >>> > > > > > > > > > > >> > > > > >> any
> >>> > > > > > > > > > > >> > > > > >> > > > > public
> >>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
> >>> think the
> >>> > > > > > question is,
> >>> > > > > > > > > first
> >>> > > > > > > > > > > >> > should we
> >>> > > > > > > > > > > >> > > > > >> let
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be
> maintaining
> >>> the
> >>> > > > names
> >>> > > > > > -> ids
> >>> > > > > > > > > mapping
> >>> > > > > > > > > > > >> > itself
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > fully
> >>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> >>> existing
> >>> > > > RPCs
> >>> > > > > > and the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> RPCs
> >>> > > > > > > > > > > >> > with
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > ids;
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we
> ever
> >>> consider
> >>> > > > > > exposing
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > topic ids
> >>> > > > > > > > > > > >> > > > > in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well
> (both
> >>> > > > > > > > > subscribe/assign, as
> >>> > > > > > > > > > > >> well
> >>> > > > > > > > > > > >> > as in
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > rebalance
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> >>> topic
> >>> > > > > > > > > > > >> > deletion-and-recreation).
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
> >>> on all
> >>> > > > other
> >>> > > > > > minor
> >>> > > > > > > > > > > >> questions
> >>> > > > > > > > > > > >> > > > > except
> >>> > > > > > > > > > > >> > > > > >> for
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> >>> question:
> >>> > > > > > personally
> >>> > > > > > > > > I think
> >>> > > > > > > > > > > >> > it's not
> >>> > > > > > > > > > > >> > > > > >> > > necessary
> >>> > > > > > > > > > > >> > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> >>> API with
> >>> > > > > > Pattern,
> >>> > > > > > > > > but
> >>> > > > > > > > > > > >> > instead we
> >>> > > > > > > > > > > >> > > > > >> still
> >>> > > > > > > > > > > >> > > > > >> > > use
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> >>> that our
> >>> > > > > > > > > subscription may
> >>> > > > > > > > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> rejected by
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > server.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> >>> case is a
> >>> > > > > > very rare
> >>> > > > > > > > > > > >> scenario
> >>> > > > > > > > > > > >> > I felt
> >>> > > > > > > > > > > >> > > > > >> > using
> >>> > > > > > > > > > > >> > > > > >> > > an
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String`
> based
> >>> > > > > > subscription may
> >>> > > > > > > > > be
> >>> > > > > > > > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> vulnerable
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > > > various
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
> >>> 5:23 AM
> >>> > > > > > David Jacot
> >>> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> >>> feedback. Let
> >>> > > > me
> >>> > > > > > answer
> >>> > > > > > > > > your
> >>> > > > > > > > > > > >> > questions
> >>> > > > > > > > > > > >> > > > > >> > inline.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> >>> premature to
> >>> > > > > > talk about
> >>> > > > > > > > > > > >> target
> >>> > > > > > > > > > > >> > > > > versions
> >>> > > > > > > > > > > >> > > > > >> for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> >>> existing group
> >>> > > > > > protocol.
> >>> > > > > > > > > > > >> Unlike
> >>> > > > > > > > > > > >> > KRaft,
> >>> > > > > > > > > > > >> > > > > >> this
> >>> > > > > > > > > > > >> > > > > >> > > > > affects a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > core
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
> >>> hence
> >>> > > > > > > > > deprecation/removal
> >>> > > > > > > > > > > >> > will be
> >>> > > > > > > > > > > >> > > > > >> > heavily
> >>> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> >>> applications
> >>> > > > migrate
> >>> > > > > > to
> >>> > > > > > > > > the new
> >>> > > > > > > > > > > >> > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I
> >>> will
> >>> > > > remove it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> >>> intend to
> >>> > > > > > release
> >>> > > > > > > > > this in
> >>> > > > > > > > > > > >> > 4.x, but
> >>> > > > > > > > > > > >> > > > > it
> >>> > > > > > > > > > > >> > > > > >> > > wasn't
> >>> > > > > > > > > > > >> > > > > >> > > > > made
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
> >>> that as a
> >>> > > > way to
> >>> > > > > > > > > estimate
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > we'd
> >>> > > > > > > > > > > >> > > > > >> > > deprecate
> >>> > > > > > > > > > > >> > > > > >> > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > remove
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol,
> >>> I also
> >>> > > > > > suggest
> >>> > > > > > > > > removing
> >>> > > > > > > > > > > >> > this part.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> >>> reasoning. As
> >>> > > > > > > > > explained, I
> >>> > > > > > > > > > > >> plan
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> rewrite
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> >>> while we
> >>> > > > > > implement
> >>> > > > > > > > > the new
> >>> > > > > > > > > > > >> > protocol.
> >>> > > > > > > > > > > >> > > > > >> This
> >>> > > > > > > > > > > >> > > > > >> > > means
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
> >>> will be
> >>> > > > slightly
> >>> > > > > > > > > different
> >>> > > > > > > > > > > >> > (e.g.
> >>> > > > > > > > > > > >> > > > > >> threading
> >>> > > > > > > > > > > >> > > > > >> > > > > model).
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted
> to
> >>> tighten
> >>> > > > the
> >>> > > > > > > > > switch from
> >>> > > > > > > > > > > >> > the old
> >>> > > > > > > > > > > >> > > > > >> group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the
> new
> >>> group
> >>> > > > > > coordinator
> >>> > > > > > > > > to a
> >>> > > > > > > > > > > >> > major
> >>> > > > > > > > > > > >> > > > > >> release.
> >>> > > > > > > > > > > >> > > > > >> > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be
> >>> to use a
> >>> > > > flag
> >>> > > > > > to do
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > switch
> >>> > > > > > > > > > > >> > > > > >> instead
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > > > relying
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
> >>> upgrade.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
> >>> out the
> >>> > > > > > details of
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > migration
> >>> > > > > > > > > > > >> > > > > >> story.
> >>> > > > > > > > > > > >> > > > > >> > > It
> >>> > > > > > > > > > > >> > > > > >> > > > > sounds
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we
> will
> >>> support
> >>> > > > > > online
> >>> > > > > > > > > > > >> migrations.
> >>> > > > > > > > > > > >> > Is
> >>> > > > > > > > > > > >> > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > correct?
> >>> > > > > > > > > > > >> > > > > >> > > > > We
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> >>> detail. It
> >>> > > > could
> >>> > > > > > also
> >>> > > > > > > > > be done
> >>> > > > > > > > > > > >> > as a
> >>> > > > > > > > > > > >> > > > > >> separate
> >>> > > > > > > > > > > >> > > > > >> > > KIP,
> >>> > > > > > > > > > > >> > > > > >> > > > > if
> >>> > > > > > > > > > > >> > > > > >> > > > > > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
> >>> online
> >>> > > > > > migrations
> >>> > > > > > > > > for the
> >>> > > > > > > > > > > >> > group.
> >>> > > > > > > > > > > >> > > > > That
> >>> > > > > > > > > > > >> > > > > >> > > means
> >>> > > > > > > > > > > >> > > > > >> > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> >>> protocol
> >>> > > > > > will be
> >>> > > > > > > > > able to
> >>> > > > > > > > > > > >> > switch to
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly
> explain
> >>> how that
> >>> > > > > > will work
> >>> > > > > > > > > > > >> though.
> >>> > > > > > > > > > > >> > It is
> >>> > > > > > > > > > > >> > > > > >> > > basically a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must
> be
> >>> > > > upgraded or
> >>> > > > > > > > > rolled to a
> >>> > > > > > > > > > > >> > software
> >>> > > > > > > > > > > >> > > > > >> > > supporting
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new group
> coordinator.
> >>> Both the
> >>> > > > > > old and
> >>> > > > > > > > > the new
> >>> > > > > > > > > > > >> > > > > >> coordinator
> >>> > > > > > > > > > > >> > > > > >> > > will
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
> >>> protocol and
> >>> > > > rely
> >>> > > > > > on the
> >>> > > > > > > > > same
> >>> > > > > > > > > > > >> > persisted
> >>> > > > > > > > > > > >> > > > > >> > > metadata so
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
> >>> together. This
> >>> > > > point
> >>> > > > > > is an
> >>> > > > > > > > > > > >> offline
> >>> > > > > > > > > > > >> > > > > >> migration.
> >>> > > > > > > > > > > >> > > > > >> > We
> >>> > > > > > > > > > > >> > > > > >> > > > > cannot
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
> >>> because it
> >>> > > > would
> >>> > > > > > require
> >>> > > > > > > > > > > >> > shutting down
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > current
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
> >>> starting up the
> >>> > > > > > new one
> >>> > > > > > > > > and
> >>> > > > > > > > > > > >> that
> >>> > > > > > > > > > > >> > would
> >>> > > > > > > > > > > >> > > > > >> cause
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
> >>> metadata
> >>> > > > > > version/IBP
> >>> > > > > > > > > must be
> >>> > > > > > > > > > > >> > upgraded
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> X
> >>> > > > > > > > > > > >> > > > > >> > in
> >>> > > > > > > > > > > >> > > > > >> > > > > order
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> >>> protocol.
> >>> > > > This
> >>> > > > > > cannot
> >>> > > > > > > > > be done
> >>> > > > > > > > > > > >> > before
> >>> > > > > > > > > > > >> > > > > 1)
> >>> > > > > > > > > > > >> > > > > >> is
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because
> the
> >>> old
> >>> > > > > > coordinator
> >>> > > > > > > > > doesn't
> >>> > > > > > > > > > > >> > support
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must
> >>> be
> >>> > > > upgraded
> >>> > > > > > to a
> >>> > > > > > > > > version
> >>> > > > > > > > > > > >> > > > > supporting
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > online
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> >>> KIP-792).
> >>> > > > If
> >>> > > > > > the
> >>> > > > > > > > > consumer
> >>> > > > > > > > > > > >> is
> >>> > > > > > > > > > > >> > > > > already
> >>> > > > > > > > > > > >> > > > > >> > > there.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done
> >>> at this
> >>> > > > point.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must
> >>> be rolled
> >>> > > > > > with the
> >>> > > > > > > > > > > >> feature
> >>> > > > > > > > > > > >> > flag
> >>> > > > > > > > > > > >> > > > > >> turned
> >>> > > > > > > > > > > >> > > > > >> > > on.
> >>> > > > > > > > > > > >> > > > > >> > > > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> >>> automatically
> >>> > > > > > > > > converted when
> >>> > > > > > > > > > > >> > the first
> >>> > > > > > > > > > > >> > > > > >> > > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new
> protocol
> >>> joins
> >>> > > > the
> >>> > > > > > group.
> >>> > > > > > > > > While
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > members
> >>> > > > > > > > > > > >> > > > > >> > > using the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are
> being
> >>> > > > upgraded,
> >>> > > > > > the old
> >>> > > > > > > > > > > >> protocol
> >>> > > > > > > > > > > >> > is
> >>> > > > > > > > > > > >> > > > > >> proxied
> >>> > > > > > > > > > > >> > > > > >> > > into
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
> >>> this in
> >>> > > > the
> >>> > > > > > KIP.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that
> >>> we are
> >>> > > > > > pushing the
> >>> > > > > > > > > pattern
> >>> > > > > > > > > > > >> > > > > >> subscriptions
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
> >>> there
> >>> > > > could be
> >>> > > > > > some
> >>> > > > > > > > > tricky
> >>> > > > > > > > > > > >> > > > > >> compatibility
> >>> > > > > > > > > > > >> > > > > >> > > > > issues.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Will
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism
> for
> >>> users to
> >>> > > > > > detect
> >>> > > > > > > > > that
> >>> > > > > > > > > > > >> they
> >>> > > > > > > > > > > >> > need to
> >>> > > > > > > > > > > >> > > > > >> > update
> >>> > > > > > > > > > > >> > > > > >> > > > > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > > regex
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
> >>> the new
> >>> > > > > > protocol?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a
> >>> bit more
> >>> > > > > > optimistic
> >>> > > > > > > > > than
> >>> > > > > > > > > > > >> you
> >>> > > > > > > > > > > >> > on this
> >>> > > > > > > > > > > >> > > > > >> > > point. I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> >>> majority of
> >>> > > > the
> >>> > > > > > cases
> >>> > > > > > > > > are
> >>> > > > > > > > > > > >> simple
> >>> > > > > > > > > > > >> > > > > regexes
> >>> > > > > > > > > > > >> > > > > >> > which
> >>> > > > > > > > > > > >> > > > > >> > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> >>> engine. The
> >>> > > > > > > > > coordinator will
> >>> > > > > > > > > > > >> > verify
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > regex
> >>> > > > > > > > > > > >> > > > > >> > > > > anyway
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
> >>> consumer if the
> >>> > > > > > regex is
> >>> > > > > > > > > not
> >>> > > > > > > > > > > >> valid.
> >>> > > > > > > > > > > >> > > > > Coming
> >>> > > > > > > > > > > >> > > > > >> > back
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in
> the
> >>> worst
> >>> > > > case,
> >>> > > > > > the
> >>> > > > > > > > > first
> >>> > > > > > > > > > > >> > upgraded
> >>> > > > > > > > > > > >> > > > > >> > consumer
> >>> > > > > > > > > > > >> > > > > >> > > > > joining
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> >>> rejected.
> >>> > > > This
> >>> > > > > > should
> >>> > > > > > > > > be used
> >>> > > > > > > > > > > >> > as the
> >>> > > > > > > > > > > >> > > > > >> last
> >>> > > > > > > > > > > >> > > > > >> > > > > defence, I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers
> >>> to
> >>> > > > validate
> >>> > > > > > their
> >>> > > > > > > > > regex
> >>> > > > > > > > > > > >> > before
> >>> > > > > > > > > > > >> > > > > >> > upgrading
> >>> > > > > > > > > > > >> > > > > >> > > > > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
> >>> them with
> >>> > > > > > another
> >>> > > > > > > > > group.
> >>> > > > > > > > > > > >> For
> >>> > > > > > > > > > > >> > > > > >> instance,
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> >>> pre-prod
> >>> > > > > > environment.
> >>> > > > > > > > > > > >> Another
> >>> > > > > > > > > > > >> > way
> >>> > > > > > > > > > > >> > > > > >> would be
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> >>> consumer-group tool
> >>> > > > to
> >>> > > > > > > > > provide a
> >>> > > > > > > > > > > >> regex
> >>> > > > > > > > > > > >> > > > > >> validation
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this
> >>> be
> >>> > > > enough in
> >>> > > > > > your
> >>> > > > > > > > > > > >> opinion?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the
> >>> last
> >>> > > > question,
> >>> > > > > > will
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> Java
> >>> > > > > > > > > > > >> > client
> >>> > > > > > > > > > > >> > > > > >> allow
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > users to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
> >>> current regex
> >>> > > > > > engine for
> >>> > > > > > > > > > > >> > compatibility
> >>> > > > > > > > > > > >> > > > > >> > > reasons?
> >>> > > > > > > > > > > >> > > > > >> > > > > For
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
> >>> keep using
> >>> > > > > > client
> >>> > > > > > > > > based
> >>> > > > > > > > > > > >> > regex at
> >>> > > > > > > > > > > >> > > > > >> first
> >>> > > > > > > > > > > >> > > > > >> > to
> >>> > > > > > > > > > > >> > > > > >> > > keep
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple
> >>> and then
> >>> > > > > > migrate to
> >>> > > > > > > > > server
> >>> > > > > > > > > > > >> > based
> >>> > > > > > > > > > > >> > > > > >> regexes
> >>> > > > > > > > > > > >> > > > > >> > > as a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > second
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
> >>> point but I
> >>> > > > am
> >>> > > > > > > > > concerned that
> >>> > > > > > > > > > > >> > this
> >>> > > > > > > > > > > >> > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > allow
> >>> > > > > > > > > > > >> > > > > >> > > > > users
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
> >>> this mode.
> >>> > > > That
> >>> > > > > > > > > would go
> >>> > > > > > > > > > > >> > against our
> >>> > > > > > > > > > > >> > > > > >> goal
> >>> > > > > > > > > > > >> > > > > >> > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the
> client
> >>> because
> >>> > > > we
> >>> > > > > > would
> >>> > > > > > > > > have to
> >>> > > > > > > > > > > >> > continue
> >>> > > > > > > > > > > >> > > > > >> > > monitoring
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> >>> client
> >>> > > > side. I
> >>> > > > > > would
> >>> > > > > > > > > rather
> >>> > > > > > > > > > > >> > not do
> >>> > > > > > > > > > > >> > > > > >> this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
> >>> the group
> >>> > > > > > > > > coordinator
> >>> > > > > > > > > > > >> will
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > > >> > > responsible for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > storing
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations
> >>> and that
> >>> > > > the
> >>> > > > > > > > > > > >> configurations
> >>> > > > > > > > > > > >> > will be
> >>> > > > > > > > > > > >> > > > > >> > > deleted
> >>> > > > > > > > > > > >> > > > > >> > > > > when
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
> >>> Will a
> >>> > > > > > transition to
> >>> > > > > > > > > DEAD
> >>> > > > > > > > > > > >> > trigger
> >>> > > > > > > > > > > >> > > > > >> > deletion
> >>> > > > > > > > > > > >> > > > > >> > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> >>> > > > configurations
> >>> > > > > > will be
> >>> > > > > > > > > > > >> deleted
> >>> > > > > > > > > > > >> > when
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > group is
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> >>> together.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice
> >>> to store
> >>> > > > the
> >>> > > > > > > > > configs in
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > group
> >>> > > > > > > > > > > >> > > > > >> > > coordinator
> >>> > > > > > > > > > > >> > > > > >> > > > > > > make it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> >>> cluster
> >>> > > > > > configs and
> >>> > > > > > > > > their
> >>> > > > > > > > > > > >> > values?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
> >>> group
> >>> > > > > > > > > configurations are
> >>> > > > > > > > > > > >> > overrides
> >>> > > > > > > > > > > >> > > > > >> of
> >>> > > > > > > > > > > >> > > > > >> > > cluster
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want
> >>> to know
> >>> > > > all
> >>> > > > > > the
> >>> > > > > > > > > overrides
> >>> > > > > > > > > > > >> > though,
> >>> > > > > > > > > > > >> > > > > you
> >>> > > > > > > > > > > >> > > > > >> > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > have
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> >>> > > > coordinators.
> >>> > > > > > You
> >>> > > > > > > > > cannot
> >>> > > > > > > > > > > >> > rely on
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > metadata
> >>> > > > > > > > > > > >> > > > > >> > > > > log
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would
> someone
> >>> > > > configure a
> >>> > > > > > group
> >>> > > > > > > > > before
> >>> > > > > > > > > > > >> > starting
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > consumers?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
> >>> allowing the
> >>> > > > > > explicit
> >>> > > > > > > > > creation
> >>> > > > > > > > > > > >> of
> >>> > > > > > > > > > > >> > > > > groups?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> >>> decoupled
> >>> > > > from
> >>> > > > > > the
> >>> > > > > > > > > group
> >>> > > > > > > > > > > >> > lifecycle.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will
> be
> >>> > > > > > automatically
> >>> > > > > > > > > created in
> >>> > > > > > > > > > > >> > this
> >>> > > > > > > > > > > >> > > > > case.
> >>> > > > > > > > > > > >> > > > > >> > > However,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
> >>> lost after
> >>> > > > the
> >>> > > > > > > > > retention
> >>> > > > > > > > > > > >> > period of
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> >>> > > > Consumer.subscribe
> >>> > > > > > method
> >>> > > > > > > > > for the
> >>> > > > > > > > > > > >> > Java
> >>> > > > > > > > > > > >> > > > > >> client
> >>> > > > > > > > > > > >> > > > > >> > > still
> >>> > > > > > > > > > > >> > > > > >> > > > > take
> >>> > > > > > > > > > > >> > > > > >> > > > > > > a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> `java.util.regex.Pattern` of
> >>> > > > do
> >>> > > > > > we
> >>> > > > > > > > > have to
> >>> > > > > > > > > > > >> > introduce
> >>> > > > > > > > > > > >> > > > > an
> >>> > > > > > > > > > > >> > > > > >> > > overload?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> >>> question. I
> >>> > > > > > forgot
> >>> > > > > > > > > about
> >>> > > > > > > > > > > >> that
> >>> > > > > > > > > > > >> > one.
> >>> > > > > > > > > > > >> > > > > As
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> `java.util.regex.Pattern` is
> >>> > > > not
> >>> > > > > > fully
> >>> > > > > > > > > > > >> compatible
> >>> > > > > > > > > > > >> > with
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > engine
> >>> > > > > > > > > > > >> > > > > >> > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
> >>> might be
> >>> > > > better
> >>> > > > > > to
> >>> > > > > > > > > deprecate
> >>> > > > > > > > > > > >> > it and
> >>> > > > > > > > > > > >> > > > > >> use an
> >>> > > > > > > > > > > >> > > > > >> > > > > overload
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string.
> >>> We would
> >>> > > > > > rely on
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> server
> >>> > > > > > > > > > > >> > side
> >>> > > > > > > > > > > >> > > > > >> > > validation.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration,
> >>> I think
> >>> > > > that
> >>> > > > > > we
> >>> > > > > > > > > could
> >>> > > > > > > > > > > >> still
> >>> > > > > > > > > > > >> > try to
> >>> > > > > > > > > > > >> > > > > >> > > toString
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it.
> That
> >>> should
> >>> > > > > > work, I
> >>> > > > > > > > > think, in
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> majority
> >>> > > > > > > > > > > >> > > > > >> > > of the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> >>> Justine
> >>> > > > that we
> >>> > > > > > > > > should be
> >>> > > > > > > > > > > >> > clearer
> >>> > > > > > > > > > > >> > > > > about
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > > > reason
> >>> > > > > > > > > > > >> > > > > >> > > > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> >>> > > > IBP/metadata.version
> >>> > > > > > from the
> >>> > > > > > > > > > > >> feature
> >>> > > > > > > > > > > >> > flag.
> >>> > > > > > > > > > > >> > > > > >> Maybe
> >>> > > > > > > > > > > >> > > > > >> > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > mean
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
> >>> default for
> >>> > > > the
> >>> > > > > > feature
> >>> > > > > > > > > flag
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > true
> >>> > > > > > > > > > > >> > > > > >> based
> >>> > > > > > > > > > > >> > > > > >> > on
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
> >>> once we
> >>> > > > want to
> >>> > > > > > make
> >>> > > > > > > > > it the
> >>> > > > > > > > > > > >> > default.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
> >>> that feature
> >>> > > > > > flag
> >>> > > > > > > > > mainly
> >>> > > > > > > > > > > >> > during the
> >>> > > > > > > > > > > >> > > > > >> > > development
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not
> >>> have
> >>> > > > mentioned
> >>> > > > > > it, I
> >>> > > > > > > > > think,
> >>> > > > > > > > > > > >> > because
> >>> > > > > > > > > > > >> > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > could
> >>> > > > > > > > > > > >> > > > > >> > > > > use
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config
> for
> >>> it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> >>> protocol APIs
> >>> > > > > > don't
> >>> > > > > > > > > mention
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > required
> >>> > > > > > > > > > > >> > > > > >> > > ACLs, it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that
> for
> >>> > > > consistency.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
> >>> that
> >>> > > > > > > > > > > >> ConsumerGroupHeartbeat
> >>> > > > > > > > > > > >> > > > > requires
> >>> > > > > > > > > > > >> > > > > >> > > "Read
> >>> > > > > > > > > > > >> > > > > >> > > > > Group"
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to
> >>> do more
> >>> > > > than
> >>> > > > > > > > > reading.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how
> >>> the
> >>> > > > current
> >>> > > > > > > > > protocol works
> >>> > > > > > > > > > > >> > though.
> >>> > > > > > > > > > > >> > > > > We
> >>> > > > > > > > > > > >> > > > > >> > only
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group"
> >>> to join a
> >>> > > > > > group. We
> >>> > > > > > > > > could
> >>> > > > > > > > > > > >> > consider
> >>> > > > > > > > > > > >> > > > > >> > > changing
> >>> > > > > > > > > > > >> > > > > >> > > > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure
> that
> >>> it is
> >>> > > > worth
> >>> > > > > > it.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> >>> recreation
> >>> > > > > > handled by
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > consumer
> >>> > > > > > > > > > > >> > > > > with
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would
> >>> be good to
> >>> > > > > > have a
> >>> > > > > > > > > section
> >>> > > > > > > > > > > >> on
> >>> > > > > > > > > > > >> > this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a
> protocol
> >>> > > > > > perspective, the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> topic
> >>> > > > > > > > > > > >> > will
> >>> > > > > > > > > > > >> > > > > >> have a
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> >>> treat it
> >>> > > > like a
> >>> > > > > > > > > topic with
> >>> > > > > > > > > > > >> a
> >>> > > > > > > > > > > >> > > > > different
> >>> > > > > > > > > > > >> > > > > >> > > name.
> >>> > > > > > > > > > > >> > > > > >> > > > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that
> the
> >>> > > > fetch/commit
> >>> > > > > > > > > offsets
> >>> > > > > > > > > > > >> APIs
> >>> > > > > > > > > > > >> > do not
> >>> > > > > > > > > > > >> > > > > >> > support
> >>> > > > > > > > > > > >> > > > > >> > > > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
> >>> would
> >>> > > > reuse the
> >>> > > > > > > > > offsets
> >>> > > > > > > > > > > >> based
> >>> > > > > > > > > > > >> > on the
> >>> > > > > > > > > > > >> > > > > >> > same.
> >>> > > > > > > > > > > >> > > > > >> > > I
> >>> > > > > > > > > > > >> > > > > >> > > > > think
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
> >>> those
> >>> > > > APIs
> >>> > > > > > as well
> >>> > > > > > > > > in
> >>> > > > > > > > > > > >> order
> >>> > > > > > > > > > > >> > to be
> >>> > > > > > > > > > > >> > > > > >> > > consistent
> >>> > > > > > > > > > > >> > > > > >> > > > > end
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> >>> strengthen
> >>> > > > the
> >>> > > > > > > > > semantics of
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> consumer.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP
> mentions
> >>> we will
> >>> > > > > > write the
> >>> > > > > > > > > new
> >>> > > > > > > > > > > >> > coordinator
> >>> > > > > > > > > > > >> > > > > >> in
> >>> > > > > > > > > > > >> > > > > >> > > Java.
> >>> > > > > > > > > > > >> > > > > >> > > > > Even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> >>> implementation
> >>> > > > > > detail, do
> >>> > > > > > > > > we plan
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > have a
> >>> > > > > > > > > > > >> > > > > >> new
> >>> > > > > > > > > > > >> > > > > >> > > gradle
> >>> > > > > > > > > > > >> > > > > >> > > > > > > module
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> >>> scalability
> >>> > > > > > goal when
> >>> > > > > > > > > it
> >>> > > > > > > > > > > >> comes
> >>> > > > > > > > > > > >> > to how
> >>> > > > > > > > > > > >> > > > > >> many
> >>> > > > > > > > > > > >> > > > > >> > > > > members
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> >>> support?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers
> >>> at the
> >>> > > > > > moment. The
> >>> > > > > > > > > > > >> protocol
> >>> > > > > > > > > > > >> > should
> >>> > > > > > > > > > > >> > > > > >> > > support
> >>> > > > > > > > > > > >> > > > > >> > > > > 1000s
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group.
> >>> We will
> >>> > > > > > measure
> >>> > > > > > > > > this when
> >>> > > > > > > > > > > >> > we have
> >>> > > > > > > > > > > >> > > > > a
> >>> > > > > > > > > > > >> > > > > >> > first
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
> >>> that we
> >>> > > > might
> >>> > > > > > have
> >>> > > > > > > > > other
> >>> > > > > > > > > > > >> > > > > bottlenecks
> >>> > > > > > > > > > > >> > > > > >> > down
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> >>> commits).
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> >>> having
> >>> > > > > > > > > SubscribedTopidIds
> >>> > > > > > > > > > > >> > instead
> >>> > > > > > > > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> SubscribedTopicNames
> >>> in
> >>> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> >>> > > > > > > > > > > >> > > > > >> Is
> >>> > > > > > > > > > > >> > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > > > idea
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> >>> resolve the
> >>> > > > > > regex on
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > server, we
> >>> > > > > > > > > > > >> > > > > >> can do
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> >>> > > > difference
> >>> > > > > > is that
> >>> > > > > > > > > > > >> sending
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> regex is
> >>> > > > > > > > > > > >> > > > > >> > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
> >>> topic
> >>> > > > names
> >>> > > > > > is less
> >>> > > > > > > > > > > >> > efficient.
> >>> > > > > > > > > > > >> > > > > >> > > Furthermore,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > delete
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is
> easier
> >>> to
> >>> > > > handle
> >>> > > > > > if we
> >>> > > > > > > > > have
> >>> > > > > > > > > > > >> topic
> >>> > > > > > > > > > > >> > ids.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> >>> consolidate the
> >>> > > > > > metadata
> >>> > > > > > > > > lookup
> >>> > > > > > > > > > > >> > on the
> >>> > > > > > > > > > > >> > > > > >> server
> >>> > > > > > > > > > > >> > > > > >> > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > both
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
> >>> with your
> >>> > > > > > point. As
> >>> > > > > > > > > a
> >>> > > > > > > > > > > >> second
> >>> > > > > > > > > > > >> > > > > though,
> >>> > > > > > > > > > > >> > > > > >> > using
> >>> > > > > > > > > > > >> > > > > >> > > > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better
> here
> >>> for the
> >>> > > > > > delete and
> >>> > > > > > > > > > > >> > recreation
> >>> > > > > > > > > > > >> > > > > case.
> >>> > > > > > > > > > > >> > > > > >> > > Also, I
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
> >>> allow
> >>> > > > users to
> >>> > > > > > > > > subscribe
> >>> > > > > > > > > > > >> with
> >>> > > > > > > > > > > >> > topic
> >>> > > > > > > > > > > >> > > > > >> ids
> >>> > > > > > > > > > > >> > > > > >> > in
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that
> is
> >>> the only
> >>> > > > > > way to be
> >>> > > > > > > > > > > >> really
> >>> > > > > > > > > > > >> > robust
> >>> > > > > > > > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > topic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022
> >>> at 1:38 PM
> >>> > > > > > David
> >>> > > > > > > > > Jacot <
> >>> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> >>> comments.
> >>> > > > Please
> >>> > > > > > find
> >>> > > > > > > > > my
> >>> > > > > > > > > > > >> answers
> >>> > > > > > > > > > > >> > > > > below.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
> >>> protocol
> >>> > > > relies on
> >>> > > > > > > > > topic IDs
> >>> > > > > > > > > > > >> > with the
> >>> > > > > > > > > > > >> > > > > >> > > exception
> >>> > > > > > > > > > > >> > > > > >> > > > > of the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based
> in
> >>> the
> >>> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> >>> > > > > > > > > > > >> > > > > >> I
> >>> > > > > > > > > > > >> > > > > >> > am
> >>> > > > > > > > > > > >> > > > > >> > > not
> >>> > > > > > > > > > > >> > > > > >> > > > > sure
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic
> names
> >>> is the
> >>> > > > > > right call
> >>> > > > > > > > > here.
> >>> > > > > > > > > > > >> I
> >>> > > > > > > > > > > >> > need to
> >>> > > > > > > > > > > >> > > > > >> > think
> >>> > > > > > > > > > > >> > > > > >> > > > > about it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> >>> Obviously, the
> >>> > > > > > KIP does
> >>> > > > > > > > > not
> >>> > > > > > > > > > > >> > change the
> >>> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
> >>> topic
> >>> > > > IDs.
> >>> > > > > > This
> >>> > > > > > > > > may be
> >>> > > > > > > > > > > >> > something
> >>> > > > > > > > > > > >> > > > > >> that
> >>> > > > > > > > > > > >> > > > > >> > we
> >>> > > > > > > > > > > >> > > > > >> > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as
> it
> >>> would
> >>> > > > give
> >>> > > > > > better
> >>> > > > > > > > > > > >> overall
> >>> > > > > > > > > > > >> > > > > >> guarantee in
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> >>> think that
> >>> > > > I
> >>> > > > > > should
> >>> > > > > > > > > not
> >>> > > > > > > > > > > >> have
> >>> > > > > > > > > > > >> > > > > mentioned
> >>> > > > > > > > > > > >> > > > > >> > this
> >>> > > > > > > > > > > >> > > > > >> > > > > flag at
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove
> >>> it. We can
> >>> > > > > > use an
> >>> > > > > > > > > internal
> >>> > > > > > > > > > > >> > > > > >> configuration
> >>> > > > > > > > > > > >> > > > > >> > > while
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
> >>> feature.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster
> types
> >>> will be
> >>> > > > > > > > > supported. The
> >>> > > > > > > > > > > >> > change is
> >>> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
> >>> that the
> >>> > > > > > cluster
> >>> > > > > > > > > uses
> >>> > > > > > > > > > > >> topic
> >>> > > > > > > > > > > >> > IDs.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11,
> 2022
> >>> at 9:53
> >>> > > > PM
> >>> > > > > > > > > Guozhang
> >>> > > > > > > > > > > >> Wang <
> >>> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> >>> feedback.
> >>> > > > Here
> >>> > > > > > are
> >>> > > > > > > > > some
> >>> > > > > > > > > > > >> replies
> >>> > > > > > > > > > > >> > > > > inlined
> >>> > > > > > > > > > > >> > > > > >> > > below:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
> >>> 2022 at
> >>> > > > 2:53 AM
> >>> > > > > > > > > Ismael Juma
> >>> > > > > > > > > > > >> <
> >>> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
> >>> KIP. This
> >>> > > > has
> >>> > > > > > the
> >>> > > > > > > > > > > >> potential
> >>> > > > > > > > > > > >> > to be a
> >>> > > > > > > > > > > >> > > > > >> > great
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> >>> > > > questions/comments:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> >>> > > > premature to
> >>> > > > > > talk
> >>> > > > > > > > > about
> >>> > > > > > > > > > > >> > target
> >>> > > > > > > > > > > >> > > > > >> versions
> >>> > > > > > > > > > > >> > > > > >> > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> >>> existing
> >>> > > > group
> >>> > > > > > > > > protocol.
> >>> > > > > > > > > > > >> > Unlike
> >>> > > > > > > > > > > >> > > > > >> KRaft,
> >>> > > > > > > > > > > >> > > > > >> > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
> >>> and hence
> >>> > > > > > > > > > > >> > deprecation/removal will
> >>> > > > > > > > > > > >> > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> > > heavily
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> >>> applications
> >>> > > > > > migrate
> >>> > > > > > > > > to the
> >>> > > > > > > > > > > >> new
> >>> > > > > > > > > > > >> > > > > >> protocol.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
> >>> you. I
> >>> > > > think
> >>> > > > > > we can
> >>> > > > > > > > > > > >> remove
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> proposed
> >>> > > > > > > > > > > >> > > > > >> > > > > timeline
> >>> > > > > > > > > > > >> > > > > >> > > > > > > in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> >>> > > > Deprecation,
> >>> > > > > > and
> >>> > > > > > > > > Migration
> >>> > > > > > > > > > > >> > Plan` and
> >>> > > > > > > > > > > >> > > > > >> > > instead
> >>> > > > > > > > > > > >> > > > > >> > > > > just
> >>> > > > > > > > > > > >> > > > > >> > > > > > > state
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
> >>> decide in the
> >>> > > > > > future
> >>> > > > > > > > > about
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > we
> >>> > > > > > > > > > > >> > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> >>> behaviors.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says
> >>> we
> >>> > > > intend to
> >>> > > > > > > > > release this
> >>> > > > > > > > > > > >> > in 4.x,
> >>> > > > > > > > > > > >> > > > > >> but
> >>> > > > > > > > > > > >> > > > > >> > it
> >>> > > > > > > > > > > >> > > > > >> > > > > wasn't
> >>> > > > > > > > > > > >> > > > > >> > > > > > > made
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we
> added
> >>> that as
> >>> > > > a
> >>> > > > > > way to
> >>> > > > > > > > > > > >> estimate
> >>> > > > > > > > > > > >> > when
> >>> > > > > > > > > > > >> > > > > we'd
> >>> > > > > > > > > > > >> > > > > >> > > > > deprecate
> >>> > > > > > > > > > > >> > > > > >> > > > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> >>> protocol, I
> >>> > > > also
> >>> > > > > > suggest
> >>> > > > > > > > > > > >> removing
> >>> > > > > > > > > > > >> > this
> >>> > > > > > > > > > > >> > > > > >> part.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's
> not
> >>> > > > specifically
> >>> > > > > > > > > related to
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
> >>> more for
> >>> > > > client
> >>> > > > > > > > > upgrades.
> >>> > > > > > > > > > > >> I.e.
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> > > broker-side
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation
> may
> >>> be done
> >>> > > > > > first,
> >>> > > > > > > > > and then
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > client
> >>> > > > > > > > > > > >> > > > > >> > side,
> >>> > > > > > > > > > > >> > > > > >> > > > > and we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> >>> "released"
> >>> > > > by
> >>> > > > > > the
> >>> > > > > > > > > time
> >>> > > > > > > > > > > >> clients
> >>> > > > > > > > > > > >> > > > > >> > > implementations
> >>> > > > > > > > > > > >> > > > > >> > > > > are
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
> >>> enable the
> >>> > > > > > feature the
> >>> > > > > > > > > > > >> clients
> >>> > > > > > > > > > > >> > need to
> >>> > > > > > > > > > > >> > > > > >> > first
> >>> > > > > > > > > > > >> > > > > >> > > > > swap-in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> >>> rolling
> >>> > > > bounce
> >>> > > > > > and
> >>> > > > > > > > > then set
> >>> > > > > > > > > > > >> > the flag
> >>> > > > > > > > > > > >> > > > > >> > with a
> >>> > > > > > > > > > > >> > > > > >> > > > > second
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence
> >>> we feel
> >>> > > > it's
> >>> > > > > > > > > better to be
> >>> > > > > > > > > > > >> > released
> >>> > > > > > > > > > > >> > > > > >> in a
> >>> > > > > > > > > > > >> > > > > >> > > major
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> >>> flesh out
> >>> > > > the
> >>> > > > > > > > > details of
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > migration
> >>> > > > > > > > > > > >> > > > > >> > > story.
> >>> > > > > > > > > > > >> > > > > >> > > > > It
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
> >>> will
> >>> > > > support
> >>> > > > > > online
> >>> > > > > > > > > > > >> > migrations. Is
> >>> > > > > > > > > > > >> > > > > >> that
> >>> > > > > > > > > > > >> > > > > >> > > > > correct?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > We
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> >>> detail.
> >>> > > > It
> >>> > > > > > could
> >>> > > > > > > > > also be
> >>> > > > > > > > > > > >> > done as a
> >>> > > > > > > > > > > >> > > > > >> > > separate
> >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think
> that's
> >>> the
> >>> > > > part we
> >>> > > > > > can
> >>> > > > > > > > > be more
> >>> > > > > > > > > > > >> > concrete
> >>> > > > > > > > > > > >> > > > > >> about
> >>> > > > > > > > > > > >> > > > > >> > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > sure
> >>> > > > > > > > > > > >> > > > > >> > > > > > > (and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related
> to
> >>> your
> >>> > > > > > question 2)
> >>> > > > > > > > > above).
> >>> > > > > > > > > > > >> > We will
> >>> > > > > > > > > > > >> > > > > >> work
> >>> > > > > > > > > > > >> > > > > >> > on
> >>> > > > > > > > > > > >> > > > > >> > > > > making
> >>> > > > > > > > > > > >> > > > > >> > > > > > > it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
> >>> parallel as we
> >>> > > > > > solicit
> >>> > > > > > > > > more
> >>> > > > > > > > > > > >> > feedback.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
> >>> that we are
> >>> > > > > > pushing
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> > pattern
> >>> > > > > > > > > > > >> > > > > >> > > subscriptions
> >>> > > > > > > > > > > >> > > > > >> > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
> >>> like there
> >>> > > > > > could be
> >>> > > > > > > > > some
> >>> > > > > > > > > > > >> tricky
> >>> > > > > > > > > > > >> > > > > >> > > compatibility
> >>> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a
> mechanism
> >>> for
> >>> > > > users to
> >>> > > > > > > > > detect that
> >>> > > > > > > > > > > >> > they
> >>> > > > > > > > > > > >> > > > > need
> >>> > > > > > > > > > > >> > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> > > update
> >>> > > > > > > > > > > >> > > > > >> > > > > > > their
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before
> switching
> >>> to the
> >>> > > > new
> >>> > > > > > > > > protocol?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
> >>> need some
> >>> > > > > > tooling for
> >>> > > > > > > > > > > >> non-java
> >>> > > > > > > > > > > >> > client
> >>> > > > > > > > > > > >> > > > > >> > users
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > sort
> >>> > > > > > > > > > > >> > > > > >> > > > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
> >>> client before
> >>> > > > > > > > > switching to
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > new
> >>> > > > > > > > > > > >> > > > > >> > protocol.
> >>> > > > > > > > > > > >> > > > > >> > > I
> >>> > > > > > > > > > > >> > > > > >> > > > > do not
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on
> >>> top of my
> >>> > > > head
> >>> > > > > > > > > though,
> >>> > > > > > > > > > > >> maybe
> >>> > > > > > > > > > > >> > others
> >>> > > > > > > > > > > >> > > > > >> like
> >>> > > > > > > > > > > >> > > > > >> > > @Matt
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to
> >>> the last
> >>> > > > > > question,
> >>> > > > > > > > > will the
> >>> > > > > > > > > > > >> > Java
> >>> > > > > > > > > > > >> > > > > >> client
> >>> > > > > > > > > > > >> > > > > >> > > allow
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> >>> current
> >>> > > > regex
> >>> > > > > > > > > engine for
> >>> > > > > > > > > > > >> > > > > >> compatibility
> >>> > > > > > > > > > > >> > > > > >> > > > > reasons?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > For
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy
> >>> to keep
> >>> > > > using
> >>> > > > > > > > > client
> >>> > > > > > > > > > > >> based
> >>> > > > > > > > > > > >> > regex
> >>> > > > > > > > > > > >> > > > > at
> >>> > > > > > > > > > > >> > > > > >> > > first to
> >>> > > > > > > > > > > >> > > > > >> > > > > keep
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
> >>> simple and
> >>> > > > then
> >>> > > > > > migrate
> >>> > > > > > > > > to
> >>> > > > > > > > > > > >> > server based
> >>> > > > > > > > > > > >> > > > > >> > > regexes
> >>> > > > > > > > > > > >> > > > > >> > > > > as a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have
> >>> not thought
> >>> > > > > > about
> >>> > > > > > > > > that for
> >>> > > > > > > > > > > >> > java
> >>> > > > > > > > > > > >> > > > > >> clients,
> >>> > > > > > > > > > > >> > > > > >> > > and
> >>> > > > > > > > > > > >> > > > > >> > > > > we can
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind
> of
> >>> > > > > > compatibility
> >>> > > > > > > > > issues do
> >>> > > > > > > > > > > >> > you have
> >>> > > > > > > > > > > >> > > > > in
> >>> > > > > > > > > > > >> > > > > >> > > mind?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
> >>> that the
> >>> > > > group
> >>> > > > > > > > > coordinator
> >>> > > > > > > > > > > >> > will be
> >>> > > > > > > > > > > >> > > > > >> > > > > responsible for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
> >>> configurations and
> >>> > > > that
> >>> > > > > > the
> >>> > > > > > > > > > > >> > configurations
> >>> > > > > > > > > > > >> > > > > will
> >>> > > > > > > > > > > >> > > > > >> be
> >>> > > > > > > > > > > >> > > > > >> > > > > deleted
> >>> > > > > > > > > > > >> > > > > >> > > > > > > when
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
> >>> deleted. Will a
> >>> > > > > > > > > transition to
> >>> > > > > > > > > > > >> DEAD
> >>> > > > > > > > > > > >> > > > > trigger
> >>> > > > > > > > > > > >> > > > > >> > > deletion
> >>> > > > > > > > > > > >> > > > > >> > > > > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
> >>> DEAD state
> >>> > > > is an
> >>> > > > > > > > > ending
> >>> > > > > > > > > > > >> state
> >>> > > > > > > > > > > >> > (we
> >>> > > > > > > > > > > >> > > > > would
> >>> > > > > > > > > > > >> > > > > >> > only
> >>> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
> >>> group is
> >>> > > > EMPTY
> >>> > > > > > and
> >>> > > > > > > > > also all
> >>> > > > > > > > > > > >> of
> >>> > > > > > > > > > > >> > its
> >>> > > > > > > > > > > >> > > > > >> > metadata
> >>> > > > > > > > > > > >> > > > > >> > > are
> >>> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
> >>> DEAD this
> >>> > > > > > group
> >>> > > > > > > > > would
> >>> > > > > > > > > > > >> never
> >>> > > > > > > > > > > >> > be
> >>> > > > > > > > > > > >> > > > > >> revived.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
> >>> choice to
> >>> > > > store
> >>> > > > > > the
> >>> > > > > > > > > configs
> >>> > > > > > > > > > > >> in
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > coordinator
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list
> >>> all
> >>> > > > cluster
> >>> > > > > > configs
> >>> > > > > > > > > and
> >>> > > > > > > > > > > >> their
> >>> > > > > > > > > > > >> > > > > values?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> >>> question,
> >>> > > > and our
> >>> > > > > > > > > thoughts
> >>> > > > > > > > > > > >> are
> >>> > > > > > > > > > > >> > that
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > so-called
> >>> > > > > > > > > > > >> > > > > >> > > > > > > "group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations"
> are
> >>> > > > overrides
> >>> > > > > > of the
> >>> > > > > > > > > > > >> > cluster-level
> >>> > > > > > > > > > > >> > > > > >> > > > > configurations
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
> >>> group so
> >>> > > > when an
> >>> > > > > > > > > admin list
> >>> > > > > > > > > > > >> > cluster
> >>> > > > > > > > > > > >> > > > > >> > configs
> >>> > > > > > > > > > > >> > > > > >> > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> >>> cluster-level
> >>> > > > > > > > > "defaults", not
> >>> > > > > > > > > > > >> > showing
> >>> > > > > > > > > > > >> > > > > >> any
> >>> > > > > > > > > > > >> > > > > >> > > > > per-group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> >>> someone
> >>> > > > > > configure a
> >>> > > > > > > > > group
> >>> > > > > > > > > > > >> > before
> >>> > > > > > > > > > > >> > > > > >> starting
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> >>> allowing
> >>> > > > the
> >>> > > > > > explicit
> >>> > > > > > > > > > > >> > creation of
> >>> > > > > > > > > > > >> > > > > >> groups?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could
> be
> >>> > > > decoupled
> >>> > > > > > from
> >>> > > > > > > > > the group
> >>> > > > > > > > > > > >> > > > > lifecycle.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can
> be
> >>> created
> >>> > > > > > before
> >>> > > > > > > > > the group
> >>> > > > > > > > > > > >> > itself
> >>> > > > > > > > > > > >> > > > > as
> >>> > > > > > > > > > > >> > > > > >> an
> >>> > > > > > > > > > > >> > > > > >> > > > > > > independent
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course,
> this
> >>> > > > requires
> >>> > > > > > the
> >>> > > > > > > > > > > >> corresponding
> >>> > > > > > > > > > > >> > > > > >> request to
> >>> > > > > > > > > > > >> > > > > >> > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> >>> based on
> >>> > > > the
> >>> > > > > > group
> >>> > > > > > > > > id ---
> >>> > > > > > > > > > > >> > the only
> >>> > > > > > > > > > > >> > > > > >> > thing
> >>> > > > > > > > > > > >> > > > > >> > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> >>> itself is
> >>> > > > gone
> >>> > > > > > we
> >>> > > > > > > > > also check
> >>> > > > > > > > > > > >> > if there
> >>> > > > > > > > > > > >> > > > > >> are
> >>> > > > > > > > > > > >> > > > > >> > > any
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related
> >>> to that
> >>> > > > group
> >>> > > > > > and
> >>> > > > > > > > > delete
> >>> > > > > > > > > > > >> as
> >>> > > > > > > > > > > >> > well.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> >>> indeed
> >>> > > > > > introduces an
> >>> > > > > > > > > > > >> > asymmetry on
> >>> > > > > > > > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > creation
> >>> > > > > > > > > > > >> > > > > >> > > > > /
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> >>> config
> >>> > > > > > entities,
> >>> > > > > > > > > and we
> >>> > > > > > > > > > > >> > would like
> >>> > > > > > > > > > > >> > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> > > hear
> >>> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether
> >>> we should
> >>> > > > aim
> >>> > > > > > for
> >>> > > > > > > > > symmetry
> >>> > > > > > > > > > > >> > i.e.
> >>> > > > > > > > > > > >> > > > > >> totally
> >>> > > > > > > > > > > >> > > > > >> > > > > decouple
> >>> > > > > > > > > > > >> > > > > >> > > > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence
> >>> not
> >>> > > > delete
> >>> > > > > > them at
> >>> > > > > > > > > all
> >>> > > > > > > > > > > >> when
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> group
> >>> > > > > > > > > > > >> > > > > >> > is
> >>> > > > > > > > > > > >> > > > > >> > > > > gone,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > but
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> >>> deletion
> >>> > > > > > operations
> >>> > > > > > > > > by
> >>> > > > > > > > > > > >> > themselves.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> >>> > > > > > Consumer.subscribe
> >>> > > > > > > > > method for
> >>> > > > > > > > > > > >> > the Java
> >>> > > > > > > > > > > >> > > > > >> > client
> >>> > > > > > > > > > > >> > > > > >> > > > > still
> >>> > > > > > > > > > > >> > > > > >> > > > > > > take
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > `java.util.regex.Pattern` of
> >>> > > > > > do we
> >>> > > > > > > > > have
> >>> > > > > > > > > > > >> to
> >>> > > > > > > > > > > >> > > > > >> introduce an
> >>> > > > > > > > > > > >> > > > > >> > > > > overload?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
> >>> need to
> >>> > > > > > introduce
> >>> > > > > > > > > an
> >>> > > > > > > > > > > >> > overload, but
> >>> > > > > > > > > > > >> > > > > >> I'm
> >>> > > > > > > > > > > >> > > > > >> > > all
> >>> > > > > > > > > > > >> > > > > >> > > > > ears
> >>> > > > > > > > > > > >> > > > > >> > > > > > > if
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> >>> compatibility
> >>> > > > > > issues
> >>> > > > > > > > > that we
> >>> > > > > > > > > > > >> may
> >>> > > > > > > > > > > >> > > > > overlook.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree
> with
> >>> Justine
> >>> > > > > > that we
> >>> > > > > > > > > should
> >>> > > > > > > > > > > >> be
> >>> > > > > > > > > > > >> > clearer
> >>> > > > > > > > > > > >> > > > > >> > about
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > reason
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> >>> > > > > > IBP/metadata.version
> >>> > > > > > > > > from the
> >>> > > > > > > > > > > >> > feature
> >>> > > > > > > > > > > >> > > > > >> flag.
> >>> > > > > > > > > > > >> > > > > >> > > Maybe
> >>> > > > > > > > > > > >> > > > > >> > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > mean
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> >>> default
> >>> > > > for
> >>> > > > > > the
> >>> > > > > > > > > feature
> >>> > > > > > > > > > > >> flag
> >>> > > > > > > > > > > >> > to
> >>> > > > > > > > > > > >> > > > > true
> >>> > > > > > > > > > > >> > > > > >> > > based on
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> metadata.version
> >>> once we
> >>> > > > > > want to
> >>> > > > > > > > > make it
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> default.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> >>> protocol
> >>> > > > APIs
> >>> > > > > > don't
> >>> > > > > > > > > mention
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > >> required
> >>> > > > > > > > > > > >> > > > > >> > > ACLs,
> >>> > > > > > > > > > > >> > > > > >> > > > > it
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add
> that
> >>> for
> >>> > > > > > consistency.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> >>> certainly do
> >>> > > > that.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
> >>> odd that
> >>> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> >>> > > > > > > > > > > >> > > > > >> > requires
> >>> > > > > > > > > > > >> > > > > >> > > > > "Read
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems
> >>> to do
> >>> > > > more
> >>> > > > > > than
> >>> > > > > > > > > reading.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that
> thought
> >>> myself
> >>> > > > as
> >>> > > > > > well,
> >>> > > > > > > > > but in
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > end we
> >>> > > > > > > > > > > >> > > > > >> > could
> >>> > > > > > > > > > > >> > > > > >> > > not
> >>> > > > > > > > > > > >> > > > > >> > > > > > > find a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
> >>> alternative: adding
> >>> > > > > > Write
> >>> > > > > > > > > Group
> >>> > > > > > > > > > > >> seems
> >>> > > > > > > > > > > >> > an
> >>> > > > > > > > > > > >> > > > > >> overkill
> >>> > > > > > > > > > > >> > > > > >> > > here
> >>> > > > > > > > > > > >> > > > > >> > > > > > > since we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> >>> elsewhere (we
> >>> > > > only
> >>> > > > > > have
> >>> > > > > > > > > Read /
> >>> > > > > > > > > > > >> > Delete
> >>> > > > > > > > > > > >> > > > > and
> >>> > > > > > > > > > > >> > > > > >> > > Describe
> >>> > > > > > > > > > > >> > > > > >> > > > > on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like
> >>> to hear
> >>> > > > others
> >>> > > > > > > > > thoughts.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is
> topic
> >>> > > > recreation
> >>> > > > > > > > > handled by
> >>> > > > > > > > > > > >> the
> >>> > > > > > > > > > > >> > > > > consumer
> >>> > > > > > > > > > > >> > > > > >> > with
> >>> > > > > > > > > > > >> > > > > >> > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
> >>> would be
> >>> > > > good
> >>> > > > > > to have
> >>> > > > > > > > > a
> >>> > > > > > > > > > > >> > section on
> >>> > > > > > > > > > > >> > > > > >> this.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with
> regex
> >>> > > > > > subscription
> >>> > > > > > > > > right? Yes
> >>> > > > > > > > > > > >> > we can
> >>> > > > > > > > > > > >> > > > > >> add a
> >>> > > > > > > > > > > >> > > > > >> > > > > section
> >>> > > > > > > > > > > >> > > > > >> > > > > > > about
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
> >>> basically the
> >>> > > > idea
> >>> > > > > > is that
> >>> > > > > > > > > > > >> consumer
> >>> > > > > > > > > > > >> > would
> >>> > > > > > > > > > > >> > > > > be
> >>> > > > > > > > > > > >> > > > > >> > > totally
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol
> >>> as it's
> >>> > > > > > handled all
> >>> > > > > > > > > by the
> >>> > > > > > > > > > > >> > brokers.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> >>> mentions we
> >>> > > > will
> >>> > > > > > write
> >>> > > > > > > > > the
> >>> > > > > > > > > > > >> new
> >>> > > > > > > > > > > >> > > > > >> coordinator
> >>> > > > > > > > > > > >> > > > > >> > in
> >>> > > > > > > > > > > >> > > > > >> > > > > Java.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > Even
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> >>> implementation
> >>> > > > > > detail,
> >>> > > > > > > > > do we
> >>> > > > > > > > > > > >> > plan to
> >>> > > > > > > > > > > >> > > > > >> have a
> >>> > > > > > > > > > > >> > > > > >> > > new
> >>> > > > > > > > > > > >> > > > > >> > > > > gradle
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
> >>> thought about
> >>> > > > > > that. But
> >>> > > > > > > > > I think
> >>> > > > > > > > > > > >> > the
> >>> > > > > > > > > > > >> > > > > answer
> >>> > > > > > > > > > > >> > > > > >> > > should
> >>> > > > > > > > > > > >> > > > > >> > > > > be
> >>> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have
> a
> >>> > > > scalability
> >>> > > > > > goal
> >>> > > > > > > > > when it
> >>> > > > > > > > > > > >> > comes to
> >>> > > > > > > > > > > >> > > > > >> how
> >>> > > > > > > > > > > >> > > > > >> > > many
> >>> > > > > > > > > > > >> > > > > >> > > > > > > members
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol
> >>> can
> >>> > > > support?
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> >>> think we
> >>> > > > > > should
> >>> > > > > > > > > shoot for
> >>> > > > > > > > > > > >> > 1000s of
> >>> > > > > > > > > > > >> > > > > >> > > members.
> >>> > > > > > > > > > > >> > > > > >> > > > > But
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
> >>> also
> >>> > > > depend
> >>> > > > > > on the
> >>> > > > > > > > > offset
> >>> > > > > > > > > > > >> > > > > management
> >>> > > > > > > > > > > >> > > > > >> > > (commit,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of
> the
> >>> > > > coordinator
> >>> > > > > > > > > which we
> >>> > > > > > > > > > > >> did
> >>> > > > > > > > > > > >> > not
> >>> > > > > > > > > > > >> > > > > >> cover in
> >>> > > > > > > > > > > >> > > > > >> > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> >>> number that
> >>> > > > > > applies
> >>> > > > > > > > > > > >> > universally.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
> >>> consider
> >>> > > > having
> >>> > > > > > > > > > > >> > SubscribedTopidIds
> >>> > > > > > > > > > > >> > > > > >> instead
> >>> > > > > > > > > > > >> > > > > >> > of
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> SubscribedTopicNames in
> >>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> >>> > > > > > > > > > > >> > > > > >> > > Is the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > idea
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have
> to
> >>> resolve
> >>> > > > the
> >>> > > > > > regex
> >>> > > > > > > > > on the
> >>> > > > > > > > > > > >> > server,
> >>> > > > > > > > > > > >> > > > > we
> >>> > > > > > > > > > > >> > > > > >> > can
> >>> > > > > > > > > > > >> > > > > >> > > do
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > same
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name?
> >>> The
> >>> > > > > > difference is
> >>> > > > > > > > > that
> >>> > > > > > > > > > > >> > sending the
> >>> > > > > > > > > > > >> > > > > >> > regex
> >>> > > > > > > > > > > >> > > > > >> > > is
> >>> > > > > > > > > > > >> > > > > >> > > > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
> >>> the topic
> >>> > > > > > names is
> >>> > > > > > > > > less
> >>> > > > > > > > > > > >> > efficient.
> >>> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> >>> easier to
> >>> > > > > > handle if
> >>> > > > > > > > > we have
> >>> > > > > > > > > > > >> > topic
> >>> > > > > > > > > > > >> > > > > ids.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason
> to
> >>> still
> >>> > > > let
> >>> > > > > > the
> >>> > > > > > > > > clients
> >>> > > > > > > > > > > >> send
> >>> > > > > > > > > > > >> > names
> >>> > > > > > > > > > > >> > > > > >> is to
> >>> > > > > > > > > > > >> > > > > >> > > keep
> >>> > > > > > > > > > > >> > > > > >> > > > > the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of
> names
> >>> -> ids
> >>> > > > on
> >>> > > > > > the
> >>> > > > > > > > > broker /
> >>> > > > > > > > > > > >> > admin
> >>> > > > > > > > > > > >> > > > > client
> >>> > > > > > > > > > > >> > > > > >> > > only.
> >>> > > > > > > > > > > >> > > > > >> > > > > Note
> >>> > > > > > > > > > > >> > > > > >> > > > > > > that
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> >>> topic id
> >>> > > > in
> >>> > > > > > > > > KIP-516, we
> >>> > > > > > > > > > > >> > never
> >>> > > > > > > > > > > >> > > > > >> > > implemented the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > logic
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> consumer/producers
> >>> > > > leveraging
> >>> > > > > > the
> >>> > > > > > > > > related
> >>> > > > > > > > > > > >> > newer
> >>> > > > > > > > > > > >> > > > > >> versioned
> >>> > > > > > > > > > > >> > > > > >> > > RPCs,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the
> topic
> >>> id as
> >>> > > > empty
> >>> > > > > > UUID.
> >>> > > > > > > > > We
> >>> > > > > > > > > > > >> want
> >>> > > > > > > > > > > >> > to keep
> >>> > > > > > > > > > > >> > > > > >> the
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and
> only
> >>> > > > delegate
> >>> > > > > > the
> >>> > > > > > > > > reasoning
> >>> > > > > > > > > > > >> on
> >>> > > > > > > > > > > >> > broker
> >>> > > > > > > > > > > >> > > > > >> and
> >>> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
> >>> 2022 at
> >>> > > > 10:45
> >>> > > > > > AM
> >>> > > > > > > > > David
> >>> > > > > > > > > > > >> Jacot
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> <dj...@confluent.io.invalid>
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like
> >>> to start a
> >>> > > > > > > > > discussion
> >>> > > > > > > > > > > >> thread
> >>> > > > > > > > > > > >> > on
> >>> > > > > > > > > > > >> > > > > >> KIP-848:
> >>> > > > > > > > > > > >> > > > > >> > > The
> >>> > > > > > > > > > > >> > > > > >> > > > > Next
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of
> >>> the
> >>> > > > Consumer
> >>> > > > > > > > > Rebalance
> >>> > > > > > > > > > > >> > Protocol.
> >>> > > > > > > > > > > >> > > > > >> With
> >>> > > > > > > > > > > >> > > > > >> > > this
> >>> > > > > > > > > > > >> > > > > >> > > > > KIP,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > we
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> >>> rebalance
> >>> > > > > > protocol
> >>> > > > > > > > > (for
> >>> > > > > > > > > > > >> > consumers)
> >>> > > > > > > > > > > >> > > > > >> more
> >>> > > > > > > > > > > >> > > > > >> > > > > reliable,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > more
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
> >>> easier to
> >>> > > > > > implement for
> >>> > > > > > > > > > > >> > clients, and
> >>> > > > > > > > > > > >> > > > > >> easier
> >>> > > > > > > > > > > >> > > > > >> > > to
> >>> > > > > > > > > > > >> > > > > >> > > > > debug
> >>> > > > > > > > > > > >> > > > > >> > > > > > > for
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
> >>> here:
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
> >>> look and
> >>> > > > let
> >>> > > > > > me
> >>> > > > > > > > > know what
> >>> > > > > > > > > > > >> > you
> >>> > > > > > > > > > > >> > > > > think.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
> >>> away from
> >>> > > > > > July
> >>> > > > > > > > > 18th to
> >>> > > > > > > > > > > >> > August 8th.
> >>> > > > > > > > > > > >> > > > > >> That
> >>> > > > > > > > > > > >> > > > > >> > > gives
> >>> > > > > > > > > > > >> > > > > >> > > > > > > you a
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
> >>> read and
> >>> > > > digest
> >>> > > > > > this
> >>> > > > > > > > > long
> >>> > > > > > > > > > > >> KIP.
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > --
> >>> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> >>> > > > > > > > > > > >> > > > > >> > > > > > >
> >>> > > > > > > > > > > >> > > > > >> > > > >
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> > >
> >>> > > > > > > > > > > >> > > > > >> >
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >> --
> >>> > > > > > > > > > > >> > > > > >> -- Guozhang
> >>> > > > > > > > > > > >> > > > > >>
> >>> > > > > > > > > > > >> > > > > >
> >>> > > > > > > > > > > >> > > > >
> >>> > > > > > > > > > > >> >
> >>> > > > > > > > > > > >>
> >>> > > > > > > > > > > >
> >>> > > > > > > > >
> >>> > > > > >
> >>> > > > >
> >>> > > > >
> >>> > > > > --
> >>> > > > > -- Guozhang
> >>> > > >
> >>> > >
> >>> > >
> >>> > > --
> >>> > > -- Guozhang
> >>>
> >>
> >>
> >> --
> >> -- Guozhang
> >>
> >
> >
> > --
> > -- Guozhang
> >
>
>
> --
> -- Guozhang
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hi David,

Another quick thing I just realized about "PartitionAssignor#onAssignment":
in the new proposal, this function could be triggered multiple time within
a group generation as it will be triggered whenever the broker gives the
client new assignments, whereas today it is only triggered once per
rebalance. I think it's worth emphasizing this a bit in the doc, especially
around:

1) how many times `onAssignment` could be triggered with a single rebalance
(especially if the new assignment is decided by the client-assignor).
2) when the Assignment metadata could change across multiple `onAssignment`
calls, and how users should handle it.
3) the order of calling `onAssignment` and `onPartitionsRevoked/Assigned`.

For example, today the Streams client relies on the following assumptions:
1) within a rebalance the onPartitionsRevoked would be called before the
onAssignment, and onAssignment is called before onPartitionsAssigned (today
we would trigger all three only once within a generation), 2) onAssignment
contains the source-of-truth of the final assignment, and hence we can
simply ignore onPartitionsAssigned and just rely on onAssignment to reason
about which tasks are removed / assigned. And each time it calls, the
corresponding metadata change are "final" for this new generation.

We can change however Streams handle the new protocol, but I think there
are other clients externally which may also make such assumptions about the
ordering / call times about those callbacks, and hence worth clarifying
those in the new proposal.



Guozhang




On Tue, Sep 13, 2022 at 12:20 PM Guozhang Wang <wa...@gmail.com> wrote:

> Hello David,
>
> Just had a few more nit questions about the public APIs:
>
> 1) The client-side "PartitionAssignor#Assignment" has an Error byte field,
> while the broker-side "PartitionAssignor#Assignment" does not. And the
> rationale seems to be assuming that we should always be able to do the
> assignment at the broker-side assignor without errors. Personally I think
> it's still potentially beneficial to add the Error field even for
> broker-side assignors, e.g. for some edge cases where some subscribed
> topics are not recognized with the current broker's metadata. What do you
> think?
>
> 2) The client-side "GroupMember" has three additional fields
> reason/version/metadata compared with the broker-side GroupMember. I agree
> that broker-side assignor would not need reason/metadata since they are
> blackbox strings/bytes to the assignor, but what about version? E.g. is it
> possible that we evolve our broker-side built-in assignor but the old
> versioned clients would not be able to work with the new version, in which
> case we need to let the broker being aware of this and upgrade its behavior
> to cooperate with the clients?
>
> 3) Also related to 2) above, for the client-side "GroupMember", instead of
> including these three fields, what about just adding the "Metadata" field
> class which has these three fields? Also, there are two "Metadata"
> currently in the APIs, the first is a class that encodes
> reason/version/metadata, and the second is just the encoded metadata bytes.
> I'm wondering what about just naming the first as memberMetadata, which
> then has a bytebuffer field Metadata, or instead naming the second
> bytebuffer field as metadataBytes?
>
> Guozhang
>
>
>
>
>
>
>
> On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com> wrote:
>
>> Hello David,
>>
>> Thanks for bringing this question up. I think the main benefits as you
>> listed is 2) above if it stays; just to clarify, we would only be able to
>> save one round trip if the rebalance is still triggered by the broker; if
>> the rebalance is triggered by the client then the num.round.trips are the
>> same:
>>
>> 1) With GroupPrepareAssignment:
>>
>> T0: client decides to do a new assignment, suppose it has already sent a
>> HB and hence has to wait for it to return first since only one request /
>> response can be inflight with the coordinator's socket.
>> T1: client receives the HB response, and then sends the
>> GroupPrepareAssignment request.
>> T2: the GroupPrepareAssignment response is returned.
>> T3: it calculates the new assignment, and sends a GroupInstallAssignment
>> request.
>>
>> In total, two round trips.
>>
>> 2) Without GroupPrepareAssignment:
>>
>> T0: client decides to do a new assignment, suppose it has already sent a
>> HB and hence has to wait for it to return first since only one request /
>> response can be inflight with the coordinator's socket.
>> T1: client receives the HB response, and then sends the new HB request
>> with the flag indicating a new rebalance needed..
>> T2: the HB response with the optional member metadata map is returned.
>> T3: it calculates the new assignment, and sends a GroupInstallAssignment
>> request.
>>
>> In total, two round trips as well.
>>
>> -----------------------------
>>
>> So to complete the full picture here, we'd need to modify both HB request
>> and response so that the client can also indicate a new rebalance via the
>> HB request as well, right?
>>
>> Assuming all above is true, I think it's okay to merge the
>> GroupPrepareAssignment into HB given that we can make the additional fields
>> encoding the full member (subscription) metadata and topic metadata as
>> optional fields.
>>
>>
>> Guozhang
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> On Mon, Sep 12, 2022 at 5:22 AM David Jacot <dj...@confluent.io.invalid>
>> wrote:
>>
>>> Hi all,
>>>
>>> During an offline conversation, someone asked why we need the
>>> ConsumerGroupPrepareAssignment API and suggested that we could instead
>>> provide the group state in the heartbeat response. This has a few
>>> advantages: 1) it does not require using a special error code to
>>> signal that a new assignment is required as the signal would be the
>>> provided group state; 2) it removes one round trip when a client side
>>> assignor is used. The downside is that it makes the heartbeat
>>> response's definition quite large. I recall that I went with the
>>> current approach due to this.
>>>
>>> Providing the group state in the heartbeat response is appealing. What
>>> do you guys think?
>>>
>>> Best,
>>> David
>>>
>>> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io> wrote:
>>> >
>>> > Hi Guozhang,
>>> >
>>> > 1. I have added a reference to the relevant chapter instead of
>>> > repeating the whole thing. Does that work for you?
>>> >
>>> > 2. The "Rebalance Triggers" section you are referring to is about when
>>> > a rebalance should be triggered for the non-upgraded members using the
>>> > old protocol. The section mentions that a rebalance must be triggered
>>> > when a new assignment is installed. This implies that the group epoch
>>> > was updated either by a native member or a non-upgraded member. For
>>> > the latter, the JoinGroup request would be the trigger. I have added a
>>> > reference to the relevant chapter in the "JoinGroup Handling" section
>>> > as well. Does that make sense?
>>> >
>>> > Thanks,
>>> > David
>>> >
>>> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
>>> wrote:
>>> > >
>>> > > Hello David,
>>> > >
>>> > > Alright I think that's sufficient. Just to make that clear in the
>>> doc,
>>> > > could we update:
>>> > >
>>> > > 1) the heartbeat request handling section, stating when coordinator
>>> will
>>> > > trigger rebalance based on the HB's member metadata / reason?
>>> > > 2) the "Rebalance Triggers" section to include what we described in
>>> "Group
>>> > > Epoch - Trigger a rebalance" section as well?
>>> > >
>>> > >
>>> > > Guozhang
>>> > >
>>> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
>>> <dj...@confluent.io.invalid>
>>> > > wrote:
>>> > >
>>> > > > Hi Guozhang,
>>> > > >
>>> > > > I thought that the assignor will always be consulted when the next
>>> > > > heartbeat request is constructed. In other words,
>>> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
>>> This
>>> > > > gives the opportunity for the assignor to enforce a rebalance by
>>> > > > setting the reason to a non-zero value or by changing the bytes. Do
>>> > > > you think that this is not sufficient? Are you concerned by the
>>> delay?
>>> > > >
>>> > > > Best,
>>> > > > David
>>> > > >
>>> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
>>> wrote:
>>> > > > >
>>> > > > > Hello David,
>>> > > > >
>>> > > > > One of Jun's comments make me thinking:
>>> > > > >
>>> > > > > ```
>>> > > > > In this case, a new assignment is triggered by the client side
>>> > > > > assignor. When constructing the HB, the consumer will always
>>> consult
>>> > > > > the client side assignor and propagate the information to the
>>> group
>>> > > > > coordinator. In other words, we don't expect users to call
>>> > > > > Consumer#enforceRebalance anymore.
>>> > > > > ```
>>> > > > >
>>> > > > > As I looked at the current PartitionAssignor's interface, we
>>> actually do
>>> > > > > not have a way yet to instruct how to construct the next HB
>>> request, e.g.
>>> > > > > when the assignor wants to enforce a new rebalance with a new
>>> assignment,
>>> > > > > we'd need some customizable APIs inside the PartitionAssignor to
>>> indicate
>>> > > > > the next HB telling broker about so. WDYT about adding such an
>>> API on the
>>> > > > > PartitionAssignor?
>>> > > > >
>>> > > > >
>>> > > > > Guozhang
>>> > > > >
>>> > > > >
>>> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
>>> <dj...@confluent.io.invalid>
>>> > > > > wrote:
>>> > > > >
>>> > > > > > Hi Jun,
>>> > > > > >
>>> > > > > > I have updated the KIP to include your feedback. I have also
>>> tried to
>>> > > > > > clarify the parts which were not cleared.
>>> > > > > >
>>> > > > > > Best,
>>> > > > > > David
>>> > > > > >
>>> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <
>>> djacot@confluent.io>
>>> > > > wrote:
>>> > > > > > >
>>> > > > > > > Hi Jun,
>>> > > > > > >
>>> > > > > > > Thanks for your feedback. Let me start by answering your
>>> questions
>>> > > > > > > inline and I will update the KIP next week.
>>> > > > > > >
>>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
>>> seem to
>>> > > > be
>>> > > > > > fewer
>>> > > > > > > > RPCs during rebalance and more efficient support of
>>> wildcard. A few
>>> > > > > > > > comments below.
>>> > > > > > >
>>> > > > > > > I would also add that the KIP removes the global sync
>>> barrier in the
>>> > > > > > > protocol which is essential to improve group stability and
>>> > > > > > > scalability, and the KIP also simplifies the client by
>>> moving most of
>>> > > > > > > the logic to the server side.
>>> > > > > > >
>>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
>>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
>>> rolling
>>> > > > > > changing
>>> > > > > > > > of the partition assignor in the consumers?
>>> > > > > > >
>>> > > > > > > Definitely. The group coordinator will use the assignor used
>>> by a
>>> > > > > > > majority of the members. This allows the group to move from
>>> one
>>> > > > > > > assignor to another by a roll. This is explained in the
>>> Assignor
>>> > > > > > > Selection chapter.
>>> > > > > > >
>>> > > > > > > > 30.2 For each field, could you explain whether it's
>>> required in
>>> > > > every
>>> > > > > > > > request or the scenarios when it needs to be filled? For
>>> example,
>>> > > > it's
>>> > > > > > not
>>> > > > > > > > clear to me when TopicPartitions needs to be filled.
>>> > > > > > >
>>> > > > > > > The client is expected to set those fields in case of a
>>> connection
>>> > > > > > > issue (e.g. timeout) or when the fields have changed since
>>> the last
>>> > > > > > > HB. The server populates those fields as long as the member
>>> is not
>>> > > > > > > fully reconciled - the member should acknowledge that it has
>>> the
>>> > > > > > > expected epoch and assignment. I will clarify this in the
>>> KIP.
>>> > > > > > >
>>> > > > > > > > 31. In the current consumer protocol, the rack affinity
>>> between the
>>> > > > > > client
>>> > > > > > > > and the broker is only considered during fetching, but not
>>> during
>>> > > > > > assigning
>>> > > > > > > > partitions to consumers. Sometimes, once the assignment is
>>> made,
>>> > > > there
>>> > > > > > is
>>> > > > > > > > no opportunity for read affinity because no replicas of
>>> assigned
>>> > > > > > partitions
>>> > > > > > > > are close to the member. I am wondering if we should use
>>> this
>>> > > > > > opportunity
>>> > > > > > > > to address this by including rack in GroupMember.
>>> > > > > > >
>>> > > > > > > That's an interesting idea. I don't see any issue with
>>> adding the
>>> > > > rack
>>> > > > > > > to the members. I will do so.
>>> > > > > > >
>>> > > > > > > > 32. On the metric side, often, it's useful to know how
>>> busy a group
>>> > > > > > > > coordinator is. By moving the event loop model, it seems
>>> that we
>>> > > > could
>>> > > > > > add
>>> > > > > > > > a metric that tracks the fraction of the time the event
>>> loop is
>>> > > > doing
>>> > > > > > the
>>> > > > > > > > actual work.
>>> > > > > > >
>>> > > > > > > That's a great idea. I will add it. Thanks.
>>> > > > > > >
>>> > > > > > > > 33. Could we add a section on coordinator failover
>>> handling? For
>>> > > > > > example,
>>> > > > > > > > does it need to trigger the check if any group with the
>>> wildcard
>>> > > > > > > > subscription now has a new matching topic?
>>> > > > > > >
>>> > > > > > > Sure. When the new group coordinator takes over, it has to:
>>> > > > > > > * Setup the session timeouts.
>>> > > > > > > * Trigger a new assignment if a client side assignor is
>>> used. We
>>> > > > don't
>>> > > > > > > store the information about the member selected to run the
>>> assignment
>>> > > > > > > so we have to start a new one.
>>> > > > > > > * Update the topics metadata, verify the wildcard
>>> subscriptions, and
>>> > > > > > > trigger a rebalance if needed.
>>> > > > > > >
>>> > > > > > > > 34. ConsumerGroupMetadataValue,
>>> > > > ConsumerGroupPartitionMetadataValue,
>>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
>>> the epoch
>>> > > > > > field
>>> > > > > > > > reflects? For example, does the epoch in
>>> ConsumerGroupMetadataValue
>>> > > > > > reflect
>>> > > > > > > > the latest group epoch? What about the one in
>>> > > > > > > > ConsumerGroupPartitionMetadataValue and
>>> > > > > > ConsumerGroupMemberMetadataValue?
>>> > > > > > >
>>> > > > > > > Sure. I will clarify that but it is always the latest group
>>> epoch.
>>> > > > > > > When the group state is updated, the group epoch is bumped
>>> so we use
>>> > > > > > > that one for all the change records related to the update.
>>> > > > > > >
>>> > > > > > > > 35. "the group coordinator will ensure that the following
>>> > > > invariants
>>> > > > > > are
>>> > > > > > > > met: ... All members exists." It's possible for a member
>>> not to
>>> > > > get any
>>> > > > > > > > assigned partitions, right?
>>> > > > > > >
>>> > > > > > > That's right. Here I meant that the members provided by the
>>> assignor
>>> > > > > > > in the assignment must exist in the group. The assignor can
>>> not make
>>> > > > > > > up new member ids.
>>> > > > > > >
>>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
>>> to 0":
>>> > > > When
>>> > > > > > would
>>> > > > > > > > a consumer rejoin and what member id would be used?
>>> > > > > > >
>>> > > > > > > A member is expected to abandon all its partitions and
>>> rejoins when
>>> > > > it
>>> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
>>> group
>>> > > > > > > coordinator will have removed the member from the group. The
>>> member
>>> > > > > > > can rejoin the group with the same member id but with 0 as
>>> epoch. Let
>>> > > > > > > me see if I can clarify this in the KIP.
>>> > > > > > >
>>> > > > > > > > 37. "Instead, power users will have the ability to trigger
>>> a
>>> > > > > > reassignment
>>> > > > > > > > by either providing a non-zero reason or by updating the
>>> assignor
>>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
>>> deprecation
>>> > > > of
>>> > > > > > > > Consumer#enforeRebalance.
>>> > > > > > >
>>> > > > > > > In this case, a new assignment is triggered by the client
>>> side
>>> > > > > > > assignor. When constructing the HB, the consumer will always
>>> consult
>>> > > > > > > the client side assignor and propagate the information to
>>> the group
>>> > > > > > > coordinator. In other words, we don't expect users to call
>>> > > > > > > Consumer#enforceRebalance anymore.
>>> > > > > > >
>>> > > > > > > > 38. The reassignment examples are nice. But the section
>>> seems to
>>> > > > have
>>> > > > > > > > multiple typos.
>>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
>>> gets into
>>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
>>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
>>> into
>>> > > > > > epoch=3,
>>> > > > > > > > partitions=[foo-1] too early.
>>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
>>> epoch=2,
>>> > > > > > > > partitions=[foo-0].
>>> > > > > > >
>>> > > > > > > Sorry for that! I will revise them.
>>> > > > > > >
>>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
>>> upgrade from
>>> > > > any
>>> > > > > > old
>>> > > > > > > > version to new one?
>>> > > > > > >
>>> > > > > > > We will support upgrading from the consumer protocol version
>>> 3,
>>> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
>>> earliest
>>> > > > > > > version is unknown at the moment. This is explained in the
>>> migration
>>> > > > > > > plan chapter.
>>> > > > > > >
>>> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
>>> based on
>>> > > > it
>>> > > > > > > next week.
>>> > > > > > >
>>> > > > > > > Best,
>>> > > > > > > David
>>> > > > > > >
>>> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
>>> <ju...@confluent.io.invalid>
>>> > > > wrote:
>>> > > > > > > >
>>> > > > > > > > Hi, David,
>>> > > > > > > >
>>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
>>> seem to
>>> > > > be
>>> > > > > > fewer
>>> > > > > > > > RPCs during rebalance and more efficient support of
>>> wildcard. A few
>>> > > > > > > > comments below.
>>> > > > > > > >
>>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
>>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
>>> rolling
>>> > > > > > changing
>>> > > > > > > > of the partition assignor in the consumers?
>>> > > > > > > > 30.2 For each field, could you explain whether it's
>>> required in
>>> > > > every
>>> > > > > > > > request or the scenarios when it needs to be filled? For
>>> example,
>>> > > > it's
>>> > > > > > not
>>> > > > > > > > clear to me when TopicPartitions needs to be filled.
>>> > > > > > > >
>>> > > > > > > > 31. In the current consumer protocol, the rack affinity
>>> between the
>>> > > > > > client
>>> > > > > > > > and the broker is only considered during fetching, but not
>>> during
>>> > > > > > assigning
>>> > > > > > > > partitions to consumers. Sometimes, once the assignment is
>>> made,
>>> > > > there
>>> > > > > > is
>>> > > > > > > > no opportunity for read affinity because no replicas of
>>> assigned
>>> > > > > > partitions
>>> > > > > > > > are close to the member. I am wondering if we should use
>>> this
>>> > > > > > opportunity
>>> > > > > > > > to address this by including rack in GroupMember.
>>> > > > > > > >
>>> > > > > > > > 32. On the metric side, often, it's useful to know how
>>> busy a group
>>> > > > > > > > coordinator is. By moving the event loop model, it seems
>>> that we
>>> > > > could
>>> > > > > > add
>>> > > > > > > > a metric that tracks the fraction of the time the event
>>> loop is
>>> > > > doing
>>> > > > > > the
>>> > > > > > > > actual work.
>>> > > > > > > >
>>> > > > > > > > 33. Could we add a section on coordinator failover
>>> handling? For
>>> > > > > > example,
>>> > > > > > > > does it need to trigger the check if any group with the
>>> wildcard
>>> > > > > > > > subscription now has a new matching topic?
>>> > > > > > > >
>>> > > > > > > > 34. ConsumerGroupMetadataValue,
>>> > > > ConsumerGroupPartitionMetadataValue,
>>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
>>> the epoch
>>> > > > > > field
>>> > > > > > > > reflects? For example, does the epoch in
>>> ConsumerGroupMetadataValue
>>> > > > > > reflect
>>> > > > > > > > the latest group epoch? What about the one in
>>> > > > > > > > ConsumerGroupPartitionMetadataValue and
>>> > > > > > ConsumerGroupMemberMetadataValue?
>>> > > > > > > >
>>> > > > > > > > 35. "the group coordinator will ensure that the following
>>> > > > invariants
>>> > > > > > are
>>> > > > > > > > met: ... All members exists." It's possible for a member
>>> not to
>>> > > > get any
>>> > > > > > > > assigned partitions, right?
>>> > > > > > > >
>>> > > > > > > > 36. "He can rejoins the group with a member epoch equals
>>> to 0":
>>> > > > When
>>> > > > > > would
>>> > > > > > > > a consumer rejoin and what member id would be used?
>>> > > > > > > >
>>> > > > > > > > 37. "Instead, power users will have the ability to trigger
>>> a
>>> > > > > > reassignment
>>> > > > > > > > by either providing a non-zero reason or by updating the
>>> assignor
>>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
>>> deprecation
>>> > > > of
>>> > > > > > > > Consumer#enforeRebalance.
>>> > > > > > > >
>>> > > > > > > > 38. The reassignment examples are nice. But the section
>>> seems to
>>> > > > have
>>> > > > > > > > multiple typos.
>>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
>>> gets into
>>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
>>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
>>> into
>>> > > > > > epoch=3,
>>> > > > > > > > partitions=[foo-1] too early.
>>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
>>> epoch=2,
>>> > > > > > > > partitions=[foo-0].
>>> > > > > > > >
>>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the
>>> upgrade from
>>> > > > any
>>> > > > > > old
>>> > > > > > > > version to new one?
>>> > > > > > > >
>>> > > > > > > > Thanks,
>>> > > > > > > >
>>> > > > > > > > Jun
>>> > > > > > > >
>>> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
>>> > > > > > <dj...@confluent.io.invalid>
>>> > > > > > > > wrote:
>>> > > > > > > >
>>> > > > > > > > > Hi all,
>>> > > > > > > > >
>>> > > > > > > > > The KIP states that we will re-implement the coordinator
>>> in
>>> > > > Java. I
>>> > > > > > > > > discussed this offline with a few folks and folks are
>>> concerned
>>> > > > that
>>> > > > > > > > > we could introduce many regressions in the old protocol
>>> if we do
>>> > > > so.
>>> > > > > > > > > Therefore, I am going to remove this statement from the
>>> KIP. It
>>> > > > is an
>>> > > > > > > > > implementation detail after all so it does not have to be
>>> > > > decided at
>>> > > > > > > > > this stage. We will likely start by trying to refactor
>>> the
>>> > > > current
>>> > > > > > > > > implementation as a first step.
>>> > > > > > > > >
>>> > > > > > > > > Cheers,
>>> > > > > > > > > David
>>> > > > > > > > >
>>> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
>>> djacot@confluent.io
>>> > > > >
>>> > > > > > wrote:
>>> > > > > > > > > >
>>> > > > > > > > > > Hi Luke,
>>> > > > > > > > > >
>>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
>>> assigning,
>>> > > > > > reconciling,
>>> > > > > > > > > stable,
>>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
>>> right?
>>> > > > > > > > > >
>>> > > > > > > > > > This sentence does not refer to those group states but
>>> rather
>>> > > > to a
>>> > > > > > > > > > state machine replication (SMR). This refers to the
>>> entire
>>> > > > state of
>>> > > > > > > > > > group coordinator which is replicated via the log
>>> layer. I will
>>> > > > > > > > > > clarify this in the KIP.
>>> > > > > > > > > >
>>> > > > > > > > > > > 1.2. What do you mean "each state machine is
>>> modelled as an
>>> > > > event
>>> > > > > > > > > loop"?
>>> > > > > > > > > >
>>> > > > > > > > > > The idea is to follow a model similar to the new quorum
>>> > > > > > controller. We
>>> > > > > > > > > > will have N threads to process events. Each
>>> __consumer_offsets
>>> > > > > > > > > > partition is assigned to a unique thread and all the
>>> events
>>> > > > (e.g.
>>> > > > > > > > > > requests, callbacks, etc.) are processed by this
>>> thread. This
>>> > > > > > simplify
>>> > > > > > > > > > concurrency and will enable us to do simulation
>>> testing for the
>>> > > > > > group
>>> > > > > > > > > > coordinator.
>>> > > > > > > > > >
>>> > > > > > > > > > > 1.3. Why do we need a state machine per
>>> *__consumer_offsets*
>>> > > > > > > > > partitions?
>>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
>>> group
>>> > > > > > coordinator?
>>> > > > > > > > > For
>>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
>>> groups, and
>>> > > > > > both
>>> > > > > > > > > exist in
>>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
>>> for it,
>>> > > > or
>>> > > > > > 2?
>>> > > > > > > > > >
>>> > > > > > > > > > See 1.1. The confusion comes from there, I think.
>>> > > > > > > > > >
>>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
>>> number
>>> > > > of
>>> > > > > > threads
>>> > > > > > > > > used
>>> > > > > > > > > > > to run the state machines. But I'm wondering if the
>>> purpose
>>> > > > of
>>> > > > > > the
>>> > > > > > > > > threads
>>> > > > > > > > > > > is only to keep the state of each consumer group (or
>>> > > > > > > > > *__consumer_offsets*
>>> > > > > > > > > > > partitions?), and no heavy computation, why should
>>> we need
>>> > > > > > > > > multi-threads
>>> > > > > > > > > > > here?
>>> > > > > > > > > >
>>> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
>>> > > > processing as
>>> > > > > > the
>>> > > > > > > > > > computation could be heavy.
>>> > > > > > > > > >
>>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
>>> default
>>> > > > session
>>> > > > > > > > > timeout not
>>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
>>> min/max
>>> > > > > > session
>>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
>>> > > > > > > > > > >
>>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
>>> timeout to
>>> > > > detect
>>> > > > > > client
>>> > > > > > > > > > > failures when using the consumer group protocol.
>>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
>>> minimum
>>> > > > > > session
>>> > > > > > > > > timeout.
>>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
>>> maximum
>>> > > > > > session
>>> > > > > > > > > timeout.
>>> > > > > > > > > >
>>> > > > > > > > > > This is indeed a mistake. The default session timeout
>>> should
>>> > > > be 45s
>>> > > > > > > > > > (the current default).
>>> > > > > > > > > >
>>> > > > > > > > > > > 2.2. The default server side assignor are [range,
>>> uniform],
>>> > > > > > which means
>>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
>>> why not
>>> > > > > > uniform
>>> > > > > > > > > one? I
>>> > > > > > > > > > > thought usually users will choose uniform assignor
>>> (former
>>> > > > sticky
>>> > > > > > > > > assinor)
>>> > > > > > > > > > > for better evenly distribution. Any other reason we
>>> choose
>>> > > > range
>>> > > > > > > > > assignor
>>> > > > > > > > > > > as default?
>>> > > > > > > > > > > group.consumer.assignors List range, uniform The
>>> server side
>>> > > > > > assignors.
>>> > > > > > > > > >
>>> > > > > > > > > > The order on the server side has no influence because
>>> the
>>> > > > client
>>> > > > > > must
>>> > > > > > > > > > chose the selector that he wants to use. There is no
>>> default
>>> > > > in the
>>> > > > > > > > > > current proposal. If the assignor is not specified by
>>> the
>>> > > > client,
>>> > > > > > the
>>> > > > > > > > > > request is rejected. The default client value for
>>> > > > > > > > > > `group.remote.assignor` is `uniform` though.
>>> > > > > > > > > >
>>> > > > > > > > > > Thanks for your very good comments, Luke. I hope that
>>> my
>>> > > > answers
>>> > > > > > help
>>> > > > > > > > > > to clarify things. I will update the KIP as well based
>>> on your
>>> > > > > > > > > > feedback.
>>> > > > > > > > > >
>>> > > > > > > > > > Cheers,
>>> > > > > > > > > > David
>>> > > > > > > > > >
>>> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
>>> showuon@gmail.com>
>>> > > > > > wrote:
>>> > > > > > > > > > >
>>> > > > > > > > > > > Hi David,
>>> > > > > > > > > > >
>>> > > > > > > > > > > Thanks for the update.
>>> > > > > > > > > > >
>>> > > > > > > > > > > Some more questions:
>>> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
>>> > > > > > > > > > > > The new group coordinator will have a state
>>> machine per
>>> > > > > > > > > > > *__consumer_offsets* partitions, where each state
>>> machine is
>>> > > > > > modelled
>>> > > > > > > > > as an
>>> > > > > > > > > > > event loop. Those state machines will be executed in
>>> > > > > > > > > > > *group.coordinator.threads* threads.
>>> > > > > > > > > > >
>>> > > > > > > > > > > 1.1. I think the state machine are: "Empty,
>>> assigning,
>>> > > > > > reconciling,
>>> > > > > > > > > stable,
>>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
>>> right?
>>> > > > > > > > > > > 1.2. What do you mean "each state machine is
>>> modelled as an
>>> > > > event
>>> > > > > > > > > loop"?
>>> > > > > > > > > > > 1.3. Why do we need a state machine per
>>> *__consumer_offsets*
>>> > > > > > > > > partitions?
>>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
>>> group
>>> > > > > > coordinator?
>>> > > > > > > > > For
>>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
>>> groups, and
>>> > > > > > both
>>> > > > > > > > > exist in
>>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
>>> for it,
>>> > > > or
>>> > > > > > 2?
>>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
>>> number
>>> > > > of
>>> > > > > > threads
>>> > > > > > > > > used
>>> > > > > > > > > > > to run the state machines. But I'm wondering if the
>>> purpose
>>> > > > of
>>> > > > > > the
>>> > > > > > > > > threads
>>> > > > > > > > > > > is only to keep the state of each consumer group (or
>>> > > > > > > > > *__consumer_offsets*
>>> > > > > > > > > > > partitions?), and no heavy computation, why should
>>> we need
>>> > > > > > > > > multi-threads
>>> > > > > > > > > > > here?
>>> > > > > > > > > > >
>>> > > > > > > > > > > 2. For the default value in the new configs:
>>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
>>> default
>>> > > > session
>>> > > > > > > > > timeout not
>>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
>>> min/max
>>> > > > > > session
>>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
>>> > > > > > > > > > >
>>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
>>> timeout to
>>> > > > detect
>>> > > > > > client
>>> > > > > > > > > > > failures when using the consumer group protocol.
>>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
>>> minimum
>>> > > > > > session
>>> > > > > > > > > timeout.
>>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
>>> maximum
>>> > > > > > session
>>> > > > > > > > > timeout.
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > > 2.2. The default server side assignor are [range,
>>> uniform],
>>> > > > > > which means
>>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
>>> why not
>>> > > > > > uniform
>>> > > > > > > > > one? I
>>> > > > > > > > > > > thought usually users will choose uniform assignor
>>> (former
>>> > > > sticky
>>> > > > > > > > > assinor)
>>> > > > > > > > > > > for better evenly distribution. Any other reason we
>>> choose
>>> > > > range
>>> > > > > > > > > assignor
>>> > > > > > > > > > > as default?
>>> > > > > > > > > > > group.consumer.assignors List range, uniform The
>>> server side
>>> > > > > > assignors.
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > > Thank you.
>>> > > > > > > > > > > Luke
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > >
>>> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
>>> showuon@gmail.com
>>> > > > >
>>> > > > > > wrote:
>>> > > > > > > > > > >
>>> > > > > > > > > > > > Hi Sagar,
>>> > > > > > > > > > > >
>>> > > > > > > > > > > > I have some thoughts about Kafka Connect
>>> integrating with
>>> > > > > > KIP-848,
>>> > > > > > > > > but I
>>> > > > > > > > > > > > think we should have a separate discussion thread
>>> for the
>>> > > > Kafka
>>> > > > > > > > > Connect
>>> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
>>> Rebalance
>>> > > > > > Protocol
>>> > > > > > > > > [1],
>>> > > > > > > > > > > > and let this discussion thread focus on consumer
>>> rebalance
>>> > > > > > protocol,
>>> > > > > > > > > WDYT?
>>> > > > > > > > > > > >
>>> > > > > > > > > > > > [1]
>>> > > > > > > > > > > >
>>> > > > > > > > >
>>> > > > > >
>>> > > >
>>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>>> > > > > > > > > > > >
>>> > > > > > > > > > > > Thank you.
>>> > > > > > > > > > > > Luke
>>> > > > > > > > > > > >
>>> > > > > > > > > > > >
>>> > > > > > > > > > > >
>>> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
>>> > > > > > sagarmeansocean@gmail.com>
>>> > > > > > > > > wrote:
>>> > > > > > > > > > > >
>>> > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
>>> like
>>> > > > there's
>>> > > > > > > > > agreement on
>>> > > > > > > > > > > >> using separate APIs for Connect. I would revisit
>>> the doc
>>> > > > and
>>> > > > > > see
>>> > > > > > > > > what
>>> > > > > > > > > > > >> changes are to be made.
>>> > > > > > > > > > > >>
>>> > > > > > > > > > > >> Thanks!
>>> > > > > > > > > > > >> Sagar.
>>> > > > > > > > > > > >>
>>> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
>>> > > > > > > > > <dj...@confluent.io.invalid>
>>> > > > > > > > > > > >> wrote:
>>> > > > > > > > > > > >>
>>> > > > > > > > > > > >> > Hi Sagar,
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >> > Thanks for the feedback and the document.
>>> That's really
>>> > > > > > helpful. I
>>> > > > > > > > > > > >> > will take a look at it.
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >> > Overall, it seems to me that both Connect and
>>> the
>>> > > > Consumer
>>> > > > > > could
>>> > > > > > > > > share
>>> > > > > > > > > > > >> > the same underlying "engine". The main
>>> difference is
>>> > > > that
>>> > > > > > the
>>> > > > > > > > > Consumer
>>> > > > > > > > > > > >> > assigns topic-partitions to members whereas
>>> Connect
>>> > > > assigns
>>> > > > > > tasks
>>> > > > > > > > > to
>>> > > > > > > > > > > >> > workers. I see two ways to move forward:
>>> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
>>> different
>>> > > > > > resource
>>> > > > > > > > > types
>>> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
>>> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
>>> dedicated
>>> > > > APIs
>>> > > > > > > > > would be
>>> > > > > > > > > > > >> > similar to the new ones but different on the
>>> > > > > > content/resources and
>>> > > > > > > > > > > >> > they would rely on the same engine on the
>>> coordinator
>>> > > > side.
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >> > I personally lean towards 2) because I am not a
>>> fan of
>>> > > > > > > > > overcharging
>>> > > > > > > > > > > >> > APIs to serve different purposes. That being
>>> said, I am
>>> > > > not
>>> > > > > > > > > opposed to
>>> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >> > I think that we can continue to discuss it here
>>> for now
>>> > > > in
>>> > > > > > order
>>> > > > > > > > > to
>>> > > > > > > > > > > >> > ensure that this KIP is compatible with what we
>>> will do
>>> > > > for
>>> > > > > > > > > Connect in
>>> > > > > > > > > > > >> > the future.
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >> > Best,
>>> > > > > > > > > > > >> > David
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
>>> > > > > > djacot@confluent.io>
>>> > > > > > > > > wrote:
>>> > > > > > > > > > > >> > >
>>> > > > > > > > > > > >> > > Hi all,
>>> > > > > > > > > > > >> > >
>>> > > > > > > > > > > >> > > I am back from vacation. I will go through
>>> and address
>>> > > > > > your
>>> > > > > > > > > comments
>>> > > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
>>> > > > > > > > > > > >> > >
>>> > > > > > > > > > > >> > > Cheers,
>>> > > > > > > > > > > >> > > David
>>> > > > > > > > > > > >> > >
>>> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory
>>> Harris <
>>> > > > > > > > > gharris1727@gmail.com
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >> > wrote:
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > Hey All!
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
>>> > > > cooperative
>>> > > > > > > > > rebalancing
>>> > > > > > > > > > > >> > making it
>>> > > > > > > > > > > >> > > > down the stack!
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > I had a few questions:
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
>>> describes how
>>> > > > > > member
>>> > > > > > > > > epoch
>>> > > > > > > > > > > >> > should
>>> > > > > > > > > > > >> > > > advance in step with the group epoch and
>>> assignment
>>> > > > > > epoch
>>> > > > > > > > > values. I
>>> > > > > > > > > > > >> > think
>>> > > > > > > > > > > >> > > > that this is a good idea for the reasons
>>> described
>>> > > > in
>>> > > > > > the
>>> > > > > > > > > KIP. When
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > protocol is incrementally assigning
>>> partitions to a
>>> > > > > > worker,
>>> > > > > > > > > what
>>> > > > > > > > > > > >> member
>>> > > > > > > > > > > >> > > > epoch does each incremental assignment use?
>>> Are
>>> > > > member
>>> > > > > > epochs
>>> > > > > > > > > > > >> re-used,
>>> > > > > > > > > > > >> > and
>>> > > > > > > > > > > >> > > > a single member epoch can correspond to
>>> multiple
>>> > > > > > different
>>> > > > > > > > > > > >> > (monotonically
>>> > > > > > > > > > > >> > > > larger) assignments?
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque
>>> to the
>>> > > > group
>>> > > > > > > > > > > >> coordinator? If
>>> > > > > > > > > > > >> > > > not, should custom client-side assignor
>>> > > > implementations
>>> > > > > > > > > interact
>>> > > > > > > > > > > >> with
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > Reason field, and how is its common meaning
>>> agreed
>>> > > > > > upon? If
>>> > > > > > > > > so, what
>>> > > > > > > > > > > >> > is the
>>> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
>>> including
>>> > > > such
>>> > > > > > > > > functionality
>>> > > > > > > > > > > >> > in the
>>> > > > > > > > > > > >> > > > opaque metadata?
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
>>> "Thanks to
>>> > > > > > this, the
>>> > > > > > > > > input
>>> > > > > > > > > > > >> of
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > client side assignor is entirely driven by
>>> the group
>>> > > > > > > > > coordinator.
>>> > > > > > > > > > > >> The
>>> > > > > > > > > > > >> > > > consumer is no longer responsible for
>>> maintaining
>>> > > > any
>>> > > > > > state
>>> > > > > > > > > besides
>>> > > > > > > > > > > >> its
>>> > > > > > > > > > > >> > > > assigned partitions." Does this mean that
>>> the
>>> > > > > > client-side
>>> > > > > > > > > assignor
>>> > > > > > > > > > > >> MAY
>>> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
>>> (such as
>>> > > > > > partition
>>> > > > > > > > > > > >> > throughput,
>>> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or
>>> that
>>> > > > > > additional
>>> > > > > > > > > > > >> > non-Metadata
>>> > > > > > > > > > > >> > > > state SHOULD NOT be used?
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > 4. I see that there are separate classes
>>> > > > > > > > > > > >> > > > for
>>> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
>>> > > > > > > > > > > >> > > > and
>>> > > > org.apache.kafka.clients.consumer.PartitionAssignor
>>> > > > > > that
>>> > > > > > > > > seem to
>>> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
>>> these two
>>> > > > > > > > > implementations
>>> > > > > > > > > > > >> to
>>> > > > > > > > > > > >> > be
>>> > > > > > > > > > > >> > > > unified? This would serve to promote
>>> feature parity
>>> > > > of
>>> > > > > > > > > server-side
>>> > > > > > > > > > > >> and
>>> > > > > > > > > > > >> > > > client-side assignors, and would also
>>> facilitate
>>> > > > > > operational
>>> > > > > > > > > > > >> > flexibility in
>>> > > > > > > > > > > >> > > > certain situations. For example, if a
>>> server-side
>>> > > > > > assignor
>>> > > > > > > > > has some
>>> > > > > > > > > > > >> > poor
>>> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
>>> patched
>>> > > > > > assignor to
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > client
>>> > > > > > > > > > > >> > > > and switching one consumer group to a
>>> client-side
>>> > > > > > assignor
>>> > > > > > > > > may be
>>> > > > > > > > > > > >> > faster
>>> > > > > > > > > > > >> > > > and less risky than patching all of the
>>> brokers.
>>> > > > With
>>> > > > > > the
>>> > > > > > > > > currently
>>> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
>>> > > > reimplementation
>>> > > > > > would
>>> > > > > > > > > have
>>> > > > > > > > > > > >> to be
>>> > > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
>>> > > > > > significantly,
>>> > > > > > > > > then it
>>> > > > > > > > > > > >> is
>>> > > > > > > > > > > >> > > > possible that a reimplementation would not
>>> be
>>> > > > possible.
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > --
>>> > > > > > > > > > > >> > > > Greg Harris
>>> > > > > > > > > > > >> > > > gharris1727@gmail.com
>>> > > > > > > > > > > >> > > > github.com/gharris1727
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
>>> > > > > > > > > sagarmeansocean@gmail.com>
>>> > > > > > > > > > > >> > wrote:
>>> > > > > > > > > > > >> > > >
>>> > > > > > > > > > > >> > > > > Hi Guozhang/David,
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > I created a confluence page to discuss
>>> how Connect
>>> > > > > > would
>>> > > > > > > > > need to
>>> > > > > > > > > > > >> > change
>>> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
>>> Here's the
>>> > > > page:
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >>
>>> > > > > > > > >
>>> > > > > >
>>> > > >
>>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > It's also pretty longish and I have tried
>>> to keep
>>> > > > a
>>> > > > > > format
>>> > > > > > > > > > > >> similar to
>>> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think.
>>> Also, do you
>>> > > > > > think this
>>> > > > > > > > > > > >> should
>>> > > > > > > > > > > >> > be
>>> > > > > > > > > > > >> > > > > moved to a separate discussion thread or
>>> is this
>>> > > > one
>>> > > > > > fine?
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > Thanks!
>>> > > > > > > > > > > >> > > > > Sagar.
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
>>> > > > > > > > > sagarmeansocean@gmail.com>
>>> > > > > > > > > > > >> > wrote:
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > > Hello Guozhang,
>>> > > > > > > > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
>>> Streams.
>>> > > > > > Sure, I
>>> > > > > > > > > would do
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > > analysis and come up with such a
>>> document.
>>> > > > > > > > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > > Thanks!
>>> > > > > > > > > > > >> > > > > > Sagar.
>>> > > > > > > > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM
>>> Guozhang Wang <
>>> > > > > > > > > > > >> wangguoz@gmail.com>
>>> > > > > > > > > > > >> > > > > wrote:
>>> > > > > > > > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> Hello Sagar,
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > > >> It would be great if you could come
>>> back with
>>> > > > some
>>> > > > > > > > > analysis on
>>> > > > > > > > > > > >> > how to
>>> > > > > > > > > > > >> > > > > >> implement the Connect side integration
>>> with
>>> > > > the new
>>> > > > > > > > > protocol;
>>> > > > > > > > > > > >> so
>>> > > > > > > > > > > >> > far
>>> > > > > > > > > > > >> > > > > >> besides leveraging on the new
>>> "protocol type"
>>> > > > we
>>> > > > > > did not
>>> > > > > > > > > yet
>>> > > > > > > > > > > >> think
>>> > > > > > > > > > > >> > > > > through
>>> > > > > > > > > > > >> > > > > >> the Connect side implementations. For
>>> Streams
>>> > > > > > here's a
>>> > > > > > > > > draft of
>>> > > > > > > > > > > >> > > > > >> integration
>>> > > > > > > > > > > >> > > > > >> plan:
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >>
>>> > > > > > > > >
>>> > > > > >
>>> > > >
>>> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
>>> > > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar
>>> <
>>> > > > > > > > > > > >> sagarmeansocean@gmail.com
>>> > > > > > > > > > > >> > >
>>> > > > > > > > > > > >> > > > > wrote:
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > > >> > Hi David,
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
>>> reason I
>>> > > > thought
>>> > > > > > > > > connect can
>>> > > > > > > > > > > >> > also fit
>>> > > > > > > > > > > >> > > > > >> into
>>> > > > > > > > > > > >> > > > > >> > this new scheme is that even today
>>> the
>>> > > > connect
>>> > > > > > uses a
>>> > > > > > > > > > > >> > > > > WorkerCoordinator
>>> > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator
>>> to empower
>>> > > > > > > > > rebalances of
>>> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
>>> WorkerCoordinator sets
>>> > > > the
>>> > > > > > > > > > > >> protocolType()
>>> > > > > > > > > > > >> > to
>>> > > > > > > > > > > >> > > > > >> connect
>>> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
>>> plumbing
>>> > > > into
>>> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> > I think the changes to support
>>> connect would
>>> > > > be
>>> > > > > > > > > similar at a
>>> > > > > > > > > > > >> > high
>>> > > > > > > > > > > >> > > > > level
>>> > > > > > > > > > > >> > > > > >> to
>>> > > > > > > > > > > >> > > > > >> > the changes in streams mainly
>>> because of the
>>> > > > > > Client
>>> > > > > > > > > side
>>> > > > > > > > > > > >> > assignors
>>> > > > > > > > > > > >> > > > > being
>>> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
>>> level, we
>>> > > > > > might
>>> > > > > > > > > need to
>>> > > > > > > > > > > >> make
>>> > > > > > > > > > > >> > a lot
>>> > > > > > > > > > > >> > > > > of
>>> > > > > > > > > > > >> > > > > >> > changes to get onto this new
>>> assignment
>>> > > > protocol
>>> > > > > > like
>>> > > > > > > > > > > >> enhancing
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
>>> SyncGroup and
>>> > > > > > using
>>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
>>> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
>>> streams (or
>>> > > > > > there
>>> > > > > > > > > might be
>>> > > > > > > > > > > >> > > > > >> deviations). I
>>> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
>>> analysis of
>>> > > > the
>>> > > > > > same
>>> > > > > > > > > and we
>>> > > > > > > > > > > >> > can have
>>> > > > > > > > > > > >> > > > > a
>>> > > > > > > > > > > >> > > > > >> > separate discussion thread for that
>>> as that
>>> > > > would
>>> > > > > > > > > derail this
>>> > > > > > > > > > > >> > > > > discussion
>>> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
>>> good to
>>> > > > you.
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> > Thanks!
>>> > > > > > > > > > > >> > > > > >> > Sagar.
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM
>>> David Jacot
>>> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> > wrote:
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
>>> `Assignment#error`.
>>> > > > > > Sure, I
>>> > > > > > > > > can
>>> > > > > > > > > > > >> > mention it.
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C
>>> from his
>>> > > > current
>>> > > > > > > > > assignment
>>> > > > > > > > > > > >> to
>>> > > > > > > > > > > >> > his
>>> > > > > > > > > > > >> > > > > >> > > target assignment when he can move
>>> to
>>> > > > epoch 3.
>>> > > > > > When
>>> > > > > > > > > that
>>> > > > > > > > > > > >> > happens,
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
>>> persisted
>>> > > > > > with all
>>> > > > > > > > > its
>>> > > > > > > > > > > >> > assigned
>>> > > > > > > > > > > >> > > > > >> > > partitions even if they are not
>>> all revoked
>>> > > > > > yet. In
>>> > > > > > > > > other
>>> > > > > > > > > > > >> > words, the
>>> > > > > > > > > > > >> > > > > >> > > member assignment becomes the
>>> target
>>> > > > > > assignment.
>>> > > > > > > > > This is
>>> > > > > > > > > > > >> > basically
>>> > > > > > > > > > > >> > > > > an
>>> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
>>> write all
>>> > > > the
>>> > > > > > > > > changes to
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > log.
>>> > > > > > > > > > > >> > > > > The
>>> > > > > > > > > > > >> > > > > >> > > examples are based on the
>>> persisted state
>>> > > > so I
>>> > > > > > > > > understand
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
>>> improve
>>> > > > this in
>>> > > > > > the
>>> > > > > > > > > > > >> > description.
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
>>> reuse the
>>> > > > > > protocol
>>> > > > > > > > > with a
>>> > > > > > > > > > > >> > client side
>>> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
>>> protocol. The
>>> > > > > > assignment
>>> > > > > > > > > is
>>> > > > > > > > > > > >> about
>>> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata,
>>> could
>>> > > > Connect
>>> > > > > > fit
>>> > > > > > > > > into this?
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> > > Best,
>>> > > > > > > > > > > >> > > > > >> > > David
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
>>> Sagar <
>>> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
>>> > > > > > > > > > > >> > > > > >> wrote:
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > Hi David,
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
>>> minor
>>> > > > > > observations:
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
>>> section in
>>> > > > Client
>>> > > > > > Side
>>> > > > > > > > > mode
>>> > > > > > > > > > > >> > Assignment
>>> > > > > > > > > > > >> > > > > >> > > process,
>>> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case,
>>> the
>>> > > > client
>>> > > > > > side
>>> > > > > > > > > assignor
>>> > > > > > > > > > > >> can
>>> > > > > > > > > > > >> > > > > return
>>> > > > > > > > > > > >> > > > > >> an
>>> > > > > > > > > > > >> > > > > >> > > > error to the group coordinator`.
>>> In this
>>> > > > > > case are
>>> > > > > > > > > you
>>> > > > > > > > > > > >> > referring to
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
>>> AssignmentError
>>> > > > that's
>>> > > > > > > > > listed down
>>> > > > > > > > > > > >> > towards
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > end?
>>> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
>>> make sense
>>> > > > to
>>> > > > > > > > > mention this
>>> > > > > > > > > > > >> > > > > explicitly
>>> > > > > > > > > > > >> > > > > >> > here?
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section,
>>> I have a
>>> > > > > > slight
>>> > > > > > > > > > > >> confusion,
>>> > > > > > > > > > > >> > not
>>> > > > > > > > > > > >> > > > > sure
>>> > > > > > > > > > > >> > > > > >> if
>>> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
>>> this step:
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
>>> coordinator
>>> > > > > > > > > transitions him
>>> > > > > > > > > > > >> to
>>> > > > > > > > > > > >> > epoch
>>> > > > > > > > > > > >> > > > > 3
>>> > > > > > > > > > > >> > > > > >> > > because
>>> > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
>>> > > > persists
>>> > > > > > the
>>> > > > > > > > > change and
>>> > > > > > > > > > > >> > reply.
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
>>> > > > > > > > > > > >> > > > > >> > > >       - A
>>> > > > > > > > > > > >> > > > > >> > > >       - B
>>> > > > > > > > > > > >> > > > > >> > > >       - C
>>> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
>>> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
>>> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
>>> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
>>> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
>>> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
>>> partitions=[foo-0,
>>> > > > > > foo-1]
>>> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
>>> partitions=[foo-2]
>>> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
>>> partitions=[foo-1]
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it
>>> transitions to
>>> > > > epoch 3
>>> > > > > > but
>>> > > > > > > > > cannot
>>> > > > > > > > > > > >> get
>>> > > > > > > > > > > >> > foo-1
>>> > > > > > > > > > > >> > > > > >> yet.
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member
>>> C can't
>>> > > > get
>>> > > > > > the
>>> > > > > > > > > foo-1
>>> > > > > > > > > > > >> > partition
>>> > > > > > > > > > > >> > > > > yet,
>>> > > > > > > > > > > >> > > > > >> > but
>>> > > > > > > > > > > >> > > > > >> > > > based on the description above,
>>> it seems
>>> > > > it
>>> > > > > > > > > already has
>>> > > > > > > > > > > >> it.
>>> > > > > > > > > > > >> > Do you
>>> > > > > > > > > > > >> > > > > >> > think
>>> > > > > > > > > > > >> > > > > >> > > it
>>> > > > > > > > > > > >> > > > > >> > > > would be better to remove it and
>>> > > > populate it
>>> > > > > > only
>>> > > > > > > > > when it
>>> > > > > > > > > > > >> > actually
>>> > > > > > > > > > > >> > > > > >> gets
>>> > > > > > > > > > > >> > > > > >> > > it?
>>> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
>>> places, so
>>> > > > have
>>> > > > > > I
>>> > > > > > > > > > > >> understood it
>>> > > > > > > > > > > >> > > > > >> > incorrectly
>>> > > > > > > > > > > >> > > > > >> > > ?
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
>>> out of
>>> > > > scope
>>> > > > > > of
>>> > > > > > > > > this
>>> > > > > > > > > > > >> > discussion,
>>> > > > > > > > > > > >> > > > > but
>>> > > > > > > > > > > >> > > > > >> > from
>>> > > > > > > > > > > >> > > > > >> > > > what I understood it would
>>> probably be
>>> > > > > > running in
>>> > > > > > > > > client
>>> > > > > > > > > > > >> > side
>>> > > > > > > > > > > >> > > > > >> assignor
>>> > > > > > > > > > > >> > > > > >> > > mode
>>> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
>>> protocol as it
>>> > > > has
>>> > > > > > its
>>> > > > > > > > > own
>>> > > > > > > > > > > >> Custom
>>> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
>>> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > Thanks!
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > Sagar.
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
>>> David
>>> > > > Jacot
>>> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
>>> move
>>> > > > forward
>>> > > > > > with
>>> > > > > > > > > > > >> > specialized API
>>> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
>>> generic API.
>>> > > > For
>>> > > > > > > > > Connect, we
>>> > > > > > > > > > > >> > can apply
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
>>> reuse/share the
>>> > > > core
>>> > > > > > > > > > > >> > implementation on
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
>>> registry, I
>>> > > > > > think
>>> > > > > > > > > that we
>>> > > > > > > > > > > >> > should
>>> > > > > > > > > > > >> > > > > >> consider
>>> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
>>> simple
>>> > > > > > > > > membership/leader
>>> > > > > > > > > > > >> > election.
>>> > > > > > > > > > > >> > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > Best,
>>> > > > > > > > > > > >> > > > > >> > > > > David
>>> > > > > > > > > > > >> > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22
>>> AM Ismael
>>> > > > > > Juma <
>>> > > > > > > > > > > >> > ismael@juma.me.uk
>>> > > > > > > > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
>>> > > > > > > > > > > >> > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
>>> java.util.regex.Pattern
>>> > > > vs
>>> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
>>> > > > > > > > > > > >> > > > > >> > > > > should
>>> > > > > > > > > > > >> > > > > >> > > > > > document the differences in
>>> more
>>> > > > detail
>>> > > > > > before
>>> > > > > > > > > > > >> deciding
>>> > > > > > > > > > > >> > one
>>> > > > > > > > > > > >> > > > > way
>>> > > > > > > > > > > >> > > > > >> or
>>> > > > > > > > > > > >> > > > > >> > > > > another.
>>> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
>>> > > > > > > > > java.util.regex.Pattern,
>>> > > > > > > > > > > >> they
>>> > > > > > > > > > > >> > expect
>>> > > > > > > > > > > >> > > > > >> > their
>>> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If
>>> we are
>>> > > > doing
>>> > > > > > > > > something
>>> > > > > > > > > > > >> > different,
>>> > > > > > > > > > > >> > > > > >> then
>>> > > > > > > > > > > >> > > > > >> > we
>>> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
>>> overload
>>> > > > with
>>> > > > > > our own
>>> > > > > > > > > > > >> Pattern
>>> > > > > > > > > > > >> > class
>>> > > > > > > > > > > >> > > > > (I
>>> > > > > > > > > > > >> > > > > >> > > don't
>>> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
>>> re2j's at
>>> > > > this
>>> > > > > > > > > point).
>>> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
>>> major new
>>> > > > > > protocol
>>> > > > > > > > > should
>>> > > > > > > > > > > >> > > > > integrate
>>> > > > > > > > > > > >> > > > > >> > fully
>>> > > > > > > > > > > >> > > > > >> > > > > with
>>> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the
>>> topic
>>> > > > > > recreation case
>>> > > > > > > > > > > >> > correctly.
>>> > > > > > > > > > > >> > > > > That's
>>> > > > > > > > > > > >> > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> > > main
>>> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
>>> agree with
>>> > > > > > David
>>> > > > > > > > > that we'd
>>> > > > > > > > > > > >> > want to
>>> > > > > > > > > > > >> > > > > add
>>> > > > > > > > > > > >> > > > > >> > > topic
>>> > > > > > > > > > > >> > > > > >> > > > > ids
>>> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
>>> that don't
>>> > > > > > have it
>>> > > > > > > > > yet and
>>> > > > > > > > > > > >> > that we
>>> > > > > > > > > > > >> > > > > can
>>> > > > > > > > > > > >> > > > > >> > > probably
>>> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals
>>> versus adding
>>> > > > new
>>> > > > > > APIs
>>> > > > > > > > > to the
>>> > > > > > > > > > > >> > Java
>>> > > > > > > > > > > >> > > > > >> Consumer
>>> > > > > > > > > > > >> > > > > >> > > > > (unless
>>> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs
>>> is
>>> > > > required
>>> > > > > > for
>>> > > > > > > > > > > >> reasonable
>>> > > > > > > > > > > >> > > > > >> semantics).
>>> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about
>>> the
>>> > > > > > coordinator
>>> > > > > > > > > storing
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > >> configs.
>>> > > > > > > > > > > >> > > > > >> > > It's
>>> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
>>> > > > centralized
>>> > > > > > in the
>>> > > > > > > > > > > >> metadata
>>> > > > > > > > > > > >> > log for
>>> > > > > > > > > > > >> > > > > >> > > various
>>> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
>>> visibility,
>>> > > > > > consistency,
>>> > > > > > > > > > > >> etc.).
>>> > > > > > > > > > > >> > > > > >> Similarly, I
>>> > > > > > > > > > > >> > > > > >> > > am
>>> > > > > > > > > > > >> > > > > >> > > > > not
>>> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
>>> deleting
>>> > > > > > configs in a
>>> > > > > > > > > way
>>> > > > > > > > > > > >> that
>>> > > > > > > > > > > >> > they
>>> > > > > > > > > > > >> > > > > >> cannot
>>> > > > > > > > > > > >> > > > > >> > > be
>>> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property
>>> for modern
>>> > > > > > systems
>>> > > > > > > > > is to
>>> > > > > > > > > > > >> > minimize
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > number
>>> > > > > > > > > > > >> > > > > >> > > > > of
>>> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
>>> scenarios.
>>> > > > > > > > > > > >> > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > Ismael
>>> > > > > > > > > > > >> > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47
>>> PM David
>>> > > > > > Jacot
>>> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
>>> > > > > > > > > > > >> > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My
>>> answers are
>>> > > > below:
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
>>> especially
>>> > > > > > the last
>>> > > > > > > > > step
>>> > > > > > > > > > > >> when
>>> > > > > > > > > > > >> > > > > clients
>>> > > > > > > > > > > >> > > > > >> > > flip the
>>> > > > > > > > > > > >> > > > > >> > > > > > > flag
>>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
>>> protocol, in
>>> > > > which
>>> > > > > > we
>>> > > > > > > > > would
>>> > > > > > > > > > > >> have a
>>> > > > > > > > > > > >> > > > > window
>>> > > > > > > > > > > >> > > > > >> > where
>>> > > > > > > > > > > >> > > > > >> > > > > both
>>> > > > > > > > > > > >> > > > > >> > > > > > > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
>>> > > > protocols /
>>> > > > > > rpcs
>>> > > > > > > > > are
>>> > > > > > > > > > > >> used
>>> > > > > > > > > > > >> > by
>>> > > > > > > > > > > >> > > > > >> members
>>> > > > > > > > > > > >> > > > > >> > of
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > same
>>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
>>> coordinator could
>>> > > > > > "mimic"
>>> > > > > > > > > the old
>>> > > > > > > > > > > >> > behavior
>>> > > > > > > > > > > >> > > > > >> while
>>> > > > > > > > > > > >> > > > > >> > > > > using the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
>>> something we
>>> > > > need to
>>> > > > > > > > > present
>>> > > > > > > > > > > >> about.
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a
>>> new
>>> > > > version
>>> > > > > > of KIP
>>> > > > > > > > > which
>>> > > > > > > > > > > >> > includes
>>> > > > > > > > > > > >> > > > > >> more
>>> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
>>> > > > "Supporting
>>> > > > > > > > > Online
>>> > > > > > > > > > > >> > Consumer
>>> > > > > > > > > > > >> > > > > Group
>>> > > > > > > > > > > >> > > > > >> > > Upgrade"
>>> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
>>> > > > Deprecation,
>>> > > > > > and
>>> > > > > > > > > Migration
>>> > > > > > > > > > > >> > Plan". I
>>> > > > > > > > > > > >> > > > > >> think
>>> > > > > > > > > > > >> > > > > >> > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
>>> few cases
>>> > > > > > now but
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > overall idea
>>> > > > > > > > > > > >> > > > > >> and
>>> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
>>> understandable.
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
>>> ids. So
>>> > > > far as
>>> > > > > > > > > KIP-516 the
>>> > > > > > > > > > > >> > topic ids
>>> > > > > > > > > > > >> > > > > >> are
>>> > > > > > > > > > > >> > > > > >> > > only
>>> > > > > > > > > > > >> > > > > >> > > > > used
>>> > > > > > > > > > > >> > > > > >> > > > > > > as
>>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
>>> client,
>>> > > > but
>>> > > > > > they
>>> > > > > > > > > are not
>>> > > > > > > > > > > >> > exposed
>>> > > > > > > > > > > >> > > > > via
>>> > > > > > > > > > > >> > > > > >> any
>>> > > > > > > > > > > >> > > > > >> > > > > public
>>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
>>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
>>> think the
>>> > > > > > question is,
>>> > > > > > > > > first
>>> > > > > > > > > > > >> > should we
>>> > > > > > > > > > > >> > > > > >> let
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
>>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
>>> the
>>> > > > names
>>> > > > > > -> ids
>>> > > > > > > > > mapping
>>> > > > > > > > > > > >> > itself
>>> > > > > > > > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> > fully
>>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
>>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
>>> existing
>>> > > > RPCs
>>> > > > > > and the
>>> > > > > > > > > new
>>> > > > > > > > > > > >> RPCs
>>> > > > > > > > > > > >> > with
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > topic
>>> > > > > > > > > > > >> > > > > >> > > > > ids;
>>> > > > > > > > > > > >> > > > > >> > > > > > > and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
>>> consider
>>> > > > > > exposing
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > topic ids
>>> > > > > > > > > > > >> > > > > in
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > consumer
>>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
>>> > > > > > > > > subscribe/assign, as
>>> > > > > > > > > > > >> well
>>> > > > > > > > > > > >> > as in
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > rebalance
>>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
>>> topic
>>> > > > > > > > > > > >> > deletion-and-recreation).
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
>>> include
>>> > > > > > converting
>>> > > > > > > > > all
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > offsets
>>> > > > > > > > > > > >> > > > > >> > > related
>>> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in
>>> this
>>> > > > KIP,
>>> > > > > > the
>>> > > > > > > > > consumer
>>> > > > > > > > > > > >> > would be
>>> > > > > > > > > > > >> > > > > >> able
>>> > > > > > > > > > > >> > > > > >> > to
>>> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
>>> ids. That
>>> > > > > > being
>>> > > > > > > > > said, it
>>> > > > > > > > > > > >> > still has
>>> > > > > > > > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> > > provide
>>> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in
>>> various APIs so
>>> > > > > > having a
>>> > > > > > > > > > > >> mapping
>>> > > > > > > > > > > >> > in the
>>> > > > > > > > > > > >> > > > > >> > > consumer
>>> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
>>> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
>>> opinion on
>>> > > > > > this.
>>> > > > > > > > > Here I
>>> > > > > > > > > > > >> > wonder if
>>> > > > > > > > > > > >> > > > > >> this
>>> > > > > > > > > > > >> > > > > >> > > goes
>>> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
>>> KIP. I
>>> > > > would
>>> > > > > > rather
>>> > > > > > > > > focus
>>> > > > > > > > > > > >> on
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> > internals
>>> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider
>>> this
>>> > > > > > separately if
>>> > > > > > > > > we see
>>> > > > > > > > > > > >> > value in
>>> > > > > > > > > > > >> > > > > >> doing
>>> > > > > > > > > > > >> > > > > >> > > it.
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
>>> point about
>>> > > > > > using
>>> > > > > > > > > topic ids
>>> > > > > > > > > > > >> > in the
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> ConsumerGroupHeartbeatRequest, I
>>> > > > think
>>> > > > > > that
>>> > > > > > > > > there
>>> > > > > > > > > > > >> is
>>> > > > > > > > > > > >> > one
>>> > > > > > > > > > > >> > > > > >> > advantage
>>> > > > > > > > > > > >> > > > > >> > > in
>>> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer
>>> will
>>> > > > have
>>> > > > > > the
>>> > > > > > > > > > > >> opportunity
>>> > > > > > > > > > > >> > to
>>> > > > > > > > > > > >> > > > > >> validate
>>> > > > > > > > > > > >> > > > > >> > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
>>> passing
>>> > > > them
>>> > > > > > into
>>> > > > > > > > > the
>>> > > > > > > > > > > >> group
>>> > > > > > > > > > > >> > > > > rebalance
>>> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
>>> > > > coordinator
>>> > > > > > will
>>> > > > > > > > > also
>>> > > > > > > > > > > >> notice
>>> > > > > > > > > > > >> > it but
>>> > > > > > > > > > > >> > > > > >> it
>>> > > > > > > > > > > >> > > > > >> > > does
>>> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
>>> reject an
>>> > > > > > invalid
>>> > > > > > > > > topic in
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> response.
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
>>> on all
>>> > > > other
>>> > > > > > minor
>>> > > > > > > > > > > >> questions
>>> > > > > > > > > > > >> > > > > except
>>> > > > > > > > > > > >> > > > > >> for
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
>>> question:
>>> > > > > > personally
>>> > > > > > > > > I think
>>> > > > > > > > > > > >> > it's not
>>> > > > > > > > > > > >> > > > > >> > > necessary
>>> > > > > > > > > > > >> > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
>>> API with
>>> > > > > > Pattern,
>>> > > > > > > > > but
>>> > > > > > > > > > > >> > instead we
>>> > > > > > > > > > > >> > > > > >> still
>>> > > > > > > > > > > >> > > > > >> > > use
>>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
>>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
>>> that our
>>> > > > > > > > > subscription may
>>> > > > > > > > > > > >> be
>>> > > > > > > > > > > >> > > > > >> rejected by
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > server.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
>>> case is a
>>> > > > > > very rare
>>> > > > > > > > > > > >> scenario
>>> > > > > > > > > > > >> > I felt
>>> > > > > > > > > > > >> > > > > >> > using
>>> > > > > > > > > > > >> > > > > >> > > an
>>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
>>> > > > > > subscription may
>>> > > > > > > > > be
>>> > > > > > > > > > > >> more
>>> > > > > > > > > > > >> > > > > >> vulnerable
>>> > > > > > > > > > > >> > > > > >> > to
>>> > > > > > > > > > > >> > > > > >> > > > > various
>>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
>>> look at
>>> > > > the
>>> > > > > > > > > differences
>>> > > > > > > > > > > >> > between
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > two
>>> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
>>> understand the
>>> > > > > > potential
>>> > > > > > > > > issues.
>>> > > > > > > > > > > >> My
>>> > > > > > > > > > > >> > > > > >> > > understanding is
>>> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all
>>> the basic
>>> > > > > > regular
>>> > > > > > > > > > > >> > expressions. The
>>> > > > > > > > > > > >> > > > > >> > > differences
>>> > > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
>>> about
>>> > > > the
>>> > > > > > various
>>> > > > > > > > > > > >> character
>>> > > > > > > > > > > >> > > > > >> classes. I
>>> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
>>> think
>>> > > > about
>>> > > > > > this.
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > Best,
>>> > > > > > > > > > > >> > > > > >> > > > > > > David
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
>>> 11:28 PM
>>> > > > > > Guozhang
>>> > > > > > > > > Wang <
>>> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
>>> > > > > > > > > > > >> > > > > >> > > >
>>> > > > > > > > > > > >> > > > > >> > > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on
>>> the high
>>> > > > > > level
>>> > > > > > > > > there are
>>> > > > > > > > > > > >> > two meta
>>> > > > > > > > > > > >> > > > > >> > > points we
>>> > > > > > > > > > > >> > > > > >> > > > > need
>>> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
>>> especially
>>> > > > > > the last
>>> > > > > > > > > step
>>> > > > > > > > > > > >> when
>>> > > > > > > > > > > >> > > > > clients
>>> > > > > > > > > > > >> > > > > >> > > flip the
>>> > > > > > > > > > > >> > > > > >> > > > > > > flag
>>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
>>> protocol, in
>>> > > > which
>>> > > > > > we
>>> > > > > > > > > would
>>> > > > > > > > > > > >> have a
>>> > > > > > > > > > > >> > > > > window
>>> > > > > > > > > > > >> > > > > >> > where
>>> > > > > > > > > > > >> > > > > >> > > > > both
>>> > > > > > > > > > > >> > > > > >> > > > > > > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
>>> > > > protocols /
>>> > > > > > rpcs
>>> > > > > > > > > are
>>> > > > > > > > > > > >> used
>>> > > > > > > > > > > >> > by
>>> > > > > > > > > > > >> > > > > >> members
>>> > > > > > > > > > > >> > > > > >> > of
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > same
>>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
>>> coordinator could
>>> > > > > > "mimic"
>>> > > > > > > > > the old
>>> > > > > > > > > > > >> > behavior
>>> > > > > > > > > > > >> > > > > >> while
>>> > > > > > > > > > > >> > > > > >> > > > > using the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is
>>> something we
>>> > > > need to
>>> > > > > > > > > present
>>> > > > > > > > > > > >> about.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
>>> ids. So
>>> > > > far as
>>> > > > > > > > > KIP-516 the
>>> > > > > > > > > > > >> > topic ids
>>> > > > > > > > > > > >> > > > > >> are
>>> > > > > > > > > > > >> > > > > >> > > only
>>> > > > > > > > > > > >> > > > > >> > > > > used
>>> > > > > > > > > > > >> > > > > >> > > > > > > as
>>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
>>> client,
>>> > > > but
>>> > > > > > they
>>> > > > > > > > > are not
>>> > > > > > > > > > > >> > exposed
>>> > > > > > > > > > > >> > > > > via
>>> > > > > > > > > > > >> > > > > >> any
>>> > > > > > > > > > > >> > > > > >> > > > > public
>>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
>>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I
>>> think the
>>> > > > > > question is,
>>> > > > > > > > > first
>>> > > > > > > > > > > >> > should we
>>> > > > > > > > > > > >> > > > > >> let
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
>>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
>>> the
>>> > > > names
>>> > > > > > -> ids
>>> > > > > > > > > mapping
>>> > > > > > > > > > > >> > itself
>>> > > > > > > > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> > fully
>>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
>>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
>>> existing
>>> > > > RPCs
>>> > > > > > and the
>>> > > > > > > > > new
>>> > > > > > > > > > > >> RPCs
>>> > > > > > > > > > > >> > with
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > topic
>>> > > > > > > > > > > >> > > > > >> > > > > ids;
>>> > > > > > > > > > > >> > > > > >> > > > > > > and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
>>> consider
>>> > > > > > exposing
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > topic ids
>>> > > > > > > > > > > >> > > > > in
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > consumer
>>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
>>> > > > > > > > > subscribe/assign, as
>>> > > > > > > > > > > >> well
>>> > > > > > > > > > > >> > as in
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > rebalance
>>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
>>> topic
>>> > > > > > > > > > > >> > deletion-and-recreation).
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
>>> on all
>>> > > > other
>>> > > > > > minor
>>> > > > > > > > > > > >> questions
>>> > > > > > > > > > > >> > > > > except
>>> > > > > > > > > > > >> > > > > >> for
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
>>> question:
>>> > > > > > personally
>>> > > > > > > > > I think
>>> > > > > > > > > > > >> > it's not
>>> > > > > > > > > > > >> > > > > >> > > necessary
>>> > > > > > > > > > > >> > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
>>> API with
>>> > > > > > Pattern,
>>> > > > > > > > > but
>>> > > > > > > > > > > >> > instead we
>>> > > > > > > > > > > >> > > > > >> still
>>> > > > > > > > > > > >> > > > > >> > > use
>>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
>>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
>>> that our
>>> > > > > > > > > subscription may
>>> > > > > > > > > > > >> be
>>> > > > > > > > > > > >> > > > > >> rejected by
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > server.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
>>> case is a
>>> > > > > > very rare
>>> > > > > > > > > > > >> scenario
>>> > > > > > > > > > > >> > I felt
>>> > > > > > > > > > > >> > > > > >> > using
>>> > > > > > > > > > > >> > > > > >> > > an
>>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
>>> > > > > > subscription may
>>> > > > > > > > > be
>>> > > > > > > > > > > >> more
>>> > > > > > > > > > > >> > > > > >> vulnerable
>>> > > > > > > > > > > >> > > > > >> > to
>>> > > > > > > > > > > >> > > > > >> > > > > various
>>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
>>> 5:23 AM
>>> > > > > > David Jacot
>>> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
>>> feedback. Let
>>> > > > me
>>> > > > > > answer
>>> > > > > > > > > your
>>> > > > > > > > > > > >> > questions
>>> > > > > > > > > > > >> > > > > >> > inline.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
>>> premature to
>>> > > > > > talk about
>>> > > > > > > > > > > >> target
>>> > > > > > > > > > > >> > > > > versions
>>> > > > > > > > > > > >> > > > > >> for
>>> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
>>> existing group
>>> > > > > > protocol.
>>> > > > > > > > > > > >> Unlike
>>> > > > > > > > > > > >> > KRaft,
>>> > > > > > > > > > > >> > > > > >> this
>>> > > > > > > > > > > >> > > > > >> > > > > affects a
>>> > > > > > > > > > > >> > > > > >> > > > > > > core
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
>>> hence
>>> > > > > > > > > deprecation/removal
>>> > > > > > > > > > > >> > will be
>>> > > > > > > > > > > >> > > > > >> > heavily
>>> > > > > > > > > > > >> > > > > >> > > > > > > dependent
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
>>> applications
>>> > > > migrate
>>> > > > > > to
>>> > > > > > > > > the new
>>> > > > > > > > > > > >> > protocol.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I
>>> will
>>> > > > remove it.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
>>> intend to
>>> > > > > > release
>>> > > > > > > > > this in
>>> > > > > > > > > > > >> > 4.x, but
>>> > > > > > > > > > > >> > > > > it
>>> > > > > > > > > > > >> > > > > >> > > wasn't
>>> > > > > > > > > > > >> > > > > >> > > > > made
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added
>>> that as a
>>> > > > way to
>>> > > > > > > > > estimate
>>> > > > > > > > > > > >> when
>>> > > > > > > > > > > >> > we'd
>>> > > > > > > > > > > >> > > > > >> > > deprecate
>>> > > > > > > > > > > >> > > > > >> > > > > and
>>> > > > > > > > > > > >> > > > > >> > > > > > > remove
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol,
>>> I also
>>> > > > > > suggest
>>> > > > > > > > > removing
>>> > > > > > > > > > > >> > this part.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
>>> reasoning. As
>>> > > > > > > > > explained, I
>>> > > > > > > > > > > >> plan
>>> > > > > > > > > > > >> > to
>>> > > > > > > > > > > >> > > > > >> rewrite
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
>>> while we
>>> > > > > > implement
>>> > > > > > > > > the new
>>> > > > > > > > > > > >> > protocol.
>>> > > > > > > > > > > >> > > > > >> This
>>> > > > > > > > > > > >> > > > > >> > > means
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals
>>> will be
>>> > > > slightly
>>> > > > > > > > > different
>>> > > > > > > > > > > >> > (e.g.
>>> > > > > > > > > > > >> > > > > >> threading
>>> > > > > > > > > > > >> > > > > >> > > > > model).
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
>>> tighten
>>> > > > the
>>> > > > > > > > > switch from
>>> > > > > > > > > > > >> > the old
>>> > > > > > > > > > > >> > > > > >> group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
>>> group
>>> > > > > > coordinator
>>> > > > > > > > > to a
>>> > > > > > > > > > > >> > major
>>> > > > > > > > > > > >> > > > > >> release.
>>> > > > > > > > > > > >> > > > > >> > > The
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be
>>> to use a
>>> > > > flag
>>> > > > > > to do
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > switch
>>> > > > > > > > > > > >> > > > > >> instead
>>> > > > > > > > > > > >> > > > > >> > of
>>> > > > > > > > > > > >> > > > > >> > > > > relying
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software
>>> upgrade.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
>>> out the
>>> > > > > > details of
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > migration
>>> > > > > > > > > > > >> > > > > >> story.
>>> > > > > > > > > > > >> > > > > >> > > It
>>> > > > > > > > > > > >> > > > > >> > > > > sounds
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > like
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
>>> support
>>> > > > > > online
>>> > > > > > > > > > > >> migrations.
>>> > > > > > > > > > > >> > Is
>>> > > > > > > > > > > >> > > > > that
>>> > > > > > > > > > > >> > > > > >> > > correct?
>>> > > > > > > > > > > >> > > > > >> > > > > We
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
>>> detail. It
>>> > > > could
>>> > > > > > also
>>> > > > > > > > > be done
>>> > > > > > > > > > > >> > as a
>>> > > > > > > > > > > >> > > > > >> separate
>>> > > > > > > > > > > >> > > > > >> > > KIP,
>>> > > > > > > > > > > >> > > > > >> > > > > if
>>> > > > > > > > > > > >> > > > > >> > > > > > > it's
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
>>> online
>>> > > > > > migrations
>>> > > > > > > > > for the
>>> > > > > > > > > > > >> > group.
>>> > > > > > > > > > > >> > > > > That
>>> > > > > > > > > > > >> > > > > >> > > means
>>> > > > > > > > > > > >> > > > > >> > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
>>> protocol
>>> > > > > > will be
>>> > > > > > > > > able to
>>> > > > > > > > > > > >> > switch to
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain
>>> how that
>>> > > > > > will work
>>> > > > > > > > > > > >> though.
>>> > > > > > > > > > > >> > It is
>>> > > > > > > > > > > >> > > > > >> > > basically a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
>>> > > > upgraded or
>>> > > > > > > > > rolled to a
>>> > > > > > > > > > > >> > software
>>> > > > > > > > > > > >> > > > > >> > > supporting
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
>>> Both the
>>> > > > > > old and
>>> > > > > > > > > the new
>>> > > > > > > > > > > >> > > > > >> coordinator
>>> > > > > > > > > > > >> > > > > >> > > will
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
>>> protocol and
>>> > > > rely
>>> > > > > > on the
>>> > > > > > > > > same
>>> > > > > > > > > > > >> > persisted
>>> > > > > > > > > > > >> > > > > >> > > metadata so
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work
>>> together. This
>>> > > > point
>>> > > > > > is an
>>> > > > > > > > > > > >> offline
>>> > > > > > > > > > > >> > > > > >> migration.
>>> > > > > > > > > > > >> > > > > >> > We
>>> > > > > > > > > > > >> > > > > >> > > > > cannot
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
>>> because it
>>> > > > would
>>> > > > > > require
>>> > > > > > > > > > > >> > shutting down
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > current
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
>>> starting up the
>>> > > > > > new one
>>> > > > > > > > > and
>>> > > > > > > > > > > >> that
>>> > > > > > > > > > > >> > would
>>> > > > > > > > > > > >> > > > > >> cause
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
>>> metadata
>>> > > > > > version/IBP
>>> > > > > > > > > must be
>>> > > > > > > > > > > >> > upgraded
>>> > > > > > > > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> X
>>> > > > > > > > > > > >> > > > > >> > in
>>> > > > > > > > > > > >> > > > > >> > > > > order
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
>>> protocol.
>>> > > > This
>>> > > > > > cannot
>>> > > > > > > > > be done
>>> > > > > > > > > > > >> > before
>>> > > > > > > > > > > >> > > > > 1)
>>> > > > > > > > > > > >> > > > > >> is
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the
>>> old
>>> > > > > > coordinator
>>> > > > > > > > > doesn't
>>> > > > > > > > > > > >> > support
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must
>>> be
>>> > > > upgraded
>>> > > > > > to a
>>> > > > > > > > > version
>>> > > > > > > > > > > >> > > > > supporting
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > online
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
>>> KIP-792).
>>> > > > If
>>> > > > > > the
>>> > > > > > > > > consumer
>>> > > > > > > > > > > >> is
>>> > > > > > > > > > > >> > > > > already
>>> > > > > > > > > > > >> > > > > >> > > there.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done
>>> at this
>>> > > > point.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must
>>> be rolled
>>> > > > > > with the
>>> > > > > > > > > > > >> feature
>>> > > > > > > > > > > >> > flag
>>> > > > > > > > > > > >> > > > > >> turned
>>> > > > > > > > > > > >> > > > > >> > > on.
>>> > > > > > > > > > > >> > > > > >> > > > > The
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
>>> automatically
>>> > > > > > > > > converted when
>>> > > > > > > > > > > >> > the first
>>> > > > > > > > > > > >> > > > > >> > > consumer
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
>>> joins
>>> > > > the
>>> > > > > > group.
>>> > > > > > > > > While
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > members
>>> > > > > > > > > > > >> > > > > >> > > using the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
>>> > > > upgraded,
>>> > > > > > the old
>>> > > > > > > > > > > >> protocol
>>> > > > > > > > > > > >> > is
>>> > > > > > > > > > > >> > > > > >> proxied
>>> > > > > > > > > > > >> > > > > >> > > into
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
>>> this in
>>> > > > the
>>> > > > > > KIP.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that
>>> we are
>>> > > > > > pushing the
>>> > > > > > > > > pattern
>>> > > > > > > > > > > >> > > > > >> subscriptions
>>> > > > > > > > > > > >> > > > > >> > > to
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
>>> there
>>> > > > could be
>>> > > > > > some
>>> > > > > > > > > tricky
>>> > > > > > > > > > > >> > > > > >> compatibility
>>> > > > > > > > > > > >> > > > > >> > > > > issues.
>>> > > > > > > > > > > >> > > > > >> > > > > > > Will
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
>>> users to
>>> > > > > > detect
>>> > > > > > > > > that
>>> > > > > > > > > > > >> they
>>> > > > > > > > > > > >> > need to
>>> > > > > > > > > > > >> > > > > >> > update
>>> > > > > > > > > > > >> > > > > >> > > > > their
>>> > > > > > > > > > > >> > > > > >> > > > > > > regex
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
>>> the new
>>> > > > > > protocol?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a
>>> bit more
>>> > > > > > optimistic
>>> > > > > > > > > than
>>> > > > > > > > > > > >> you
>>> > > > > > > > > > > >> > on this
>>> > > > > > > > > > > >> > > > > >> > > point. I
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
>>> majority of
>>> > > > the
>>> > > > > > cases
>>> > > > > > > > > are
>>> > > > > > > > > > > >> simple
>>> > > > > > > > > > > >> > > > > regexes
>>> > > > > > > > > > > >> > > > > >> > which
>>> > > > > > > > > > > >> > > > > >> > > > > should
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
>>> engine. The
>>> > > > > > > > > coordinator will
>>> > > > > > > > > > > >> > verify
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > regex
>>> > > > > > > > > > > >> > > > > >> > > > > anyway
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the
>>> consumer if the
>>> > > > > > regex is
>>> > > > > > > > > not
>>> > > > > > > > > > > >> valid.
>>> > > > > > > > > > > >> > > > > Coming
>>> > > > > > > > > > > >> > > > > >> > back
>>> > > > > > > > > > > >> > > > > >> > > to
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
>>> worst
>>> > > > case,
>>> > > > > > the
>>> > > > > > > > > first
>>> > > > > > > > > > > >> > upgraded
>>> > > > > > > > > > > >> > > > > >> > consumer
>>> > > > > > > > > > > >> > > > > >> > > > > joining
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
>>> rejected.
>>> > > > This
>>> > > > > > should
>>> > > > > > > > > be used
>>> > > > > > > > > > > >> > as the
>>> > > > > > > > > > > >> > > > > >> last
>>> > > > > > > > > > > >> > > > > >> > > > > defence, I
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers
>>> to
>>> > > > validate
>>> > > > > > their
>>> > > > > > > > > regex
>>> > > > > > > > > > > >> > before
>>> > > > > > > > > > > >> > > > > >> > upgrading
>>> > > > > > > > > > > >> > > > > >> > > > > their
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
>>> them with
>>> > > > > > another
>>> > > > > > > > > group.
>>> > > > > > > > > > > >> For
>>> > > > > > > > > > > >> > > > > >> instance,
>>> > > > > > > > > > > >> > > > > >> > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
>>> pre-prod
>>> > > > > > environment.
>>> > > > > > > > > > > >> Another
>>> > > > > > > > > > > >> > way
>>> > > > > > > > > > > >> > > > > >> would be
>>> > > > > > > > > > > >> > > > > >> > > to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
>>> consumer-group tool
>>> > > > to
>>> > > > > > > > > provide a
>>> > > > > > > > > > > >> regex
>>> > > > > > > > > > > >> > > > > >> validation
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this
>>> be
>>> > > > enough in
>>> > > > > > your
>>> > > > > > > > > > > >> opinion?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the
>>> last
>>> > > > question,
>>> > > > > > will
>>> > > > > > > > > the
>>> > > > > > > > > > > >> Java
>>> > > > > > > > > > > >> > client
>>> > > > > > > > > > > >> > > > > >> allow
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > users to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
>>> current regex
>>> > > > > > engine for
>>> > > > > > > > > > > >> > compatibility
>>> > > > > > > > > > > >> > > > > >> > > reasons?
>>> > > > > > > > > > > >> > > > > >> > > > > For
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
>>> keep using
>>> > > > > > client
>>> > > > > > > > > based
>>> > > > > > > > > > > >> > regex at
>>> > > > > > > > > > > >> > > > > >> first
>>> > > > > > > > > > > >> > > > > >> > to
>>> > > > > > > > > > > >> > > > > >> > > keep
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple
>>> and then
>>> > > > > > migrate to
>>> > > > > > > > > server
>>> > > > > > > > > > > >> > based
>>> > > > > > > > > > > >> > > > > >> regexes
>>> > > > > > > > > > > >> > > > > >> > > as a
>>> > > > > > > > > > > >> > > > > >> > > > > > > second
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your
>>> point but I
>>> > > > am
>>> > > > > > > > > concerned that
>>> > > > > > > > > > > >> > this
>>> > > > > > > > > > > >> > > > > would
>>> > > > > > > > > > > >> > > > > >> > > allow
>>> > > > > > > > > > > >> > > > > >> > > > > users
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
>>> this mode.
>>> > > > That
>>> > > > > > > > > would go
>>> > > > > > > > > > > >> > against our
>>> > > > > > > > > > > >> > > > > >> goal
>>> > > > > > > > > > > >> > > > > >> > > of
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
>>> because
>>> > > > we
>>> > > > > > would
>>> > > > > > > > > have to
>>> > > > > > > > > > > >> > continue
>>> > > > > > > > > > > >> > > > > >> > > monitoring
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
>>> client
>>> > > > side. I
>>> > > > > > would
>>> > > > > > > > > rather
>>> > > > > > > > > > > >> > not do
>>> > > > > > > > > > > >> > > > > >> this.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
>>> the group
>>> > > > > > > > > coordinator
>>> > > > > > > > > > > >> will
>>> > > > > > > > > > > >> > be
>>> > > > > > > > > > > >> > > > > >> > > responsible for
>>> > > > > > > > > > > >> > > > > >> > > > > > > storing
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations
>>> and that
>>> > > > the
>>> > > > > > > > > > > >> configurations
>>> > > > > > > > > > > >> > will be
>>> > > > > > > > > > > >> > > > > >> > > deleted
>>> > > > > > > > > > > >> > > > > >> > > > > when
>>> > > > > > > > > > > >> > > > > >> > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
>>> Will a
>>> > > > > > transition to
>>> > > > > > > > > DEAD
>>> > > > > > > > > > > >> > trigger
>>> > > > > > > > > > > >> > > > > >> > deletion
>>> > > > > > > > > > > >> > > > > >> > > of
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
>>> > > > configurations
>>> > > > > > will be
>>> > > > > > > > > > > >> deleted
>>> > > > > > > > > > > >> > when
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > group is
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
>>> together.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice
>>> to store
>>> > > > the
>>> > > > > > > > > configs in
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > group
>>> > > > > > > > > > > >> > > > > >> > > coordinator
>>> > > > > > > > > > > >> > > > > >> > > > > > > make it
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
>>> cluster
>>> > > > > > configs and
>>> > > > > > > > > their
>>> > > > > > > > > > > >> > values?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
>>> group
>>> > > > > > > > > configurations are
>>> > > > > > > > > > > >> > overrides
>>> > > > > > > > > > > >> > > > > >> of
>>> > > > > > > > > > > >> > > > > >> > > cluster
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want
>>> to know
>>> > > > all
>>> > > > > > the
>>> > > > > > > > > overrides
>>> > > > > > > > > > > >> > though,
>>> > > > > > > > > > > >> > > > > you
>>> > > > > > > > > > > >> > > > > >> > > would
>>> > > > > > > > > > > >> > > > > >> > > > > have
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
>>> > > > coordinators.
>>> > > > > > You
>>> > > > > > > > > cannot
>>> > > > > > > > > > > >> > rely on
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > metadata
>>> > > > > > > > > > > >> > > > > >> > > > > log
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
>>> > > > configure a
>>> > > > > > group
>>> > > > > > > > > before
>>> > > > > > > > > > > >> > starting
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > consumers?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
>>> allowing the
>>> > > > > > explicit
>>> > > > > > > > > creation
>>> > > > > > > > > > > >> of
>>> > > > > > > > > > > >> > > > > groups?
>>> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
>>> decoupled
>>> > > > from
>>> > > > > > the
>>> > > > > > > > > group
>>> > > > > > > > > > > >> > lifecycle.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
>>> > > > > > automatically
>>> > > > > > > > > created in
>>> > > > > > > > > > > >> > this
>>> > > > > > > > > > > >> > > > > case.
>>> > > > > > > > > > > >> > > > > >> > > However,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
>>> lost after
>>> > > > the
>>> > > > > > > > > retention
>>> > > > > > > > > > > >> > period of
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
>>> > > > Consumer.subscribe
>>> > > > > > method
>>> > > > > > > > > for the
>>> > > > > > > > > > > >> > Java
>>> > > > > > > > > > > >> > > > > >> client
>>> > > > > > > > > > > >> > > > > >> > > still
>>> > > > > > > > > > > >> > > > > >> > > > > take
>>> > > > > > > > > > > >> > > > > >> > > > > > > a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>>> `java.util.regex.Pattern` of
>>> > > > do
>>> > > > > > we
>>> > > > > > > > > have to
>>> > > > > > > > > > > >> > introduce
>>> > > > > > > > > > > >> > > > > an
>>> > > > > > > > > > > >> > > > > >> > > overload?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
>>> question. I
>>> > > > > > forgot
>>> > > > > > > > > about
>>> > > > > > > > > > > >> that
>>> > > > > > > > > > > >> > one.
>>> > > > > > > > > > > >> > > > > As
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> `java.util.regex.Pattern` is
>>> > > > not
>>> > > > > > fully
>>> > > > > > > > > > > >> compatible
>>> > > > > > > > > > > >> > with
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > engine
>>> > > > > > > > > > > >> > > > > >> > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
>>> might be
>>> > > > better
>>> > > > > > to
>>> > > > > > > > > deprecate
>>> > > > > > > > > > > >> > it and
>>> > > > > > > > > > > >> > > > > >> use an
>>> > > > > > > > > > > >> > > > > >> > > > > overload
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string.
>>> We would
>>> > > > > > rely on
>>> > > > > > > > > the
>>> > > > > > > > > > > >> server
>>> > > > > > > > > > > >> > side
>>> > > > > > > > > > > >> > > > > >> > > validation.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration,
>>> I think
>>> > > > that
>>> > > > > > we
>>> > > > > > > > > could
>>> > > > > > > > > > > >> still
>>> > > > > > > > > > > >> > try to
>>> > > > > > > > > > > >> > > > > >> > > toString
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
>>> should
>>> > > > > > work, I
>>> > > > > > > > > think, in
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> majority
>>> > > > > > > > > > > >> > > > > >> > > of the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
>>> Justine
>>> > > > that we
>>> > > > > > > > > should be
>>> > > > > > > > > > > >> > clearer
>>> > > > > > > > > > > >> > > > > about
>>> > > > > > > > > > > >> > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> > > > > reason
>>> > > > > > > > > > > >> > > > > >> > > > > > > to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
>>> > > > IBP/metadata.version
>>> > > > > > from the
>>> > > > > > > > > > > >> feature
>>> > > > > > > > > > > >> > flag.
>>> > > > > > > > > > > >> > > > > >> Maybe
>>> > > > > > > > > > > >> > > > > >> > > we
>>> > > > > > > > > > > >> > > > > >> > > > > mean
>>> > > > > > > > > > > >> > > > > >> > > > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
>>> default for
>>> > > > the
>>> > > > > > feature
>>> > > > > > > > > flag
>>> > > > > > > > > > > >> to
>>> > > > > > > > > > > >> > true
>>> > > > > > > > > > > >> > > > > >> based
>>> > > > > > > > > > > >> > > > > >> > on
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version
>>> once we
>>> > > > want to
>>> > > > > > make
>>> > > > > > > > > it the
>>> > > > > > > > > > > >> > default.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use
>>> that feature
>>> > > > > > flag
>>> > > > > > > > > mainly
>>> > > > > > > > > > > >> > during the
>>> > > > > > > > > > > >> > > > > >> > > development
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not
>>> have
>>> > > > mentioned
>>> > > > > > it, I
>>> > > > > > > > > think,
>>> > > > > > > > > > > >> > because
>>> > > > > > > > > > > >> > > > > we
>>> > > > > > > > > > > >> > > > > >> > > could
>>> > > > > > > > > > > >> > > > > >> > > > > use
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for
>>> it.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
>>> protocol APIs
>>> > > > > > don't
>>> > > > > > > > > mention
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > required
>>> > > > > > > > > > > >> > > > > >> > > ACLs, it
>>> > > > > > > > > > > >> > > > > >> > > > > > > would
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > be
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
>>> > > > consistency.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
>>> that
>>> > > > > > > > > > > >> ConsumerGroupHeartbeat
>>> > > > > > > > > > > >> > > > > requires
>>> > > > > > > > > > > >> > > > > >> > > "Read
>>> > > > > > > > > > > >> > > > > >> > > > > Group"
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > even
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to
>>> do more
>>> > > > than
>>> > > > > > > > > reading.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how
>>> the
>>> > > > current
>>> > > > > > > > > protocol works
>>> > > > > > > > > > > >> > though.
>>> > > > > > > > > > > >> > > > > We
>>> > > > > > > > > > > >> > > > > >> > only
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group"
>>> to join a
>>> > > > > > group. We
>>> > > > > > > > > could
>>> > > > > > > > > > > >> > consider
>>> > > > > > > > > > > >> > > > > >> > > changing
>>> > > > > > > > > > > >> > > > > >> > > > > this
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that
>>> it is
>>> > > > worth
>>> > > > > > it.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
>>> recreation
>>> > > > > > handled by
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > consumer
>>> > > > > > > > > > > >> > > > > with
>>> > > > > > > > > > > >> > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> > > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would
>>> be good to
>>> > > > > > have a
>>> > > > > > > > > section
>>> > > > > > > > > > > >> on
>>> > > > > > > > > > > >> > this.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
>>> > > > > > perspective, the
>>> > > > > > > > > new
>>> > > > > > > > > > > >> topic
>>> > > > > > > > > > > >> > will
>>> > > > > > > > > > > >> > > > > >> have a
>>> > > > > > > > > > > >> > > > > >> > > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
>>> treat it
>>> > > > like a
>>> > > > > > > > > topic with
>>> > > > > > > > > > > >> a
>>> > > > > > > > > > > >> > > > > different
>>> > > > > > > > > > > >> > > > > >> > > name.
>>> > > > > > > > > > > >> > > > > >> > > > > The
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
>>> > > > fetch/commit
>>> > > > > > > > > offsets
>>> > > > > > > > > > > >> APIs
>>> > > > > > > > > > > >> > do not
>>> > > > > > > > > > > >> > > > > >> > support
>>> > > > > > > > > > > >> > > > > >> > > > > topic
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
>>> would
>>> > > > reuse the
>>> > > > > > > > > offsets
>>> > > > > > > > > > > >> based
>>> > > > > > > > > > > >> > on the
>>> > > > > > > > > > > >> > > > > >> > same.
>>> > > > > > > > > > > >> > > > > >> > > I
>>> > > > > > > > > > > >> > > > > >> > > > > think
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
>>> those
>>> > > > APIs
>>> > > > > > as well
>>> > > > > > > > > in
>>> > > > > > > > > > > >> order
>>> > > > > > > > > > > >> > to be
>>> > > > > > > > > > > >> > > > > >> > > consistent
>>> > > > > > > > > > > >> > > > > >> > > > > end
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
>>> strengthen
>>> > > > the
>>> > > > > > > > > semantics of
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > >> consumer.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions
>>> we will
>>> > > > > > write the
>>> > > > > > > > > new
>>> > > > > > > > > > > >> > coordinator
>>> > > > > > > > > > > >> > > > > >> in
>>> > > > > > > > > > > >> > > > > >> > > Java.
>>> > > > > > > > > > > >> > > > > >> > > > > Even
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
>>> implementation
>>> > > > > > detail, do
>>> > > > > > > > > we plan
>>> > > > > > > > > > > >> to
>>> > > > > > > > > > > >> > have a
>>> > > > > > > > > > > >> > > > > >> new
>>> > > > > > > > > > > >> > > > > >> > > gradle
>>> > > > > > > > > > > >> > > > > >> > > > > > > module
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
>>> scalability
>>> > > > > > goal when
>>> > > > > > > > > it
>>> > > > > > > > > > > >> comes
>>> > > > > > > > > > > >> > to how
>>> > > > > > > > > > > >> > > > > >> many
>>> > > > > > > > > > > >> > > > > >> > > > > members
>>> > > > > > > > > > > >> > > > > >> > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
>>> support?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers
>>> at the
>>> > > > > > moment. The
>>> > > > > > > > > > > >> protocol
>>> > > > > > > > > > > >> > should
>>> > > > > > > > > > > >> > > > > >> > > support
>>> > > > > > > > > > > >> > > > > >> > > > > 1000s
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group.
>>> We will
>>> > > > > > measure
>>> > > > > > > > > this when
>>> > > > > > > > > > > >> > we have
>>> > > > > > > > > > > >> > > > > a
>>> > > > > > > > > > > >> > > > > >> > first
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
>>> that we
>>> > > > might
>>> > > > > > have
>>> > > > > > > > > other
>>> > > > > > > > > > > >> > > > > bottlenecks
>>> > > > > > > > > > > >> > > > > >> > down
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
>>> commits).
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
>>> having
>>> > > > > > > > > SubscribedTopidIds
>>> > > > > > > > > > > >> > instead
>>> > > > > > > > > > > >> > > > > of
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames
>>> in
>>> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
>>> > > > > > > > > > > >> > > > > >> Is
>>> > > > > > > > > > > >> > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> > > > > idea
>>> > > > > > > > > > > >> > > > > >> > > > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
>>> resolve the
>>> > > > > > regex on
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > server, we
>>> > > > > > > > > > > >> > > > > >> can do
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > same
>>> > > > > > > > > > > >> > > > > >> > > > > > > for
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
>>> > > > difference
>>> > > > > > is that
>>> > > > > > > > > > > >> sending
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> regex is
>>> > > > > > > > > > > >> > > > > >> > > more
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
>>> topic
>>> > > > names
>>> > > > > > is less
>>> > > > > > > > > > > >> > efficient.
>>> > > > > > > > > > > >> > > > > >> > > Furthermore,
>>> > > > > > > > > > > >> > > > > >> > > > > > > delete
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier
>>> to
>>> > > > handle
>>> > > > > > if we
>>> > > > > > > > > have
>>> > > > > > > > > > > >> topic
>>> > > > > > > > > > > >> > ids.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
>>> consolidate the
>>> > > > > > metadata
>>> > > > > > > > > lookup
>>> > > > > > > > > > > >> > on the
>>> > > > > > > > > > > >> > > > > >> server
>>> > > > > > > > > > > >> > > > > >> > > for
>>> > > > > > > > > > > >> > > > > >> > > > > both
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
>>> with your
>>> > > > > > point. As
>>> > > > > > > > > a
>>> > > > > > > > > > > >> second
>>> > > > > > > > > > > >> > > > > though,
>>> > > > > > > > > > > >> > > > > >> > using
>>> > > > > > > > > > > >> > > > > >> > > > > topic
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here
>>> for the
>>> > > > > > delete and
>>> > > > > > > > > > > >> > recreation
>>> > > > > > > > > > > >> > > > > case.
>>> > > > > > > > > > > >> > > > > >> > > Also, I
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
>>> allow
>>> > > > users to
>>> > > > > > > > > subscribe
>>> > > > > > > > > > > >> with
>>> > > > > > > > > > > >> > topic
>>> > > > > > > > > > > >> > > > > >> ids
>>> > > > > > > > > > > >> > > > > >> > in
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that is
>>> the only
>>> > > > > > way to be
>>> > > > > > > > > > > >> really
>>> > > > > > > > > > > >> > robust
>>> > > > > > > > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> > > topic
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > David
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022
>>> at 1:38 PM
>>> > > > > > David
>>> > > > > > > > > Jacot <
>>> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
>>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
>>> comments.
>>> > > > Please
>>> > > > > > find
>>> > > > > > > > > my
>>> > > > > > > > > > > >> answers
>>> > > > > > > > > > > >> > > > > below.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
>>> protocol
>>> > > > relies on
>>> > > > > > > > > topic IDs
>>> > > > > > > > > > > >> > with the
>>> > > > > > > > > > > >> > > > > >> > > exception
>>> > > > > > > > > > > >> > > > > >> > > > > of the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in
>>> the
>>> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
>>> > > > > > > > > > > >> > > > > >> I
>>> > > > > > > > > > > >> > > > > >> > am
>>> > > > > > > > > > > >> > > > > >> > > not
>>> > > > > > > > > > > >> > > > > >> > > > > sure
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names
>>> is the
>>> > > > > > right call
>>> > > > > > > > > here.
>>> > > > > > > > > > > >> I
>>> > > > > > > > > > > >> > need to
>>> > > > > > > > > > > >> > > > > >> > think
>>> > > > > > > > > > > >> > > > > >> > > > > about it
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
>>> Obviously, the
>>> > > > > > KIP does
>>> > > > > > > > > not
>>> > > > > > > > > > > >> > change the
>>> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
>>> topic
>>> > > > IDs.
>>> > > > > > This
>>> > > > > > > > > may be
>>> > > > > > > > > > > >> > something
>>> > > > > > > > > > > >> > > > > >> that
>>> > > > > > > > > > > >> > > > > >> > we
>>> > > > > > > > > > > >> > > > > >> > > > > should
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
>>> would
>>> > > > give
>>> > > > > > better
>>> > > > > > > > > > > >> overall
>>> > > > > > > > > > > >> > > > > >> guarantee in
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
>>> think that
>>> > > > I
>>> > > > > > should
>>> > > > > > > > > not
>>> > > > > > > > > > > >> have
>>> > > > > > > > > > > >> > > > > mentioned
>>> > > > > > > > > > > >> > > > > >> > this
>>> > > > > > > > > > > >> > > > > >> > > > > flag at
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove
>>> it. We can
>>> > > > > > use an
>>> > > > > > > > > internal
>>> > > > > > > > > > > >> > > > > >> configuration
>>> > > > > > > > > > > >> > > > > >> > > while
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
>>> feature.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
>>> will be
>>> > > > > > > > > supported. The
>>> > > > > > > > > > > >> > change is
>>> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
>>> that the
>>> > > > > > cluster
>>> > > > > > > > > uses
>>> > > > > > > > > > > >> topic
>>> > > > > > > > > > > >> > IDs.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022
>>> at 9:53
>>> > > > PM
>>> > > > > > > > > Guozhang
>>> > > > > > > > > > > >> Wang <
>>> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
>>> feedback.
>>> > > > Here
>>> > > > > > are
>>> > > > > > > > > some
>>> > > > > > > > > > > >> replies
>>> > > > > > > > > > > >> > > > > inlined
>>> > > > > > > > > > > >> > > > > >> > > below:
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9,
>>> 2022 at
>>> > > > 2:53 AM
>>> > > > > > > > > Ismael Juma
>>> > > > > > > > > > > >> <
>>> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
>>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
>>> KIP. This
>>> > > > has
>>> > > > > > the
>>> > > > > > > > > > > >> potential
>>> > > > > > > > > > > >> > to be a
>>> > > > > > > > > > > >> > > > > >> > great
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
>>> > > > questions/comments:
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
>>> > > > premature to
>>> > > > > > talk
>>> > > > > > > > > about
>>> > > > > > > > > > > >> > target
>>> > > > > > > > > > > >> > > > > >> versions
>>> > > > > > > > > > > >> > > > > >> > > for
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
>>> existing
>>> > > > group
>>> > > > > > > > > protocol.
>>> > > > > > > > > > > >> > Unlike
>>> > > > > > > > > > > >> > > > > >> KRaft,
>>> > > > > > > > > > > >> > > > > >> > > this
>>> > > > > > > > > > > >> > > > > >> > > > > > > affects a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > core
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
>>> and hence
>>> > > > > > > > > > > >> > deprecation/removal will
>>> > > > > > > > > > > >> > > > > >> be
>>> > > > > > > > > > > >> > > > > >> > > heavily
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
>>> applications
>>> > > > > > migrate
>>> > > > > > > > > to the
>>> > > > > > > > > > > >> new
>>> > > > > > > > > > > >> > > > > >> protocol.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
>>> you. I
>>> > > > think
>>> > > > > > we can
>>> > > > > > > > > > > >> remove
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> proposed
>>> > > > > > > > > > > >> > > > > >> > > > > timeline
>>> > > > > > > > > > > >> > > > > >> > > > > > > in
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
>>> > > > Deprecation,
>>> > > > > > and
>>> > > > > > > > > Migration
>>> > > > > > > > > > > >> > Plan` and
>>> > > > > > > > > > > >> > > > > >> > > instead
>>> > > > > > > > > > > >> > > > > >> > > > > just
>>> > > > > > > > > > > >> > > > > >> > > > > > > state
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will
>>> decide in the
>>> > > > > > future
>>> > > > > > > > > about
>>> > > > > > > > > > > >> when
>>> > > > > > > > > > > >> > we
>>> > > > > > > > > > > >> > > > > would
>>> > > > > > > > > > > >> > > > > >> > > > > deprecate old
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
>>> behaviors.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says
>>> we
>>> > > > intend to
>>> > > > > > > > > release this
>>> > > > > > > > > > > >> > in 4.x,
>>> > > > > > > > > > > >> > > > > >> but
>>> > > > > > > > > > > >> > > > > >> > it
>>> > > > > > > > > > > >> > > > > >> > > > > wasn't
>>> > > > > > > > > > > >> > > > > >> > > > > > > made
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
>>> that as
>>> > > > a
>>> > > > > > way to
>>> > > > > > > > > > > >> estimate
>>> > > > > > > > > > > >> > when
>>> > > > > > > > > > > >> > > > > we'd
>>> > > > > > > > > > > >> > > > > >> > > > > deprecate
>>> > > > > > > > > > > >> > > > > >> > > > > > > and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
>>> protocol, I
>>> > > > also
>>> > > > > > suggest
>>> > > > > > > > > > > >> removing
>>> > > > > > > > > > > >> > this
>>> > > > > > > > > > > >> > > > > >> part.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
>>> > > > specifically
>>> > > > > > > > > related to
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's
>>> more for
>>> > > > client
>>> > > > > > > > > upgrades.
>>> > > > > > > > > > > >> I.e.
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> > > broker-side
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may
>>> be done
>>> > > > > > first,
>>> > > > > > > > > and then
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > client
>>> > > > > > > > > > > >> > > > > >> > side,
>>> > > > > > > > > > > >> > > > > >> > > > > and we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > would
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
>>> "released"
>>> > > > by
>>> > > > > > the
>>> > > > > > > > > time
>>> > > > > > > > > > > >> clients
>>> > > > > > > > > > > >> > > > > >> > > implementations
>>> > > > > > > > > > > >> > > > > >> > > > > are
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
>>> enable the
>>> > > > > > feature the
>>> > > > > > > > > > > >> clients
>>> > > > > > > > > > > >> > need to
>>> > > > > > > > > > > >> > > > > >> > first
>>> > > > > > > > > > > >> > > > > >> > > > > swap-in
>>> > > > > > > > > > > >> > > > > >> > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
>>> rolling
>>> > > > bounce
>>> > > > > > and
>>> > > > > > > > > then set
>>> > > > > > > > > > > >> > the flag
>>> > > > > > > > > > > >> > > > > >> > with a
>>> > > > > > > > > > > >> > > > > >> > > > > second
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence
>>> we feel
>>> > > > it's
>>> > > > > > > > > better to be
>>> > > > > > > > > > > >> > released
>>> > > > > > > > > > > >> > > > > >> in a
>>> > > > > > > > > > > >> > > > > >> > > major
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
>>> flesh out
>>> > > > the
>>> > > > > > > > > details of
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > migration
>>> > > > > > > > > > > >> > > > > >> > > story.
>>> > > > > > > > > > > >> > > > > >> > > > > It
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
>>> will
>>> > > > support
>>> > > > > > online
>>> > > > > > > > > > > >> > migrations. Is
>>> > > > > > > > > > > >> > > > > >> that
>>> > > > > > > > > > > >> > > > > >> > > > > correct?
>>> > > > > > > > > > > >> > > > > >> > > > > > > We
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
>>> detail.
>>> > > > It
>>> > > > > > could
>>> > > > > > > > > also be
>>> > > > > > > > > > > >> > done as a
>>> > > > > > > > > > > >> > > > > >> > > separate
>>> > > > > > > > > > > >> > > > > >> > > > > KIP,
>>> > > > > > > > > > > >> > > > > >> > > > > > > if
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's
>>> the
>>> > > > part we
>>> > > > > > can
>>> > > > > > > > > be more
>>> > > > > > > > > > > >> > concrete
>>> > > > > > > > > > > >> > > > > >> about
>>> > > > > > > > > > > >> > > > > >> > > for
>>> > > > > > > > > > > >> > > > > >> > > > > sure
>>> > > > > > > > > > > >> > > > > >> > > > > > > (and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to
>>> your
>>> > > > > > question 2)
>>> > > > > > > > > above).
>>> > > > > > > > > > > >> > We will
>>> > > > > > > > > > > >> > > > > >> work
>>> > > > > > > > > > > >> > > > > >> > on
>>> > > > > > > > > > > >> > > > > >> > > > > making
>>> > > > > > > > > > > >> > > > > >> > > > > > > it
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > more
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
>>> parallel as we
>>> > > > > > solicit
>>> > > > > > > > > more
>>> > > > > > > > > > > >> > feedback.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
>>> that we are
>>> > > > > > pushing
>>> > > > > > > > > the
>>> > > > > > > > > > > >> > pattern
>>> > > > > > > > > > > >> > > > > >> > > subscriptions
>>> > > > > > > > > > > >> > > > > >> > > > > to
>>> > > > > > > > > > > >> > > > > >> > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems
>>> like there
>>> > > > > > could be
>>> > > > > > > > > some
>>> > > > > > > > > > > >> tricky
>>> > > > > > > > > > > >> > > > > >> > > compatibility
>>> > > > > > > > > > > >> > > > > >> > > > > > > issues.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism
>>> for
>>> > > > users to
>>> > > > > > > > > detect that
>>> > > > > > > > > > > >> > they
>>> > > > > > > > > > > >> > > > > need
>>> > > > > > > > > > > >> > > > > >> to
>>> > > > > > > > > > > >> > > > > >> > > update
>>> > > > > > > > > > > >> > > > > >> > > > > > > their
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching
>>> to the
>>> > > > new
>>> > > > > > > > > protocol?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we
>>> need some
>>> > > > > > tooling for
>>> > > > > > > > > > > >> non-java
>>> > > > > > > > > > > >> > client
>>> > > > > > > > > > > >> > > > > >> > users
>>> > > > > > > > > > > >> > > > > >> > > to
>>> > > > > > > > > > > >> > > > > >> > > > > sort
>>> > > > > > > > > > > >> > > > > >> > > > > > > of
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
>>> client before
>>> > > > > > > > > switching to
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > new
>>> > > > > > > > > > > >> > > > > >> > protocol.
>>> > > > > > > > > > > >> > > > > >> > > I
>>> > > > > > > > > > > >> > > > > >> > > > > do not
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on
>>> top of my
>>> > > > head
>>> > > > > > > > > though,
>>> > > > > > > > > > > >> maybe
>>> > > > > > > > > > > >> > others
>>> > > > > > > > > > > >> > > > > >> like
>>> > > > > > > > > > > >> > > > > >> > > @Matt
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to
>>> the last
>>> > > > > > question,
>>> > > > > > > > > will the
>>> > > > > > > > > > > >> > Java
>>> > > > > > > > > > > >> > > > > >> client
>>> > > > > > > > > > > >> > > > > >> > > allow
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
>>> current
>>> > > > regex
>>> > > > > > > > > engine for
>>> > > > > > > > > > > >> > > > > >> compatibility
>>> > > > > > > > > > > >> > > > > >> > > > > reasons?
>>> > > > > > > > > > > >> > > > > >> > > > > > > For
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy
>>> to keep
>>> > > > using
>>> > > > > > > > > client
>>> > > > > > > > > > > >> based
>>> > > > > > > > > > > >> > regex
>>> > > > > > > > > > > >> > > > > at
>>> > > > > > > > > > > >> > > > > >> > > first to
>>> > > > > > > > > > > >> > > > > >> > > > > keep
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations
>>> simple and
>>> > > > then
>>> > > > > > migrate
>>> > > > > > > > > to
>>> > > > > > > > > > > >> > server based
>>> > > > > > > > > > > >> > > > > >> > > regexes
>>> > > > > > > > > > > >> > > > > >> > > > > as a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > second
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have
>>> not thought
>>> > > > > > about
>>> > > > > > > > > that for
>>> > > > > > > > > > > >> > java
>>> > > > > > > > > > > >> > > > > >> clients,
>>> > > > > > > > > > > >> > > > > >> > > and
>>> > > > > > > > > > > >> > > > > >> > > > > we can
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
>>> > > > > > compatibility
>>> > > > > > > > > issues do
>>> > > > > > > > > > > >> > you have
>>> > > > > > > > > > > >> > > > > in
>>> > > > > > > > > > > >> > > > > >> > > mind?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
>>> that the
>>> > > > group
>>> > > > > > > > > coordinator
>>> > > > > > > > > > > >> > will be
>>> > > > > > > > > > > >> > > > > >> > > > > responsible for
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
>>> configurations and
>>> > > > that
>>> > > > > > the
>>> > > > > > > > > > > >> > configurations
>>> > > > > > > > > > > >> > > > > will
>>> > > > > > > > > > > >> > > > > >> be
>>> > > > > > > > > > > >> > > > > >> > > > > deleted
>>> > > > > > > > > > > >> > > > > >> > > > > > > when
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is
>>> deleted. Will a
>>> > > > > > > > > transition to
>>> > > > > > > > > > > >> DEAD
>>> > > > > > > > > > > >> > > > > trigger
>>> > > > > > > > > > > >> > > > > >> > > deletion
>>> > > > > > > > > > > >> > > > > >> > > > > of
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the
>>> DEAD state
>>> > > > is an
>>> > > > > > > > > ending
>>> > > > > > > > > > > >> state
>>> > > > > > > > > > > >> > (we
>>> > > > > > > > > > > >> > > > > would
>>> > > > > > > > > > > >> > > > > >> > only
>>> > > > > > > > > > > >> > > > > >> > > > > > > transit to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
>>> group is
>>> > > > EMPTY
>>> > > > > > and
>>> > > > > > > > > also all
>>> > > > > > > > > > > >> of
>>> > > > > > > > > > > >> > its
>>> > > > > > > > > > > >> > > > > >> > metadata
>>> > > > > > > > > > > >> > > > > >> > > are
>>> > > > > > > > > > > >> > > > > >> > > > > > > gone),
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > once
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
>>> DEAD this
>>> > > > > > group
>>> > > > > > > > > would
>>> > > > > > > > > > > >> never
>>> > > > > > > > > > > >> > be
>>> > > > > > > > > > > >> > > > > >> revived.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
>>> choice to
>>> > > > store
>>> > > > > > the
>>> > > > > > > > > configs
>>> > > > > > > > > > > >> in
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > group
>>> > > > > > > > > > > >> > > > > >> > > > > coordinator
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list
>>> all
>>> > > > cluster
>>> > > > > > configs
>>> > > > > > > > > and
>>> > > > > > > > > > > >> their
>>> > > > > > > > > > > >> > > > > values?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
>>> question,
>>> > > > and our
>>> > > > > > > > > thoughts
>>> > > > > > > > > > > >> are
>>> > > > > > > > > > > >> > that
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > so-called
>>> > > > > > > > > > > >> > > > > >> > > > > > > "group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
>>> > > > overrides
>>> > > > > > of the
>>> > > > > > > > > > > >> > cluster-level
>>> > > > > > > > > > > >> > > > > >> > > > > configurations
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
>>> group so
>>> > > > when an
>>> > > > > > > > > admin list
>>> > > > > > > > > > > >> > cluster
>>> > > > > > > > > > > >> > > > > >> > configs
>>> > > > > > > > > > > >> > > > > >> > > it's
>>> > > > > > > > > > > >> > > > > >> > > > > > > okay to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
>>> cluster-level
>>> > > > > > > > > "defaults", not
>>> > > > > > > > > > > >> > showing
>>> > > > > > > > > > > >> > > > > >> any
>>> > > > > > > > > > > >> > > > > >> > > > > per-group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
>>> someone
>>> > > > > > configure a
>>> > > > > > > > > group
>>> > > > > > > > > > > >> > before
>>> > > > > > > > > > > >> > > > > >> starting
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
>>> allowing
>>> > > > the
>>> > > > > > explicit
>>> > > > > > > > > > > >> > creation of
>>> > > > > > > > > > > >> > > > > >> groups?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
>>> > > > decoupled
>>> > > > > > from
>>> > > > > > > > > the group
>>> > > > > > > > > > > >> > > > > lifecycle.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
>>> created
>>> > > > > > before
>>> > > > > > > > > the group
>>> > > > > > > > > > > >> > itself
>>> > > > > > > > > > > >> > > > > as
>>> > > > > > > > > > > >> > > > > >> an
>>> > > > > > > > > > > >> > > > > >> > > > > > > independent
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
>>> > > > requires
>>> > > > > > the
>>> > > > > > > > > > > >> corresponding
>>> > > > > > > > > > > >> > > > > >> request to
>>> > > > > > > > > > > >> > > > > >> > > be
>>> > > > > > > > > > > >> > > > > >> > > > > > > routed to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
>>> based on
>>> > > > the
>>> > > > > > group
>>> > > > > > > > > id ---
>>> > > > > > > > > > > >> > the only
>>> > > > > > > > > > > >> > > > > >> > thing
>>> > > > > > > > > > > >> > > > > >> > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
>>> itself is
>>> > > > gone
>>> > > > > > we
>>> > > > > > > > > also check
>>> > > > > > > > > > > >> > if there
>>> > > > > > > > > > > >> > > > > >> are
>>> > > > > > > > > > > >> > > > > >> > > any
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related
>>> to that
>>> > > > group
>>> > > > > > and
>>> > > > > > > > > delete
>>> > > > > > > > > > > >> as
>>> > > > > > > > > > > >> > well.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
>>> indeed
>>> > > > > > introduces an
>>> > > > > > > > > > > >> > asymmetry on
>>> > > > > > > > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > creation
>>> > > > > > > > > > > >> > > > > >> > > > > /
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
>>> config
>>> > > > > > entities,
>>> > > > > > > > > and we
>>> > > > > > > > > > > >> > would like
>>> > > > > > > > > > > >> > > > > >> to
>>> > > > > > > > > > > >> > > > > >> > > hear
>>> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether
>>> we should
>>> > > > aim
>>> > > > > > for
>>> > > > > > > > > symmetry
>>> > > > > > > > > > > >> > i.e.
>>> > > > > > > > > > > >> > > > > >> totally
>>> > > > > > > > > > > >> > > > > >> > > > > decouple
>>> > > > > > > > > > > >> > > > > >> > > > > > > group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence
>>> not
>>> > > > delete
>>> > > > > > them at
>>> > > > > > > > > all
>>> > > > > > > > > > > >> when
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> group
>>> > > > > > > > > > > >> > > > > >> > is
>>> > > > > > > > > > > >> > > > > >> > > > > gone,
>>> > > > > > > > > > > >> > > > > >> > > > > > > but
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > always
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
>>> deletion
>>> > > > > > operations
>>> > > > > > > > > by
>>> > > > > > > > > > > >> > themselves.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
>>> > > > > > Consumer.subscribe
>>> > > > > > > > > method for
>>> > > > > > > > > > > >> > the Java
>>> > > > > > > > > > > >> > > > > >> > client
>>> > > > > > > > > > > >> > > > > >> > > > > still
>>> > > > > > > > > > > >> > > > > >> > > > > > > take
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > `java.util.regex.Pattern` of
>>> > > > > > do we
>>> > > > > > > > > have
>>> > > > > > > > > > > >> to
>>> > > > > > > > > > > >> > > > > >> introduce an
>>> > > > > > > > > > > >> > > > > >> > > > > overload?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
>>> need to
>>> > > > > > introduce
>>> > > > > > > > > an
>>> > > > > > > > > > > >> > overload, but
>>> > > > > > > > > > > >> > > > > >> I'm
>>> > > > > > > > > > > >> > > > > >> > > all
>>> > > > > > > > > > > >> > > > > >> > > > > ears
>>> > > > > > > > > > > >> > > > > >> > > > > > > if
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > there
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
>>> compatibility
>>> > > > > > issues
>>> > > > > > > > > that we
>>> > > > > > > > > > > >> may
>>> > > > > > > > > > > >> > > > > overlook.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
>>> Justine
>>> > > > > > that we
>>> > > > > > > > > should
>>> > > > > > > > > > > >> be
>>> > > > > > > > > > > >> > clearer
>>> > > > > > > > > > > >> > > > > >> > about
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > reason
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > to
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
>>> > > > > > IBP/metadata.version
>>> > > > > > > > > from the
>>> > > > > > > > > > > >> > feature
>>> > > > > > > > > > > >> > > > > >> flag.
>>> > > > > > > > > > > >> > > > > >> > > Maybe
>>> > > > > > > > > > > >> > > > > >> > > > > we
>>> > > > > > > > > > > >> > > > > >> > > > > > > mean
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
>>> default
>>> > > > for
>>> > > > > > the
>>> > > > > > > > > feature
>>> > > > > > > > > > > >> flag
>>> > > > > > > > > > > >> > to
>>> > > > > > > > > > > >> > > > > true
>>> > > > > > > > > > > >> > > > > >> > > based on
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
>>> once we
>>> > > > > > want to
>>> > > > > > > > > make it
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > >> default.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
>>> protocol
>>> > > > APIs
>>> > > > > > don't
>>> > > > > > > > > mention
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > >> required
>>> > > > > > > > > > > >> > > > > >> > > ACLs,
>>> > > > > > > > > > > >> > > > > >> > > > > it
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that
>>> for
>>> > > > > > consistency.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
>>> certainly do
>>> > > > that.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
>>> odd that
>>> > > > > > > > > > > >> > ConsumerGroupHeartbeat
>>> > > > > > > > > > > >> > > > > >> > requires
>>> > > > > > > > > > > >> > > > > >> > > > > "Read
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems
>>> to do
>>> > > > more
>>> > > > > > than
>>> > > > > > > > > reading.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
>>> myself
>>> > > > as
>>> > > > > > well,
>>> > > > > > > > > but in
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > end we
>>> > > > > > > > > > > >> > > > > >> > could
>>> > > > > > > > > > > >> > > > > >> > > not
>>> > > > > > > > > > > >> > > > > >> > > > > > > find a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better
>>> alternative: adding
>>> > > > > > Write
>>> > > > > > > > > Group
>>> > > > > > > > > > > >> seems
>>> > > > > > > > > > > >> > an
>>> > > > > > > > > > > >> > > > > >> overkill
>>> > > > > > > > > > > >> > > > > >> > > here
>>> > > > > > > > > > > >> > > > > >> > > > > > > since we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > do
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
>>> elsewhere (we
>>> > > > only
>>> > > > > > have
>>> > > > > > > > > Read /
>>> > > > > > > > > > > >> > Delete
>>> > > > > > > > > > > >> > > > > and
>>> > > > > > > > > > > >> > > > > >> > > Describe
>>> > > > > > > > > > > >> > > > > >> > > > > on
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like
>>> to hear
>>> > > > others
>>> > > > > > > > > thoughts.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
>>> > > > recreation
>>> > > > > > > > > handled by
>>> > > > > > > > > > > >> the
>>> > > > > > > > > > > >> > > > > consumer
>>> > > > > > > > > > > >> > > > > >> > with
>>> > > > > > > > > > > >> > > > > >> > > the
>>> > > > > > > > > > > >> > > > > >> > > > > new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > group
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
>>> would be
>>> > > > good
>>> > > > > > to have
>>> > > > > > > > > a
>>> > > > > > > > > > > >> > section on
>>> > > > > > > > > > > >> > > > > >> this.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
>>> > > > > > subscription
>>> > > > > > > > > right? Yes
>>> > > > > > > > > > > >> > we can
>>> > > > > > > > > > > >> > > > > >> add a
>>> > > > > > > > > > > >> > > > > >> > > > > section
>>> > > > > > > > > > > >> > > > > >> > > > > > > about
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but
>>> basically the
>>> > > > idea
>>> > > > > > is that
>>> > > > > > > > > > > >> consumer
>>> > > > > > > > > > > >> > would
>>> > > > > > > > > > > >> > > > > be
>>> > > > > > > > > > > >> > > > > >> > > totally
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol
>>> as it's
>>> > > > > > handled all
>>> > > > > > > > > by the
>>> > > > > > > > > > > >> > brokers.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
>>> mentions we
>>> > > > will
>>> > > > > > write
>>> > > > > > > > > the
>>> > > > > > > > > > > >> new
>>> > > > > > > > > > > >> > > > > >> coordinator
>>> > > > > > > > > > > >> > > > > >> > in
>>> > > > > > > > > > > >> > > > > >> > > > > Java.
>>> > > > > > > > > > > >> > > > > >> > > > > > > Even
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
>>> implementation
>>> > > > > > detail,
>>> > > > > > > > > do we
>>> > > > > > > > > > > >> > plan to
>>> > > > > > > > > > > >> > > > > >> have a
>>> > > > > > > > > > > >> > > > > >> > > new
>>> > > > > > > > > > > >> > > > > >> > > > > gradle
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > module
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not
>>> thought about
>>> > > > > > that. But
>>> > > > > > > > > I think
>>> > > > > > > > > > > >> > the
>>> > > > > > > > > > > >> > > > > answer
>>> > > > > > > > > > > >> > > > > >> > > should
>>> > > > > > > > > > > >> > > > > >> > > > > be
>>> > > > > > > > > > > >> > > > > >> > > > > > > yes.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
>>> > > > scalability
>>> > > > > > goal
>>> > > > > > > > > when it
>>> > > > > > > > > > > >> > comes to
>>> > > > > > > > > > > >> > > > > >> how
>>> > > > > > > > > > > >> > > > > >> > > many
>>> > > > > > > > > > > >> > > > > >> > > > > > > members
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol
>>> can
>>> > > > support?
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
>>> think we
>>> > > > > > should
>>> > > > > > > > > shoot for
>>> > > > > > > > > > > >> > 1000s of
>>> > > > > > > > > > > >> > > > > >> > > members.
>>> > > > > > > > > > > >> > > > > >> > > > > But
>>> > > > > > > > > > > >> > > > > >> > > > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
>>> also
>>> > > > depend
>>> > > > > > on the
>>> > > > > > > > > offset
>>> > > > > > > > > > > >> > > > > management
>>> > > > > > > > > > > >> > > > > >> > > (commit,
>>> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
>>> > > > coordinator
>>> > > > > > > > > which we
>>> > > > > > > > > > > >> did
>>> > > > > > > > > > > >> > not
>>> > > > > > > > > > > >> > > > > >> cover in
>>> > > > > > > > > > > >> > > > > >> > > this
>>> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
>>> number that
>>> > > > > > applies
>>> > > > > > > > > > > >> > universally.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
>>> consider
>>> > > > having
>>> > > > > > > > > > > >> > SubscribedTopidIds
>>> > > > > > > > > > > >> > > > > >> instead
>>> > > > > > > > > > > >> > > > > >> > of
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> SubscribedTopicNames in
>>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
>>> > > > > > > > > > > >> > > > > >> > > Is the
>>> > > > > > > > > > > >> > > > > >> > > > > > > idea
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
>>> resolve
>>> > > > the
>>> > > > > > regex
>>> > > > > > > > > on the
>>> > > > > > > > > > > >> > server,
>>> > > > > > > > > > > >> > > > > we
>>> > > > > > > > > > > >> > > > > >> > can
>>> > > > > > > > > > > >> > > > > >> > > do
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > same
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > for
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name?
>>> The
>>> > > > > > difference is
>>> > > > > > > > > that
>>> > > > > > > > > > > >> > sending the
>>> > > > > > > > > > > >> > > > > >> > regex
>>> > > > > > > > > > > >> > > > > >> > > is
>>> > > > > > > > > > > >> > > > > >> > > > > more
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
>>> the topic
>>> > > > > > names is
>>> > > > > > > > > less
>>> > > > > > > > > > > >> > efficient.
>>> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
>>> easier to
>>> > > > > > handle if
>>> > > > > > > > > we have
>>> > > > > > > > > > > >> > topic
>>> > > > > > > > > > > >> > > > > ids.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
>>> still
>>> > > > let
>>> > > > > > the
>>> > > > > > > > > clients
>>> > > > > > > > > > > >> send
>>> > > > > > > > > > > >> > names
>>> > > > > > > > > > > >> > > > > >> is to
>>> > > > > > > > > > > >> > > > > >> > > keep
>>> > > > > > > > > > > >> > > > > >> > > > > the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names
>>> -> ids
>>> > > > on
>>> > > > > > the
>>> > > > > > > > > broker /
>>> > > > > > > > > > > >> > admin
>>> > > > > > > > > > > >> > > > > client
>>> > > > > > > > > > > >> > > > > >> > > only.
>>> > > > > > > > > > > >> > > > > >> > > > > Note
>>> > > > > > > > > > > >> > > > > >> > > > > > > that
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
>>> topic id
>>> > > > in
>>> > > > > > > > > KIP-516, we
>>> > > > > > > > > > > >> > never
>>> > > > > > > > > > > >> > > > > >> > > implemented the
>>> > > > > > > > > > > >> > > > > >> > > > > > > logic
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
>>> > > > leveraging
>>> > > > > > the
>>> > > > > > > > > related
>>> > > > > > > > > > > >> > newer
>>> > > > > > > > > > > >> > > > > >> versioned
>>> > > > > > > > > > > >> > > > > >> > > RPCs,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic
>>> id as
>>> > > > empty
>>> > > > > > UUID.
>>> > > > > > > > > We
>>> > > > > > > > > > > >> want
>>> > > > > > > > > > > >> > to keep
>>> > > > > > > > > > > >> > > > > >> the
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
>>> > > > delegate
>>> > > > > > the
>>> > > > > > > > > reasoning
>>> > > > > > > > > > > >> on
>>> > > > > > > > > > > >> > broker
>>> > > > > > > > > > > >> > > > > >> and
>>> > > > > > > > > > > >> > > > > >> > > > > > > potentially
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
>>> 2022 at
>>> > > > 10:45
>>> > > > > > AM
>>> > > > > > > > > David
>>> > > > > > > > > > > >> Jacot
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> <dj...@confluent.io.invalid>
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like
>>> to start a
>>> > > > > > > > > discussion
>>> > > > > > > > > > > >> thread
>>> > > > > > > > > > > >> > on
>>> > > > > > > > > > > >> > > > > >> KIP-848:
>>> > > > > > > > > > > >> > > > > >> > > The
>>> > > > > > > > > > > >> > > > > >> > > > > Next
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of
>>> the
>>> > > > Consumer
>>> > > > > > > > > Rebalance
>>> > > > > > > > > > > >> > Protocol.
>>> > > > > > > > > > > >> > > > > >> With
>>> > > > > > > > > > > >> > > > > >> > > this
>>> > > > > > > > > > > >> > > > > >> > > > > KIP,
>>> > > > > > > > > > > >> > > > > >> > > > > > > we
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
>>> rebalance
>>> > > > > > protocol
>>> > > > > > > > > (for
>>> > > > > > > > > > > >> > consumers)
>>> > > > > > > > > > > >> > > > > >> more
>>> > > > > > > > > > > >> > > > > >> > > > > reliable,
>>> > > > > > > > > > > >> > > > > >> > > > > > > more
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
>>> easier to
>>> > > > > > implement for
>>> > > > > > > > > > > >> > clients, and
>>> > > > > > > > > > > >> > > > > >> easier
>>> > > > > > > > > > > >> > > > > >> > > to
>>> > > > > > > > > > > >> > > > > >> > > > > debug
>>> > > > > > > > > > > >> > > > > >> > > > > > > for
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is
>>> here:
>>> > > > > > > > > > > >> > > > > >> > > > >
>>> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
>>> look and
>>> > > > let
>>> > > > > > me
>>> > > > > > > > > know what
>>> > > > > > > > > > > >> > you
>>> > > > > > > > > > > >> > > > > think.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
>>> away from
>>> > > > > > July
>>> > > > > > > > > 18th to
>>> > > > > > > > > > > >> > August 8th.
>>> > > > > > > > > > > >> > > > > >> That
>>> > > > > > > > > > > >> > > > > >> > > gives
>>> > > > > > > > > > > >> > > > > >> > > > > > > you a
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to
>>> read and
>>> > > > digest
>>> > > > > > this
>>> > > > > > > > > long
>>> > > > > > > > > > > >> KIP.
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
>>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
>>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > > > > > --
>>> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
>>> > > > > > > > > > > >> > > > > >> > > > > > >
>>> > > > > > > > > > > >> > > > > >> > > > >
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> > >
>>> > > > > > > > > > > >> > > > > >> >
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > > >> --
>>> > > > > > > > > > > >> > > > > >> -- Guozhang
>>> > > > > > > > > > > >> > > > > >>
>>> > > > > > > > > > > >> > > > > >
>>> > > > > > > > > > > >> > > > >
>>> > > > > > > > > > > >> >
>>> > > > > > > > > > > >>
>>> > > > > > > > > > > >
>>> > > > > > > > >
>>> > > > > >
>>> > > > >
>>> > > > >
>>> > > > > --
>>> > > > > -- Guozhang
>>> > > >
>>> > >
>>> > >
>>> > > --
>>> > > -- Guozhang
>>>
>>
>>
>> --
>> -- Guozhang
>>
>
>
> --
> -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hello David,

Just had a few more nit questions about the public APIs:

1) The client-side "PartitionAssignor#Assignment" has an Error byte field,
while the broker-side "PartitionAssignor#Assignment" does not. And the
rationale seems to be assuming that we should always be able to do the
assignment at the broker-side assignor without errors. Personally I think
it's still potentially beneficial to add the Error field even for
broker-side assignors, e.g. for some edge cases where some subscribed
topics are not recognized with the current broker's metadata. What do you
think?

2) The client-side "GroupMember" has three additional fields
reason/version/metadata compared with the broker-side GroupMember. I agree
that broker-side assignor would not need reason/metadata since they are
blackbox strings/bytes to the assignor, but what about version? E.g. is it
possible that we evolve our broker-side built-in assignor but the old
versioned clients would not be able to work with the new version, in which
case we need to let the broker being aware of this and upgrade its behavior
to cooperate with the clients?

3) Also related to 2) above, for the client-side "GroupMember", instead of
including these three fields, what about just adding the "Metadata" field
class which has these three fields? Also, there are two "Metadata"
currently in the APIs, the first is a class that encodes
reason/version/metadata, and the second is just the encoded metadata bytes.
I'm wondering what about just naming the first as memberMetadata, which
then has a bytebuffer field Metadata, or instead naming the second
bytebuffer field as metadataBytes?

Guozhang







On Tue, Sep 13, 2022 at 12:08 PM Guozhang Wang <wa...@gmail.com> wrote:

> Hello David,
>
> Thanks for bringing this question up. I think the main benefits as you
> listed is 2) above if it stays; just to clarify, we would only be able to
> save one round trip if the rebalance is still triggered by the broker; if
> the rebalance is triggered by the client then the num.round.trips are the
> same:
>
> 1) With GroupPrepareAssignment:
>
> T0: client decides to do a new assignment, suppose it has already sent a
> HB and hence has to wait for it to return first since only one request /
> response can be inflight with the coordinator's socket.
> T1: client receives the HB response, and then sends the
> GroupPrepareAssignment request.
> T2: the GroupPrepareAssignment response is returned.
> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> request.
>
> In total, two round trips.
>
> 2) Without GroupPrepareAssignment:
>
> T0: client decides to do a new assignment, suppose it has already sent a
> HB and hence has to wait for it to return first since only one request /
> response can be inflight with the coordinator's socket.
> T1: client receives the HB response, and then sends the new HB request
> with the flag indicating a new rebalance needed..
> T2: the HB response with the optional member metadata map is returned.
> T3: it calculates the new assignment, and sends a GroupInstallAssignment
> request.
>
> In total, two round trips as well.
>
> -----------------------------
>
> So to complete the full picture here, we'd need to modify both HB request
> and response so that the client can also indicate a new rebalance via the
> HB request as well, right?
>
> Assuming all above is true, I think it's okay to merge the
> GroupPrepareAssignment into HB given that we can make the additional fields
> encoding the full member (subscription) metadata and topic metadata as
> optional fields.
>
>
> Guozhang
>
>
>
>
>
>
>
>
>
>
> On Mon, Sep 12, 2022 at 5:22 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
>> Hi all,
>>
>> During an offline conversation, someone asked why we need the
>> ConsumerGroupPrepareAssignment API and suggested that we could instead
>> provide the group state in the heartbeat response. This has a few
>> advantages: 1) it does not require using a special error code to
>> signal that a new assignment is required as the signal would be the
>> provided group state; 2) it removes one round trip when a client side
>> assignor is used. The downside is that it makes the heartbeat
>> response's definition quite large. I recall that I went with the
>> current approach due to this.
>>
>> Providing the group state in the heartbeat response is appealing. What
>> do you guys think?
>>
>> Best,
>> David
>>
>> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io> wrote:
>> >
>> > Hi Guozhang,
>> >
>> > 1. I have added a reference to the relevant chapter instead of
>> > repeating the whole thing. Does that work for you?
>> >
>> > 2. The "Rebalance Triggers" section you are referring to is about when
>> > a rebalance should be triggered for the non-upgraded members using the
>> > old protocol. The section mentions that a rebalance must be triggered
>> > when a new assignment is installed. This implies that the group epoch
>> > was updated either by a native member or a non-upgraded member. For
>> > the latter, the JoinGroup request would be the trigger. I have added a
>> > reference to the relevant chapter in the "JoinGroup Handling" section
>> > as well. Does that make sense?
>> >
>> > Thanks,
>> > David
>> >
>> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
>> wrote:
>> > >
>> > > Hello David,
>> > >
>> > > Alright I think that's sufficient. Just to make that clear in the doc,
>> > > could we update:
>> > >
>> > > 1) the heartbeat request handling section, stating when coordinator
>> will
>> > > trigger rebalance based on the HB's member metadata / reason?
>> > > 2) the "Rebalance Triggers" section to include what we described in
>> "Group
>> > > Epoch - Trigger a rebalance" section as well?
>> > >
>> > >
>> > > Guozhang
>> > >
>> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot
>> <dj...@confluent.io.invalid>
>> > > wrote:
>> > >
>> > > > Hi Guozhang,
>> > > >
>> > > > I thought that the assignor will always be consulted when the next
>> > > > heartbeat request is constructed. In other words,
>> > > > `PartitionAssignor#metadata` will be called for every heartbeat.
>> This
>> > > > gives the opportunity for the assignor to enforce a rebalance by
>> > > > setting the reason to a non-zero value or by changing the bytes. Do
>> > > > you think that this is not sufficient? Are you concerned by the
>> delay?
>> > > >
>> > > > Best,
>> > > > David
>> > > >
>> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
>> wrote:
>> > > > >
>> > > > > Hello David,
>> > > > >
>> > > > > One of Jun's comments make me thinking:
>> > > > >
>> > > > > ```
>> > > > > In this case, a new assignment is triggered by the client side
>> > > > > assignor. When constructing the HB, the consumer will always
>> consult
>> > > > > the client side assignor and propagate the information to the
>> group
>> > > > > coordinator. In other words, we don't expect users to call
>> > > > > Consumer#enforceRebalance anymore.
>> > > > > ```
>> > > > >
>> > > > > As I looked at the current PartitionAssignor's interface, we
>> actually do
>> > > > > not have a way yet to instruct how to construct the next HB
>> request, e.g.
>> > > > > when the assignor wants to enforce a new rebalance with a new
>> assignment,
>> > > > > we'd need some customizable APIs inside the PartitionAssignor to
>> indicate
>> > > > > the next HB telling broker about so. WDYT about adding such an
>> API on the
>> > > > > PartitionAssignor?
>> > > > >
>> > > > >
>> > > > > Guozhang
>> > > > >
>> > > > >
>> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
>> <dj...@confluent.io.invalid>
>> > > > > wrote:
>> > > > >
>> > > > > > Hi Jun,
>> > > > > >
>> > > > > > I have updated the KIP to include your feedback. I have also
>> tried to
>> > > > > > clarify the parts which were not cleared.
>> > > > > >
>> > > > > > Best,
>> > > > > > David
>> > > > > >
>> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <djacot@confluent.io
>> >
>> > > > wrote:
>> > > > > > >
>> > > > > > > Hi Jun,
>> > > > > > >
>> > > > > > > Thanks for your feedback. Let me start by answering your
>> questions
>> > > > > > > inline and I will update the KIP next week.
>> > > > > > >
>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
>> seem to
>> > > > be
>> > > > > > fewer
>> > > > > > > > RPCs during rebalance and more efficient support of
>> wildcard. A few
>> > > > > > > > comments below.
>> > > > > > >
>> > > > > > > I would also add that the KIP removes the global sync barrier
>> in the
>> > > > > > > protocol which is essential to improve group stability and
>> > > > > > > scalability, and the KIP also simplifies the client by moving
>> most of
>> > > > > > > the logic to the server side.
>> > > > > > >
>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
>> rolling
>> > > > > > changing
>> > > > > > > > of the partition assignor in the consumers?
>> > > > > > >
>> > > > > > > Definitely. The group coordinator will use the assignor used
>> by a
>> > > > > > > majority of the members. This allows the group to move from
>> one
>> > > > > > > assignor to another by a roll. This is explained in the
>> Assignor
>> > > > > > > Selection chapter.
>> > > > > > >
>> > > > > > > > 30.2 For each field, could you explain whether it's
>> required in
>> > > > every
>> > > > > > > > request or the scenarios when it needs to be filled? For
>> example,
>> > > > it's
>> > > > > > not
>> > > > > > > > clear to me when TopicPartitions needs to be filled.
>> > > > > > >
>> > > > > > > The client is expected to set those fields in case of a
>> connection
>> > > > > > > issue (e.g. timeout) or when the fields have changed since
>> the last
>> > > > > > > HB. The server populates those fields as long as the member
>> is not
>> > > > > > > fully reconciled - the member should acknowledge that it has
>> the
>> > > > > > > expected epoch and assignment. I will clarify this in the KIP.
>> > > > > > >
>> > > > > > > > 31. In the current consumer protocol, the rack affinity
>> between the
>> > > > > > client
>> > > > > > > > and the broker is only considered during fetching, but not
>> during
>> > > > > > assigning
>> > > > > > > > partitions to consumers. Sometimes, once the assignment is
>> made,
>> > > > there
>> > > > > > is
>> > > > > > > > no opportunity for read affinity because no replicas of
>> assigned
>> > > > > > partitions
>> > > > > > > > are close to the member. I am wondering if we should use
>> this
>> > > > > > opportunity
>> > > > > > > > to address this by including rack in GroupMember.
>> > > > > > >
>> > > > > > > That's an interesting idea. I don't see any issue with adding
>> the
>> > > > rack
>> > > > > > > to the members. I will do so.
>> > > > > > >
>> > > > > > > > 32. On the metric side, often, it's useful to know how busy
>> a group
>> > > > > > > > coordinator is. By moving the event loop model, it seems
>> that we
>> > > > could
>> > > > > > add
>> > > > > > > > a metric that tracks the fraction of the time the event
>> loop is
>> > > > doing
>> > > > > > the
>> > > > > > > > actual work.
>> > > > > > >
>> > > > > > > That's a great idea. I will add it. Thanks.
>> > > > > > >
>> > > > > > > > 33. Could we add a section on coordinator failover
>> handling? For
>> > > > > > example,
>> > > > > > > > does it need to trigger the check if any group with the
>> wildcard
>> > > > > > > > subscription now has a new matching topic?
>> > > > > > >
>> > > > > > > Sure. When the new group coordinator takes over, it has to:
>> > > > > > > * Setup the session timeouts.
>> > > > > > > * Trigger a new assignment if a client side assignor is used.
>> We
>> > > > don't
>> > > > > > > store the information about the member selected to run the
>> assignment
>> > > > > > > so we have to start a new one.
>> > > > > > > * Update the topics metadata, verify the wildcard
>> subscriptions, and
>> > > > > > > trigger a rebalance if needed.
>> > > > > > >
>> > > > > > > > 34. ConsumerGroupMetadataValue,
>> > > > ConsumerGroupPartitionMetadataValue,
>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
>> the epoch
>> > > > > > field
>> > > > > > > > reflects? For example, does the epoch in
>> ConsumerGroupMetadataValue
>> > > > > > reflect
>> > > > > > > > the latest group epoch? What about the one in
>> > > > > > > > ConsumerGroupPartitionMetadataValue and
>> > > > > > ConsumerGroupMemberMetadataValue?
>> > > > > > >
>> > > > > > > Sure. I will clarify that but it is always the latest group
>> epoch.
>> > > > > > > When the group state is updated, the group epoch is bumped so
>> we use
>> > > > > > > that one for all the change records related to the update.
>> > > > > > >
>> > > > > > > > 35. "the group coordinator will ensure that the following
>> > > > invariants
>> > > > > > are
>> > > > > > > > met: ... All members exists." It's possible for a member
>> not to
>> > > > get any
>> > > > > > > > assigned partitions, right?
>> > > > > > >
>> > > > > > > That's right. Here I meant that the members provided by the
>> assignor
>> > > > > > > in the assignment must exist in the group. The assignor can
>> not make
>> > > > > > > up new member ids.
>> > > > > > >
>> > > > > > > > 36. "He can rejoins the group with a member epoch equals to
>> 0":
>> > > > When
>> > > > > > would
>> > > > > > > > a consumer rejoin and what member id would be used?
>> > > > > > >
>> > > > > > > A member is expected to abandon all its partitions and
>> rejoins when
>> > > > it
>> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the
>> group
>> > > > > > > coordinator will have removed the member from the group. The
>> member
>> > > > > > > can rejoin the group with the same member id but with 0 as
>> epoch. Let
>> > > > > > > me see if I can clarify this in the KIP.
>> > > > > > >
>> > > > > > > > 37. "Instead, power users will have the ability to trigger a
>> > > > > > reassignment
>> > > > > > > > by either providing a non-zero reason or by updating the
>> assignor
>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
>> deprecation
>> > > > of
>> > > > > > > > Consumer#enforeRebalance.
>> > > > > > >
>> > > > > > > In this case, a new assignment is triggered by the client side
>> > > > > > > assignor. When constructing the HB, the consumer will always
>> consult
>> > > > > > > the client side assignor and propagate the information to the
>> group
>> > > > > > > coordinator. In other words, we don't expect users to call
>> > > > > > > Consumer#enforceRebalance anymore.
>> > > > > > >
>> > > > > > > > 38. The reassignment examples are nice. But the section
>> seems to
>> > > > have
>> > > > > > > > multiple typos.
>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
>> gets into
>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
>> into
>> > > > > > epoch=3,
>> > > > > > > > partitions=[foo-1] too early.
>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
>> epoch=2,
>> > > > > > > > partitions=[foo-0].
>> > > > > > >
>> > > > > > > Sorry for that! I will revise them.
>> > > > > > >
>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
>> from
>> > > > any
>> > > > > > old
>> > > > > > > > version to new one?
>> > > > > > >
>> > > > > > > We will support upgrading from the consumer protocol version
>> 3,
>> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
>> earliest
>> > > > > > > version is unknown at the moment. This is explained in the
>> migration
>> > > > > > > plan chapter.
>> > > > > > >
>> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
>> based on
>> > > > it
>> > > > > > > next week.
>> > > > > > >
>> > > > > > > Best,
>> > > > > > > David
>> > > > > > >
>> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
>> <ju...@confluent.io.invalid>
>> > > > wrote:
>> > > > > > > >
>> > > > > > > > Hi, David,
>> > > > > > > >
>> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
>> seem to
>> > > > be
>> > > > > > fewer
>> > > > > > > > RPCs during rebalance and more efficient support of
>> wildcard. A few
>> > > > > > > > comments below.
>> > > > > > > >
>> > > > > > > > 30. ConsumerGroupHeartbeatRequest
>> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
>> rolling
>> > > > > > changing
>> > > > > > > > of the partition assignor in the consumers?
>> > > > > > > > 30.2 For each field, could you explain whether it's
>> required in
>> > > > every
>> > > > > > > > request or the scenarios when it needs to be filled? For
>> example,
>> > > > it's
>> > > > > > not
>> > > > > > > > clear to me when TopicPartitions needs to be filled.
>> > > > > > > >
>> > > > > > > > 31. In the current consumer protocol, the rack affinity
>> between the
>> > > > > > client
>> > > > > > > > and the broker is only considered during fetching, but not
>> during
>> > > > > > assigning
>> > > > > > > > partitions to consumers. Sometimes, once the assignment is
>> made,
>> > > > there
>> > > > > > is
>> > > > > > > > no opportunity for read affinity because no replicas of
>> assigned
>> > > > > > partitions
>> > > > > > > > are close to the member. I am wondering if we should use
>> this
>> > > > > > opportunity
>> > > > > > > > to address this by including rack in GroupMember.
>> > > > > > > >
>> > > > > > > > 32. On the metric side, often, it's useful to know how busy
>> a group
>> > > > > > > > coordinator is. By moving the event loop model, it seems
>> that we
>> > > > could
>> > > > > > add
>> > > > > > > > a metric that tracks the fraction of the time the event
>> loop is
>> > > > doing
>> > > > > > the
>> > > > > > > > actual work.
>> > > > > > > >
>> > > > > > > > 33. Could we add a section on coordinator failover
>> handling? For
>> > > > > > example,
>> > > > > > > > does it need to trigger the check if any group with the
>> wildcard
>> > > > > > > > subscription now has a new matching topic?
>> > > > > > > >
>> > > > > > > > 34. ConsumerGroupMetadataValue,
>> > > > ConsumerGroupPartitionMetadataValue,
>> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what
>> the epoch
>> > > > > > field
>> > > > > > > > reflects? For example, does the epoch in
>> ConsumerGroupMetadataValue
>> > > > > > reflect
>> > > > > > > > the latest group epoch? What about the one in
>> > > > > > > > ConsumerGroupPartitionMetadataValue and
>> > > > > > ConsumerGroupMemberMetadataValue?
>> > > > > > > >
>> > > > > > > > 35. "the group coordinator will ensure that the following
>> > > > invariants
>> > > > > > are
>> > > > > > > > met: ... All members exists." It's possible for a member
>> not to
>> > > > get any
>> > > > > > > > assigned partitions, right?
>> > > > > > > >
>> > > > > > > > 36. "He can rejoins the group with a member epoch equals to
>> 0":
>> > > > When
>> > > > > > would
>> > > > > > > > a consumer rejoin and what member id would be used?
>> > > > > > > >
>> > > > > > > > 37. "Instead, power users will have the ability to trigger a
>> > > > > > reassignment
>> > > > > > > > by either providing a non-zero reason or by updating the
>> assignor
>> > > > > > > > metadata." Hmm, this seems to be conflicting with the
>> deprecation
>> > > > of
>> > > > > > > > Consumer#enforeRebalance.
>> > > > > > > >
>> > > > > > > > 38. The reassignment examples are nice. But the section
>> seems to
>> > > > have
>> > > > > > > > multiple typos.
>> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
>> gets into
>> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
>> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
>> into
>> > > > > > epoch=3,
>> > > > > > > > partitions=[foo-1] too early.
>> > > > > > > > 38.3 After A transitions to epoch 3, C still has A -
>> epoch=2,
>> > > > > > > > partitions=[foo-0].
>> > > > > > > >
>> > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
>> from
>> > > > any
>> > > > > > old
>> > > > > > > > version to new one?
>> > > > > > > >
>> > > > > > > > Thanks,
>> > > > > > > >
>> > > > > > > > Jun
>> > > > > > > >
>> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
>> > > > > > <dj...@confluent.io.invalid>
>> > > > > > > > wrote:
>> > > > > > > >
>> > > > > > > > > Hi all,
>> > > > > > > > >
>> > > > > > > > > The KIP states that we will re-implement the coordinator
>> in
>> > > > Java. I
>> > > > > > > > > discussed this offline with a few folks and folks are
>> concerned
>> > > > that
>> > > > > > > > > we could introduce many regressions in the old protocol
>> if we do
>> > > > so.
>> > > > > > > > > Therefore, I am going to remove this statement from the
>> KIP. It
>> > > > is an
>> > > > > > > > > implementation detail after all so it does not have to be
>> > > > decided at
>> > > > > > > > > this stage. We will likely start by trying to refactor the
>> > > > current
>> > > > > > > > > implementation as a first step.
>> > > > > > > > >
>> > > > > > > > > Cheers,
>> > > > > > > > > David
>> > > > > > > > >
>> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
>> djacot@confluent.io
>> > > > >
>> > > > > > wrote:
>> > > > > > > > > >
>> > > > > > > > > > Hi Luke,
>> > > > > > > > > >
>> > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
>> > > > > > reconciling,
>> > > > > > > > > stable,
>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
>> right?
>> > > > > > > > > >
>> > > > > > > > > > This sentence does not refer to those group states but
>> rather
>> > > > to a
>> > > > > > > > > > state machine replication (SMR). This refers to the
>> entire
>> > > > state of
>> > > > > > > > > > group coordinator which is replicated via the log
>> layer. I will
>> > > > > > > > > > clarify this in the KIP.
>> > > > > > > > > >
>> > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
>> as an
>> > > > event
>> > > > > > > > > loop"?
>> > > > > > > > > >
>> > > > > > > > > > The idea is to follow a model similar to the new quorum
>> > > > > > controller. We
>> > > > > > > > > > will have N threads to process events. Each
>> __consumer_offsets
>> > > > > > > > > > partition is assigned to a unique thread and all the
>> events
>> > > > (e.g.
>> > > > > > > > > > requests, callbacks, etc.) are processed by this
>> thread. This
>> > > > > > simplify
>> > > > > > > > > > concurrency and will enable us to do simulation testing
>> for the
>> > > > > > group
>> > > > > > > > > > coordinator.
>> > > > > > > > > >
>> > > > > > > > > > > 1.3. Why do we need a state machine per
>> *__consumer_offsets*
>> > > > > > > > > partitions?
>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
>> group
>> > > > > > coordinator?
>> > > > > > > > > For
>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
>> groups, and
>> > > > > > both
>> > > > > > > > > exist in
>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
>> for it,
>> > > > or
>> > > > > > 2?
>> > > > > > > > > >
>> > > > > > > > > > See 1.1. The confusion comes from there, I think.
>> > > > > > > > > >
>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
>> number
>> > > > of
>> > > > > > threads
>> > > > > > > > > used
>> > > > > > > > > > > to run the state machines. But I'm wondering if the
>> purpose
>> > > > of
>> > > > > > the
>> > > > > > > > > threads
>> > > > > > > > > > > is only to keep the state of each consumer group (or
>> > > > > > > > > *__consumer_offsets*
>> > > > > > > > > > > partitions?), and no heavy computation, why should we
>> need
>> > > > > > > > > multi-threads
>> > > > > > > > > > > here?
>> > > > > > > > > >
>> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
>> > > > processing as
>> > > > > > the
>> > > > > > > > > > computation could be heavy.
>> > > > > > > > > >
>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
>> default
>> > > > session
>> > > > > > > > > timeout not
>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
>> min/max
>> > > > > > session
>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
>> > > > > > > > > > >
>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
>> timeout to
>> > > > detect
>> > > > > > client
>> > > > > > > > > > > failures when using the consumer group protocol.
>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
>> minimum
>> > > > > > session
>> > > > > > > > > timeout.
>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
>> maximum
>> > > > > > session
>> > > > > > > > > timeout.
>> > > > > > > > > >
>> > > > > > > > > > This is indeed a mistake. The default session timeout
>> should
>> > > > be 45s
>> > > > > > > > > > (the current default).
>> > > > > > > > > >
>> > > > > > > > > > > 2.2. The default server side assignor are [range,
>> uniform],
>> > > > > > which means
>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
>> why not
>> > > > > > uniform
>> > > > > > > > > one? I
>> > > > > > > > > > > thought usually users will choose uniform assignor
>> (former
>> > > > sticky
>> > > > > > > > > assinor)
>> > > > > > > > > > > for better evenly distribution. Any other reason we
>> choose
>> > > > range
>> > > > > > > > > assignor
>> > > > > > > > > > > as default?
>> > > > > > > > > > > group.consumer.assignors List range, uniform The
>> server side
>> > > > > > assignors.
>> > > > > > > > > >
>> > > > > > > > > > The order on the server side has no influence because
>> the
>> > > > client
>> > > > > > must
>> > > > > > > > > > chose the selector that he wants to use. There is no
>> default
>> > > > in the
>> > > > > > > > > > current proposal. If the assignor is not specified by
>> the
>> > > > client,
>> > > > > > the
>> > > > > > > > > > request is rejected. The default client value for
>> > > > > > > > > > `group.remote.assignor` is `uniform` though.
>> > > > > > > > > >
>> > > > > > > > > > Thanks for your very good comments, Luke. I hope that my
>> > > > answers
>> > > > > > help
>> > > > > > > > > > to clarify things. I will update the KIP as well based
>> on your
>> > > > > > > > > > feedback.
>> > > > > > > > > >
>> > > > > > > > > > Cheers,
>> > > > > > > > > > David
>> > > > > > > > > >
>> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
>> showuon@gmail.com>
>> > > > > > wrote:
>> > > > > > > > > > >
>> > > > > > > > > > > Hi David,
>> > > > > > > > > > >
>> > > > > > > > > > > Thanks for the update.
>> > > > > > > > > > >
>> > > > > > > > > > > Some more questions:
>> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
>> > > > > > > > > > > > The new group coordinator will have a state machine
>> per
>> > > > > > > > > > > *__consumer_offsets* partitions, where each state
>> machine is
>> > > > > > modelled
>> > > > > > > > > as an
>> > > > > > > > > > > event loop. Those state machines will be executed in
>> > > > > > > > > > > *group.coordinator.threads* threads.
>> > > > > > > > > > >
>> > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
>> > > > > > reconciling,
>> > > > > > > > > stable,
>> > > > > > > > > > > dead" mentioned in Consumer Group States section,
>> right?
>> > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
>> as an
>> > > > event
>> > > > > > > > > loop"?
>> > > > > > > > > > > 1.3. Why do we need a state machine per
>> *__consumer_offsets*
>> > > > > > > > > partitions?
>> > > > > > > > > > > Not a state machine "per consumer group" owned by a
>> group
>> > > > > > coordinator?
>> > > > > > > > > For
>> > > > > > > > > > > example, if one group coordinator owns 2 consumer
>> groups, and
>> > > > > > both
>> > > > > > > > > exist in
>> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
>> for it,
>> > > > or
>> > > > > > 2?
>> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
>> number
>> > > > of
>> > > > > > threads
>> > > > > > > > > used
>> > > > > > > > > > > to run the state machines. But I'm wondering if the
>> purpose
>> > > > of
>> > > > > > the
>> > > > > > > > > threads
>> > > > > > > > > > > is only to keep the state of each consumer group (or
>> > > > > > > > > *__consumer_offsets*
>> > > > > > > > > > > partitions?), and no heavy computation, why should we
>> need
>> > > > > > > > > multi-threads
>> > > > > > > > > > > here?
>> > > > > > > > > > >
>> > > > > > > > > > > 2. For the default value in the new configs:
>> > > > > > > > > > > 2.1. The consumer session timeout, why does the
>> default
>> > > > session
>> > > > > > > > > timeout not
>> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
>> min/max
>> > > > > > session
>> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
>> > > > > > > > > > >
>> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The
>> timeout to
>> > > > detect
>> > > > > > client
>> > > > > > > > > > > failures when using the consumer group protocol.
>> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
>> minimum
>> > > > > > session
>> > > > > > > > > timeout.
>> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
>> maximum
>> > > > > > session
>> > > > > > > > > timeout.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > 2.2. The default server side assignor are [range,
>> uniform],
>> > > > > > which means
>> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
>> why not
>> > > > > > uniform
>> > > > > > > > > one? I
>> > > > > > > > > > > thought usually users will choose uniform assignor
>> (former
>> > > > sticky
>> > > > > > > > > assinor)
>> > > > > > > > > > > for better evenly distribution. Any other reason we
>> choose
>> > > > range
>> > > > > > > > > assignor
>> > > > > > > > > > > as default?
>> > > > > > > > > > > group.consumer.assignors List range, uniform The
>> server side
>> > > > > > assignors.
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > Thank you.
>> > > > > > > > > > > Luke
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > >
>> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
>> showuon@gmail.com
>> > > > >
>> > > > > > wrote:
>> > > > > > > > > > >
>> > > > > > > > > > > > Hi Sagar,
>> > > > > > > > > > > >
>> > > > > > > > > > > > I have some thoughts about Kafka Connect
>> integrating with
>> > > > > > KIP-848,
>> > > > > > > > > but I
>> > > > > > > > > > > > think we should have a separate discussion thread
>> for the
>> > > > Kafka
>> > > > > > > > > Connect
>> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
>> Rebalance
>> > > > > > Protocol
>> > > > > > > > > [1],
>> > > > > > > > > > > > and let this discussion thread focus on consumer
>> rebalance
>> > > > > > protocol,
>> > > > > > > > > WDYT?
>> > > > > > > > > > > >
>> > > > > > > > > > > > [1]
>> > > > > > > > > > > >
>> > > > > > > > >
>> > > > > >
>> > > >
>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>> > > > > > > > > > > >
>> > > > > > > > > > > > Thank you.
>> > > > > > > > > > > > Luke
>> > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > > > >
>> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
>> > > > > > sagarmeansocean@gmail.com>
>> > > > > > > > > wrote:
>> > > > > > > > > > > >
>> > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
>> like
>> > > > there's
>> > > > > > > > > agreement on
>> > > > > > > > > > > >> using separate APIs for Connect. I would revisit
>> the doc
>> > > > and
>> > > > > > see
>> > > > > > > > > what
>> > > > > > > > > > > >> changes are to be made.
>> > > > > > > > > > > >>
>> > > > > > > > > > > >> Thanks!
>> > > > > > > > > > > >> Sagar.
>> > > > > > > > > > > >>
>> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
>> > > > > > > > > <dj...@confluent.io.invalid>
>> > > > > > > > > > > >> wrote:
>> > > > > > > > > > > >>
>> > > > > > > > > > > >> > Hi Sagar,
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >> > Thanks for the feedback and the document. That's
>> really
>> > > > > > helpful. I
>> > > > > > > > > > > >> > will take a look at it.
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >> > Overall, it seems to me that both Connect and the
>> > > > Consumer
>> > > > > > could
>> > > > > > > > > share
>> > > > > > > > > > > >> > the same underlying "engine". The main
>> difference is
>> > > > that
>> > > > > > the
>> > > > > > > > > Consumer
>> > > > > > > > > > > >> > assigns topic-partitions to members whereas
>> Connect
>> > > > assigns
>> > > > > > tasks
>> > > > > > > > > to
>> > > > > > > > > > > >> > workers. I see two ways to move forward:
>> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
>> different
>> > > > > > resource
>> > > > > > > > > types
>> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
>> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
>> dedicated
>> > > > APIs
>> > > > > > > > > would be
>> > > > > > > > > > > >> > similar to the new ones but different on the
>> > > > > > content/resources and
>> > > > > > > > > > > >> > they would rely on the same engine on the
>> coordinator
>> > > > side.
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >> > I personally lean towards 2) because I am not a
>> fan of
>> > > > > > > > > overcharging
>> > > > > > > > > > > >> > APIs to serve different purposes. That being
>> said, I am
>> > > > not
>> > > > > > > > > opposed to
>> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >> > I think that we can continue to discuss it here
>> for now
>> > > > in
>> > > > > > order
>> > > > > > > > > to
>> > > > > > > > > > > >> > ensure that this KIP is compatible with what we
>> will do
>> > > > for
>> > > > > > > > > Connect in
>> > > > > > > > > > > >> > the future.
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >> > Best,
>> > > > > > > > > > > >> > David
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
>> > > > > > djacot@confluent.io>
>> > > > > > > > > wrote:
>> > > > > > > > > > > >> > >
>> > > > > > > > > > > >> > > Hi all,
>> > > > > > > > > > > >> > >
>> > > > > > > > > > > >> > > I am back from vacation. I will go through and
>> address
>> > > > > > your
>> > > > > > > > > comments
>> > > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
>> > > > > > > > > > > >> > >
>> > > > > > > > > > > >> > > Cheers,
>> > > > > > > > > > > >> > > David
>> > > > > > > > > > > >> > >
>> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris
>> <
>> > > > > > > > > gharris1727@gmail.com
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >> > wrote:
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > Hey All!
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
>> > > > cooperative
>> > > > > > > > > rebalancing
>> > > > > > > > > > > >> > making it
>> > > > > > > > > > > >> > > > down the stack!
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > I had a few questions:
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
>> describes how
>> > > > > > member
>> > > > > > > > > epoch
>> > > > > > > > > > > >> > should
>> > > > > > > > > > > >> > > > advance in step with the group epoch and
>> assignment
>> > > > > > epoch
>> > > > > > > > > values. I
>> > > > > > > > > > > >> > think
>> > > > > > > > > > > >> > > > that this is a good idea for the reasons
>> described
>> > > > in
>> > > > > > the
>> > > > > > > > > KIP. When
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > protocol is incrementally assigning
>> partitions to a
>> > > > > > worker,
>> > > > > > > > > what
>> > > > > > > > > > > >> member
>> > > > > > > > > > > >> > > > epoch does each incremental assignment use?
>> Are
>> > > > member
>> > > > > > epochs
>> > > > > > > > > > > >> re-used,
>> > > > > > > > > > > >> > and
>> > > > > > > > > > > >> > > > a single member epoch can correspond to
>> multiple
>> > > > > > different
>> > > > > > > > > > > >> > (monotonically
>> > > > > > > > > > > >> > > > larger) assignments?
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque
>> to the
>> > > > group
>> > > > > > > > > > > >> coordinator? If
>> > > > > > > > > > > >> > > > not, should custom client-side assignor
>> > > > implementations
>> > > > > > > > > interact
>> > > > > > > > > > > >> with
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > Reason field, and how is its common meaning
>> agreed
>> > > > > > upon? If
>> > > > > > > > > so, what
>> > > > > > > > > > > >> > is the
>> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
>> including
>> > > > such
>> > > > > > > > > functionality
>> > > > > > > > > > > >> > in the
>> > > > > > > > > > > >> > > > opaque metadata?
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
>> "Thanks to
>> > > > > > this, the
>> > > > > > > > > input
>> > > > > > > > > > > >> of
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > client side assignor is entirely driven by
>> the group
>> > > > > > > > > coordinator.
>> > > > > > > > > > > >> The
>> > > > > > > > > > > >> > > > consumer is no longer responsible for
>> maintaining
>> > > > any
>> > > > > > state
>> > > > > > > > > besides
>> > > > > > > > > > > >> its
>> > > > > > > > > > > >> > > > assigned partitions." Does this mean that the
>> > > > > > client-side
>> > > > > > > > > assignor
>> > > > > > > > > > > >> MAY
>> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
>> (such as
>> > > > > > partition
>> > > > > > > > > > > >> > throughput,
>> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or
>> that
>> > > > > > additional
>> > > > > > > > > > > >> > non-Metadata
>> > > > > > > > > > > >> > > > state SHOULD NOT be used?
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > 4. I see that there are separate classes
>> > > > > > > > > > > >> > > > for
>> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
>> > > > > > > > > > > >> > > > and
>> > > > org.apache.kafka.clients.consumer.PartitionAssignor
>> > > > > > that
>> > > > > > > > > seem to
>> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
>> these two
>> > > > > > > > > implementations
>> > > > > > > > > > > >> to
>> > > > > > > > > > > >> > be
>> > > > > > > > > > > >> > > > unified? This would serve to promote feature
>> parity
>> > > > of
>> > > > > > > > > server-side
>> > > > > > > > > > > >> and
>> > > > > > > > > > > >> > > > client-side assignors, and would also
>> facilitate
>> > > > > > operational
>> > > > > > > > > > > >> > flexibility in
>> > > > > > > > > > > >> > > > certain situations. For example, if a
>> server-side
>> > > > > > assignor
>> > > > > > > > > has some
>> > > > > > > > > > > >> > poor
>> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
>> patched
>> > > > > > assignor to
>> > > > > > > > > the
>> > > > > > > > > > > >> > client
>> > > > > > > > > > > >> > > > and switching one consumer group to a
>> client-side
>> > > > > > assignor
>> > > > > > > > > may be
>> > > > > > > > > > > >> > faster
>> > > > > > > > > > > >> > > > and less risky than patching all of the
>> brokers.
>> > > > With
>> > > > > > the
>> > > > > > > > > currently
>> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
>> > > > reimplementation
>> > > > > > would
>> > > > > > > > > have
>> > > > > > > > > > > >> to be
>> > > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
>> > > > > > significantly,
>> > > > > > > > > then it
>> > > > > > > > > > > >> is
>> > > > > > > > > > > >> > > > possible that a reimplementation would not be
>> > > > possible.
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > --
>> > > > > > > > > > > >> > > > Greg Harris
>> > > > > > > > > > > >> > > > gharris1727@gmail.com
>> > > > > > > > > > > >> > > > github.com/gharris1727
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
>> > > > > > > > > sagarmeansocean@gmail.com>
>> > > > > > > > > > > >> > wrote:
>> > > > > > > > > > > >> > > >
>> > > > > > > > > > > >> > > > > Hi Guozhang/David,
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> > > > > I created a confluence page to discuss how
>> Connect
>> > > > > > would
>> > > > > > > > > need to
>> > > > > > > > > > > >> > change
>> > > > > > > > > > > >> > > > > based on the new rebalance protocol.
>> Here's the
>> > > > page:
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >>
>> > > > > > > > >
>> > > > > >
>> > > >
>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> > > > > It's also pretty longish and I have tried
>> to keep
>> > > > a
>> > > > > > format
>> > > > > > > > > > > >> similar to
>> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also,
>> do you
>> > > > > > think this
>> > > > > > > > > > > >> should
>> > > > > > > > > > > >> > be
>> > > > > > > > > > > >> > > > > moved to a separate discussion thread or
>> is this
>> > > > one
>> > > > > > fine?
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> > > > > Thanks!
>> > > > > > > > > > > >> > > > > Sagar.
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
>> > > > > > > > > sagarmeansocean@gmail.com>
>> > > > > > > > > > > >> > wrote:
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> > > > > > Hello Guozhang,
>> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
>> Streams.
>> > > > > > Sure, I
>> > > > > > > > > would do
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > > analysis and come up with such a
>> document.
>> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > > Thanks!
>> > > > > > > > > > > >> > > > > > Sagar.
>> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang
>> Wang <
>> > > > > > > > > > > >> wangguoz@gmail.com>
>> > > > > > > > > > > >> > > > > wrote:
>> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> Hello Sagar,
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > > >> It would be great if you could come
>> back with
>> > > > some
>> > > > > > > > > analysis on
>> > > > > > > > > > > >> > how to
>> > > > > > > > > > > >> > > > > >> implement the Connect side integration
>> with
>> > > > the new
>> > > > > > > > > protocol;
>> > > > > > > > > > > >> so
>> > > > > > > > > > > >> > far
>> > > > > > > > > > > >> > > > > >> besides leveraging on the new "protocol
>> type"
>> > > > we
>> > > > > > did not
>> > > > > > > > > yet
>> > > > > > > > > > > >> think
>> > > > > > > > > > > >> > > > > through
>> > > > > > > > > > > >> > > > > >> the Connect side implementations. For
>> Streams
>> > > > > > here's a
>> > > > > > > > > draft of
>> > > > > > > > > > > >> > > > > >> integration
>> > > > > > > > > > > >> > > > > >> plan:
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >>
>> > > > > > > > >
>> > > > > >
>> > > >
>> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
>> > > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
>> > > > > > > > > > > >> sagarmeansocean@gmail.com
>> > > > > > > > > > > >> > >
>> > > > > > > > > > > >> > > > > wrote:
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > > >> > Hi David,
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
>> reason I
>> > > > thought
>> > > > > > > > > connect can
>> > > > > > > > > > > >> > also fit
>> > > > > > > > > > > >> > > > > >> into
>> > > > > > > > > > > >> > > > > >> > this new scheme is that even today the
>> > > > connect
>> > > > > > uses a
>> > > > > > > > > > > >> > > > > WorkerCoordinator
>> > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to
>> empower
>> > > > > > > > > rebalances of
>> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
>> WorkerCoordinator sets
>> > > > the
>> > > > > > > > > > > >> protocolType()
>> > > > > > > > > > > >> > to
>> > > > > > > > > > > >> > > > > >> connect
>> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
>> plumbing
>> > > > into
>> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> > I think the changes to support
>> connect would
>> > > > be
>> > > > > > > > > similar at a
>> > > > > > > > > > > >> > high
>> > > > > > > > > > > >> > > > > level
>> > > > > > > > > > > >> > > > > >> to
>> > > > > > > > > > > >> > > > > >> > the changes in streams mainly because
>> of the
>> > > > > > Client
>> > > > > > > > > side
>> > > > > > > > > > > >> > assignors
>> > > > > > > > > > > >> > > > > being
>> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
>> level, we
>> > > > > > might
>> > > > > > > > > need to
>> > > > > > > > > > > >> make
>> > > > > > > > > > > >> > a lot
>> > > > > > > > > > > >> > > > > of
>> > > > > > > > > > > >> > > > > >> > changes to get onto this new
>> assignment
>> > > > protocol
>> > > > > > like
>> > > > > > > > > > > >> enhancing
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
>> SyncGroup and
>> > > > > > using
>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
>> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
>> streams (or
>> > > > > > there
>> > > > > > > > > might be
>> > > > > > > > > > > >> > > > > >> deviations). I
>> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
>> analysis of
>> > > > the
>> > > > > > same
>> > > > > > > > > and we
>> > > > > > > > > > > >> > can have
>> > > > > > > > > > > >> > > > > a
>> > > > > > > > > > > >> > > > > >> > separate discussion thread for that
>> as that
>> > > > would
>> > > > > > > > > derail this
>> > > > > > > > > > > >> > > > > discussion
>> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
>> good to
>> > > > you.
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> > Thanks!
>> > > > > > > > > > > >> > > > > >> > Sagar.
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David
>> Jacot
>> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> > wrote:
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
>> `Assignment#error`.
>> > > > > > Sure, I
>> > > > > > > > > can
>> > > > > > > > > > > >> > mention it.
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from
>> his
>> > > > current
>> > > > > > > > > assignment
>> > > > > > > > > > > >> to
>> > > > > > > > > > > >> > his
>> > > > > > > > > > > >> > > > > >> > > target assignment when he can move
>> to
>> > > > epoch 3.
>> > > > > > When
>> > > > > > > > > that
>> > > > > > > > > > > >> > happens,
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
>> persisted
>> > > > > > with all
>> > > > > > > > > its
>> > > > > > > > > > > >> > assigned
>> > > > > > > > > > > >> > > > > >> > > partitions even if they are not all
>> revoked
>> > > > > > yet. In
>> > > > > > > > > other
>> > > > > > > > > > > >> > words, the
>> > > > > > > > > > > >> > > > > >> > > member assignment becomes the target
>> > > > > > assignment.
>> > > > > > > > > This is
>> > > > > > > > > > > >> > basically
>> > > > > > > > > > > >> > > > > an
>> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
>> write all
>> > > > the
>> > > > > > > > > changes to
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > log.
>> > > > > > > > > > > >> > > > > The
>> > > > > > > > > > > >> > > > > >> > > examples are based on the persisted
>> state
>> > > > so I
>> > > > > > > > > understand
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
>> improve
>> > > > this in
>> > > > > > the
>> > > > > > > > > > > >> > description.
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could
>> reuse the
>> > > > > > protocol
>> > > > > > > > > with a
>> > > > > > > > > > > >> > client side
>> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the
>> protocol. The
>> > > > > > assignment
>> > > > > > > > > is
>> > > > > > > > > > > >> about
>> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
>> > > > Connect
>> > > > > > fit
>> > > > > > > > > into this?
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> > > Best,
>> > > > > > > > > > > >> > > > > >> > > David
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
>> Sagar <
>> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
>> > > > > > > > > > > >> > > > > >> wrote:
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > Hi David,
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
>> minor
>> > > > > > observations:
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error
>> section in
>> > > > Client
>> > > > > > Side
>> > > > > > > > > mode
>> > > > > > > > > > > >> > Assignment
>> > > > > > > > > > > >> > > > > >> > > process,
>> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case,
>> the
>> > > > client
>> > > > > > side
>> > > > > > > > > assignor
>> > > > > > > > > > > >> can
>> > > > > > > > > > > >> > > > > return
>> > > > > > > > > > > >> > > > > >> an
>> > > > > > > > > > > >> > > > > >> > > > error to the group coordinator`.
>> In this
>> > > > > > case are
>> > > > > > > > > you
>> > > > > > > > > > > >> > referring to
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
>> AssignmentError
>> > > > that's
>> > > > > > > > > listed down
>> > > > > > > > > > > >> > towards
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > end?
>> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would
>> make sense
>> > > > to
>> > > > > > > > > mention this
>> > > > > > > > > > > >> > > > > explicitly
>> > > > > > > > > > > >> > > > > >> > here?
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I
>> have a
>> > > > > > slight
>> > > > > > > > > > > >> confusion,
>> > > > > > > > > > > >> > not
>> > > > > > > > > > > >> > > > > sure
>> > > > > > > > > > > >> > > > > >> if
>> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
>> this step:
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
>> coordinator
>> > > > > > > > > transitions him
>> > > > > > > > > > > >> to
>> > > > > > > > > > > >> > epoch
>> > > > > > > > > > > >> > > > > 3
>> > > > > > > > > > > >> > > > > >> > > because
>> > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
>> > > > persists
>> > > > > > the
>> > > > > > > > > change and
>> > > > > > > > > > > >> > reply.
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
>> > > > > > > > > > > >> > > > > >> > > >       - A
>> > > > > > > > > > > >> > > > > >> > > >       - B
>> > > > > > > > > > > >> > > > > >> > > >       - C
>> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
>> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
>> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
>> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
>> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
>> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
>> partitions=[foo-0,
>> > > > > > foo-1]
>> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
>> partitions=[foo-2]
>> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
>> partitions=[foo-1]
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions
>> to
>> > > > epoch 3
>> > > > > > but
>> > > > > > > > > cannot
>> > > > > > > > > > > >> get
>> > > > > > > > > > > >> > foo-1
>> > > > > > > > > > > >> > > > > >> yet.
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C
>> can't
>> > > > get
>> > > > > > the
>> > > > > > > > > foo-1
>> > > > > > > > > > > >> > partition
>> > > > > > > > > > > >> > > > > yet,
>> > > > > > > > > > > >> > > > > >> > but
>> > > > > > > > > > > >> > > > > >> > > > based on the description above,
>> it seems
>> > > > it
>> > > > > > > > > already has
>> > > > > > > > > > > >> it.
>> > > > > > > > > > > >> > Do you
>> > > > > > > > > > > >> > > > > >> > think
>> > > > > > > > > > > >> > > > > >> > > it
>> > > > > > > > > > > >> > > > > >> > > > would be better to remove it and
>> > > > populate it
>> > > > > > only
>> > > > > > > > > when it
>> > > > > > > > > > > >> > actually
>> > > > > > > > > > > >> > > > > >> gets
>> > > > > > > > > > > >> > > > > >> > > it?
>> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
>> places, so
>> > > > have
>> > > > > > I
>> > > > > > > > > > > >> understood it
>> > > > > > > > > > > >> > > > > >> > incorrectly
>> > > > > > > > > > > >> > > > > >> > > ?
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
>> out of
>> > > > scope
>> > > > > > of
>> > > > > > > > > this
>> > > > > > > > > > > >> > discussion,
>> > > > > > > > > > > >> > > > > but
>> > > > > > > > > > > >> > > > > >> > from
>> > > > > > > > > > > >> > > > > >> > > > what I understood it would
>> probably be
>> > > > > > running in
>> > > > > > > > > client
>> > > > > > > > > > > >> > side
>> > > > > > > > > > > >> > > > > >> assignor
>> > > > > > > > > > > >> > > > > >> > > mode
>> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance
>> protocol as it
>> > > > has
>> > > > > > its
>> > > > > > > > > own
>> > > > > > > > > > > >> Custom
>> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
>> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > Thanks!
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > Sagar.
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
>> David
>> > > > Jacot
>> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
>> move
>> > > > forward
>> > > > > > with
>> > > > > > > > > > > >> > specialized API
>> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
>> generic API.
>> > > > For
>> > > > > > > > > Connect, we
>> > > > > > > > > > > >> > can apply
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
>> reuse/share the
>> > > > core
>> > > > > > > > > > > >> > implementation on
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
>> registry, I
>> > > > > > think
>> > > > > > > > > that we
>> > > > > > > > > > > >> > should
>> > > > > > > > > > > >> > > > > >> consider
>> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
>> simple
>> > > > > > > > > membership/leader
>> > > > > > > > > > > >> > election.
>> > > > > > > > > > > >> > > > > >> > > > >
>> > > > > > > > > > > >> > > > > >> > > > > Best,
>> > > > > > > > > > > >> > > > > >> > > > > David
>> > > > > > > > > > > >> > > > > >> > > > >
>> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22
>> AM Ismael
>> > > > > > Juma <
>> > > > > > > > > > > >> > ismael@juma.me.uk
>> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
>> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
>> java.util.regex.Pattern
>> > > > vs
>> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
>> > > > > > > > > > > >> > > > > >> > > > > should
>> > > > > > > > > > > >> > > > > >> > > > > > document the differences in
>> more
>> > > > detail
>> > > > > > before
>> > > > > > > > > > > >> deciding
>> > > > > > > > > > > >> > one
>> > > > > > > > > > > >> > > > > way
>> > > > > > > > > > > >> > > > > >> or
>> > > > > > > > > > > >> > > > > >> > > > > another.
>> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
>> > > > > > > > > java.util.regex.Pattern,
>> > > > > > > > > > > >> they
>> > > > > > > > > > > >> > expect
>> > > > > > > > > > > >> > > > > >> > their
>> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If
>> we are
>> > > > doing
>> > > > > > > > > something
>> > > > > > > > > > > >> > different,
>> > > > > > > > > > > >> > > > > >> then
>> > > > > > > > > > > >> > > > > >> > we
>> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
>> overload
>> > > > with
>> > > > > > our own
>> > > > > > > > > > > >> Pattern
>> > > > > > > > > > > >> > class
>> > > > > > > > > > > >> > > > > (I
>> > > > > > > > > > > >> > > > > >> > > don't
>> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
>> re2j's at
>> > > > this
>> > > > > > > > > point).
>> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
>> major new
>> > > > > > protocol
>> > > > > > > > > should
>> > > > > > > > > > > >> > > > > integrate
>> > > > > > > > > > > >> > > > > >> > fully
>> > > > > > > > > > > >> > > > > >> > > > > with
>> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
>> > > > > > recreation case
>> > > > > > > > > > > >> > correctly.
>> > > > > > > > > > > >> > > > > That's
>> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > >> > > > > >> > > main
>> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
>> agree with
>> > > > > > David
>> > > > > > > > > that we'd
>> > > > > > > > > > > >> > want to
>> > > > > > > > > > > >> > > > > add
>> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > >> > > > > >> > > > > ids
>> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols
>> that don't
>> > > > > > have it
>> > > > > > > > > yet and
>> > > > > > > > > > > >> > that we
>> > > > > > > > > > > >> > > > > can
>> > > > > > > > > > > >> > > > > >> > > probably
>> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals versus
>> adding
>> > > > new
>> > > > > > APIs
>> > > > > > > > > to the
>> > > > > > > > > > > >> > Java
>> > > > > > > > > > > >> > > > > >> Consumer
>> > > > > > > > > > > >> > > > > >> > > > > (unless
>> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs
>> is
>> > > > required
>> > > > > > for
>> > > > > > > > > > > >> reasonable
>> > > > > > > > > > > >> > > > > >> semantics).
>> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about
>> the
>> > > > > > coordinator
>> > > > > > > > > storing
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > >> configs.
>> > > > > > > > > > > >> > > > > >> > > It's
>> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
>> > > > centralized
>> > > > > > in the
>> > > > > > > > > > > >> metadata
>> > > > > > > > > > > >> > log for
>> > > > > > > > > > > >> > > > > >> > > various
>> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
>> visibility,
>> > > > > > consistency,
>> > > > > > > > > > > >> etc.).
>> > > > > > > > > > > >> > > > > >> Similarly, I
>> > > > > > > > > > > >> > > > > >> > > am
>> > > > > > > > > > > >> > > > > >> > > > > not
>> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
>> deleting
>> > > > > > configs in a
>> > > > > > > > > way
>> > > > > > > > > > > >> that
>> > > > > > > > > > > >> > they
>> > > > > > > > > > > >> > > > > >> cannot
>> > > > > > > > > > > >> > > > > >> > > be
>> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property
>> for modern
>> > > > > > systems
>> > > > > > > > > is to
>> > > > > > > > > > > >> > minimize
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > number
>> > > > > > > > > > > >> > > > > >> > > > > of
>> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
>> scenarios.
>> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > Ismael
>> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47
>> PM David
>> > > > > > Jacot
>> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
>> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers
>> are
>> > > > below:
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
>> especially
>> > > > > > the last
>> > > > > > > > > step
>> > > > > > > > > > > >> when
>> > > > > > > > > > > >> > > > > clients
>> > > > > > > > > > > >> > > > > >> > > flip the
>> > > > > > > > > > > >> > > > > >> > > > > > > flag
>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
>> protocol, in
>> > > > which
>> > > > > > we
>> > > > > > > > > would
>> > > > > > > > > > > >> have a
>> > > > > > > > > > > >> > > > > window
>> > > > > > > > > > > >> > > > > >> > where
>> > > > > > > > > > > >> > > > > >> > > > > both
>> > > > > > > > > > > >> > > > > >> > > > > > > new
>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
>> > > > protocols /
>> > > > > > rpcs
>> > > > > > > > > are
>> > > > > > > > > > > >> used
>> > > > > > > > > > > >> > by
>> > > > > > > > > > > >> > > > > >> members
>> > > > > > > > > > > >> > > > > >> > of
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > same
>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
>> coordinator could
>> > > > > > "mimic"
>> > > > > > > > > the old
>> > > > > > > > > > > >> > behavior
>> > > > > > > > > > > >> > > > > >> while
>> > > > > > > > > > > >> > > > > >> > > > > using the
>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
>> we
>> > > > need to
>> > > > > > > > > present
>> > > > > > > > > > > >> about.
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a
>> new
>> > > > version
>> > > > > > of KIP
>> > > > > > > > > which
>> > > > > > > > > > > >> > includes
>> > > > > > > > > > > >> > > > > >> more
>> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
>> > > > "Supporting
>> > > > > > > > > Online
>> > > > > > > > > > > >> > Consumer
>> > > > > > > > > > > >> > > > > Group
>> > > > > > > > > > > >> > > > > >> > > Upgrade"
>> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
>> > > > Deprecation,
>> > > > > > and
>> > > > > > > > > Migration
>> > > > > > > > > > > >> > Plan". I
>> > > > > > > > > > > >> > > > > >> think
>> > > > > > > > > > > >> > > > > >> > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
>> few cases
>> > > > > > now but
>> > > > > > > > > the
>> > > > > > > > > > > >> > overall idea
>> > > > > > > > > > > >> > > > > >> and
>> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
>> understandable.
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
>> ids. So
>> > > > far as
>> > > > > > > > > KIP-516 the
>> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > >> > > > > >> are
>> > > > > > > > > > > >> > > > > >> > > only
>> > > > > > > > > > > >> > > > > >> > > > > used
>> > > > > > > > > > > >> > > > > >> > > > > > > as
>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
>> client,
>> > > > but
>> > > > > > they
>> > > > > > > > > are not
>> > > > > > > > > > > >> > exposed
>> > > > > > > > > > > >> > > > > via
>> > > > > > > > > > > >> > > > > >> any
>> > > > > > > > > > > >> > > > > >> > > > > public
>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
>> the
>> > > > > > question is,
>> > > > > > > > > first
>> > > > > > > > > > > >> > should we
>> > > > > > > > > > > >> > > > > >> let
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
>> the
>> > > > names
>> > > > > > -> ids
>> > > > > > > > > mapping
>> > > > > > > > > > > >> > itself
>> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> > fully
>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
>> existing
>> > > > RPCs
>> > > > > > and the
>> > > > > > > > > new
>> > > > > > > > > > > >> RPCs
>> > > > > > > > > > > >> > with
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > >> > > > > >> > > > > ids;
>> > > > > > > > > > > >> > > > > >> > > > > > > and
>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
>> consider
>> > > > > > exposing
>> > > > > > > > > the
>> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > >> > > > > in
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > consumer
>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
>> > > > > > > > > subscribe/assign, as
>> > > > > > > > > > > >> well
>> > > > > > > > > > > >> > as in
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > rebalance
>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
>> topic
>> > > > > > > > > > > >> > deletion-and-recreation).
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
>> include
>> > > > > > converting
>> > > > > > > > > all
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > offsets
>> > > > > > > > > > > >> > > > > >> > > related
>> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in
>> this
>> > > > KIP,
>> > > > > > the
>> > > > > > > > > consumer
>> > > > > > > > > > > >> > would be
>> > > > > > > > > > > >> > > > > >> able
>> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
>> ids. That
>> > > > > > being
>> > > > > > > > > said, it
>> > > > > > > > > > > >> > still has
>> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> > > provide
>> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in various
>> APIs so
>> > > > > > having a
>> > > > > > > > > > > >> mapping
>> > > > > > > > > > > >> > in the
>> > > > > > > > > > > >> > > > > >> > > consumer
>> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
>> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
>> opinion on
>> > > > > > this.
>> > > > > > > > > Here I
>> > > > > > > > > > > >> > wonder if
>> > > > > > > > > > > >> > > > > >> this
>> > > > > > > > > > > >> > > > > >> > > goes
>> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
>> KIP. I
>> > > > would
>> > > > > > rather
>> > > > > > > > > focus
>> > > > > > > > > > > >> on
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> > internals
>> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider
>> this
>> > > > > > separately if
>> > > > > > > > > we see
>> > > > > > > > > > > >> > value in
>> > > > > > > > > > > >> > > > > >> doing
>> > > > > > > > > > > >> > > > > >> > > it.
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
>> point about
>> > > > > > using
>> > > > > > > > > topic ids
>> > > > > > > > > > > >> > in the
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> ConsumerGroupHeartbeatRequest, I
>> > > > think
>> > > > > > that
>> > > > > > > > > there
>> > > > > > > > > > > >> is
>> > > > > > > > > > > >> > one
>> > > > > > > > > > > >> > > > > >> > advantage
>> > > > > > > > > > > >> > > > > >> > > in
>> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer
>> will
>> > > > have
>> > > > > > the
>> > > > > > > > > > > >> opportunity
>> > > > > > > > > > > >> > to
>> > > > > > > > > > > >> > > > > >> validate
>> > > > > > > > > > > >> > > > > >> > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
>> passing
>> > > > them
>> > > > > > into
>> > > > > > > > > the
>> > > > > > > > > > > >> group
>> > > > > > > > > > > >> > > > > rebalance
>> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
>> > > > coordinator
>> > > > > > will
>> > > > > > > > > also
>> > > > > > > > > > > >> notice
>> > > > > > > > > > > >> > it but
>> > > > > > > > > > > >> > > > > >> it
>> > > > > > > > > > > >> > > > > >> > > does
>> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
>> reject an
>> > > > > > invalid
>> > > > > > > > > topic in
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> response.
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
>> on all
>> > > > other
>> > > > > > minor
>> > > > > > > > > > > >> questions
>> > > > > > > > > > > >> > > > > except
>> > > > > > > > > > > >> > > > > >> for
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
>> question:
>> > > > > > personally
>> > > > > > > > > I think
>> > > > > > > > > > > >> > it's not
>> > > > > > > > > > > >> > > > > >> > > necessary
>> > > > > > > > > > > >> > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
>> API with
>> > > > > > Pattern,
>> > > > > > > > > but
>> > > > > > > > > > > >> > instead we
>> > > > > > > > > > > >> > > > > >> still
>> > > > > > > > > > > >> > > > > >> > > use
>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
>> that our
>> > > > > > > > > subscription may
>> > > > > > > > > > > >> be
>> > > > > > > > > > > >> > > > > >> rejected by
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > server.
>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
>> case is a
>> > > > > > very rare
>> > > > > > > > > > > >> scenario
>> > > > > > > > > > > >> > I felt
>> > > > > > > > > > > >> > > > > >> > using
>> > > > > > > > > > > >> > > > > >> > > an
>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
>> > > > > > subscription may
>> > > > > > > > > be
>> > > > > > > > > > > >> more
>> > > > > > > > > > > >> > > > > >> vulnerable
>> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > >> > > > > >> > > > > various
>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
>> look at
>> > > > the
>> > > > > > > > > differences
>> > > > > > > > > > > >> > between
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > two
>> > > > > > > > > > > >> > > > > >> > > > > > > engines to better
>> understand the
>> > > > > > potential
>> > > > > > > > > issues.
>> > > > > > > > > > > >> My
>> > > > > > > > > > > >> > > > > >> > > understanding is
>> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all the
>> basic
>> > > > > > regular
>> > > > > > > > > > > >> > expressions. The
>> > > > > > > > > > > >> > > > > >> > > differences
>> > > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
>> about
>> > > > the
>> > > > > > various
>> > > > > > > > > > > >> character
>> > > > > > > > > > > >> > > > > >> classes. I
>> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
>> think
>> > > > about
>> > > > > > this.
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > Best,
>> > > > > > > > > > > >> > > > > >> > > > > > > David
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
>> 11:28 PM
>> > > > > > Guozhang
>> > > > > > > > > Wang <
>> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
>> > > > > > > > > > > >> > > > > >> > > >
>> > > > > > > > > > > >> > > > > >> > > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on
>> the high
>> > > > > > level
>> > > > > > > > > there are
>> > > > > > > > > > > >> > two meta
>> > > > > > > > > > > >> > > > > >> > > points we
>> > > > > > > > > > > >> > > > > >> > > > > need
>> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
>> especially
>> > > > > > the last
>> > > > > > > > > step
>> > > > > > > > > > > >> when
>> > > > > > > > > > > >> > > > > clients
>> > > > > > > > > > > >> > > > > >> > > flip the
>> > > > > > > > > > > >> > > > > >> > > > > > > flag
>> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
>> protocol, in
>> > > > which
>> > > > > > we
>> > > > > > > > > would
>> > > > > > > > > > > >> have a
>> > > > > > > > > > > >> > > > > window
>> > > > > > > > > > > >> > > > > >> > where
>> > > > > > > > > > > >> > > > > >> > > > > both
>> > > > > > > > > > > >> > > > > >> > > > > > > new
>> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
>> > > > protocols /
>> > > > > > rpcs
>> > > > > > > > > are
>> > > > > > > > > > > >> used
>> > > > > > > > > > > >> > by
>> > > > > > > > > > > >> > > > > >> members
>> > > > > > > > > > > >> > > > > >> > of
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > same
>> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the
>> coordinator could
>> > > > > > "mimic"
>> > > > > > > > > the old
>> > > > > > > > > > > >> > behavior
>> > > > > > > > > > > >> > > > > >> while
>> > > > > > > > > > > >> > > > > >> > > > > using the
>> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
>> we
>> > > > need to
>> > > > > > > > > present
>> > > > > > > > > > > >> about.
>> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic
>> ids. So
>> > > > far as
>> > > > > > > > > KIP-516 the
>> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > >> > > > > >> are
>> > > > > > > > > > > >> > > > > >> > > only
>> > > > > > > > > > > >> > > > > >> > > > > used
>> > > > > > > > > > > >> > > > > >> > > > > > > as
>> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
>> client,
>> > > > but
>> > > > > > they
>> > > > > > > > > are not
>> > > > > > > > > > > >> > exposed
>> > > > > > > > > > > >> > > > > via
>> > > > > > > > > > > >> > > > > >> any
>> > > > > > > > > > > >> > > > > >> > > > > public
>> > > > > > > > > > > >> > > > > >> > > > > > > APIs
>> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
>> the
>> > > > > > question is,
>> > > > > > > > > first
>> > > > > > > > > > > >> > should we
>> > > > > > > > > > > >> > > > > >> let
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > consumer
>> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
>> the
>> > > > names
>> > > > > > -> ids
>> > > > > > > > > mapping
>> > > > > > > > > > > >> > itself
>> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> > fully
>> > > > > > > > > > > >> > > > > >> > > > > > > leverage
>> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
>> existing
>> > > > RPCs
>> > > > > > and the
>> > > > > > > > > new
>> > > > > > > > > > > >> RPCs
>> > > > > > > > > > > >> > with
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > >> > > > > >> > > > > ids;
>> > > > > > > > > > > >> > > > > >> > > > > > > and
>> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
>> consider
>> > > > > > exposing
>> > > > > > > > > the
>> > > > > > > > > > > >> > topic ids
>> > > > > > > > > > > >> > > > > in
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > consumer
>> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
>> > > > > > > > > subscribe/assign, as
>> > > > > > > > > > > >> well
>> > > > > > > > > > > >> > as in
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > rebalance
>> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
>> topic
>> > > > > > > > > > > >> > deletion-and-recreation).
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David
>> on all
>> > > > other
>> > > > > > minor
>> > > > > > > > > > > >> questions
>> > > > > > > > > > > >> > > > > except
>> > > > > > > > > > > >> > > > > >> for
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
>> question:
>> > > > > > personally
>> > > > > > > > > I think
>> > > > > > > > > > > >> > it's not
>> > > > > > > > > > > >> > > > > >> > > necessary
>> > > > > > > > > > > >> > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
>> API with
>> > > > > > Pattern,
>> > > > > > > > > but
>> > > > > > > > > > > >> > instead we
>> > > > > > > > > > > >> > > > > >> still
>> > > > > > > > > > > >> > > > > >> > > use
>> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
>> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
>> that our
>> > > > > > > > > subscription may
>> > > > > > > > > > > >> be
>> > > > > > > > > > > >> > > > > >> rejected by
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > server.
>> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
>> case is a
>> > > > > > very rare
>> > > > > > > > > > > >> scenario
>> > > > > > > > > > > >> > I felt
>> > > > > > > > > > > >> > > > > >> > using
>> > > > > > > > > > > >> > > > > >> > > an
>> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
>> > > > > > subscription may
>> > > > > > > > > be
>> > > > > > > > > > > >> more
>> > > > > > > > > > > >> > > > > >> vulnerable
>> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > >> > > > > >> > > > > various
>> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
>> 5:23 AM
>> > > > > > David Jacot
>> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
>> feedback. Let
>> > > > me
>> > > > > > answer
>> > > > > > > > > your
>> > > > > > > > > > > >> > questions
>> > > > > > > > > > > >> > > > > >> > inline.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
>> premature to
>> > > > > > talk about
>> > > > > > > > > > > >> target
>> > > > > > > > > > > >> > > > > versions
>> > > > > > > > > > > >> > > > > >> for
>> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
>> existing group
>> > > > > > protocol.
>> > > > > > > > > > > >> Unlike
>> > > > > > > > > > > >> > KRaft,
>> > > > > > > > > > > >> > > > > >> this
>> > > > > > > > > > > >> > > > > >> > > > > affects a
>> > > > > > > > > > > >> > > > > >> > > > > > > core
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
>> hence
>> > > > > > > > > deprecation/removal
>> > > > > > > > > > > >> > will be
>> > > > > > > > > > > >> > > > > >> > heavily
>> > > > > > > > > > > >> > > > > >> > > > > > > dependent
>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
>> applications
>> > > > migrate
>> > > > > > to
>> > > > > > > > > the new
>> > > > > > > > > > > >> > protocol.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
>> > > > remove it.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
>> intend to
>> > > > > > release
>> > > > > > > > > this in
>> > > > > > > > > > > >> > 4.x, but
>> > > > > > > > > > > >> > > > > it
>> > > > > > > > > > > >> > > > > >> > > wasn't
>> > > > > > > > > > > >> > > > > >> > > > > made
>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that
>> as a
>> > > > way to
>> > > > > > > > > estimate
>> > > > > > > > > > > >> when
>> > > > > > > > > > > >> > we'd
>> > > > > > > > > > > >> > > > > >> > > deprecate
>> > > > > > > > > > > >> > > > > >> > > > > and
>> > > > > > > > > > > >> > > > > >> > > > > > > remove
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I
>> also
>> > > > > > suggest
>> > > > > > > > > removing
>> > > > > > > > > > > >> > this part.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
>> reasoning. As
>> > > > > > > > > explained, I
>> > > > > > > > > > > >> plan
>> > > > > > > > > > > >> > to
>> > > > > > > > > > > >> > > > > >> rewrite
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
>> while we
>> > > > > > implement
>> > > > > > > > > the new
>> > > > > > > > > > > >> > protocol.
>> > > > > > > > > > > >> > > > > >> This
>> > > > > > > > > > > >> > > > > >> > > means
>> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals will
>> be
>> > > > slightly
>> > > > > > > > > different
>> > > > > > > > > > > >> > (e.g.
>> > > > > > > > > > > >> > > > > >> threading
>> > > > > > > > > > > >> > > > > >> > > > > model).
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
>> tighten
>> > > > the
>> > > > > > > > > switch from
>> > > > > > > > > > > >> > the old
>> > > > > > > > > > > >> > > > > >> group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
>> group
>> > > > > > coordinator
>> > > > > > > > > to a
>> > > > > > > > > > > >> > major
>> > > > > > > > > > > >> > > > > >> release.
>> > > > > > > > > > > >> > > > > >> > > The
>> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to
>> use a
>> > > > flag
>> > > > > > to do
>> > > > > > > > > the
>> > > > > > > > > > > >> > switch
>> > > > > > > > > > > >> > > > > >> instead
>> > > > > > > > > > > >> > > > > >> > of
>> > > > > > > > > > > >> > > > > >> > > > > relying
>> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
>> out the
>> > > > > > details of
>> > > > > > > > > the
>> > > > > > > > > > > >> > migration
>> > > > > > > > > > > >> > > > > >> story.
>> > > > > > > > > > > >> > > > > >> > > It
>> > > > > > > > > > > >> > > > > >> > > > > sounds
>> > > > > > > > > > > >> > > > > >> > > > > > > > > like
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
>> support
>> > > > > > online
>> > > > > > > > > > > >> migrations.
>> > > > > > > > > > > >> > Is
>> > > > > > > > > > > >> > > > > that
>> > > > > > > > > > > >> > > > > >> > > correct?
>> > > > > > > > > > > >> > > > > >> > > > > We
>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
>> detail. It
>> > > > could
>> > > > > > also
>> > > > > > > > > be done
>> > > > > > > > > > > >> > as a
>> > > > > > > > > > > >> > > > > >> separate
>> > > > > > > > > > > >> > > > > >> > > KIP,
>> > > > > > > > > > > >> > > > > >> > > > > if
>> > > > > > > > > > > >> > > > > >> > > > > > > it's
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
>> online
>> > > > > > migrations
>> > > > > > > > > for the
>> > > > > > > > > > > >> > group.
>> > > > > > > > > > > >> > > > > That
>> > > > > > > > > > > >> > > > > >> > > means
>> > > > > > > > > > > >> > > > > >> > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
>> protocol
>> > > > > > will be
>> > > > > > > > > able to
>> > > > > > > > > > > >> > switch to
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > new
>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain
>> how that
>> > > > > > will work
>> > > > > > > > > > > >> though.
>> > > > > > > > > > > >> > It is
>> > > > > > > > > > > >> > > > > >> > > basically a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
>> > > > upgraded or
>> > > > > > > > > rolled to a
>> > > > > > > > > > > >> > software
>> > > > > > > > > > > >> > > > > >> > > supporting
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
>> Both the
>> > > > > > old and
>> > > > > > > > > the new
>> > > > > > > > > > > >> > > > > >> coordinator
>> > > > > > > > > > > >> > > > > >> > > will
>> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old
>> protocol and
>> > > > rely
>> > > > > > on the
>> > > > > > > > > same
>> > > > > > > > > > > >> > persisted
>> > > > > > > > > > > >> > > > > >> > > metadata so
>> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work together.
>> This
>> > > > point
>> > > > > > is an
>> > > > > > > > > > > >> offline
>> > > > > > > > > > > >> > > > > >> migration.
>> > > > > > > > > > > >> > > > > >> > We
>> > > > > > > > > > > >> > > > > >> > > > > cannot
>> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live
>> because it
>> > > > would
>> > > > > > require
>> > > > > > > > > > > >> > shutting down
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > current
>> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and
>> starting up the
>> > > > > > new one
>> > > > > > > > > and
>> > > > > > > > > > > >> that
>> > > > > > > > > > > >> > would
>> > > > > > > > > > > >> > > > > >> cause
>> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's
>> metadata
>> > > > > > version/IBP
>> > > > > > > > > must be
>> > > > > > > > > > > >> > upgraded
>> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> X
>> > > > > > > > > > > >> > > > > >> > in
>> > > > > > > > > > > >> > > > > >> > > > > order
>> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
>> protocol.
>> > > > This
>> > > > > > cannot
>> > > > > > > > > be done
>> > > > > > > > > > > >> > before
>> > > > > > > > > > > >> > > > > 1)
>> > > > > > > > > > > >> > > > > >> is
>> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the
>> old
>> > > > > > coordinator
>> > > > > > > > > doesn't
>> > > > > > > > > > > >> > support
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > new
>> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
>> > > > upgraded
>> > > > > > to a
>> > > > > > > > > version
>> > > > > > > > > > > >> > > > > supporting
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > online
>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
>> KIP-792).
>> > > > If
>> > > > > > the
>> > > > > > > > > consumer
>> > > > > > > > > > > >> is
>> > > > > > > > > > > >> > > > > already
>> > > > > > > > > > > >> > > > > >> > > there.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at
>> this
>> > > > point.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must
>> be rolled
>> > > > > > with the
>> > > > > > > > > > > >> feature
>> > > > > > > > > > > >> > flag
>> > > > > > > > > > > >> > > > > >> turned
>> > > > > > > > > > > >> > > > > >> > > on.
>> > > > > > > > > > > >> > > > > >> > > > > The
>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
>> automatically
>> > > > > > > > > converted when
>> > > > > > > > > > > >> > the first
>> > > > > > > > > > > >> > > > > >> > > consumer
>> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
>> joins
>> > > > the
>> > > > > > group.
>> > > > > > > > > While
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > members
>> > > > > > > > > > > >> > > > > >> > > using the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
>> > > > upgraded,
>> > > > > > the old
>> > > > > > > > > > > >> protocol
>> > > > > > > > > > > >> > is
>> > > > > > > > > > > >> > > > > >> proxied
>> > > > > > > > > > > >> > > > > >> > > into
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
>> this in
>> > > > the
>> > > > > > KIP.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we
>> are
>> > > > > > pushing the
>> > > > > > > > > pattern
>> > > > > > > > > > > >> > > > > >> subscriptions
>> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like
>> there
>> > > > could be
>> > > > > > some
>> > > > > > > > > tricky
>> > > > > > > > > > > >> > > > > >> compatibility
>> > > > > > > > > > > >> > > > > >> > > > > issues.
>> > > > > > > > > > > >> > > > > >> > > > > > > Will
>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
>> users to
>> > > > > > detect
>> > > > > > > > > that
>> > > > > > > > > > > >> they
>> > > > > > > > > > > >> > need to
>> > > > > > > > > > > >> > > > > >> > update
>> > > > > > > > > > > >> > > > > >> > > > > their
>> > > > > > > > > > > >> > > > > >> > > > > > > regex
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
>> the new
>> > > > > > protocol?
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit
>> more
>> > > > > > optimistic
>> > > > > > > > > than
>> > > > > > > > > > > >> you
>> > > > > > > > > > > >> > on this
>> > > > > > > > > > > >> > > > > >> > > point. I
>> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
>> majority of
>> > > > the
>> > > > > > cases
>> > > > > > > > > are
>> > > > > > > > > > > >> simple
>> > > > > > > > > > > >> > > > > regexes
>> > > > > > > > > > > >> > > > > >> > which
>> > > > > > > > > > > >> > > > > >> > > > > should
>> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
>> engine. The
>> > > > > > > > > coordinator will
>> > > > > > > > > > > >> > verify
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > regex
>> > > > > > > > > > > >> > > > > >> > > > > anyway
>> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer
>> if the
>> > > > > > regex is
>> > > > > > > > > not
>> > > > > > > > > > > >> valid.
>> > > > > > > > > > > >> > > > > Coming
>> > > > > > > > > > > >> > > > > >> > back
>> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
>> worst
>> > > > case,
>> > > > > > the
>> > > > > > > > > first
>> > > > > > > > > > > >> > upgraded
>> > > > > > > > > > > >> > > > > >> > consumer
>> > > > > > > > > > > >> > > > > >> > > > > joining
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
>> rejected.
>> > > > This
>> > > > > > should
>> > > > > > > > > be used
>> > > > > > > > > > > >> > as the
>> > > > > > > > > > > >> > > > > >> last
>> > > > > > > > > > > >> > > > > >> > > > > defence, I
>> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
>> > > > validate
>> > > > > > their
>> > > > > > > > > regex
>> > > > > > > > > > > >> > before
>> > > > > > > > > > > >> > > > > >> > upgrading
>> > > > > > > > > > > >> > > > > >> > > > > their
>> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
>> them with
>> > > > > > another
>> > > > > > > > > group.
>> > > > > > > > > > > >> For
>> > > > > > > > > > > >> > > > > >> instance,
>> > > > > > > > > > > >> > > > > >> > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
>> pre-prod
>> > > > > > environment.
>> > > > > > > > > > > >> Another
>> > > > > > > > > > > >> > way
>> > > > > > > > > > > >> > > > > >> would be
>> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
>> consumer-group tool
>> > > > to
>> > > > > > > > > provide a
>> > > > > > > > > > > >> regex
>> > > > > > > > > > > >> > > > > >> validation
>> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
>> > > > enough in
>> > > > > > your
>> > > > > > > > > > > >> opinion?
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
>> > > > question,
>> > > > > > will
>> > > > > > > > > the
>> > > > > > > > > > > >> Java
>> > > > > > > > > > > >> > client
>> > > > > > > > > > > >> > > > > >> allow
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > users to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the
>> current regex
>> > > > > > engine for
>> > > > > > > > > > > >> > compatibility
>> > > > > > > > > > > >> > > > > >> > > reasons?
>> > > > > > > > > > > >> > > > > >> > > > > For
>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
>> keep using
>> > > > > > client
>> > > > > > > > > based
>> > > > > > > > > > > >> > regex at
>> > > > > > > > > > > >> > > > > >> first
>> > > > > > > > > > > >> > > > > >> > to
>> > > > > > > > > > > >> > > > > >> > > keep
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and
>> then
>> > > > > > migrate to
>> > > > > > > > > server
>> > > > > > > > > > > >> > based
>> > > > > > > > > > > >> > > > > >> regexes
>> > > > > > > > > > > >> > > > > >> > > as a
>> > > > > > > > > > > >> > > > > >> > > > > > > second
>> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your point
>> but I
>> > > > am
>> > > > > > > > > concerned that
>> > > > > > > > > > > >> > this
>> > > > > > > > > > > >> > > > > would
>> > > > > > > > > > > >> > > > > >> > > allow
>> > > > > > > > > > > >> > > > > >> > > > > users
>> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in
>> this mode.
>> > > > That
>> > > > > > > > > would go
>> > > > > > > > > > > >> > against our
>> > > > > > > > > > > >> > > > > >> goal
>> > > > > > > > > > > >> > > > > >> > > of
>> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
>> because
>> > > > we
>> > > > > > would
>> > > > > > > > > have to
>> > > > > > > > > > > >> > continue
>> > > > > > > > > > > >> > > > > >> > > monitoring
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
>> client
>> > > > side. I
>> > > > > > would
>> > > > > > > > > rather
>> > > > > > > > > > > >> > not do
>> > > > > > > > > > > >> > > > > >> this.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
>> the group
>> > > > > > > > > coordinator
>> > > > > > > > > > > >> will
>> > > > > > > > > > > >> > be
>> > > > > > > > > > > >> > > > > >> > > responsible for
>> > > > > > > > > > > >> > > > > >> > > > > > > storing
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations
>> and that
>> > > > the
>> > > > > > > > > > > >> configurations
>> > > > > > > > > > > >> > will be
>> > > > > > > > > > > >> > > > > >> > > deleted
>> > > > > > > > > > > >> > > > > >> > > > > when
>> > > > > > > > > > > >> > > > > >> > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted.
>> Will a
>> > > > > > transition to
>> > > > > > > > > DEAD
>> > > > > > > > > > > >> > trigger
>> > > > > > > > > > > >> > > > > >> > deletion
>> > > > > > > > > > > >> > > > > >> > > of
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
>> > > > configurations
>> > > > > > will be
>> > > > > > > > > > > >> deleted
>> > > > > > > > > > > >> > when
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > group is
>> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
>> together.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to
>> store
>> > > > the
>> > > > > > > > > configs in
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > group
>> > > > > > > > > > > >> > > > > >> > > coordinator
>> > > > > > > > > > > >> > > > > >> > > > > > > make it
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
>> cluster
>> > > > > > configs and
>> > > > > > > > > their
>> > > > > > > > > > > >> > values?
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
>> group
>> > > > > > > > > configurations are
>> > > > > > > > > > > >> > overrides
>> > > > > > > > > > > >> > > > > >> of
>> > > > > > > > > > > >> > > > > >> > > cluster
>> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to
>> know
>> > > > all
>> > > > > > the
>> > > > > > > > > overrides
>> > > > > > > > > > > >> > though,
>> > > > > > > > > > > >> > > > > you
>> > > > > > > > > > > >> > > > > >> > > would
>> > > > > > > > > > > >> > > > > >> > > > > have
>> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
>> > > > coordinators.
>> > > > > > You
>> > > > > > > > > cannot
>> > > > > > > > > > > >> > rely on
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > metadata
>> > > > > > > > > > > >> > > > > >> > > > > log
>> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
>> > > > configure a
>> > > > > > group
>> > > > > > > > > before
>> > > > > > > > > > > >> > starting
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > consumers?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered
>> allowing the
>> > > > > > explicit
>> > > > > > > > > creation
>> > > > > > > > > > > >> of
>> > > > > > > > > > > >> > > > > groups?
>> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
>> decoupled
>> > > > from
>> > > > > > the
>> > > > > > > > > group
>> > > > > > > > > > > >> > lifecycle.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
>> > > > > > automatically
>> > > > > > > > > created in
>> > > > > > > > > > > >> > this
>> > > > > > > > > > > >> > > > > case.
>> > > > > > > > > > > >> > > > > >> > > However,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be
>> lost after
>> > > > the
>> > > > > > > > > retention
>> > > > > > > > > > > >> > period of
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
>> > > > Consumer.subscribe
>> > > > > > method
>> > > > > > > > > for the
>> > > > > > > > > > > >> > Java
>> > > > > > > > > > > >> > > > > >> client
>> > > > > > > > > > > >> > > > > >> > > still
>> > > > > > > > > > > >> > > > > >> > > > > take
>> > > > > > > > > > > >> > > > > >> > > > > > > a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>> `java.util.regex.Pattern` of
>> > > > do
>> > > > > > we
>> > > > > > > > > have to
>> > > > > > > > > > > >> > introduce
>> > > > > > > > > > > >> > > > > an
>> > > > > > > > > > > >> > > > > >> > > overload?
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
>> question. I
>> > > > > > forgot
>> > > > > > > > > about
>> > > > > > > > > > > >> that
>> > > > > > > > > > > >> > one.
>> > > > > > > > > > > >> > > > > As
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> `java.util.regex.Pattern` is
>> > > > not
>> > > > > > fully
>> > > > > > > > > > > >> compatible
>> > > > > > > > > > > >> > with
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > engine
>> > > > > > > > > > > >> > > > > >> > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it
>> might be
>> > > > better
>> > > > > > to
>> > > > > > > > > deprecate
>> > > > > > > > > > > >> > it and
>> > > > > > > > > > > >> > > > > >> use an
>> > > > > > > > > > > >> > > > > >> > > > > overload
>> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string.
>> We would
>> > > > > > rely on
>> > > > > > > > > the
>> > > > > > > > > > > >> server
>> > > > > > > > > > > >> > side
>> > > > > > > > > > > >> > > > > >> > > validation.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I
>> think
>> > > > that
>> > > > > > we
>> > > > > > > > > could
>> > > > > > > > > > > >> still
>> > > > > > > > > > > >> > try to
>> > > > > > > > > > > >> > > > > >> > > toString
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
>> should
>> > > > > > work, I
>> > > > > > > > > think, in
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> majority
>> > > > > > > > > > > >> > > > > >> > > of the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
>> Justine
>> > > > that we
>> > > > > > > > > should be
>> > > > > > > > > > > >> > clearer
>> > > > > > > > > > > >> > > > > about
>> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > >> > > > > >> > > > > reason
>> > > > > > > > > > > >> > > > > >> > > > > > > to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
>> > > > IBP/metadata.version
>> > > > > > from the
>> > > > > > > > > > > >> feature
>> > > > > > > > > > > >> > flag.
>> > > > > > > > > > > >> > > > > >> Maybe
>> > > > > > > > > > > >> > > > > >> > > we
>> > > > > > > > > > > >> > > > > >> > > > > mean
>> > > > > > > > > > > >> > > > > >> > > > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the
>> default for
>> > > > the
>> > > > > > feature
>> > > > > > > > > flag
>> > > > > > > > > > > >> to
>> > > > > > > > > > > >> > true
>> > > > > > > > > > > >> > > > > >> based
>> > > > > > > > > > > >> > > > > >> > on
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once
>> we
>> > > > want to
>> > > > > > make
>> > > > > > > > > it the
>> > > > > > > > > > > >> > default.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that
>> feature
>> > > > > > flag
>> > > > > > > > > mainly
>> > > > > > > > > > > >> > during the
>> > > > > > > > > > > >> > > > > >> > > development
>> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
>> > > > mentioned
>> > > > > > it, I
>> > > > > > > > > think,
>> > > > > > > > > > > >> > because
>> > > > > > > > > > > >> > > > > we
>> > > > > > > > > > > >> > > > > >> > > could
>> > > > > > > > > > > >> > > > > >> > > > > use
>> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for
>> it.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
>> protocol APIs
>> > > > > > don't
>> > > > > > > > > mention
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > required
>> > > > > > > > > > > >> > > > > >> > > ACLs, it
>> > > > > > > > > > > >> > > > > >> > > > > > > would
>> > > > > > > > > > > >> > > > > >> > > > > > > > > be
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
>> > > > consistency.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
>> that
>> > > > > > > > > > > >> ConsumerGroupHeartbeat
>> > > > > > > > > > > >> > > > > requires
>> > > > > > > > > > > >> > > > > >> > > "Read
>> > > > > > > > > > > >> > > > > >> > > > > Group"
>> > > > > > > > > > > >> > > > > >> > > > > > > > > even
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do
>> more
>> > > > than
>> > > > > > > > > reading.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
>> > > > current
>> > > > > > > > > protocol works
>> > > > > > > > > > > >> > though.
>> > > > > > > > > > > >> > > > > We
>> > > > > > > > > > > >> > > > > >> > only
>> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to
>> join a
>> > > > > > group. We
>> > > > > > > > > could
>> > > > > > > > > > > >> > consider
>> > > > > > > > > > > >> > > > > >> > > changing
>> > > > > > > > > > > >> > > > > >> > > > > this
>> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that
>> it is
>> > > > worth
>> > > > > > it.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
>> recreation
>> > > > > > handled by
>> > > > > > > > > the
>> > > > > > > > > > > >> > consumer
>> > > > > > > > > > > >> > > > > with
>> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > >> > > > > >> > > new
>> > > > > > > > > > > >> > > > > >> > > > > > > group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be
>> good to
>> > > > > > have a
>> > > > > > > > > section
>> > > > > > > > > > > >> on
>> > > > > > > > > > > >> > this.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
>> > > > > > perspective, the
>> > > > > > > > > new
>> > > > > > > > > > > >> topic
>> > > > > > > > > > > >> > will
>> > > > > > > > > > > >> > > > > >> have a
>> > > > > > > > > > > >> > > > > >> > > new
>> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
>> treat it
>> > > > like a
>> > > > > > > > > topic with
>> > > > > > > > > > > >> a
>> > > > > > > > > > > >> > > > > different
>> > > > > > > > > > > >> > > > > >> > > name.
>> > > > > > > > > > > >> > > > > >> > > > > The
>> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
>> > > > fetch/commit
>> > > > > > > > > offsets
>> > > > > > > > > > > >> APIs
>> > > > > > > > > > > >> > do not
>> > > > > > > > > > > >> > > > > >> > support
>> > > > > > > > > > > >> > > > > >> > > > > topic
>> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer
>> would
>> > > > reuse the
>> > > > > > > > > offsets
>> > > > > > > > > > > >> based
>> > > > > > > > > > > >> > on the
>> > > > > > > > > > > >> > > > > >> > same.
>> > > > > > > > > > > >> > > > > >> > > I
>> > > > > > > > > > > >> > > > > >> > > > > think
>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
>> those
>> > > > APIs
>> > > > > > as well
>> > > > > > > > > in
>> > > > > > > > > > > >> order
>> > > > > > > > > > > >> > to be
>> > > > > > > > > > > >> > > > > >> > > consistent
>> > > > > > > > > > > >> > > > > >> > > > > end
>> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
>> strengthen
>> > > > the
>> > > > > > > > > semantics of
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > >> consumer.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions
>> we will
>> > > > > > write the
>> > > > > > > > > new
>> > > > > > > > > > > >> > coordinator
>> > > > > > > > > > > >> > > > > >> in
>> > > > > > > > > > > >> > > > > >> > > Java.
>> > > > > > > > > > > >> > > > > >> > > > > Even
>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
>> implementation
>> > > > > > detail, do
>> > > > > > > > > we plan
>> > > > > > > > > > > >> to
>> > > > > > > > > > > >> > have a
>> > > > > > > > > > > >> > > > > >> new
>> > > > > > > > > > > >> > > > > >> > > gradle
>> > > > > > > > > > > >> > > > > >> > > > > > > module
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
>> scalability
>> > > > > > goal when
>> > > > > > > > > it
>> > > > > > > > > > > >> comes
>> > > > > > > > > > > >> > to how
>> > > > > > > > > > > >> > > > > >> many
>> > > > > > > > > > > >> > > > > >> > > > > members
>> > > > > > > > > > > >> > > > > >> > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > new
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
>> support?
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers
>> at the
>> > > > > > moment. The
>> > > > > > > > > > > >> protocol
>> > > > > > > > > > > >> > should
>> > > > > > > > > > > >> > > > > >> > > support
>> > > > > > > > > > > >> > > > > >> > > > > 1000s
>> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group.
>> We will
>> > > > > > measure
>> > > > > > > > > this when
>> > > > > > > > > > > >> > we have
>> > > > > > > > > > > >> > > > > a
>> > > > > > > > > > > >> > > > > >> > first
>> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
>> that we
>> > > > might
>> > > > > > have
>> > > > > > > > > other
>> > > > > > > > > > > >> > > > > bottlenecks
>> > > > > > > > > > > >> > > > > >> > down
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
>> commits).
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
>> having
>> > > > > > > > > SubscribedTopidIds
>> > > > > > > > > > > >> > instead
>> > > > > > > > > > > >> > > > > of
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames
>> in
>> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
>> > > > > > > > > > > >> > > > > >> Is
>> > > > > > > > > > > >> > > > > >> > the
>> > > > > > > > > > > >> > > > > >> > > > > idea
>> > > > > > > > > > > >> > > > > >> > > > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
>> resolve the
>> > > > > > regex on
>> > > > > > > > > the
>> > > > > > > > > > > >> > server, we
>> > > > > > > > > > > >> > > > > >> can do
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > same
>> > > > > > > > > > > >> > > > > >> > > > > > > for
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
>> > > > difference
>> > > > > > is that
>> > > > > > > > > > > >> sending
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> regex is
>> > > > > > > > > > > >> > > > > >> > > more
>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
>> topic
>> > > > names
>> > > > > > is less
>> > > > > > > > > > > >> > efficient.
>> > > > > > > > > > > >> > > > > >> > > Furthermore,
>> > > > > > > > > > > >> > > > > >> > > > > > > delete
>> > > > > > > > > > > >> > > > > >> > > > > > > > > and
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier
>> to
>> > > > handle
>> > > > > > if we
>> > > > > > > > > have
>> > > > > > > > > > > >> topic
>> > > > > > > > > > > >> > ids.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
>> consolidate the
>> > > > > > metadata
>> > > > > > > > > lookup
>> > > > > > > > > > > >> > on the
>> > > > > > > > > > > >> > > > > >> server
>> > > > > > > > > > > >> > > > > >> > > for
>> > > > > > > > > > > >> > > > > >> > > > > both
>> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
>> with your
>> > > > > > point. As
>> > > > > > > > > a
>> > > > > > > > > > > >> second
>> > > > > > > > > > > >> > > > > though,
>> > > > > > > > > > > >> > > > > >> > using
>> > > > > > > > > > > >> > > > > >> > > > > topic
>> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here
>> for the
>> > > > > > delete and
>> > > > > > > > > > > >> > recreation
>> > > > > > > > > > > >> > > > > case.
>> > > > > > > > > > > >> > > > > >> > > Also, I
>> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may
>> allow
>> > > > users to
>> > > > > > > > > subscribe
>> > > > > > > > > > > >> with
>> > > > > > > > > > > >> > topic
>> > > > > > > > > > > >> > > > > >> ids
>> > > > > > > > > > > >> > > > > >> > in
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that is
>> the only
>> > > > > > way to be
>> > > > > > > > > > > >> really
>> > > > > > > > > > > >> > robust
>> > > > > > > > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> > > topic
>> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > David
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at
>> 1:38 PM
>> > > > > > David
>> > > > > > > > > Jacot <
>> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
>> comments.
>> > > > Please
>> > > > > > find
>> > > > > > > > > my
>> > > > > > > > > > > >> answers
>> > > > > > > > > > > >> > > > > below.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new
>> protocol
>> > > > relies on
>> > > > > > > > > topic IDs
>> > > > > > > > > > > >> > with the
>> > > > > > > > > > > >> > > > > >> > > exception
>> > > > > > > > > > > >> > > > > >> > > > > of the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in
>> the
>> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
>> > > > > > > > > > > >> > > > > >> I
>> > > > > > > > > > > >> > > > > >> > am
>> > > > > > > > > > > >> > > > > >> > > not
>> > > > > > > > > > > >> > > > > >> > > > > sure
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names
>> is the
>> > > > > > right call
>> > > > > > > > > here.
>> > > > > > > > > > > >> I
>> > > > > > > > > > > >> > need to
>> > > > > > > > > > > >> > > > > >> > think
>> > > > > > > > > > > >> > > > > >> > > > > about it
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
>> Obviously, the
>> > > > > > KIP does
>> > > > > > > > > not
>> > > > > > > > > > > >> > change the
>> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
>> topic
>> > > > IDs.
>> > > > > > This
>> > > > > > > > > may be
>> > > > > > > > > > > >> > something
>> > > > > > > > > > > >> > > > > >> that
>> > > > > > > > > > > >> > > > > >> > we
>> > > > > > > > > > > >> > > > > >> > > > > should
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
>> would
>> > > > give
>> > > > > > better
>> > > > > > > > > > > >> overall
>> > > > > > > > > > > >> > > > > >> guarantee in
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
>> think that
>> > > > I
>> > > > > > should
>> > > > > > > > > not
>> > > > > > > > > > > >> have
>> > > > > > > > > > > >> > > > > mentioned
>> > > > > > > > > > > >> > > > > >> > this
>> > > > > > > > > > > >> > > > > >> > > > > flag at
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove
>> it. We can
>> > > > > > use an
>> > > > > > > > > internal
>> > > > > > > > > > > >> > > > > >> configuration
>> > > > > > > > > > > >> > > > > >> > > while
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the
>> feature.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
>> will be
>> > > > > > > > > supported. The
>> > > > > > > > > > > >> > change is
>> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
>> that the
>> > > > > > cluster
>> > > > > > > > > uses
>> > > > > > > > > > > >> topic
>> > > > > > > > > > > >> > IDs.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
>> > > > > > > > > > > >> > > > > >> > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022
>> at 9:53
>> > > > PM
>> > > > > > > > > Guozhang
>> > > > > > > > > > > >> Wang <
>> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
>> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
>> feedback.
>> > > > Here
>> > > > > > are
>> > > > > > > > > some
>> > > > > > > > > > > >> replies
>> > > > > > > > > > > >> > > > > inlined
>> > > > > > > > > > > >> > > > > >> > > below:
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022
>> at
>> > > > 2:53 AM
>> > > > > > > > > Ismael Juma
>> > > > > > > > > > > >> <
>> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
>> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
>> KIP. This
>> > > > has
>> > > > > > the
>> > > > > > > > > > > >> potential
>> > > > > > > > > > > >> > to be a
>> > > > > > > > > > > >> > > > > >> > great
>> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
>> > > > questions/comments:
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
>> > > > premature to
>> > > > > > talk
>> > > > > > > > > about
>> > > > > > > > > > > >> > target
>> > > > > > > > > > > >> > > > > >> versions
>> > > > > > > > > > > >> > > > > >> > > for
>> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
>> existing
>> > > > group
>> > > > > > > > > protocol.
>> > > > > > > > > > > >> > Unlike
>> > > > > > > > > > > >> > > > > >> KRaft,
>> > > > > > > > > > > >> > > > > >> > > this
>> > > > > > > > > > > >> > > > > >> > > > > > > affects a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > core
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
>> and hence
>> > > > > > > > > > > >> > deprecation/removal will
>> > > > > > > > > > > >> > > > > >> be
>> > > > > > > > > > > >> > > > > >> > > heavily
>> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
>> applications
>> > > > > > migrate
>> > > > > > > > > to the
>> > > > > > > > > > > >> new
>> > > > > > > > > > > >> > > > > >> protocol.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
>> you. I
>> > > > think
>> > > > > > we can
>> > > > > > > > > > > >> remove
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> proposed
>> > > > > > > > > > > >> > > > > >> > > > > timeline
>> > > > > > > > > > > >> > > > > >> > > > > > > in
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
>> > > > Deprecation,
>> > > > > > and
>> > > > > > > > > Migration
>> > > > > > > > > > > >> > Plan` and
>> > > > > > > > > > > >> > > > > >> > > instead
>> > > > > > > > > > > >> > > > > >> > > > > just
>> > > > > > > > > > > >> > > > > >> > > > > > > state
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide
>> in the
>> > > > > > future
>> > > > > > > > > about
>> > > > > > > > > > > >> when
>> > > > > > > > > > > >> > we
>> > > > > > > > > > > >> > > > > would
>> > > > > > > > > > > >> > > > > >> > > > > deprecate old
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
>> behaviors.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
>> > > > intend to
>> > > > > > > > > release this
>> > > > > > > > > > > >> > in 4.x,
>> > > > > > > > > > > >> > > > > >> but
>> > > > > > > > > > > >> > > > > >> > it
>> > > > > > > > > > > >> > > > > >> > > > > wasn't
>> > > > > > > > > > > >> > > > > >> > > > > > > made
>> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
>> that as
>> > > > a
>> > > > > > way to
>> > > > > > > > > > > >> estimate
>> > > > > > > > > > > >> > when
>> > > > > > > > > > > >> > > > > we'd
>> > > > > > > > > > > >> > > > > >> > > > > deprecate
>> > > > > > > > > > > >> > > > > >> > > > > > > and
>> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
>> protocol, I
>> > > > also
>> > > > > > suggest
>> > > > > > > > > > > >> removing
>> > > > > > > > > > > >> > this
>> > > > > > > > > > > >> > > > > >> part.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
>> > > > specifically
>> > > > > > > > > related to
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
>> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more
>> for
>> > > > client
>> > > > > > > > > upgrades.
>> > > > > > > > > > > >> I.e.
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> > > broker-side
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may
>> be done
>> > > > > > first,
>> > > > > > > > > and then
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > client
>> > > > > > > > > > > >> > > > > >> > side,
>> > > > > > > > > > > >> > > > > >> > > > > and we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > would
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
>> "released"
>> > > > by
>> > > > > > the
>> > > > > > > > > time
>> > > > > > > > > > > >> clients
>> > > > > > > > > > > >> > > > > >> > > implementations
>> > > > > > > > > > > >> > > > > >> > > > > are
>> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to
>> enable the
>> > > > > > feature the
>> > > > > > > > > > > >> clients
>> > > > > > > > > > > >> > need to
>> > > > > > > > > > > >> > > > > >> > first
>> > > > > > > > > > > >> > > > > >> > > > > swap-in
>> > > > > > > > > > > >> > > > > >> > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
>> rolling
>> > > > bounce
>> > > > > > and
>> > > > > > > > > then set
>> > > > > > > > > > > >> > the flag
>> > > > > > > > > > > >> > > > > >> > with a
>> > > > > > > > > > > >> > > > > >> > > > > second
>> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence
>> we feel
>> > > > it's
>> > > > > > > > > better to be
>> > > > > > > > > > > >> > released
>> > > > > > > > > > > >> > > > > >> in a
>> > > > > > > > > > > >> > > > > >> > > major
>> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
>> flesh out
>> > > > the
>> > > > > > > > > details of
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > migration
>> > > > > > > > > > > >> > > > > >> > > story.
>> > > > > > > > > > > >> > > > > >> > > > > It
>> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
>> will
>> > > > support
>> > > > > > online
>> > > > > > > > > > > >> > migrations. Is
>> > > > > > > > > > > >> > > > > >> that
>> > > > > > > > > > > >> > > > > >> > > > > correct?
>> > > > > > > > > > > >> > > > > >> > > > > > > We
>> > > > > > > > > > > >> > > > > >> > > > > > > > > should
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
>> detail.
>> > > > It
>> > > > > > could
>> > > > > > > > > also be
>> > > > > > > > > > > >> > done as a
>> > > > > > > > > > > >> > > > > >> > > separate
>> > > > > > > > > > > >> > > > > >> > > > > KIP,
>> > > > > > > > > > > >> > > > > >> > > > > > > if
>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's
>> the
>> > > > part we
>> > > > > > can
>> > > > > > > > > be more
>> > > > > > > > > > > >> > concrete
>> > > > > > > > > > > >> > > > > >> about
>> > > > > > > > > > > >> > > > > >> > > for
>> > > > > > > > > > > >> > > > > >> > > > > sure
>> > > > > > > > > > > >> > > > > >> > > > > > > (and
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to
>> your
>> > > > > > question 2)
>> > > > > > > > > above).
>> > > > > > > > > > > >> > We will
>> > > > > > > > > > > >> > > > > >> work
>> > > > > > > > > > > >> > > > > >> > on
>> > > > > > > > > > > >> > > > > >> > > > > making
>> > > > > > > > > > > >> > > > > >> > > > > > > it
>> > > > > > > > > > > >> > > > > >> > > > > > > > > more
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in
>> parallel as we
>> > > > > > solicit
>> > > > > > > > > more
>> > > > > > > > > > > >> > feedback.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy
>> that we are
>> > > > > > pushing
>> > > > > > > > > the
>> > > > > > > > > > > >> > pattern
>> > > > > > > > > > > >> > > > > >> > > subscriptions
>> > > > > > > > > > > >> > > > > >> > > > > to
>> > > > > > > > > > > >> > > > > >> > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like
>> there
>> > > > > > could be
>> > > > > > > > > some
>> > > > > > > > > > > >> tricky
>> > > > > > > > > > > >> > > > > >> > > compatibility
>> > > > > > > > > > > >> > > > > >> > > > > > > issues.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism
>> for
>> > > > users to
>> > > > > > > > > detect that
>> > > > > > > > > > > >> > they
>> > > > > > > > > > > >> > > > > need
>> > > > > > > > > > > >> > > > > >> to
>> > > > > > > > > > > >> > > > > >> > > update
>> > > > > > > > > > > >> > > > > >> > > > > > > their
>> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching
>> to the
>> > > > new
>> > > > > > > > > protocol?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need
>> some
>> > > > > > tooling for
>> > > > > > > > > > > >> non-java
>> > > > > > > > > > > >> > client
>> > > > > > > > > > > >> > > > > >> > users
>> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > >> > > > > >> > > > > sort
>> > > > > > > > > > > >> > > > > >> > > > > > > of
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the
>> client before
>> > > > > > > > > switching to
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > new
>> > > > > > > > > > > >> > > > > >> > protocol.
>> > > > > > > > > > > >> > > > > >> > > I
>> > > > > > > > > > > >> > > > > >> > > > > do not
>> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on
>> top of my
>> > > > head
>> > > > > > > > > though,
>> > > > > > > > > > > >> maybe
>> > > > > > > > > > > >> > others
>> > > > > > > > > > > >> > > > > >> like
>> > > > > > > > > > > >> > > > > >> > > @Matt
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the
>> last
>> > > > > > question,
>> > > > > > > > > will the
>> > > > > > > > > > > >> > Java
>> > > > > > > > > > > >> > > > > >> client
>> > > > > > > > > > > >> > > > > >> > > allow
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
>> current
>> > > > regex
>> > > > > > > > > engine for
>> > > > > > > > > > > >> > > > > >> compatibility
>> > > > > > > > > > > >> > > > > >> > > > > reasons?
>> > > > > > > > > > > >> > > > > >> > > > > > > For
>> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy
>> to keep
>> > > > using
>> > > > > > > > > client
>> > > > > > > > > > > >> based
>> > > > > > > > > > > >> > regex
>> > > > > > > > > > > >> > > > > at
>> > > > > > > > > > > >> > > > > >> > > first to
>> > > > > > > > > > > >> > > > > >> > > > > keep
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple
>> and
>> > > > then
>> > > > > > migrate
>> > > > > > > > > to
>> > > > > > > > > > > >> > server based
>> > > > > > > > > > > >> > > > > >> > > regexes
>> > > > > > > > > > > >> > > > > >> > > > > as a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > second
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not
>> thought
>> > > > > > about
>> > > > > > > > > that for
>> > > > > > > > > > > >> > java
>> > > > > > > > > > > >> > > > > >> clients,
>> > > > > > > > > > > >> > > > > >> > > and
>> > > > > > > > > > > >> > > > > >> > > > > we can
>> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
>> > > > > > compatibility
>> > > > > > > > > issues do
>> > > > > > > > > > > >> > you have
>> > > > > > > > > > > >> > > > > in
>> > > > > > > > > > > >> > > > > >> > > mind?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
>> that the
>> > > > group
>> > > > > > > > > coordinator
>> > > > > > > > > > > >> > will be
>> > > > > > > > > > > >> > > > > >> > > > > responsible for
>> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the
>> configurations and
>> > > > that
>> > > > > > the
>> > > > > > > > > > > >> > configurations
>> > > > > > > > > > > >> > > > > will
>> > > > > > > > > > > >> > > > > >> be
>> > > > > > > > > > > >> > > > > >> > > > > deleted
>> > > > > > > > > > > >> > > > > >> > > > > > > when
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted.
>> Will a
>> > > > > > > > > transition to
>> > > > > > > > > > > >> DEAD
>> > > > > > > > > > > >> > > > > trigger
>> > > > > > > > > > > >> > > > > >> > > deletion
>> > > > > > > > > > > >> > > > > >> > > > > of
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD
>> state
>> > > > is an
>> > > > > > > > > ending
>> > > > > > > > > > > >> state
>> > > > > > > > > > > >> > (we
>> > > > > > > > > > > >> > > > > would
>> > > > > > > > > > > >> > > > > >> > only
>> > > > > > > > > > > >> > > > > >> > > > > > > transit to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the
>> group is
>> > > > EMPTY
>> > > > > > and
>> > > > > > > > > also all
>> > > > > > > > > > > >> of
>> > > > > > > > > > > >> > its
>> > > > > > > > > > > >> > > > > >> > metadata
>> > > > > > > > > > > >> > > > > >> > > are
>> > > > > > > > > > > >> > > > > >> > > > > > > gone),
>> > > > > > > > > > > >> > > > > >> > > > > > > > > once
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
>> DEAD this
>> > > > > > group
>> > > > > > > > > would
>> > > > > > > > > > > >> never
>> > > > > > > > > > > >> > be
>> > > > > > > > > > > >> > > > > >> revived.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the
>> choice to
>> > > > store
>> > > > > > the
>> > > > > > > > > configs
>> > > > > > > > > > > >> in
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > group
>> > > > > > > > > > > >> > > > > >> > > > > coordinator
>> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
>> > > > cluster
>> > > > > > configs
>> > > > > > > > > and
>> > > > > > > > > > > >> their
>> > > > > > > > > > > >> > > > > values?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
>> question,
>> > > > and our
>> > > > > > > > > thoughts
>> > > > > > > > > > > >> are
>> > > > > > > > > > > >> > that
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > so-called
>> > > > > > > > > > > >> > > > > >> > > > > > > "group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
>> > > > overrides
>> > > > > > of the
>> > > > > > > > > > > >> > cluster-level
>> > > > > > > > > > > >> > > > > >> > > > > configurations
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per
>> group so
>> > > > when an
>> > > > > > > > > admin list
>> > > > > > > > > > > >> > cluster
>> > > > > > > > > > > >> > > > > >> > configs
>> > > > > > > > > > > >> > > > > >> > > it's
>> > > > > > > > > > > >> > > > > >> > > > > > > okay to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
>> cluster-level
>> > > > > > > > > "defaults", not
>> > > > > > > > > > > >> > showing
>> > > > > > > > > > > >> > > > > >> any
>> > > > > > > > > > > >> > > > > >> > > > > per-group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
>> someone
>> > > > > > configure a
>> > > > > > > > > group
>> > > > > > > > > > > >> > before
>> > > > > > > > > > > >> > > > > >> starting
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
>> allowing
>> > > > the
>> > > > > > explicit
>> > > > > > > > > > > >> > creation of
>> > > > > > > > > > > >> > > > > >> groups?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
>> > > > decoupled
>> > > > > > from
>> > > > > > > > > the group
>> > > > > > > > > > > >> > > > > lifecycle.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
>> created
>> > > > > > before
>> > > > > > > > > the group
>> > > > > > > > > > > >> > itself
>> > > > > > > > > > > >> > > > > as
>> > > > > > > > > > > >> > > > > >> an
>> > > > > > > > > > > >> > > > > >> > > > > > > independent
>> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
>> > > > requires
>> > > > > > the
>> > > > > > > > > > > >> corresponding
>> > > > > > > > > > > >> > > > > >> request to
>> > > > > > > > > > > >> > > > > >> > > be
>> > > > > > > > > > > >> > > > > >> > > > > > > routed to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
>> based on
>> > > > the
>> > > > > > group
>> > > > > > > > > id ---
>> > > > > > > > > > > >> > the only
>> > > > > > > > > > > >> > > > > >> > thing
>> > > > > > > > > > > >> > > > > >> > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
>> itself is
>> > > > gone
>> > > > > > we
>> > > > > > > > > also check
>> > > > > > > > > > > >> > if there
>> > > > > > > > > > > >> > > > > >> are
>> > > > > > > > > > > >> > > > > >> > > any
>> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to
>> that
>> > > > group
>> > > > > > and
>> > > > > > > > > delete
>> > > > > > > > > > > >> as
>> > > > > > > > > > > >> > well.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
>> indeed
>> > > > > > introduces an
>> > > > > > > > > > > >> > asymmetry on
>> > > > > > > > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > creation
>> > > > > > > > > > > >> > > > > >> > > > > /
>> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
>> config
>> > > > > > entities,
>> > > > > > > > > and we
>> > > > > > > > > > > >> > would like
>> > > > > > > > > > > >> > > > > >> to
>> > > > > > > > > > > >> > > > > >> > > hear
>> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we
>> should
>> > > > aim
>> > > > > > for
>> > > > > > > > > symmetry
>> > > > > > > > > > > >> > i.e.
>> > > > > > > > > > > >> > > > > >> totally
>> > > > > > > > > > > >> > > > > >> > > > > decouple
>> > > > > > > > > > > >> > > > > >> > > > > > > group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence
>> not
>> > > > delete
>> > > > > > them at
>> > > > > > > > > all
>> > > > > > > > > > > >> when
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> group
>> > > > > > > > > > > >> > > > > >> > is
>> > > > > > > > > > > >> > > > > >> > > > > gone,
>> > > > > > > > > > > >> > > > > >> > > > > > > but
>> > > > > > > > > > > >> > > > > >> > > > > > > > > always
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
>> deletion
>> > > > > > operations
>> > > > > > > > > by
>> > > > > > > > > > > >> > themselves.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
>> > > > > > Consumer.subscribe
>> > > > > > > > > method for
>> > > > > > > > > > > >> > the Java
>> > > > > > > > > > > >> > > > > >> > client
>> > > > > > > > > > > >> > > > > >> > > > > still
>> > > > > > > > > > > >> > > > > >> > > > > > > take
>> > > > > > > > > > > >> > > > > >> > > > > > > > > a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > `java.util.regex.Pattern` of
>> > > > > > do we
>> > > > > > > > > have
>> > > > > > > > > > > >> to
>> > > > > > > > > > > >> > > > > >> introduce an
>> > > > > > > > > > > >> > > > > >> > > > > overload?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
>> need to
>> > > > > > introduce
>> > > > > > > > > an
>> > > > > > > > > > > >> > overload, but
>> > > > > > > > > > > >> > > > > >> I'm
>> > > > > > > > > > > >> > > > > >> > > all
>> > > > > > > > > > > >> > > > > >> > > > > ears
>> > > > > > > > > > > >> > > > > >> > > > > > > if
>> > > > > > > > > > > >> > > > > >> > > > > > > > > there
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
>> compatibility
>> > > > > > issues
>> > > > > > > > > that we
>> > > > > > > > > > > >> may
>> > > > > > > > > > > >> > > > > overlook.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
>> Justine
>> > > > > > that we
>> > > > > > > > > should
>> > > > > > > > > > > >> be
>> > > > > > > > > > > >> > clearer
>> > > > > > > > > > > >> > > > > >> > about
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > reason
>> > > > > > > > > > > >> > > > > >> > > > > > > > > to
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
>> > > > > > IBP/metadata.version
>> > > > > > > > > from the
>> > > > > > > > > > > >> > feature
>> > > > > > > > > > > >> > > > > >> flag.
>> > > > > > > > > > > >> > > > > >> > > Maybe
>> > > > > > > > > > > >> > > > > >> > > > > we
>> > > > > > > > > > > >> > > > > >> > > > > > > mean
>> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
>> default
>> > > > for
>> > > > > > the
>> > > > > > > > > feature
>> > > > > > > > > > > >> flag
>> > > > > > > > > > > >> > to
>> > > > > > > > > > > >> > > > > true
>> > > > > > > > > > > >> > > > > >> > > based on
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
>> once we
>> > > > > > want to
>> > > > > > > > > make it
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > >> default.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
>> protocol
>> > > > APIs
>> > > > > > don't
>> > > > > > > > > mention
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > >> required
>> > > > > > > > > > > >> > > > > >> > > ACLs,
>> > > > > > > > > > > >> > > > > >> > > > > it
>> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that
>> for
>> > > > > > consistency.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
>> certainly do
>> > > > that.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
>> odd that
>> > > > > > > > > > > >> > ConsumerGroupHeartbeat
>> > > > > > > > > > > >> > > > > >> > requires
>> > > > > > > > > > > >> > > > > >> > > > > "Read
>> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems
>> to do
>> > > > more
>> > > > > > than
>> > > > > > > > > reading.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
>> myself
>> > > > as
>> > > > > > well,
>> > > > > > > > > but in
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > end we
>> > > > > > > > > > > >> > > > > >> > could
>> > > > > > > > > > > >> > > > > >> > > not
>> > > > > > > > > > > >> > > > > >> > > > > > > find a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative:
>> adding
>> > > > > > Write
>> > > > > > > > > Group
>> > > > > > > > > > > >> seems
>> > > > > > > > > > > >> > an
>> > > > > > > > > > > >> > > > > >> overkill
>> > > > > > > > > > > >> > > > > >> > > here
>> > > > > > > > > > > >> > > > > >> > > > > > > since we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > do
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
>> elsewhere (we
>> > > > only
>> > > > > > have
>> > > > > > > > > Read /
>> > > > > > > > > > > >> > Delete
>> > > > > > > > > > > >> > > > > and
>> > > > > > > > > > > >> > > > > >> > > Describe
>> > > > > > > > > > > >> > > > > >> > > > > on
>> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to
>> hear
>> > > > others
>> > > > > > > > > thoughts.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
>> > > > recreation
>> > > > > > > > > handled by
>> > > > > > > > > > > >> the
>> > > > > > > > > > > >> > > > > consumer
>> > > > > > > > > > > >> > > > > >> > with
>> > > > > > > > > > > >> > > > > >> > > the
>> > > > > > > > > > > >> > > > > >> > > > > new
>> > > > > > > > > > > >> > > > > >> > > > > > > > > group
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It
>> would be
>> > > > good
>> > > > > > to have
>> > > > > > > > > a
>> > > > > > > > > > > >> > section on
>> > > > > > > > > > > >> > > > > >> this.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
>> > > > > > subscription
>> > > > > > > > > right? Yes
>> > > > > > > > > > > >> > we can
>> > > > > > > > > > > >> > > > > >> add a
>> > > > > > > > > > > >> > > > > >> > > > > section
>> > > > > > > > > > > >> > > > > >> > > > > > > about
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically
>> the
>> > > > idea
>> > > > > > is that
>> > > > > > > > > > > >> consumer
>> > > > > > > > > > > >> > would
>> > > > > > > > > > > >> > > > > be
>> > > > > > > > > > > >> > > > > >> > > totally
>> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as
>> it's
>> > > > > > handled all
>> > > > > > > > > by the
>> > > > > > > > > > > >> > brokers.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
>> mentions we
>> > > > will
>> > > > > > write
>> > > > > > > > > the
>> > > > > > > > > > > >> new
>> > > > > > > > > > > >> > > > > >> coordinator
>> > > > > > > > > > > >> > > > > >> > in
>> > > > > > > > > > > >> > > > > >> > > > > Java.
>> > > > > > > > > > > >> > > > > >> > > > > > > Even
>> > > > > > > > > > > >> > > > > >> > > > > > > > > though
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
>> implementation
>> > > > > > detail,
>> > > > > > > > > do we
>> > > > > > > > > > > >> > plan to
>> > > > > > > > > > > >> > > > > >> have a
>> > > > > > > > > > > >> > > > > >> > > new
>> > > > > > > > > > > >> > > > > >> > > > > gradle
>> > > > > > > > > > > >> > > > > >> > > > > > > > > module
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought
>> about
>> > > > > > that. But
>> > > > > > > > > I think
>> > > > > > > > > > > >> > the
>> > > > > > > > > > > >> > > > > answer
>> > > > > > > > > > > >> > > > > >> > > should
>> > > > > > > > > > > >> > > > > >> > > > > be
>> > > > > > > > > > > >> > > > > >> > > > > > > yes.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
>> > > > scalability
>> > > > > > goal
>> > > > > > > > > when it
>> > > > > > > > > > > >> > comes to
>> > > > > > > > > > > >> > > > > >> how
>> > > > > > > > > > > >> > > > > >> > > many
>> > > > > > > > > > > >> > > > > >> > > > > > > members
>> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
>> > > > support?
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
>> think we
>> > > > > > should
>> > > > > > > > > shoot for
>> > > > > > > > > > > >> > 1000s of
>> > > > > > > > > > > >> > > > > >> > > members.
>> > > > > > > > > > > >> > > > > >> > > > > But
>> > > > > > > > > > > >> > > > > >> > > > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
>> also
>> > > > depend
>> > > > > > on the
>> > > > > > > > > offset
>> > > > > > > > > > > >> > > > > management
>> > > > > > > > > > > >> > > > > >> > > (commit,
>> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
>> > > > coordinator
>> > > > > > > > > which we
>> > > > > > > > > > > >> did
>> > > > > > > > > > > >> > not
>> > > > > > > > > > > >> > > > > >> cover in
>> > > > > > > > > > > >> > > > > >> > > this
>> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
>> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
>> number that
>> > > > > > applies
>> > > > > > > > > > > >> > universally.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we
>> consider
>> > > > having
>> > > > > > > > > > > >> > SubscribedTopidIds
>> > > > > > > > > > > >> > > > > >> instead
>> > > > > > > > > > > >> > > > > >> > of
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> SubscribedTopicNames in
>> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
>> > > > > > > > > > > >> > > > > >> > > Is the
>> > > > > > > > > > > >> > > > > >> > > > > > > idea
>> > > > > > > > > > > >> > > > > >> > > > > > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
>> resolve
>> > > > the
>> > > > > > regex
>> > > > > > > > > on the
>> > > > > > > > > > > >> > server,
>> > > > > > > > > > > >> > > > > we
>> > > > > > > > > > > >> > > > > >> > can
>> > > > > > > > > > > >> > > > > >> > > do
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > same
>> > > > > > > > > > > >> > > > > >> > > > > > > > > for
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name?
>> The
>> > > > > > difference is
>> > > > > > > > > that
>> > > > > > > > > > > >> > sending the
>> > > > > > > > > > > >> > > > > >> > regex
>> > > > > > > > > > > >> > > > > >> > > is
>> > > > > > > > > > > >> > > > > >> > > > > more
>> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
>> the topic
>> > > > > > names is
>> > > > > > > > > less
>> > > > > > > > > > > >> > efficient.
>> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
>> easier to
>> > > > > > handle if
>> > > > > > > > > we have
>> > > > > > > > > > > >> > topic
>> > > > > > > > > > > >> > > > > ids.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
>> still
>> > > > let
>> > > > > > the
>> > > > > > > > > clients
>> > > > > > > > > > > >> send
>> > > > > > > > > > > >> > names
>> > > > > > > > > > > >> > > > > >> is to
>> > > > > > > > > > > >> > > > > >> > > keep
>> > > > > > > > > > > >> > > > > >> > > > > the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names
>> -> ids
>> > > > on
>> > > > > > the
>> > > > > > > > > broker /
>> > > > > > > > > > > >> > admin
>> > > > > > > > > > > >> > > > > client
>> > > > > > > > > > > >> > > > > >> > > only.
>> > > > > > > > > > > >> > > > > >> > > > > Note
>> > > > > > > > > > > >> > > > > >> > > > > > > that
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
>> topic id
>> > > > in
>> > > > > > > > > KIP-516, we
>> > > > > > > > > > > >> > never
>> > > > > > > > > > > >> > > > > >> > > implemented the
>> > > > > > > > > > > >> > > > > >> > > > > > > logic
>> > > > > > > > > > > >> > > > > >> > > > > > > > > on
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
>> > > > leveraging
>> > > > > > the
>> > > > > > > > > related
>> > > > > > > > > > > >> > newer
>> > > > > > > > > > > >> > > > > >> versioned
>> > > > > > > > > > > >> > > > > >> > > RPCs,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic
>> id as
>> > > > empty
>> > > > > > UUID.
>> > > > > > > > > We
>> > > > > > > > > > > >> want
>> > > > > > > > > > > >> > to keep
>> > > > > > > > > > > >> > > > > >> the
>> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
>> > > > delegate
>> > > > > > the
>> > > > > > > > > reasoning
>> > > > > > > > > > > >> on
>> > > > > > > > > > > >> > broker
>> > > > > > > > > > > >> > > > > >> and
>> > > > > > > > > > > >> > > > > >> > > > > > > potentially
>> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
>> 2022 at
>> > > > 10:45
>> > > > > > AM
>> > > > > > > > > David
>> > > > > > > > > > > >> Jacot
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> <dj...@confluent.io.invalid>
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to
>> start a
>> > > > > > > > > discussion
>> > > > > > > > > > > >> thread
>> > > > > > > > > > > >> > on
>> > > > > > > > > > > >> > > > > >> KIP-848:
>> > > > > > > > > > > >> > > > > >> > > The
>> > > > > > > > > > > >> > > > > >> > > > > Next
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of
>> the
>> > > > Consumer
>> > > > > > > > > Rebalance
>> > > > > > > > > > > >> > Protocol.
>> > > > > > > > > > > >> > > > > >> With
>> > > > > > > > > > > >> > > > > >> > > this
>> > > > > > > > > > > >> > > > > >> > > > > KIP,
>> > > > > > > > > > > >> > > > > >> > > > > > > we
>> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
>> rebalance
>> > > > > > protocol
>> > > > > > > > > (for
>> > > > > > > > > > > >> > consumers)
>> > > > > > > > > > > >> > > > > >> more
>> > > > > > > > > > > >> > > > > >> > > > > reliable,
>> > > > > > > > > > > >> > > > > >> > > > > > > more
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable,
>> easier to
>> > > > > > implement for
>> > > > > > > > > > > >> > clients, and
>> > > > > > > > > > > >> > > > > >> easier
>> > > > > > > > > > > >> > > > > >> > > to
>> > > > > > > > > > > >> > > > > >> > > > > debug
>> > > > > > > > > > > >> > > > > >> > > > > > > for
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
>> > > > > > > > > > > >> > > > > >> > > > >
>> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
>> look and
>> > > > let
>> > > > > > me
>> > > > > > > > > know what
>> > > > > > > > > > > >> > you
>> > > > > > > > > > > >> > > > > think.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
>> away from
>> > > > > > July
>> > > > > > > > > 18th to
>> > > > > > > > > > > >> > August 8th.
>> > > > > > > > > > > >> > > > > >> That
>> > > > > > > > > > > >> > > > > >> > > gives
>> > > > > > > > > > > >> > > > > >> > > > > > > you a
>> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read
>> and
>> > > > digest
>> > > > > > this
>> > > > > > > > > long
>> > > > > > > > > > > >> KIP.
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
>> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
>> > > > > > > > > > > >> > > > > >> > > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > > > > > --
>> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
>> > > > > > > > > > > >> > > > > >> > > > > > >
>> > > > > > > > > > > >> > > > > >> > > > >
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> > >
>> > > > > > > > > > > >> > > > > >> >
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > > >> --
>> > > > > > > > > > > >> > > > > >> -- Guozhang
>> > > > > > > > > > > >> > > > > >>
>> > > > > > > > > > > >> > > > > >
>> > > > > > > > > > > >> > > > >
>> > > > > > > > > > > >> >
>> > > > > > > > > > > >>
>> > > > > > > > > > > >
>> > > > > > > > >
>> > > > > >
>> > > > >
>> > > > >
>> > > > > --
>> > > > > -- Guozhang
>> > > >
>> > >
>> > >
>> > > --
>> > > -- Guozhang
>>
>
>
> --
> -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hello David,

Thanks for bringing this question up. I think the main benefits as you
listed is 2) above if it stays; just to clarify, we would only be able to
save one round trip if the rebalance is still triggered by the broker; if
the rebalance is triggered by the client then the num.round.trips are the
same:

1) With GroupPrepareAssignment:

T0: client decides to do a new assignment, suppose it has already sent a HB
and hence has to wait for it to return first since only one request /
response can be inflight with the coordinator's socket.
T1: client receives the HB response, and then sends the
GroupPrepareAssignment request.
T2: the GroupPrepareAssignment response is returned.
T3: it calculates the new assignment, and sends a GroupInstallAssignment
request.

In total, two round trips.

2) Without GroupPrepareAssignment:

T0: client decides to do a new assignment, suppose it has already sent a HB
and hence has to wait for it to return first since only one request /
response can be inflight with the coordinator's socket.
T1: client receives the HB response, and then sends the new HB request with
the flag indicating a new rebalance needed..
T2: the HB response with the optional member metadata map is returned.
T3: it calculates the new assignment, and sends a GroupInstallAssignment
request.

In total, two round trips as well.

-----------------------------

So to complete the full picture here, we'd need to modify both HB request
and response so that the client can also indicate a new rebalance via the
HB request as well, right?

Assuming all above is true, I think it's okay to merge the
GroupPrepareAssignment into HB given that we can make the additional fields
encoding the full member (subscription) metadata and topic metadata as
optional fields.


Guozhang










On Mon, Sep 12, 2022 at 5:22 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi all,
>
> During an offline conversation, someone asked why we need the
> ConsumerGroupPrepareAssignment API and suggested that we could instead
> provide the group state in the heartbeat response. This has a few
> advantages: 1) it does not require using a special error code to
> signal that a new assignment is required as the signal would be the
> provided group state; 2) it removes one round trip when a client side
> assignor is used. The downside is that it makes the heartbeat
> response's definition quite large. I recall that I went with the
> current approach due to this.
>
> Providing the group state in the heartbeat response is appealing. What
> do you guys think?
>
> Best,
> David
>
> On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io> wrote:
> >
> > Hi Guozhang,
> >
> > 1. I have added a reference to the relevant chapter instead of
> > repeating the whole thing. Does that work for you?
> >
> > 2. The "Rebalance Triggers" section you are referring to is about when
> > a rebalance should be triggered for the non-upgraded members using the
> > old protocol. The section mentions that a rebalance must be triggered
> > when a new assignment is installed. This implies that the group epoch
> > was updated either by a native member or a non-upgraded member. For
> > the latter, the JoinGroup request would be the trigger. I have added a
> > reference to the relevant chapter in the "JoinGroup Handling" section
> > as well. Does that make sense?
> >
> > Thanks,
> > David
> >
> > On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com>
> wrote:
> > >
> > > Hello David,
> > >
> > > Alright I think that's sufficient. Just to make that clear in the doc,
> > > could we update:
> > >
> > > 1) the heartbeat request handling section, stating when coordinator
> will
> > > trigger rebalance based on the HB's member metadata / reason?
> > > 2) the "Rebalance Triggers" section to include what we described in
> "Group
> > > Epoch - Trigger a rebalance" section as well?
> > >
> > >
> > > Guozhang
> > >
> > > On Fri, Sep 9, 2022 at 1:28 AM David Jacot <djacot@confluent.io.invalid
> >
> > > wrote:
> > >
> > > > Hi Guozhang,
> > > >
> > > > I thought that the assignor will always be consulted when the next
> > > > heartbeat request is constructed. In other words,
> > > > `PartitionAssignor#metadata` will be called for every heartbeat. This
> > > > gives the opportunity for the assignor to enforce a rebalance by
> > > > setting the reason to a non-zero value or by changing the bytes. Do
> > > > you think that this is not sufficient? Are you concerned by the
> delay?
> > > >
> > > > Best,
> > > > David
> > > >
> > > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > > >
> > > > > Hello David,
> > > > >
> > > > > One of Jun's comments make me thinking:
> > > > >
> > > > > ```
> > > > > In this case, a new assignment is triggered by the client side
> > > > > assignor. When constructing the HB, the consumer will always
> consult
> > > > > the client side assignor and propagate the information to the group
> > > > > coordinator. In other words, we don't expect users to call
> > > > > Consumer#enforceRebalance anymore.
> > > > > ```
> > > > >
> > > > > As I looked at the current PartitionAssignor's interface, we
> actually do
> > > > > not have a way yet to instruct how to construct the next HB
> request, e.g.
> > > > > when the assignor wants to enforce a new rebalance with a new
> assignment,
> > > > > we'd need some customizable APIs inside the PartitionAssignor to
> indicate
> > > > > the next HB telling broker about so. WDYT about adding such an API
> on the
> > > > > PartitionAssignor?
> > > > >
> > > > >
> > > > > Guozhang
> > > > >
> > > > >
> > > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> <dj...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > > > Hi Jun,
> > > > > >
> > > > > > I have updated the KIP to include your feedback. I have also
> tried to
> > > > > > clarify the parts which were not cleared.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io>
> > > > wrote:
> > > > > > >
> > > > > > > Hi Jun,
> > > > > > >
> > > > > > > Thanks for your feedback. Let me start by answering your
> questions
> > > > > > > inline and I will update the KIP next week.
> > > > > > >
> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> seem to
> > > > be
> > > > > > fewer
> > > > > > > > RPCs during rebalance and more efficient support of
> wildcard. A few
> > > > > > > > comments below.
> > > > > > >
> > > > > > > I would also add that the KIP removes the global sync barrier
> in the
> > > > > > > protocol which is essential to improve group stability and
> > > > > > > scalability, and the KIP also simplifies the client by moving
> most of
> > > > > > > the logic to the server side.
> > > > > > >
> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> rolling
> > > > > > changing
> > > > > > > > of the partition assignor in the consumers?
> > > > > > >
> > > > > > > Definitely. The group coordinator will use the assignor used
> by a
> > > > > > > majority of the members. This allows the group to move from one
> > > > > > > assignor to another by a roll. This is explained in the
> Assignor
> > > > > > > Selection chapter.
> > > > > > >
> > > > > > > > 30.2 For each field, could you explain whether it's required
> in
> > > > every
> > > > > > > > request or the scenarios when it needs to be filled? For
> example,
> > > > it's
> > > > > > not
> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > > >
> > > > > > > The client is expected to set those fields in case of a
> connection
> > > > > > > issue (e.g. timeout) or when the fields have changed since the
> last
> > > > > > > HB. The server populates those fields as long as the member is
> not
> > > > > > > fully reconciled - the member should acknowledge that it has
> the
> > > > > > > expected epoch and assignment. I will clarify this in the KIP.
> > > > > > >
> > > > > > > > 31. In the current consumer protocol, the rack affinity
> between the
> > > > > > client
> > > > > > > > and the broker is only considered during fetching, but not
> during
> > > > > > assigning
> > > > > > > > partitions to consumers. Sometimes, once the assignment is
> made,
> > > > there
> > > > > > is
> > > > > > > > no opportunity for read affinity because no replicas of
> assigned
> > > > > > partitions
> > > > > > > > are close to the member. I am wondering if we should use this
> > > > > > opportunity
> > > > > > > > to address this by including rack in GroupMember.
> > > > > > >
> > > > > > > That's an interesting idea. I don't see any issue with adding
> the
> > > > rack
> > > > > > > to the members. I will do so.
> > > > > > >
> > > > > > > > 32. On the metric side, often, it's useful to know how busy
> a group
> > > > > > > > coordinator is. By moving the event loop model, it seems
> that we
> > > > could
> > > > > > add
> > > > > > > > a metric that tracks the fraction of the time the event loop
> is
> > > > doing
> > > > > > the
> > > > > > > > actual work.
> > > > > > >
> > > > > > > That's a great idea. I will add it. Thanks.
> > > > > > >
> > > > > > > > 33. Could we add a section on coordinator failover handling?
> For
> > > > > > example,
> > > > > > > > does it need to trigger the check if any group with the
> wildcard
> > > > > > > > subscription now has a new matching topic?
> > > > > > >
> > > > > > > Sure. When the new group coordinator takes over, it has to:
> > > > > > > * Setup the session timeouts.
> > > > > > > * Trigger a new assignment if a client side assignor is used.
> We
> > > > don't
> > > > > > > store the information about the member selected to run the
> assignment
> > > > > > > so we have to start a new one.
> > > > > > > * Update the topics metadata, verify the wildcard
> subscriptions, and
> > > > > > > trigger a rebalance if needed.
> > > > > > >
> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> epoch
> > > > > > field
> > > > > > > > reflects? For example, does the epoch in
> ConsumerGroupMetadataValue
> > > > > > reflect
> > > > > > > > the latest group epoch? What about the one in
> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >
> > > > > > > Sure. I will clarify that but it is always the latest group
> epoch.
> > > > > > > When the group state is updated, the group epoch is bumped so
> we use
> > > > > > > that one for all the change records related to the update.
> > > > > > >
> > > > > > > > 35. "the group coordinator will ensure that the following
> > > > invariants
> > > > > > are
> > > > > > > > met: ... All members exists." It's possible for a member not
> to
> > > > get any
> > > > > > > > assigned partitions, right?
> > > > > > >
> > > > > > > That's right. Here I meant that the members provided by the
> assignor
> > > > > > > in the assignment must exist in the group. The assignor can
> not make
> > > > > > > up new member ids.
> > > > > > >
> > > > > > > > 36. "He can rejoins the group with a member epoch equals to
> 0":
> > > > When
> > > > > > would
> > > > > > > > a consumer rejoin and what member id would be used?
> > > > > > >
> > > > > > > A member is expected to abandon all its partitions and rejoins
> when
> > > > it
> > > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > > > > > coordinator will have removed the member from the group. The
> member
> > > > > > > can rejoin the group with the same member id but with 0 as
> epoch. Let
> > > > > > > me see if I can clarify this in the KIP.
> > > > > > >
> > > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > > reassignment
> > > > > > > > by either providing a non-zero reason or by updating the
> assignor
> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> deprecation
> > > > of
> > > > > > > > Consumer#enforeRebalance.
> > > > > > >
> > > > > > > In this case, a new assignment is triggered by the client side
> > > > > > > assignor. When constructing the HB, the consumer will always
> consult
> > > > > > > the client side assignor and propagate the information to the
> group
> > > > > > > coordinator. In other words, we don't expect users to call
> > > > > > > Consumer#enforceRebalance anymore.
> > > > > > >
> > > > > > > > 38. The reassignment examples are nice. But the section
> seems to
> > > > have
> > > > > > > > multiple typos.
> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> gets into
> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> into
> > > > > > epoch=3,
> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > > partitions=[foo-0].
> > > > > > >
> > > > > > > Sorry for that! I will revise them.
> > > > > > >
> > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> from
> > > > any
> > > > > > old
> > > > > > > > version to new one?
> > > > > > >
> > > > > > > We will support upgrading from the consumer protocol version 3,
> > > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
> earliest
> > > > > > > version is unknown at the moment. This is explained in the
> migration
> > > > > > > plan chapter.
> > > > > > >
> > > > > > > Thanks again for your feedback, Jun. I will update the KIP
> based on
> > > > it
> > > > > > > next week.
> > > > > > >
> > > > > > > Best,
> > > > > > > David
> > > > > > >
> > > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao
> <ju...@confluent.io.invalid>
> > > > wrote:
> > > > > > > >
> > > > > > > > Hi, David,
> > > > > > > >
> > > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP
> seem to
> > > > be
> > > > > > fewer
> > > > > > > > RPCs during rebalance and more efficient support of
> wildcard. A few
> > > > > > > > comments below.
> > > > > > > >
> > > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> rolling
> > > > > > changing
> > > > > > > > of the partition assignor in the consumers?
> > > > > > > > 30.2 For each field, could you explain whether it's required
> in
> > > > every
> > > > > > > > request or the scenarios when it needs to be filled? For
> example,
> > > > it's
> > > > > > not
> > > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > > > >
> > > > > > > > 31. In the current consumer protocol, the rack affinity
> between the
> > > > > > client
> > > > > > > > and the broker is only considered during fetching, but not
> during
> > > > > > assigning
> > > > > > > > partitions to consumers. Sometimes, once the assignment is
> made,
> > > > there
> > > > > > is
> > > > > > > > no opportunity for read affinity because no replicas of
> assigned
> > > > > > partitions
> > > > > > > > are close to the member. I am wondering if we should use this
> > > > > > opportunity
> > > > > > > > to address this by including rack in GroupMember.
> > > > > > > >
> > > > > > > > 32. On the metric side, often, it's useful to know how busy
> a group
> > > > > > > > coordinator is. By moving the event loop model, it seems
> that we
> > > > could
> > > > > > add
> > > > > > > > a metric that tracks the fraction of the time the event loop
> is
> > > > doing
> > > > > > the
> > > > > > > > actual work.
> > > > > > > >
> > > > > > > > 33. Could we add a section on coordinator failover handling?
> For
> > > > > > example,
> > > > > > > > does it need to trigger the check if any group with the
> wildcard
> > > > > > > > subscription now has a new matching topic?
> > > > > > > >
> > > > > > > > 34. ConsumerGroupMetadataValue,
> > > > ConsumerGroupPartitionMetadataValue,
> > > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> epoch
> > > > > > field
> > > > > > > > reflects? For example, does the epoch in
> ConsumerGroupMetadataValue
> > > > > > reflect
> > > > > > > > the latest group epoch? What about the one in
> > > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > > ConsumerGroupMemberMetadataValue?
> > > > > > > >
> > > > > > > > 35. "the group coordinator will ensure that the following
> > > > invariants
> > > > > > are
> > > > > > > > met: ... All members exists." It's possible for a member not
> to
> > > > get any
> > > > > > > > assigned partitions, right?
> > > > > > > >
> > > > > > > > 36. "He can rejoins the group with a member epoch equals to
> 0":
> > > > When
> > > > > > would
> > > > > > > > a consumer rejoin and what member id would be used?
> > > > > > > >
> > > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > > reassignment
> > > > > > > > by either providing a non-zero reason or by updating the
> assignor
> > > > > > > > metadata." Hmm, this seems to be conflicting with the
> deprecation
> > > > of
> > > > > > > > Consumer#enforeRebalance.
> > > > > > > >
> > > > > > > > 38. The reassignment examples are nice. But the section
> seems to
> > > > have
> > > > > > > > multiple typos.
> > > > > > > > 38.1 When the group transitions to epoch 2, B immediately
> gets into
> > > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > > 38.2 When the group transitions to epoch 3, C seems to get
> into
> > > > > > epoch=3,
> > > > > > > > partitions=[foo-1] too early.
> > > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > > partitions=[foo-0].
> > > > > > > >
> > > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> from
> > > > any
> > > > > > old
> > > > > > > > version to new one?
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > >
> > > > > > > > Jun
> > > > > > > >
> > > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi all,
> > > > > > > > >
> > > > > > > > > The KIP states that we will re-implement the coordinator in
> > > > Java. I
> > > > > > > > > discussed this offline with a few folks and folks are
> concerned
> > > > that
> > > > > > > > > we could introduce many regressions in the old protocol if
> we do
> > > > so.
> > > > > > > > > Therefore, I am going to remove this statement from the
> KIP. It
> > > > is an
> > > > > > > > > implementation detail after all so it does not have to be
> > > > decided at
> > > > > > > > > this stage. We will likely start by trying to refactor the
> > > > current
> > > > > > > > > implementation as a first step.
> > > > > > > > >
> > > > > > > > > Cheers,
> > > > > > > > > David
> > > > > > > > >
> > > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> djacot@confluent.io
> > > > >
> > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Hi Luke,
> > > > > > > > > >
> > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > > reconciling,
> > > > > > > > > stable,
> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> right?
> > > > > > > > > >
> > > > > > > > > > This sentence does not refer to those group states but
> rather
> > > > to a
> > > > > > > > > > state machine replication (SMR). This refers to the
> entire
> > > > state of
> > > > > > > > > > group coordinator which is replicated via the log layer.
> I will
> > > > > > > > > > clarify this in the KIP.
> > > > > > > > > >
> > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
> as an
> > > > event
> > > > > > > > > loop"?
> > > > > > > > > >
> > > > > > > > > > The idea is to follow a model similar to the new quorum
> > > > > > controller. We
> > > > > > > > > > will have N threads to process events. Each
> __consumer_offsets
> > > > > > > > > > partition is assigned to a unique thread and all the
> events
> > > > (e.g.
> > > > > > > > > > requests, callbacks, etc.) are processed by this thread.
> This
> > > > > > simplify
> > > > > > > > > > concurrency and will enable us to do simulation testing
> for the
> > > > > > group
> > > > > > > > > > coordinator.
> > > > > > > > > >
> > > > > > > > > > > 1.3. Why do we need a state machine per
> *__consumer_offsets*
> > > > > > > > > partitions?
> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> group
> > > > > > coordinator?
> > > > > > > > > For
> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> groups, and
> > > > > > both
> > > > > > > > > exist in
> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> for it,
> > > > or
> > > > > > 2?
> > > > > > > > > >
> > > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > > > > > > > >
> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> number
> > > > of
> > > > > > threads
> > > > > > > > > used
> > > > > > > > > > > to run the state machines. But I'm wondering if the
> purpose
> > > > of
> > > > > > the
> > > > > > > > > threads
> > > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > partitions?), and no heavy computation, why should we
> need
> > > > > > > > > multi-threads
> > > > > > > > > > > here?
> > > > > > > > > >
> > > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > > > processing as
> > > > > > the
> > > > > > > > > > computation could be heavy.
> > > > > > > > > >
> > > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > > session
> > > > > > > > > timeout not
> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> min/max
> > > > > > session
> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > > >
> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout
> to
> > > > detect
> > > > > > client
> > > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> minimum
> > > > > > session
> > > > > > > > > timeout.
> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> maximum
> > > > > > session
> > > > > > > > > timeout.
> > > > > > > > > >
> > > > > > > > > > This is indeed a mistake. The default session timeout
> should
> > > > be 45s
> > > > > > > > > > (the current default).
> > > > > > > > > >
> > > > > > > > > > > 2.2. The default server side assignor are [range,
> uniform],
> > > > > > which means
> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> why not
> > > > > > uniform
> > > > > > > > > one? I
> > > > > > > > > > > thought usually users will choose uniform assignor
> (former
> > > > sticky
> > > > > > > > > assinor)
> > > > > > > > > > > for better evenly distribution. Any other reason we
> choose
> > > > range
> > > > > > > > > assignor
> > > > > > > > > > > as default?
> > > > > > > > > > > group.consumer.assignors List range, uniform The
> server side
> > > > > > assignors.
> > > > > > > > > >
> > > > > > > > > > The order on the server side has no influence because the
> > > > client
> > > > > > must
> > > > > > > > > > chose the selector that he wants to use. There is no
> default
> > > > in the
> > > > > > > > > > current proposal. If the assignor is not specified by the
> > > > client,
> > > > > > the
> > > > > > > > > > request is rejected. The default client value for
> > > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > > > >
> > > > > > > > > > Thanks for your very good comments, Luke. I hope that my
> > > > answers
> > > > > > help
> > > > > > > > > > to clarify things. I will update the KIP as well based
> on your
> > > > > > > > > > feedback.
> > > > > > > > > >
> > > > > > > > > > Cheers,
> > > > > > > > > > David
> > > > > > > > > >
> > > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> showuon@gmail.com>
> > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > Hi David,
> > > > > > > > > > >
> > > > > > > > > > > Thanks for the update.
> > > > > > > > > > >
> > > > > > > > > > > Some more questions:
> > > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > > > > > The new group coordinator will have a state machine
> per
> > > > > > > > > > > *__consumer_offsets* partitions, where each state
> machine is
> > > > > > modelled
> > > > > > > > > as an
> > > > > > > > > > > event loop. Those state machines will be executed in
> > > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > > >
> > > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > > reconciling,
> > > > > > > > > stable,
> > > > > > > > > > > dead" mentioned in Consumer Group States section,
> right?
> > > > > > > > > > > 1.2. What do you mean "each state machine is modelled
> as an
> > > > event
> > > > > > > > > loop"?
> > > > > > > > > > > 1.3. Why do we need a state machine per
> *__consumer_offsets*
> > > > > > > > > partitions?
> > > > > > > > > > > Not a state machine "per consumer group" owned by a
> group
> > > > > > coordinator?
> > > > > > > > > For
> > > > > > > > > > > example, if one group coordinator owns 2 consumer
> groups, and
> > > > > > both
> > > > > > > > > exist in
> > > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine
> for it,
> > > > or
> > > > > > 2?
> > > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> number
> > > > of
> > > > > > threads
> > > > > > > > > used
> > > > > > > > > > > to run the state machines. But I'm wondering if the
> purpose
> > > > of
> > > > > > the
> > > > > > > > > threads
> > > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > > *__consumer_offsets*
> > > > > > > > > > > partitions?), and no heavy computation, why should we
> need
> > > > > > > > > multi-threads
> > > > > > > > > > > here?
> > > > > > > > > > >
> > > > > > > > > > > 2. For the default value in the new configs:
> > > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > > session
> > > > > > > > > timeout not
> > > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> min/max
> > > > > > session
> > > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > > >
> > > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout
> to
> > > > detect
> > > > > > client
> > > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> minimum
> > > > > > session
> > > > > > > > > timeout.
> > > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> maximum
> > > > > > session
> > > > > > > > > timeout.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > 2.2. The default server side assignor are [range,
> uniform],
> > > > > > which means
> > > > > > > > > > > we'll default to "range" assignor. I'd like to know
> why not
> > > > > > uniform
> > > > > > > > > one? I
> > > > > > > > > > > thought usually users will choose uniform assignor
> (former
> > > > sticky
> > > > > > > > > assinor)
> > > > > > > > > > > for better evenly distribution. Any other reason we
> choose
> > > > range
> > > > > > > > > assignor
> > > > > > > > > > > as default?
> > > > > > > > > > > group.consumer.assignors List range, uniform The
> server side
> > > > > > assignors.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Thank you.
> > > > > > > > > > > Luke
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> showuon@gmail.com
> > > > >
> > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > > Hi Sagar,
> > > > > > > > > > > >
> > > > > > > > > > > > I have some thoughts about Kafka Connect integrating
> with
> > > > > > KIP-848,
> > > > > > > > > but I
> > > > > > > > > > > > think we should have a separate discussion thread
> for the
> > > > Kafka
> > > > > > > > > Connect
> > > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> Rebalance
> > > > > > Protocol
> > > > > > > > > [1],
> > > > > > > > > > > > and let this discussion thread focus on consumer
> rebalance
> > > > > > protocol,
> > > > > > > > > WDYT?
> > > > > > > > > > > >
> > > > > > > > > > > > [1]
> > > > > > > > > > > >
> > > > > > > > >
> > > > > >
> > > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > >
> > > > > > > > > > > > Thank you.
> > > > > > > > > > > > Luke
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > wrote:
> > > > > > > > > > > >
> > > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks
> like
> > > > there's
> > > > > > > > > agreement on
> > > > > > > > > > > >> using separate APIs for Connect. I would revisit
> the doc
> > > > and
> > > > > > see
> > > > > > > > > what
> > > > > > > > > > > >> changes are to be made.
> > > > > > > > > > > >>
> > > > > > > > > > > >> Thanks!
> > > > > > > > > > > >> Sagar.
> > > > > > > > > > > >>
> > > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > > >> wrote:
> > > > > > > > > > > >>
> > > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > Thanks for the feedback and the document. That's
> really
> > > > > > helpful. I
> > > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > Overall, it seems to me that both Connect and the
> > > > Consumer
> > > > > > could
> > > > > > > > > share
> > > > > > > > > > > >> > the same underlying "engine". The main difference
> is
> > > > that
> > > > > > the
> > > > > > > > > Consumer
> > > > > > > > > > > >> > assigns topic-partitions to members whereas
> Connect
> > > > assigns
> > > > > > tasks
> > > > > > > > > to
> > > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> different
> > > > > > resource
> > > > > > > > > types
> > > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> dedicated
> > > > APIs
> > > > > > > > > would be
> > > > > > > > > > > >> > similar to the new ones but different on the
> > > > > > content/resources and
> > > > > > > > > > > >> > they would rely on the same engine on the
> coordinator
> > > > side.
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > I personally lean towards 2) because I am not a
> fan of
> > > > > > > > > overcharging
> > > > > > > > > > > >> > APIs to serve different purposes. That being
> said, I am
> > > > not
> > > > > > > > > opposed to
> > > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > I think that we can continue to discuss it here
> for now
> > > > in
> > > > > > order
> > > > > > > > > to
> > > > > > > > > > > >> > ensure that this KIP is compatible with what we
> will do
> > > > for
> > > > > > > > > Connect in
> > > > > > > > > > > >> > the future.
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > Best,
> > > > > > > > > > > >> > David
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > > > > djacot@confluent.io>
> > > > > > > > > wrote:
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Hi all,
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > I am back from vacation. I will go through and
> address
> > > > > > your
> > > > > > > > > comments
> > > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > Cheers,
> > > > > > > > > > > >> > > David
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > > > > > gharris1727@gmail.com
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > > > cooperative
> > > > > > > > > rebalancing
> > > > > > > > > > > >> > making it
> > > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> describes how
> > > > > > member
> > > > > > > > > epoch
> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > advance in step with the group epoch and
> assignment
> > > > > > epoch
> > > > > > > > > values. I
> > > > > > > > > > > >> > think
> > > > > > > > > > > >> > > > that this is a good idea for the reasons
> described
> > > > in
> > > > > > the
> > > > > > > > > KIP. When
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > protocol is incrementally assigning
> partitions to a
> > > > > > worker,
> > > > > > > > > what
> > > > > > > > > > > >> member
> > > > > > > > > > > >> > > > epoch does each incremental assignment use?
> Are
> > > > member
> > > > > > epochs
> > > > > > > > > > > >> re-used,
> > > > > > > > > > > >> > and
> > > > > > > > > > > >> > > > a single member epoch can correspond to
> multiple
> > > > > > different
> > > > > > > > > > > >> > (monotonically
> > > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to
> the
> > > > group
> > > > > > > > > > > >> coordinator? If
> > > > > > > > > > > >> > > > not, should custom client-side assignor
> > > > implementations
> > > > > > > > > interact
> > > > > > > > > > > >> with
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > Reason field, and how is its common meaning
> agreed
> > > > > > upon? If
> > > > > > > > > so, what
> > > > > > > > > > > >> > is the
> > > > > > > > > > > >> > > > benefit of a distinct Reason field over
> including
> > > > such
> > > > > > > > > functionality
> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > 3. The following is included in the KIP:
> "Thanks to
> > > > > > this, the
> > > > > > > > > input
> > > > > > > > > > > >> of
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > client side assignor is entirely driven by
> the group
> > > > > > > > > coordinator.
> > > > > > > > > > > >> The
> > > > > > > > > > > >> > > > consumer is no longer responsible for
> maintaining
> > > > any
> > > > > > state
> > > > > > > > > besides
> > > > > > > > > > > >> its
> > > > > > > > > > > >> > > > assigned partitions." Does this mean that the
> > > > > > client-side
> > > > > > > > > assignor
> > > > > > > > > > > >> MAY
> > > > > > > > > > > >> > > > incorporate additional non-Metadata state
> (such as
> > > > > > partition
> > > > > > > > > > > >> > throughput,
> > > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or
> that
> > > > > > additional
> > > > > > > > > > > >> > non-Metadata
> > > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > > > > > >> > > > for
> > > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > > >> > > > and
> > > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > > that
> > > > > > > > > seem to
> > > > > > > > > > > >> > > > overlap significantly. Is it possible for
> these two
> > > > > > > > > implementations
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > unified? This would serve to promote feature
> parity
> > > > of
> > > > > > > > > server-side
> > > > > > > > > > > >> and
> > > > > > > > > > > >> > > > client-side assignors, and would also
> facilitate
> > > > > > operational
> > > > > > > > > > > >> > flexibility in
> > > > > > > > > > > >> > > > certain situations. For example, if a
> server-side
> > > > > > assignor
> > > > > > > > > has some
> > > > > > > > > > > >> > poor
> > > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> patched
> > > > > > assignor to
> > > > > > > > > the
> > > > > > > > > > > >> > client
> > > > > > > > > > > >> > > > and switching one consumer group to a
> client-side
> > > > > > assignor
> > > > > > > > > may be
> > > > > > > > > > > >> > faster
> > > > > > > > > > > >> > > > and less risky than patching all of the
> brokers.
> > > > With
> > > > > > the
> > > > > > > > > currently
> > > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > > reimplementation
> > > > > > would
> > > > > > > > > have
> > > > > > > > > > > >> to be
> > > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> > > > > > significantly,
> > > > > > > > > then it
> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > possible that a reimplementation would not be
> > > > possible.
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > --
> > > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > >
> > > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > I created a confluence page to discuss how
> Connect
> > > > > > would
> > > > > > > > > need to
> > > > > > > > > > > >> > change
> > > > > > > > > > > >> > > > > based on the new rebalance protocol. Here's
> the
> > > > page:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> >
> > > > > > > > > > > >>
> > > > > > > > >
> > > > > >
> > > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > It's also pretty longish and I have tried
> to keep
> > > > a
> > > > > > format
> > > > > > > > > > > >> similar to
> > > > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also,
> do you
> > > > > > think this
> > > > > > > > > > > >> should
> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > moved to a separate discussion thread or is
> this
> > > > one
> > > > > > fine?
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > wrote:
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> Streams.
> > > > > > Sure, I
> > > > > > > > > would do
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang
> Wang <
> > > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > >> It would be great if you could come back
> with
> > > > some
> > > > > > > > > analysis on
> > > > > > > > > > > >> > how to
> > > > > > > > > > > >> > > > > >> implement the Connect side integration
> with
> > > > the new
> > > > > > > > > protocol;
> > > > > > > > > > > >> so
> > > > > > > > > > > >> > far
> > > > > > > > > > > >> > > > > >> besides leveraging on the new "protocol
> type"
> > > > we
> > > > > > did not
> > > > > > > > > yet
> > > > > > > > > > > >> think
> > > > > > > > > > > >> > > > > through
> > > > > > > > > > > >> > > > > >> the Connect side implementations. For
> Streams
> > > > > > here's a
> > > > > > > > > draft of
> > > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> >
> > > > > > > > > > > >>
> > > > > > > > >
> > > > > >
> > > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > > >> > >
> > > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> > Thank you for your response. The
> reason I
> > > > thought
> > > > > > > > > connect can
> > > > > > > > > > > >> > also fit
> > > > > > > > > > > >> > > > > >> into
> > > > > > > > > > > >> > > > > >> > this new scheme is that even today the
> > > > connect
> > > > > > uses a
> > > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to
> empower
> > > > > > > > > rebalances of
> > > > > > > > > > > >> > > > > >> > tasks/connectors. The
> WorkerCoordinator sets
> > > > the
> > > > > > > > > > > >> protocolType()
> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> plumbing
> > > > into
> > > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> > I think the changes to support connect
> would
> > > > be
> > > > > > > > > similar at a
> > > > > > > > > > > >> > high
> > > > > > > > > > > >> > > > > level
> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > >> > > > > >> > the changes in streams mainly because
> of the
> > > > > > Client
> > > > > > > > > side
> > > > > > > > > > > >> > assignors
> > > > > > > > > > > >> > > > > being
> > > > > > > > > > > >> > > > > >> > used in both. At an implementation
> level, we
> > > > > > might
> > > > > > > > > need to
> > > > > > > > > > > >> make
> > > > > > > > > > > >> > a lot
> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > >> > > > > >> > changes to get onto this new assignment
> > > > protocol
> > > > > > like
> > > > > > > > > > > >> enhancing
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> > JoinGroup request/response and
> SyncGroup and
> > > > > > using
> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> streams (or
> > > > > > there
> > > > > > > > > might be
> > > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > > >> > > > > >> > would try to perform a detailed
> analysis of
> > > > the
> > > > > > same
> > > > > > > > > and we
> > > > > > > > > > > >> > can have
> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > >> > > > > >> > separate discussion thread for that as
> that
> > > > would
> > > > > > > > > derail this
> > > > > > > > > > > >> > > > > discussion
> > > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds
> good to
> > > > you.
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David
> Jacot
> > > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> `Assignment#error`.
> > > > > > Sure, I
> > > > > > > > > can
> > > > > > > > > > > >> > mention it.
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from
> his
> > > > current
> > > > > > > > > assignment
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > his
> > > > > > > > > > > >> > > > > >> > > target assignment when he can move to
> > > > epoch 3.
> > > > > > When
> > > > > > > > > that
> > > > > > > > > > > >> > happens,
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > member assignment is updated and
> persisted
> > > > > > with all
> > > > > > > > > its
> > > > > > > > > > > >> > assigned
> > > > > > > > > > > >> > > > > >> > > partitions even if they are not all
> revoked
> > > > > > yet. In
> > > > > > > > > other
> > > > > > > > > > > >> > words, the
> > > > > > > > > > > >> > > > > >> > > member assignment becomes the target
> > > > > > assignment.
> > > > > > > > > This is
> > > > > > > > > > > >> > basically
> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > >> > > > > >> > > optimization to avoid having to
> write all
> > > > the
> > > > > > > > > changes to
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > log.
> > > > > > > > > > > >> > > > > The
> > > > > > > > > > > >> > > > > >> > > examples are based on the persisted
> state
> > > > so I
> > > > > > > > > understand
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can
> improve
> > > > this in
> > > > > > the
> > > > > > > > > > > >> > description.
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse
> the
> > > > > > protocol
> > > > > > > > > with a
> > > > > > > > > > > >> > client side
> > > > > > > > > > > >> > > > > >> > > assignor if it fits in the protocol.
> The
> > > > > > assignment
> > > > > > > > > is
> > > > > > > > > > > >> about
> > > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> > > > Connect
> > > > > > fit
> > > > > > > > > into this?
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM
> Sagar <
> > > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had
> minor
> > > > > > observations:
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error section
> in
> > > > Client
> > > > > > Side
> > > > > > > > > mode
> > > > > > > > > > > >> > Assignment
> > > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case, the
> > > > client
> > > > > > side
> > > > > > > > > assignor
> > > > > > > > > > > >> can
> > > > > > > > > > > >> > > > > return
> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > >> > > > > >> > > > error to the group coordinator`.
> In this
> > > > > > case are
> > > > > > > > > you
> > > > > > > > > > > >> > referring to
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > Assignor returning an
> AssignmentError
> > > > that's
> > > > > > > > > listed down
> > > > > > > > > > > >> > towards
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > > >> > > > > >> > > > If yes, do you think it would make
> sense
> > > > to
> > > > > > > > > mention this
> > > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I
> have a
> > > > > > slight
> > > > > > > > > > > >> confusion,
> > > > > > > > > > > >> > not
> > > > > > > > > > > >> > > > > sure
> > > > > > > > > > > >> > > > > >> if
> > > > > > > > > > > >> > > > > >> > > > others have the same. Consider
> this step:
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> coordinator
> > > > > > > > > transitions him
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > epoch
> > > > > > > > > > > >> > > > > 3
> > > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> > > > persists
> > > > > > the
> > > > > > > > > change and
> > > > > > > > > > > >> > reply.
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> partitions=[foo-0,
> > > > > > foo-1]
> > > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> partitions=[foo-2]
> > > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> partitions=[foo-1]
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions
> to
> > > > epoch 3
> > > > > > but
> > > > > > > > > cannot
> > > > > > > > > > > >> get
> > > > > > > > > > > >> > foo-1
> > > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C
> can't
> > > > get
> > > > > > the
> > > > > > > > > foo-1
> > > > > > > > > > > >> > partition
> > > > > > > > > > > >> > > > > yet,
> > > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > > >> > > > > >> > > > based on the description above, it
> seems
> > > > it
> > > > > > > > > already has
> > > > > > > > > > > >> it.
> > > > > > > > > > > >> > Do you
> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > > >> > > > > >> > > > would be better to remove it and
> > > > populate it
> > > > > > only
> > > > > > > > > when it
> > > > > > > > > > > >> > actually
> > > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > > >> > > > > >> > > > I see this in a lot of other
> places, so
> > > > have
> > > > > > I
> > > > > > > > > > > >> understood it
> > > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be
> out of
> > > > scope
> > > > > > of
> > > > > > > > > this
> > > > > > > > > > > >> > discussion,
> > > > > > > > > > > >> > > > > but
> > > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > > >> > > > > >> > > > what I understood it would
> probably be
> > > > > > running in
> > > > > > > > > client
> > > > > > > > > > > >> > side
> > > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > > >> > > > > >> > > > even on the new rebalance protocol
> as it
> > > > has
> > > > > > its
> > > > > > > > > own
> > > > > > > > > > > >> Custom
> > > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM
> David
> > > > Jacot
> > > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to
> move
> > > > forward
> > > > > > with
> > > > > > > > > > > >> > specialized API
> > > > > > > > > > > >> > > > > >> > > > > instead of relying on one
> generic API.
> > > > For
> > > > > > > > > Connect, we
> > > > > > > > > > > >> > can apply
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > exact same pattern and
> reuse/share the
> > > > core
> > > > > > > > > > > >> > implementation on
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> registry, I
> > > > > > think
> > > > > > > > > that we
> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > > >> > > > > >> > > > > having a tailored API to do
> simple
> > > > > > > > > membership/leader
> > > > > > > > > > > >> > election.
> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM
> Ismael
> > > > > > Juma <
> > > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> java.util.regex.Pattern
> > > > vs
> > > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > >> > > > > >> > > > > > document the differences in
> more
> > > > detail
> > > > > > before
> > > > > > > > > > > >> deciding
> > > > > > > > > > > >> > one
> > > > > > > > > > > >> > > > > way
> > > > > > > > > > > >> > > > > >> or
> > > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > > >> they
> > > > > > > > > > > >> > expect
> > > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If we
> are
> > > > doing
> > > > > > > > > something
> > > > > > > > > > > >> > different,
> > > > > > > > > > > >> > > > > >> then
> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> overload
> > > > with
> > > > > > our own
> > > > > > > > > > > >> Pattern
> > > > > > > > > > > >> > class
> > > > > > > > > > > >> > > > > (I
> > > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose
> re2j's at
> > > > this
> > > > > > > > > point).
> > > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> major new
> > > > > > protocol
> > > > > > > > > should
> > > > > > > > > > > >> > > > > integrate
> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > > > > > recreation case
> > > > > > > > > > > >> > correctly.
> > > > > > > > > > > >> > > > > That's
> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I
> agree with
> > > > > > David
> > > > > > > > > that we'd
> > > > > > > > > > > >> > want to
> > > > > > > > > > > >> > > > > add
> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols that
> don't
> > > > > > have it
> > > > > > > > > yet and
> > > > > > > > > > > >> > that we
> > > > > > > > > > > >> > > > > can
> > > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > > >> > > > > >> > > > > > focus on the internals versus
> adding
> > > > new
> > > > > > APIs
> > > > > > > > > to the
> > > > > > > > > > > >> > Java
> > > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs is
> > > > required
> > > > > > for
> > > > > > > > > > > >> reasonable
> > > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about
> the
> > > > > > coordinator
> > > > > > > > > storing
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > > > centralized
> > > > > > in the
> > > > > > > > > > > >> metadata
> > > > > > > > > > > >> > log for
> > > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> visibility,
> > > > > > consistency,
> > > > > > > > > > > >> etc.).
> > > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > > >> > > > > >> > > > > > sure about automatically
> deleting
> > > > > > configs in a
> > > > > > > > > way
> > > > > > > > > > > >> that
> > > > > > > > > > > >> > they
> > > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > >> > > > > >> > > > > > recovered. A good property for
> modern
> > > > > > systems
> > > > > > > > > is to
> > > > > > > > > > > >> > minimize
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> scenarios.
> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47
> PM David
> > > > > > Jacot
> > > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers
> are
> > > > below:
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> especially
> > > > > > the last
> > > > > > > > > step
> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> protocol, in
> > > > which
> > > > > > we
> > > > > > > > > would
> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > > protocols /
> > > > > > rpcs
> > > > > > > > > are
> > > > > > > > > > > >> used
> > > > > > > > > > > >> > by
> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> could
> > > > > > "mimic"
> > > > > > > > > the old
> > > > > > > > > > > >> > behavior
> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
> we
> > > > need to
> > > > > > > > > present
> > > > > > > > > > > >> about.
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a new
> > > > version
> > > > > > of KIP
> > > > > > > > > which
> > > > > > > > > > > >> > includes
> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> > > > "Supporting
> > > > > > > > > Online
> > > > > > > > > > > >> > Consumer
> > > > > > > > > > > >> > > > > Group
> > > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > > > Deprecation,
> > > > > > and
> > > > > > > > > Migration
> > > > > > > > > > > >> > Plan". I
> > > > > > > > > > > >> > > > > >> think
> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > >> > > > > > > I have to think through a
> few cases
> > > > > > now but
> > > > > > > > > the
> > > > > > > > > > > >> > overall idea
> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> understandable.
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids.
> So
> > > > far as
> > > > > > > > > KIP-516 the
> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> client,
> > > > but
> > > > > > they
> > > > > > > > > are not
> > > > > > > > > > > >> > exposed
> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
> the
> > > > > > question is,
> > > > > > > > > first
> > > > > > > > > > > >> > should we
> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> the
> > > > names
> > > > > > -> ids
> > > > > > > > > mapping
> > > > > > > > > > > >> > itself
> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> existing
> > > > RPCs
> > > > > > and the
> > > > > > > > > new
> > > > > > > > > > > >> RPCs
> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> consider
> > > > > > exposing
> > > > > > > > > the
> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > >> well
> > > > > > > > > > > >> > as in
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> topic
> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> include
> > > > > > converting
> > > > > > > > > all
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > offsets
> > > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in
> this
> > > > KIP,
> > > > > > the
> > > > > > > > > consumer
> > > > > > > > > > > >> > would be
> > > > > > > > > > > >> > > > > >> able
> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic
> ids. That
> > > > > > being
> > > > > > > > > said, it
> > > > > > > > > > > >> > still has
> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > > >> > > > > >> > > > > > > the topics names in various
> APIs so
> > > > > > having a
> > > > > > > > > > > >> mapping
> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> opinion on
> > > > > > this.
> > > > > > > > > Here I
> > > > > > > > > > > >> > wonder if
> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this
> KIP. I
> > > > would
> > > > > > rather
> > > > > > > > > focus
> > > > > > > > > > > >> on
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > > > > > separately if
> > > > > > > > > we see
> > > > > > > > > > > >> > value in
> > > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's
> point about
> > > > > > using
> > > > > > > > > topic ids
> > > > > > > > > > > >> > in the
> > > > > > > > > > > >> > > > > >> > > > > > >
> ConsumerGroupHeartbeatRequest, I
> > > > think
> > > > > > that
> > > > > > > > > there
> > > > > > > > > > > >> is
> > > > > > > > > > > >> > one
> > > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer
> will
> > > > have
> > > > > > the
> > > > > > > > > > > >> opportunity
> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> passing
> > > > them
> > > > > > into
> > > > > > > > > the
> > > > > > > > > > > >> group
> > > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > > > coordinator
> > > > > > will
> > > > > > > > > also
> > > > > > > > > > > >> notice
> > > > > > > > > > > >> > it but
> > > > > > > > > > > >> > > > > >> it
> > > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> reject an
> > > > > > invalid
> > > > > > > > > topic in
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> all
> > > > other
> > > > > > minor
> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> question:
> > > > > > personally
> > > > > > > > > I think
> > > > > > > > > > > >> > it's not
> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> API with
> > > > > > Pattern,
> > > > > > > > > but
> > > > > > > > > > > >> > instead we
> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> that our
> > > > > > > > > subscription may
> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> case is a
> > > > > > very rare
> > > > > > > > > > > >> scenario
> > > > > > > > > > > >> > I felt
> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > > subscription may
> > > > > > > > > be
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
> look at
> > > > the
> > > > > > > > > differences
> > > > > > > > > > > >> > between
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > > >> > > > > >> > > > > > > engines to better understand
> the
> > > > > > potential
> > > > > > > > > issues.
> > > > > > > > > > > >> My
> > > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > > >> > > > > >> > > > > > > that would work for all the
> basic
> > > > > > regular
> > > > > > > > > > > >> > expressions. The
> > > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
> about
> > > > the
> > > > > > various
> > > > > > > > > > > >> character
> > > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > > >> > > > > >> > > > > > > wonder what other people
> think
> > > > about
> > > > > > this.
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at
> 11:28 PM
> > > > > > Guozhang
> > > > > > > > > Wang <
> > > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on
> the high
> > > > > > level
> > > > > > > > > there are
> > > > > > > > > > > >> > two meta
> > > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> especially
> > > > > > the last
> > > > > > > > > step
> > > > > > > > > > > >> when
> > > > > > > > > > > >> > > > > clients
> > > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > > >> > > > > >> > > > > > > > to enable the new
> protocol, in
> > > > which
> > > > > > we
> > > > > > > > > would
> > > > > > > > > > > >> have a
> > > > > > > > > > > >> > > > > window
> > > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > > protocols /
> > > > > > rpcs
> > > > > > > > > are
> > > > > > > > > > > >> used
> > > > > > > > > > > >> > by
> > > > > > > > > > > >> > > > > >> members
> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> could
> > > > > > "mimic"
> > > > > > > > > the old
> > > > > > > > > > > >> > behavior
> > > > > > > > > > > >> > > > > >> while
> > > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something
> we
> > > > need to
> > > > > > > > > present
> > > > > > > > > > > >> about.
> > > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids.
> So
> > > > far as
> > > > > > > > > KIP-516 the
> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> client,
> > > > but
> > > > > > they
> > > > > > > > > are not
> > > > > > > > > > > >> > exposed
> > > > > > > > > > > >> > > > > via
> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think
> the
> > > > > > question is,
> > > > > > > > > first
> > > > > > > > > > > >> > should we
> > > > > > > > > > > >> > > > > >> let
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining
> the
> > > > names
> > > > > > -> ids
> > > > > > > > > mapping
> > > > > > > > > > > >> > itself
> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented
> existing
> > > > RPCs
> > > > > > and the
> > > > > > > > > new
> > > > > > > > > > > >> RPCs
> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> consider
> > > > > > exposing
> > > > > > > > > the
> > > > > > > > > > > >> > topic ids
> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > > subscribe/assign, as
> > > > > > > > > > > >> well
> > > > > > > > > > > >> > as in
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like
> topic
> > > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> all
> > > > other
> > > > > > minor
> > > > > > > > > > > >> questions
> > > > > > > > > > > >> > > > > except
> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> question:
> > > > > > personally
> > > > > > > > > I think
> > > > > > > > > > > >> > it's not
> > > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe
> API with
> > > > > > Pattern,
> > > > > > > > > but
> > > > > > > > > > > >> > instead we
> > > > > > > > > > > >> > > > > >> still
> > > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > > >> > > > > >> > > > > > > > while just documenting
> that our
> > > > > > > > > subscription may
> > > > > > > > > > > >> be
> > > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible
> case is a
> > > > > > very rare
> > > > > > > > > > > >> scenario
> > > > > > > > > > > >> > I felt
> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > > subscription may
> > > > > > > > > be
> > > > > > > > > > > >> more
> > > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at
> 5:23 AM
> > > > > > David Jacot
> > > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your
> feedback. Let
> > > > me
> > > > > > answer
> > > > > > > > > your
> > > > > > > > > > > >> > questions
> > > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> premature to
> > > > > > talk about
> > > > > > > > > > > >> target
> > > > > > > > > > > >> > > > > versions
> > > > > > > > > > > >> > > > > >> for
> > > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the
> existing group
> > > > > > protocol.
> > > > > > > > > > > >> Unlike
> > > > > > > > > > > >> > KRaft,
> > > > > > > > > > > >> > > > > >> this
> > > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and
> hence
> > > > > > > > > deprecation/removal
> > > > > > > > > > > >> > will be
> > > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly
> applications
> > > > migrate
> > > > > > to
> > > > > > > > > the new
> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> > > > remove it.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> intend to
> > > > > > release
> > > > > > > > > this in
> > > > > > > > > > > >> > 4.x, but
> > > > > > > > > > > >> > > > > it
> > > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that
> as a
> > > > way to
> > > > > > > > > estimate
> > > > > > > > > > > >> when
> > > > > > > > > > > >> > we'd
> > > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I
> also
> > > > > > suggest
> > > > > > > > > removing
> > > > > > > > > > > >> > this part.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> reasoning. As
> > > > > > > > > explained, I
> > > > > > > > > > > >> plan
> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java
> while we
> > > > > > implement
> > > > > > > > > the new
> > > > > > > > > > > >> > protocol.
> > > > > > > > > > > >> > > > > >> This
> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > >> > > > > >> > > > > > > > > that the internals will
> be
> > > > slightly
> > > > > > > > > different
> > > > > > > > > > > >> > (e.g.
> > > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
> tighten
> > > > the
> > > > > > > > > switch from
> > > > > > > > > > > >> > the old
> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
> group
> > > > > > coordinator
> > > > > > > > > to a
> > > > > > > > > > > >> > major
> > > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to
> use a
> > > > flag
> > > > > > to do
> > > > > > > > > the
> > > > > > > > > > > >> > switch
> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh
> out the
> > > > > > details of
> > > > > > > > > the
> > > > > > > > > > > >> > migration
> > > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
> support
> > > > > > online
> > > > > > > > > > > >> migrations.
> > > > > > > > > > > >> > Is
> > > > > > > > > > > >> > > > > that
> > > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in
> detail. It
> > > > could
> > > > > > also
> > > > > > > > > be done
> > > > > > > > > > > >> > as a
> > > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support
> online
> > > > > > migrations
> > > > > > > > > for the
> > > > > > > > > > > >> > group.
> > > > > > > > > > > >> > > > > That
> > > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> protocol
> > > > > > will be
> > > > > > > > > able to
> > > > > > > > > > > >> > switch to
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain
> how that
> > > > > > will work
> > > > > > > > > > > >> though.
> > > > > > > > > > > >> > It is
> > > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> > > > upgraded or
> > > > > > > > > rolled to a
> > > > > > > > > > > >> > software
> > > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
> Both the
> > > > > > old and
> > > > > > > > > the new
> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > > >> > > > > >> > > > > > > > > support the old protocol
> and
> > > > rely
> > > > > > on the
> > > > > > > > > same
> > > > > > > > > > > >> > persisted
> > > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > > >> > > > > >> > > > > > > > > they can work together.
> This
> > > > point
> > > > > > is an
> > > > > > > > > > > >> offline
> > > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > > >> > > > > >> > > > > > > > > do this one live because
> it
> > > > would
> > > > > > require
> > > > > > > > > > > >> > shutting down
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and starting
> up the
> > > > > > new one
> > > > > > > > > and
> > > > > > > > > > > >> that
> > > > > > > > > > > >> > would
> > > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > > > > > version/IBP
> > > > > > > > > must be
> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> X
> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new
> protocol.
> > > > This
> > > > > > cannot
> > > > > > > > > be done
> > > > > > > > > > > >> > before
> > > > > > > > > > > >> > > > > 1)
> > > > > > > > > > > >> > > > > >> is
> > > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the
> old
> > > > > > coordinator
> > > > > > > > > doesn't
> > > > > > > > > > > >> > support
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> > > > upgraded
> > > > > > to a
> > > > > > > > > version
> > > > > > > > > > > >> > > > > supporting
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> KIP-792).
> > > > If
> > > > > > the
> > > > > > > > > consumer
> > > > > > > > > > > >> is
> > > > > > > > > > > >> > > > > already
> > > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at
> this
> > > > point.
> > > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be
> rolled
> > > > > > with the
> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > flag
> > > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> automatically
> > > > > > > > > converted when
> > > > > > > > > > > >> > the first
> > > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
> joins
> > > > the
> > > > > > group.
> > > > > > > > > While
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > members
> > > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> > > > upgraded,
> > > > > > the old
> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > is
> > > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of
> this in
> > > > the
> > > > > > KIP.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we
> are
> > > > > > pushing the
> > > > > > > > > pattern
> > > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like there
> > > > could be
> > > > > > some
> > > > > > > > > tricky
> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
> users to
> > > > > > detect
> > > > > > > > > that
> > > > > > > > > > > >> they
> > > > > > > > > > > >> > need to
> > > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to
> the new
> > > > > > protocol?
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit
> more
> > > > > > optimistic
> > > > > > > > > than
> > > > > > > > > > > >> you
> > > > > > > > > > > >> > on this
> > > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > > >> > > > > >> > > > > > > > > believe that the
> majority of
> > > > the
> > > > > > cases
> > > > > > > > > are
> > > > > > > > > > > >> simple
> > > > > > > > > > > >> > > > > regexes
> > > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > >> > > > > >> > > > > > > > > work with the new
> engine. The
> > > > > > > > > coordinator will
> > > > > > > > > > > >> > verify
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer
> if the
> > > > > > regex is
> > > > > > > > > not
> > > > > > > > > > > >> valid.
> > > > > > > > > > > >> > > > > Coming
> > > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
> worst
> > > > case,
> > > > > > the
> > > > > > > > > first
> > > > > > > > > > > >> > upgraded
> > > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > > >> > > > > >> > > > > > > > > the group will be
> rejected.
> > > > This
> > > > > > should
> > > > > > > > > be used
> > > > > > > > > > > >> > as the
> > > > > > > > > > > >> > > > > >> last
> > > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> > > > validate
> > > > > > their
> > > > > > > > > regex
> > > > > > > > > > > >> > before
> > > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test
> them with
> > > > > > another
> > > > > > > > > group.
> > > > > > > > > > > >> For
> > > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a
> pre-prod
> > > > > > environment.
> > > > > > > > > > > >> Another
> > > > > > > > > > > >> > way
> > > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > >> > > > > > > > > extend the
> consumer-group tool
> > > > to
> > > > > > > > > provide a
> > > > > > > > > > > >> regex
> > > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> > > > enough in
> > > > > > your
> > > > > > > > > > > >> opinion?
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> > > > question,
> > > > > > will
> > > > > > > > > the
> > > > > > > > > > > >> Java
> > > > > > > > > > > >> > client
> > > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the current
> regex
> > > > > > engine for
> > > > > > > > > > > >> > compatibility
> > > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to
> keep using
> > > > > > client
> > > > > > > > > based
> > > > > > > > > > > >> > regex at
> > > > > > > > > > > >> > > > > >> first
> > > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and
> then
> > > > > > migrate to
> > > > > > > > > server
> > > > > > > > > > > >> > based
> > > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > I understand your point
> but I
> > > > am
> > > > > > > > > concerned that
> > > > > > > > > > > >> > this
> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in this
> mode.
> > > > That
> > > > > > > > > would go
> > > > > > > > > > > >> > against our
> > > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
> because
> > > > we
> > > > > > would
> > > > > > > > > have to
> > > > > > > > > > > >> > continue
> > > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the
> client
> > > > side. I
> > > > > > would
> > > > > > > > > rather
> > > > > > > > > > > >> > not do
> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that
> the group
> > > > > > > > > coordinator
> > > > > > > > > > > >> will
> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations and
> that
> > > > the
> > > > > > > > > > > >> configurations
> > > > > > > > > > > >> > will be
> > > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will
> a
> > > > > > transition to
> > > > > > > > > DEAD
> > > > > > > > > > > >> > trigger
> > > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > > configurations
> > > > > > will be
> > > > > > > > > > > >> deleted
> > > > > > > > > > > >> > when
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go
> together.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to
> store
> > > > the
> > > > > > > > > configs in
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > group
> > > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> cluster
> > > > > > configs and
> > > > > > > > > their
> > > > > > > > > > > >> > values?
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The
> group
> > > > > > > > > configurations are
> > > > > > > > > > > >> > overrides
> > > > > > > > > > > >> > > > > >> of
> > > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to
> know
> > > > all
> > > > > > the
> > > > > > > > > overrides
> > > > > > > > > > > >> > though,
> > > > > > > > > > > >> > > > > you
> > > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > > > coordinators.
> > > > > > You
> > > > > > > > > cannot
> > > > > > > > > > > >> > rely on
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> > > > configure a
> > > > > > group
> > > > > > > > > before
> > > > > > > > > > > >> > starting
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > > >> > > > > >> > > > > > > > > > we considered allowing
> the
> > > > > > explicit
> > > > > > > > > creation
> > > > > > > > > > > >> of
> > > > > > > > > > > >> > > > > groups?
> > > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> decoupled
> > > > from
> > > > > > the
> > > > > > > > > group
> > > > > > > > > > > >> > lifecycle.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > > > > > automatically
> > > > > > > > > created in
> > > > > > > > > > > >> > this
> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be lost
> after
> > > > the
> > > > > > > > > retention
> > > > > > > > > > > >> > period of
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > > Consumer.subscribe
> > > > > > method
> > > > > > > > > for the
> > > > > > > > > > > >> > Java
> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> `java.util.regex.Pattern` of
> > > > do
> > > > > > we
> > > > > > > > > have to
> > > > > > > > > > > >> > introduce
> > > > > > > > > > > >> > > > > an
> > > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> question. I
> > > > > > forgot
> > > > > > > > > about
> > > > > > > > > > > >> that
> > > > > > > > > > > >> > one.
> > > > > > > > > > > >> > > > > As
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> `java.util.regex.Pattern` is
> > > > not
> > > > > > fully
> > > > > > > > > > > >> compatible
> > > > > > > > > > > >> > with
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might
> be
> > > > better
> > > > > > to
> > > > > > > > > deprecate
> > > > > > > > > > > >> > it and
> > > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string. We
> would
> > > > > > rely on
> > > > > > > > > the
> > > > > > > > > > > >> server
> > > > > > > > > > > >> > side
> > > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I
> think
> > > > that
> > > > > > we
> > > > > > > > > could
> > > > > > > > > > > >> still
> > > > > > > > > > > >> > try to
> > > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
> should
> > > > > > work, I
> > > > > > > > > think, in
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with
> Justine
> > > > that we
> > > > > > > > > should be
> > > > > > > > > > > >> > clearer
> > > > > > > > > > > >> > > > > about
> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > > IBP/metadata.version
> > > > > > from the
> > > > > > > > > > > >> feature
> > > > > > > > > > > >> > flag.
> > > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the default
> for
> > > > the
> > > > > > feature
> > > > > > > > > flag
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > true
> > > > > > > > > > > >> > > > > >> based
> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once
> we
> > > > want to
> > > > > > make
> > > > > > > > > it the
> > > > > > > > > > > >> > default.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that
> feature
> > > > > > flag
> > > > > > > > > mainly
> > > > > > > > > > > >> > during the
> > > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> > > > mentioned
> > > > > > it, I
> > > > > > > > > think,
> > > > > > > > > > > >> > because
> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for
> it.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the
> protocol APIs
> > > > > > don't
> > > > > > > > > mention
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > required
> > > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> > > > consistency.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd
> that
> > > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > > >> > > > > requires
> > > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do
> more
> > > > than
> > > > > > > > > reading.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> > > > current
> > > > > > > > > protocol works
> > > > > > > > > > > >> > though.
> > > > > > > > > > > >> > > > > We
> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to
> join a
> > > > > > group. We
> > > > > > > > > could
> > > > > > > > > > > >> > consider
> > > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that
> it is
> > > > worth
> > > > > > it.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> recreation
> > > > > > handled by
> > > > > > > > > the
> > > > > > > > > > > >> > consumer
> > > > > > > > > > > >> > > > > with
> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be
> good to
> > > > > > have a
> > > > > > > > > section
> > > > > > > > > > > >> on
> > > > > > > > > > > >> > this.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > > > > > perspective, the
> > > > > > > > > new
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > will
> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will
> treat it
> > > > like a
> > > > > > > > > topic with
> > > > > > > > > > > >> a
> > > > > > > > > > > >> > > > > different
> > > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> > > > fetch/commit
> > > > > > > > > offsets
> > > > > > > > > > > >> APIs
> > > > > > > > > > > >> > do not
> > > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would
> > > > reuse the
> > > > > > > > > offsets
> > > > > > > > > > > >> based
> > > > > > > > > > > >> > on the
> > > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > > >> > > > > >> > > > > > > > > that we should update
> those
> > > > APIs
> > > > > > as well
> > > > > > > > > in
> > > > > > > > > > > >> order
> > > > > > > > > > > >> > to be
> > > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> strengthen
> > > > the
> > > > > > > > > semantics of
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions
> we will
> > > > > > write the
> > > > > > > > > new
> > > > > > > > > > > >> > coordinator
> > > > > > > > > > > >> > > > > >> in
> > > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > > >> > > > > >> > > > > > > > > > this is an
> implementation
> > > > > > detail, do
> > > > > > > > > we plan
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > have a
> > > > > > > > > > > >> > > > > >> new
> > > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> scalability
> > > > > > goal when
> > > > > > > > > it
> > > > > > > > > > > >> comes
> > > > > > > > > > > >> > to how
> > > > > > > > > > > >> > > > > >> many
> > > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> support?
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at
> the
> > > > > > moment. The
> > > > > > > > > > > >> protocol
> > > > > > > > > > > >> > should
> > > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > > >> > > > > >> > > > > > > > > of members per group. We
> will
> > > > > > measure
> > > > > > > > > this when
> > > > > > > > > > > >> > we have
> > > > > > > > > > > >> > > > > a
> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note
> that we
> > > > might
> > > > > > have
> > > > > > > > > other
> > > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset
> commits).
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> having
> > > > > > > > > SubscribedTopidIds
> > > > > > > > > > > >> > instead
> > > > > > > > > > > >> > > > > of
> > > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to
> resolve the
> > > > > > regex on
> > > > > > > > > the
> > > > > > > > > > > >> > server, we
> > > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> > > > difference
> > > > > > is that
> > > > > > > > > > > >> sending
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the
> topic
> > > > names
> > > > > > is less
> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to
> > > > handle
> > > > > > if we
> > > > > > > > > have
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > ids.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> consolidate the
> > > > > > metadata
> > > > > > > > > lookup
> > > > > > > > > > > >> > on the
> > > > > > > > > > > >> > > > > >> server
> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree
> with your
> > > > > > point. As
> > > > > > > > > a
> > > > > > > > > > > >> second
> > > > > > > > > > > >> > > > > though,
> > > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here
> for the
> > > > > > delete and
> > > > > > > > > > > >> > recreation
> > > > > > > > > > > >> > > > > case.
> > > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow
> > > > users to
> > > > > > > > > subscribe
> > > > > > > > > > > >> with
> > > > > > > > > > > >> > topic
> > > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > future because that is
> the only
> > > > > > way to be
> > > > > > > > > > > >> really
> > > > > > > > > > > >> > robust
> > > > > > > > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at
> 1:38 PM
> > > > > > David
> > > > > > > > > Jacot <
> > > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your
> comments.
> > > > Please
> > > > > > find
> > > > > > > > > my
> > > > > > > > > > > >> answers
> > > > > > > > > > > >> > > > > below.
> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol
> > > > relies on
> > > > > > > > > topic IDs
> > > > > > > > > > > >> > with the
> > > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in
> the
> > > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > > >> > > > > >> I
> > > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names
> is the
> > > > > > right call
> > > > > > > > > here.
> > > > > > > > > > > >> I
> > > > > > > > > > > >> > need to
> > > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> Obviously, the
> > > > > > KIP does
> > > > > > > > > not
> > > > > > > > > > > >> > change the
> > > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use
> topic
> > > > IDs.
> > > > > > This
> > > > > > > > > may be
> > > > > > > > > > > >> > something
> > > > > > > > > > > >> > > > > >> that
> > > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
> would
> > > > give
> > > > > > better
> > > > > > > > > > > >> overall
> > > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I
> think that
> > > > I
> > > > > > should
> > > > > > > > > not
> > > > > > > > > > > >> have
> > > > > > > > > > > >> > > > > mentioned
> > > > > > > > > > > >> > > > > >> > this
> > > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it.
> We can
> > > > > > use an
> > > > > > > > > internal
> > > > > > > > > > > >> > > > > >> configuration
> > > > > > > > > > > >> > > > > >> > > while
> > > > > > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
> will be
> > > > > > > > > supported. The
> > > > > > > > > > > >> > change is
> > > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is
> that the
> > > > > > cluster
> > > > > > > > > uses
> > > > > > > > > > > >> topic
> > > > > > > > > > > >> > IDs.
> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022
> at 9:53
> > > > PM
> > > > > > > > > Guozhang
> > > > > > > > > > > >> Wang <
> > > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> feedback.
> > > > Here
> > > > > > are
> > > > > > > > > some
> > > > > > > > > > > >> replies
> > > > > > > > > > > >> > > > > inlined
> > > > > > > > > > > >> > > > > >> > > below:
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022
> at
> > > > 2:53 AM
> > > > > > > > > Ismael Juma
> > > > > > > > > > > >> <
> > > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the
> KIP. This
> > > > has
> > > > > > the
> > > > > > > > > > > >> potential
> > > > > > > > > > > >> > to be a
> > > > > > > > > > > >> > > > > >> > great
> > > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > > questions/comments:
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> > > > premature to
> > > > > > talk
> > > > > > > > > about
> > > > > > > > > > > >> > target
> > > > > > > > > > > >> > > > > >> versions
> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> existing
> > > > group
> > > > > > > > > protocol.
> > > > > > > > > > > >> > Unlike
> > > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol
> and hence
> > > > > > > > > > > >> > deprecation/removal will
> > > > > > > > > > > >> > > > > >> be
> > > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> applications
> > > > > > migrate
> > > > > > > > > to the
> > > > > > > > > > > >> new
> > > > > > > > > > > >> > > > > >> protocol.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with
> you. I
> > > > think
> > > > > > we can
> > > > > > > > > > > >> remove
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> proposed
> > > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > > > Deprecation,
> > > > > > and
> > > > > > > > > Migration
> > > > > > > > > > > >> > Plan` and
> > > > > > > > > > > >> > > > > >> > > instead
> > > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide
> in the
> > > > > > future
> > > > > > > > > about
> > > > > > > > > > > >> when
> > > > > > > > > > > >> > we
> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and
> behaviors.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> > > > intend to
> > > > > > > > > release this
> > > > > > > > > > > >> > in 4.x,
> > > > > > > > > > > >> > > > > >> but
> > > > > > > > > > > >> > > > > >> > it
> > > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
> that as
> > > > a
> > > > > > way to
> > > > > > > > > > > >> estimate
> > > > > > > > > > > >> > when
> > > > > > > > > > > >> > > > > we'd
> > > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group
> protocol, I
> > > > also
> > > > > > suggest
> > > > > > > > > > > >> removing
> > > > > > > > > > > >> > this
> > > > > > > > > > > >> > > > > >> part.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> > > > specifically
> > > > > > > > > related to
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more
> for
> > > > client
> > > > > > > > > upgrades.
> > > > > > > > > > > >> I.e.
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may
> be done
> > > > > > first,
> > > > > > > > > and then
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > client
> > > > > > > > > > > >> > > > > >> > side,
> > > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> "released"
> > > > by
> > > > > > the
> > > > > > > > > time
> > > > > > > > > > > >> clients
> > > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable
> the
> > > > > > feature the
> > > > > > > > > > > >> clients
> > > > > > > > > > > >> > need to
> > > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a
> rolling
> > > > bounce
> > > > > > and
> > > > > > > > > then set
> > > > > > > > > > > >> > the flag
> > > > > > > > > > > >> > > > > >> > with a
> > > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we
> feel
> > > > it's
> > > > > > > > > better to be
> > > > > > > > > > > >> > released
> > > > > > > > > > > >> > > > > >> in a
> > > > > > > > > > > >> > > > > >> > > major
> > > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to
> flesh out
> > > > the
> > > > > > > > > details of
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > migration
> > > > > > > > > > > >> > > > > >> > > story.
> > > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we
> will
> > > > support
> > > > > > online
> > > > > > > > > > > >> > migrations. Is
> > > > > > > > > > > >> > > > > >> that
> > > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> detail.
> > > > It
> > > > > > could
> > > > > > > > > also be
> > > > > > > > > > > >> > done as a
> > > > > > > > > > > >> > > > > >> > > separate
> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's
> the
> > > > part we
> > > > > > can
> > > > > > > > > be more
> > > > > > > > > > > >> > concrete
> > > > > > > > > > > >> > > > > >> about
> > > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to
> your
> > > > > > question 2)
> > > > > > > > > above).
> > > > > > > > > > > >> > We will
> > > > > > > > > > > >> > > > > >> work
> > > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel
> as we
> > > > > > solicit
> > > > > > > > > more
> > > > > > > > > > > >> > feedback.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that
> we are
> > > > > > pushing
> > > > > > > > > the
> > > > > > > > > > > >> > pattern
> > > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like
> there
> > > > > > could be
> > > > > > > > > some
> > > > > > > > > > > >> tricky
> > > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism
> for
> > > > users to
> > > > > > > > > detect that
> > > > > > > > > > > >> > they
> > > > > > > > > > > >> > > > > need
> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > >> > > > > >> > > update
> > > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching
> to the
> > > > new
> > > > > > > > > protocol?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need
> some
> > > > > > tooling for
> > > > > > > > > > > >> non-java
> > > > > > > > > > > >> > client
> > > > > > > > > > > >> > > > > >> > users
> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client
> before
> > > > > > > > > switching to
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > new
> > > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top
> of my
> > > > head
> > > > > > > > > though,
> > > > > > > > > > > >> maybe
> > > > > > > > > > > >> > others
> > > > > > > > > > > >> > > > > >> like
> > > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the
> last
> > > > > > question,
> > > > > > > > > will the
> > > > > > > > > > > >> > Java
> > > > > > > > > > > >> > > > > >> client
> > > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> current
> > > > regex
> > > > > > > > > engine for
> > > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to
> keep
> > > > using
> > > > > > > > > client
> > > > > > > > > > > >> based
> > > > > > > > > > > >> > regex
> > > > > > > > > > > >> > > > > at
> > > > > > > > > > > >> > > > > >> > > first to
> > > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple
> and
> > > > then
> > > > > > migrate
> > > > > > > > > to
> > > > > > > > > > > >> > server based
> > > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not
> thought
> > > > > > about
> > > > > > > > > that for
> > > > > > > > > > > >> > java
> > > > > > > > > > > >> > > > > >> clients,
> > > > > > > > > > > >> > > > > >> > > and
> > > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > > > > > compatibility
> > > > > > > > > issues do
> > > > > > > > > > > >> > you have
> > > > > > > > > > > >> > > > > in
> > > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say
> that the
> > > > group
> > > > > > > > > coordinator
> > > > > > > > > > > >> > will be
> > > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the configurations
> and
> > > > that
> > > > > > the
> > > > > > > > > > > >> > configurations
> > > > > > > > > > > >> > > > > will
> > > > > > > > > > > >> > > > > >> be
> > > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted.
> Will a
> > > > > > > > > transition to
> > > > > > > > > > > >> DEAD
> > > > > > > > > > > >> > > > > trigger
> > > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD
> state
> > > > is an
> > > > > > > > > ending
> > > > > > > > > > > >> state
> > > > > > > > > > > >> > (we
> > > > > > > > > > > >> > > > > would
> > > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the group
> is
> > > > EMPTY
> > > > > > and
> > > > > > > > > also all
> > > > > > > > > > > >> of
> > > > > > > > > > > >> > its
> > > > > > > > > > > >> > > > > >> > metadata
> > > > > > > > > > > >> > > > > >> > > are
> > > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to
> DEAD this
> > > > > > group
> > > > > > > > > would
> > > > > > > > > > > >> never
> > > > > > > > > > > >> > be
> > > > > > > > > > > >> > > > > >> revived.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice
> to
> > > > store
> > > > > > the
> > > > > > > > > configs
> > > > > > > > > > > >> in
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > group
> > > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> > > > cluster
> > > > > > configs
> > > > > > > > > and
> > > > > > > > > > > >> their
> > > > > > > > > > > >> > > > > values?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good
> question,
> > > > and our
> > > > > > > > > thoughts
> > > > > > > > > > > >> are
> > > > > > > > > > > >> > that
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> > > > overrides
> > > > > > of the
> > > > > > > > > > > >> > cluster-level
> > > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per group
> so
> > > > when an
> > > > > > > > > admin list
> > > > > > > > > > > >> > cluster
> > > > > > > > > > > >> > > > > >> > configs
> > > > > > > > > > > >> > > > > >> > > it's
> > > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> cluster-level
> > > > > > > > > "defaults", not
> > > > > > > > > > > >> > showing
> > > > > > > > > > > >> > > > > >> any
> > > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would
> someone
> > > > > > configure a
> > > > > > > > > group
> > > > > > > > > > > >> > before
> > > > > > > > > > > >> > > > > >> starting
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> allowing
> > > > the
> > > > > > explicit
> > > > > > > > > > > >> > creation of
> > > > > > > > > > > >> > > > > >> groups?
> > > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> > > > decoupled
> > > > > > from
> > > > > > > > > the group
> > > > > > > > > > > >> > > > > lifecycle.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
> created
> > > > > > before
> > > > > > > > > the group
> > > > > > > > > > > >> > itself
> > > > > > > > > > > >> > > > > as
> > > > > > > > > > > >> > > > > >> an
> > > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> > > > requires
> > > > > > the
> > > > > > > > > > > >> corresponding
> > > > > > > > > > > >> > > > > >> request to
> > > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> based on
> > > > the
> > > > > > group
> > > > > > > > > id ---
> > > > > > > > > > > >> > the only
> > > > > > > > > > > >> > > > > >> > thing
> > > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group
> itself is
> > > > gone
> > > > > > we
> > > > > > > > > also check
> > > > > > > > > > > >> > if there
> > > > > > > > > > > >> > > > > >> are
> > > > > > > > > > > >> > > > > >> > > any
> > > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to
> that
> > > > group
> > > > > > and
> > > > > > > > > delete
> > > > > > > > > > > >> as
> > > > > > > > > > > >> > well.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this
> indeed
> > > > > > introduces an
> > > > > > > > > > > >> > asymmetry on
> > > > > > > > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > creation
> > > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> config
> > > > > > entities,
> > > > > > > > > and we
> > > > > > > > > > > >> > would like
> > > > > > > > > > > >> > > > > >> to
> > > > > > > > > > > >> > > > > >> > > hear
> > > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we
> should
> > > > aim
> > > > > > for
> > > > > > > > > symmetry
> > > > > > > > > > > >> > i.e.
> > > > > > > > > > > >> > > > > >> totally
> > > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not
> > > > delete
> > > > > > them at
> > > > > > > > > all
> > > > > > > > > > > >> when
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> group
> > > > > > > > > > > >> > > > > >> > is
> > > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> deletion
> > > > > > operations
> > > > > > > > > by
> > > > > > > > > > > >> > themselves.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > > > > Consumer.subscribe
> > > > > > > > > method for
> > > > > > > > > > > >> > the Java
> > > > > > > > > > > >> > > > > >> > client
> > > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > `java.util.regex.Pattern` of
> > > > > > do we
> > > > > > > > > have
> > > > > > > > > > > >> to
> > > > > > > > > > > >> > > > > >> introduce an
> > > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not
> need to
> > > > > > introduce
> > > > > > > > > an
> > > > > > > > > > > >> > overload, but
> > > > > > > > > > > >> > > > > >> I'm
> > > > > > > > > > > >> > > > > >> > > all
> > > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> compatibility
> > > > > > issues
> > > > > > > > > that we
> > > > > > > > > > > >> may
> > > > > > > > > > > >> > > > > overlook.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
> Justine
> > > > > > that we
> > > > > > > > > should
> > > > > > > > > > > >> be
> > > > > > > > > > > >> > clearer
> > > > > > > > > > > >> > > > > >> > about
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > > > IBP/metadata.version
> > > > > > > > > from the
> > > > > > > > > > > >> > feature
> > > > > > > > > > > >> > > > > >> flag.
> > > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> default
> > > > for
> > > > > > the
> > > > > > > > > feature
> > > > > > > > > > > >> flag
> > > > > > > > > > > >> > to
> > > > > > > > > > > >> > > > > true
> > > > > > > > > > > >> > > > > >> > > based on
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
> once we
> > > > > > want to
> > > > > > > > > make it
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > >> default.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> protocol
> > > > APIs
> > > > > > don't
> > > > > > > > > mention
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > >> required
> > > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that
> for
> > > > > > consistency.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can
> certainly do
> > > > that.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit
> odd that
> > > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > > > > >> > > > > >> > requires
> > > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to
> do
> > > > more
> > > > > > than
> > > > > > > > > reading.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
> myself
> > > > as
> > > > > > well,
> > > > > > > > > but in
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > end we
> > > > > > > > > > > >> > > > > >> > could
> > > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative:
> adding
> > > > > > Write
> > > > > > > > > Group
> > > > > > > > > > > >> seems
> > > > > > > > > > > >> > an
> > > > > > > > > > > >> > > > > >> overkill
> > > > > > > > > > > >> > > > > >> > > here
> > > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it
> elsewhere (we
> > > > only
> > > > > > have
> > > > > > > > > Read /
> > > > > > > > > > > >> > Delete
> > > > > > > > > > > >> > > > > and
> > > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to
> hear
> > > > others
> > > > > > > > > thoughts.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> > > > recreation
> > > > > > > > > handled by
> > > > > > > > > > > >> the
> > > > > > > > > > > >> > > > > consumer
> > > > > > > > > > > >> > > > > >> > with
> > > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would
> be
> > > > good
> > > > > > to have
> > > > > > > > > a
> > > > > > > > > > > >> > section on
> > > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > > > > > subscription
> > > > > > > > > right? Yes
> > > > > > > > > > > >> > we can
> > > > > > > > > > > >> > > > > >> add a
> > > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically
> the
> > > > idea
> > > > > > is that
> > > > > > > > > > > >> consumer
> > > > > > > > > > > >> > would
> > > > > > > > > > > >> > > > > be
> > > > > > > > > > > >> > > > > >> > > totally
> > > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as
> it's
> > > > > > handled all
> > > > > > > > > by the
> > > > > > > > > > > >> > brokers.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP
> mentions we
> > > > will
> > > > > > write
> > > > > > > > > the
> > > > > > > > > > > >> new
> > > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> implementation
> > > > > > detail,
> > > > > > > > > do we
> > > > > > > > > > > >> > plan to
> > > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought
> about
> > > > > > that. But
> > > > > > > > > I think
> > > > > > > > > > > >> > the
> > > > > > > > > > > >> > > > > answer
> > > > > > > > > > > >> > > > > >> > > should
> > > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> > > > scalability
> > > > > > goal
> > > > > > > > > when it
> > > > > > > > > > > >> > comes to
> > > > > > > > > > > >> > > > > >> how
> > > > > > > > > > > >> > > > > >> > > many
> > > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> > > > support?
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> think we
> > > > > > should
> > > > > > > > > shoot for
> > > > > > > > > > > >> > 1000s of
> > > > > > > > > > > >> > > > > >> > > members.
> > > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals
> also
> > > > depend
> > > > > > on the
> > > > > > > > > offset
> > > > > > > > > > > >> > > > > management
> > > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> > > > coordinator
> > > > > > > > > which we
> > > > > > > > > > > >> did
> > > > > > > > > > > >> > not
> > > > > > > > > > > >> > > > > >> cover in
> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a
> number that
> > > > > > applies
> > > > > > > > > > > >> > universally.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider
> > > > having
> > > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> SubscribedTopicNames in
> > > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
> resolve
> > > > the
> > > > > > regex
> > > > > > > > > on the
> > > > > > > > > > > >> > server,
> > > > > > > > > > > >> > > > > we
> > > > > > > > > > > >> > > > > >> > can
> > > > > > > > > > > >> > > > > >> > > do
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > > > > > difference is
> > > > > > > > > that
> > > > > > > > > > > >> > sending the
> > > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > > >> > > > > >> > > is
> > > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending
> the topic
> > > > > > names is
> > > > > > > > > less
> > > > > > > > > > > >> > efficient.
> > > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is
> easier to
> > > > > > handle if
> > > > > > > > > we have
> > > > > > > > > > > >> > topic
> > > > > > > > > > > >> > > > > ids.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
> still
> > > > let
> > > > > > the
> > > > > > > > > clients
> > > > > > > > > > > >> send
> > > > > > > > > > > >> > names
> > > > > > > > > > > >> > > > > >> is to
> > > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names
> -> ids
> > > > on
> > > > > > the
> > > > > > > > > broker /
> > > > > > > > > > > >> > admin
> > > > > > > > > > > >> > > > > client
> > > > > > > > > > > >> > > > > >> > > only.
> > > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> topic id
> > > > in
> > > > > > > > > KIP-516, we
> > > > > > > > > > > >> > never
> > > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> > > > leveraging
> > > > > > the
> > > > > > > > > related
> > > > > > > > > > > >> > newer
> > > > > > > > > > > >> > > > > >> versioned
> > > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic
> id as
> > > > empty
> > > > > > UUID.
> > > > > > > > > We
> > > > > > > > > > > >> want
> > > > > > > > > > > >> > to keep
> > > > > > > > > > > >> > > > > >> the
> > > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> > > > delegate
> > > > > > the
> > > > > > > > > reasoning
> > > > > > > > > > > >> on
> > > > > > > > > > > >> > broker
> > > > > > > > > > > >> > > > > >> and
> > > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6,
> 2022 at
> > > > 10:45
> > > > > > AM
> > > > > > > > > David
> > > > > > > > > > > >> Jacot
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> <dj...@confluent.io.invalid>
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to
> start a
> > > > > > > > > discussion
> > > > > > > > > > > >> thread
> > > > > > > > > > > >> > on
> > > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the
> > > > Consumer
> > > > > > > > > Rebalance
> > > > > > > > > > > >> > Protocol.
> > > > > > > > > > > >> > > > > >> With
> > > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> rebalance
> > > > > > protocol
> > > > > > > > > (for
> > > > > > > > > > > >> > consumers)
> > > > > > > > > > > >> > > > > >> more
> > > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier
> to
> > > > > > implement for
> > > > > > > > > > > >> > clients, and
> > > > > > > > > > > >> > > > > >> easier
> > > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a
> look and
> > > > let
> > > > > > me
> > > > > > > > > know what
> > > > > > > > > > > >> > you
> > > > > > > > > > > >> > > > > think.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be
> away from
> > > > > > July
> > > > > > > > > 18th to
> > > > > > > > > > > >> > August 8th.
> > > > > > > > > > > >> > > > > >> That
> > > > > > > > > > > >> > > > > >> > > gives
> > > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read
> and
> > > > digest
> > > > > > this
> > > > > > > > > long
> > > > > > > > > > > >> KIP.
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > > >> > > > > >> >
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > >> --
> > > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > > > > >> > > > > >>
> > > > > > > > > > > >> > > > > >
> > > > > > > > > > > >> > > > >
> > > > > > > > > > > >> >
> > > > > > > > > > > >>
> > > > > > > > > > > >
> > > > > > > > >
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > > -- Guozhang
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi all,

During an offline conversation, someone asked why we need the
ConsumerGroupPrepareAssignment API and suggested that we could instead
provide the group state in the heartbeat response. This has a few
advantages: 1) it does not require using a special error code to
signal that a new assignment is required as the signal would be the
provided group state; 2) it removes one round trip when a client side
assignor is used. The downside is that it makes the heartbeat
response's definition quite large. I recall that I went with the
current approach due to this.

Providing the group state in the heartbeat response is appealing. What
do you guys think?

Best,
David

On Mon, Sep 12, 2022 at 2:17 PM David Jacot <dj...@confluent.io> wrote:
>
> Hi Guozhang,
>
> 1. I have added a reference to the relevant chapter instead of
> repeating the whole thing. Does that work for you?
>
> 2. The "Rebalance Triggers" section you are referring to is about when
> a rebalance should be triggered for the non-upgraded members using the
> old protocol. The section mentions that a rebalance must be triggered
> when a new assignment is installed. This implies that the group epoch
> was updated either by a native member or a non-upgraded member. For
> the latter, the JoinGroup request would be the trigger. I have added a
> reference to the relevant chapter in the "JoinGroup Handling" section
> as well. Does that make sense?
>
> Thanks,
> David
>
> On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello David,
> >
> > Alright I think that's sufficient. Just to make that clear in the doc,
> > could we update:
> >
> > 1) the heartbeat request handling section, stating when coordinator will
> > trigger rebalance based on the HB's member metadata / reason?
> > 2) the "Rebalance Triggers" section to include what we described in "Group
> > Epoch - Trigger a rebalance" section as well?
> >
> >
> > Guozhang
> >
> > On Fri, Sep 9, 2022 at 1:28 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > I thought that the assignor will always be consulted when the next
> > > heartbeat request is constructed. In other words,
> > > `PartitionAssignor#metadata` will be called for every heartbeat. This
> > > gives the opportunity for the assignor to enforce a rebalance by
> > > setting the reason to a non-zero value or by changing the bytes. Do
> > > you think that this is not sufficient? Are you concerned by the delay?
> > >
> > > Best,
> > > David
> > >
> > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com> wrote:
> > > >
> > > > Hello David,
> > > >
> > > > One of Jun's comments make me thinking:
> > > >
> > > > ```
> > > > In this case, a new assignment is triggered by the client side
> > > > assignor. When constructing the HB, the consumer will always consult
> > > > the client side assignor and propagate the information to the group
> > > > coordinator. In other words, we don't expect users to call
> > > > Consumer#enforceRebalance anymore.
> > > > ```
> > > >
> > > > As I looked at the current PartitionAssignor's interface, we actually do
> > > > not have a way yet to instruct how to construct the next HB request, e.g.
> > > > when the assignor wants to enforce a new rebalance with a new assignment,
> > > > we'd need some customizable APIs inside the PartitionAssignor to indicate
> > > > the next HB telling broker about so. WDYT about adding such an API on the
> > > > PartitionAssignor?
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi Jun,
> > > > >
> > > > > I have updated the KIP to include your feedback. I have also tried to
> > > > > clarify the parts which were not cleared.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > > > >
> > > > > > Hi Jun,
> > > > > >
> > > > > > Thanks for your feedback. Let me start by answering your questions
> > > > > > inline and I will update the KIP next week.
> > > > > >
> > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to
> > > be
> > > > > fewer
> > > > > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > > > > comments below.
> > > > > >
> > > > > > I would also add that the KIP removes the global sync barrier in the
> > > > > > protocol which is essential to improve group stability and
> > > > > > scalability, and the KIP also simplifies the client by moving most of
> > > > > > the logic to the server side.
> > > > > >
> > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > > > > changing
> > > > > > > of the partition assignor in the consumers?
> > > > > >
> > > > > > Definitely. The group coordinator will use the assignor used by a
> > > > > > majority of the members. This allows the group to move from one
> > > > > > assignor to another by a roll. This is explained in the Assignor
> > > > > > Selection chapter.
> > > > > >
> > > > > > > 30.2 For each field, could you explain whether it's required in
> > > every
> > > > > > > request or the scenarios when it needs to be filled? For example,
> > > it's
> > > > > not
> > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >
> > > > > > The client is expected to set those fields in case of a connection
> > > > > > issue (e.g. timeout) or when the fields have changed since the last
> > > > > > HB. The server populates those fields as long as the member is not
> > > > > > fully reconciled - the member should acknowledge that it has the
> > > > > > expected epoch and assignment. I will clarify this in the KIP.
> > > > > >
> > > > > > > 31. In the current consumer protocol, the rack affinity between the
> > > > > client
> > > > > > > and the broker is only considered during fetching, but not during
> > > > > assigning
> > > > > > > partitions to consumers. Sometimes, once the assignment is made,
> > > there
> > > > > is
> > > > > > > no opportunity for read affinity because no replicas of assigned
> > > > > partitions
> > > > > > > are close to the member. I am wondering if we should use this
> > > > > opportunity
> > > > > > > to address this by including rack in GroupMember.
> > > > > >
> > > > > > That's an interesting idea. I don't see any issue with adding the
> > > rack
> > > > > > to the members. I will do so.
> > > > > >
> > > > > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > > > > coordinator is. By moving the event loop model, it seems that we
> > > could
> > > > > add
> > > > > > > a metric that tracks the fraction of the time the event loop is
> > > doing
> > > > > the
> > > > > > > actual work.
> > > > > >
> > > > > > That's a great idea. I will add it. Thanks.
> > > > > >
> > > > > > > 33. Could we add a section on coordinator failover handling? For
> > > > > example,
> > > > > > > does it need to trigger the check if any group with the wildcard
> > > > > > > subscription now has a new matching topic?
> > > > > >
> > > > > > Sure. When the new group coordinator takes over, it has to:
> > > > > > * Setup the session timeouts.
> > > > > > * Trigger a new assignment if a client side assignor is used. We
> > > don't
> > > > > > store the information about the member selected to run the assignment
> > > > > > so we have to start a new one.
> > > > > > * Update the topics metadata, verify the wildcard subscriptions, and
> > > > > > trigger a rebalance if needed.
> > > > > >
> > > > > > > 34. ConsumerGroupMetadataValue,
> > > ConsumerGroupPartitionMetadataValue,
> > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > > > > field
> > > > > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > > > > reflect
> > > > > > > the latest group epoch? What about the one in
> > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > ConsumerGroupMemberMetadataValue?
> > > > > >
> > > > > > Sure. I will clarify that but it is always the latest group epoch.
> > > > > > When the group state is updated, the group epoch is bumped so we use
> > > > > > that one for all the change records related to the update.
> > > > > >
> > > > > > > 35. "the group coordinator will ensure that the following
> > > invariants
> > > > > are
> > > > > > > met: ... All members exists." It's possible for a member not to
> > > get any
> > > > > > > assigned partitions, right?
> > > > > >
> > > > > > That's right. Here I meant that the members provided by the assignor
> > > > > > in the assignment must exist in the group. The assignor can not make
> > > > > > up new member ids.
> > > > > >
> > > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > > When
> > > > > would
> > > > > > > a consumer rejoin and what member id would be used?
> > > > > >
> > > > > > A member is expected to abandon all its partitions and rejoins when
> > > it
> > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > > > > coordinator will have removed the member from the group. The member
> > > > > > can rejoin the group with the same member id but with 0 as epoch. Let
> > > > > > me see if I can clarify this in the KIP.
> > > > > >
> > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > reassignment
> > > > > > > by either providing a non-zero reason or by updating the assignor
> > > > > > > metadata." Hmm, this seems to be conflicting with the deprecation
> > > of
> > > > > > > Consumer#enforeRebalance.
> > > > > >
> > > > > > In this case, a new assignment is triggered by the client side
> > > > > > assignor. When constructing the HB, the consumer will always consult
> > > > > > the client side assignor and propagate the information to the group
> > > > > > coordinator. In other words, we don't expect users to call
> > > > > > Consumer#enforceRebalance anymore.
> > > > > >
> > > > > > > 38. The reassignment examples are nice. But the section seems to
> > > have
> > > > > > > multiple typos.
> > > > > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > > epoch=3,
> > > > > > > partitions=[foo-1] too early.
> > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > partitions=[foo-0].
> > > > > >
> > > > > > Sorry for that! I will revise them.
> > > > > >
> > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade from
> > > any
> > > > > old
> > > > > > > version to new one?
> > > > > >
> > > > > > We will support upgrading from the consumer protocol version 3,
> > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the earliest
> > > > > > version is unknown at the moment. This is explained in the migration
> > > > > > plan chapter.
> > > > > >
> > > > > > Thanks again for your feedback, Jun. I will update the KIP based on
> > > it
> > > > > > next week.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <ju...@confluent.io.invalid>
> > > wrote:
> > > > > > >
> > > > > > > Hi, David,
> > > > > > >
> > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to
> > > be
> > > > > fewer
> > > > > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > > > > comments below.
> > > > > > >
> > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > > > > changing
> > > > > > > of the partition assignor in the consumers?
> > > > > > > 30.2 For each field, could you explain whether it's required in
> > > every
> > > > > > > request or the scenarios when it needs to be filled? For example,
> > > it's
> > > > > not
> > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > > >
> > > > > > > 31. In the current consumer protocol, the rack affinity between the
> > > > > client
> > > > > > > and the broker is only considered during fetching, but not during
> > > > > assigning
> > > > > > > partitions to consumers. Sometimes, once the assignment is made,
> > > there
> > > > > is
> > > > > > > no opportunity for read affinity because no replicas of assigned
> > > > > partitions
> > > > > > > are close to the member. I am wondering if we should use this
> > > > > opportunity
> > > > > > > to address this by including rack in GroupMember.
> > > > > > >
> > > > > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > > > > coordinator is. By moving the event loop model, it seems that we
> > > could
> > > > > add
> > > > > > > a metric that tracks the fraction of the time the event loop is
> > > doing
> > > > > the
> > > > > > > actual work.
> > > > > > >
> > > > > > > 33. Could we add a section on coordinator failover handling? For
> > > > > example,
> > > > > > > does it need to trigger the check if any group with the wildcard
> > > > > > > subscription now has a new matching topic?
> > > > > > >
> > > > > > > 34. ConsumerGroupMetadataValue,
> > > ConsumerGroupPartitionMetadataValue,
> > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > > > > field
> > > > > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > > > > reflect
> > > > > > > the latest group epoch? What about the one in
> > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >
> > > > > > > 35. "the group coordinator will ensure that the following
> > > invariants
> > > > > are
> > > > > > > met: ... All members exists." It's possible for a member not to
> > > get any
> > > > > > > assigned partitions, right?
> > > > > > >
> > > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > > When
> > > > > would
> > > > > > > a consumer rejoin and what member id would be used?
> > > > > > >
> > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > reassignment
> > > > > > > by either providing a non-zero reason or by updating the assignor
> > > > > > > metadata." Hmm, this seems to be conflicting with the deprecation
> > > of
> > > > > > > Consumer#enforeRebalance.
> > > > > > >
> > > > > > > 38. The reassignment examples are nice. But the section seems to
> > > have
> > > > > > > multiple typos.
> > > > > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > > epoch=3,
> > > > > > > partitions=[foo-1] too early.
> > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > partitions=[foo-0].
> > > > > > >
> > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade from
> > > any
> > > > > old
> > > > > > > version to new one?
> > > > > > >
> > > > > > > Thanks,
> > > > > > >
> > > > > > > Jun
> > > > > > >
> > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hi all,
> > > > > > > >
> > > > > > > > The KIP states that we will re-implement the coordinator in
> > > Java. I
> > > > > > > > discussed this offline with a few folks and folks are concerned
> > > that
> > > > > > > > we could introduce many regressions in the old protocol if we do
> > > so.
> > > > > > > > Therefore, I am going to remove this statement from the KIP. It
> > > is an
> > > > > > > > implementation detail after all so it does not have to be
> > > decided at
> > > > > > > > this stage. We will likely start by trying to refactor the
> > > current
> > > > > > > > implementation as a first step.
> > > > > > > >
> > > > > > > > Cheers,
> > > > > > > > David
> > > > > > > >
> > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <djacot@confluent.io
> > > >
> > > > > wrote:
> > > > > > > > >
> > > > > > > > > Hi Luke,
> > > > > > > > >
> > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > reconciling,
> > > > > > > > stable,
> > > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > >
> > > > > > > > > This sentence does not refer to those group states but rather
> > > to a
> > > > > > > > > state machine replication (SMR). This refers to the entire
> > > state of
> > > > > > > > > group coordinator which is replicated via the log layer. I will
> > > > > > > > > clarify this in the KIP.
> > > > > > > > >
> > > > > > > > > > 1.2. What do you mean "each state machine is modelled as an
> > > event
> > > > > > > > loop"?
> > > > > > > > >
> > > > > > > > > The idea is to follow a model similar to the new quorum
> > > > > controller. We
> > > > > > > > > will have N threads to process events. Each __consumer_offsets
> > > > > > > > > partition is assigned to a unique thread and all the events
> > > (e.g.
> > > > > > > > > requests, callbacks, etc.) are processed by this thread. This
> > > > > simplify
> > > > > > > > > concurrency and will enable us to do simulation testing for the
> > > > > group
> > > > > > > > > coordinator.
> > > > > > > > >
> > > > > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > > > > partitions?
> > > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > > coordinator?
> > > > > > > > For
> > > > > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > > > > both
> > > > > > > > exist in
> > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it,
> > > or
> > > > > 2?
> > > > > > > > >
> > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > > > > > > >
> > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number
> > > of
> > > > > threads
> > > > > > > > used
> > > > > > > > > > to run the state machines. But I'm wondering if the purpose
> > > of
> > > > > the
> > > > > > > > threads
> > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > *__consumer_offsets*
> > > > > > > > > > partitions?), and no heavy computation, why should we need
> > > > > > > > multi-threads
> > > > > > > > > > here?
> > > > > > > > >
> > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > > processing as
> > > > > the
> > > > > > > > > computation could be heavy.
> > > > > > > > >
> > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > session
> > > > > > > > timeout not
> > > > > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > > > > session
> > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > >
> > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > > detect
> > > > > client
> > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > >
> > > > > > > > > This is indeed a mistake. The default session timeout should
> > > be 45s
> > > > > > > > > (the current default).
> > > > > > > > >
> > > > > > > > > > 2.2. The default server side assignor are [range, uniform],
> > > > > which means
> > > > > > > > > > we'll default to "range" assignor. I'd like to know why not
> > > > > uniform
> > > > > > > > one? I
> > > > > > > > > > thought usually users will choose uniform assignor (former
> > > sticky
> > > > > > > > assinor)
> > > > > > > > > > for better evenly distribution. Any other reason we choose
> > > range
> > > > > > > > assignor
> > > > > > > > > > as default?
> > > > > > > > > > group.consumer.assignors List range, uniform The server side
> > > > > assignors.
> > > > > > > > >
> > > > > > > > > The order on the server side has no influence because the
> > > client
> > > > > must
> > > > > > > > > chose the selector that he wants to use. There is no default
> > > in the
> > > > > > > > > current proposal. If the assignor is not specified by the
> > > client,
> > > > > the
> > > > > > > > > request is rejected. The default client value for
> > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > > >
> > > > > > > > > Thanks for your very good comments, Luke. I hope that my
> > > answers
> > > > > help
> > > > > > > > > to clarify things. I will update the KIP as well based on your
> > > > > > > > > feedback.
> > > > > > > > >
> > > > > > > > > Cheers,
> > > > > > > > > David
> > > > > > > > >
> > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com>
> > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Hi David,
> > > > > > > > > >
> > > > > > > > > > Thanks for the update.
> > > > > > > > > >
> > > > > > > > > > Some more questions:
> > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > > > > The new group coordinator will have a state machine per
> > > > > > > > > > *__consumer_offsets* partitions, where each state machine is
> > > > > modelled
> > > > > > > > as an
> > > > > > > > > > event loop. Those state machines will be executed in
> > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > >
> > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > reconciling,
> > > > > > > > stable,
> > > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > > > 1.2. What do you mean "each state machine is modelled as an
> > > event
> > > > > > > > loop"?
> > > > > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > > > > partitions?
> > > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > > coordinator?
> > > > > > > > For
> > > > > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > > > > both
> > > > > > > > exist in
> > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it,
> > > or
> > > > > 2?
> > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number
> > > of
> > > > > threads
> > > > > > > > used
> > > > > > > > > > to run the state machines. But I'm wondering if the purpose
> > > of
> > > > > the
> > > > > > > > threads
> > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > *__consumer_offsets*
> > > > > > > > > > partitions?), and no heavy computation, why should we need
> > > > > > > > multi-threads
> > > > > > > > > > here?
> > > > > > > > > >
> > > > > > > > > > 2. For the default value in the new configs:
> > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > session
> > > > > > > > timeout not
> > > > > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > > > > session
> > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > >
> > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > > detect
> > > > > client
> > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > 2.2. The default server side assignor are [range, uniform],
> > > > > which means
> > > > > > > > > > we'll default to "range" assignor. I'd like to know why not
> > > > > uniform
> > > > > > > > one? I
> > > > > > > > > > thought usually users will choose uniform assignor (former
> > > sticky
> > > > > > > > assinor)
> > > > > > > > > > for better evenly distribution. Any other reason we choose
> > > range
> > > > > > > > assignor
> > > > > > > > > > as default?
> > > > > > > > > > group.consumer.assignors List range, uniform The server side
> > > > > assignors.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Thank you.
> > > > > > > > > > Luke
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <showuon@gmail.com
> > > >
> > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hi Sagar,
> > > > > > > > > > >
> > > > > > > > > > > I have some thoughts about Kafka Connect integrating with
> > > > > KIP-848,
> > > > > > > > but I
> > > > > > > > > > > think we should have a separate discussion thread for the
> > > Kafka
> > > > > > > > Connect
> > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer Rebalance
> > > > > Protocol
> > > > > > > > [1],
> > > > > > > > > > > and let this discussion thread focus on consumer rebalance
> > > > > protocol,
> > > > > > > > WDYT?
> > > > > > > > > > >
> > > > > > > > > > > [1]
> > > > > > > > > > >
> > > > > > > >
> > > > >
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >
> > > > > > > > > > > Thank you.
> > > > > > > > > > > Luke
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > sagarmeansocean@gmail.com>
> > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks like
> > > there's
> > > > > > > > agreement on
> > > > > > > > > > >> using separate APIs for Connect. I would revisit the doc
> > > and
> > > > > see
> > > > > > > > what
> > > > > > > > > > >> changes are to be made.
> > > > > > > > > > >>
> > > > > > > > > > >> Thanks!
> > > > > > > > > > >> Sagar.
> > > > > > > > > > >>
> > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> wrote:
> > > > > > > > > > >>
> > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > >> >
> > > > > > > > > > >> > Thanks for the feedback and the document. That's really
> > > > > helpful. I
> > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > >> >
> > > > > > > > > > >> > Overall, it seems to me that both Connect and the
> > > Consumer
> > > > > could
> > > > > > > > share
> > > > > > > > > > >> > the same underlying "engine". The main difference is
> > > that
> > > > > the
> > > > > > > > Consumer
> > > > > > > > > > >> > assigns topic-partitions to members whereas Connect
> > > assigns
> > > > > tasks
> > > > > > > > to
> > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > > > > >> > 1) We extend the new proposed APIs to support different
> > > > > resource
> > > > > > > > types
> > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The dedicated
> > > APIs
> > > > > > > > would be
> > > > > > > > > > >> > similar to the new ones but different on the
> > > > > content/resources and
> > > > > > > > > > >> > they would rely on the same engine on the coordinator
> > > side.
> > > > > > > > > > >> >
> > > > > > > > > > >> > I personally lean towards 2) because I am not a fan of
> > > > > > > > overcharging
> > > > > > > > > > >> > APIs to serve different purposes. That being said, I am
> > > not
> > > > > > > > opposed to
> > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > > > > >> >
> > > > > > > > > > >> > I think that we can continue to discuss it here for now
> > > in
> > > > > order
> > > > > > > > to
> > > > > > > > > > >> > ensure that this KIP is compatible with what we will do
> > > for
> > > > > > > > Connect in
> > > > > > > > > > >> > the future.
> > > > > > > > > > >> >
> > > > > > > > > > >> > Best,
> > > > > > > > > > >> > David
> > > > > > > > > > >> >
> > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > > > djacot@confluent.io>
> > > > > > > > wrote:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Hi all,
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > I am back from vacation. I will go through and address
> > > > > your
> > > > > > > > comments
> > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Cheers,
> > > > > > > > > > >> > > David
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > > > > gharris1727@gmail.com
> > > > > > > > > > >> >
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > > cooperative
> > > > > > > > rebalancing
> > > > > > > > > > >> > making it
> > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section describes how
> > > > > member
> > > > > > > > epoch
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > advance in step with the group epoch and assignment
> > > > > epoch
> > > > > > > > values. I
> > > > > > > > > > >> > think
> > > > > > > > > > >> > > > that this is a good idea for the reasons described
> > > in
> > > > > the
> > > > > > > > KIP. When
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > protocol is incrementally assigning partitions to a
> > > > > worker,
> > > > > > > > what
> > > > > > > > > > >> member
> > > > > > > > > > >> > > > epoch does each incremental assignment use? Are
> > > member
> > > > > epochs
> > > > > > > > > > >> re-used,
> > > > > > > > > > >> > and
> > > > > > > > > > >> > > > a single member epoch can correspond to multiple
> > > > > different
> > > > > > > > > > >> > (monotonically
> > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the
> > > group
> > > > > > > > > > >> coordinator? If
> > > > > > > > > > >> > > > not, should custom client-side assignor
> > > implementations
> > > > > > > > interact
> > > > > > > > > > >> with
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > Reason field, and how is its common meaning agreed
> > > > > upon? If
> > > > > > > > so, what
> > > > > > > > > > >> > is the
> > > > > > > > > > >> > > > benefit of a distinct Reason field over including
> > > such
> > > > > > > > functionality
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 3. The following is included in the KIP: "Thanks to
> > > > > this, the
> > > > > > > > input
> > > > > > > > > > >> of
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > client side assignor is entirely driven by the group
> > > > > > > > coordinator.
> > > > > > > > > > >> The
> > > > > > > > > > >> > > > consumer is no longer responsible for maintaining
> > > any
> > > > > state
> > > > > > > > besides
> > > > > > > > > > >> its
> > > > > > > > > > >> > > > assigned partitions." Does this mean that the
> > > > > client-side
> > > > > > > > assignor
> > > > > > > > > > >> MAY
> > > > > > > > > > >> > > > incorporate additional non-Metadata state (such as
> > > > > partition
> > > > > > > > > > >> > throughput,
> > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or that
> > > > > additional
> > > > > > > > > > >> > non-Metadata
> > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > > > > >> > > > for
> > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > >> > > > and
> > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > that
> > > > > > > > seem to
> > > > > > > > > > >> > > > overlap significantly. Is it possible for these two
> > > > > > > > implementations
> > > > > > > > > > >> to
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > unified? This would serve to promote feature parity
> > > of
> > > > > > > > server-side
> > > > > > > > > > >> and
> > > > > > > > > > >> > > > client-side assignors, and would also facilitate
> > > > > operational
> > > > > > > > > > >> > flexibility in
> > > > > > > > > > >> > > > certain situations. For example, if a server-side
> > > > > assignor
> > > > > > > > has some
> > > > > > > > > > >> > poor
> > > > > > > > > > >> > > > behavior and needs a patch, deploying the patched
> > > > > assignor to
> > > > > > > > the
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > and switching one consumer group to a client-side
> > > > > assignor
> > > > > > > > may be
> > > > > > > > > > >> > faster
> > > > > > > > > > >> > > > and less risky than patching all of the brokers.
> > > With
> > > > > the
> > > > > > > > currently
> > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > reimplementation
> > > > > would
> > > > > > > > have
> > > > > > > > > > >> to be
> > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> > > > > significantly,
> > > > > > > > then it
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > possible that a reimplementation would not be
> > > possible.
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > --
> > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > I created a confluence page to discuss how Connect
> > > > > would
> > > > > > > > need to
> > > > > > > > > > >> > change
> > > > > > > > > > >> > > > > based on the new rebalance protocol. Here's the
> > > page:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > >
> > > > >
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > It's also pretty longish and I have tried to keep
> > > a
> > > > > format
> > > > > > > > > > >> similar to
> > > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also, do you
> > > > > think this
> > > > > > > > > > >> should
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > moved to a separate discussion thread or is this
> > > one
> > > > > fine?
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka Streams.
> > > > > Sure, I
> > > > > > > > would do
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> It would be great if you could come back with
> > > some
> > > > > > > > analysis on
> > > > > > > > > > >> > how to
> > > > > > > > > > >> > > > > >> implement the Connect side integration with
> > > the new
> > > > > > > > protocol;
> > > > > > > > > > >> so
> > > > > > > > > > >> > far
> > > > > > > > > > >> > > > > >> besides leveraging on the new "protocol type"
> > > we
> > > > > did not
> > > > > > > > yet
> > > > > > > > > > >> think
> > > > > > > > > > >> > > > > through
> > > > > > > > > > >> > > > > >> the Connect side implementations. For Streams
> > > > > here's a
> > > > > > > > draft of
> > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > >
> > > > >
> > > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > Thank you for your response. The reason I
> > > thought
> > > > > > > > connect can
> > > > > > > > > > >> > also fit
> > > > > > > > > > >> > > > > >> into
> > > > > > > > > > >> > > > > >> > this new scheme is that even today the
> > > connect
> > > > > uses a
> > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to empower
> > > > > > > > rebalances of
> > > > > > > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets
> > > the
> > > > > > > > > > >> protocolType()
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > >> > > > > >> > and uses the metadata() method by plumbing
> > > into
> > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > I think the changes to support connect would
> > > be
> > > > > > > > similar at a
> > > > > > > > > > >> > high
> > > > > > > > > > >> > > > > level
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > the changes in streams mainly because of the
> > > > > Client
> > > > > > > > side
> > > > > > > > > > >> > assignors
> > > > > > > > > > >> > > > > being
> > > > > > > > > > >> > > > > >> > used in both. At an implementation level, we
> > > > > might
> > > > > > > > need to
> > > > > > > > > > >> make
> > > > > > > > > > >> > a lot
> > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > changes to get onto this new assignment
> > > protocol
> > > > > like
> > > > > > > > > > >> enhancing
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > JoinGroup request/response and SyncGroup and
> > > > > using
> > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > >> > API etc again on similar lines to streams (or
> > > > > there
> > > > > > > > might be
> > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > >> > > > > >> > would try to perform a detailed analysis of
> > > the
> > > > > same
> > > > > > > > and we
> > > > > > > > > > >> > can have
> > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > >> > separate discussion thread for that as that
> > > would
> > > > > > > > derail this
> > > > > > > > > > >> > > > > discussion
> > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds good to
> > > you.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`.
> > > > > Sure, I
> > > > > > > > can
> > > > > > > > > > >> > mention it.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from his
> > > current
> > > > > > > > assignment
> > > > > > > > > > >> to
> > > > > > > > > > >> > his
> > > > > > > > > > >> > > > > >> > > target assignment when he can move to
> > > epoch 3.
> > > > > When
> > > > > > > > that
> > > > > > > > > > >> > happens,
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > member assignment is updated and persisted
> > > > > with all
> > > > > > > > its
> > > > > > > > > > >> > assigned
> > > > > > > > > > >> > > > > >> > > partitions even if they are not all revoked
> > > > > yet. In
> > > > > > > > other
> > > > > > > > > > >> > words, the
> > > > > > > > > > >> > > > > >> > > member assignment becomes the target
> > > > > assignment.
> > > > > > > > This is
> > > > > > > > > > >> > basically
> > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > >> > > optimization to avoid having to write all
> > > the
> > > > > > > > changes to
> > > > > > > > > > >> the
> > > > > > > > > > >> > log.
> > > > > > > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > examples are based on the persisted state
> > > so I
> > > > > > > > understand
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can improve
> > > this in
> > > > > the
> > > > > > > > > > >> > description.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse the
> > > > > protocol
> > > > > > > > with a
> > > > > > > > > > >> > client side
> > > > > > > > > > >> > > > > >> > > assignor if it fits in the protocol. The
> > > > > assignment
> > > > > > > > is
> > > > > > > > > > >> about
> > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> > > Connect
> > > > > fit
> > > > > > > > into this?
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor
> > > > > observations:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error section in
> > > Client
> > > > > Side
> > > > > > > > mode
> > > > > > > > > > >> > Assignment
> > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case, the
> > > client
> > > > > side
> > > > > > > > assignor
> > > > > > > > > > >> can
> > > > > > > > > > >> > > > > return
> > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > >> > > > error to the group coordinator`. In this
> > > > > case are
> > > > > > > > you
> > > > > > > > > > >> > referring to
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > Assignor returning an AssignmentError
> > > that's
> > > > > > > > listed down
> > > > > > > > > > >> > towards
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > >> > > > > >> > > > If yes, do you think it would make sense
> > > to
> > > > > > > > mention this
> > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I have a
> > > > > slight
> > > > > > > > > > >> confusion,
> > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> if
> > > > > > > > > > >> > > > > >> > > > others have the same. Consider this step:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group coordinator
> > > > > > > > transitions him
> > > > > > > > > > >> to
> > > > > > > > > > >> > epoch
> > > > > > > > > > >> > > > > 3
> > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> > > persists
> > > > > the
> > > > > > > > change and
> > > > > > > > > > >> > reply.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > > > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0,
> > > > > foo-1]
> > > > > > > > > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > > > > > > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions to
> > > epoch 3
> > > > > but
> > > > > > > > cannot
> > > > > > > > > > >> get
> > > > > > > > > > >> > foo-1
> > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C can't
> > > get
> > > > > the
> > > > > > > > foo-1
> > > > > > > > > > >> > partition
> > > > > > > > > > >> > > > > yet,
> > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > >> > > > > >> > > > based on the description above, it seems
> > > it
> > > > > > > > already has
> > > > > > > > > > >> it.
> > > > > > > > > > >> > Do you
> > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > >> > > > > >> > > > would be better to remove it and
> > > populate it
> > > > > only
> > > > > > > > when it
> > > > > > > > > > >> > actually
> > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > >> > > > > >> > > > I see this in a lot of other places, so
> > > have
> > > > > I
> > > > > > > > > > >> understood it
> > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be out of
> > > scope
> > > > > of
> > > > > > > > this
> > > > > > > > > > >> > discussion,
> > > > > > > > > > >> > > > > but
> > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > >> > > > > >> > > > what I understood it would probably be
> > > > > running in
> > > > > > > > client
> > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > >> > > > > >> > > > even on the new rebalance protocol as it
> > > has
> > > > > its
> > > > > > > > own
> > > > > > > > > > >> Custom
> > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David
> > > Jacot
> > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move
> > > forward
> > > > > with
> > > > > > > > > > >> > specialized API
> > > > > > > > > > >> > > > > >> > > > > instead of relying on one generic API.
> > > For
> > > > > > > > Connect, we
> > > > > > > > > > >> > can apply
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > exact same pattern and reuse/share the
> > > core
> > > > > > > > > > >> > implementation on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > server side. For the schema registry, I
> > > > > think
> > > > > > > > that we
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > > > > > > membership/leader
> > > > > > > > > > >> > election.
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael
> > > > > Juma <
> > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern
> > > vs
> > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > document the differences in more
> > > detail
> > > > > before
> > > > > > > > > > >> deciding
> > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > way
> > > > > > > > > > >> > > > > >> or
> > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > >> they
> > > > > > > > > > >> > expect
> > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If we are
> > > doing
> > > > > > > > something
> > > > > > > > > > >> > different,
> > > > > > > > > > >> > > > > >> then
> > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > >> > > > > > should consider adding an overload
> > > with
> > > > > our own
> > > > > > > > > > >> Pattern
> > > > > > > > > > >> > class
> > > > > > > > > > >> > > > > (I
> > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose re2j's at
> > > this
> > > > > > > > point).
> > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new
> > > > > protocol
> > > > > > > > should
> > > > > > > > > > >> > > > > integrate
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > > > > recreation case
> > > > > > > > > > >> > correctly.
> > > > > > > > > > >> > > > > That's
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I agree with
> > > > > David
> > > > > > > > that we'd
> > > > > > > > > > >> > want to
> > > > > > > > > > >> > > > > add
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols that don't
> > > > > have it
> > > > > > > > yet and
> > > > > > > > > > >> > that we
> > > > > > > > > > >> > > > > can
> > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > >> > > > > >> > > > > > focus on the internals versus adding
> > > new
> > > > > APIs
> > > > > > > > to the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs is
> > > required
> > > > > for
> > > > > > > > > > >> reasonable
> > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about the
> > > > > coordinator
> > > > > > > > storing
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > > centralized
> > > > > in the
> > > > > > > > > > >> metadata
> > > > > > > > > > >> > log for
> > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > >> > > > > >> > > > > > reasons (auditability, visibility,
> > > > > consistency,
> > > > > > > > > > >> etc.).
> > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > >> > > > > >> > > > > > sure about automatically deleting
> > > > > configs in a
> > > > > > > > way
> > > > > > > > > > >> that
> > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > >> > > > > > recovered. A good property for modern
> > > > > systems
> > > > > > > > is to
> > > > > > > > > > >> > minimize
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David
> > > > > Jacot
> > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are
> > > below:
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > > > > the last
> > > > > > > > step
> > > > > > > > > > >> when
> > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in
> > > which
> > > > > we
> > > > > > > > would
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > protocols /
> > > > > rpcs
> > > > > > > > are
> > > > > > > > > > >> used
> > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > > > > "mimic"
> > > > > > > > the old
> > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > > need to
> > > > > > > > present
> > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a new
> > > version
> > > > > of KIP
> > > > > > > > which
> > > > > > > > > > >> > includes
> > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> > > "Supporting
> > > > > > > > Online
> > > > > > > > > > >> > Consumer
> > > > > > > > > > >> > > > > Group
> > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > > Deprecation,
> > > > > and
> > > > > > > > Migration
> > > > > > > > > > >> > Plan". I
> > > > > > > > > > >> > > > > >> think
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > I have to think through a few cases
> > > > > now but
> > > > > > > > the
> > > > > > > > > > >> > overall idea
> > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > > far as
> > > > > > > > KIP-516 the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client,
> > > but
> > > > > they
> > > > > > > > are not
> > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > > question is,
> > > > > > > > first
> > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > > names
> > > > > -> ids
> > > > > > > > mapping
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > > RPCs
> > > > > and the
> > > > > > > > new
> > > > > > > > > > >> RPCs
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > > > > exposing
> > > > > > > > the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > subscribe/assign, as
> > > > > > > > > > >> well
> > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would include
> > > > > converting
> > > > > > > > all
> > > > > > > > > > >> the
> > > > > > > > > > >> > offsets
> > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this
> > > KIP,
> > > > > the
> > > > > > > > consumer
> > > > > > > > > > >> > would be
> > > > > > > > > > >> > > > > >> able
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic ids. That
> > > > > being
> > > > > > > > said, it
> > > > > > > > > > >> > still has
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > >> > > > > >> > > > > > > the topics names in various APIs so
> > > > > having a
> > > > > > > > > > >> mapping
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong opinion on
> > > > > this.
> > > > > > > > Here I
> > > > > > > > > > >> > wonder if
> > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I
> > > would
> > > > > rather
> > > > > > > > focus
> > > > > > > > > > >> on
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > > > > separately if
> > > > > > > > we see
> > > > > > > > > > >> > value in
> > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's point about
> > > > > using
> > > > > > > > topic ids
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I
> > > think
> > > > > that
> > > > > > > > there
> > > > > > > > > > >> is
> > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer will
> > > have
> > > > > the
> > > > > > > > > > >> opportunity
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > the topics exists before passing
> > > them
> > > > > into
> > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > > coordinator
> > > > > will
> > > > > > > > also
> > > > > > > > > > >> notice
> > > > > > > > > > >> > it but
> > > > > > > > > > >> > > > > >> it
> > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > >> > > > > >> > > > > > > not really have a way to reject an
> > > > > invalid
> > > > > > > > topic in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all
> > > other
> > > > > minor
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > > > > personally
> > > > > > > > I think
> > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > > > > Pattern,
> > > > > > > > but
> > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > > > > subscription may
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > > > > very rare
> > > > > > > > > > >> scenario
> > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > subscription may
> > > > > > > > be
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to look at
> > > the
> > > > > > > > differences
> > > > > > > > > > >> > between
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > >> > > > > >> > > > > > > engines to better understand the
> > > > > potential
> > > > > > > > issues.
> > > > > > > > > > >> My
> > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > >> > > > > >> > > > > > > that would work for all the basic
> > > > > regular
> > > > > > > > > > >> > expressions. The
> > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly about
> > > the
> > > > > various
> > > > > > > > > > >> character
> > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > >> > > > > >> > > > > > > wonder what other people think
> > > about
> > > > > this.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM
> > > > > Guozhang
> > > > > > > > Wang <
> > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on the high
> > > > > level
> > > > > > > > there are
> > > > > > > > > > >> > two meta
> > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > > > > the last
> > > > > > > > step
> > > > > > > > > > >> when
> > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in
> > > which
> > > > > we
> > > > > > > > would
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > protocols /
> > > > > rpcs
> > > > > > > > are
> > > > > > > > > > >> used
> > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > > > > "mimic"
> > > > > > > > the old
> > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > > need to
> > > > > > > > present
> > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > > far as
> > > > > > > > KIP-516 the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client,
> > > but
> > > > > they
> > > > > > > > are not
> > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > > question is,
> > > > > > > > first
> > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > > names
> > > > > -> ids
> > > > > > > > mapping
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > > RPCs
> > > > > and the
> > > > > > > > new
> > > > > > > > > > >> RPCs
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > > > > exposing
> > > > > > > > the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > subscribe/assign, as
> > > > > > > > > > >> well
> > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all
> > > other
> > > > > minor
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > > > > personally
> > > > > > > > I think
> > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > > > > Pattern,
> > > > > > > > but
> > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > > > > subscription may
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > > > > very rare
> > > > > > > > > > >> scenario
> > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > subscription may
> > > > > > > > be
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM
> > > > > David Jacot
> > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let
> > > me
> > > > > answer
> > > > > > > > your
> > > > > > > > > > >> > questions
> > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's premature to
> > > > > talk about
> > > > > > > > > > >> target
> > > > > > > > > > >> > > > > versions
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the existing group
> > > > > protocol.
> > > > > > > > > > >> Unlike
> > > > > > > > > > >> > KRaft,
> > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > > > > > > deprecation/removal
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly applications
> > > migrate
> > > > > to
> > > > > > > > the new
> > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> > > remove it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to
> > > > > release
> > > > > > > > this in
> > > > > > > > > > >> > 4.x, but
> > > > > > > > > > >> > > > > it
> > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that as a
> > > way to
> > > > > > > > estimate
> > > > > > > > > > >> when
> > > > > > > > > > >> > we'd
> > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I also
> > > > > suggest
> > > > > > > > removing
> > > > > > > > > > >> > this part.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> > > > > > > > explained, I
> > > > > > > > > > >> plan
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java while we
> > > > > implement
> > > > > > > > the new
> > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> This
> > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > >> > > > > > > > > that the internals will be
> > > slightly
> > > > > > > > different
> > > > > > > > > > >> > (e.g.
> > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten
> > > the
> > > > > > > > switch from
> > > > > > > > > > >> > the old
> > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new group
> > > > > coordinator
> > > > > > > > to a
> > > > > > > > > > >> > major
> > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to use a
> > > flag
> > > > > to do
> > > > > > > > the
> > > > > > > > > > >> > switch
> > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the
> > > > > details of
> > > > > > > > the
> > > > > > > > > > >> > migration
> > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will support
> > > > > online
> > > > > > > > > > >> migrations.
> > > > > > > > > > >> > Is
> > > > > > > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in detail. It
> > > could
> > > > > also
> > > > > > > > be done
> > > > > > > > > > >> > as a
> > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support online
> > > > > migrations
> > > > > > > > for the
> > > > > > > > > > >> > group.
> > > > > > > > > > >> > > > > That
> > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old protocol
> > > > > will be
> > > > > > > > able to
> > > > > > > > > > >> > switch to
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain how that
> > > > > will work
> > > > > > > > > > >> though.
> > > > > > > > > > >> > It is
> > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> > > upgraded or
> > > > > > > > rolled to a
> > > > > > > > > > >> > software
> > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator. Both the
> > > > > old and
> > > > > > > > the new
> > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > >> > > > > >> > > > > > > > > support the old protocol and
> > > rely
> > > > > on the
> > > > > > > > same
> > > > > > > > > > >> > persisted
> > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > >> > > > > >> > > > > > > > > they can work together. This
> > > point
> > > > > is an
> > > > > > > > > > >> offline
> > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > >> > > > > >> > > > > > > > > do this one live because it
> > > would
> > > > > require
> > > > > > > > > > >> > shutting down
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and starting up the
> > > > > new one
> > > > > > > > and
> > > > > > > > > > >> that
> > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > > > > version/IBP
> > > > > > > > must be
> > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> X
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new protocol.
> > > This
> > > > > cannot
> > > > > > > > be done
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > 1)
> > > > > > > > > > >> > > > > >> is
> > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the old
> > > > > coordinator
> > > > > > > > doesn't
> > > > > > > > > > >> > support
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> > > upgraded
> > > > > to a
> > > > > > > > version
> > > > > > > > > > >> > > > > supporting
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have KIP-792).
> > > If
> > > > > the
> > > > > > > > consumer
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > > already
> > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at this
> > > point.
> > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled
> > > > > with the
> > > > > > > > > > >> feature
> > > > > > > > > > >> > flag
> > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is automatically
> > > > > > > > converted when
> > > > > > > > > > >> > the first
> > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol joins
> > > the
> > > > > group.
> > > > > > > > While
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > members
> > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> > > upgraded,
> > > > > the old
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > is
> > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of this in
> > > the
> > > > > KIP.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are
> > > > > pushing the
> > > > > > > > pattern
> > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like there
> > > could be
> > > > > some
> > > > > > > > tricky
> > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for users to
> > > > > detect
> > > > > > > > that
> > > > > > > > > > >> they
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to the new
> > > > > protocol?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit more
> > > > > optimistic
> > > > > > > > than
> > > > > > > > > > >> you
> > > > > > > > > > >> > on this
> > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > >> > > > > >> > > > > > > > > believe that the majority of
> > > the
> > > > > cases
> > > > > > > > are
> > > > > > > > > > >> simple
> > > > > > > > > > >> > > > > regexes
> > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > work with the new engine. The
> > > > > > > > coordinator will
> > > > > > > > > > >> > verify
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer if the
> > > > > regex is
> > > > > > > > not
> > > > > > > > > > >> valid.
> > > > > > > > > > >> > > > > Coming
> > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the worst
> > > case,
> > > > > the
> > > > > > > > first
> > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > >> > > > > >> > > > > > > > > the group will be rejected.
> > > This
> > > > > should
> > > > > > > > be used
> > > > > > > > > > >> > as the
> > > > > > > > > > >> > > > > >> last
> > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> > > validate
> > > > > their
> > > > > > > > regex
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test them with
> > > > > another
> > > > > > > > group.
> > > > > > > > > > >> For
> > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod
> > > > > environment.
> > > > > > > > > > >> Another
> > > > > > > > > > >> > way
> > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > extend the consumer-group tool
> > > to
> > > > > > > > provide a
> > > > > > > > > > >> regex
> > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> > > enough in
> > > > > your
> > > > > > > > > > >> opinion?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> > > question,
> > > > > will
> > > > > > > > the
> > > > > > > > > > >> Java
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the current regex
> > > > > engine for
> > > > > > > > > > >> > compatibility
> > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to keep using
> > > > > client
> > > > > > > > based
> > > > > > > > > > >> > regex at
> > > > > > > > > > >> > > > > >> first
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and then
> > > > > migrate to
> > > > > > > > server
> > > > > > > > > > >> > based
> > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I understand your point but I
> > > am
> > > > > > > > concerned that
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in this mode.
> > > That
> > > > > > > > would go
> > > > > > > > > > >> > against our
> > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client because
> > > we
> > > > > would
> > > > > > > > have to
> > > > > > > > > > >> > continue
> > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the client
> > > side. I
> > > > > would
> > > > > > > > rather
> > > > > > > > > > >> > not do
> > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> > > > > > > > coordinator
> > > > > > > > > > >> will
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations and that
> > > the
> > > > > > > > > > >> configurations
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a
> > > > > transition to
> > > > > > > > DEAD
> > > > > > > > > > >> > trigger
> > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > configurations
> > > > > will be
> > > > > > > > > > >> deleted
> > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to store
> > > the
> > > > > > > > configs in
> > > > > > > > > > >> the
> > > > > > > > > > >> > group
> > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all cluster
> > > > > configs and
> > > > > > > > their
> > > > > > > > > > >> > values?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > > > > > > configurations are
> > > > > > > > > > >> > overrides
> > > > > > > > > > >> > > > > >> of
> > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to know
> > > all
> > > > > the
> > > > > > > > overrides
> > > > > > > > > > >> > though,
> > > > > > > > > > >> > > > > you
> > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > > coordinators.
> > > > > You
> > > > > > > > cannot
> > > > > > > > > > >> > rely on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> > > configure a
> > > > > group
> > > > > > > > before
> > > > > > > > > > >> > starting
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > >> > > > > >> > > > > > > > > > we considered allowing the
> > > > > explicit
> > > > > > > > creation
> > > > > > > > > > >> of
> > > > > > > > > > >> > > > > groups?
> > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be decoupled
> > > from
> > > > > the
> > > > > > > > group
> > > > > > > > > > >> > lifecycle.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > > > > automatically
> > > > > > > > created in
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be lost after
> > > the
> > > > > > > > retention
> > > > > > > > > > >> > period of
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > Consumer.subscribe
> > > > > method
> > > > > > > > for the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of
> > > do
> > > > > we
> > > > > > > > have to
> > > > > > > > > > >> > introduce
> > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group question. I
> > > > > forgot
> > > > > > > > about
> > > > > > > > > > >> that
> > > > > > > > > > >> > one.
> > > > > > > > > > >> > > > > As
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is
> > > not
> > > > > fully
> > > > > > > > > > >> compatible
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might be
> > > better
> > > > > to
> > > > > > > > deprecate
> > > > > > > > > > >> > it and
> > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string. We would
> > > > > rely on
> > > > > > > > the
> > > > > > > > > > >> server
> > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I think
> > > that
> > > > > we
> > > > > > > > could
> > > > > > > > > > >> still
> > > > > > > > > > >> > try to
> > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That should
> > > > > work, I
> > > > > > > > think, in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine
> > > that we
> > > > > > > > should be
> > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > about
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > IBP/metadata.version
> > > > > from the
> > > > > > > > > > >> feature
> > > > > > > > > > >> > flag.
> > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the default for
> > > the
> > > > > feature
> > > > > > > > flag
> > > > > > > > > > >> to
> > > > > > > > > > >> > true
> > > > > > > > > > >> > > > > >> based
> > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once we
> > > want to
> > > > > make
> > > > > > > > it the
> > > > > > > > > > >> > default.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that feature
> > > > > flag
> > > > > > > > mainly
> > > > > > > > > > >> > during the
> > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> > > mentioned
> > > > > it, I
> > > > > > > > think,
> > > > > > > > > > >> > because
> > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs
> > > > > don't
> > > > > > > > mention
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > required
> > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> > > consistency.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > requires
> > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do more
> > > than
> > > > > > > > reading.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> > > current
> > > > > > > > protocol works
> > > > > > > > > > >> > though.
> > > > > > > > > > >> > > > > We
> > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to join a
> > > > > group. We
> > > > > > > > could
> > > > > > > > > > >> > consider
> > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that it is
> > > worth
> > > > > it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic recreation
> > > > > handled by
> > > > > > > > the
> > > > > > > > > > >> > consumer
> > > > > > > > > > >> > > > > with
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be good to
> > > > > have a
> > > > > > > > section
> > > > > > > > > > >> on
> > > > > > > > > > >> > this.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > > > > perspective, the
> > > > > > > > new
> > > > > > > > > > >> topic
> > > > > > > > > > >> > will
> > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will treat it
> > > like a
> > > > > > > > topic with
> > > > > > > > > > >> a
> > > > > > > > > > >> > > > > different
> > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> > > fetch/commit
> > > > > > > > offsets
> > > > > > > > > > >> APIs
> > > > > > > > > > >> > do not
> > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would
> > > reuse the
> > > > > > > > offsets
> > > > > > > > > > >> based
> > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > >> > > > > >> > > > > > > > > that we should update those
> > > APIs
> > > > > as well
> > > > > > > > in
> > > > > > > > > > >> order
> > > > > > > > > > >> > to be
> > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would strengthen
> > > the
> > > > > > > > semantics of
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will
> > > > > write the
> > > > > > > > new
> > > > > > > > > > >> > coordinator
> > > > > > > > > > >> > > > > >> in
> > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > >> > > > > >> > > > > > > > > > this is an implementation
> > > > > detail, do
> > > > > > > > we plan
> > > > > > > > > > >> to
> > > > > > > > > > >> > have a
> > > > > > > > > > >> > > > > >> new
> > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability
> > > > > goal when
> > > > > > > > it
> > > > > > > > > > >> comes
> > > > > > > > > > >> > to how
> > > > > > > > > > >> > > > > >> many
> > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at the
> > > > > moment. The
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > >> > > > > >> > > > > > > > > of members per group. We will
> > > > > measure
> > > > > > > > this when
> > > > > > > > > > >> > we have
> > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note that we
> > > might
> > > > > have
> > > > > > > > other
> > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> > > > > > > > SubscribedTopidIds
> > > > > > > > > > >> > instead
> > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to resolve the
> > > > > regex on
> > > > > > > > the
> > > > > > > > > > >> > server, we
> > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> > > difference
> > > > > is that
> > > > > > > > > > >> sending
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the topic
> > > names
> > > > > is less
> > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to
> > > handle
> > > > > if we
> > > > > > > > have
> > > > > > > > > > >> topic
> > > > > > > > > > >> > ids.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to consolidate the
> > > > > metadata
> > > > > > > > lookup
> > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > >> server
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree with your
> > > > > point. As
> > > > > > > > a
> > > > > > > > > > >> second
> > > > > > > > > > >> > > > > though,
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here for the
> > > > > delete and
> > > > > > > > > > >> > recreation
> > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow
> > > users to
> > > > > > > > subscribe
> > > > > > > > > > >> with
> > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > future because that is the only
> > > > > way to be
> > > > > > > > > > >> really
> > > > > > > > > > >> > robust
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM
> > > > > David
> > > > > > > > Jacot <
> > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your comments.
> > > Please
> > > > > find
> > > > > > > > my
> > > > > > > > > > >> answers
> > > > > > > > > > >> > > > > below.
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol
> > > relies on
> > > > > > > > topic IDs
> > > > > > > > > > >> > with the
> > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > >> > > > > >> I
> > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names is the
> > > > > right call
> > > > > > > > here.
> > > > > > > > > > >> I
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > >> > > > > >> > > > > > > > > > a little more. Obviously, the
> > > > > KIP does
> > > > > > > > not
> > > > > > > > > > >> > change the
> > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic
> > > IDs.
> > > > > This
> > > > > > > > may be
> > > > > > > > > > >> > something
> > > > > > > > > > >> > > > > >> that
> > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it would
> > > give
> > > > > better
> > > > > > > > > > >> overall
> > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I think that
> > > I
> > > > > should
> > > > > > > > not
> > > > > > > > > > >> have
> > > > > > > > > > >> > > > > mentioned
> > > > > > > > > > >> > > > > >> > this
> > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it. We can
> > > > > use an
> > > > > > > > internal
> > > > > > > > > > >> > > > > >> configuration
> > > > > > > > > > >> > > > > >> > > while
> > > > > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> > > > > > > > supported. The
> > > > > > > > > > >> > change is
> > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is that the
> > > > > cluster
> > > > > > > > uses
> > > > > > > > > > >> topic
> > > > > > > > > > >> > IDs.
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53
> > > PM
> > > > > > > > Guozhang
> > > > > > > > > > >> Wang <
> > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the feedback.
> > > Here
> > > > > are
> > > > > > > > some
> > > > > > > > > > >> replies
> > > > > > > > > > >> > > > > inlined
> > > > > > > > > > >> > > > > >> > > below:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at
> > > 2:53 AM
> > > > > > > > Ismael Juma
> > > > > > > > > > >> <
> > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This
> > > has
> > > > > the
> > > > > > > > > > >> potential
> > > > > > > > > > >> > to be a
> > > > > > > > > > >> > > > > >> > great
> > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > questions/comments:
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> > > premature to
> > > > > talk
> > > > > > > > about
> > > > > > > > > > >> > target
> > > > > > > > > > >> > > > > >> versions
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the existing
> > > group
> > > > > > > > protocol.
> > > > > > > > > > >> > Unlike
> > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > > > > > > > > >> > deprecation/removal will
> > > > > > > > > > >> > > > > >> be
> > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly applications
> > > > > migrate
> > > > > > > > to the
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > > >> protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I
> > > think
> > > > > we can
> > > > > > > > > > >> remove
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> proposed
> > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > > Deprecation,
> > > > > and
> > > > > > > > Migration
> > > > > > > > > > >> > Plan` and
> > > > > > > > > > >> > > > > >> > > instead
> > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide in the
> > > > > future
> > > > > > > > about
> > > > > > > > > > >> when
> > > > > > > > > > >> > we
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> > > intend to
> > > > > > > > release this
> > > > > > > > > > >> > in 4.x,
> > > > > > > > > > >> > > > > >> but
> > > > > > > > > > >> > > > > >> > it
> > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added that as
> > > a
> > > > > way to
> > > > > > > > > > >> estimate
> > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > we'd
> > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I
> > > also
> > > > > suggest
> > > > > > > > > > >> removing
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > >> part.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> > > specifically
> > > > > > > > related to
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for
> > > client
> > > > > > > > upgrades.
> > > > > > > > > > >> I.e.
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may be done
> > > > > first,
> > > > > > > > and then
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > client
> > > > > > > > > > >> > > > > >> > side,
> > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as "released"
> > > by
> > > > > the
> > > > > > > > time
> > > > > > > > > > >> clients
> > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable the
> > > > > feature the
> > > > > > > > > > >> clients
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling
> > > bounce
> > > > > and
> > > > > > > > then set
> > > > > > > > > > >> > the flag
> > > > > > > > > > >> > > > > >> > with a
> > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel
> > > it's
> > > > > > > > better to be
> > > > > > > > > > >> > released
> > > > > > > > > > >> > > > > >> in a
> > > > > > > > > > >> > > > > >> > > major
> > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out
> > > the
> > > > > > > > details of
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > migration
> > > > > > > > > > >> > > > > >> > > story.
> > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we will
> > > support
> > > > > online
> > > > > > > > > > >> > migrations. Is
> > > > > > > > > > >> > > > > >> that
> > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in detail.
> > > It
> > > > > could
> > > > > > > > also be
> > > > > > > > > > >> > done as a
> > > > > > > > > > >> > > > > >> > > separate
> > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the
> > > part we
> > > > > can
> > > > > > > > be more
> > > > > > > > > > >> > concrete
> > > > > > > > > > >> > > > > >> about
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to your
> > > > > question 2)
> > > > > > > > above).
> > > > > > > > > > >> > We will
> > > > > > > > > > >> > > > > >> work
> > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel as we
> > > > > solicit
> > > > > > > > more
> > > > > > > > > > >> > feedback.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are
> > > > > pushing
> > > > > > > > the
> > > > > > > > > > >> > pattern
> > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like there
> > > > > could be
> > > > > > > > some
> > > > > > > > > > >> tricky
> > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for
> > > users to
> > > > > > > > detect that
> > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > need
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > > update
> > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching to the
> > > new
> > > > > > > > protocol?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need some
> > > > > tooling for
> > > > > > > > > > >> non-java
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > >> > users
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> > > > > > > > switching to
> > > > > > > > > > >> the
> > > > > > > > > > >> > new
> > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top of my
> > > head
> > > > > > > > though,
> > > > > > > > > > >> maybe
> > > > > > > > > > >> > others
> > > > > > > > > > >> > > > > >> like
> > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the last
> > > > > question,
> > > > > > > > will the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the current
> > > regex
> > > > > > > > engine for
> > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to keep
> > > using
> > > > > > > > client
> > > > > > > > > > >> based
> > > > > > > > > > >> > regex
> > > > > > > > > > >> > > > > at
> > > > > > > > > > >> > > > > >> > > first to
> > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple and
> > > then
> > > > > migrate
> > > > > > > > to
> > > > > > > > > > >> > server based
> > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not thought
> > > > > about
> > > > > > > > that for
> > > > > > > > > > >> > java
> > > > > > > > > > >> > > > > >> clients,
> > > > > > > > > > >> > > > > >> > > and
> > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > > > > compatibility
> > > > > > > > issues do
> > > > > > > > > > >> > you have
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that the
> > > group
> > > > > > > > coordinator
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the configurations and
> > > that
> > > > > the
> > > > > > > > > > >> > configurations
> > > > > > > > > > >> > > > > will
> > > > > > > > > > >> > > > > >> be
> > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> > > > > > > > transition to
> > > > > > > > > > >> DEAD
> > > > > > > > > > >> > > > > trigger
> > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state
> > > is an
> > > > > > > > ending
> > > > > > > > > > >> state
> > > > > > > > > > >> > (we
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the group is
> > > EMPTY
> > > > > and
> > > > > > > > also all
> > > > > > > > > > >> of
> > > > > > > > > > >> > its
> > > > > > > > > > >> > > > > >> > metadata
> > > > > > > > > > >> > > > > >> > > are
> > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this
> > > > > group
> > > > > > > > would
> > > > > > > > > > >> never
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > >> revived.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to
> > > store
> > > > > the
> > > > > > > > configs
> > > > > > > > > > >> in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > group
> > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> > > cluster
> > > > > configs
> > > > > > > > and
> > > > > > > > > > >> their
> > > > > > > > > > >> > > > > values?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good question,
> > > and our
> > > > > > > > thoughts
> > > > > > > > > > >> are
> > > > > > > > > > >> > that
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> > > overrides
> > > > > of the
> > > > > > > > > > >> > cluster-level
> > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per group so
> > > when an
> > > > > > > > admin list
> > > > > > > > > > >> > cluster
> > > > > > > > > > >> > > > > >> > configs
> > > > > > > > > > >> > > > > >> > > it's
> > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> > > > > > > > "defaults", not
> > > > > > > > > > >> > showing
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone
> > > > > configure a
> > > > > > > > group
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > >> starting
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered allowing
> > > the
> > > > > explicit
> > > > > > > > > > >> > creation of
> > > > > > > > > > >> > > > > >> groups?
> > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> > > decoupled
> > > > > from
> > > > > > > > the group
> > > > > > > > > > >> > > > > lifecycle.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be created
> > > > > before
> > > > > > > > the group
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > as
> > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> > > requires
> > > > > the
> > > > > > > > > > >> corresponding
> > > > > > > > > > >> > > > > >> request to
> > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator based on
> > > the
> > > > > group
> > > > > > > > id ---
> > > > > > > > > > >> > the only
> > > > > > > > > > >> > > > > >> > thing
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group itself is
> > > gone
> > > > > we
> > > > > > > > also check
> > > > > > > > > > >> > if there
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > any
> > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to that
> > > group
> > > > > and
> > > > > > > > delete
> > > > > > > > > > >> as
> > > > > > > > > > >> > well.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed
> > > > > introduces an
> > > > > > > > > > >> > asymmetry on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > creation
> > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the config
> > > > > entities,
> > > > > > > > and we
> > > > > > > > > > >> > would like
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > > hear
> > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we should
> > > aim
> > > > > for
> > > > > > > > symmetry
> > > > > > > > > > >> > i.e.
> > > > > > > > > > >> > > > > >> totally
> > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not
> > > delete
> > > > > them at
> > > > > > > > all
> > > > > > > > > > >> when
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > >> > is
> > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit deletion
> > > > > operations
> > > > > > > > by
> > > > > > > > > > >> > themselves.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > > > Consumer.subscribe
> > > > > > > > method for
> > > > > > > > > > >> > the Java
> > > > > > > > > > >> > > > > >> > client
> > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > `java.util.regex.Pattern` of
> > > > > do we
> > > > > > > > have
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > > >> introduce an
> > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not need to
> > > > > introduce
> > > > > > > > an
> > > > > > > > > > >> > overload, but
> > > > > > > > > > >> > > > > >> I'm
> > > > > > > > > > >> > > > > >> > > all
> > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some compatibility
> > > > > issues
> > > > > > > > that we
> > > > > > > > > > >> may
> > > > > > > > > > >> > > > > overlook.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine
> > > > > that we
> > > > > > > > should
> > > > > > > > > > >> be
> > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > >> > about
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > > IBP/metadata.version
> > > > > > > > from the
> > > > > > > > > > >> > feature
> > > > > > > > > > >> > > > > >> flag.
> > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the default
> > > for
> > > > > the
> > > > > > > > feature
> > > > > > > > > > >> flag
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > true
> > > > > > > > > > >> > > > > >> > > based on
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version once we
> > > > > want to
> > > > > > > > make it
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> default.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol
> > > APIs
> > > > > don't
> > > > > > > > mention
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> required
> > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that for
> > > > > consistency.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do
> > > that.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > >> > requires
> > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to do
> > > more
> > > > > than
> > > > > > > > reading.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought myself
> > > as
> > > > > well,
> > > > > > > > but in
> > > > > > > > > > >> the
> > > > > > > > > > >> > end we
> > > > > > > > > > >> > > > > >> > could
> > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative: adding
> > > > > Write
> > > > > > > > Group
> > > > > > > > > > >> seems
> > > > > > > > > > >> > an
> > > > > > > > > > >> > > > > >> overkill
> > > > > > > > > > >> > > > > >> > > here
> > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we
> > > only
> > > > > have
> > > > > > > > Read /
> > > > > > > > > > >> > Delete
> > > > > > > > > > >> > > > > and
> > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to hear
> > > others
> > > > > > > > thoughts.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> > > recreation
> > > > > > > > handled by
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > with
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be
> > > good
> > > > > to have
> > > > > > > > a
> > > > > > > > > > >> > section on
> > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > > > > subscription
> > > > > > > > right? Yes
> > > > > > > > > > >> > we can
> > > > > > > > > > >> > > > > >> add a
> > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically the
> > > idea
> > > > > is that
> > > > > > > > > > >> consumer
> > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > be
> > > > > > > > > > >> > > > > >> > > totally
> > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as it's
> > > > > handled all
> > > > > > > > by the
> > > > > > > > > > >> > brokers.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we
> > > will
> > > > > write
> > > > > > > > the
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an implementation
> > > > > detail,
> > > > > > > > do we
> > > > > > > > > > >> > plan to
> > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought about
> > > > > that. But
> > > > > > > > I think
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > answer
> > > > > > > > > > >> > > > > >> > > should
> > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> > > scalability
> > > > > goal
> > > > > > > > when it
> > > > > > > > > > >> > comes to
> > > > > > > > > > >> > > > > >> how
> > > > > > > > > > >> > > > > >> > > many
> > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> > > support?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I think we
> > > > > should
> > > > > > > > shoot for
> > > > > > > > > > >> > 1000s of
> > > > > > > > > > >> > > > > >> > > members.
> > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals also
> > > depend
> > > > > on the
> > > > > > > > offset
> > > > > > > > > > >> > > > > management
> > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> > > coordinator
> > > > > > > > which we
> > > > > > > > > > >> did
> > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > >> cover in
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a number that
> > > > > applies
> > > > > > > > > > >> > universally.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider
> > > having
> > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to resolve
> > > the
> > > > > regex
> > > > > > > > on the
> > > > > > > > > > >> > server,
> > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > can
> > > > > > > > > > >> > > > > >> > > do
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > > > > difference is
> > > > > > > > that
> > > > > > > > > > >> > sending the
> > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > >> > > is
> > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending the topic
> > > > > names is
> > > > > > > > less
> > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is easier to
> > > > > handle if
> > > > > > > > we have
> > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > ids.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to still
> > > let
> > > > > the
> > > > > > > > clients
> > > > > > > > > > >> send
> > > > > > > > > > >> > names
> > > > > > > > > > >> > > > > >> is to
> > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids
> > > on
> > > > > the
> > > > > > > > broker /
> > > > > > > > > > >> > admin
> > > > > > > > > > >> > > > > client
> > > > > > > > > > >> > > > > >> > > only.
> > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added topic id
> > > in
> > > > > > > > KIP-516, we
> > > > > > > > > > >> > never
> > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> > > leveraging
> > > > > the
> > > > > > > > related
> > > > > > > > > > >> > newer
> > > > > > > > > > >> > > > > >> versioned
> > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic id as
> > > empty
> > > > > UUID.
> > > > > > > > We
> > > > > > > > > > >> want
> > > > > > > > > > >> > to keep
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> > > delegate
> > > > > the
> > > > > > > > reasoning
> > > > > > > > > > >> on
> > > > > > > > > > >> > broker
> > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at
> > > 10:45
> > > > > AM
> > > > > > > > David
> > > > > > > > > > >> Jacot
> > > > > > > > > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> > > > > > > > discussion
> > > > > > > > > > >> thread
> > > > > > > > > > >> > on
> > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the
> > > Consumer
> > > > > > > > Rebalance
> > > > > > > > > > >> > Protocol.
> > > > > > > > > > >> > > > > >> With
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the rebalance
> > > > > protocol
> > > > > > > > (for
> > > > > > > > > > >> > consumers)
> > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to
> > > > > implement for
> > > > > > > > > > >> > clients, and
> > > > > > > > > > >> > > > > >> easier
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > > > > >> > > > > >> > > > >
> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a look and
> > > let
> > > > > me
> > > > > > > > know what
> > > > > > > > > > >> > you
> > > > > > > > > > >> > > > > think.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from
> > > > > July
> > > > > > > > 18th to
> > > > > > > > > > >> > August 8th.
> > > > > > > > > > >> > > > > >> That
> > > > > > > > > > >> > > > > >> > > gives
> > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read and
> > > digest
> > > > > this
> > > > > > > > long
> > > > > > > > > > >> KIP.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> --
> > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > > > > >
> > > > > > > >
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Sagar <sa...@gmail.com>]

Hi Hector,

FWIW, I had created a draft page for Connect here:
https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol.
I am looking for reviews/comments on this(TBH I haven't spent much time on
this since the last edit I made).

Also, @David/@Guozhang, would it make sense to start the connect discussion
on a separate thread since this thread is about to go to Voting phase?

Thanks!
Sagar.


On Mon, Sep 12, 2022 at 5:47 PM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Guozhang,
>
> 1. I have added a reference to the relevant chapter instead of
> repeating the whole thing. Does that work for you?
>
> 2. The "Rebalance Triggers" section you are referring to is about when
> a rebalance should be triggered for the non-upgraded members using the
> old protocol. The section mentions that a rebalance must be triggered
> when a new assignment is installed. This implies that the group epoch
> was updated either by a native member or a non-upgraded member. For
> the latter, the JoinGroup request would be the trigger. I have added a
> reference to the relevant chapter in the "JoinGroup Handling" section
> as well. Does that make sense?
>
> Thanks,
> David
>
> On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello David,
> >
> > Alright I think that's sufficient. Just to make that clear in the doc,
> > could we update:
> >
> > 1) the heartbeat request handling section, stating when coordinator will
> > trigger rebalance based on the HB's member metadata / reason?
> > 2) the "Rebalance Triggers" section to include what we described in
> "Group
> > Epoch - Trigger a rebalance" section as well?
> >
> >
> > Guozhang
> >
> > On Fri, Sep 9, 2022 at 1:28 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > I thought that the assignor will always be consulted when the next
> > > heartbeat request is constructed. In other words,
> > > `PartitionAssignor#metadata` will be called for every heartbeat. This
> > > gives the opportunity for the assignor to enforce a rebalance by
> > > setting the reason to a non-zero value or by changing the bytes. Do
> > > you think that this is not sufficient? Are you concerned by the delay?
> > >
> > > Best,
> > > David
> > >
> > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > >
> > > > Hello David,
> > > >
> > > > One of Jun's comments make me thinking:
> > > >
> > > > ```
> > > > In this case, a new assignment is triggered by the client side
> > > > assignor. When constructing the HB, the consumer will always consult
> > > > the client side assignor and propagate the information to the group
> > > > coordinator. In other words, we don't expect users to call
> > > > Consumer#enforceRebalance anymore.
> > > > ```
> > > >
> > > > As I looked at the current PartitionAssignor's interface, we
> actually do
> > > > not have a way yet to instruct how to construct the next HB request,
> e.g.
> > > > when the assignor wants to enforce a new rebalance with a new
> assignment,
> > > > we'd need some customizable APIs inside the PartitionAssignor to
> indicate
> > > > the next HB telling broker about so. WDYT about adding such an API
> on the
> > > > PartitionAssignor?
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi Jun,
> > > > >
> > > > > I have updated the KIP to include your feedback. I have also tried
> to
> > > > > clarify the parts which were not cleared.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > > > >
> > > > > > Hi Jun,
> > > > > >
> > > > > > Thanks for your feedback. Let me start by answering your
> questions
> > > > > > inline and I will update the KIP next week.
> > > > > >
> > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem
> to
> > > be
> > > > > fewer
> > > > > > > RPCs during rebalance and more efficient support of wildcard.
> A few
> > > > > > > comments below.
> > > > > >
> > > > > > I would also add that the KIP removes the global sync barrier in
> the
> > > > > > protocol which is essential to improve group stability and
> > > > > > scalability, and the KIP also simplifies the client by moving
> most of
> > > > > > the logic to the server side.
> > > > > >
> > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> rolling
> > > > > changing
> > > > > > > of the partition assignor in the consumers?
> > > > > >
> > > > > > Definitely. The group coordinator will use the assignor used by a
> > > > > > majority of the members. This allows the group to move from one
> > > > > > assignor to another by a roll. This is explained in the Assignor
> > > > > > Selection chapter.
> > > > > >
> > > > > > > 30.2 For each field, could you explain whether it's required in
> > > every
> > > > > > > request or the scenarios when it needs to be filled? For
> example,
> > > it's
> > > > > not
> > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >
> > > > > > The client is expected to set those fields in case of a
> connection
> > > > > > issue (e.g. timeout) or when the fields have changed since the
> last
> > > > > > HB. The server populates those fields as long as the member is
> not
> > > > > > fully reconciled - the member should acknowledge that it has the
> > > > > > expected epoch and assignment. I will clarify this in the KIP.
> > > > > >
> > > > > > > 31. In the current consumer protocol, the rack affinity
> between the
> > > > > client
> > > > > > > and the broker is only considered during fetching, but not
> during
> > > > > assigning
> > > > > > > partitions to consumers. Sometimes, once the assignment is
> made,
> > > there
> > > > > is
> > > > > > > no opportunity for read affinity because no replicas of
> assigned
> > > > > partitions
> > > > > > > are close to the member. I am wondering if we should use this
> > > > > opportunity
> > > > > > > to address this by including rack in GroupMember.
> > > > > >
> > > > > > That's an interesting idea. I don't see any issue with adding the
> > > rack
> > > > > > to the members. I will do so.
> > > > > >
> > > > > > > 32. On the metric side, often, it's useful to know how busy a
> group
> > > > > > > coordinator is. By moving the event loop model, it seems that
> we
> > > could
> > > > > add
> > > > > > > a metric that tracks the fraction of the time the event loop is
> > > doing
> > > > > the
> > > > > > > actual work.
> > > > > >
> > > > > > That's a great idea. I will add it. Thanks.
> > > > > >
> > > > > > > 33. Could we add a section on coordinator failover handling?
> For
> > > > > example,
> > > > > > > does it need to trigger the check if any group with the
> wildcard
> > > > > > > subscription now has a new matching topic?
> > > > > >
> > > > > > Sure. When the new group coordinator takes over, it has to:
> > > > > > * Setup the session timeouts.
> > > > > > * Trigger a new assignment if a client side assignor is used. We
> > > don't
> > > > > > store the information about the member selected to run the
> assignment
> > > > > > so we have to start a new one.
> > > > > > * Update the topics metadata, verify the wildcard subscriptions,
> and
> > > > > > trigger a rebalance if needed.
> > > > > >
> > > > > > > 34. ConsumerGroupMetadataValue,
> > > ConsumerGroupPartitionMetadataValue,
> > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> epoch
> > > > > field
> > > > > > > reflects? For example, does the epoch in
> ConsumerGroupMetadataValue
> > > > > reflect
> > > > > > > the latest group epoch? What about the one in
> > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > ConsumerGroupMemberMetadataValue?
> > > > > >
> > > > > > Sure. I will clarify that but it is always the latest group
> epoch.
> > > > > > When the group state is updated, the group epoch is bumped so we
> use
> > > > > > that one for all the change records related to the update.
> > > > > >
> > > > > > > 35. "the group coordinator will ensure that the following
> > > invariants
> > > > > are
> > > > > > > met: ... All members exists." It's possible for a member not to
> > > get any
> > > > > > > assigned partitions, right?
> > > > > >
> > > > > > That's right. Here I meant that the members provided by the
> assignor
> > > > > > in the assignment must exist in the group. The assignor can not
> make
> > > > > > up new member ids.
> > > > > >
> > > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > > When
> > > > > would
> > > > > > > a consumer rejoin and what member id would be used?
> > > > > >
> > > > > > A member is expected to abandon all its partitions and rejoins
> when
> > > it
> > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > > > > coordinator will have removed the member from the group. The
> member
> > > > > > can rejoin the group with the same member id but with 0 as
> epoch. Let
> > > > > > me see if I can clarify this in the KIP.
> > > > > >
> > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > reassignment
> > > > > > > by either providing a non-zero reason or by updating the
> assignor
> > > > > > > metadata." Hmm, this seems to be conflicting with the
> deprecation
> > > of
> > > > > > > Consumer#enforeRebalance.
> > > > > >
> > > > > > In this case, a new assignment is triggered by the client side
> > > > > > assignor. When constructing the HB, the consumer will always
> consult
> > > > > > the client side assignor and propagate the information to the
> group
> > > > > > coordinator. In other words, we don't expect users to call
> > > > > > Consumer#enforceRebalance anymore.
> > > > > >
> > > > > > > 38. The reassignment examples are nice. But the section seems
> to
> > > have
> > > > > > > multiple typos.
> > > > > > > 38.1 When the group transitions to epoch 2, B immediately gets
> into
> > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > > epoch=3,
> > > > > > > partitions=[foo-1] too early.
> > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > partitions=[foo-0].
> > > > > >
> > > > > > Sorry for that! I will revise them.
> > > > > >
> > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> from
> > > any
> > > > > old
> > > > > > > version to new one?
> > > > > >
> > > > > > We will support upgrading from the consumer protocol version 3,
> > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
> earliest
> > > > > > version is unknown at the moment. This is explained in the
> migration
> > > > > > plan chapter.
> > > > > >
> > > > > > Thanks again for your feedback, Jun. I will update the KIP based
> on
> > > it
> > > > > > next week.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <jun@confluent.io.invalid
> >
> > > wrote:
> > > > > > >
> > > > > > > Hi, David,
> > > > > > >
> > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem
> to
> > > be
> > > > > fewer
> > > > > > > RPCs during rebalance and more efficient support of wildcard.
> A few
> > > > > > > comments below.
> > > > > > >
> > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> rolling
> > > > > changing
> > > > > > > of the partition assignor in the consumers?
> > > > > > > 30.2 For each field, could you explain whether it's required in
> > > every
> > > > > > > request or the scenarios when it needs to be filled? For
> example,
> > > it's
> > > > > not
> > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > > >
> > > > > > > 31. In the current consumer protocol, the rack affinity
> between the
> > > > > client
> > > > > > > and the broker is only considered during fetching, but not
> during
> > > > > assigning
> > > > > > > partitions to consumers. Sometimes, once the assignment is
> made,
> > > there
> > > > > is
> > > > > > > no opportunity for read affinity because no replicas of
> assigned
> > > > > partitions
> > > > > > > are close to the member. I am wondering if we should use this
> > > > > opportunity
> > > > > > > to address this by including rack in GroupMember.
> > > > > > >
> > > > > > > 32. On the metric side, often, it's useful to know how busy a
> group
> > > > > > > coordinator is. By moving the event loop model, it seems that
> we
> > > could
> > > > > add
> > > > > > > a metric that tracks the fraction of the time the event loop is
> > > doing
> > > > > the
> > > > > > > actual work.
> > > > > > >
> > > > > > > 33. Could we add a section on coordinator failover handling?
> For
> > > > > example,
> > > > > > > does it need to trigger the check if any group with the
> wildcard
> > > > > > > subscription now has a new matching topic?
> > > > > > >
> > > > > > > 34. ConsumerGroupMetadataValue,
> > > ConsumerGroupPartitionMetadataValue,
> > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> epoch
> > > > > field
> > > > > > > reflects? For example, does the epoch in
> ConsumerGroupMetadataValue
> > > > > reflect
> > > > > > > the latest group epoch? What about the one in
> > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >
> > > > > > > 35. "the group coordinator will ensure that the following
> > > invariants
> > > > > are
> > > > > > > met: ... All members exists." It's possible for a member not to
> > > get any
> > > > > > > assigned partitions, right?
> > > > > > >
> > > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > > When
> > > > > would
> > > > > > > a consumer rejoin and what member id would be used?
> > > > > > >
> > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > reassignment
> > > > > > > by either providing a non-zero reason or by updating the
> assignor
> > > > > > > metadata." Hmm, this seems to be conflicting with the
> deprecation
> > > of
> > > > > > > Consumer#enforeRebalance.
> > > > > > >
> > > > > > > 38. The reassignment examples are nice. But the section seems
> to
> > > have
> > > > > > > multiple typos.
> > > > > > > 38.1 When the group transitions to epoch 2, B immediately gets
> into
> > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > > epoch=3,
> > > > > > > partitions=[foo-1] too early.
> > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > partitions=[foo-0].
> > > > > > >
> > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> from
> > > any
> > > > > old
> > > > > > > version to new one?
> > > > > > >
> > > > > > > Thanks,
> > > > > > >
> > > > > > > Jun
> > > > > > >
> > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hi all,
> > > > > > > >
> > > > > > > > The KIP states that we will re-implement the coordinator in
> > > Java. I
> > > > > > > > discussed this offline with a few folks and folks are
> concerned
> > > that
> > > > > > > > we could introduce many regressions in the old protocol if
> we do
> > > so.
> > > > > > > > Therefore, I am going to remove this statement from the KIP.
> It
> > > is an
> > > > > > > > implementation detail after all so it does not have to be
> > > decided at
> > > > > > > > this stage. We will likely start by trying to refactor the
> > > current
> > > > > > > > implementation as a first step.
> > > > > > > >
> > > > > > > > Cheers,
> > > > > > > > David
> > > > > > > >
> > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> djacot@confluent.io
> > > >
> > > > > wrote:
> > > > > > > > >
> > > > > > > > > Hi Luke,
> > > > > > > > >
> > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > reconciling,
> > > > > > > > stable,
> > > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > >
> > > > > > > > > This sentence does not refer to those group states but
> rather
> > > to a
> > > > > > > > > state machine replication (SMR). This refers to the entire
> > > state of
> > > > > > > > > group coordinator which is replicated via the log layer. I
> will
> > > > > > > > > clarify this in the KIP.
> > > > > > > > >
> > > > > > > > > > 1.2. What do you mean "each state machine is modelled as
> an
> > > event
> > > > > > > > loop"?
> > > > > > > > >
> > > > > > > > > The idea is to follow a model similar to the new quorum
> > > > > controller. We
> > > > > > > > > will have N threads to process events. Each
> __consumer_offsets
> > > > > > > > > partition is assigned to a unique thread and all the events
> > > (e.g.
> > > > > > > > > requests, callbacks, etc.) are processed by this thread.
> This
> > > > > simplify
> > > > > > > > > concurrency and will enable us to do simulation testing
> for the
> > > > > group
> > > > > > > > > coordinator.
> > > > > > > > >
> > > > > > > > > > 1.3. Why do we need a state machine per
> *__consumer_offsets*
> > > > > > > > partitions?
> > > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > > coordinator?
> > > > > > > > For
> > > > > > > > > > example, if one group coordinator owns 2 consumer
> groups, and
> > > > > both
> > > > > > > > exist in
> > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for
> it,
> > > or
> > > > > 2?
> > > > > > > > >
> > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > > > > > > >
> > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> number
> > > of
> > > > > threads
> > > > > > > > used
> > > > > > > > > > to run the state machines. But I'm wondering if the
> purpose
> > > of
> > > > > the
> > > > > > > > threads
> > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > *__consumer_offsets*
> > > > > > > > > > partitions?), and no heavy computation, why should we
> need
> > > > > > > > multi-threads
> > > > > > > > > > here?
> > > > > > > > >
> > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > > processing as
> > > > > the
> > > > > > > > > computation could be heavy.
> > > > > > > > >
> > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > session
> > > > > > > > timeout not
> > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> min/max
> > > > > session
> > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > >
> > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > > detect
> > > > > client
> > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> minimum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> maximum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > >
> > > > > > > > > This is indeed a mistake. The default session timeout
> should
> > > be 45s
> > > > > > > > > (the current default).
> > > > > > > > >
> > > > > > > > > > 2.2. The default server side assignor are [range,
> uniform],
> > > > > which means
> > > > > > > > > > we'll default to "range" assignor. I'd like to know why
> not
> > > > > uniform
> > > > > > > > one? I
> > > > > > > > > > thought usually users will choose uniform assignor
> (former
> > > sticky
> > > > > > > > assinor)
> > > > > > > > > > for better evenly distribution. Any other reason we
> choose
> > > range
> > > > > > > > assignor
> > > > > > > > > > as default?
> > > > > > > > > > group.consumer.assignors List range, uniform The server
> side
> > > > > assignors.
> > > > > > > > >
> > > > > > > > > The order on the server side has no influence because the
> > > client
> > > > > must
> > > > > > > > > chose the selector that he wants to use. There is no
> default
> > > in the
> > > > > > > > > current proposal. If the assignor is not specified by the
> > > client,
> > > > > the
> > > > > > > > > request is rejected. The default client value for
> > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > > >
> > > > > > > > > Thanks for your very good comments, Luke. I hope that my
> > > answers
> > > > > help
> > > > > > > > > to clarify things. I will update the KIP as well based on
> your
> > > > > > > > > feedback.
> > > > > > > > >
> > > > > > > > > Cheers,
> > > > > > > > > David
> > > > > > > > >
> > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> showuon@gmail.com>
> > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Hi David,
> > > > > > > > > >
> > > > > > > > > > Thanks for the update.
> > > > > > > > > >
> > > > > > > > > > Some more questions:
> > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > > > > The new group coordinator will have a state machine per
> > > > > > > > > > *__consumer_offsets* partitions, where each state
> machine is
> > > > > modelled
> > > > > > > > as an
> > > > > > > > > > event loop. Those state machines will be executed in
> > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > >
> > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > reconciling,
> > > > > > > > stable,
> > > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > > > 1.2. What do you mean "each state machine is modelled as
> an
> > > event
> > > > > > > > loop"?
> > > > > > > > > > 1.3. Why do we need a state machine per
> *__consumer_offsets*
> > > > > > > > partitions?
> > > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > > coordinator?
> > > > > > > > For
> > > > > > > > > > example, if one group coordinator owns 2 consumer
> groups, and
> > > > > both
> > > > > > > > exist in
> > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for
> it,
> > > or
> > > > > 2?
> > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> number
> > > of
> > > > > threads
> > > > > > > > used
> > > > > > > > > > to run the state machines. But I'm wondering if the
> purpose
> > > of
> > > > > the
> > > > > > > > threads
> > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > *__consumer_offsets*
> > > > > > > > > > partitions?), and no heavy computation, why should we
> need
> > > > > > > > multi-threads
> > > > > > > > > > here?
> > > > > > > > > >
> > > > > > > > > > 2. For the default value in the new configs:
> > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > session
> > > > > > > > timeout not
> > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> min/max
> > > > > session
> > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > >
> > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > > detect
> > > > > client
> > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> minimum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> maximum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > 2.2. The default server side assignor are [range,
> uniform],
> > > > > which means
> > > > > > > > > > we'll default to "range" assignor. I'd like to know why
> not
> > > > > uniform
> > > > > > > > one? I
> > > > > > > > > > thought usually users will choose uniform assignor
> (former
> > > sticky
> > > > > > > > assinor)
> > > > > > > > > > for better evenly distribution. Any other reason we
> choose
> > > range
> > > > > > > > assignor
> > > > > > > > > > as default?
> > > > > > > > > > group.consumer.assignors List range, uniform The server
> side
> > > > > assignors.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Thank you.
> > > > > > > > > > Luke
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> showuon@gmail.com
> > > >
> > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hi Sagar,
> > > > > > > > > > >
> > > > > > > > > > > I have some thoughts about Kafka Connect integrating
> with
> > > > > KIP-848,
> > > > > > > > but I
> > > > > > > > > > > think we should have a separate discussion thread for
> the
> > > Kafka
> > > > > > > > Connect
> > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> Rebalance
> > > > > Protocol
> > > > > > > > [1],
> > > > > > > > > > > and let this discussion thread focus on consumer
> rebalance
> > > > > protocol,
> > > > > > > > WDYT?
> > > > > > > > > > >
> > > > > > > > > > > [1]
> > > > > > > > > > >
> > > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >
> > > > > > > > > > > Thank you.
> > > > > > > > > > > Luke
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > sagarmeansocean@gmail.com>
> > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks like
> > > there's
> > > > > > > > agreement on
> > > > > > > > > > >> using separate APIs for Connect. I would revisit the
> doc
> > > and
> > > > > see
> > > > > > > > what
> > > > > > > > > > >> changes are to be made.
> > > > > > > > > > >>
> > > > > > > > > > >> Thanks!
> > > > > > > > > > >> Sagar.
> > > > > > > > > > >>
> > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> wrote:
> > > > > > > > > > >>
> > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > >> >
> > > > > > > > > > >> > Thanks for the feedback and the document. That's
> really
> > > > > helpful. I
> > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > >> >
> > > > > > > > > > >> > Overall, it seems to me that both Connect and the
> > > Consumer
> > > > > could
> > > > > > > > share
> > > > > > > > > > >> > the same underlying "engine". The main difference is
> > > that
> > > > > the
> > > > > > > > Consumer
> > > > > > > > > > >> > assigns topic-partitions to members whereas Connect
> > > assigns
> > > > > tasks
> > > > > > > > to
> > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> different
> > > > > resource
> > > > > > > > types
> > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> dedicated
> > > APIs
> > > > > > > > would be
> > > > > > > > > > >> > similar to the new ones but different on the
> > > > > content/resources and
> > > > > > > > > > >> > they would rely on the same engine on the
> coordinator
> > > side.
> > > > > > > > > > >> >
> > > > > > > > > > >> > I personally lean towards 2) because I am not a fan
> of
> > > > > > > > overcharging
> > > > > > > > > > >> > APIs to serve different purposes. That being said,
> I am
> > > not
> > > > > > > > opposed to
> > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > > > > >> >
> > > > > > > > > > >> > I think that we can continue to discuss it here for
> now
> > > in
> > > > > order
> > > > > > > > to
> > > > > > > > > > >> > ensure that this KIP is compatible with what we
> will do
> > > for
> > > > > > > > Connect in
> > > > > > > > > > >> > the future.
> > > > > > > > > > >> >
> > > > > > > > > > >> > Best,
> > > > > > > > > > >> > David
> > > > > > > > > > >> >
> > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > > > djacot@confluent.io>
> > > > > > > > wrote:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Hi all,
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > I am back from vacation. I will go through and
> address
> > > > > your
> > > > > > > > comments
> > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Cheers,
> > > > > > > > > > >> > > David
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > > > > gharris1727@gmail.com
> > > > > > > > > > >> >
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > > cooperative
> > > > > > > > rebalancing
> > > > > > > > > > >> > making it
> > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> describes how
> > > > > member
> > > > > > > > epoch
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > advance in step with the group epoch and
> assignment
> > > > > epoch
> > > > > > > > values. I
> > > > > > > > > > >> > think
> > > > > > > > > > >> > > > that this is a good idea for the reasons
> described
> > > in
> > > > > the
> > > > > > > > KIP. When
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > protocol is incrementally assigning partitions
> to a
> > > > > worker,
> > > > > > > > what
> > > > > > > > > > >> member
> > > > > > > > > > >> > > > epoch does each incremental assignment use? Are
> > > member
> > > > > epochs
> > > > > > > > > > >> re-used,
> > > > > > > > > > >> > and
> > > > > > > > > > >> > > > a single member epoch can correspond to multiple
> > > > > different
> > > > > > > > > > >> > (monotonically
> > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to
> the
> > > group
> > > > > > > > > > >> coordinator? If
> > > > > > > > > > >> > > > not, should custom client-side assignor
> > > implementations
> > > > > > > > interact
> > > > > > > > > > >> with
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > Reason field, and how is its common meaning
> agreed
> > > > > upon? If
> > > > > > > > so, what
> > > > > > > > > > >> > is the
> > > > > > > > > > >> > > > benefit of a distinct Reason field over
> including
> > > such
> > > > > > > > functionality
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 3. The following is included in the KIP:
> "Thanks to
> > > > > this, the
> > > > > > > > input
> > > > > > > > > > >> of
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > client side assignor is entirely driven by the
> group
> > > > > > > > coordinator.
> > > > > > > > > > >> The
> > > > > > > > > > >> > > > consumer is no longer responsible for
> maintaining
> > > any
> > > > > state
> > > > > > > > besides
> > > > > > > > > > >> its
> > > > > > > > > > >> > > > assigned partitions." Does this mean that the
> > > > > client-side
> > > > > > > > assignor
> > > > > > > > > > >> MAY
> > > > > > > > > > >> > > > incorporate additional non-Metadata state (such
> as
> > > > > partition
> > > > > > > > > > >> > throughput,
> > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or that
> > > > > additional
> > > > > > > > > > >> > non-Metadata
> > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > > > > >> > > > for
> > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > >> > > > and
> > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > that
> > > > > > > > seem to
> > > > > > > > > > >> > > > overlap significantly. Is it possible for these
> two
> > > > > > > > implementations
> > > > > > > > > > >> to
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > unified? This would serve to promote feature
> parity
> > > of
> > > > > > > > server-side
> > > > > > > > > > >> and
> > > > > > > > > > >> > > > client-side assignors, and would also facilitate
> > > > > operational
> > > > > > > > > > >> > flexibility in
> > > > > > > > > > >> > > > certain situations. For example, if a
> server-side
> > > > > assignor
> > > > > > > > has some
> > > > > > > > > > >> > poor
> > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> patched
> > > > > assignor to
> > > > > > > > the
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > and switching one consumer group to a
> client-side
> > > > > assignor
> > > > > > > > may be
> > > > > > > > > > >> > faster
> > > > > > > > > > >> > > > and less risky than patching all of the brokers.
> > > With
> > > > > the
> > > > > > > > currently
> > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > reimplementation
> > > > > would
> > > > > > > > have
> > > > > > > > > > >> to be
> > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> > > > > significantly,
> > > > > > > > then it
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > possible that a reimplementation would not be
> > > possible.
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > --
> > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > I created a confluence page to discuss how
> Connect
> > > > > would
> > > > > > > > need to
> > > > > > > > > > >> > change
> > > > > > > > > > >> > > > > based on the new rebalance protocol. Here's
> the
> > > page:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > It's also pretty longish and I have tried to
> keep
> > > a
> > > > > format
> > > > > > > > > > >> similar to
> > > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also, do
> you
> > > > > think this
> > > > > > > > > > >> should
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > moved to a separate discussion thread or is
> this
> > > one
> > > > > fine?
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> Streams.
> > > > > Sure, I
> > > > > > > > would do
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang
> Wang <
> > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> It would be great if you could come back
> with
> > > some
> > > > > > > > analysis on
> > > > > > > > > > >> > how to
> > > > > > > > > > >> > > > > >> implement the Connect side integration with
> > > the new
> > > > > > > > protocol;
> > > > > > > > > > >> so
> > > > > > > > > > >> > far
> > > > > > > > > > >> > > > > >> besides leveraging on the new "protocol
> type"
> > > we
> > > > > did not
> > > > > > > > yet
> > > > > > > > > > >> think
> > > > > > > > > > >> > > > > through
> > > > > > > > > > >> > > > > >> the Connect side implementations. For
> Streams
> > > > > here's a
> > > > > > > > draft of
> > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > >
> > > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > Thank you for your response. The reason I
> > > thought
> > > > > > > > connect can
> > > > > > > > > > >> > also fit
> > > > > > > > > > >> > > > > >> into
> > > > > > > > > > >> > > > > >> > this new scheme is that even today the
> > > connect
> > > > > uses a
> > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to
> empower
> > > > > > > > rebalances of
> > > > > > > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator
> sets
> > > the
> > > > > > > > > > >> protocolType()
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> plumbing
> > > into
> > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > I think the changes to support connect
> would
> > > be
> > > > > > > > similar at a
> > > > > > > > > > >> > high
> > > > > > > > > > >> > > > > level
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > the changes in streams mainly because of
> the
> > > > > Client
> > > > > > > > side
> > > > > > > > > > >> > assignors
> > > > > > > > > > >> > > > > being
> > > > > > > > > > >> > > > > >> > used in both. At an implementation
> level, we
> > > > > might
> > > > > > > > need to
> > > > > > > > > > >> make
> > > > > > > > > > >> > a lot
> > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > changes to get onto this new assignment
> > > protocol
> > > > > like
> > > > > > > > > > >> enhancing
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > JoinGroup request/response and SyncGroup
> and
> > > > > using
> > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> streams (or
> > > > > there
> > > > > > > > might be
> > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > >> > > > > >> > would try to perform a detailed analysis
> of
> > > the
> > > > > same
> > > > > > > > and we
> > > > > > > > > > >> > can have
> > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > >> > separate discussion thread for that as
> that
> > > would
> > > > > > > > derail this
> > > > > > > > > > >> > > > > discussion
> > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds good
> to
> > > you.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David
> Jacot
> > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> `Assignment#error`.
> > > > > Sure, I
> > > > > > > > can
> > > > > > > > > > >> > mention it.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from his
> > > current
> > > > > > > > assignment
> > > > > > > > > > >> to
> > > > > > > > > > >> > his
> > > > > > > > > > >> > > > > >> > > target assignment when he can move to
> > > epoch 3.
> > > > > When
> > > > > > > > that
> > > > > > > > > > >> > happens,
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > member assignment is updated and
> persisted
> > > > > with all
> > > > > > > > its
> > > > > > > > > > >> > assigned
> > > > > > > > > > >> > > > > >> > > partitions even if they are not all
> revoked
> > > > > yet. In
> > > > > > > > other
> > > > > > > > > > >> > words, the
> > > > > > > > > > >> > > > > >> > > member assignment becomes the target
> > > > > assignment.
> > > > > > > > This is
> > > > > > > > > > >> > basically
> > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > >> > > optimization to avoid having to write
> all
> > > the
> > > > > > > > changes to
> > > > > > > > > > >> the
> > > > > > > > > > >> > log.
> > > > > > > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > examples are based on the persisted
> state
> > > so I
> > > > > > > > understand
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can improve
> > > this in
> > > > > the
> > > > > > > > > > >> > description.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse
> the
> > > > > protocol
> > > > > > > > with a
> > > > > > > > > > >> > client side
> > > > > > > > > > >> > > > > >> > > assignor if it fits in the protocol.
> The
> > > > > assignment
> > > > > > > > is
> > > > > > > > > > >> about
> > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> > > Connect
> > > > > fit
> > > > > > > > into this?
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor
> > > > > observations:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error section in
> > > Client
> > > > > Side
> > > > > > > > mode
> > > > > > > > > > >> > Assignment
> > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case, the
> > > client
> > > > > side
> > > > > > > > assignor
> > > > > > > > > > >> can
> > > > > > > > > > >> > > > > return
> > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > >> > > > error to the group coordinator`. In
> this
> > > > > case are
> > > > > > > > you
> > > > > > > > > > >> > referring to
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > Assignor returning an AssignmentError
> > > that's
> > > > > > > > listed down
> > > > > > > > > > >> > towards
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > >> > > > > >> > > > If yes, do you think it would make
> sense
> > > to
> > > > > > > > mention this
> > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I
> have a
> > > > > slight
> > > > > > > > > > >> confusion,
> > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> if
> > > > > > > > > > >> > > > > >> > > > others have the same. Consider this
> step:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> coordinator
> > > > > > > > transitions him
> > > > > > > > > > >> to
> > > > > > > > > > >> > epoch
> > > > > > > > > > >> > > > > 3
> > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> > > persists
> > > > > the
> > > > > > > > change and
> > > > > > > > > > >> > reply.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> partitions=[foo-0,
> > > > > foo-1]
> > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> partitions=[foo-2]
> > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> partitions=[foo-1]
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions to
> > > epoch 3
> > > > > but
> > > > > > > > cannot
> > > > > > > > > > >> get
> > > > > > > > > > >> > foo-1
> > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C
> can't
> > > get
> > > > > the
> > > > > > > > foo-1
> > > > > > > > > > >> > partition
> > > > > > > > > > >> > > > > yet,
> > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > >> > > > > >> > > > based on the description above, it
> seems
> > > it
> > > > > > > > already has
> > > > > > > > > > >> it.
> > > > > > > > > > >> > Do you
> > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > >> > > > > >> > > > would be better to remove it and
> > > populate it
> > > > > only
> > > > > > > > when it
> > > > > > > > > > >> > actually
> > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > >> > > > > >> > > > I see this in a lot of other places,
> so
> > > have
> > > > > I
> > > > > > > > > > >> understood it
> > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be out
> of
> > > scope
> > > > > of
> > > > > > > > this
> > > > > > > > > > >> > discussion,
> > > > > > > > > > >> > > > > but
> > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > >> > > > > >> > > > what I understood it would probably
> be
> > > > > running in
> > > > > > > > client
> > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > >> > > > > >> > > > even on the new rebalance protocol
> as it
> > > has
> > > > > its
> > > > > > > > own
> > > > > > > > > > >> Custom
> > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David
> > > Jacot
> > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move
> > > forward
> > > > > with
> > > > > > > > > > >> > specialized API
> > > > > > > > > > >> > > > > >> > > > > instead of relying on one generic
> API.
> > > For
> > > > > > > > Connect, we
> > > > > > > > > > >> > can apply
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > exact same pattern and reuse/share
> the
> > > core
> > > > > > > > > > >> > implementation on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> registry, I
> > > > > think
> > > > > > > > that we
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > > > > > > membership/leader
> > > > > > > > > > >> > election.
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM
> Ismael
> > > > > Juma <
> > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> java.util.regex.Pattern
> > > vs
> > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > document the differences in more
> > > detail
> > > > > before
> > > > > > > > > > >> deciding
> > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > way
> > > > > > > > > > >> > > > > >> or
> > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > >> they
> > > > > > > > > > >> > expect
> > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If we
> are
> > > doing
> > > > > > > > something
> > > > > > > > > > >> > different,
> > > > > > > > > > >> > > > > >> then
> > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> overload
> > > with
> > > > > our own
> > > > > > > > > > >> Pattern
> > > > > > > > > > >> > class
> > > > > > > > > > >> > > > > (I
> > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose re2j's
> at
> > > this
> > > > > > > > point).
> > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> major new
> > > > > protocol
> > > > > > > > should
> > > > > > > > > > >> > > > > integrate
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > > > > recreation case
> > > > > > > > > > >> > correctly.
> > > > > > > > > > >> > > > > That's
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I agree
> with
> > > > > David
> > > > > > > > that we'd
> > > > > > > > > > >> > want to
> > > > > > > > > > >> > > > > add
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols that
> don't
> > > > > have it
> > > > > > > > yet and
> > > > > > > > > > >> > that we
> > > > > > > > > > >> > > > > can
> > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > >> > > > > >> > > > > > focus on the internals versus
> adding
> > > new
> > > > > APIs
> > > > > > > > to the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs is
> > > required
> > > > > for
> > > > > > > > > > >> reasonable
> > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about the
> > > > > coordinator
> > > > > > > > storing
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > > centralized
> > > > > in the
> > > > > > > > > > >> metadata
> > > > > > > > > > >> > log for
> > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> visibility,
> > > > > consistency,
> > > > > > > > > > >> etc.).
> > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > >> > > > > >> > > > > > sure about automatically deleting
> > > > > configs in a
> > > > > > > > way
> > > > > > > > > > >> that
> > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > >> > > > > > recovered. A good property for
> modern
> > > > > systems
> > > > > > > > is to
> > > > > > > > > > >> > minimize
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> scenarios.
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM
> David
> > > > > Jacot
> > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are
> > > below:
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> especially
> > > > > the last
> > > > > > > > step
> > > > > > > > > > >> when
> > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol,
> in
> > > which
> > > > > we
> > > > > > > > would
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > protocols /
> > > > > rpcs
> > > > > > > > are
> > > > > > > > > > >> used
> > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> could
> > > > > "mimic"
> > > > > > > > the old
> > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > > need to
> > > > > > > > present
> > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a new
> > > version
> > > > > of KIP
> > > > > > > > which
> > > > > > > > > > >> > includes
> > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> > > "Supporting
> > > > > > > > Online
> > > > > > > > > > >> > Consumer
> > > > > > > > > > >> > > > > Group
> > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > > Deprecation,
> > > > > and
> > > > > > > > Migration
> > > > > > > > > > >> > Plan". I
> > > > > > > > > > >> > > > > >> think
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > I have to think through a few
> cases
> > > > > now but
> > > > > > > > the
> > > > > > > > > > >> > overall idea
> > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> understandable.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > > far as
> > > > > > > > KIP-516 the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> client,
> > > but
> > > > > they
> > > > > > > > are not
> > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > > question is,
> > > > > > > > first
> > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > > names
> > > > > -> ids
> > > > > > > > mapping
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > > RPCs
> > > > > and the
> > > > > > > > new
> > > > > > > > > > >> RPCs
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> consider
> > > > > exposing
> > > > > > > > the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > subscribe/assign, as
> > > > > > > > > > >> well
> > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> include
> > > > > converting
> > > > > > > > all
> > > > > > > > > > >> the
> > > > > > > > > > >> > offsets
> > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this
> > > KIP,
> > > > > the
> > > > > > > > consumer
> > > > > > > > > > >> > would be
> > > > > > > > > > >> > > > > >> able
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic ids.
> That
> > > > > being
> > > > > > > > said, it
> > > > > > > > > > >> > still has
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > >> > > > > >> > > > > > > the topics names in various
> APIs so
> > > > > having a
> > > > > > > > > > >> mapping
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> opinion on
> > > > > this.
> > > > > > > > Here I
> > > > > > > > > > >> > wonder if
> > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I
> > > would
> > > > > rather
> > > > > > > > focus
> > > > > > > > > > >> on
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > > > > separately if
> > > > > > > > we see
> > > > > > > > > > >> > value in
> > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's point
> about
> > > > > using
> > > > > > > > topic ids
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest,
> I
> > > think
> > > > > that
> > > > > > > > there
> > > > > > > > > > >> is
> > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer will
> > > have
> > > > > the
> > > > > > > > > > >> opportunity
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> passing
> > > them
> > > > > into
> > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > > coordinator
> > > > > will
> > > > > > > > also
> > > > > > > > > > >> notice
> > > > > > > > > > >> > it but
> > > > > > > > > > >> > > > > >> it
> > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> reject an
> > > > > invalid
> > > > > > > > topic in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> all
> > > other
> > > > > minor
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> question:
> > > > > personally
> > > > > > > > I think
> > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API
> with
> > > > > Pattern,
> > > > > > > > but
> > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > >> > > > > > > > while just documenting that
> our
> > > > > > > > subscription may
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case
> is a
> > > > > very rare
> > > > > > > > > > >> scenario
> > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > subscription may
> > > > > > > > be
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
> look at
> > > the
> > > > > > > > differences
> > > > > > > > > > >> > between
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > >> > > > > >> > > > > > > engines to better understand
> the
> > > > > potential
> > > > > > > > issues.
> > > > > > > > > > >> My
> > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > >> > > > > >> > > > > > > that would work for all the
> basic
> > > > > regular
> > > > > > > > > > >> > expressions. The
> > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
> about
> > > the
> > > > > various
> > > > > > > > > > >> character
> > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > >> > > > > >> > > > > > > wonder what other people think
> > > about
> > > > > this.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28
> PM
> > > > > Guozhang
> > > > > > > > Wang <
> > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on the
> high
> > > > > level
> > > > > > > > there are
> > > > > > > > > > >> > two meta
> > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> especially
> > > > > the last
> > > > > > > > step
> > > > > > > > > > >> when
> > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol,
> in
> > > which
> > > > > we
> > > > > > > > would
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > protocols /
> > > > > rpcs
> > > > > > > > are
> > > > > > > > > > >> used
> > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> could
> > > > > "mimic"
> > > > > > > > the old
> > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > > need to
> > > > > > > > present
> > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > > far as
> > > > > > > > KIP-516 the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> client,
> > > but
> > > > > they
> > > > > > > > are not
> > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > > question is,
> > > > > > > > first
> > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > > names
> > > > > -> ids
> > > > > > > > mapping
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > > RPCs
> > > > > and the
> > > > > > > > new
> > > > > > > > > > >> RPCs
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> consider
> > > > > exposing
> > > > > > > > the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > subscribe/assign, as
> > > > > > > > > > >> well
> > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> all
> > > other
> > > > > minor
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> question:
> > > > > personally
> > > > > > > > I think
> > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API
> with
> > > > > Pattern,
> > > > > > > > but
> > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > >> > > > > > > > while just documenting that
> our
> > > > > > > > subscription may
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case
> is a
> > > > > very rare
> > > > > > > > > > >> scenario
> > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > subscription may
> > > > > > > > be
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23
> AM
> > > > > David Jacot
> > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your feedback.
> Let
> > > me
> > > > > answer
> > > > > > > > your
> > > > > > > > > > >> > questions
> > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> premature to
> > > > > talk about
> > > > > > > > > > >> target
> > > > > > > > > > >> > > > > versions
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the existing
> group
> > > > > protocol.
> > > > > > > > > > >> Unlike
> > > > > > > > > > >> > KRaft,
> > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > > > > > > deprecation/removal
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly applications
> > > migrate
> > > > > to
> > > > > > > > the new
> > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> > > remove it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> intend to
> > > > > release
> > > > > > > > this in
> > > > > > > > > > >> > 4.x, but
> > > > > > > > > > >> > > > > it
> > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that as
> a
> > > way to
> > > > > > > > estimate
> > > > > > > > > > >> when
> > > > > > > > > > >> > we'd
> > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I
> also
> > > > > suggest
> > > > > > > > removing
> > > > > > > > > > >> > this part.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> reasoning. As
> > > > > > > > explained, I
> > > > > > > > > > >> plan
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java while
> we
> > > > > implement
> > > > > > > > the new
> > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> This
> > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > >> > > > > > > > > that the internals will be
> > > slightly
> > > > > > > > different
> > > > > > > > > > >> > (e.g.
> > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
> tighten
> > > the
> > > > > > > > switch from
> > > > > > > > > > >> > the old
> > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
> group
> > > > > coordinator
> > > > > > > > to a
> > > > > > > > > > >> > major
> > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to
> use a
> > > flag
> > > > > to do
> > > > > > > > the
> > > > > > > > > > >> > switch
> > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out
> the
> > > > > details of
> > > > > > > > the
> > > > > > > > > > >> > migration
> > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
> support
> > > > > online
> > > > > > > > > > >> migrations.
> > > > > > > > > > >> > Is
> > > > > > > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in detail.
> It
> > > could
> > > > > also
> > > > > > > > be done
> > > > > > > > > > >> > as a
> > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support online
> > > > > migrations
> > > > > > > > for the
> > > > > > > > > > >> > group.
> > > > > > > > > > >> > > > > That
> > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> protocol
> > > > > will be
> > > > > > > > able to
> > > > > > > > > > >> > switch to
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain how
> that
> > > > > will work
> > > > > > > > > > >> though.
> > > > > > > > > > >> > It is
> > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> > > upgraded or
> > > > > > > > rolled to a
> > > > > > > > > > >> > software
> > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
> Both the
> > > > > old and
> > > > > > > > the new
> > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > >> > > > > >> > > > > > > > > support the old protocol
> and
> > > rely
> > > > > on the
> > > > > > > > same
> > > > > > > > > > >> > persisted
> > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > >> > > > > >> > > > > > > > > they can work together.
> This
> > > point
> > > > > is an
> > > > > > > > > > >> offline
> > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > >> > > > > >> > > > > > > > > do this one live because it
> > > would
> > > > > require
> > > > > > > > > > >> > shutting down
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and starting
> up the
> > > > > new one
> > > > > > > > and
> > > > > > > > > > >> that
> > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > > > > version/IBP
> > > > > > > > must be
> > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> X
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new protocol.
> > > This
> > > > > cannot
> > > > > > > > be done
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > 1)
> > > > > > > > > > >> > > > > >> is
> > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the old
> > > > > coordinator
> > > > > > > > doesn't
> > > > > > > > > > >> > support
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> > > upgraded
> > > > > to a
> > > > > > > > version
> > > > > > > > > > >> > > > > supporting
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> KIP-792).
> > > If
> > > > > the
> > > > > > > > consumer
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > > already
> > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at
> this
> > > point.
> > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be
> rolled
> > > > > with the
> > > > > > > > > > >> feature
> > > > > > > > > > >> > flag
> > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> automatically
> > > > > > > > converted when
> > > > > > > > > > >> > the first
> > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
> joins
> > > the
> > > > > group.
> > > > > > > > While
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > members
> > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> > > upgraded,
> > > > > the old
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > is
> > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of this
> in
> > > the
> > > > > KIP.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are
> > > > > pushing the
> > > > > > > > pattern
> > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like there
> > > could be
> > > > > some
> > > > > > > > tricky
> > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
> users to
> > > > > detect
> > > > > > > > that
> > > > > > > > > > >> they
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to the
> new
> > > > > protocol?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit
> more
> > > > > optimistic
> > > > > > > > than
> > > > > > > > > > >> you
> > > > > > > > > > >> > on this
> > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > >> > > > > >> > > > > > > > > believe that the majority
> of
> > > the
> > > > > cases
> > > > > > > > are
> > > > > > > > > > >> simple
> > > > > > > > > > >> > > > > regexes
> > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > work with the new engine.
> The
> > > > > > > > coordinator will
> > > > > > > > > > >> > verify
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer if
> the
> > > > > regex is
> > > > > > > > not
> > > > > > > > > > >> valid.
> > > > > > > > > > >> > > > > Coming
> > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
> worst
> > > case,
> > > > > the
> > > > > > > > first
> > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > >> > > > > >> > > > > > > > > the group will be rejected.
> > > This
> > > > > should
> > > > > > > > be used
> > > > > > > > > > >> > as the
> > > > > > > > > > >> > > > > >> last
> > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> > > validate
> > > > > their
> > > > > > > > regex
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test them
> with
> > > > > another
> > > > > > > > group.
> > > > > > > > > > >> For
> > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod
> > > > > environment.
> > > > > > > > > > >> Another
> > > > > > > > > > >> > way
> > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > extend the consumer-group
> tool
> > > to
> > > > > > > > provide a
> > > > > > > > > > >> regex
> > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> > > enough in
> > > > > your
> > > > > > > > > > >> opinion?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> > > question,
> > > > > will
> > > > > > > > the
> > > > > > > > > > >> Java
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the current
> regex
> > > > > engine for
> > > > > > > > > > >> > compatibility
> > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to keep
> using
> > > > > client
> > > > > > > > based
> > > > > > > > > > >> > regex at
> > > > > > > > > > >> > > > > >> first
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and
> then
> > > > > migrate to
> > > > > > > > server
> > > > > > > > > > >> > based
> > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I understand your point
> but I
> > > am
> > > > > > > > concerned that
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in this
> mode.
> > > That
> > > > > > > > would go
> > > > > > > > > > >> > against our
> > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
> because
> > > we
> > > > > would
> > > > > > > > have to
> > > > > > > > > > >> > continue
> > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the client
> > > side. I
> > > > > would
> > > > > > > > rather
> > > > > > > > > > >> > not do
> > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that the
> group
> > > > > > > > coordinator
> > > > > > > > > > >> will
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations and
> that
> > > the
> > > > > > > > > > >> configurations
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a
> > > > > transition to
> > > > > > > > DEAD
> > > > > > > > > > >> > trigger
> > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > configurations
> > > > > will be
> > > > > > > > > > >> deleted
> > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to
> store
> > > the
> > > > > > > > configs in
> > > > > > > > > > >> the
> > > > > > > > > > >> > group
> > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> cluster
> > > > > configs and
> > > > > > > > their
> > > > > > > > > > >> > values?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > > > > > > configurations are
> > > > > > > > > > >> > overrides
> > > > > > > > > > >> > > > > >> of
> > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to
> know
> > > all
> > > > > the
> > > > > > > > overrides
> > > > > > > > > > >> > though,
> > > > > > > > > > >> > > > > you
> > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > > coordinators.
> > > > > You
> > > > > > > > cannot
> > > > > > > > > > >> > rely on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> > > configure a
> > > > > group
> > > > > > > > before
> > > > > > > > > > >> > starting
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > >> > > > > >> > > > > > > > > > we considered allowing
> the
> > > > > explicit
> > > > > > > > creation
> > > > > > > > > > >> of
> > > > > > > > > > >> > > > > groups?
> > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> decoupled
> > > from
> > > > > the
> > > > > > > > group
> > > > > > > > > > >> > lifecycle.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > > > > automatically
> > > > > > > > created in
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be lost
> after
> > > the
> > > > > > > > retention
> > > > > > > > > > >> > period of
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > Consumer.subscribe
> > > > > method
> > > > > > > > for the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> `java.util.regex.Pattern` of
> > > do
> > > > > we
> > > > > > > > have to
> > > > > > > > > > >> > introduce
> > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> question. I
> > > > > forgot
> > > > > > > > about
> > > > > > > > > > >> that
> > > > > > > > > > >> > one.
> > > > > > > > > > >> > > > > As
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern`
> is
> > > not
> > > > > fully
> > > > > > > > > > >> compatible
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might be
> > > better
> > > > > to
> > > > > > > > deprecate
> > > > > > > > > > >> > it and
> > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string. We
> would
> > > > > rely on
> > > > > > > > the
> > > > > > > > > > >> server
> > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I
> think
> > > that
> > > > > we
> > > > > > > > could
> > > > > > > > > > >> still
> > > > > > > > > > >> > try to
> > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
> should
> > > > > work, I
> > > > > > > > think, in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine
> > > that we
> > > > > > > > should be
> > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > about
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > IBP/metadata.version
> > > > > from the
> > > > > > > > > > >> feature
> > > > > > > > > > >> > flag.
> > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the default
> for
> > > the
> > > > > feature
> > > > > > > > flag
> > > > > > > > > > >> to
> > > > > > > > > > >> > true
> > > > > > > > > > >> > > > > >> based
> > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once we
> > > want to
> > > > > make
> > > > > > > > it the
> > > > > > > > > > >> > default.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that
> feature
> > > > > flag
> > > > > > > > mainly
> > > > > > > > > > >> > during the
> > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> > > mentioned
> > > > > it, I
> > > > > > > > think,
> > > > > > > > > > >> > because
> > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol
> APIs
> > > > > don't
> > > > > > > > mention
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > required
> > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> > > consistency.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > requires
> > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do
> more
> > > than
> > > > > > > > reading.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> > > current
> > > > > > > > protocol works
> > > > > > > > > > >> > though.
> > > > > > > > > > >> > > > > We
> > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to
> join a
> > > > > group. We
> > > > > > > > could
> > > > > > > > > > >> > consider
> > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that it
> is
> > > worth
> > > > > it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> recreation
> > > > > handled by
> > > > > > > > the
> > > > > > > > > > >> > consumer
> > > > > > > > > > >> > > > > with
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be
> good to
> > > > > have a
> > > > > > > > section
> > > > > > > > > > >> on
> > > > > > > > > > >> > this.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > > > > perspective, the
> > > > > > > > new
> > > > > > > > > > >> topic
> > > > > > > > > > >> > will
> > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will treat
> it
> > > like a
> > > > > > > > topic with
> > > > > > > > > > >> a
> > > > > > > > > > >> > > > > different
> > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> > > fetch/commit
> > > > > > > > offsets
> > > > > > > > > > >> APIs
> > > > > > > > > > >> > do not
> > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would
> > > reuse the
> > > > > > > > offsets
> > > > > > > > > > >> based
> > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > >> > > > > >> > > > > > > > > that we should update those
> > > APIs
> > > > > as well
> > > > > > > > in
> > > > > > > > > > >> order
> > > > > > > > > > >> > to be
> > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> strengthen
> > > the
> > > > > > > > semantics of
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we
> will
> > > > > write the
> > > > > > > > new
> > > > > > > > > > >> > coordinator
> > > > > > > > > > >> > > > > >> in
> > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > >> > > > > >> > > > > > > > > > this is an implementation
> > > > > detail, do
> > > > > > > > we plan
> > > > > > > > > > >> to
> > > > > > > > > > >> > have a
> > > > > > > > > > >> > > > > >> new
> > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> scalability
> > > > > goal when
> > > > > > > > it
> > > > > > > > > > >> comes
> > > > > > > > > > >> > to how
> > > > > > > > > > >> > > > > >> many
> > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> support?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at
> the
> > > > > moment. The
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > >> > > > > >> > > > > > > > > of members per group. We
> will
> > > > > measure
> > > > > > > > this when
> > > > > > > > > > >> > we have
> > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note that
> we
> > > might
> > > > > have
> > > > > > > > other
> > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> having
> > > > > > > > SubscribedTopidIds
> > > > > > > > > > >> > instead
> > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to resolve
> the
> > > > > regex on
> > > > > > > > the
> > > > > > > > > > >> > server, we
> > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> > > difference
> > > > > is that
> > > > > > > > > > >> sending
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the topic
> > > names
> > > > > is less
> > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to
> > > handle
> > > > > if we
> > > > > > > > have
> > > > > > > > > > >> topic
> > > > > > > > > > >> > ids.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> consolidate the
> > > > > metadata
> > > > > > > > lookup
> > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > >> server
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree with
> your
> > > > > point. As
> > > > > > > > a
> > > > > > > > > > >> second
> > > > > > > > > > >> > > > > though,
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here for
> the
> > > > > delete and
> > > > > > > > > > >> > recreation
> > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow
> > > users to
> > > > > > > > subscribe
> > > > > > > > > > >> with
> > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > future because that is the
> only
> > > > > way to be
> > > > > > > > > > >> really
> > > > > > > > > > >> > robust
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at
> 1:38 PM
> > > > > David
> > > > > > > > Jacot <
> > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your comments.
> > > Please
> > > > > find
> > > > > > > > my
> > > > > > > > > > >> answers
> > > > > > > > > > >> > > > > below.
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol
> > > relies on
> > > > > > > > topic IDs
> > > > > > > > > > >> > with the
> > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > >> > > > > >> I
> > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names is
> the
> > > > > right call
> > > > > > > > here.
> > > > > > > > > > >> I
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> Obviously, the
> > > > > KIP does
> > > > > > > > not
> > > > > > > > > > >> > change the
> > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic
> > > IDs.
> > > > > This
> > > > > > > > may be
> > > > > > > > > > >> > something
> > > > > > > > > > >> > > > > >> that
> > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
> would
> > > give
> > > > > better
> > > > > > > > > > >> overall
> > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I think
> that
> > > I
> > > > > should
> > > > > > > > not
> > > > > > > > > > >> have
> > > > > > > > > > >> > > > > mentioned
> > > > > > > > > > >> > > > > >> > this
> > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it.
> We can
> > > > > use an
> > > > > > > > internal
> > > > > > > > > > >> > > > > >> configuration
> > > > > > > > > > >> > > > > >> > > while
> > > > > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
> will be
> > > > > > > > supported. The
> > > > > > > > > > >> > change is
> > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is that
> the
> > > > > cluster
> > > > > > > > uses
> > > > > > > > > > >> topic
> > > > > > > > > > >> > IDs.
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at
> 9:53
> > > PM
> > > > > > > > Guozhang
> > > > > > > > > > >> Wang <
> > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> feedback.
> > > Here
> > > > > are
> > > > > > > > some
> > > > > > > > > > >> replies
> > > > > > > > > > >> > > > > inlined
> > > > > > > > > > >> > > > > >> > > below:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at
> > > 2:53 AM
> > > > > > > > Ismael Juma
> > > > > > > > > > >> <
> > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP.
> This
> > > has
> > > > > the
> > > > > > > > > > >> potential
> > > > > > > > > > >> > to be a
> > > > > > > > > > >> > > > > >> > great
> > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > questions/comments:
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> > > premature to
> > > > > talk
> > > > > > > > about
> > > > > > > > > > >> > target
> > > > > > > > > > >> > > > > >> versions
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> existing
> > > group
> > > > > > > > protocol.
> > > > > > > > > > >> > Unlike
> > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol and
> hence
> > > > > > > > > > >> > deprecation/removal will
> > > > > > > > > > >> > > > > >> be
> > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> applications
> > > > > migrate
> > > > > > > > to the
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > > >> protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you.
> I
> > > think
> > > > > we can
> > > > > > > > > > >> remove
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> proposed
> > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > > Deprecation,
> > > > > and
> > > > > > > > Migration
> > > > > > > > > > >> > Plan` and
> > > > > > > > > > >> > > > > >> > > instead
> > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide in
> the
> > > > > future
> > > > > > > > about
> > > > > > > > > > >> when
> > > > > > > > > > >> > we
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> > > intend to
> > > > > > > > release this
> > > > > > > > > > >> > in 4.x,
> > > > > > > > > > >> > > > > >> but
> > > > > > > > > > >> > > > > >> > it
> > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
> that as
> > > a
> > > > > way to
> > > > > > > > > > >> estimate
> > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > we'd
> > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I
> > > also
> > > > > suggest
> > > > > > > > > > >> removing
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > >> part.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> > > specifically
> > > > > > > > related to
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for
> > > client
> > > > > > > > upgrades.
> > > > > > > > > > >> I.e.
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may be
> done
> > > > > first,
> > > > > > > > and then
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > client
> > > > > > > > > > >> > > > > >> > side,
> > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> "released"
> > > by
> > > > > the
> > > > > > > > time
> > > > > > > > > > >> clients
> > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable
> the
> > > > > feature the
> > > > > > > > > > >> clients
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling
> > > bounce
> > > > > and
> > > > > > > > then set
> > > > > > > > > > >> > the flag
> > > > > > > > > > >> > > > > >> > with a
> > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we
> feel
> > > it's
> > > > > > > > better to be
> > > > > > > > > > >> > released
> > > > > > > > > > >> > > > > >> in a
> > > > > > > > > > >> > > > > >> > > major
> > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh
> out
> > > the
> > > > > > > > details of
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > migration
> > > > > > > > > > >> > > > > >> > > story.
> > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we will
> > > support
> > > > > online
> > > > > > > > > > >> > migrations. Is
> > > > > > > > > > >> > > > > >> that
> > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> detail.
> > > It
> > > > > could
> > > > > > > > also be
> > > > > > > > > > >> > done as a
> > > > > > > > > > >> > > > > >> > > separate
> > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the
> > > part we
> > > > > can
> > > > > > > > be more
> > > > > > > > > > >> > concrete
> > > > > > > > > > >> > > > > >> about
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to your
> > > > > question 2)
> > > > > > > > above).
> > > > > > > > > > >> > We will
> > > > > > > > > > >> > > > > >> work
> > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel
> as we
> > > > > solicit
> > > > > > > > more
> > > > > > > > > > >> > feedback.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that
> we are
> > > > > pushing
> > > > > > > > the
> > > > > > > > > > >> > pattern
> > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like
> there
> > > > > could be
> > > > > > > > some
> > > > > > > > > > >> tricky
> > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for
> > > users to
> > > > > > > > detect that
> > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > need
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > > update
> > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching to
> the
> > > new
> > > > > > > > protocol?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need
> some
> > > > > tooling for
> > > > > > > > > > >> non-java
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > >> > users
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client
> before
> > > > > > > > switching to
> > > > > > > > > > >> the
> > > > > > > > > > >> > new
> > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top
> of my
> > > head
> > > > > > > > though,
> > > > > > > > > > >> maybe
> > > > > > > > > > >> > others
> > > > > > > > > > >> > > > > >> like
> > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the
> last
> > > > > question,
> > > > > > > > will the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> current
> > > regex
> > > > > > > > engine for
> > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to
> keep
> > > using
> > > > > > > > client
> > > > > > > > > > >> based
> > > > > > > > > > >> > regex
> > > > > > > > > > >> > > > > at
> > > > > > > > > > >> > > > > >> > > first to
> > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple and
> > > then
> > > > > migrate
> > > > > > > > to
> > > > > > > > > > >> > server based
> > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not
> thought
> > > > > about
> > > > > > > > that for
> > > > > > > > > > >> > java
> > > > > > > > > > >> > > > > >> clients,
> > > > > > > > > > >> > > > > >> > > and
> > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > > > > compatibility
> > > > > > > > issues do
> > > > > > > > > > >> > you have
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that
> the
> > > group
> > > > > > > > coordinator
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the configurations
> and
> > > that
> > > > > the
> > > > > > > > > > >> > configurations
> > > > > > > > > > >> > > > > will
> > > > > > > > > > >> > > > > >> be
> > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted.
> Will a
> > > > > > > > transition to
> > > > > > > > > > >> DEAD
> > > > > > > > > > >> > > > > trigger
> > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD
> state
> > > is an
> > > > > > > > ending
> > > > > > > > > > >> state
> > > > > > > > > > >> > (we
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the group is
> > > EMPTY
> > > > > and
> > > > > > > > also all
> > > > > > > > > > >> of
> > > > > > > > > > >> > its
> > > > > > > > > > >> > > > > >> > metadata
> > > > > > > > > > >> > > > > >> > > are
> > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD
> this
> > > > > group
> > > > > > > > would
> > > > > > > > > > >> never
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > >> revived.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to
> > > store
> > > > > the
> > > > > > > > configs
> > > > > > > > > > >> in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > group
> > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> > > cluster
> > > > > configs
> > > > > > > > and
> > > > > > > > > > >> their
> > > > > > > > > > >> > > > > values?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good question,
> > > and our
> > > > > > > > thoughts
> > > > > > > > > > >> are
> > > > > > > > > > >> > that
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> > > overrides
> > > > > of the
> > > > > > > > > > >> > cluster-level
> > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per group so
> > > when an
> > > > > > > > admin list
> > > > > > > > > > >> > cluster
> > > > > > > > > > >> > > > > >> > configs
> > > > > > > > > > >> > > > > >> > > it's
> > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> cluster-level
> > > > > > > > "defaults", not
> > > > > > > > > > >> > showing
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone
> > > > > configure a
> > > > > > > > group
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > >> starting
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> allowing
> > > the
> > > > > explicit
> > > > > > > > > > >> > creation of
> > > > > > > > > > >> > > > > >> groups?
> > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> > > decoupled
> > > > > from
> > > > > > > > the group
> > > > > > > > > > >> > > > > lifecycle.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
> created
> > > > > before
> > > > > > > > the group
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > as
> > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> > > requires
> > > > > the
> > > > > > > > > > >> corresponding
> > > > > > > > > > >> > > > > >> request to
> > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> based on
> > > the
> > > > > group
> > > > > > > > id ---
> > > > > > > > > > >> > the only
> > > > > > > > > > >> > > > > >> > thing
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group itself
> is
> > > gone
> > > > > we
> > > > > > > > also check
> > > > > > > > > > >> > if there
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > any
> > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to
> that
> > > group
> > > > > and
> > > > > > > > delete
> > > > > > > > > > >> as
> > > > > > > > > > >> > well.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed
> > > > > introduces an
> > > > > > > > > > >> > asymmetry on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > creation
> > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> config
> > > > > entities,
> > > > > > > > and we
> > > > > > > > > > >> > would like
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > > hear
> > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we
> should
> > > aim
> > > > > for
> > > > > > > > symmetry
> > > > > > > > > > >> > i.e.
> > > > > > > > > > >> > > > > >> totally
> > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not
> > > delete
> > > > > them at
> > > > > > > > all
> > > > > > > > > > >> when
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > >> > is
> > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> deletion
> > > > > operations
> > > > > > > > by
> > > > > > > > > > >> > themselves.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > > > Consumer.subscribe
> > > > > > > > method for
> > > > > > > > > > >> > the Java
> > > > > > > > > > >> > > > > >> > client
> > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > `java.util.regex.Pattern` of
> > > > > do we
> > > > > > > > have
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > > >> introduce an
> > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not need
> to
> > > > > introduce
> > > > > > > > an
> > > > > > > > > > >> > overload, but
> > > > > > > > > > >> > > > > >> I'm
> > > > > > > > > > >> > > > > >> > > all
> > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> compatibility
> > > > > issues
> > > > > > > > that we
> > > > > > > > > > >> may
> > > > > > > > > > >> > > > > overlook.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
> Justine
> > > > > that we
> > > > > > > > should
> > > > > > > > > > >> be
> > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > >> > about
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > > IBP/metadata.version
> > > > > > > > from the
> > > > > > > > > > >> > feature
> > > > > > > > > > >> > > > > >> flag.
> > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> default
> > > for
> > > > > the
> > > > > > > > feature
> > > > > > > > > > >> flag
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > true
> > > > > > > > > > >> > > > > >> > > based on
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
> once we
> > > > > want to
> > > > > > > > make it
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> default.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> protocol
> > > APIs
> > > > > don't
> > > > > > > > mention
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> required
> > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that for
> > > > > consistency.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly
> do
> > > that.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd
> that
> > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > >> > requires
> > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to do
> > > more
> > > > > than
> > > > > > > > reading.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
> myself
> > > as
> > > > > well,
> > > > > > > > but in
> > > > > > > > > > >> the
> > > > > > > > > > >> > end we
> > > > > > > > > > >> > > > > >> > could
> > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative:
> adding
> > > > > Write
> > > > > > > > Group
> > > > > > > > > > >> seems
> > > > > > > > > > >> > an
> > > > > > > > > > >> > > > > >> overkill
> > > > > > > > > > >> > > > > >> > > here
> > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere
> (we
> > > only
> > > > > have
> > > > > > > > Read /
> > > > > > > > > > >> > Delete
> > > > > > > > > > >> > > > > and
> > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to
> hear
> > > others
> > > > > > > > thoughts.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> > > recreation
> > > > > > > > handled by
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > with
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be
> > > good
> > > > > to have
> > > > > > > > a
> > > > > > > > > > >> > section on
> > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > > > > subscription
> > > > > > > > right? Yes
> > > > > > > > > > >> > we can
> > > > > > > > > > >> > > > > >> add a
> > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically the
> > > idea
> > > > > is that
> > > > > > > > > > >> consumer
> > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > be
> > > > > > > > > > >> > > > > >> > > totally
> > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as
> it's
> > > > > handled all
> > > > > > > > by the
> > > > > > > > > > >> > brokers.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions
> we
> > > will
> > > > > write
> > > > > > > > the
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> implementation
> > > > > detail,
> > > > > > > > do we
> > > > > > > > > > >> > plan to
> > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought
> about
> > > > > that. But
> > > > > > > > I think
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > answer
> > > > > > > > > > >> > > > > >> > > should
> > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> > > scalability
> > > > > goal
> > > > > > > > when it
> > > > > > > > > > >> > comes to
> > > > > > > > > > >> > > > > >> how
> > > > > > > > > > >> > > > > >> > > many
> > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> > > support?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> think we
> > > > > should
> > > > > > > > shoot for
> > > > > > > > > > >> > 1000s of
> > > > > > > > > > >> > > > > >> > > members.
> > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals also
> > > depend
> > > > > on the
> > > > > > > > offset
> > > > > > > > > > >> > > > > management
> > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> > > coordinator
> > > > > > > > which we
> > > > > > > > > > >> did
> > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > >> cover in
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a number
> that
> > > > > applies
> > > > > > > > > > >> > universally.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider
> > > having
> > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames
> in
> > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
> resolve
> > > the
> > > > > regex
> > > > > > > > on the
> > > > > > > > > > >> > server,
> > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > can
> > > > > > > > > > >> > > > > >> > > do
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > > > > difference is
> > > > > > > > that
> > > > > > > > > > >> > sending the
> > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > >> > > is
> > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending the
> topic
> > > > > names is
> > > > > > > > less
> > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is easier
> to
> > > > > handle if
> > > > > > > > we have
> > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > ids.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
> still
> > > let
> > > > > the
> > > > > > > > clients
> > > > > > > > > > >> send
> > > > > > > > > > >> > names
> > > > > > > > > > >> > > > > >> is to
> > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names ->
> ids
> > > on
> > > > > the
> > > > > > > > broker /
> > > > > > > > > > >> > admin
> > > > > > > > > > >> > > > > client
> > > > > > > > > > >> > > > > >> > > only.
> > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> topic id
> > > in
> > > > > > > > KIP-516, we
> > > > > > > > > > >> > never
> > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> > > leveraging
> > > > > the
> > > > > > > > related
> > > > > > > > > > >> > newer
> > > > > > > > > > >> > > > > >> versioned
> > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic id
> as
> > > empty
> > > > > UUID.
> > > > > > > > We
> > > > > > > > > > >> want
> > > > > > > > > > >> > to keep
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> > > delegate
> > > > > the
> > > > > > > > reasoning
> > > > > > > > > > >> on
> > > > > > > > > > >> > broker
> > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022
> at
> > > 10:45
> > > > > AM
> > > > > > > > David
> > > > > > > > > > >> Jacot
> > > > > > > > > > >> > > > > >> > > > > > > > >
> <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to
> start a
> > > > > > > > discussion
> > > > > > > > > > >> thread
> > > > > > > > > > >> > on
> > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the
> > > Consumer
> > > > > > > > Rebalance
> > > > > > > > > > >> > Protocol.
> > > > > > > > > > >> > > > > >> With
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> rebalance
> > > > > protocol
> > > > > > > > (for
> > > > > > > > > > >> > consumers)
> > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to
> > > > > implement for
> > > > > > > > > > >> > clients, and
> > > > > > > > > > >> > > > > >> easier
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > > > > >> > > > > >> > > > >
> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a look
> and
> > > let
> > > > > me
> > > > > > > > know what
> > > > > > > > > > >> > you
> > > > > > > > > > >> > > > > think.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away
> from
> > > > > July
> > > > > > > > 18th to
> > > > > > > > > > >> > August 8th.
> > > > > > > > > > >> > > > > >> That
> > > > > > > > > > >> > > > > >> > > gives
> > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read and
> > > digest
> > > > > this
> > > > > > > > long
> > > > > > > > > > >> KIP.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> --
> > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > > > > >
> > > > > > > >
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
>
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Makes sense, thanks David.

On Mon, Sep 12, 2022 at 5:17 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Guozhang,
>
> 1. I have added a reference to the relevant chapter instead of
> repeating the whole thing. Does that work for you?
>
> 2. The "Rebalance Triggers" section you are referring to is about when
> a rebalance should be triggered for the non-upgraded members using the
> old protocol. The section mentions that a rebalance must be triggered
> when a new assignment is installed. This implies that the group epoch
> was updated either by a native member or a non-upgraded member. For
> the latter, the JoinGroup request would be the trigger. I have added a
> reference to the relevant chapter in the "JoinGroup Handling" section
> as well. Does that make sense?
>
> Thanks,
> David
>
> On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello David,
> >
> > Alright I think that's sufficient. Just to make that clear in the doc,
> > could we update:
> >
> > 1) the heartbeat request handling section, stating when coordinator will
> > trigger rebalance based on the HB's member metadata / reason?
> > 2) the "Rebalance Triggers" section to include what we described in
> "Group
> > Epoch - Trigger a rebalance" section as well?
> >
> >
> > Guozhang
> >
> > On Fri, Sep 9, 2022 at 1:28 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi Guozhang,
> > >
> > > I thought that the assignor will always be consulted when the next
> > > heartbeat request is constructed. In other words,
> > > `PartitionAssignor#metadata` will be called for every heartbeat. This
> > > gives the opportunity for the assignor to enforce a rebalance by
> > > setting the reason to a non-zero value or by changing the bytes. Do
> > > you think that this is not sufficient? Are you concerned by the delay?
> > >
> > > Best,
> > > David
> > >
> > > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com>
> wrote:
> > > >
> > > > Hello David,
> > > >
> > > > One of Jun's comments make me thinking:
> > > >
> > > > ```
> > > > In this case, a new assignment is triggered by the client side
> > > > assignor. When constructing the HB, the consumer will always consult
> > > > the client side assignor and propagate the information to the group
> > > > coordinator. In other words, we don't expect users to call
> > > > Consumer#enforceRebalance anymore.
> > > > ```
> > > >
> > > > As I looked at the current PartitionAssignor's interface, we
> actually do
> > > > not have a way yet to instruct how to construct the next HB request,
> e.g.
> > > > when the assignor wants to enforce a new rebalance with a new
> assignment,
> > > > we'd need some customizable APIs inside the PartitionAssignor to
> indicate
> > > > the next HB telling broker about so. WDYT about adding such an API
> on the
> > > > PartitionAssignor?
> > > >
> > > >
> > > > Guozhang
> > > >
> > > >
> > > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot
> <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi Jun,
> > > > >
> > > > > I have updated the KIP to include your feedback. I have also tried
> to
> > > > > clarify the parts which were not cleared.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > > > >
> > > > > > Hi Jun,
> > > > > >
> > > > > > Thanks for your feedback. Let me start by answering your
> questions
> > > > > > inline and I will update the KIP next week.
> > > > > >
> > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem
> to
> > > be
> > > > > fewer
> > > > > > > RPCs during rebalance and more efficient support of wildcard.
> A few
> > > > > > > comments below.
> > > > > >
> > > > > > I would also add that the KIP removes the global sync barrier in
> the
> > > > > > protocol which is essential to improve group stability and
> > > > > > scalability, and the KIP also simplifies the client by moving
> most of
> > > > > > the logic to the server side.
> > > > > >
> > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> rolling
> > > > > changing
> > > > > > > of the partition assignor in the consumers?
> > > > > >
> > > > > > Definitely. The group coordinator will use the assignor used by a
> > > > > > majority of the members. This allows the group to move from one
> > > > > > assignor to another by a roll. This is explained in the Assignor
> > > > > > Selection chapter.
> > > > > >
> > > > > > > 30.2 For each field, could you explain whether it's required in
> > > every
> > > > > > > request or the scenarios when it needs to be filled? For
> example,
> > > it's
> > > > > not
> > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >
> > > > > > The client is expected to set those fields in case of a
> connection
> > > > > > issue (e.g. timeout) or when the fields have changed since the
> last
> > > > > > HB. The server populates those fields as long as the member is
> not
> > > > > > fully reconciled - the member should acknowledge that it has the
> > > > > > expected epoch and assignment. I will clarify this in the KIP.
> > > > > >
> > > > > > > 31. In the current consumer protocol, the rack affinity
> between the
> > > > > client
> > > > > > > and the broker is only considered during fetching, but not
> during
> > > > > assigning
> > > > > > > partitions to consumers. Sometimes, once the assignment is
> made,
> > > there
> > > > > is
> > > > > > > no opportunity for read affinity because no replicas of
> assigned
> > > > > partitions
> > > > > > > are close to the member. I am wondering if we should use this
> > > > > opportunity
> > > > > > > to address this by including rack in GroupMember.
> > > > > >
> > > > > > That's an interesting idea. I don't see any issue with adding the
> > > rack
> > > > > > to the members. I will do so.
> > > > > >
> > > > > > > 32. On the metric side, often, it's useful to know how busy a
> group
> > > > > > > coordinator is. By moving the event loop model, it seems that
> we
> > > could
> > > > > add
> > > > > > > a metric that tracks the fraction of the time the event loop is
> > > doing
> > > > > the
> > > > > > > actual work.
> > > > > >
> > > > > > That's a great idea. I will add it. Thanks.
> > > > > >
> > > > > > > 33. Could we add a section on coordinator failover handling?
> For
> > > > > example,
> > > > > > > does it need to trigger the check if any group with the
> wildcard
> > > > > > > subscription now has a new matching topic?
> > > > > >
> > > > > > Sure. When the new group coordinator takes over, it has to:
> > > > > > * Setup the session timeouts.
> > > > > > * Trigger a new assignment if a client side assignor is used. We
> > > don't
> > > > > > store the information about the member selected to run the
> assignment
> > > > > > so we have to start a new one.
> > > > > > * Update the topics metadata, verify the wildcard subscriptions,
> and
> > > > > > trigger a rebalance if needed.
> > > > > >
> > > > > > > 34. ConsumerGroupMetadataValue,
> > > ConsumerGroupPartitionMetadataValue,
> > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> epoch
> > > > > field
> > > > > > > reflects? For example, does the epoch in
> ConsumerGroupMetadataValue
> > > > > reflect
> > > > > > > the latest group epoch? What about the one in
> > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > ConsumerGroupMemberMetadataValue?
> > > > > >
> > > > > > Sure. I will clarify that but it is always the latest group
> epoch.
> > > > > > When the group state is updated, the group epoch is bumped so we
> use
> > > > > > that one for all the change records related to the update.
> > > > > >
> > > > > > > 35. "the group coordinator will ensure that the following
> > > invariants
> > > > > are
> > > > > > > met: ... All members exists." It's possible for a member not to
> > > get any
> > > > > > > assigned partitions, right?
> > > > > >
> > > > > > That's right. Here I meant that the members provided by the
> assignor
> > > > > > in the assignment must exist in the group. The assignor can not
> make
> > > > > > up new member ids.
> > > > > >
> > > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > > When
> > > > > would
> > > > > > > a consumer rejoin and what member id would be used?
> > > > > >
> > > > > > A member is expected to abandon all its partitions and rejoins
> when
> > > it
> > > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > > > > coordinator will have removed the member from the group. The
> member
> > > > > > can rejoin the group with the same member id but with 0 as
> epoch. Let
> > > > > > me see if I can clarify this in the KIP.
> > > > > >
> > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > reassignment
> > > > > > > by either providing a non-zero reason or by updating the
> assignor
> > > > > > > metadata." Hmm, this seems to be conflicting with the
> deprecation
> > > of
> > > > > > > Consumer#enforeRebalance.
> > > > > >
> > > > > > In this case, a new assignment is triggered by the client side
> > > > > > assignor. When constructing the HB, the consumer will always
> consult
> > > > > > the client side assignor and propagate the information to the
> group
> > > > > > coordinator. In other words, we don't expect users to call
> > > > > > Consumer#enforceRebalance anymore.
> > > > > >
> > > > > > > 38. The reassignment examples are nice. But the section seems
> to
> > > have
> > > > > > > multiple typos.
> > > > > > > 38.1 When the group transitions to epoch 2, B immediately gets
> into
> > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > > epoch=3,
> > > > > > > partitions=[foo-1] too early.
> > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > partitions=[foo-0].
> > > > > >
> > > > > > Sorry for that! I will revise them.
> > > > > >
> > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> from
> > > any
> > > > > old
> > > > > > > version to new one?
> > > > > >
> > > > > > We will support upgrading from the consumer protocol version 3,
> > > > > > introduced in KIP-792. KIP-792 is not implemented yet so the
> earliest
> > > > > > version is unknown at the moment. This is explained in the
> migration
> > > > > > plan chapter.
> > > > > >
> > > > > > Thanks again for your feedback, Jun. I will update the KIP based
> on
> > > it
> > > > > > next week.
> > > > > >
> > > > > > Best,
> > > > > > David
> > > > > >
> > > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <jun@confluent.io.invalid
> >
> > > wrote:
> > > > > > >
> > > > > > > Hi, David,
> > > > > > >
> > > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem
> to
> > > be
> > > > > fewer
> > > > > > > RPCs during rebalance and more efficient support of wildcard.
> A few
> > > > > > > comments below.
> > > > > > >
> > > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support
> rolling
> > > > > changing
> > > > > > > of the partition assignor in the consumers?
> > > > > > > 30.2 For each field, could you explain whether it's required in
> > > every
> > > > > > > request or the scenarios when it needs to be filled? For
> example,
> > > it's
> > > > > not
> > > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > > >
> > > > > > > 31. In the current consumer protocol, the rack affinity
> between the
> > > > > client
> > > > > > > and the broker is only considered during fetching, but not
> during
> > > > > assigning
> > > > > > > partitions to consumers. Sometimes, once the assignment is
> made,
> > > there
> > > > > is
> > > > > > > no opportunity for read affinity because no replicas of
> assigned
> > > > > partitions
> > > > > > > are close to the member. I am wondering if we should use this
> > > > > opportunity
> > > > > > > to address this by including rack in GroupMember.
> > > > > > >
> > > > > > > 32. On the metric side, often, it's useful to know how busy a
> group
> > > > > > > coordinator is. By moving the event loop model, it seems that
> we
> > > could
> > > > > add
> > > > > > > a metric that tracks the fraction of the time the event loop is
> > > doing
> > > > > the
> > > > > > > actual work.
> > > > > > >
> > > > > > > 33. Could we add a section on coordinator failover handling?
> For
> > > > > example,
> > > > > > > does it need to trigger the check if any group with the
> wildcard
> > > > > > > subscription now has a new matching topic?
> > > > > > >
> > > > > > > 34. ConsumerGroupMetadataValue,
> > > ConsumerGroupPartitionMetadataValue,
> > > > > > > ConsumerGroupMemberMetadataValue: Could we document what the
> epoch
> > > > > field
> > > > > > > reflects? For example, does the epoch in
> ConsumerGroupMetadataValue
> > > > > reflect
> > > > > > > the latest group epoch? What about the one in
> > > > > > > ConsumerGroupPartitionMetadataValue and
> > > > > ConsumerGroupMemberMetadataValue?
> > > > > > >
> > > > > > > 35. "the group coordinator will ensure that the following
> > > invariants
> > > > > are
> > > > > > > met: ... All members exists." It's possible for a member not to
> > > get any
> > > > > > > assigned partitions, right?
> > > > > > >
> > > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > > When
> > > > > would
> > > > > > > a consumer rejoin and what member id would be used?
> > > > > > >
> > > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > > reassignment
> > > > > > > by either providing a non-zero reason or by updating the
> assignor
> > > > > > > metadata." Hmm, this seems to be conflicting with the
> deprecation
> > > of
> > > > > > > Consumer#enforeRebalance.
> > > > > > >
> > > > > > > 38. The reassignment examples are nice. But the section seems
> to
> > > have
> > > > > > > multiple typos.
> > > > > > > 38.1 When the group transitions to epoch 2, B immediately gets
> into
> > > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > > epoch=3,
> > > > > > > partitions=[foo-1] too early.
> > > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > > partitions=[foo-0].
> > > > > > >
> > > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade
> from
> > > any
> > > > > old
> > > > > > > version to new one?
> > > > > > >
> > > > > > > Thanks,
> > > > > > >
> > > > > > > Jun
> > > > > > >
> > > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hi all,
> > > > > > > >
> > > > > > > > The KIP states that we will re-implement the coordinator in
> > > Java. I
> > > > > > > > discussed this offline with a few folks and folks are
> concerned
> > > that
> > > > > > > > we could introduce many regressions in the old protocol if
> we do
> > > so.
> > > > > > > > Therefore, I am going to remove this statement from the KIP.
> It
> > > is an
> > > > > > > > implementation detail after all so it does not have to be
> > > decided at
> > > > > > > > this stage. We will likely start by trying to refactor the
> > > current
> > > > > > > > implementation as a first step.
> > > > > > > >
> > > > > > > > Cheers,
> > > > > > > > David
> > > > > > > >
> > > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <
> djacot@confluent.io
> > > >
> > > > > wrote:
> > > > > > > > >
> > > > > > > > > Hi Luke,
> > > > > > > > >
> > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > reconciling,
> > > > > > > > stable,
> > > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > >
> > > > > > > > > This sentence does not refer to those group states but
> rather
> > > to a
> > > > > > > > > state machine replication (SMR). This refers to the entire
> > > state of
> > > > > > > > > group coordinator which is replicated via the log layer. I
> will
> > > > > > > > > clarify this in the KIP.
> > > > > > > > >
> > > > > > > > > > 1.2. What do you mean "each state machine is modelled as
> an
> > > event
> > > > > > > > loop"?
> > > > > > > > >
> > > > > > > > > The idea is to follow a model similar to the new quorum
> > > > > controller. We
> > > > > > > > > will have N threads to process events. Each
> __consumer_offsets
> > > > > > > > > partition is assigned to a unique thread and all the events
> > > (e.g.
> > > > > > > > > requests, callbacks, etc.) are processed by this thread.
> This
> > > > > simplify
> > > > > > > > > concurrency and will enable us to do simulation testing
> for the
> > > > > group
> > > > > > > > > coordinator.
> > > > > > > > >
> > > > > > > > > > 1.3. Why do we need a state machine per
> *__consumer_offsets*
> > > > > > > > partitions?
> > > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > > coordinator?
> > > > > > > > For
> > > > > > > > > > example, if one group coordinator owns 2 consumer
> groups, and
> > > > > both
> > > > > > > > exist in
> > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for
> it,
> > > or
> > > > > 2?
> > > > > > > > >
> > > > > > > > > See 1.1. The confusion comes from there, I think.
> > > > > > > > >
> > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> number
> > > of
> > > > > threads
> > > > > > > > used
> > > > > > > > > > to run the state machines. But I'm wondering if the
> purpose
> > > of
> > > > > the
> > > > > > > > threads
> > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > *__consumer_offsets*
> > > > > > > > > > partitions?), and no heavy computation, why should we
> need
> > > > > > > > multi-threads
> > > > > > > > > > here?
> > > > > > > > >
> > > > > > > > > See 1.2. The idea is to have an ability to shard the
> > > processing as
> > > > > the
> > > > > > > > > computation could be heavy.
> > > > > > > > >
> > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > session
> > > > > > > > timeout not
> > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> min/max
> > > > > session
> > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > >
> > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > > detect
> > > > > client
> > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> minimum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> maximum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > >
> > > > > > > > > This is indeed a mistake. The default session timeout
> should
> > > be 45s
> > > > > > > > > (the current default).
> > > > > > > > >
> > > > > > > > > > 2.2. The default server side assignor are [range,
> uniform],
> > > > > which means
> > > > > > > > > > we'll default to "range" assignor. I'd like to know why
> not
> > > > > uniform
> > > > > > > > one? I
> > > > > > > > > > thought usually users will choose uniform assignor
> (former
> > > sticky
> > > > > > > > assinor)
> > > > > > > > > > for better evenly distribution. Any other reason we
> choose
> > > range
> > > > > > > > assignor
> > > > > > > > > > as default?
> > > > > > > > > > group.consumer.assignors List range, uniform The server
> side
> > > > > assignors.
> > > > > > > > >
> > > > > > > > > The order on the server side has no influence because the
> > > client
> > > > > must
> > > > > > > > > chose the selector that he wants to use. There is no
> default
> > > in the
> > > > > > > > > current proposal. If the assignor is not specified by the
> > > client,
> > > > > the
> > > > > > > > > request is rejected. The default client value for
> > > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > > >
> > > > > > > > > Thanks for your very good comments, Luke. I hope that my
> > > answers
> > > > > help
> > > > > > > > > to clarify things. I will update the KIP as well based on
> your
> > > > > > > > > feedback.
> > > > > > > > >
> > > > > > > > > Cheers,
> > > > > > > > > David
> > > > > > > > >
> > > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <
> showuon@gmail.com>
> > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Hi David,
> > > > > > > > > >
> > > > > > > > > > Thanks for the update.
> > > > > > > > > >
> > > > > > > > > > Some more questions:
> > > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > > > > The new group coordinator will have a state machine per
> > > > > > > > > > *__consumer_offsets* partitions, where each state
> machine is
> > > > > modelled
> > > > > > > > as an
> > > > > > > > > > event loop. Those state machines will be executed in
> > > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > > >
> > > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > > reconciling,
> > > > > > > > stable,
> > > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > > > 1.2. What do you mean "each state machine is modelled as
> an
> > > event
> > > > > > > > loop"?
> > > > > > > > > > 1.3. Why do we need a state machine per
> *__consumer_offsets*
> > > > > > > > partitions?
> > > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > > coordinator?
> > > > > > > > For
> > > > > > > > > > example, if one group coordinator owns 2 consumer
> groups, and
> > > > > both
> > > > > > > > exist in
> > > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for
> it,
> > > or
> > > > > 2?
> > > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the
> number
> > > of
> > > > > threads
> > > > > > > > used
> > > > > > > > > > to run the state machines. But I'm wondering if the
> purpose
> > > of
> > > > > the
> > > > > > > > threads
> > > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > > *__consumer_offsets*
> > > > > > > > > > partitions?), and no heavy computation, why should we
> need
> > > > > > > > multi-threads
> > > > > > > > > > here?
> > > > > > > > > >
> > > > > > > > > > 2. For the default value in the new configs:
> > > > > > > > > > 2.1. The consumer session timeout, why does the default
> > > session
> > > > > > > > timeout not
> > > > > > > > > > locate between min (45s) and max(60s)? I thought the
> min/max
> > > > > session
> > > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > > >
> > > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > > detect
> > > > > client
> > > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > > group.consumer.min.session.timeout.ms int 45s The
> minimum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > > group.consumer.max.session.timeout.ms int 60s The
> maximum
> > > > > session
> > > > > > > > timeout.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > 2.2. The default server side assignor are [range,
> uniform],
> > > > > which means
> > > > > > > > > > we'll default to "range" assignor. I'd like to know why
> not
> > > > > uniform
> > > > > > > > one? I
> > > > > > > > > > thought usually users will choose uniform assignor
> (former
> > > sticky
> > > > > > > > assinor)
> > > > > > > > > > for better evenly distribution. Any other reason we
> choose
> > > range
> > > > > > > > assignor
> > > > > > > > > > as default?
> > > > > > > > > > group.consumer.assignors List range, uniform The server
> side
> > > > > assignors.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Thank you.
> > > > > > > > > > Luke
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <
> showuon@gmail.com
> > > >
> > > > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Hi Sagar,
> > > > > > > > > > >
> > > > > > > > > > > I have some thoughts about Kafka Connect integrating
> with
> > > > > KIP-848,
> > > > > > > > but I
> > > > > > > > > > > think we should have a separate discussion thread for
> the
> > > Kafka
> > > > > > > > Connect
> > > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer
> Rebalance
> > > > > Protocol
> > > > > > > > [1],
> > > > > > > > > > > and let this discussion thread focus on consumer
> rebalance
> > > > > protocol,
> > > > > > > > WDYT?
> > > > > > > > > > >
> > > > > > > > > > > [1]
> > > > > > > > > > >
> > > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >
> > > > > > > > > > > Thank you.
> > > > > > > > > > > Luke
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > > sagarmeansocean@gmail.com>
> > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks like
> > > there's
> > > > > > > > agreement on
> > > > > > > > > > >> using separate APIs for Connect. I would revisit the
> doc
> > > and
> > > > > see
> > > > > > > > what
> > > > > > > > > > >> changes are to be made.
> > > > > > > > > > >>
> > > > > > > > > > >> Thanks!
> > > > > > > > > > >> Sagar.
> > > > > > > > > > >>
> > > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > > >> wrote:
> > > > > > > > > > >>
> > > > > > > > > > >> > Hi Sagar,
> > > > > > > > > > >> >
> > > > > > > > > > >> > Thanks for the feedback and the document. That's
> really
> > > > > helpful. I
> > > > > > > > > > >> > will take a look at it.
> > > > > > > > > > >> >
> > > > > > > > > > >> > Overall, it seems to me that both Connect and the
> > > Consumer
> > > > > could
> > > > > > > > share
> > > > > > > > > > >> > the same underlying "engine". The main difference is
> > > that
> > > > > the
> > > > > > > > Consumer
> > > > > > > > > > >> > assigns topic-partitions to members whereas Connect
> > > assigns
> > > > > tasks
> > > > > > > > to
> > > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > > > > >> > 1) We extend the new proposed APIs to support
> different
> > > > > resource
> > > > > > > > types
> > > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The
> dedicated
> > > APIs
> > > > > > > > would be
> > > > > > > > > > >> > similar to the new ones but different on the
> > > > > content/resources and
> > > > > > > > > > >> > they would rely on the same engine on the
> coordinator
> > > side.
> > > > > > > > > > >> >
> > > > > > > > > > >> > I personally lean towards 2) because I am not a fan
> of
> > > > > > > > overcharging
> > > > > > > > > > >> > APIs to serve different purposes. That being said,
> I am
> > > not
> > > > > > > > opposed to
> > > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > > > > >> >
> > > > > > > > > > >> > I think that we can continue to discuss it here for
> now
> > > in
> > > > > order
> > > > > > > > to
> > > > > > > > > > >> > ensure that this KIP is compatible with what we
> will do
> > > for
> > > > > > > > Connect in
> > > > > > > > > > >> > the future.
> > > > > > > > > > >> >
> > > > > > > > > > >> > Best,
> > > > > > > > > > >> > David
> > > > > > > > > > >> >
> > > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > > > djacot@confluent.io>
> > > > > > > > wrote:
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Hi all,
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > I am back from vacation. I will go through and
> address
> > > > > your
> > > > > > > > comments
> > > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > Cheers,
> > > > > > > > > > >> > > David
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > > > > gharris1727@gmail.com
> > > > > > > > > > >> >
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Hey All!
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > > cooperative
> > > > > > > > rebalancing
> > > > > > > > > > >> > making it
> > > > > > > > > > >> > > > down the stack!
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section
> describes how
> > > > > member
> > > > > > > > epoch
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > advance in step with the group epoch and
> assignment
> > > > > epoch
> > > > > > > > values. I
> > > > > > > > > > >> > think
> > > > > > > > > > >> > > > that this is a good idea for the reasons
> described
> > > in
> > > > > the
> > > > > > > > KIP. When
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > protocol is incrementally assigning partitions
> to a
> > > > > worker,
> > > > > > > > what
> > > > > > > > > > >> member
> > > > > > > > > > >> > > > epoch does each incremental assignment use? Are
> > > member
> > > > > epochs
> > > > > > > > > > >> re-used,
> > > > > > > > > > >> > and
> > > > > > > > > > >> > > > a single member epoch can correspond to multiple
> > > > > different
> > > > > > > > > > >> > (monotonically
> > > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to
> the
> > > group
> > > > > > > > > > >> coordinator? If
> > > > > > > > > > >> > > > not, should custom client-side assignor
> > > implementations
> > > > > > > > interact
> > > > > > > > > > >> with
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > Reason field, and how is its common meaning
> agreed
> > > > > upon? If
> > > > > > > > so, what
> > > > > > > > > > >> > is the
> > > > > > > > > > >> > > > benefit of a distinct Reason field over
> including
> > > such
> > > > > > > > functionality
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 3. The following is included in the KIP:
> "Thanks to
> > > > > this, the
> > > > > > > > input
> > > > > > > > > > >> of
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > client side assignor is entirely driven by the
> group
> > > > > > > > coordinator.
> > > > > > > > > > >> The
> > > > > > > > > > >> > > > consumer is no longer responsible for
> maintaining
> > > any
> > > > > state
> > > > > > > > besides
> > > > > > > > > > >> its
> > > > > > > > > > >> > > > assigned partitions." Does this mean that the
> > > > > client-side
> > > > > > > > assignor
> > > > > > > > > > >> MAY
> > > > > > > > > > >> > > > incorporate additional non-Metadata state (such
> as
> > > > > partition
> > > > > > > > > > >> > throughput,
> > > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or that
> > > > > additional
> > > > > > > > > > >> > non-Metadata
> > > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > > > > >> > > > for
> > > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > > >> > > > and
> > > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > > that
> > > > > > > > seem to
> > > > > > > > > > >> > > > overlap significantly. Is it possible for these
> two
> > > > > > > > implementations
> > > > > > > > > > >> to
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > unified? This would serve to promote feature
> parity
> > > of
> > > > > > > > server-side
> > > > > > > > > > >> and
> > > > > > > > > > >> > > > client-side assignors, and would also facilitate
> > > > > operational
> > > > > > > > > > >> > flexibility in
> > > > > > > > > > >> > > > certain situations. For example, if a
> server-side
> > > > > assignor
> > > > > > > > has some
> > > > > > > > > > >> > poor
> > > > > > > > > > >> > > > behavior and needs a patch, deploying the
> patched
> > > > > assignor to
> > > > > > > > the
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > and switching one consumer group to a
> client-side
> > > > > assignor
> > > > > > > > may be
> > > > > > > > > > >> > faster
> > > > > > > > > > >> > > > and less risky than patching all of the brokers.
> > > With
> > > > > the
> > > > > > > > currently
> > > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > > reimplementation
> > > > > would
> > > > > > > > have
> > > > > > > > > > >> to be
> > > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> > > > > significantly,
> > > > > > > > then it
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > possible that a reimplementation would not be
> > > possible.
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > --
> > > > > > > > > > >> > > > Greg Harris
> > > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > >
> > > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > I created a confluence page to discuss how
> Connect
> > > > > would
> > > > > > > > need to
> > > > > > > > > > >> > change
> > > > > > > > > > >> > > > > based on the new rebalance protocol. Here's
> the
> > > page:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > >
> > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > It's also pretty longish and I have tried to
> keep
> > > a
> > > > > format
> > > > > > > > > > >> similar to
> > > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also, do
> you
> > > > > think this
> > > > > > > > > > >> should
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > moved to a separate discussion thread or is
> this
> > > one
> > > > > fine?
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > Thanks!
> > > > > > > > > > >> > > > > Sagar.
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > wrote:
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka
> Streams.
> > > > > Sure, I
> > > > > > > > would do
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang
> Wang <
> > > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> It would be great if you could come back
> with
> > > some
> > > > > > > > analysis on
> > > > > > > > > > >> > how to
> > > > > > > > > > >> > > > > >> implement the Connect side integration with
> > > the new
> > > > > > > > protocol;
> > > > > > > > > > >> so
> > > > > > > > > > >> > far
> > > > > > > > > > >> > > > > >> besides leveraging on the new "protocol
> type"
> > > we
> > > > > did not
> > > > > > > > yet
> > > > > > > > > > >> think
> > > > > > > > > > >> > > > > through
> > > > > > > > > > >> > > > > >> the Connect side implementations. For
> Streams
> > > > > here's a
> > > > > > > > draft of
> > > > > > > > > > >> > > > > >> integration
> > > > > > > > > > >> > > > > >> plan:
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > >
> > > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > > >> > >
> > > > > > > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > Thank you for your response. The reason I
> > > thought
> > > > > > > > connect can
> > > > > > > > > > >> > also fit
> > > > > > > > > > >> > > > > >> into
> > > > > > > > > > >> > > > > >> > this new scheme is that even today the
> > > connect
> > > > > uses a
> > > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to
> empower
> > > > > > > > rebalances of
> > > > > > > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator
> sets
> > > the
> > > > > > > > > > >> protocolType()
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> connect
> > > > > > > > > > >> > > > > >> > and uses the metadata() method by
> plumbing
> > > into
> > > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > I think the changes to support connect
> would
> > > be
> > > > > > > > similar at a
> > > > > > > > > > >> > high
> > > > > > > > > > >> > > > > level
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > the changes in streams mainly because of
> the
> > > > > Client
> > > > > > > > side
> > > > > > > > > > >> > assignors
> > > > > > > > > > >> > > > > being
> > > > > > > > > > >> > > > > >> > used in both. At an implementation
> level, we
> > > > > might
> > > > > > > > need to
> > > > > > > > > > >> make
> > > > > > > > > > >> > a lot
> > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > changes to get onto this new assignment
> > > protocol
> > > > > like
> > > > > > > > > > >> enhancing
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > JoinGroup request/response and SyncGroup
> and
> > > > > using
> > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > >> > API etc again on similar lines to
> streams (or
> > > > > there
> > > > > > > > might be
> > > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > > >> > > > > >> > would try to perform a detailed analysis
> of
> > > the
> > > > > same
> > > > > > > > and we
> > > > > > > > > > >> > can have
> > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > >> > separate discussion thread for that as
> that
> > > would
> > > > > > > > derail this
> > > > > > > > > > >> > > > > discussion
> > > > > > > > > > >> > > > > >> > thread. Let me know if that sounds good
> to
> > > you.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David
> Jacot
> > > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 1) Yes. That refers to
> `Assignment#error`.
> > > > > Sure, I
> > > > > > > > can
> > > > > > > > > > >> > mention it.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from his
> > > current
> > > > > > > > assignment
> > > > > > > > > > >> to
> > > > > > > > > > >> > his
> > > > > > > > > > >> > > > > >> > > target assignment when he can move to
> > > epoch 3.
> > > > > When
> > > > > > > > that
> > > > > > > > > > >> > happens,
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > member assignment is updated and
> persisted
> > > > > with all
> > > > > > > > its
> > > > > > > > > > >> > assigned
> > > > > > > > > > >> > > > > >> > > partitions even if they are not all
> revoked
> > > > > yet. In
> > > > > > > > other
> > > > > > > > > > >> > words, the
> > > > > > > > > > >> > > > > >> > > member assignment becomes the target
> > > > > assignment.
> > > > > > > > This is
> > > > > > > > > > >> > basically
> > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > >> > > optimization to avoid having to write
> all
> > > the
> > > > > > > > changes to
> > > > > > > > > > >> the
> > > > > > > > > > >> > log.
> > > > > > > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > examples are based on the persisted
> state
> > > so I
> > > > > > > > understand
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> > > confusion. Let me see if I can improve
> > > this in
> > > > > the
> > > > > > > > > > >> > description.
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse
> the
> > > > > protocol
> > > > > > > > with a
> > > > > > > > > > >> > client side
> > > > > > > > > > >> > > > > >> > > assignor if it fits in the protocol.
> The
> > > > > assignment
> > > > > > > > is
> > > > > > > > > > >> about
> > > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> > > Connect
> > > > > fit
> > > > > > > > into this?
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > > >> > > > > >> > > David
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor
> > > > > observations:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error section in
> > > Client
> > > > > Side
> > > > > > > > mode
> > > > > > > > > > >> > Assignment
> > > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > > >> > > > > >> > > > you mentioned => `In this case, the
> > > client
> > > > > side
> > > > > > > > assignor
> > > > > > > > > > >> can
> > > > > > > > > > >> > > > > return
> > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > >> > > > error to the group coordinator`. In
> this
> > > > > case are
> > > > > > > > you
> > > > > > > > > > >> > referring to
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > Assignor returning an AssignmentError
> > > that's
> > > > > > > > listed down
> > > > > > > > > > >> > towards
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > end?
> > > > > > > > > > >> > > > > >> > > > If yes, do you think it would make
> sense
> > > to
> > > > > > > > mention this
> > > > > > > > > > >> > > > > explicitly
> > > > > > > > > > >> > > > > >> > here?
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I
> have a
> > > > > slight
> > > > > > > > > > >> confusion,
> > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> if
> > > > > > > > > > >> > > > > >> > > > others have the same. Consider this
> step:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > When B heartbeats, the group
> coordinator
> > > > > > > > transitions him
> > > > > > > > > > >> to
> > > > > > > > > > >> > epoch
> > > > > > > > > > >> > > > > 3
> > > > > > > > > > >> > > > > >> > > because
> > > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> > > persists
> > > > > the
> > > > > > > > change and
> > > > > > > > > > >> > reply.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > > > > >> > > > > >> > > >       - A - epoch=2,
> partitions=[foo-0,
> > > > > foo-1]
> > > > > > > > > > >> > > > > >> > > >       - B - epoch=3,
> partitions=[foo-2]
> > > > > > > > > > >> > > > > >> > > >       - C - epoch=3,
> partitions=[foo-1]
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions to
> > > epoch 3
> > > > > but
> > > > > > > > cannot
> > > > > > > > > > >> get
> > > > > > > > > > >> > foo-1
> > > > > > > > > > >> > > > > >> yet.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C
> can't
> > > get
> > > > > the
> > > > > > > > foo-1
> > > > > > > > > > >> > partition
> > > > > > > > > > >> > > > > yet,
> > > > > > > > > > >> > > > > >> > but
> > > > > > > > > > >> > > > > >> > > > based on the description above, it
> seems
> > > it
> > > > > > > > already has
> > > > > > > > > > >> it.
> > > > > > > > > > >> > Do you
> > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > >> > > it
> > > > > > > > > > >> > > > > >> > > > would be better to remove it and
> > > populate it
> > > > > only
> > > > > > > > when it
> > > > > > > > > > >> > actually
> > > > > > > > > > >> > > > > >> gets
> > > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > > >> > > > > >> > > > I see this in a lot of other places,
> so
> > > have
> > > > > I
> > > > > > > > > > >> understood it
> > > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Regarding connect , it might be out
> of
> > > scope
> > > > > of
> > > > > > > > this
> > > > > > > > > > >> > discussion,
> > > > > > > > > > >> > > > > but
> > > > > > > > > > >> > > > > >> > from
> > > > > > > > > > >> > > > > >> > > > what I understood it would probably
> be
> > > > > running in
> > > > > > > > client
> > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > >> assignor
> > > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > > >> > > > > >> > > > even on the new rebalance protocol
> as it
> > > has
> > > > > its
> > > > > > > > own
> > > > > > > > > > >> Custom
> > > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David
> > > Jacot
> > > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move
> > > forward
> > > > > with
> > > > > > > > > > >> > specialized API
> > > > > > > > > > >> > > > > >> > > > > instead of relying on one generic
> API.
> > > For
> > > > > > > > Connect, we
> > > > > > > > > > >> > can apply
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > exact same pattern and reuse/share
> the
> > > core
> > > > > > > > > > >> > implementation on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > server side. For the schema
> registry, I
> > > > > think
> > > > > > > > that we
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > >> consider
> > > > > > > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > > > > > > membership/leader
> > > > > > > > > > >> > election.
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM
> Ismael
> > > > > Juma <
> > > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > 1. Regarding
> java.util.regex.Pattern
> > > vs
> > > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > document the differences in more
> > > detail
> > > > > before
> > > > > > > > > > >> deciding
> > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > way
> > > > > > > > > > >> > > > > >> or
> > > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > > > > java.util.regex.Pattern,
> > > > > > > > > > >> they
> > > > > > > > > > >> > expect
> > > > > > > > > > >> > > > > >> > their
> > > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If we
> are
> > > doing
> > > > > > > > something
> > > > > > > > > > >> > different,
> > > > > > > > > > >> > > > > >> then
> > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > >> > > > > > should consider adding an
> overload
> > > with
> > > > > our own
> > > > > > > > > > >> Pattern
> > > > > > > > > > >> > class
> > > > > > > > > > >> > > > > (I
> > > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > > >> > > > > >> > > > > > think we'd want to expose re2j's
> at
> > > this
> > > > > > > > point).
> > > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any
> major new
> > > > > protocol
> > > > > > > > should
> > > > > > > > > > >> > > > > integrate
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > > > > recreation case
> > > > > > > > > > >> > correctly.
> > > > > > > > > > >> > > > > That's
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > main
> > > > > > > > > > >> > > > > >> > > > > > part we need to handle. I agree
> with
> > > > > David
> > > > > > > > that we'd
> > > > > > > > > > >> > want to
> > > > > > > > > > >> > > > > add
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > > >> > > > > >> > > > > > to the relevant protocols that
> don't
> > > > > have it
> > > > > > > > yet and
> > > > > > > > > > >> > that we
> > > > > > > > > > >> > > > > can
> > > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > > >> > > > > >> > > > > > focus on the internals versus
> adding
> > > new
> > > > > APIs
> > > > > > > > to the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs is
> > > required
> > > > > for
> > > > > > > > > > >> reasonable
> > > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about the
> > > > > coordinator
> > > > > > > > storing
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> configs.
> > > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > > centralized
> > > > > in the
> > > > > > > > > > >> metadata
> > > > > > > > > > >> > log for
> > > > > > > > > > >> > > > > >> > > various
> > > > > > > > > > >> > > > > >> > > > > > reasons (auditability,
> visibility,
> > > > > consistency,
> > > > > > > > > > >> etc.).
> > > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > > >> > > > > >> > > am
> > > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > > >> > > > > >> > > > > > sure about automatically deleting
> > > > > configs in a
> > > > > > > > way
> > > > > > > > > > >> that
> > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > >> cannot
> > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > >> > > > > > recovered. A good property for
> modern
> > > > > systems
> > > > > > > > is to
> > > > > > > > > > >> > minimize
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > number
> > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss
> scenarios.
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM
> David
> > > > > Jacot
> > > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are
> > > below:
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> especially
> > > > > the last
> > > > > > > > step
> > > > > > > > > > >> when
> > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol,
> in
> > > which
> > > > > we
> > > > > > > > would
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > protocols /
> > > > > rpcs
> > > > > > > > are
> > > > > > > > > > >> used
> > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> could
> > > > > "mimic"
> > > > > > > > the old
> > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > > need to
> > > > > > > > present
> > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a new
> > > version
> > > > > of KIP
> > > > > > > > which
> > > > > > > > > > >> > includes
> > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> > > "Supporting
> > > > > > > > Online
> > > > > > > > > > >> > Consumer
> > > > > > > > > > >> > > > > Group
> > > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > > Deprecation,
> > > > > and
> > > > > > > > Migration
> > > > > > > > > > >> > Plan". I
> > > > > > > > > > >> > > > > >> think
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > I have to think through a few
> cases
> > > > > now but
> > > > > > > > the
> > > > > > > > > > >> > overall idea
> > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > >> > > > > > > mechanism should be
> understandable.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > > far as
> > > > > > > > KIP-516 the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> client,
> > > but
> > > > > they
> > > > > > > > are not
> > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > > question is,
> > > > > > > > first
> > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > > names
> > > > > -> ids
> > > > > > > > mapping
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > > RPCs
> > > > > and the
> > > > > > > > new
> > > > > > > > > > >> RPCs
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> consider
> > > > > exposing
> > > > > > > > the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > subscribe/assign, as
> > > > > > > > > > >> well
> > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would
> include
> > > > > converting
> > > > > > > > all
> > > > > > > > > > >> the
> > > > > > > > > > >> > offsets
> > > > > > > > > > >> > > > > >> > > related
> > > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this
> > > KIP,
> > > > > the
> > > > > > > > consumer
> > > > > > > > > > >> > would be
> > > > > > > > > > >> > > > > >> able
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > > > fully operate with topic ids.
> That
> > > > > being
> > > > > > > > said, it
> > > > > > > > > > >> > still has
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > > >> > > > > >> > > > > > > the topics names in various
> APIs so
> > > > > having a
> > > > > > > > > > >> mapping
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong
> opinion on
> > > > > this.
> > > > > > > > Here I
> > > > > > > > > > >> > wonder if
> > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I
> > > would
> > > > > rather
> > > > > > > > focus
> > > > > > > > > > >> on
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > internals
> > > > > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > > > > separately if
> > > > > > > > we see
> > > > > > > > > > >> > value in
> > > > > > > > > > >> > > > > >> doing
> > > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's point
> about
> > > > > using
> > > > > > > > topic ids
> > > > > > > > > > >> > in the
> > > > > > > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest,
> I
> > > think
> > > > > that
> > > > > > > > there
> > > > > > > > > > >> is
> > > > > > > > > > >> > one
> > > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > > >> > > > > >> > > in
> > > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer will
> > > have
> > > > > the
> > > > > > > > > > >> opportunity
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> validate
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > the topics exists before
> passing
> > > them
> > > > > into
> > > > > > > > the
> > > > > > > > > > >> group
> > > > > > > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > > coordinator
> > > > > will
> > > > > > > > also
> > > > > > > > > > >> notice
> > > > > > > > > > >> > it but
> > > > > > > > > > >> > > > > >> it
> > > > > > > > > > >> > > > > >> > > does
> > > > > > > > > > >> > > > > >> > > > > > > not really have a way to
> reject an
> > > > > invalid
> > > > > > > > topic in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> response.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> all
> > > other
> > > > > minor
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> question:
> > > > > personally
> > > > > > > > I think
> > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API
> with
> > > > > Pattern,
> > > > > > > > but
> > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > >> > > > > > > > while just documenting that
> our
> > > > > > > > subscription may
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case
> is a
> > > > > very rare
> > > > > > > > > > >> scenario
> > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > subscription may
> > > > > > > > be
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > That could work. I have to
> look at
> > > the
> > > > > > > > differences
> > > > > > > > > > >> > between
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > two
> > > > > > > > > > >> > > > > >> > > > > > > engines to better understand
> the
> > > > > potential
> > > > > > > > issues.
> > > > > > > > > > >> My
> > > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > > >> > > > > >> > > > > > > that would work for all the
> basic
> > > > > regular
> > > > > > > > > > >> > expressions. The
> > > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > > >> > > > > >> > > > > > > between the two are mainly
> about
> > > the
> > > > > various
> > > > > > > > > > >> character
> > > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > > >> > > > > >> > > > > > > wonder what other people think
> > > about
> > > > > this.
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28
> PM
> > > > > Guozhang
> > > > > > > > Wang <
> > > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > > >> > > > > >> > > >
> > > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on the
> high
> > > > > level
> > > > > > > > there are
> > > > > > > > > > >> > two meta
> > > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path,
> especially
> > > > > the last
> > > > > > > > step
> > > > > > > > > > >> when
> > > > > > > > > > >> > > > > clients
> > > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol,
> in
> > > which
> > > > > we
> > > > > > > > would
> > > > > > > > > > >> have a
> > > > > > > > > > >> > > > > window
> > > > > > > > > > >> > > > > >> > where
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > > protocols /
> > > > > rpcs
> > > > > > > > are
> > > > > > > > > > >> used
> > > > > > > > > > >> > by
> > > > > > > > > > >> > > > > >> members
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator
> could
> > > > > "mimic"
> > > > > > > > the old
> > > > > > > > > > >> > behavior
> > > > > > > > > > >> > > > > >> while
> > > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > > need to
> > > > > > > > present
> > > > > > > > > > >> about.
> > > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > > far as
> > > > > > > > KIP-516 the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > only
> > > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin
> client,
> > > but
> > > > > they
> > > > > > > > are not
> > > > > > > > > > >> > exposed
> > > > > > > > > > >> > > > > via
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > > question is,
> > > > > > > > first
> > > > > > > > > > >> > should we
> > > > > > > > > > >> > > > > >> let
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > > names
> > > > > -> ids
> > > > > > > > mapping
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > fully
> > > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > > RPCs
> > > > > and the
> > > > > > > > new
> > > > > > > > > > >> RPCs
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever
> consider
> > > > > exposing
> > > > > > > > the
> > > > > > > > > > >> > topic ids
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > > subscribe/assign, as
> > > > > > > > > > >> well
> > > > > > > > > > >> > as in
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on
> all
> > > other
> > > > > minor
> > > > > > > > > > >> questions
> > > > > > > > > > >> > > > > except
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)`
> question:
> > > > > personally
> > > > > > > > I think
> > > > > > > > > > >> > it's not
> > > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API
> with
> > > > > Pattern,
> > > > > > > > but
> > > > > > > > > > >> > instead we
> > > > > > > > > > >> > > > > >> still
> > > > > > > > > > >> > > > > >> > > use
> > > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > > >> > > > > >> > > > > > > > while just documenting that
> our
> > > > > > > > subscription may
> > > > > > > > > > >> be
> > > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case
> is a
> > > > > very rare
> > > > > > > > > > >> scenario
> > > > > > > > > > >> > I felt
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > an
> > > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > > subscription may
> > > > > > > > be
> > > > > > > > > > >> more
> > > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23
> AM
> > > > > David Jacot
> > > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your feedback.
> Let
> > > me
> > > > > answer
> > > > > > > > your
> > > > > > > > > > >> > questions
> > > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's
> premature to
> > > > > talk about
> > > > > > > > > > >> target
> > > > > > > > > > >> > > > > versions
> > > > > > > > > > >> > > > > >> for
> > > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > > removal of the existing
> group
> > > > > protocol.
> > > > > > > > > > >> Unlike
> > > > > > > > > > >> > KRaft,
> > > > > > > > > > >> > > > > >> this
> > > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > > > > > > deprecation/removal
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > > how quickly applications
> > > migrate
> > > > > to
> > > > > > > > the new
> > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> > > remove it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we
> intend to
> > > > > release
> > > > > > > > this in
> > > > > > > > > > >> > 4.x, but
> > > > > > > > > > >> > > > > it
> > > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that as
> a
> > > way to
> > > > > > > > estimate
> > > > > > > > > > >> when
> > > > > > > > > > >> > we'd
> > > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I
> also
> > > > > suggest
> > > > > > > > removing
> > > > > > > > > > >> > this part.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my
> reasoning. As
> > > > > > > > explained, I
> > > > > > > > > > >> plan
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java while
> we
> > > > > implement
> > > > > > > > the new
> > > > > > > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> This
> > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > >> > > > > > > > > that the internals will be
> > > slightly
> > > > > > > > different
> > > > > > > > > > >> > (e.g.
> > > > > > > > > > >> > > > > >> threading
> > > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to
> tighten
> > > the
> > > > > > > > switch from
> > > > > > > > > > >> > the old
> > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new
> group
> > > > > coordinator
> > > > > > > > to a
> > > > > > > > > > >> > major
> > > > > > > > > > >> > > > > >> release.
> > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to
> use a
> > > flag
> > > > > to do
> > > > > > > > the
> > > > > > > > > > >> > switch
> > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out
> the
> > > > > details of
> > > > > > > > the
> > > > > > > > > > >> > migration
> > > > > > > > > > >> > > > > >> story.
> > > > > > > > > > >> > > > > >> > > It
> > > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will
> support
> > > > > online
> > > > > > > > > > >> migrations.
> > > > > > > > > > >> > Is
> > > > > > > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > explain this in detail.
> It
> > > could
> > > > > also
> > > > > > > > be done
> > > > > > > > > > >> > as a
> > > > > > > > > > >> > > > > >> separate
> > > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support online
> > > > > migrations
> > > > > > > > for the
> > > > > > > > > > >> > group.
> > > > > > > > > > >> > > > > That
> > > > > > > > > > >> > > > > >> > > means
> > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > a group using the old
> protocol
> > > > > will be
> > > > > > > > able to
> > > > > > > > > > >> > switch to
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain how
> that
> > > > > will work
> > > > > > > > > > >> though.
> > > > > > > > > > >> > It is
> > > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> > > upgraded or
> > > > > > > > rolled to a
> > > > > > > > > > >> > software
> > > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator.
> Both the
> > > > > old and
> > > > > > > > the new
> > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > >> > > will
> > > > > > > > > > >> > > > > >> > > > > > > > > support the old protocol
> and
> > > rely
> > > > > on the
> > > > > > > > same
> > > > > > > > > > >> > persisted
> > > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > > >> > > > > >> > > > > > > > > they can work together.
> This
> > > point
> > > > > is an
> > > > > > > > > > >> offline
> > > > > > > > > > >> > > > > >> migration.
> > > > > > > > > > >> > > > > >> > We
> > > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > > >> > > > > >> > > > > > > > > do this one live because it
> > > would
> > > > > require
> > > > > > > > > > >> > shutting down
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > current
> > > > > > > > > > >> > > > > >> > > > > > > > > coordinator and starting
> up the
> > > > > new one
> > > > > > > > and
> > > > > > > > > > >> that
> > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > >> cause
> > > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > > > > version/IBP
> > > > > > > > must be
> > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> X
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > > >> > > > > >> > > > > > > > > to enable the new protocol.
> > > This
> > > > > cannot
> > > > > > > > be done
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > 1)
> > > > > > > > > > >> > > > > >> is
> > > > > > > > > > >> > > > > >> > > > > > > > > terminated because the old
> > > > > coordinator
> > > > > > > > doesn't
> > > > > > > > > > >> > support
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > new
> > > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> > > upgraded
> > > > > to a
> > > > > > > > version
> > > > > > > > > > >> > > > > supporting
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > > >> > > > > >> > > > > > > > > migration (must have
> KIP-792).
> > > If
> > > > > the
> > > > > > > > consumer
> > > > > > > > > > >> is
> > > > > > > > > > >> > > > > already
> > > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at
> this
> > > point.
> > > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be
> rolled
> > > > > with the
> > > > > > > > > > >> feature
> > > > > > > > > > >> > flag
> > > > > > > > > > >> > > > > >> turned
> > > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > consumer group is
> automatically
> > > > > > > > converted when
> > > > > > > > > > >> > the first
> > > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol
> joins
> > > the
> > > > > group.
> > > > > > > > While
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > members
> > > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> > > upgraded,
> > > > > the old
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > is
> > > > > > > > > > >> > > > > >> proxied
> > > > > > > > > > >> > > > > >> > > into
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of this
> in
> > > the
> > > > > KIP.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are
> > > > > pushing the
> > > > > > > > pattern
> > > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like there
> > > could be
> > > > > some
> > > > > > > > tricky
> > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for
> users to
> > > > > detect
> > > > > > > > that
> > > > > > > > > > >> they
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > update
> > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > > >> > > > > >> > > > > > > > > > before switching to the
> new
> > > > > protocol?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit
> more
> > > > > optimistic
> > > > > > > > than
> > > > > > > > > > >> you
> > > > > > > > > > >> > on this
> > > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > > >> > > > > >> > > > > > > > > believe that the majority
> of
> > > the
> > > > > cases
> > > > > > > > are
> > > > > > > > > > >> simple
> > > > > > > > > > >> > > > > regexes
> > > > > > > > > > >> > > > > >> > which
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > work with the new engine.
> The
> > > > > > > > coordinator will
> > > > > > > > > > >> > verify
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer if
> the
> > > > > regex is
> > > > > > > > not
> > > > > > > > > > >> valid.
> > > > > > > > > > >> > > > > Coming
> > > > > > > > > > >> > > > > >> > back
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the
> worst
> > > case,
> > > > > the
> > > > > > > > first
> > > > > > > > > > >> > upgraded
> > > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > > >> > > > > >> > > > > > > > > the group will be rejected.
> > > This
> > > > > should
> > > > > > > > be used
> > > > > > > > > > >> > as the
> > > > > > > > > > >> > > > > >> last
> > > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> > > validate
> > > > > their
> > > > > > > > regex
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test them
> with
> > > > > another
> > > > > > > > group.
> > > > > > > > > > >> For
> > > > > > > > > > >> > > > > >> instance,
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod
> > > > > environment.
> > > > > > > > > > >> Another
> > > > > > > > > > >> > way
> > > > > > > > > > >> > > > > >> would be
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > extend the consumer-group
> tool
> > > to
> > > > > > > > provide a
> > > > > > > > > > >> regex
> > > > > > > > > > >> > > > > >> validation
> > > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> > > enough in
> > > > > your
> > > > > > > > > > >> opinion?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> > > question,
> > > > > will
> > > > > > > > the
> > > > > > > > > > >> Java
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > >> allow
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > > >> > > > > >> > > > > > > > > > stick with the current
> regex
> > > > > engine for
> > > > > > > > > > >> > compatibility
> > > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to keep
> using
> > > > > client
> > > > > > > > based
> > > > > > > > > > >> > regex at
> > > > > > > > > > >> > > > > >> first
> > > > > > > > > > >> > > > > >> > to
> > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and
> then
> > > > > migrate to
> > > > > > > > server
> > > > > > > > > > >> > based
> > > > > > > > > > >> > > > > >> regexes
> > > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I understand your point
> but I
> > > am
> > > > > > > > concerned that
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in this
> mode.
> > > That
> > > > > > > > would go
> > > > > > > > > > >> > against our
> > > > > > > > > > >> > > > > >> goal
> > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client
> because
> > > we
> > > > > would
> > > > > > > > have to
> > > > > > > > > > >> > continue
> > > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the client
> > > side. I
> > > > > would
> > > > > > > > rather
> > > > > > > > > > >> > not do
> > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that the
> group
> > > > > > > > coordinator
> > > > > > > > > > >> will
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > > >> > > > > >> > > > > > > > > > the configurations and
> that
> > > the
> > > > > > > > > > >> configurations
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a
> > > > > transition to
> > > > > > > > DEAD
> > > > > > > > > > >> > trigger
> > > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > > >> > > > > >> > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > > configurations
> > > > > will be
> > > > > > > > > > >> deleted
> > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to
> store
> > > the
> > > > > > > > configs in
> > > > > > > > > > >> the
> > > > > > > > > > >> > group
> > > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all
> cluster
> > > > > configs and
> > > > > > > > their
> > > > > > > > > > >> > values?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > > > > > > configurations are
> > > > > > > > > > >> > overrides
> > > > > > > > > > >> > > > > >> of
> > > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to
> know
> > > all
> > > > > the
> > > > > > > > overrides
> > > > > > > > > > >> > though,
> > > > > > > > > > >> > > > > you
> > > > > > > > > > >> > > > > >> > > would
> > > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > > coordinators.
> > > > > You
> > > > > > > > cannot
> > > > > > > > > > >> > rely on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> > > configure a
> > > > > group
> > > > > > > > before
> > > > > > > > > > >> > starting
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > > >> > > > > >> > > > > > > > > > we considered allowing
> the
> > > > > explicit
> > > > > > > > creation
> > > > > > > > > > >> of
> > > > > > > > > > >> > > > > groups?
> > > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > configs could be
> decoupled
> > > from
> > > > > the
> > > > > > > > group
> > > > > > > > > > >> > lifecycle.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > > > > automatically
> > > > > > > > created in
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > > >> > > > > >> > > > > > > > > the configs will be lost
> after
> > > the
> > > > > > > > retention
> > > > > > > > > > >> > period of
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > > Consumer.subscribe
> > > > > method
> > > > > > > > for the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > >> > > still
> > > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> `java.util.regex.Pattern` of
> > > do
> > > > > we
> > > > > > > > have to
> > > > > > > > > > >> > introduce
> > > > > > > > > > >> > > > > an
> > > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > That's a very group
> question. I
> > > > > forgot
> > > > > > > > about
> > > > > > > > > > >> that
> > > > > > > > > > >> > one.
> > > > > > > > > > >> > > > > As
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern`
> is
> > > not
> > > > > fully
> > > > > > > > > > >> compatible
> > > > > > > > > > >> > with
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might be
> > > better
> > > > > to
> > > > > > > > deprecate
> > > > > > > > > > >> > it and
> > > > > > > > > > >> > > > > >> use an
> > > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > > >> > > > > >> > > > > > > > > which takes a string. We
> would
> > > > > rely on
> > > > > > > > the
> > > > > > > > > > >> server
> > > > > > > > > > >> > side
> > > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I
> think
> > > that
> > > > > we
> > > > > > > > could
> > > > > > > > > > >> still
> > > > > > > > > > >> > try to
> > > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That
> should
> > > > > work, I
> > > > > > > > think, in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> majority
> > > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine
> > > that we
> > > > > > > > should be
> > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > about
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > > IBP/metadata.version
> > > > > from the
> > > > > > > > > > >> feature
> > > > > > > > > > >> > flag.
> > > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > > >> > > > > >> > > we
> > > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > > can switch the default
> for
> > > the
> > > > > feature
> > > > > > > > flag
> > > > > > > > > > >> to
> > > > > > > > > > >> > true
> > > > > > > > > > >> > > > > >> based
> > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once we
> > > want to
> > > > > make
> > > > > > > > it the
> > > > > > > > > > >> > default.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that
> feature
> > > > > flag
> > > > > > > > mainly
> > > > > > > > > > >> > during the
> > > > > > > > > > >> > > > > >> > > development
> > > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> > > mentioned
> > > > > it, I
> > > > > > > > think,
> > > > > > > > > > >> > because
> > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > > could
> > > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol
> APIs
> > > > > don't
> > > > > > > > mention
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > required
> > > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> > > consistency.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > requires
> > > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do
> more
> > > than
> > > > > > > > reading.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> > > current
> > > > > > > > protocol works
> > > > > > > > > > >> > though.
> > > > > > > > > > >> > > > > We
> > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to
> join a
> > > > > group. We
> > > > > > > > could
> > > > > > > > > > >> > consider
> > > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that it
> is
> > > worth
> > > > > it.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic
> recreation
> > > > > handled by
> > > > > > > > the
> > > > > > > > > > >> > consumer
> > > > > > > > > > >> > > > > with
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be
> good to
> > > > > have a
> > > > > > > > section
> > > > > > > > > > >> on
> > > > > > > > > > >> > this.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > > > > perspective, the
> > > > > > > > new
> > > > > > > > > > >> topic
> > > > > > > > > > >> > will
> > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will treat
> it
> > > like a
> > > > > > > > topic with
> > > > > > > > > > >> a
> > > > > > > > > > >> > > > > different
> > > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> > > fetch/commit
> > > > > > > > offsets
> > > > > > > > > > >> APIs
> > > > > > > > > > >> > do not
> > > > > > > > > > >> > > > > >> > support
> > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would
> > > reuse the
> > > > > > > > offsets
> > > > > > > > > > >> based
> > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > >> > same.
> > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > > >> > > > > >> > > > > > > > > that we should update those
> > > APIs
> > > > > as well
> > > > > > > > in
> > > > > > > > > > >> order
> > > > > > > > > > >> > to be
> > > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > > >> > > > > >> > > > > > > > > to end. That would
> strengthen
> > > the
> > > > > > > > semantics of
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we
> will
> > > > > write the
> > > > > > > > new
> > > > > > > > > > >> > coordinator
> > > > > > > > > > >> > > > > >> in
> > > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > >> > > > > >> > > > > > > > > > this is an implementation
> > > > > detail, do
> > > > > > > > we plan
> > > > > > > > > > >> to
> > > > > > > > > > >> > have a
> > > > > > > > > > >> > > > > >> new
> > > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a
> scalability
> > > > > goal when
> > > > > > > > it
> > > > > > > > > > >> comes
> > > > > > > > > > >> > to how
> > > > > > > > > > >> > > > > >> many
> > > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can
> support?
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at
> the
> > > > > moment. The
> > > > > > > > > > >> protocol
> > > > > > > > > > >> > should
> > > > > > > > > > >> > > > > >> > > support
> > > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > > >> > > > > >> > > > > > > > > of members per group. We
> will
> > > > > measure
> > > > > > > > this when
> > > > > > > > > > >> > we have
> > > > > > > > > > >> > > > > a
> > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note that
> we
> > > might
> > > > > have
> > > > > > > > other
> > > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > > >> > > > > >> > down
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider
> having
> > > > > > > > SubscribedTopidIds
> > > > > > > > > > >> > instead
> > > > > > > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > >> Is
> > > > > > > > > > >> > > > > >> > the
> > > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > since we have to resolve
> the
> > > > > regex on
> > > > > > > > the
> > > > > > > > > > >> > server, we
> > > > > > > > > > >> > > > > >> can do
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> > > difference
> > > > > is that
> > > > > > > > > > >> sending
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> regex is
> > > > > > > > > > >> > > > > >> > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the topic
> > > names
> > > > > is less
> > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to
> > > handle
> > > > > if we
> > > > > > > > have
> > > > > > > > > > >> topic
> > > > > > > > > > >> > ids.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > The idea was to
> consolidate the
> > > > > metadata
> > > > > > > > lookup
> > > > > > > > > > >> > on the
> > > > > > > > > > >> > > > > >> server
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree with
> your
> > > > > point. As
> > > > > > > > a
> > > > > > > > > > >> second
> > > > > > > > > > >> > > > > though,
> > > > > > > > > > >> > > > > >> > using
> > > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here for
> the
> > > > > delete and
> > > > > > > > > > >> > recreation
> > > > > > > > > > >> > > > > case.
> > > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow
> > > users to
> > > > > > > > subscribe
> > > > > > > > > > >> with
> > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > >> ids
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > future because that is the
> only
> > > > > way to be
> > > > > > > > > > >> really
> > > > > > > > > > >> > robust
> > > > > > > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at
> 1:38 PM
> > > > > David
> > > > > > > > Jacot <
> > > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your comments.
> > > Please
> > > > > find
> > > > > > > > my
> > > > > > > > > > >> answers
> > > > > > > > > > >> > > > > below.
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol
> > > relies on
> > > > > > > > topic IDs
> > > > > > > > > > >> > with the
> > > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > > >> > > > > >> I
> > > > > > > > > > >> > > > > >> > am
> > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names is
> the
> > > > > right call
> > > > > > > > here.
> > > > > > > > > > >> I
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > think
> > > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > > >> > > > > >> > > > > > > > > > a little more.
> Obviously, the
> > > > > KIP does
> > > > > > > > not
> > > > > > > > > > >> > change the
> > > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic
> > > IDs.
> > > > > This
> > > > > > > > may be
> > > > > > > > > > >> > something
> > > > > > > > > > >> > > > > >> that
> > > > > > > > > > >> > > > > >> > we
> > > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > include though as it
> would
> > > give
> > > > > better
> > > > > > > > > > >> overall
> > > > > > > > > > >> > > > > >> guarantee in
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I think
> that
> > > I
> > > > > should
> > > > > > > > not
> > > > > > > > > > >> have
> > > > > > > > > > >> > > > > mentioned
> > > > > > > > > > >> > > > > >> > this
> > > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it.
> We can
> > > > > use an
> > > > > > > > internal
> > > > > > > > > > >> > > > > >> configuration
> > > > > > > > > > >> > > > > >> > > while
> > > > > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types
> will be
> > > > > > > > supported. The
> > > > > > > > > > >> > change is
> > > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is that
> the
> > > > > cluster
> > > > > > > > uses
> > > > > > > > > > >> topic
> > > > > > > > > > >> > IDs.
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at
> 9:53
> > > PM
> > > > > > > > Guozhang
> > > > > > > > > > >> Wang <
> > > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the
> feedback.
> > > Here
> > > > > are
> > > > > > > > some
> > > > > > > > > > >> replies
> > > > > > > > > > >> > > > > inlined
> > > > > > > > > > >> > > > > >> > > below:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at
> > > 2:53 AM
> > > > > > > > Ismael Juma
> > > > > > > > > > >> <
> > > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP.
> This
> > > has
> > > > > the
> > > > > > > > > > >> potential
> > > > > > > > > > >> > to be a
> > > > > > > > > > >> > > > > >> > great
> > > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > > questions/comments:
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> > > premature to
> > > > > talk
> > > > > > > > about
> > > > > > > > > > >> > target
> > > > > > > > > > >> > > > > >> versions
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the
> existing
> > > group
> > > > > > > > protocol.
> > > > > > > > > > >> > Unlike
> > > > > > > > > > >> > > > > >> KRaft,
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol and
> hence
> > > > > > > > > > >> > deprecation/removal will
> > > > > > > > > > >> > > > > >> be
> > > > > > > > > > >> > > > > >> > > heavily
> > > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly
> applications
> > > > > migrate
> > > > > > > > to the
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > > >> protocol.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you.
> I
> > > think
> > > > > we can
> > > > > > > > > > >> remove
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> proposed
> > > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > > Deprecation,
> > > > > and
> > > > > > > > Migration
> > > > > > > > > > >> > Plan` and
> > > > > > > > > > >> > > > > >> > > instead
> > > > > > > > > > >> > > > > >> > > > > just
> > > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide in
> the
> > > > > future
> > > > > > > > about
> > > > > > > > > > >> when
> > > > > > > > > > >> > we
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> > > intend to
> > > > > > > > release this
> > > > > > > > > > >> > in 4.x,
> > > > > > > > > > >> > > > > >> but
> > > > > > > > > > >> > > > > >> > it
> > > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added
> that as
> > > a
> > > > > way to
> > > > > > > > > > >> estimate
> > > > > > > > > > >> > when
> > > > > > > > > > >> > > > > we'd
> > > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I
> > > also
> > > > > suggest
> > > > > > > > > > >> removing
> > > > > > > > > > >> > this
> > > > > > > > > > >> > > > > >> part.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> > > specifically
> > > > > > > > related to
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for
> > > client
> > > > > > > > upgrades.
> > > > > > > > > > >> I.e.
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may be
> done
> > > > > first,
> > > > > > > > and then
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > client
> > > > > > > > > > >> > > > > >> > side,
> > > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as
> "released"
> > > by
> > > > > the
> > > > > > > > time
> > > > > > > > > > >> clients
> > > > > > > > > > >> > > > > >> > > implementations
> > > > > > > > > > >> > > > > >> > > > > are
> > > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable
> the
> > > > > feature the
> > > > > > > > > > >> clients
> > > > > > > > > > >> > need to
> > > > > > > > > > >> > > > > >> > first
> > > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling
> > > bounce
> > > > > and
> > > > > > > > then set
> > > > > > > > > > >> > the flag
> > > > > > > > > > >> > > > > >> > with a
> > > > > > > > > > >> > > > > >> > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we
> feel
> > > it's
> > > > > > > > better to be
> > > > > > > > > > >> > released
> > > > > > > > > > >> > > > > >> in a
> > > > > > > > > > >> > > > > >> > > major
> > > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh
> out
> > > the
> > > > > > > > details of
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > migration
> > > > > > > > > > >> > > > > >> > > story.
> > > > > > > > > > >> > > > > >> > > > > It
> > > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we will
> > > support
> > > > > online
> > > > > > > > > > >> > migrations. Is
> > > > > > > > > > >> > > > > >> that
> > > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in
> detail.
> > > It
> > > > > could
> > > > > > > > also be
> > > > > > > > > > >> > done as a
> > > > > > > > > > >> > > > > >> > > separate
> > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the
> > > part we
> > > > > can
> > > > > > > > be more
> > > > > > > > > > >> > concrete
> > > > > > > > > > >> > > > > >> about
> > > > > > > > > > >> > > > > >> > > for
> > > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to your
> > > > > question 2)
> > > > > > > > above).
> > > > > > > > > > >> > We will
> > > > > > > > > > >> > > > > >> work
> > > > > > > > > > >> > > > > >> > on
> > > > > > > > > > >> > > > > >> > > > > making
> > > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel
> as we
> > > > > solicit
> > > > > > > > more
> > > > > > > > > > >> > feedback.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that
> we are
> > > > > pushing
> > > > > > > > the
> > > > > > > > > > >> > pattern
> > > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like
> there
> > > > > could be
> > > > > > > > some
> > > > > > > > > > >> tricky
> > > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for
> > > users to
> > > > > > > > detect that
> > > > > > > > > > >> > they
> > > > > > > > > > >> > > > > need
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > > update
> > > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching to
> the
> > > new
> > > > > > > > protocol?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need
> some
> > > > > tooling for
> > > > > > > > > > >> non-java
> > > > > > > > > > >> > client
> > > > > > > > > > >> > > > > >> > users
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client
> before
> > > > > > > > switching to
> > > > > > > > > > >> the
> > > > > > > > > > >> > new
> > > > > > > > > > >> > > > > >> > protocol.
> > > > > > > > > > >> > > > > >> > > I
> > > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top
> of my
> > > head
> > > > > > > > though,
> > > > > > > > > > >> maybe
> > > > > > > > > > >> > others
> > > > > > > > > > >> > > > > >> like
> > > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the
> last
> > > > > question,
> > > > > > > > will the
> > > > > > > > > > >> > Java
> > > > > > > > > > >> > > > > >> client
> > > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the
> current
> > > regex
> > > > > > > > engine for
> > > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to
> keep
> > > using
> > > > > > > > client
> > > > > > > > > > >> based
> > > > > > > > > > >> > regex
> > > > > > > > > > >> > > > > at
> > > > > > > > > > >> > > > > >> > > first to
> > > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple and
> > > then
> > > > > migrate
> > > > > > > > to
> > > > > > > > > > >> > server based
> > > > > > > > > > >> > > > > >> > > regexes
> > > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not
> thought
> > > > > about
> > > > > > > > that for
> > > > > > > > > > >> > java
> > > > > > > > > > >> > > > > >> clients,
> > > > > > > > > > >> > > > > >> > > and
> > > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > > > > compatibility
> > > > > > > > issues do
> > > > > > > > > > >> > you have
> > > > > > > > > > >> > > > > in
> > > > > > > > > > >> > > > > >> > > mind?
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that
> the
> > > group
> > > > > > > > coordinator
> > > > > > > > > > >> > will be
> > > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the configurations
> and
> > > that
> > > > > the
> > > > > > > > > > >> > configurations
> > > > > > > > > > >> > > > > will
> > > > > > > > > > >> > > > > >> be
> > > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted.
> Will a
> > > > > > > > transition to
> > > > > > > > > > >> DEAD
> > > > > > > > > > >> > > > > trigger
> > > > > > > > > > >> > > > > >> > > deletion
> > > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD
> state
> > > is an
> > > > > > > > ending
> > > > > > > > > > >> state
> > > > > > > > > > >> > (we
> > > > > > > > > > >> > > > > would
> > > > > > > > > > >> > > > > >> > only
> > > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > state when the group is
> > > EMPTY
> > > > > and
> > > > > > > > also all
> > > > > > > > > > >> of
> > > > > > > > > > >> > its
> > > > > > > > > > >> > > > > >> > metadata
> > > > > > > > > > >> > > > > >> > > are
> > > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD
> this
> > > > > group
> > > > > > > > would
> > > > > > > > > > >> never
> > > > > > > > > > >> > be
> > > > > > > > > > >> > > > > >> revived.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to
> > > store
> > > > > the
> > > > > > > > configs
> > > > > > > > > > >> in
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > group
> > > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> > > cluster
> > > > > configs
> > > > > > > > and
> > > > > > > > > > >> their
> > > > > > > > > > >> > > > > values?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good question,
> > > and our
> > > > > > > > thoughts
> > > > > > > > > > >> are
> > > > > > > > > > >> > that
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > so-called
> > > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> > > overrides
> > > > > of the
> > > > > > > > > > >> > cluster-level
> > > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > > > >> > > > > >> > > > > > > > > > > customized per group so
> > > when an
> > > > > > > > admin list
> > > > > > > > > > >> > cluster
> > > > > > > > > > >> > > > > >> > configs
> > > > > > > > > > >> > > > > >> > > it's
> > > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > list just the
> cluster-level
> > > > > > > > "defaults", not
> > > > > > > > > > >> > showing
> > > > > > > > > > >> > > > > >> any
> > > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone
> > > > > configure a
> > > > > > > > group
> > > > > > > > > > >> > before
> > > > > > > > > > >> > > > > >> starting
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered
> allowing
> > > the
> > > > > explicit
> > > > > > > > > > >> > creation of
> > > > > > > > > > >> > > > > >> groups?
> > > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> > > decoupled
> > > > > from
> > > > > > > > the group
> > > > > > > > > > >> > > > > lifecycle.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be
> created
> > > > > before
> > > > > > > > the group
> > > > > > > > > > >> > itself
> > > > > > > > > > >> > > > > as
> > > > > > > > > > >> > > > > >> an
> > > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> > > requires
> > > > > the
> > > > > > > > > > >> corresponding
> > > > > > > > > > >> > > > > >> request to
> > > > > > > > > > >> > > > > >> > > be
> > > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator
> based on
> > > the
> > > > > group
> > > > > > > > id ---
> > > > > > > > > > >> > the only
> > > > > > > > > > >> > > > > >> > thing
> > > > > > > > > > >> > > > > >> > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > when the group itself
> is
> > > gone
> > > > > we
> > > > > > > > also check
> > > > > > > > > > >> > if there
> > > > > > > > > > >> > > > > >> are
> > > > > > > > > > >> > > > > >> > > any
> > > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to
> that
> > > group
> > > > > and
> > > > > > > > delete
> > > > > > > > > > >> as
> > > > > > > > > > >> > well.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed
> > > > > introduces an
> > > > > > > > > > >> > asymmetry on
> > > > > > > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > creation
> > > > > > > > > > >> > > > > >> > > > > /
> > > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the
> config
> > > > > entities,
> > > > > > > > and we
> > > > > > > > > > >> > would like
> > > > > > > > > > >> > > > > >> to
> > > > > > > > > > >> > > > > >> > > hear
> > > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we
> should
> > > aim
> > > > > for
> > > > > > > > symmetry
> > > > > > > > > > >> > i.e.
> > > > > > > > > > >> > > > > >> totally
> > > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not
> > > delete
> > > > > them at
> > > > > > > > all
> > > > > > > > > > >> when
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> group
> > > > > > > > > > >> > > > > >> > is
> > > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit
> deletion
> > > > > operations
> > > > > > > > by
> > > > > > > > > > >> > themselves.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > > > Consumer.subscribe
> > > > > > > > method for
> > > > > > > > > > >> > the Java
> > > > > > > > > > >> > > > > >> > client
> > > > > > > > > > >> > > > > >> > > > > still
> > > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > `java.util.regex.Pattern` of
> > > > > do we
> > > > > > > > have
> > > > > > > > > > >> to
> > > > > > > > > > >> > > > > >> introduce an
> > > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not need
> to
> > > > > introduce
> > > > > > > > an
> > > > > > > > > > >> > overload, but
> > > > > > > > > > >> > > > > >> I'm
> > > > > > > > > > >> > > > > >> > > all
> > > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > > > >> > > > > >> > > > > > > > > > > may be some
> compatibility
> > > > > issues
> > > > > > > > that we
> > > > > > > > > > >> may
> > > > > > > > > > >> > > > > overlook.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with
> Justine
> > > > > that we
> > > > > > > > should
> > > > > > > > > > >> be
> > > > > > > > > > >> > clearer
> > > > > > > > > > >> > > > > >> > about
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > > IBP/metadata.version
> > > > > > > > from the
> > > > > > > > > > >> > feature
> > > > > > > > > > >> > > > > >> flag.
> > > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > > > >> > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the
> default
> > > for
> > > > > the
> > > > > > > > feature
> > > > > > > > > > >> flag
> > > > > > > > > > >> > to
> > > > > > > > > > >> > > > > true
> > > > > > > > > > >> > > > > >> > > based on
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version
> once we
> > > > > want to
> > > > > > > > make it
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > >> default.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the
> protocol
> > > APIs
> > > > > don't
> > > > > > > > mention
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > >> required
> > > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > > > >> > > > > >> > > > > it
> > > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that for
> > > > > consistency.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly
> do
> > > that.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd
> that
> > > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > > > >> > > > > >> > requires
> > > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to do
> > > more
> > > > > than
> > > > > > > > reading.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought
> myself
> > > as
> > > > > well,
> > > > > > > > but in
> > > > > > > > > > >> the
> > > > > > > > > > >> > end we
> > > > > > > > > > >> > > > > >> > could
> > > > > > > > > > >> > > > > >> > > not
> > > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative:
> adding
> > > > > Write
> > > > > > > > Group
> > > > > > > > > > >> seems
> > > > > > > > > > >> > an
> > > > > > > > > > >> > > > > >> overkill
> > > > > > > > > > >> > > > > >> > > here
> > > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere
> (we
> > > only
> > > > > have
> > > > > > > > Read /
> > > > > > > > > > >> > Delete
> > > > > > > > > > >> > > > > and
> > > > > > > > > > >> > > > > >> > > Describe
> > > > > > > > > > >> > > > > >> > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to
> hear
> > > others
> > > > > > > > thoughts.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> > > recreation
> > > > > > > > handled by
> > > > > > > > > > >> the
> > > > > > > > > > >> > > > > consumer
> > > > > > > > > > >> > > > > >> > with
> > > > > > > > > > >> > > > > >> > > the
> > > > > > > > > > >> > > > > >> > > > > new
> > > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be
> > > good
> > > > > to have
> > > > > > > > a
> > > > > > > > > > >> > section on
> > > > > > > > > > >> > > > > >> this.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > > > > subscription
> > > > > > > > right? Yes
> > > > > > > > > > >> > we can
> > > > > > > > > > >> > > > > >> add a
> > > > > > > > > > >> > > > > >> > > > > section
> > > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically the
> > > idea
> > > > > is that
> > > > > > > > > > >> consumer
> > > > > > > > > > >> > would
> > > > > > > > > > >> > > > > be
> > > > > > > > > > >> > > > > >> > > totally
> > > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as
> it's
> > > > > handled all
> > > > > > > > by the
> > > > > > > > > > >> > brokers.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions
> we
> > > will
> > > > > write
> > > > > > > > the
> > > > > > > > > > >> new
> > > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > > >> > > > > >> > in
> > > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an
> implementation
> > > > > detail,
> > > > > > > > do we
> > > > > > > > > > >> > plan to
> > > > > > > > > > >> > > > > >> have a
> > > > > > > > > > >> > > > > >> > > new
> > > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought
> about
> > > > > that. But
> > > > > > > > I think
> > > > > > > > > > >> > the
> > > > > > > > > > >> > > > > answer
> > > > > > > > > > >> > > > > >> > > should
> > > > > > > > > > >> > > > > >> > > > > be
> > > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> > > scalability
> > > > > goal
> > > > > > > > when it
> > > > > > > > > > >> > comes to
> > > > > > > > > > >> > > > > >> how
> > > > > > > > > > >> > > > > >> > > many
> > > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> > > support?
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I
> think we
> > > > > should
> > > > > > > > shoot for
> > > > > > > > > > >> > 1000s of
> > > > > > > > > > >> > > > > >> > > members.
> > > > > > > > > > >> > > > > >> > > > > But
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals also
> > > depend
> > > > > on the
> > > > > > > > offset
> > > > > > > > > > >> > > > > management
> > > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> > > coordinator
> > > > > > > > which we
> > > > > > > > > > >> did
> > > > > > > > > > >> > not
> > > > > > > > > > >> > > > > >> cover in
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a number
> that
> > > > > applies
> > > > > > > > > > >> > universally.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider
> > > having
> > > > > > > > > > >> > SubscribedTopidIds
> > > > > > > > > > >> > > > > >> instead
> > > > > > > > > > >> > > > > >> > of
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames
> in
> > > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > > > > >> > > > > >> > > Is the
> > > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to
> resolve
> > > the
> > > > > regex
> > > > > > > > on the
> > > > > > > > > > >> > server,
> > > > > > > > > > >> > > > > we
> > > > > > > > > > >> > > > > >> > can
> > > > > > > > > > >> > > > > >> > > do
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > > > > difference is
> > > > > > > > that
> > > > > > > > > > >> > sending the
> > > > > > > > > > >> > > > > >> > regex
> > > > > > > > > > >> > > > > >> > > is
> > > > > > > > > > >> > > > > >> > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending the
> topic
> > > > > names is
> > > > > > > > less
> > > > > > > > > > >> > efficient.
> > > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is easier
> to
> > > > > handle if
> > > > > > > > we have
> > > > > > > > > > >> > topic
> > > > > > > > > > >> > > > > ids.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to
> still
> > > let
> > > > > the
> > > > > > > > clients
> > > > > > > > > > >> send
> > > > > > > > > > >> > names
> > > > > > > > > > >> > > > > >> is to
> > > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names ->
> ids
> > > on
> > > > > the
> > > > > > > > broker /
> > > > > > > > > > >> > admin
> > > > > > > > > > >> > > > > client
> > > > > > > > > > >> > > > > >> > > only.
> > > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > > >> > > > > >> > > > > > > > > > > although we added
> topic id
> > > in
> > > > > > > > KIP-516, we
> > > > > > > > > > >> > never
> > > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> > > leveraging
> > > > > the
> > > > > > > > related
> > > > > > > > > > >> > newer
> > > > > > > > > > >> > > > > >> versioned
> > > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic id
> as
> > > empty
> > > > > UUID.
> > > > > > > > We
> > > > > > > > > > >> want
> > > > > > > > > > >> > to keep
> > > > > > > > > > >> > > > > >> the
> > > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> > > delegate
> > > > > the
> > > > > > > > reasoning
> > > > > > > > > > >> on
> > > > > > > > > > >> > broker
> > > > > > > > > > >> > > > > >> and
> > > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022
> at
> > > 10:45
> > > > > AM
> > > > > > > > David
> > > > > > > > > > >> Jacot
> > > > > > > > > > >> > > > > >> > > > > > > > >
> <dj...@confluent.io.invalid>
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to
> start a
> > > > > > > > discussion
> > > > > > > > > > >> thread
> > > > > > > > > > >> > on
> > > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > > > >> > > > > >> > > The
> > > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the
> > > Consumer
> > > > > > > > Rebalance
> > > > > > > > > > >> > Protocol.
> > > > > > > > > > >> > > > > >> With
> > > > > > > > > > >> > > > > >> > > this
> > > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the
> rebalance
> > > > > protocol
> > > > > > > > (for
> > > > > > > > > > >> > consumers)
> > > > > > > > > > >> > > > > >> more
> > > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to
> > > > > implement for
> > > > > > > > > > >> > clients, and
> > > > > > > > > > >> > > > > >> easier
> > > > > > > > > > >> > > > > >> > > to
> > > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > > > > >> > > > > >> > > > >
> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a look
> and
> > > let
> > > > > me
> > > > > > > > know what
> > > > > > > > > > >> > you
> > > > > > > > > > >> > > > > think.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away
> from
> > > > > July
> > > > > > > > 18th to
> > > > > > > > > > >> > August 8th.
> > > > > > > > > > >> > > > > >> That
> > > > > > > > > > >> > > > > >> > > gives
> > > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read and
> > > digest
> > > > > this
> > > > > > > > long
> > > > > > > > > > >> KIP.
> > > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> > >
> > > > > > > > > > >> > > > > >> >
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >> --
> > > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > > > >> > > > > >>
> > > > > > > > > > >> > > > > >
> > > > > > > > > > >> > > > >
> > > > > > > > > > >> >
> > > > > > > > > > >>
> > > > > > > > > > >
> > > > > > > >
> > > > >
> > > >
> > > >
> > > > --
> > > > -- Guozhang
> > >
> >
> >
> > --
> > -- Guozhang
>
>

-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Guozhang,

1. I have added a reference to the relevant chapter instead of
repeating the whole thing. Does that work for you?

2. The "Rebalance Triggers" section you are referring to is about when
a rebalance should be triggered for the non-upgraded members using the
old protocol. The section mentions that a rebalance must be triggered
when a new assignment is installed. This implies that the group epoch
was updated either by a native member or a non-upgraded member. For
the latter, the JoinGroup request would be the trigger. I have added a
reference to the relevant chapter in the "JoinGroup Handling" section
as well. Does that make sense?

Thanks,
David

On Fri, Sep 9, 2022 at 10:35 PM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hello David,
>
> Alright I think that's sufficient. Just to make that clear in the doc,
> could we update:
>
> 1) the heartbeat request handling section, stating when coordinator will
> trigger rebalance based on the HB's member metadata / reason?
> 2) the "Rebalance Triggers" section to include what we described in "Group
> Epoch - Trigger a rebalance" section as well?
>
>
> Guozhang
>
> On Fri, Sep 9, 2022 at 1:28 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Guozhang,
> >
> > I thought that the assignor will always be consulted when the next
> > heartbeat request is constructed. In other words,
> > `PartitionAssignor#metadata` will be called for every heartbeat. This
> > gives the opportunity for the assignor to enforce a rebalance by
> > setting the reason to a non-zero value or by changing the bytes. Do
> > you think that this is not sufficient? Are you concerned by the delay?
> >
> > Best,
> > David
> >
> > On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com> wrote:
> > >
> > > Hello David,
> > >
> > > One of Jun's comments make me thinking:
> > >
> > > ```
> > > In this case, a new assignment is triggered by the client side
> > > assignor. When constructing the HB, the consumer will always consult
> > > the client side assignor and propagate the information to the group
> > > coordinator. In other words, we don't expect users to call
> > > Consumer#enforceRebalance anymore.
> > > ```
> > >
> > > As I looked at the current PartitionAssignor's interface, we actually do
> > > not have a way yet to instruct how to construct the next HB request, e.g.
> > > when the assignor wants to enforce a new rebalance with a new assignment,
> > > we'd need some customizable APIs inside the PartitionAssignor to indicate
> > > the next HB telling broker about so. WDYT about adding such an API on the
> > > PartitionAssignor?
> > >
> > >
> > > Guozhang
> > >
> > >
> > > On Tue, Sep 6, 2022 at 6:09 AM David Jacot <dj...@confluent.io.invalid>
> > > wrote:
> > >
> > > > Hi Jun,
> > > >
> > > > I have updated the KIP to include your feedback. I have also tried to
> > > > clarify the parts which were not cleared.
> > > >
> > > > Best,
> > > > David
> > > >
> > > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io>
> > wrote:
> > > > >
> > > > > Hi Jun,
> > > > >
> > > > > Thanks for your feedback. Let me start by answering your questions
> > > > > inline and I will update the KIP next week.
> > > > >
> > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to
> > be
> > > > fewer
> > > > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > > > comments below.
> > > > >
> > > > > I would also add that the KIP removes the global sync barrier in the
> > > > > protocol which is essential to improve group stability and
> > > > > scalability, and the KIP also simplifies the client by moving most of
> > > > > the logic to the server side.
> > > > >
> > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > > > changing
> > > > > > of the partition assignor in the consumers?
> > > > >
> > > > > Definitely. The group coordinator will use the assignor used by a
> > > > > majority of the members. This allows the group to move from one
> > > > > assignor to another by a roll. This is explained in the Assignor
> > > > > Selection chapter.
> > > > >
> > > > > > 30.2 For each field, could you explain whether it's required in
> > every
> > > > > > request or the scenarios when it needs to be filled? For example,
> > it's
> > > > not
> > > > > > clear to me when TopicPartitions needs to be filled.
> > > > >
> > > > > The client is expected to set those fields in case of a connection
> > > > > issue (e.g. timeout) or when the fields have changed since the last
> > > > > HB. The server populates those fields as long as the member is not
> > > > > fully reconciled - the member should acknowledge that it has the
> > > > > expected epoch and assignment. I will clarify this in the KIP.
> > > > >
> > > > > > 31. In the current consumer protocol, the rack affinity between the
> > > > client
> > > > > > and the broker is only considered during fetching, but not during
> > > > assigning
> > > > > > partitions to consumers. Sometimes, once the assignment is made,
> > there
> > > > is
> > > > > > no opportunity for read affinity because no replicas of assigned
> > > > partitions
> > > > > > are close to the member. I am wondering if we should use this
> > > > opportunity
> > > > > > to address this by including rack in GroupMember.
> > > > >
> > > > > That's an interesting idea. I don't see any issue with adding the
> > rack
> > > > > to the members. I will do so.
> > > > >
> > > > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > > > coordinator is. By moving the event loop model, it seems that we
> > could
> > > > add
> > > > > > a metric that tracks the fraction of the time the event loop is
> > doing
> > > > the
> > > > > > actual work.
> > > > >
> > > > > That's a great idea. I will add it. Thanks.
> > > > >
> > > > > > 33. Could we add a section on coordinator failover handling? For
> > > > example,
> > > > > > does it need to trigger the check if any group with the wildcard
> > > > > > subscription now has a new matching topic?
> > > > >
> > > > > Sure. When the new group coordinator takes over, it has to:
> > > > > * Setup the session timeouts.
> > > > > * Trigger a new assignment if a client side assignor is used. We
> > don't
> > > > > store the information about the member selected to run the assignment
> > > > > so we have to start a new one.
> > > > > * Update the topics metadata, verify the wildcard subscriptions, and
> > > > > trigger a rebalance if needed.
> > > > >
> > > > > > 34. ConsumerGroupMetadataValue,
> > ConsumerGroupPartitionMetadataValue,
> > > > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > > > field
> > > > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > > > reflect
> > > > > > the latest group epoch? What about the one in
> > > > > > ConsumerGroupPartitionMetadataValue and
> > > > ConsumerGroupMemberMetadataValue?
> > > > >
> > > > > Sure. I will clarify that but it is always the latest group epoch.
> > > > > When the group state is updated, the group epoch is bumped so we use
> > > > > that one for all the change records related to the update.
> > > > >
> > > > > > 35. "the group coordinator will ensure that the following
> > invariants
> > > > are
> > > > > > met: ... All members exists." It's possible for a member not to
> > get any
> > > > > > assigned partitions, right?
> > > > >
> > > > > That's right. Here I meant that the members provided by the assignor
> > > > > in the assignment must exist in the group. The assignor can not make
> > > > > up new member ids.
> > > > >
> > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > When
> > > > would
> > > > > > a consumer rejoin and what member id would be used?
> > > > >
> > > > > A member is expected to abandon all its partitions and rejoins when
> > it
> > > > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > > > coordinator will have removed the member from the group. The member
> > > > > can rejoin the group with the same member id but with 0 as epoch. Let
> > > > > me see if I can clarify this in the KIP.
> > > > >
> > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > reassignment
> > > > > > by either providing a non-zero reason or by updating the assignor
> > > > > > metadata." Hmm, this seems to be conflicting with the deprecation
> > of
> > > > > > Consumer#enforeRebalance.
> > > > >
> > > > > In this case, a new assignment is triggered by the client side
> > > > > assignor. When constructing the HB, the consumer will always consult
> > > > > the client side assignor and propagate the information to the group
> > > > > coordinator. In other words, we don't expect users to call
> > > > > Consumer#enforceRebalance anymore.
> > > > >
> > > > > > 38. The reassignment examples are nice. But the section seems to
> > have
> > > > > > multiple typos.
> > > > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > epoch=3,
> > > > > > partitions=[foo-1] too early.
> > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > partitions=[foo-0].
> > > > >
> > > > > Sorry for that! I will revise them.
> > > > >
> > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade from
> > any
> > > > old
> > > > > > version to new one?
> > > > >
> > > > > We will support upgrading from the consumer protocol version 3,
> > > > > introduced in KIP-792. KIP-792 is not implemented yet so the earliest
> > > > > version is unknown at the moment. This is explained in the migration
> > > > > plan chapter.
> > > > >
> > > > > Thanks again for your feedback, Jun. I will update the KIP based on
> > it
> > > > > next week.
> > > > >
> > > > > Best,
> > > > > David
> > > > >
> > > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <ju...@confluent.io.invalid>
> > wrote:
> > > > > >
> > > > > > Hi, David,
> > > > > >
> > > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to
> > be
> > > > fewer
> > > > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > > > comments below.
> > > > > >
> > > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > > > changing
> > > > > > of the partition assignor in the consumers?
> > > > > > 30.2 For each field, could you explain whether it's required in
> > every
> > > > > > request or the scenarios when it needs to be filled? For example,
> > it's
> > > > not
> > > > > > clear to me when TopicPartitions needs to be filled.
> > > > > >
> > > > > > 31. In the current consumer protocol, the rack affinity between the
> > > > client
> > > > > > and the broker is only considered during fetching, but not during
> > > > assigning
> > > > > > partitions to consumers. Sometimes, once the assignment is made,
> > there
> > > > is
> > > > > > no opportunity for read affinity because no replicas of assigned
> > > > partitions
> > > > > > are close to the member. I am wondering if we should use this
> > > > opportunity
> > > > > > to address this by including rack in GroupMember.
> > > > > >
> > > > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > > > coordinator is. By moving the event loop model, it seems that we
> > could
> > > > add
> > > > > > a metric that tracks the fraction of the time the event loop is
> > doing
> > > > the
> > > > > > actual work.
> > > > > >
> > > > > > 33. Could we add a section on coordinator failover handling? For
> > > > example,
> > > > > > does it need to trigger the check if any group with the wildcard
> > > > > > subscription now has a new matching topic?
> > > > > >
> > > > > > 34. ConsumerGroupMetadataValue,
> > ConsumerGroupPartitionMetadataValue,
> > > > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > > > field
> > > > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > > > reflect
> > > > > > the latest group epoch? What about the one in
> > > > > > ConsumerGroupPartitionMetadataValue and
> > > > ConsumerGroupMemberMetadataValue?
> > > > > >
> > > > > > 35. "the group coordinator will ensure that the following
> > invariants
> > > > are
> > > > > > met: ... All members exists." It's possible for a member not to
> > get any
> > > > > > assigned partitions, right?
> > > > > >
> > > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> > When
> > > > would
> > > > > > a consumer rejoin and what member id would be used?
> > > > > >
> > > > > > 37. "Instead, power users will have the ability to trigger a
> > > > reassignment
> > > > > > by either providing a non-zero reason or by updating the assignor
> > > > > > metadata." Hmm, this seems to be conflicting with the deprecation
> > of
> > > > > > Consumer#enforeRebalance.
> > > > > >
> > > > > > 38. The reassignment examples are nice. But the section seems to
> > have
> > > > > > multiple typos.
> > > > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > > epoch=3,
> > > > > > partitions=[foo-1] too early.
> > > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > > partitions=[foo-0].
> > > > > >
> > > > > > 39. Rolling upgrade of consumers: Do we support the upgrade from
> > any
> > > > old
> > > > > > version to new one?
> > > > > >
> > > > > > Thanks,
> > > > > >
> > > > > > Jun
> > > > > >
> > > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > > <dj...@confluent.io.invalid>
> > > > > > wrote:
> > > > > >
> > > > > > > Hi all,
> > > > > > >
> > > > > > > The KIP states that we will re-implement the coordinator in
> > Java. I
> > > > > > > discussed this offline with a few folks and folks are concerned
> > that
> > > > > > > we could introduce many regressions in the old protocol if we do
> > so.
> > > > > > > Therefore, I am going to remove this statement from the KIP. It
> > is an
> > > > > > > implementation detail after all so it does not have to be
> > decided at
> > > > > > > this stage. We will likely start by trying to refactor the
> > current
> > > > > > > implementation as a first step.
> > > > > > >
> > > > > > > Cheers,
> > > > > > > David
> > > > > > >
> > > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <djacot@confluent.io
> > >
> > > > wrote:
> > > > > > > >
> > > > > > > > Hi Luke,
> > > > > > > >
> > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > reconciling,
> > > > > > > stable,
> > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > >
> > > > > > > > This sentence does not refer to those group states but rather
> > to a
> > > > > > > > state machine replication (SMR). This refers to the entire
> > state of
> > > > > > > > group coordinator which is replicated via the log layer. I will
> > > > > > > > clarify this in the KIP.
> > > > > > > >
> > > > > > > > > 1.2. What do you mean "each state machine is modelled as an
> > event
> > > > > > > loop"?
> > > > > > > >
> > > > > > > > The idea is to follow a model similar to the new quorum
> > > > controller. We
> > > > > > > > will have N threads to process events. Each __consumer_offsets
> > > > > > > > partition is assigned to a unique thread and all the events
> > (e.g.
> > > > > > > > requests, callbacks, etc.) are processed by this thread. This
> > > > simplify
> > > > > > > > concurrency and will enable us to do simulation testing for the
> > > > group
> > > > > > > > coordinator.
> > > > > > > >
> > > > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > > > partitions?
> > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > coordinator?
> > > > > > > For
> > > > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > > > both
> > > > > > > exist in
> > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it,
> > or
> > > > 2?
> > > > > > > >
> > > > > > > > See 1.1. The confusion comes from there, I think.
> > > > > > > >
> > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number
> > of
> > > > threads
> > > > > > > used
> > > > > > > > > to run the state machines. But I'm wondering if the purpose
> > of
> > > > the
> > > > > > > threads
> > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > *__consumer_offsets*
> > > > > > > > > partitions?), and no heavy computation, why should we need
> > > > > > > multi-threads
> > > > > > > > > here?
> > > > > > > >
> > > > > > > > See 1.2. The idea is to have an ability to shard the
> > processing as
> > > > the
> > > > > > > > computation could be heavy.
> > > > > > > >
> > > > > > > > > 2.1. The consumer session timeout, why does the default
> > session
> > > > > > > timeout not
> > > > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > > > session
> > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > >
> > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > detect
> > > > client
> > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > > > session
> > > > > > > timeout.
> > > > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > > > session
> > > > > > > timeout.
> > > > > > > >
> > > > > > > > This is indeed a mistake. The default session timeout should
> > be 45s
> > > > > > > > (the current default).
> > > > > > > >
> > > > > > > > > 2.2. The default server side assignor are [range, uniform],
> > > > which means
> > > > > > > > > we'll default to "range" assignor. I'd like to know why not
> > > > uniform
> > > > > > > one? I
> > > > > > > > > thought usually users will choose uniform assignor (former
> > sticky
> > > > > > > assinor)
> > > > > > > > > for better evenly distribution. Any other reason we choose
> > range
> > > > > > > assignor
> > > > > > > > > as default?
> > > > > > > > > group.consumer.assignors List range, uniform The server side
> > > > assignors.
> > > > > > > >
> > > > > > > > The order on the server side has no influence because the
> > client
> > > > must
> > > > > > > > chose the selector that he wants to use. There is no default
> > in the
> > > > > > > > current proposal. If the assignor is not specified by the
> > client,
> > > > the
> > > > > > > > request is rejected. The default client value for
> > > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > > >
> > > > > > > > Thanks for your very good comments, Luke. I hope that my
> > answers
> > > > help
> > > > > > > > to clarify things. I will update the KIP as well based on your
> > > > > > > > feedback.
> > > > > > > >
> > > > > > > > Cheers,
> > > > > > > > David
> > > > > > > >
> > > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com>
> > > > wrote:
> > > > > > > > >
> > > > > > > > > Hi David,
> > > > > > > > >
> > > > > > > > > Thanks for the update.
> > > > > > > > >
> > > > > > > > > Some more questions:
> > > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > > > The new group coordinator will have a state machine per
> > > > > > > > > *__consumer_offsets* partitions, where each state machine is
> > > > modelled
> > > > > > > as an
> > > > > > > > > event loop. Those state machines will be executed in
> > > > > > > > > *group.coordinator.threads* threads.
> > > > > > > > >
> > > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > > reconciling,
> > > > > > > stable,
> > > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > > 1.2. What do you mean "each state machine is modelled as an
> > event
> > > > > > > loop"?
> > > > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > > > partitions?
> > > > > > > > > Not a state machine "per consumer group" owned by a group
> > > > coordinator?
> > > > > > > For
> > > > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > > > both
> > > > > > > exist in
> > > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it,
> > or
> > > > 2?
> > > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number
> > of
> > > > threads
> > > > > > > used
> > > > > > > > > to run the state machines. But I'm wondering if the purpose
> > of
> > > > the
> > > > > > > threads
> > > > > > > > > is only to keep the state of each consumer group (or
> > > > > > > *__consumer_offsets*
> > > > > > > > > partitions?), and no heavy computation, why should we need
> > > > > > > multi-threads
> > > > > > > > > here?
> > > > > > > > >
> > > > > > > > > 2. For the default value in the new configs:
> > > > > > > > > 2.1. The consumer session timeout, why does the default
> > session
> > > > > > > timeout not
> > > > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > > > session
> > > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > > >
> > > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> > detect
> > > > client
> > > > > > > > > failures when using the consumer group protocol.
> > > > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > > > session
> > > > > > > timeout.
> > > > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > > > session
> > > > > > > timeout.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > 2.2. The default server side assignor are [range, uniform],
> > > > which means
> > > > > > > > > we'll default to "range" assignor. I'd like to know why not
> > > > uniform
> > > > > > > one? I
> > > > > > > > > thought usually users will choose uniform assignor (former
> > sticky
> > > > > > > assinor)
> > > > > > > > > for better evenly distribution. Any other reason we choose
> > range
> > > > > > > assignor
> > > > > > > > > as default?
> > > > > > > > > group.consumer.assignors List range, uniform The server side
> > > > assignors.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Thank you.
> > > > > > > > > Luke
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <showuon@gmail.com
> > >
> > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hi Sagar,
> > > > > > > > > >
> > > > > > > > > > I have some thoughts about Kafka Connect integrating with
> > > > KIP-848,
> > > > > > > but I
> > > > > > > > > > think we should have a separate discussion thread for the
> > Kafka
> > > > > > > Connect
> > > > > > > > > > KIP: Integrating Kafka Connect With New Consumer Rebalance
> > > > Protocol
> > > > > > > [1],
> > > > > > > > > > and let this discussion thread focus on consumer rebalance
> > > > protocol,
> > > > > > > WDYT?
> > > > > > > > > >
> > > > > > > > > > [1]
> > > > > > > > > >
> > > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > >
> > > > > > > > > > Thank you.
> > > > > > > > > > Luke
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > > sagarmeansocean@gmail.com>
> > > > > > > wrote:
> > > > > > > > > >
> > > > > > > > > >> Thank you Guozhang/David for the feedback. Looks like
> > there's
> > > > > > > agreement on
> > > > > > > > > >> using separate APIs for Connect. I would revisit the doc
> > and
> > > > see
> > > > > > > what
> > > > > > > > > >> changes are to be made.
> > > > > > > > > >>
> > > > > > > > > >> Thanks!
> > > > > > > > > >> Sagar.
> > > > > > > > > >>
> > > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > >> wrote:
> > > > > > > > > >>
> > > > > > > > > >> > Hi Sagar,
> > > > > > > > > >> >
> > > > > > > > > >> > Thanks for the feedback and the document. That's really
> > > > helpful. I
> > > > > > > > > >> > will take a look at it.
> > > > > > > > > >> >
> > > > > > > > > >> > Overall, it seems to me that both Connect and the
> > Consumer
> > > > could
> > > > > > > share
> > > > > > > > > >> > the same underlying "engine". The main difference is
> > that
> > > > the
> > > > > > > Consumer
> > > > > > > > > >> > assigns topic-partitions to members whereas Connect
> > assigns
> > > > tasks
> > > > > > > to
> > > > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > > > >> > 1) We extend the new proposed APIs to support different
> > > > resource
> > > > > > > types
> > > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > > >> > 2) We use new dedicated APIs for Connect. The dedicated
> > APIs
> > > > > > > would be
> > > > > > > > > >> > similar to the new ones but different on the
> > > > content/resources and
> > > > > > > > > >> > they would rely on the same engine on the coordinator
> > side.
> > > > > > > > > >> >
> > > > > > > > > >> > I personally lean towards 2) because I am not a fan of
> > > > > > > overcharging
> > > > > > > > > >> > APIs to serve different purposes. That being said, I am
> > not
> > > > > > > opposed to
> > > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > > > >> >
> > > > > > > > > >> > I think that we can continue to discuss it here for now
> > in
> > > > order
> > > > > > > to
> > > > > > > > > >> > ensure that this KIP is compatible with what we will do
> > for
> > > > > > > Connect in
> > > > > > > > > >> > the future.
> > > > > > > > > >> >
> > > > > > > > > >> > Best,
> > > > > > > > > >> > David
> > > > > > > > > >> >
> > > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > > djacot@confluent.io>
> > > > > > > wrote:
> > > > > > > > > >> > >
> > > > > > > > > >> > > Hi all,
> > > > > > > > > >> > >
> > > > > > > > > >> > > I am back from vacation. I will go through and address
> > > > your
> > > > > > > comments
> > > > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > > > >> > >
> > > > > > > > > >> > > Cheers,
> > > > > > > > > >> > > David
> > > > > > > > > >> > >
> > > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > > > gharris1727@gmail.com
> > > > > > > > > >> >
> > > > > > > > > >> > wrote:
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > Hey All!
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> > cooperative
> > > > > > > rebalancing
> > > > > > > > > >> > making it
> > > > > > > > > >> > > > down the stack!
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > I had a few questions:
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > 1. The 'Rejected Alternatives' section describes how
> > > > member
> > > > > > > epoch
> > > > > > > > > >> > should
> > > > > > > > > >> > > > advance in step with the group epoch and assignment
> > > > epoch
> > > > > > > values. I
> > > > > > > > > >> > think
> > > > > > > > > >> > > > that this is a good idea for the reasons described
> > in
> > > > the
> > > > > > > KIP. When
> > > > > > > > > >> the
> > > > > > > > > >> > > > protocol is incrementally assigning partitions to a
> > > > worker,
> > > > > > > what
> > > > > > > > > >> member
> > > > > > > > > >> > > > epoch does each incremental assignment use? Are
> > member
> > > > epochs
> > > > > > > > > >> re-used,
> > > > > > > > > >> > and
> > > > > > > > > >> > > > a single member epoch can correspond to multiple
> > > > different
> > > > > > > > > >> > (monotonically
> > > > > > > > > >> > > > larger) assignments?
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the
> > group
> > > > > > > > > >> coordinator? If
> > > > > > > > > >> > > > not, should custom client-side assignor
> > implementations
> > > > > > > interact
> > > > > > > > > >> with
> > > > > > > > > >> > the
> > > > > > > > > >> > > > Reason field, and how is its common meaning agreed
> > > > upon? If
> > > > > > > so, what
> > > > > > > > > >> > is the
> > > > > > > > > >> > > > benefit of a distinct Reason field over including
> > such
> > > > > > > functionality
> > > > > > > > > >> > in the
> > > > > > > > > >> > > > opaque metadata?
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > 3. The following is included in the KIP: "Thanks to
> > > > this, the
> > > > > > > input
> > > > > > > > > >> of
> > > > > > > > > >> > the
> > > > > > > > > >> > > > client side assignor is entirely driven by the group
> > > > > > > coordinator.
> > > > > > > > > >> The
> > > > > > > > > >> > > > consumer is no longer responsible for maintaining
> > any
> > > > state
> > > > > > > besides
> > > > > > > > > >> its
> > > > > > > > > >> > > > assigned partitions." Does this mean that the
> > > > client-side
> > > > > > > assignor
> > > > > > > > > >> MAY
> > > > > > > > > >> > > > incorporate additional non-Metadata state (such as
> > > > partition
> > > > > > > > > >> > throughput,
> > > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or that
> > > > additional
> > > > > > > > > >> > non-Metadata
> > > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > > > >> > > > for
> > > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > > >> > > > and
> > org.apache.kafka.clients.consumer.PartitionAssignor
> > > > that
> > > > > > > seem to
> > > > > > > > > >> > > > overlap significantly. Is it possible for these two
> > > > > > > implementations
> > > > > > > > > >> to
> > > > > > > > > >> > be
> > > > > > > > > >> > > > unified? This would serve to promote feature parity
> > of
> > > > > > > server-side
> > > > > > > > > >> and
> > > > > > > > > >> > > > client-side assignors, and would also facilitate
> > > > operational
> > > > > > > > > >> > flexibility in
> > > > > > > > > >> > > > certain situations. For example, if a server-side
> > > > assignor
> > > > > > > has some
> > > > > > > > > >> > poor
> > > > > > > > > >> > > > behavior and needs a patch, deploying the patched
> > > > assignor to
> > > > > > > the
> > > > > > > > > >> > client
> > > > > > > > > >> > > > and switching one consumer group to a client-side
> > > > assignor
> > > > > > > may be
> > > > > > > > > >> > faster
> > > > > > > > > >> > > > and less risky than patching all of the brokers.
> > With
> > > > the
> > > > > > > currently
> > > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> > reimplementation
> > > > would
> > > > > > > have
> > > > > > > > > >> to be
> > > > > > > > > >> > > > assembled, and if the two APIs have diverged
> > > > significantly,
> > > > > > > then it
> > > > > > > > > >> is
> > > > > > > > > >> > > > possible that a reimplementation would not be
> > possible.
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > --
> > > > > > > > > >> > > > Greg Harris
> > > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > >> > wrote:
> > > > > > > > > >> > > >
> > > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > I created a confluence page to discuss how Connect
> > > > would
> > > > > > > need to
> > > > > > > > > >> > change
> > > > > > > > > >> > > > > based on the new rebalance protocol. Here's the
> > page:
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > >
> > > > > > > > > >> >
> > > > > > > > > >>
> > > > > > >
> > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > It's also pretty longish and I have tried to keep
> > a
> > > > format
> > > > > > > > > >> similar to
> > > > > > > > > >> > > > > KIP-848. Let me know what you think. Also, do you
> > > > think this
> > > > > > > > > >> should
> > > > > > > > > >> > be
> > > > > > > > > >> > > > > moved to a separate discussion thread or is this
> > one
> > > > fine?
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > Thanks!
> > > > > > > > > >> > > > > Sagar.
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > > > sagarmeansocean@gmail.com>
> > > > > > > > > >> > wrote:
> > > > > > > > > >> > > > >
> > > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > Thank you so much for the doc on Kafka Streams.
> > > > Sure, I
> > > > > > > would do
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > Thanks!
> > > > > > > > > >> > > > > > Sagar.
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > > >> > > > > wrote:
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > >> It would be great if you could come back with
> > some
> > > > > > > analysis on
> > > > > > > > > >> > how to
> > > > > > > > > >> > > > > >> implement the Connect side integration with
> > the new
> > > > > > > protocol;
> > > > > > > > > >> so
> > > > > > > > > >> > far
> > > > > > > > > >> > > > > >> besides leveraging on the new "protocol type"
> > we
> > > > did not
> > > > > > > yet
> > > > > > > > > >> think
> > > > > > > > > >> > > > > through
> > > > > > > > > >> > > > > >> the Connect side implementations. For Streams
> > > > here's a
> > > > > > > draft of
> > > > > > > > > >> > > > > >> integration
> > > > > > > > > >> > > > > >> plan:
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > >
> > > > > > > > > >> >
> > > > > > > > > >>
> > > > > > >
> > > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > > >> > >
> > > > > > > > > >> > > > > wrote:
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> > Thank you for your response. The reason I
> > thought
> > > > > > > connect can
> > > > > > > > > >> > also fit
> > > > > > > > > >> > > > > >> into
> > > > > > > > > >> > > > > >> > this new scheme is that even today the
> > connect
> > > > uses a
> > > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > > >> > > > > >> > extending from AbstractCoordinator to empower
> > > > > > > rebalances of
> > > > > > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets
> > the
> > > > > > > > > >> protocolType()
> > > > > > > > > >> > to
> > > > > > > > > >> > > > > >> connect
> > > > > > > > > >> > > > > >> > and uses the metadata() method by plumbing
> > into
> > > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> > I think the changes to support connect would
> > be
> > > > > > > similar at a
> > > > > > > > > >> > high
> > > > > > > > > >> > > > > level
> > > > > > > > > >> > > > > >> to
> > > > > > > > > >> > > > > >> > the changes in streams mainly because of the
> > > > Client
> > > > > > > side
> > > > > > > > > >> > assignors
> > > > > > > > > >> > > > > being
> > > > > > > > > >> > > > > >> > used in both. At an implementation level, we
> > > > might
> > > > > > > need to
> > > > > > > > > >> make
> > > > > > > > > >> > a lot
> > > > > > > > > >> > > > > of
> > > > > > > > > >> > > > > >> > changes to get onto this new assignment
> > protocol
> > > > like
> > > > > > > > > >> enhancing
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> > JoinGroup request/response and SyncGroup and
> > > > using
> > > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > >> > > > > >> > API etc again on similar lines to streams (or
> > > > there
> > > > > > > might be
> > > > > > > > > >> > > > > >> deviations). I
> > > > > > > > > >> > > > > >> > would try to perform a detailed analysis of
> > the
> > > > same
> > > > > > > and we
> > > > > > > > > >> > can have
> > > > > > > > > >> > > > > a
> > > > > > > > > >> > > > > >> > separate discussion thread for that as that
> > would
> > > > > > > derail this
> > > > > > > > > >> > > > > discussion
> > > > > > > > > >> > > > > >> > thread. Let me know if that sounds good to
> > you.
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> > wrote:
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`.
> > > > Sure, I
> > > > > > > can
> > > > > > > > > >> > mention it.
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> > > 2) The idea is to transition C from his
> > current
> > > > > > > assignment
> > > > > > > > > >> to
> > > > > > > > > >> > his
> > > > > > > > > >> > > > > >> > > target assignment when he can move to
> > epoch 3.
> > > > When
> > > > > > > that
> > > > > > > > > >> > happens,
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > member assignment is updated and persisted
> > > > with all
> > > > > > > its
> > > > > > > > > >> > assigned
> > > > > > > > > >> > > > > >> > > partitions even if they are not all revoked
> > > > yet. In
> > > > > > > other
> > > > > > > > > >> > words, the
> > > > > > > > > >> > > > > >> > > member assignment becomes the target
> > > > assignment.
> > > > > > > This is
> > > > > > > > > >> > basically
> > > > > > > > > >> > > > > an
> > > > > > > > > >> > > > > >> > > optimization to avoid having to write all
> > the
> > > > > > > changes to
> > > > > > > > > >> the
> > > > > > > > > >> > log.
> > > > > > > > > >> > > > > The
> > > > > > > > > >> > > > > >> > > examples are based on the persisted state
> > so I
> > > > > > > understand
> > > > > > > > > >> the
> > > > > > > > > >> > > > > >> > > confusion. Let me see if I can improve
> > this in
> > > > the
> > > > > > > > > >> > description.
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse the
> > > > protocol
> > > > > > > with a
> > > > > > > > > >> > client side
> > > > > > > > > >> > > > > >> > > assignor if it fits in the protocol. The
> > > > assignment
> > > > > > > is
> > > > > > > > > >> about
> > > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> > Connect
> > > > fit
> > > > > > > into this?
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> > > Best,
> > > > > > > > > >> > > > > >> > > David
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > > >> > > > > >> wrote:
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor
> > > > observations:
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > 1) In the Assignment Error section in
> > Client
> > > > Side
> > > > > > > mode
> > > > > > > > > >> > Assignment
> > > > > > > > > >> > > > > >> > > process,
> > > > > > > > > >> > > > > >> > > > you mentioned => `In this case, the
> > client
> > > > side
> > > > > > > assignor
> > > > > > > > > >> can
> > > > > > > > > >> > > > > return
> > > > > > > > > >> > > > > >> an
> > > > > > > > > >> > > > > >> > > > error to the group coordinator`. In this
> > > > case are
> > > > > > > you
> > > > > > > > > >> > referring to
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > Assignor returning an AssignmentError
> > that's
> > > > > > > listed down
> > > > > > > > > >> > towards
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > end?
> > > > > > > > > >> > > > > >> > > > If yes, do you think it would make sense
> > to
> > > > > > > mention this
> > > > > > > > > >> > > > > explicitly
> > > > > > > > > >> > > > > >> > here?
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I have a
> > > > slight
> > > > > > > > > >> confusion,
> > > > > > > > > >> > not
> > > > > > > > > >> > > > > sure
> > > > > > > > > >> > > > > >> if
> > > > > > > > > >> > > > > >> > > > others have the same. Consider this step:
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > When B heartbeats, the group coordinator
> > > > > > > transitions him
> > > > > > > > > >> to
> > > > > > > > > >> > epoch
> > > > > > > > > >> > > > > 3
> > > > > > > > > >> > > > > >> > > because
> > > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> > persists
> > > > the
> > > > > > > change and
> > > > > > > > > >> > reply.
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0,
> > > > foo-1]
> > > > > > > > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > > > > > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions to
> > epoch 3
> > > > but
> > > > > > > cannot
> > > > > > > > > >> get
> > > > > > > > > >> > foo-1
> > > > > > > > > >> > > > > >> yet.
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C can't
> > get
> > > > the
> > > > > > > foo-1
> > > > > > > > > >> > partition
> > > > > > > > > >> > > > > yet,
> > > > > > > > > >> > > > > >> > but
> > > > > > > > > >> > > > > >> > > > based on the description above, it seems
> > it
> > > > > > > already has
> > > > > > > > > >> it.
> > > > > > > > > >> > Do you
> > > > > > > > > >> > > > > >> > think
> > > > > > > > > >> > > > > >> > > it
> > > > > > > > > >> > > > > >> > > > would be better to remove it and
> > populate it
> > > > only
> > > > > > > when it
> > > > > > > > > >> > actually
> > > > > > > > > >> > > > > >> gets
> > > > > > > > > >> > > > > >> > > it?
> > > > > > > > > >> > > > > >> > > > I see this in a lot of other places, so
> > have
> > > > I
> > > > > > > > > >> understood it
> > > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > > >> > > > > >> > > ?
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > Regarding connect , it might be out of
> > scope
> > > > of
> > > > > > > this
> > > > > > > > > >> > discussion,
> > > > > > > > > >> > > > > but
> > > > > > > > > >> > > > > >> > from
> > > > > > > > > >> > > > > >> > > > what I understood it would probably be
> > > > running in
> > > > > > > client
> > > > > > > > > >> > side
> > > > > > > > > >> > > > > >> assignor
> > > > > > > > > >> > > > > >> > > mode
> > > > > > > > > >> > > > > >> > > > even on the new rebalance protocol as it
> > has
> > > > its
> > > > > > > own
> > > > > > > > > >> Custom
> > > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David
> > Jacot
> > > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move
> > forward
> > > > with
> > > > > > > > > >> > specialized API
> > > > > > > > > >> > > > > >> > > > > instead of relying on one generic API.
> > For
> > > > > > > Connect, we
> > > > > > > > > >> > can apply
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > exact same pattern and reuse/share the
> > core
> > > > > > > > > >> > implementation on
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > server side. For the schema registry, I
> > > > think
> > > > > > > that we
> > > > > > > > > >> > should
> > > > > > > > > >> > > > > >> consider
> > > > > > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > > > > > membership/leader
> > > > > > > > > >> > election.
> > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > > >> > > > > >> > > > > David
> > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael
> > > > Juma <
> > > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern
> > vs
> > > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > >> > > > > >> > > > > > document the differences in more
> > detail
> > > > before
> > > > > > > > > >> deciding
> > > > > > > > > >> > one
> > > > > > > > > >> > > > > way
> > > > > > > > > >> > > > > >> or
> > > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > > > java.util.regex.Pattern,
> > > > > > > > > >> they
> > > > > > > > > >> > expect
> > > > > > > > > >> > > > > >> > their
> > > > > > > > > >> > > > > >> > > > > > semantics to be honored. If we are
> > doing
> > > > > > > something
> > > > > > > > > >> > different,
> > > > > > > > > >> > > > > >> then
> > > > > > > > > >> > > > > >> > we
> > > > > > > > > >> > > > > >> > > > > > should consider adding an overload
> > with
> > > > our own
> > > > > > > > > >> Pattern
> > > > > > > > > >> > class
> > > > > > > > > >> > > > > (I
> > > > > > > > > >> > > > > >> > > don't
> > > > > > > > > >> > > > > >> > > > > > think we'd want to expose re2j's at
> > this
> > > > > > > point).
> > > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new
> > > > protocol
> > > > > > > should
> > > > > > > > > >> > > > > integrate
> > > > > > > > > >> > > > > >> > fully
> > > > > > > > > >> > > > > >> > > > > with
> > > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > > > recreation case
> > > > > > > > > >> > correctly.
> > > > > > > > > >> > > > > That's
> > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > >> > > main
> > > > > > > > > >> > > > > >> > > > > > part we need to handle. I agree with
> > > > David
> > > > > > > that we'd
> > > > > > > > > >> > want to
> > > > > > > > > >> > > > > add
> > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > > >> > > > > >> > > > > > to the relevant protocols that don't
> > > > have it
> > > > > > > yet and
> > > > > > > > > >> > that we
> > > > > > > > > >> > > > > can
> > > > > > > > > >> > > > > >> > > probably
> > > > > > > > > >> > > > > >> > > > > > focus on the internals versus adding
> > new
> > > > APIs
> > > > > > > to the
> > > > > > > > > >> > Java
> > > > > > > > > >> > > > > >> Consumer
> > > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > > >> > > > > >> > > > > > we find that adding new APIs is
> > required
> > > > for
> > > > > > > > > >> reasonable
> > > > > > > > > >> > > > > >> semantics).
> > > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about the
> > > > coordinator
> > > > > > > storing
> > > > > > > > > >> the
> > > > > > > > > >> > > > > >> configs.
> > > > > > > > > >> > > > > >> > > It's
> > > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> > centralized
> > > > in the
> > > > > > > > > >> metadata
> > > > > > > > > >> > log for
> > > > > > > > > >> > > > > >> > > various
> > > > > > > > > >> > > > > >> > > > > > reasons (auditability, visibility,
> > > > consistency,
> > > > > > > > > >> etc.).
> > > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > > >> > > > > >> > > am
> > > > > > > > > >> > > > > >> > > > > not
> > > > > > > > > >> > > > > >> > > > > > sure about automatically deleting
> > > > configs in a
> > > > > > > way
> > > > > > > > > >> that
> > > > > > > > > >> > they
> > > > > > > > > >> > > > > >> cannot
> > > > > > > > > >> > > > > >> > > be
> > > > > > > > > >> > > > > >> > > > > > recovered. A good property for modern
> > > > systems
> > > > > > > is to
> > > > > > > > > >> > minimize
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > number
> > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David
> > > > Jacot
> > > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > > >> > > > > >> > > > > >
> > > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are
> > below:
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > > > the last
> > > > > > > step
> > > > > > > > > >> when
> > > > > > > > > >> > > > > clients
> > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in
> > which
> > > > we
> > > > > > > would
> > > > > > > > > >> have a
> > > > > > > > > >> > > > > window
> > > > > > > > > >> > > > > >> > where
> > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > protocols /
> > > > rpcs
> > > > > > > are
> > > > > > > > > >> used
> > > > > > > > > >> > by
> > > > > > > > > >> > > > > >> members
> > > > > > > > > >> > > > > >> > of
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > > > "mimic"
> > > > > > > the old
> > > > > > > > > >> > behavior
> > > > > > > > > >> > > > > >> while
> > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > need to
> > > > > > > present
> > > > > > > > > >> about.
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > Noted. I just published a new
> > version
> > > > of KIP
> > > > > > > which
> > > > > > > > > >> > includes
> > > > > > > > > >> > > > > >> more
> > > > > > > > > >> > > > > >> > > > > > > details about this. See the
> > "Supporting
> > > > > > > Online
> > > > > > > > > >> > Consumer
> > > > > > > > > >> > > > > Group
> > > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> > Deprecation,
> > > > and
> > > > > > > Migration
> > > > > > > > > >> > Plan". I
> > > > > > > > > >> > > > > >> think
> > > > > > > > > >> > > > > >> > > that
> > > > > > > > > >> > > > > >> > > > > > > I have to think through a few cases
> > > > now but
> > > > > > > the
> > > > > > > > > >> > overall idea
> > > > > > > > > >> > > > > >> and
> > > > > > > > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > far as
> > > > > > > KIP-516 the
> > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > >> are
> > > > > > > > > >> > > > > >> > > only
> > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client,
> > but
> > > > they
> > > > > > > are not
> > > > > > > > > >> > exposed
> > > > > > > > > >> > > > > via
> > > > > > > > > >> > > > > >> any
> > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > question is,
> > > > > > > first
> > > > > > > > > >> > should we
> > > > > > > > > >> > > > > >> let
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > names
> > > > -> ids
> > > > > > > mapping
> > > > > > > > > >> > itself
> > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > >> > fully
> > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > RPCs
> > > > and the
> > > > > > > new
> > > > > > > > > >> RPCs
> > > > > > > > > >> > with
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > > > exposing
> > > > > > > the
> > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > in
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > subscribe/assign, as
> > > > > > > > > >> well
> > > > > > > > > >> > as in
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would include
> > > > converting
> > > > > > > all
> > > > > > > > > >> the
> > > > > > > > > >> > offsets
> > > > > > > > > >> > > > > >> > > related
> > > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this
> > KIP,
> > > > the
> > > > > > > consumer
> > > > > > > > > >> > would be
> > > > > > > > > >> > > > > >> able
> > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > >> > > > > > > fully operate with topic ids. That
> > > > being
> > > > > > > said, it
> > > > > > > > > >> > still has
> > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > >> > > provide
> > > > > > > > > >> > > > > >> > > > > > > the topics names in various APIs so
> > > > having a
> > > > > > > > > >> mapping
> > > > > > > > > >> > in the
> > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong opinion on
> > > > this.
> > > > > > > Here I
> > > > > > > > > >> > wonder if
> > > > > > > > > >> > > > > >> this
> > > > > > > > > >> > > > > >> > > goes
> > > > > > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I
> > would
> > > > rather
> > > > > > > focus
> > > > > > > > > >> on
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> > internals
> > > > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > > > separately if
> > > > > > > we see
> > > > > > > > > >> > value in
> > > > > > > > > >> > > > > >> doing
> > > > > > > > > >> > > > > >> > > it.
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's point about
> > > > using
> > > > > > > topic ids
> > > > > > > > > >> > in the
> > > > > > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I
> > think
> > > > that
> > > > > > > there
> > > > > > > > > >> is
> > > > > > > > > >> > one
> > > > > > > > > >> > > > > >> > advantage
> > > > > > > > > >> > > > > >> > > in
> > > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer will
> > have
> > > > the
> > > > > > > > > >> opportunity
> > > > > > > > > >> > to
> > > > > > > > > >> > > > > >> validate
> > > > > > > > > >> > > > > >> > > that
> > > > > > > > > >> > > > > >> > > > > > > the topics exists before passing
> > them
> > > > into
> > > > > > > the
> > > > > > > > > >> group
> > > > > > > > > >> > > > > rebalance
> > > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> > coordinator
> > > > will
> > > > > > > also
> > > > > > > > > >> notice
> > > > > > > > > >> > it but
> > > > > > > > > >> > > > > >> it
> > > > > > > > > >> > > > > >> > > does
> > > > > > > > > >> > > > > >> > > > > > > not really have a way to reject an
> > > > invalid
> > > > > > > topic in
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> response.
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all
> > other
> > > > minor
> > > > > > > > > >> questions
> > > > > > > > > >> > > > > except
> > > > > > > > > >> > > > > >> for
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > > > personally
> > > > > > > I think
> > > > > > > > > >> > it's not
> > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > > > Pattern,
> > > > > > > but
> > > > > > > > > >> > instead we
> > > > > > > > > >> > > > > >> still
> > > > > > > > > >> > > > > >> > > use
> > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > > > subscription may
> > > > > > > > > >> be
> > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > > > very rare
> > > > > > > > > >> scenario
> > > > > > > > > >> > I felt
> > > > > > > > > >> > > > > >> > using
> > > > > > > > > >> > > > > >> > > an
> > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > subscription may
> > > > > > > be
> > > > > > > > > >> more
> > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > That could work. I have to look at
> > the
> > > > > > > differences
> > > > > > > > > >> > between
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > two
> > > > > > > > > >> > > > > >> > > > > > > engines to better understand the
> > > > potential
> > > > > > > issues.
> > > > > > > > > >> My
> > > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > > >> > > > > >> > > > > > > that would work for all the basic
> > > > regular
> > > > > > > > > >> > expressions. The
> > > > > > > > > >> > > > > >> > > differences
> > > > > > > > > >> > > > > >> > > > > > > between the two are mainly about
> > the
> > > > various
> > > > > > > > > >> character
> > > > > > > > > >> > > > > >> classes. I
> > > > > > > > > >> > > > > >> > > > > > > wonder what other people think
> > about
> > > > this.
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM
> > > > Guozhang
> > > > > > > Wang <
> > > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > > >> > > > > >> > > >
> > > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on the high
> > > > level
> > > > > > > there are
> > > > > > > > > >> > two meta
> > > > > > > > > >> > > > > >> > > points we
> > > > > > > > > >> > > > > >> > > > > need
> > > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > > > the last
> > > > > > > step
> > > > > > > > > >> when
> > > > > > > > > >> > > > > clients
> > > > > > > > > >> > > > > >> > > flip the
> > > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in
> > which
> > > > we
> > > > > > > would
> > > > > > > > > >> have a
> > > > > > > > > >> > > > > window
> > > > > > > > > >> > > > > >> > where
> > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> > protocols /
> > > > rpcs
> > > > > > > are
> > > > > > > > > >> used
> > > > > > > > > >> > by
> > > > > > > > > >> > > > > >> members
> > > > > > > > > >> > > > > >> > of
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > > > "mimic"
> > > > > > > the old
> > > > > > > > > >> > behavior
> > > > > > > > > >> > > > > >> while
> > > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> > need to
> > > > > > > present
> > > > > > > > > >> about.
> > > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> > far as
> > > > > > > KIP-516 the
> > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > >> are
> > > > > > > > > >> > > > > >> > > only
> > > > > > > > > >> > > > > >> > > > > used
> > > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client,
> > but
> > > > they
> > > > > > > are not
> > > > > > > > > >> > exposed
> > > > > > > > > >> > > > > via
> > > > > > > > > >> > > > > >> any
> > > > > > > > > >> > > > > >> > > > > public
> > > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > > question is,
> > > > > > > first
> > > > > > > > > >> > should we
> > > > > > > > > >> > > > > >> let
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> > names
> > > > -> ids
> > > > > > > mapping
> > > > > > > > > >> > itself
> > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > >> > fully
> > > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> > RPCs
> > > > and the
> > > > > > > new
> > > > > > > > > >> RPCs
> > > > > > > > > >> > with
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > > > exposing
> > > > > > > the
> > > > > > > > > >> > topic ids
> > > > > > > > > >> > > > > in
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > > subscribe/assign, as
> > > > > > > > > >> well
> > > > > > > > > >> > as in
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > > >> > deletion-and-recreation).
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all
> > other
> > > > minor
> > > > > > > > > >> questions
> > > > > > > > > >> > > > > except
> > > > > > > > > >> > > > > >> for
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > > > personally
> > > > > > > I think
> > > > > > > > > >> > it's not
> > > > > > > > > >> > > > > >> > > necessary
> > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > > > Pattern,
> > > > > > > but
> > > > > > > > > >> > instead we
> > > > > > > > > >> > > > > >> still
> > > > > > > > > >> > > > > >> > > use
> > > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > > > subscription may
> > > > > > > > > >> be
> > > > > > > > > >> > > > > >> rejected by
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > > > very rare
> > > > > > > > > >> scenario
> > > > > > > > > >> > I felt
> > > > > > > > > >> > > > > >> > using
> > > > > > > > > >> > > > > >> > > an
> > > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > > subscription may
> > > > > > > be
> > > > > > > > > >> more
> > > > > > > > > >> > > > > >> vulnerable
> > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > >> > > > > various
> > > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM
> > > > David Jacot
> > > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let
> > me
> > > > answer
> > > > > > > your
> > > > > > > > > >> > questions
> > > > > > > > > >> > > > > >> > inline.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's premature to
> > > > talk about
> > > > > > > > > >> target
> > > > > > > > > >> > > > > versions
> > > > > > > > > >> > > > > >> for
> > > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > >> > > > > >> > > > > > > > > > removal of the existing group
> > > > protocol.
> > > > > > > > > >> Unlike
> > > > > > > > > >> > KRaft,
> > > > > > > > > >> > > > > >> this
> > > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > > > > > deprecation/removal
> > > > > > > > > >> > will be
> > > > > > > > > >> > > > > >> > heavily
> > > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > >> > > > > >> > > > > > > > > > how quickly applications
> > migrate
> > > > to
> > > > > > > the new
> > > > > > > > > >> > protocol.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> > remove it.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to
> > > > release
> > > > > > > this in
> > > > > > > > > >> > 4.x, but
> > > > > > > > > >> > > > > it
> > > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > > >> > > > > >> > > > > made
> > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that as a
> > way to
> > > > > > > estimate
> > > > > > > > > >> when
> > > > > > > > > >> > we'd
> > > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > > >> > > > > >> > > > > and
> > > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I also
> > > > suggest
> > > > > > > removing
> > > > > > > > > >> > this part.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> > > > > > > explained, I
> > > > > > > > > >> plan
> > > > > > > > > >> > to
> > > > > > > > > >> > > > > >> rewrite
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > group
> > > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java while we
> > > > implement
> > > > > > > the new
> > > > > > > > > >> > protocol.
> > > > > > > > > >> > > > > >> This
> > > > > > > > > >> > > > > >> > > means
> > > > > > > > > >> > > > > >> > > > > > > > > that the internals will be
> > slightly
> > > > > > > different
> > > > > > > > > >> > (e.g.
> > > > > > > > > >> > > > > >> threading
> > > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten
> > the
> > > > > > > switch from
> > > > > > > > > >> > the old
> > > > > > > > > >> > > > > >> group
> > > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new group
> > > > coordinator
> > > > > > > to a
> > > > > > > > > >> > major
> > > > > > > > > >> > > > > >> release.
> > > > > > > > > >> > > > > >> > > The
> > > > > > > > > >> > > > > >> > > > > > > > > alternative would be to use a
> > flag
> > > > to do
> > > > > > > the
> > > > > > > > > >> > switch
> > > > > > > > > >> > > > > >> instead
> > > > > > > > > >> > > > > >> > of
> > > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the
> > > > details of
> > > > > > > the
> > > > > > > > > >> > migration
> > > > > > > > > >> > > > > >> story.
> > > > > > > > > >> > > > > >> > > It
> > > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will support
> > > > online
> > > > > > > > > >> migrations.
> > > > > > > > > >> > Is
> > > > > > > > > >> > > > > that
> > > > > > > > > >> > > > > >> > > correct?
> > > > > > > > > >> > > > > >> > > > > We
> > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > >> > > > > >> > > > > > > > > > explain this in detail. It
> > could
> > > > also
> > > > > > > be done
> > > > > > > > > >> > as a
> > > > > > > > > >> > > > > >> separate
> > > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > > >> > > > > >> > > > > if
> > > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support online
> > > > migrations
> > > > > > > for the
> > > > > > > > > >> > group.
> > > > > > > > > >> > > > > That
> > > > > > > > > >> > > > > >> > > means
> > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > a group using the old protocol
> > > > will be
> > > > > > > able to
> > > > > > > > > >> > switch to
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > new
> > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain how that
> > > > will work
> > > > > > > > > >> though.
> > > > > > > > > >> > It is
> > > > > > > > > >> > > > > >> > > basically a
> > > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> > upgraded or
> > > > > > > rolled to a
> > > > > > > > > >> > software
> > > > > > > > > >> > > > > >> > > supporting
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > new group coordinator. Both the
> > > > old and
> > > > > > > the new
> > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > >> > > > > >> > > will
> > > > > > > > > >> > > > > >> > > > > > > > > support the old protocol and
> > rely
> > > > on the
> > > > > > > same
> > > > > > > > > >> > persisted
> > > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > > >> > > > > >> > > > > > > > > they can work together. This
> > point
> > > > is an
> > > > > > > > > >> offline
> > > > > > > > > >> > > > > >> migration.
> > > > > > > > > >> > > > > >> > We
> > > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > > >> > > > > >> > > > > > > > > do this one live because it
> > would
> > > > require
> > > > > > > > > >> > shutting down
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > current
> > > > > > > > > >> > > > > >> > > > > > > > > coordinator and starting up the
> > > > new one
> > > > > > > and
> > > > > > > > > >> that
> > > > > > > > > >> > would
> > > > > > > > > >> > > > > >> cause
> > > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > > > version/IBP
> > > > > > > must be
> > > > > > > > > >> > upgraded
> > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > >> X
> > > > > > > > > >> > > > > >> > in
> > > > > > > > > >> > > > > >> > > > > order
> > > > > > > > > >> > > > > >> > > > > > > > > to enable the new protocol.
> > This
> > > > cannot
> > > > > > > be done
> > > > > > > > > >> > before
> > > > > > > > > >> > > > > 1)
> > > > > > > > > >> > > > > >> is
> > > > > > > > > >> > > > > >> > > > > > > > > terminated because the old
> > > > coordinator
> > > > > > > doesn't
> > > > > > > > > >> > support
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > new
> > > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> > upgraded
> > > > to a
> > > > > > > version
> > > > > > > > > >> > > > > supporting
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > online
> > > > > > > > > >> > > > > >> > > > > > > > > migration (must have KIP-792).
> > If
> > > > the
> > > > > > > consumer
> > > > > > > > > >> is
> > > > > > > > > >> > > > > already
> > > > > > > > > >> > > > > >> > > there.
> > > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at this
> > point.
> > > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled
> > > > with the
> > > > > > > > > >> feature
> > > > > > > > > >> > flag
> > > > > > > > > >> > > > > >> turned
> > > > > > > > > >> > > > > >> > > on.
> > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > >> > > > > >> > > > > > > > > consumer group is automatically
> > > > > > > converted when
> > > > > > > > > >> > the first
> > > > > > > > > >> > > > > >> > > consumer
> > > > > > > > > >> > > > > >> > > > > > > > > using the new protocol joins
> > the
> > > > group.
> > > > > > > While
> > > > > > > > > >> the
> > > > > > > > > >> > > > > members
> > > > > > > > > >> > > > > >> > > using the
> > > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> > upgraded,
> > > > the old
> > > > > > > > > >> protocol
> > > > > > > > > >> > is
> > > > > > > > > >> > > > > >> proxied
> > > > > > > > > >> > > > > >> > > into
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of this in
> > the
> > > > KIP.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are
> > > > pushing the
> > > > > > > pattern
> > > > > > > > > >> > > > > >> subscriptions
> > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > >> > > > > >> > > > > > > > > > but it seems like there
> > could be
> > > > some
> > > > > > > tricky
> > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for users to
> > > > detect
> > > > > > > that
> > > > > > > > > >> they
> > > > > > > > > >> > need to
> > > > > > > > > >> > > > > >> > update
> > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > > >> > > > > >> > > > > > > > > > before switching to the new
> > > > protocol?
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit more
> > > > optimistic
> > > > > > > than
> > > > > > > > > >> you
> > > > > > > > > >> > on this
> > > > > > > > > >> > > > > >> > > point. I
> > > > > > > > > >> > > > > >> > > > > > > > > believe that the majority of
> > the
> > > > cases
> > > > > > > are
> > > > > > > > > >> simple
> > > > > > > > > >> > > > > regexes
> > > > > > > > > >> > > > > >> > which
> > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > >> > > > > >> > > > > > > > > work with the new engine. The
> > > > > > > coordinator will
> > > > > > > > > >> > verify
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > regex
> > > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer if the
> > > > regex is
> > > > > > > not
> > > > > > > > > >> valid.
> > > > > > > > > >> > > > > Coming
> > > > > > > > > >> > > > > >> > back
> > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > migration path, in the worst
> > case,
> > > > the
> > > > > > > first
> > > > > > > > > >> > upgraded
> > > > > > > > > >> > > > > >> > consumer
> > > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > > >> > > > > >> > > > > > > > > the group will be rejected.
> > This
> > > > should
> > > > > > > be used
> > > > > > > > > >> > as the
> > > > > > > > > >> > > > > >> last
> > > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> > validate
> > > > their
> > > > > > > regex
> > > > > > > > > >> > before
> > > > > > > > > >> > > > > >> > upgrading
> > > > > > > > > >> > > > > >> > > > > their
> > > > > > > > > >> > > > > >> > > > > > > > > prod would be to test them with
> > > > another
> > > > > > > group.
> > > > > > > > > >> For
> > > > > > > > > >> > > > > >> instance,
> > > > > > > > > >> > > > > >> > > that
> > > > > > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod
> > > > environment.
> > > > > > > > > >> Another
> > > > > > > > > >> > way
> > > > > > > > > >> > > > > >> would be
> > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > >> > > > > > > > > extend the consumer-group tool
> > to
> > > > > > > provide a
> > > > > > > > > >> regex
> > > > > > > > > >> > > > > >> validation
> > > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> > enough in
> > > > your
> > > > > > > > > >> opinion?
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> > question,
> > > > will
> > > > > > > the
> > > > > > > > > >> Java
> > > > > > > > > >> > client
> > > > > > > > > >> > > > > >> allow
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > > >> > > > > >> > > > > > > > > > stick with the current regex
> > > > engine for
> > > > > > > > > >> > compatibility
> > > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > > >> > > > > >> > > > > For
> > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to keep using
> > > > client
> > > > > > > based
> > > > > > > > > >> > regex at
> > > > > > > > > >> > > > > >> first
> > > > > > > > > >> > > > > >> > to
> > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and then
> > > > migrate to
> > > > > > > server
> > > > > > > > > >> > based
> > > > > > > > > >> > > > > >> regexes
> > > > > > > > > >> > > > > >> > > as a
> > > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > I understand your point but I
> > am
> > > > > > > concerned that
> > > > > > > > > >> > this
> > > > > > > > > >> > > > > would
> > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > >> > > > > >> > > > > users
> > > > > > > > > >> > > > > >> > > > > > > > > to actually stay in this mode.
> > That
> > > > > > > would go
> > > > > > > > > >> > against our
> > > > > > > > > >> > > > > >> goal
> > > > > > > > > >> > > > > >> > > of
> > > > > > > > > >> > > > > >> > > > > > > > > simplifying the client because
> > we
> > > > would
> > > > > > > have to
> > > > > > > > > >> > continue
> > > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > > >> > > > > >> > > > > > > > > the metadata on the client
> > side. I
> > > > would
> > > > > > > rather
> > > > > > > > > >> > not do
> > > > > > > > > >> > > > > >> this.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> > > > > > > coordinator
> > > > > > > > > >> will
> > > > > > > > > >> > be
> > > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > > >> > > > > >> > > > > > > > > > the configurations and that
> > the
> > > > > > > > > >> configurations
> > > > > > > > > >> > will be
> > > > > > > > > >> > > > > >> > > deleted
> > > > > > > > > >> > > > > >> > > > > when
> > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a
> > > > transition to
> > > > > > > DEAD
> > > > > > > > > >> > trigger
> > > > > > > > > >> > > > > >> > deletion
> > > > > > > > > >> > > > > >> > > of
> > > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> > configurations
> > > > will be
> > > > > > > > > >> deleted
> > > > > > > > > >> > when
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > group is
> > > > > > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to store
> > the
> > > > > > > configs in
> > > > > > > > > >> the
> > > > > > > > > >> > group
> > > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > > >> > > > > >> > > > > > > > > > harder to list all cluster
> > > > configs and
> > > > > > > their
> > > > > > > > > >> > values?
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > > > > > configurations are
> > > > > > > > > >> > overrides
> > > > > > > > > >> > > > > >> of
> > > > > > > > > >> > > > > >> > > cluster
> > > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to know
> > all
> > > > the
> > > > > > > overrides
> > > > > > > > > >> > though,
> > > > > > > > > >> > > > > you
> > > > > > > > > >> > > > > >> > > would
> > > > > > > > > >> > > > > >> > > > > have
> > > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> > coordinators.
> > > > You
> > > > > > > cannot
> > > > > > > > > >> > rely on
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > metadata
> > > > > > > > > >> > > > > >> > > > > log
> > > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> > configure a
> > > > group
> > > > > > > before
> > > > > > > > > >> > starting
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > > >> > > > > >> > > > > > > > > > we considered allowing the
> > > > explicit
> > > > > > > creation
> > > > > > > > > >> of
> > > > > > > > > >> > > > > groups?
> > > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > configs could be decoupled
> > from
> > > > the
> > > > > > > group
> > > > > > > > > >> > lifecycle.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > > > automatically
> > > > > > > created in
> > > > > > > > > >> > this
> > > > > > > > > >> > > > > case.
> > > > > > > > > >> > > > > >> > > However,
> > > > > > > > > >> > > > > >> > > > > > > > > the configs will be lost after
> > the
> > > > > > > retention
> > > > > > > > > >> > period of
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > group
> > > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> > Consumer.subscribe
> > > > method
> > > > > > > for the
> > > > > > > > > >> > Java
> > > > > > > > > >> > > > > >> client
> > > > > > > > > >> > > > > >> > > still
> > > > > > > > > >> > > > > >> > > > > take
> > > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of
> > do
> > > > we
> > > > > > > have to
> > > > > > > > > >> > introduce
> > > > > > > > > >> > > > > an
> > > > > > > > > >> > > > > >> > > overload?
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > That's a very group question. I
> > > > forgot
> > > > > > > about
> > > > > > > > > >> that
> > > > > > > > > >> > one.
> > > > > > > > > >> > > > > As
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is
> > not
> > > > fully
> > > > > > > > > >> compatible
> > > > > > > > > >> > with
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > engine
> > > > > > > > > >> > > > > >> > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might be
> > better
> > > > to
> > > > > > > deprecate
> > > > > > > > > >> > it and
> > > > > > > > > >> > > > > >> use an
> > > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > > >> > > > > >> > > > > > > > > which takes a string. We would
> > > > rely on
> > > > > > > the
> > > > > > > > > >> server
> > > > > > > > > >> > side
> > > > > > > > > >> > > > > >> > > validation.
> > > > > > > > > >> > > > > >> > > > > > > > > During the migration, I think
> > that
> > > > we
> > > > > > > could
> > > > > > > > > >> still
> > > > > > > > > >> > try to
> > > > > > > > > >> > > > > >> > > toString
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That should
> > > > work, I
> > > > > > > think, in
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> majority
> > > > > > > > > >> > > > > >> > > of the
> > > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine
> > that we
> > > > > > > should be
> > > > > > > > > >> > clearer
> > > > > > > > > >> > > > > about
> > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > > >> > > > > >> > > > > > > > > > switch to
> > IBP/metadata.version
> > > > from the
> > > > > > > > > >> feature
> > > > > > > > > >> > flag.
> > > > > > > > > >> > > > > >> Maybe
> > > > > > > > > >> > > > > >> > > we
> > > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > > >> > > > > >> > > > > > > > > > can switch the default for
> > the
> > > > feature
> > > > > > > flag
> > > > > > > > > >> to
> > > > > > > > > >> > true
> > > > > > > > > >> > > > > >> based
> > > > > > > > > >> > > > > >> > on
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once we
> > want to
> > > > make
> > > > > > > it the
> > > > > > > > > >> > default.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that feature
> > > > flag
> > > > > > > mainly
> > > > > > > > > >> > during the
> > > > > > > > > >> > > > > >> > > development
> > > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> > mentioned
> > > > it, I
> > > > > > > think,
> > > > > > > > > >> > because
> > > > > > > > > >> > > > > we
> > > > > > > > > >> > > > > >> > > could
> > > > > > > > > >> > > > > >> > > > > use
> > > > > > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs
> > > > don't
> > > > > > > mention
> > > > > > > > > >> the
> > > > > > > > > >> > > > > required
> > > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> > consistency.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > > >> > > > > requires
> > > > > > > > > >> > > > > >> > > "Read
> > > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do more
> > than
> > > > > > > reading.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> > current
> > > > > > > protocol works
> > > > > > > > > >> > though.
> > > > > > > > > >> > > > > We
> > > > > > > > > >> > > > > >> > only
> > > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to join a
> > > > group. We
> > > > > > > could
> > > > > > > > > >> > consider
> > > > > > > > > >> > > > > >> > > changing
> > > > > > > > > >> > > > > >> > > > > this
> > > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that it is
> > worth
> > > > it.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic recreation
> > > > handled by
> > > > > > > the
> > > > > > > > > >> > consumer
> > > > > > > > > >> > > > > with
> > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > >> > > new
> > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be good to
> > > > have a
> > > > > > > section
> > > > > > > > > >> on
> > > > > > > > > >> > this.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > > > perspective, the
> > > > > > > new
> > > > > > > > > >> topic
> > > > > > > > > >> > will
> > > > > > > > > >> > > > > >> have a
> > > > > > > > > >> > > > > >> > > new
> > > > > > > > > >> > > > > >> > > > > > > > > topic id so it will treat it
> > like a
> > > > > > > topic with
> > > > > > > > > >> a
> > > > > > > > > >> > > > > different
> > > > > > > > > >> > > > > >> > > name.
> > > > > > > > > >> > > > > >> > > > > The
> > > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> > fetch/commit
> > > > > > > offsets
> > > > > > > > > >> APIs
> > > > > > > > > >> > do not
> > > > > > > > > >> > > > > >> > support
> > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would
> > reuse the
> > > > > > > offsets
> > > > > > > > > >> based
> > > > > > > > > >> > on the
> > > > > > > > > >> > > > > >> > same.
> > > > > > > > > >> > > > > >> > > I
> > > > > > > > > >> > > > > >> > > > > think
> > > > > > > > > >> > > > > >> > > > > > > > > that we should update those
> > APIs
> > > > as well
> > > > > > > in
> > > > > > > > > >> order
> > > > > > > > > >> > to be
> > > > > > > > > >> > > > > >> > > consistent
> > > > > > > > > >> > > > > >> > > > > end
> > > > > > > > > >> > > > > >> > > > > > > > > to end. That would strengthen
> > the
> > > > > > > semantics of
> > > > > > > > > >> the
> > > > > > > > > >> > > > > >> consumer.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will
> > > > write the
> > > > > > > new
> > > > > > > > > >> > coordinator
> > > > > > > > > >> > > > > >> in
> > > > > > > > > >> > > > > >> > > Java.
> > > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > >> > > > > >> > > > > > > > > > this is an implementation
> > > > detail, do
> > > > > > > we plan
> > > > > > > > > >> to
> > > > > > > > > >> > have a
> > > > > > > > > >> > > > > >> new
> > > > > > > > > >> > > > > >> > > gradle
> > > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability
> > > > goal when
> > > > > > > it
> > > > > > > > > >> comes
> > > > > > > > > >> > to how
> > > > > > > > > >> > > > > >> many
> > > > > > > > > >> > > > > >> > > > > members
> > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at the
> > > > moment. The
> > > > > > > > > >> protocol
> > > > > > > > > >> > should
> > > > > > > > > >> > > > > >> > > support
> > > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > > >> > > > > >> > > > > > > > > of members per group. We will
> > > > measure
> > > > > > > this when
> > > > > > > > > >> > we have
> > > > > > > > > >> > > > > a
> > > > > > > > > >> > > > > >> > first
> > > > > > > > > >> > > > > >> > > > > > > > > implementation. Note that we
> > might
> > > > have
> > > > > > > other
> > > > > > > > > >> > > > > bottlenecks
> > > > > > > > > >> > > > > >> > down
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> > > > > > > SubscribedTopidIds
> > > > > > > > > >> > instead
> > > > > > > > > >> > > > > of
> > > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > > >> > > > > >> Is
> > > > > > > > > >> > > > > >> > the
> > > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > > since we have to resolve the
> > > > regex on
> > > > > > > the
> > > > > > > > > >> > server, we
> > > > > > > > > >> > > > > >> can do
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > same
> > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> > difference
> > > > is that
> > > > > > > > > >> sending
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> regex is
> > > > > > > > > >> > > > > >> > > more
> > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the topic
> > names
> > > > is less
> > > > > > > > > >> > efficient.
> > > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to
> > handle
> > > > if we
> > > > > > > have
> > > > > > > > > >> topic
> > > > > > > > > >> > ids.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > The idea was to consolidate the
> > > > metadata
> > > > > > > lookup
> > > > > > > > > >> > on the
> > > > > > > > > >> > > > > >> server
> > > > > > > > > >> > > > > >> > > for
> > > > > > > > > >> > > > > >> > > > > both
> > > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree with your
> > > > point. As
> > > > > > > a
> > > > > > > > > >> second
> > > > > > > > > >> > > > > though,
> > > > > > > > > >> > > > > >> > using
> > > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > > >> > > > > >> > > > > > > > > ids may be better here for the
> > > > delete and
> > > > > > > > > >> > recreation
> > > > > > > > > >> > > > > case.
> > > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow
> > users to
> > > > > > > subscribe
> > > > > > > > > >> with
> > > > > > > > > >> > topic
> > > > > > > > > >> > > > > >> ids
> > > > > > > > > >> > > > > >> > in
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > > > future because that is the only
> > > > way to be
> > > > > > > > > >> really
> > > > > > > > > >> > robust
> > > > > > > > > >> > > > > to
> > > > > > > > > >> > > > > >> > > topic
> > > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM
> > > > David
> > > > > > > Jacot <
> > > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your comments.
> > Please
> > > > find
> > > > > > > my
> > > > > > > > > >> answers
> > > > > > > > > >> > > > > below.
> > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol
> > relies on
> > > > > > > topic IDs
> > > > > > > > > >> > with the
> > > > > > > > > >> > > > > >> > > exception
> > > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > > >> > > > > >> I
> > > > > > > > > >> > > > > >> > am
> > > > > > > > > >> > > > > >> > > not
> > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > >> > > > > >> > > > > > > > > > if using topic names is the
> > > > right call
> > > > > > > here.
> > > > > > > > > >> I
> > > > > > > > > >> > need to
> > > > > > > > > >> > > > > >> > think
> > > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > > >> > > > > >> > > > > > > > > > a little more. Obviously, the
> > > > KIP does
> > > > > > > not
> > > > > > > > > >> > change the
> > > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic
> > IDs.
> > > > This
> > > > > > > may be
> > > > > > > > > >> > something
> > > > > > > > > >> > > > > >> that
> > > > > > > > > >> > > > > >> > we
> > > > > > > > > >> > > > > >> > > > > should
> > > > > > > > > >> > > > > >> > > > > > > > > > include though as it would
> > give
> > > > better
> > > > > > > > > >> overall
> > > > > > > > > >> > > > > >> guarantee in
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I think that
> > I
> > > > should
> > > > > > > not
> > > > > > > > > >> have
> > > > > > > > > >> > > > > mentioned
> > > > > > > > > >> > > > > >> > this
> > > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it. We can
> > > > use an
> > > > > > > internal
> > > > > > > > > >> > > > > >> configuration
> > > > > > > > > >> > > > > >> > > while
> > > > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> > > > > > > supported. The
> > > > > > > > > >> > change is
> > > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > > >> > > > > >> > > > > > > > > > only requirement is that the
> > > > cluster
> > > > > > > uses
> > > > > > > > > >> topic
> > > > > > > > > >> > IDs.
> > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53
> > PM
> > > > > > > Guozhang
> > > > > > > > > >> Wang <
> > > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the feedback.
> > Here
> > > > are
> > > > > > > some
> > > > > > > > > >> replies
> > > > > > > > > >> > > > > inlined
> > > > > > > > > >> > > > > >> > > below:
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at
> > 2:53 AM
> > > > > > > Ismael Juma
> > > > > > > > > >> <
> > > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This
> > has
> > > > the
> > > > > > > > > >> potential
> > > > > > > > > >> > to be a
> > > > > > > > > >> > > > > >> > great
> > > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> > questions/comments:
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> > premature to
> > > > talk
> > > > > > > about
> > > > > > > > > >> > target
> > > > > > > > > >> > > > > >> versions
> > > > > > > > > >> > > > > >> > > for
> > > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the existing
> > group
> > > > > > > protocol.
> > > > > > > > > >> > Unlike
> > > > > > > > > >> > > > > >> KRaft,
> > > > > > > > > >> > > > > >> > > this
> > > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > > > > > > > >> > deprecation/removal will
> > > > > > > > > >> > > > > >> be
> > > > > > > > > >> > > > > >> > > heavily
> > > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly applications
> > > > migrate
> > > > > > > to the
> > > > > > > > > >> new
> > > > > > > > > >> > > > > >> protocol.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I
> > think
> > > > we can
> > > > > > > > > >> remove
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> proposed
> > > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> > Deprecation,
> > > > and
> > > > > > > Migration
> > > > > > > > > >> > Plan` and
> > > > > > > > > >> > > > > >> > > instead
> > > > > > > > > >> > > > > >> > > > > just
> > > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide in the
> > > > future
> > > > > > > about
> > > > > > > > > >> when
> > > > > > > > > >> > we
> > > > > > > > > >> > > > > would
> > > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> > intend to
> > > > > > > release this
> > > > > > > > > >> > in 4.x,
> > > > > > > > > >> > > > > >> but
> > > > > > > > > >> > > > > >> > it
> > > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added that as
> > a
> > > > way to
> > > > > > > > > >> estimate
> > > > > > > > > >> > when
> > > > > > > > > >> > > > > we'd
> > > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I
> > also
> > > > suggest
> > > > > > > > > >> removing
> > > > > > > > > >> > this
> > > > > > > > > >> > > > > >> part.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> > specifically
> > > > > > > related to
> > > > > > > > > >> the
> > > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for
> > client
> > > > > > > upgrades.
> > > > > > > > > >> I.e.
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > > >> > > > > >> > > > > > > > > > > implementation may be done
> > > > first,
> > > > > > > and then
> > > > > > > > > >> the
> > > > > > > > > >> > > > > client
> > > > > > > > > >> > > > > >> > side,
> > > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as "released"
> > by
> > > > the
> > > > > > > time
> > > > > > > > > >> clients
> > > > > > > > > >> > > > > >> > > implementations
> > > > > > > > > >> > > > > >> > > > > are
> > > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable the
> > > > feature the
> > > > > > > > > >> clients
> > > > > > > > > >> > need to
> > > > > > > > > >> > > > > >> > first
> > > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling
> > bounce
> > > > and
> > > > > > > then set
> > > > > > > > > >> > the flag
> > > > > > > > > >> > > > > >> > with a
> > > > > > > > > >> > > > > >> > > > > second
> > > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel
> > it's
> > > > > > > better to be
> > > > > > > > > >> > released
> > > > > > > > > >> > > > > >> in a
> > > > > > > > > >> > > > > >> > > major
> > > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out
> > the
> > > > > > > details of
> > > > > > > > > >> the
> > > > > > > > > >> > > > > migration
> > > > > > > > > >> > > > > >> > > story.
> > > > > > > > > >> > > > > >> > > > > It
> > > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we will
> > support
> > > > online
> > > > > > > > > >> > migrations. Is
> > > > > > > > > >> > > > > >> that
> > > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in detail.
> > It
> > > > could
> > > > > > > also be
> > > > > > > > > >> > done as a
> > > > > > > > > >> > > > > >> > > separate
> > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the
> > part we
> > > > can
> > > > > > > be more
> > > > > > > > > >> > concrete
> > > > > > > > > >> > > > > >> about
> > > > > > > > > >> > > > > >> > > for
> > > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > > >> > > > > >> > > > > > > > > > > this is related to your
> > > > question 2)
> > > > > > > above).
> > > > > > > > > >> > We will
> > > > > > > > > >> > > > > >> work
> > > > > > > > > >> > > > > >> > on
> > > > > > > > > >> > > > > >> > > > > making
> > > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel as we
> > > > solicit
> > > > > > > more
> > > > > > > > > >> > feedback.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are
> > > > pushing
> > > > > > > the
> > > > > > > > > >> > pattern
> > > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > > >> > > > > >> > > > > to
> > > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like there
> > > > could be
> > > > > > > some
> > > > > > > > > >> tricky
> > > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for
> > users to
> > > > > > > detect that
> > > > > > > > > >> > they
> > > > > > > > > >> > > > > need
> > > > > > > > > >> > > > > >> to
> > > > > > > > > >> > > > > >> > > update
> > > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > > >> > > > > >> > > > > > > > > > > > before switching to the
> > new
> > > > > > > protocol?
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need some
> > > > tooling for
> > > > > > > > > >> non-java
> > > > > > > > > >> > client
> > > > > > > > > >> > > > > >> > users
> > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> > > > > > > switching to
> > > > > > > > > >> the
> > > > > > > > > >> > new
> > > > > > > > > >> > > > > >> > protocol.
> > > > > > > > > >> > > > > >> > > I
> > > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top of my
> > head
> > > > > > > though,
> > > > > > > > > >> maybe
> > > > > > > > > >> > others
> > > > > > > > > >> > > > > >> like
> > > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the last
> > > > question,
> > > > > > > will the
> > > > > > > > > >> > Java
> > > > > > > > > >> > > > > >> client
> > > > > > > > > >> > > > > >> > > allow
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the current
> > regex
> > > > > > > engine for
> > > > > > > > > >> > > > > >> compatibility
> > > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to keep
> > using
> > > > > > > client
> > > > > > > > > >> based
> > > > > > > > > >> > regex
> > > > > > > > > >> > > > > at
> > > > > > > > > >> > > > > >> > > first to
> > > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple and
> > then
> > > > migrate
> > > > > > > to
> > > > > > > > > >> > server based
> > > > > > > > > >> > > > > >> > > regexes
> > > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not thought
> > > > about
> > > > > > > that for
> > > > > > > > > >> > java
> > > > > > > > > >> > > > > >> clients,
> > > > > > > > > >> > > > > >> > > and
> > > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > > > compatibility
> > > > > > > issues do
> > > > > > > > > >> > you have
> > > > > > > > > >> > > > > in
> > > > > > > > > >> > > > > >> > > mind?
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that the
> > group
> > > > > > > coordinator
> > > > > > > > > >> > will be
> > > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > > >> > > > > >> > > > > > > > > > > > the configurations and
> > that
> > > > the
> > > > > > > > > >> > configurations
> > > > > > > > > >> > > > > will
> > > > > > > > > >> > > > > >> be
> > > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> > > > > > > transition to
> > > > > > > > > >> DEAD
> > > > > > > > > >> > > > > trigger
> > > > > > > > > >> > > > > >> > > deletion
> > > > > > > > > >> > > > > >> > > > > of
> > > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state
> > is an
> > > > > > > ending
> > > > > > > > > >> state
> > > > > > > > > >> > (we
> > > > > > > > > >> > > > > would
> > > > > > > > > >> > > > > >> > only
> > > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > > > state when the group is
> > EMPTY
> > > > and
> > > > > > > also all
> > > > > > > > > >> of
> > > > > > > > > >> > its
> > > > > > > > > >> > > > > >> > metadata
> > > > > > > > > >> > > > > >> > > are
> > > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this
> > > > group
> > > > > > > would
> > > > > > > > > >> never
> > > > > > > > > >> > be
> > > > > > > > > >> > > > > >> revived.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to
> > store
> > > > the
> > > > > > > configs
> > > > > > > > > >> in
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > group
> > > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> > cluster
> > > > configs
> > > > > > > and
> > > > > > > > > >> their
> > > > > > > > > >> > > > > values?
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > That's a good question,
> > and our
> > > > > > > thoughts
> > > > > > > > > >> are
> > > > > > > > > >> > that
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > so-called
> > > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> > overrides
> > > > of the
> > > > > > > > > >> > cluster-level
> > > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > > >> > > > > >> > > > > > > > > > > customized per group so
> > when an
> > > > > > > admin list
> > > > > > > > > >> > cluster
> > > > > > > > > >> > > > > >> > configs
> > > > > > > > > >> > > > > >> > > it's
> > > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> > > > > > > "defaults", not
> > > > > > > > > >> > showing
> > > > > > > > > >> > > > > >> any
> > > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone
> > > > configure a
> > > > > > > group
> > > > > > > > > >> > before
> > > > > > > > > >> > > > > >> starting
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > > > >> > > > > >> > > > > > > > > > > > we considered allowing
> > the
> > > > explicit
> > > > > > > > > >> > creation of
> > > > > > > > > >> > > > > >> groups?
> > > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> > decoupled
> > > > from
> > > > > > > the group
> > > > > > > > > >> > > > > lifecycle.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be created
> > > > before
> > > > > > > the group
> > > > > > > > > >> > itself
> > > > > > > > > >> > > > > as
> > > > > > > > > >> > > > > >> an
> > > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> > requires
> > > > the
> > > > > > > > > >> corresponding
> > > > > > > > > >> > > > > >> request to
> > > > > > > > > >> > > > > >> > > be
> > > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator based on
> > the
> > > > group
> > > > > > > id ---
> > > > > > > > > >> > the only
> > > > > > > > > >> > > > > >> > thing
> > > > > > > > > >> > > > > >> > > that
> > > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > > > >> > > > > >> > > > > > > > > > > when the group itself is
> > gone
> > > > we
> > > > > > > also check
> > > > > > > > > >> > if there
> > > > > > > > > >> > > > > >> are
> > > > > > > > > >> > > > > >> > > any
> > > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > > > >> > > > > >> > > > > > > > > > > entities related to that
> > group
> > > > and
> > > > > > > delete
> > > > > > > > > >> as
> > > > > > > > > >> > well.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed
> > > > introduces an
> > > > > > > > > >> > asymmetry on
> > > > > > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > creation
> > > > > > > > > >> > > > > >> > > > > /
> > > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the config
> > > > entities,
> > > > > > > and we
> > > > > > > > > >> > would like
> > > > > > > > > >> > > > > >> to
> > > > > > > > > >> > > > > >> > > hear
> > > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we should
> > aim
> > > > for
> > > > > > > symmetry
> > > > > > > > > >> > i.e.
> > > > > > > > > >> > > > > >> totally
> > > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not
> > delete
> > > > them at
> > > > > > > all
> > > > > > > > > >> when
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> group
> > > > > > > > > >> > > > > >> > is
> > > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > > >> > > > > >> > > > > > > > > > > require explicit deletion
> > > > operations
> > > > > > > by
> > > > > > > > > >> > themselves.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > > Consumer.subscribe
> > > > > > > method for
> > > > > > > > > >> > the Java
> > > > > > > > > >> > > > > >> > client
> > > > > > > > > >> > > > > >> > > > > still
> > > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > `java.util.regex.Pattern` of
> > > > do we
> > > > > > > have
> > > > > > > > > >> to
> > > > > > > > > >> > > > > >> introduce an
> > > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not need to
> > > > introduce
> > > > > > > an
> > > > > > > > > >> > overload, but
> > > > > > > > > >> > > > > >> I'm
> > > > > > > > > >> > > > > >> > > all
> > > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > > >> > > > > >> > > > > > > > > > > may be some compatibility
> > > > issues
> > > > > > > that we
> > > > > > > > > >> may
> > > > > > > > > >> > > > > overlook.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine
> > > > that we
> > > > > > > should
> > > > > > > > > >> be
> > > > > > > > > >> > clearer
> > > > > > > > > >> > > > > >> > about
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > > IBP/metadata.version
> > > > > > > from the
> > > > > > > > > >> > feature
> > > > > > > > > >> > > > > >> flag.
> > > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > > >> > > > > >> > > > > we
> > > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the default
> > for
> > > > the
> > > > > > > feature
> > > > > > > > > >> flag
> > > > > > > > > >> > to
> > > > > > > > > >> > > > > true
> > > > > > > > > >> > > > > >> > > based on
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version once we
> > > > want to
> > > > > > > make it
> > > > > > > > > >> the
> > > > > > > > > >> > > > > >> default.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol
> > APIs
> > > > don't
> > > > > > > mention
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > >> required
> > > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > > >> > > > > >> > > > > it
> > > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that for
> > > > consistency.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do
> > that.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > > >> > > > > >> > requires
> > > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to do
> > more
> > > > than
> > > > > > > reading.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought myself
> > as
> > > > well,
> > > > > > > but in
> > > > > > > > > >> the
> > > > > > > > > >> > end we
> > > > > > > > > >> > > > > >> > could
> > > > > > > > > >> > > > > >> > > not
> > > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > > >> > > > > >> > > > > > > > > > > better alternative: adding
> > > > Write
> > > > > > > Group
> > > > > > > > > >> seems
> > > > > > > > > >> > an
> > > > > > > > > >> > > > > >> overkill
> > > > > > > > > >> > > > > >> > > here
> > > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we
> > only
> > > > have
> > > > > > > Read /
> > > > > > > > > >> > Delete
> > > > > > > > > >> > > > > and
> > > > > > > > > >> > > > > >> > > Describe
> > > > > > > > > >> > > > > >> > > > > on
> > > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to hear
> > others
> > > > > > > thoughts.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> > recreation
> > > > > > > handled by
> > > > > > > > > >> the
> > > > > > > > > >> > > > > consumer
> > > > > > > > > >> > > > > >> > with
> > > > > > > > > >> > > > > >> > > the
> > > > > > > > > >> > > > > >> > > > > new
> > > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be
> > good
> > > > to have
> > > > > > > a
> > > > > > > > > >> > section on
> > > > > > > > > >> > > > > >> this.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > > > subscription
> > > > > > > right? Yes
> > > > > > > > > >> > we can
> > > > > > > > > >> > > > > >> add a
> > > > > > > > > >> > > > > >> > > > > section
> > > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically the
> > idea
> > > > is that
> > > > > > > > > >> consumer
> > > > > > > > > >> > would
> > > > > > > > > >> > > > > be
> > > > > > > > > >> > > > > >> > > totally
> > > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as it's
> > > > handled all
> > > > > > > by the
> > > > > > > > > >> > brokers.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we
> > will
> > > > write
> > > > > > > the
> > > > > > > > > >> new
> > > > > > > > > >> > > > > >> coordinator
> > > > > > > > > >> > > > > >> > in
> > > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > > >> > > > > >> > > > > > > > > > > > this is an implementation
> > > > detail,
> > > > > > > do we
> > > > > > > > > >> > plan to
> > > > > > > > > >> > > > > >> have a
> > > > > > > > > >> > > > > >> > > new
> > > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought about
> > > > that. But
> > > > > > > I think
> > > > > > > > > >> > the
> > > > > > > > > >> > > > > answer
> > > > > > > > > >> > > > > >> > > should
> > > > > > > > > >> > > > > >> > > > > be
> > > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> > scalability
> > > > goal
> > > > > > > when it
> > > > > > > > > >> > comes to
> > > > > > > > > >> > > > > >> how
> > > > > > > > > >> > > > > >> > > many
> > > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> > support?
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I think we
> > > > should
> > > > > > > shoot for
> > > > > > > > > >> > 1000s of
> > > > > > > > > >> > > > > >> > > members.
> > > > > > > > > >> > > > > >> > > > > But
> > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals also
> > depend
> > > > on the
> > > > > > > offset
> > > > > > > > > >> > > > > management
> > > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> > coordinator
> > > > > > > which we
> > > > > > > > > >> did
> > > > > > > > > >> > not
> > > > > > > > > >> > > > > >> cover in
> > > > > > > > > >> > > > > >> > > this
> > > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a number that
> > > > applies
> > > > > > > > > >> > universally.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider
> > having
> > > > > > > > > >> > SubscribedTopidIds
> > > > > > > > > >> > > > > >> instead
> > > > > > > > > >> > > > > >> > of
> > > > > > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > > > >> > > > > >> > > Is the
> > > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to resolve
> > the
> > > > regex
> > > > > > > on the
> > > > > > > > > >> > server,
> > > > > > > > > >> > > > > we
> > > > > > > > > >> > > > > >> > can
> > > > > > > > > >> > > > > >> > > do
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > > > difference is
> > > > > > > that
> > > > > > > > > >> > sending the
> > > > > > > > > >> > > > > >> > regex
> > > > > > > > > >> > > > > >> > > is
> > > > > > > > > >> > > > > >> > > > > more
> > > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending the topic
> > > > names is
> > > > > > > less
> > > > > > > > > >> > efficient.
> > > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is easier to
> > > > handle if
> > > > > > > we have
> > > > > > > > > >> > topic
> > > > > > > > > >> > > > > ids.
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to still
> > let
> > > > the
> > > > > > > clients
> > > > > > > > > >> send
> > > > > > > > > >> > names
> > > > > > > > > >> > > > > >> is to
> > > > > > > > > >> > > > > >> > > keep
> > > > > > > > > >> > > > > >> > > > > the
> > > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids
> > on
> > > > the
> > > > > > > broker /
> > > > > > > > > >> > admin
> > > > > > > > > >> > > > > client
> > > > > > > > > >> > > > > >> > > only.
> > > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > > >> > > > > >> > > > > > > > > > > although we added topic id
> > in
> > > > > > > KIP-516, we
> > > > > > > > > >> > never
> > > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> > leveraging
> > > > the
> > > > > > > related
> > > > > > > > > >> > newer
> > > > > > > > > >> > > > > >> versioned
> > > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic id as
> > empty
> > > > UUID.
> > > > > > > We
> > > > > > > > > >> want
> > > > > > > > > >> > to keep
> > > > > > > > > >> > > > > >> the
> > > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> > delegate
> > > > the
> > > > > > > reasoning
> > > > > > > > > >> on
> > > > > > > > > >> > broker
> > > > > > > > > >> > > > > >> and
> > > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at
> > 10:45
> > > > AM
> > > > > > > David
> > > > > > > > > >> Jacot
> > > > > > > > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> > > > > > > discussion
> > > > > > > > > >> thread
> > > > > > > > > >> > on
> > > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > > >> > > > > >> > > The
> > > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the
> > Consumer
> > > > > > > Rebalance
> > > > > > > > > >> > Protocol.
> > > > > > > > > >> > > > > >> With
> > > > > > > > > >> > > > > >> > > this
> > > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the rebalance
> > > > protocol
> > > > > > > (for
> > > > > > > > > >> > consumers)
> > > > > > > > > >> > > > > >> more
> > > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to
> > > > implement for
> > > > > > > > > >> > clients, and
> > > > > > > > > >> > > > > >> easier
> > > > > > > > > >> > > > > >> > > to
> > > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > > > >> > > > > >> > > > >
> > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a look and
> > let
> > > > me
> > > > > > > know what
> > > > > > > > > >> > you
> > > > > > > > > >> > > > > think.
> > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from
> > > > July
> > > > > > > 18th to
> > > > > > > > > >> > August 8th.
> > > > > > > > > >> > > > > >> That
> > > > > > > > > >> > > > > >> > > gives
> > > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read and
> > digest
> > > > this
> > > > > > > long
> > > > > > > > > >> KIP.
> > > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > > >> > > > > >> > > > >
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> > >
> > > > > > > > > >> > > > > >> >
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > >> --
> > > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > > >> > > > > >>
> > > > > > > > > >> > > > > >
> > > > > > > > > >> > > > >
> > > > > > > > > >> >
> > > > > > > > > >>
> > > > > > > > > >
> > > > > > >
> > > >
> > >
> > >
> > > --
> > > -- Guozhang
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hello David,

Alright I think that's sufficient. Just to make that clear in the doc,
could we update:

1) the heartbeat request handling section, stating when coordinator will
trigger rebalance based on the HB's member metadata / reason?
2) the "Rebalance Triggers" section to include what we described in "Group
Epoch - Trigger a rebalance" section as well?


Guozhang

On Fri, Sep 9, 2022 at 1:28 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Guozhang,
>
> I thought that the assignor will always be consulted when the next
> heartbeat request is constructed. In other words,
> `PartitionAssignor#metadata` will be called for every heartbeat. This
> gives the opportunity for the assignor to enforce a rebalance by
> setting the reason to a non-zero value or by changing the bytes. Do
> you think that this is not sufficient? Are you concerned by the delay?
>
> Best,
> David
>
> On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com> wrote:
> >
> > Hello David,
> >
> > One of Jun's comments make me thinking:
> >
> > ```
> > In this case, a new assignment is triggered by the client side
> > assignor. When constructing the HB, the consumer will always consult
> > the client side assignor and propagate the information to the group
> > coordinator. In other words, we don't expect users to call
> > Consumer#enforceRebalance anymore.
> > ```
> >
> > As I looked at the current PartitionAssignor's interface, we actually do
> > not have a way yet to instruct how to construct the next HB request, e.g.
> > when the assignor wants to enforce a new rebalance with a new assignment,
> > we'd need some customizable APIs inside the PartitionAssignor to indicate
> > the next HB telling broker about so. WDYT about adding such an API on the
> > PartitionAssignor?
> >
> >
> > Guozhang
> >
> >
> > On Tue, Sep 6, 2022 at 6:09 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi Jun,
> > >
> > > I have updated the KIP to include your feedback. I have also tried to
> > > clarify the parts which were not cleared.
> > >
> > > Best,
> > > David
> > >
> > > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io>
> wrote:
> > > >
> > > > Hi Jun,
> > > >
> > > > Thanks for your feedback. Let me start by answering your questions
> > > > inline and I will update the KIP next week.
> > > >
> > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to
> be
> > > fewer
> > > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > > comments below.
> > > >
> > > > I would also add that the KIP removes the global sync barrier in the
> > > > protocol which is essential to improve group stability and
> > > > scalability, and the KIP also simplifies the client by moving most of
> > > > the logic to the server side.
> > > >
> > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > > changing
> > > > > of the partition assignor in the consumers?
> > > >
> > > > Definitely. The group coordinator will use the assignor used by a
> > > > majority of the members. This allows the group to move from one
> > > > assignor to another by a roll. This is explained in the Assignor
> > > > Selection chapter.
> > > >
> > > > > 30.2 For each field, could you explain whether it's required in
> every
> > > > > request or the scenarios when it needs to be filled? For example,
> it's
> > > not
> > > > > clear to me when TopicPartitions needs to be filled.
> > > >
> > > > The client is expected to set those fields in case of a connection
> > > > issue (e.g. timeout) or when the fields have changed since the last
> > > > HB. The server populates those fields as long as the member is not
> > > > fully reconciled - the member should acknowledge that it has the
> > > > expected epoch and assignment. I will clarify this in the KIP.
> > > >
> > > > > 31. In the current consumer protocol, the rack affinity between the
> > > client
> > > > > and the broker is only considered during fetching, but not during
> > > assigning
> > > > > partitions to consumers. Sometimes, once the assignment is made,
> there
> > > is
> > > > > no opportunity for read affinity because no replicas of assigned
> > > partitions
> > > > > are close to the member. I am wondering if we should use this
> > > opportunity
> > > > > to address this by including rack in GroupMember.
> > > >
> > > > That's an interesting idea. I don't see any issue with adding the
> rack
> > > > to the members. I will do so.
> > > >
> > > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > > coordinator is. By moving the event loop model, it seems that we
> could
> > > add
> > > > > a metric that tracks the fraction of the time the event loop is
> doing
> > > the
> > > > > actual work.
> > > >
> > > > That's a great idea. I will add it. Thanks.
> > > >
> > > > > 33. Could we add a section on coordinator failover handling? For
> > > example,
> > > > > does it need to trigger the check if any group with the wildcard
> > > > > subscription now has a new matching topic?
> > > >
> > > > Sure. When the new group coordinator takes over, it has to:
> > > > * Setup the session timeouts.
> > > > * Trigger a new assignment if a client side assignor is used. We
> don't
> > > > store the information about the member selected to run the assignment
> > > > so we have to start a new one.
> > > > * Update the topics metadata, verify the wildcard subscriptions, and
> > > > trigger a rebalance if needed.
> > > >
> > > > > 34. ConsumerGroupMetadataValue,
> ConsumerGroupPartitionMetadataValue,
> > > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > > field
> > > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > > reflect
> > > > > the latest group epoch? What about the one in
> > > > > ConsumerGroupPartitionMetadataValue and
> > > ConsumerGroupMemberMetadataValue?
> > > >
> > > > Sure. I will clarify that but it is always the latest group epoch.
> > > > When the group state is updated, the group epoch is bumped so we use
> > > > that one for all the change records related to the update.
> > > >
> > > > > 35. "the group coordinator will ensure that the following
> invariants
> > > are
> > > > > met: ... All members exists." It's possible for a member not to
> get any
> > > > > assigned partitions, right?
> > > >
> > > > That's right. Here I meant that the members provided by the assignor
> > > > in the assignment must exist in the group. The assignor can not make
> > > > up new member ids.
> > > >
> > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> When
> > > would
> > > > > a consumer rejoin and what member id would be used?
> > > >
> > > > A member is expected to abandon all its partitions and rejoins when
> it
> > > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > > coordinator will have removed the member from the group. The member
> > > > can rejoin the group with the same member id but with 0 as epoch. Let
> > > > me see if I can clarify this in the KIP.
> > > >
> > > > > 37. "Instead, power users will have the ability to trigger a
> > > reassignment
> > > > > by either providing a non-zero reason or by updating the assignor
> > > > > metadata." Hmm, this seems to be conflicting with the deprecation
> of
> > > > > Consumer#enforeRebalance.
> > > >
> > > > In this case, a new assignment is triggered by the client side
> > > > assignor. When constructing the HB, the consumer will always consult
> > > > the client side assignor and propagate the information to the group
> > > > coordinator. In other words, we don't expect users to call
> > > > Consumer#enforceRebalance anymore.
> > > >
> > > > > 38. The reassignment examples are nice. But the section seems to
> have
> > > > > multiple typos.
> > > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > epoch=3,
> > > > > partitions=[foo-1] too early.
> > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > partitions=[foo-0].
> > > >
> > > > Sorry for that! I will revise them.
> > > >
> > > > > 39. Rolling upgrade of consumers: Do we support the upgrade from
> any
> > > old
> > > > > version to new one?
> > > >
> > > > We will support upgrading from the consumer protocol version 3,
> > > > introduced in KIP-792. KIP-792 is not implemented yet so the earliest
> > > > version is unknown at the moment. This is explained in the migration
> > > > plan chapter.
> > > >
> > > > Thanks again for your feedback, Jun. I will update the KIP based on
> it
> > > > next week.
> > > >
> > > > Best,
> > > > David
> > > >
> > > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <ju...@confluent.io.invalid>
> wrote:
> > > > >
> > > > > Hi, David,
> > > > >
> > > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to
> be
> > > fewer
> > > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > > comments below.
> > > > >
> > > > > 30. ConsumerGroupHeartbeatRequest
> > > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > > changing
> > > > > of the partition assignor in the consumers?
> > > > > 30.2 For each field, could you explain whether it's required in
> every
> > > > > request or the scenarios when it needs to be filled? For example,
> it's
> > > not
> > > > > clear to me when TopicPartitions needs to be filled.
> > > > >
> > > > > 31. In the current consumer protocol, the rack affinity between the
> > > client
> > > > > and the broker is only considered during fetching, but not during
> > > assigning
> > > > > partitions to consumers. Sometimes, once the assignment is made,
> there
> > > is
> > > > > no opportunity for read affinity because no replicas of assigned
> > > partitions
> > > > > are close to the member. I am wondering if we should use this
> > > opportunity
> > > > > to address this by including rack in GroupMember.
> > > > >
> > > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > > coordinator is. By moving the event loop model, it seems that we
> could
> > > add
> > > > > a metric that tracks the fraction of the time the event loop is
> doing
> > > the
> > > > > actual work.
> > > > >
> > > > > 33. Could we add a section on coordinator failover handling? For
> > > example,
> > > > > does it need to trigger the check if any group with the wildcard
> > > > > subscription now has a new matching topic?
> > > > >
> > > > > 34. ConsumerGroupMetadataValue,
> ConsumerGroupPartitionMetadataValue,
> > > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > > field
> > > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > > reflect
> > > > > the latest group epoch? What about the one in
> > > > > ConsumerGroupPartitionMetadataValue and
> > > ConsumerGroupMemberMetadataValue?
> > > > >
> > > > > 35. "the group coordinator will ensure that the following
> invariants
> > > are
> > > > > met: ... All members exists." It's possible for a member not to
> get any
> > > > > assigned partitions, right?
> > > > >
> > > > > 36. "He can rejoins the group with a member epoch equals to 0":
> When
> > > would
> > > > > a consumer rejoin and what member id would be used?
> > > > >
> > > > > 37. "Instead, power users will have the ability to trigger a
> > > reassignment
> > > > > by either providing a non-zero reason or by updating the assignor
> > > > > metadata." Hmm, this seems to be conflicting with the deprecation
> of
> > > > > Consumer#enforeRebalance.
> > > > >
> > > > > 38. The reassignment examples are nice. But the section seems to
> have
> > > > > multiple typos.
> > > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > > epoch=3,
> > > > > partitions=[foo-1] too early.
> > > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > > partitions=[foo-0].
> > > > >
> > > > > 39. Rolling upgrade of consumers: Do we support the upgrade from
> any
> > > old
> > > > > version to new one?
> > > > >
> > > > > Thanks,
> > > > >
> > > > > Jun
> > > > >
> > > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > > <dj...@confluent.io.invalid>
> > > > > wrote:
> > > > >
> > > > > > Hi all,
> > > > > >
> > > > > > The KIP states that we will re-implement the coordinator in
> Java. I
> > > > > > discussed this offline with a few folks and folks are concerned
> that
> > > > > > we could introduce many regressions in the old protocol if we do
> so.
> > > > > > Therefore, I am going to remove this statement from the KIP. It
> is an
> > > > > > implementation detail after all so it does not have to be
> decided at
> > > > > > this stage. We will likely start by trying to refactor the
> current
> > > > > > implementation as a first step.
> > > > > >
> > > > > > Cheers,
> > > > > > David
> > > > > >
> > > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <djacot@confluent.io
> >
> > > wrote:
> > > > > > >
> > > > > > > Hi Luke,
> > > > > > >
> > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > reconciling,
> > > > > > stable,
> > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > >
> > > > > > > This sentence does not refer to those group states but rather
> to a
> > > > > > > state machine replication (SMR). This refers to the entire
> state of
> > > > > > > group coordinator which is replicated via the log layer. I will
> > > > > > > clarify this in the KIP.
> > > > > > >
> > > > > > > > 1.2. What do you mean "each state machine is modelled as an
> event
> > > > > > loop"?
> > > > > > >
> > > > > > > The idea is to follow a model similar to the new quorum
> > > controller. We
> > > > > > > will have N threads to process events. Each __consumer_offsets
> > > > > > > partition is assigned to a unique thread and all the events
> (e.g.
> > > > > > > requests, callbacks, etc.) are processed by this thread. This
> > > simplify
> > > > > > > concurrency and will enable us to do simulation testing for the
> > > group
> > > > > > > coordinator.
> > > > > > >
> > > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > > partitions?
> > > > > > > > Not a state machine "per consumer group" owned by a group
> > > coordinator?
> > > > > > For
> > > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > > both
> > > > > > exist in
> > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it,
> or
> > > 2?
> > > > > > >
> > > > > > > See 1.1. The confusion comes from there, I think.
> > > > > > >
> > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number
> of
> > > threads
> > > > > > used
> > > > > > > > to run the state machines. But I'm wondering if the purpose
> of
> > > the
> > > > > > threads
> > > > > > > > is only to keep the state of each consumer group (or
> > > > > > *__consumer_offsets*
> > > > > > > > partitions?), and no heavy computation, why should we need
> > > > > > multi-threads
> > > > > > > > here?
> > > > > > >
> > > > > > > See 1.2. The idea is to have an ability to shard the
> processing as
> > > the
> > > > > > > computation could be heavy.
> > > > > > >
> > > > > > > > 2.1. The consumer session timeout, why does the default
> session
> > > > > > timeout not
> > > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > > session
> > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > >
> > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> detect
> > > client
> > > > > > > > failures when using the consumer group protocol.
> > > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > > session
> > > > > > timeout.
> > > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > > session
> > > > > > timeout.
> > > > > > >
> > > > > > > This is indeed a mistake. The default session timeout should
> be 45s
> > > > > > > (the current default).
> > > > > > >
> > > > > > > > 2.2. The default server side assignor are [range, uniform],
> > > which means
> > > > > > > > we'll default to "range" assignor. I'd like to know why not
> > > uniform
> > > > > > one? I
> > > > > > > > thought usually users will choose uniform assignor (former
> sticky
> > > > > > assinor)
> > > > > > > > for better evenly distribution. Any other reason we choose
> range
> > > > > > assignor
> > > > > > > > as default?
> > > > > > > > group.consumer.assignors List range, uniform The server side
> > > assignors.
> > > > > > >
> > > > > > > The order on the server side has no influence because the
> client
> > > must
> > > > > > > chose the selector that he wants to use. There is no default
> in the
> > > > > > > current proposal. If the assignor is not specified by the
> client,
> > > the
> > > > > > > request is rejected. The default client value for
> > > > > > > `group.remote.assignor` is `uniform` though.
> > > > > > >
> > > > > > > Thanks for your very good comments, Luke. I hope that my
> answers
> > > help
> > > > > > > to clarify things. I will update the KIP as well based on your
> > > > > > > feedback.
> > > > > > >
> > > > > > > Cheers,
> > > > > > > David
> > > > > > >
> > > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com>
> > > wrote:
> > > > > > > >
> > > > > > > > Hi David,
> > > > > > > >
> > > > > > > > Thanks for the update.
> > > > > > > >
> > > > > > > > Some more questions:
> > > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > > The new group coordinator will have a state machine per
> > > > > > > > *__consumer_offsets* partitions, where each state machine is
> > > modelled
> > > > > > as an
> > > > > > > > event loop. Those state machines will be executed in
> > > > > > > > *group.coordinator.threads* threads.
> > > > > > > >
> > > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > > reconciling,
> > > > > > stable,
> > > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > > 1.2. What do you mean "each state machine is modelled as an
> event
> > > > > > loop"?
> > > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > > partitions?
> > > > > > > > Not a state machine "per consumer group" owned by a group
> > > coordinator?
> > > > > > For
> > > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > > both
> > > > > > exist in
> > > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it,
> or
> > > 2?
> > > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number
> of
> > > threads
> > > > > > used
> > > > > > > > to run the state machines. But I'm wondering if the purpose
> of
> > > the
> > > > > > threads
> > > > > > > > is only to keep the state of each consumer group (or
> > > > > > *__consumer_offsets*
> > > > > > > > partitions?), and no heavy computation, why should we need
> > > > > > multi-threads
> > > > > > > > here?
> > > > > > > >
> > > > > > > > 2. For the default value in the new configs:
> > > > > > > > 2.1. The consumer session timeout, why does the default
> session
> > > > > > timeout not
> > > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > > session
> > > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > > >
> > > > > > > > group.consumer.session.timeout.ms int 30s The timeout to
> detect
> > > client
> > > > > > > > failures when using the consumer group protocol.
> > > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > > session
> > > > > > timeout.
> > > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > > session
> > > > > > timeout.
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > 2.2. The default server side assignor are [range, uniform],
> > > which means
> > > > > > > > we'll default to "range" assignor. I'd like to know why not
> > > uniform
> > > > > > one? I
> > > > > > > > thought usually users will choose uniform assignor (former
> sticky
> > > > > > assinor)
> > > > > > > > for better evenly distribution. Any other reason we choose
> range
> > > > > > assignor
> > > > > > > > as default?
> > > > > > > > group.consumer.assignors List range, uniform The server side
> > > assignors.
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > Thank you.
> > > > > > > > Luke
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <showuon@gmail.com
> >
> > > wrote:
> > > > > > > >
> > > > > > > > > Hi Sagar,
> > > > > > > > >
> > > > > > > > > I have some thoughts about Kafka Connect integrating with
> > > KIP-848,
> > > > > > but I
> > > > > > > > > think we should have a separate discussion thread for the
> Kafka
> > > > > > Connect
> > > > > > > > > KIP: Integrating Kafka Connect With New Consumer Rebalance
> > > Protocol
> > > > > > [1],
> > > > > > > > > and let this discussion thread focus on consumer rebalance
> > > protocol,
> > > > > > WDYT?
> > > > > > > > >
> > > > > > > > > [1]
> > > > > > > > >
> > > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >
> > > > > > > > > Thank you.
> > > > > > > > > Luke
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > > sagarmeansocean@gmail.com>
> > > > > > wrote:
> > > > > > > > >
> > > > > > > > >> Thank you Guozhang/David for the feedback. Looks like
> there's
> > > > > > agreement on
> > > > > > > > >> using separate APIs for Connect. I would revisit the doc
> and
> > > see
> > > > > > what
> > > > > > > > >> changes are to be made.
> > > > > > > > >>
> > > > > > > > >> Thanks!
> > > > > > > > >> Sagar.
> > > > > > > > >>
> > > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> wrote:
> > > > > > > > >>
> > > > > > > > >> > Hi Sagar,
> > > > > > > > >> >
> > > > > > > > >> > Thanks for the feedback and the document. That's really
> > > helpful. I
> > > > > > > > >> > will take a look at it.
> > > > > > > > >> >
> > > > > > > > >> > Overall, it seems to me that both Connect and the
> Consumer
> > > could
> > > > > > share
> > > > > > > > >> > the same underlying "engine". The main difference is
> that
> > > the
> > > > > > Consumer
> > > > > > > > >> > assigns topic-partitions to members whereas Connect
> assigns
> > > tasks
> > > > > > to
> > > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > > >> > 1) We extend the new proposed APIs to support different
> > > resource
> > > > > > types
> > > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > > >> > 2) We use new dedicated APIs for Connect. The dedicated
> APIs
> > > > > > would be
> > > > > > > > >> > similar to the new ones but different on the
> > > content/resources and
> > > > > > > > >> > they would rely on the same engine on the coordinator
> side.
> > > > > > > > >> >
> > > > > > > > >> > I personally lean towards 2) because I am not a fan of
> > > > > > overcharging
> > > > > > > > >> > APIs to serve different purposes. That being said, I am
> not
> > > > > > opposed to
> > > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > > >> >
> > > > > > > > >> > I think that we can continue to discuss it here for now
> in
> > > order
> > > > > > to
> > > > > > > > >> > ensure that this KIP is compatible with what we will do
> for
> > > > > > Connect in
> > > > > > > > >> > the future.
> > > > > > > > >> >
> > > > > > > > >> > Best,
> > > > > > > > >> > David
> > > > > > > > >> >
> > > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > > djacot@confluent.io>
> > > > > > wrote:
> > > > > > > > >> > >
> > > > > > > > >> > > Hi all,
> > > > > > > > >> > >
> > > > > > > > >> > > I am back from vacation. I will go through and address
> > > your
> > > > > > comments
> > > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > > >> > >
> > > > > > > > >> > > Cheers,
> > > > > > > > >> > > David
> > > > > > > > >> > >
> > > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > > gharris1727@gmail.com
> > > > > > > > >> >
> > > > > > > > >> > wrote:
> > > > > > > > >> > > >
> > > > > > > > >> > > > Hey All!
> > > > > > > > >> > > >
> > > > > > > > >> > > > Thanks for the KIP, it's wonderful to see
> cooperative
> > > > > > rebalancing
> > > > > > > > >> > making it
> > > > > > > > >> > > > down the stack!
> > > > > > > > >> > > >
> > > > > > > > >> > > > I had a few questions:
> > > > > > > > >> > > >
> > > > > > > > >> > > > 1. The 'Rejected Alternatives' section describes how
> > > member
> > > > > > epoch
> > > > > > > > >> > should
> > > > > > > > >> > > > advance in step with the group epoch and assignment
> > > epoch
> > > > > > values. I
> > > > > > > > >> > think
> > > > > > > > >> > > > that this is a good idea for the reasons described
> in
> > > the
> > > > > > KIP. When
> > > > > > > > >> the
> > > > > > > > >> > > > protocol is incrementally assigning partitions to a
> > > worker,
> > > > > > what
> > > > > > > > >> member
> > > > > > > > >> > > > epoch does each incremental assignment use? Are
> member
> > > epochs
> > > > > > > > >> re-used,
> > > > > > > > >> > and
> > > > > > > > >> > > > a single member epoch can correspond to multiple
> > > different
> > > > > > > > >> > (monotonically
> > > > > > > > >> > > > larger) assignments?
> > > > > > > > >> > > >
> > > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the
> group
> > > > > > > > >> coordinator? If
> > > > > > > > >> > > > not, should custom client-side assignor
> implementations
> > > > > > interact
> > > > > > > > >> with
> > > > > > > > >> > the
> > > > > > > > >> > > > Reason field, and how is its common meaning agreed
> > > upon? If
> > > > > > so, what
> > > > > > > > >> > is the
> > > > > > > > >> > > > benefit of a distinct Reason field over including
> such
> > > > > > functionality
> > > > > > > > >> > in the
> > > > > > > > >> > > > opaque metadata?
> > > > > > > > >> > > >
> > > > > > > > >> > > > 3. The following is included in the KIP: "Thanks to
> > > this, the
> > > > > > input
> > > > > > > > >> of
> > > > > > > > >> > the
> > > > > > > > >> > > > client side assignor is entirely driven by the group
> > > > > > coordinator.
> > > > > > > > >> The
> > > > > > > > >> > > > consumer is no longer responsible for maintaining
> any
> > > state
> > > > > > besides
> > > > > > > > >> its
> > > > > > > > >> > > > assigned partitions." Does this mean that the
> > > client-side
> > > > > > assignor
> > > > > > > > >> MAY
> > > > > > > > >> > > > incorporate additional non-Metadata state (such as
> > > partition
> > > > > > > > >> > throughput,
> > > > > > > > >> > > > cpu/memory metrics, config topics, etc), or that
> > > additional
> > > > > > > > >> > non-Metadata
> > > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > > >> > > >
> > > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > > >> > > > for
> > > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > > >> > > > and
> org.apache.kafka.clients.consumer.PartitionAssignor
> > > that
> > > > > > seem to
> > > > > > > > >> > > > overlap significantly. Is it possible for these two
> > > > > > implementations
> > > > > > > > >> to
> > > > > > > > >> > be
> > > > > > > > >> > > > unified? This would serve to promote feature parity
> of
> > > > > > server-side
> > > > > > > > >> and
> > > > > > > > >> > > > client-side assignors, and would also facilitate
> > > operational
> > > > > > > > >> > flexibility in
> > > > > > > > >> > > > certain situations. For example, if a server-side
> > > assignor
> > > > > > has some
> > > > > > > > >> > poor
> > > > > > > > >> > > > behavior and needs a patch, deploying the patched
> > > assignor to
> > > > > > the
> > > > > > > > >> > client
> > > > > > > > >> > > > and switching one consumer group to a client-side
> > > assignor
> > > > > > may be
> > > > > > > > >> > faster
> > > > > > > > >> > > > and less risky than patching all of the brokers.
> With
> > > the
> > > > > > currently
> > > > > > > > >> > > > proposed distinct APIs, a non-trivial
> reimplementation
> > > would
> > > > > > have
> > > > > > > > >> to be
> > > > > > > > >> > > > assembled, and if the two APIs have diverged
> > > significantly,
> > > > > > then it
> > > > > > > > >> is
> > > > > > > > >> > > > possible that a reimplementation would not be
> possible.
> > > > > > > > >> > > >
> > > > > > > > >> > > > --
> > > > > > > > >> > > > Greg Harris
> > > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > > >> > > > github.com/gharris1727
> > > > > > > > >> > > >
> > > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > wrote:
> > > > > > > > >> > > >
> > > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > I created a confluence page to discuss how Connect
> > > would
> > > > > > need to
> > > > > > > > >> > change
> > > > > > > > >> > > > > based on the new rebalance protocol. Here's the
> page:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> >
> > > > > > > > >>
> > > > > >
> > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > It's also pretty longish and I have tried to keep
> a
> > > format
> > > > > > > > >> similar to
> > > > > > > > >> > > > > KIP-848. Let me know what you think. Also, do you
> > > think this
> > > > > > > > >> should
> > > > > > > > >> > be
> > > > > > > > >> > > > > moved to a separate discussion thread or is this
> one
> > > fine?
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > Thanks!
> > > > > > > > >> > > > > Sagar.
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > > sagarmeansocean@gmail.com>
> > > > > > > > >> > wrote:
> > > > > > > > >> > > > >
> > > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > Thank you so much for the doc on Kafka Streams.
> > > Sure, I
> > > > > > would do
> > > > > > > > >> > the
> > > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > Thanks!
> > > > > > > > >> > > > > > Sagar.
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > > > > > > >> wangguoz@gmail.com>
> > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > >> It would be great if you could come back with
> some
> > > > > > analysis on
> > > > > > > > >> > how to
> > > > > > > > >> > > > > >> implement the Connect side integration with
> the new
> > > > > > protocol;
> > > > > > > > >> so
> > > > > > > > >> > far
> > > > > > > > >> > > > > >> besides leveraging on the new "protocol type"
> we
> > > did not
> > > > > > yet
> > > > > > > > >> think
> > > > > > > > >> > > > > through
> > > > > > > > >> > > > > >> the Connect side implementations. For Streams
> > > here's a
> > > > > > draft of
> > > > > > > > >> > > > > >> integration
> > > > > > > > >> > > > > >> plan:
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > >
> > > > > > > > >> >
> > > > > > > > >>
> > > > > >
> > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > > >> > >
> > > > > > > > >> > > > > wrote:
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > >> > Hi David,
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> > Thank you for your response. The reason I
> thought
> > > > > > connect can
> > > > > > > > >> > also fit
> > > > > > > > >> > > > > >> into
> > > > > > > > >> > > > > >> > this new scheme is that even today the
> connect
> > > uses a
> > > > > > > > >> > > > > WorkerCoordinator
> > > > > > > > >> > > > > >> > extending from AbstractCoordinator to empower
> > > > > > rebalances of
> > > > > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets
> the
> > > > > > > > >> protocolType()
> > > > > > > > >> > to
> > > > > > > > >> > > > > >> connect
> > > > > > > > >> > > > > >> > and uses the metadata() method by plumbing
> into
> > > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> > I think the changes to support connect would
> be
> > > > > > similar at a
> > > > > > > > >> > high
> > > > > > > > >> > > > > level
> > > > > > > > >> > > > > >> to
> > > > > > > > >> > > > > >> > the changes in streams mainly because of the
> > > Client
> > > > > > side
> > > > > > > > >> > assignors
> > > > > > > > >> > > > > being
> > > > > > > > >> > > > > >> > used in both. At an implementation level, we
> > > might
> > > > > > need to
> > > > > > > > >> make
> > > > > > > > >> > a lot
> > > > > > > > >> > > > > of
> > > > > > > > >> > > > > >> > changes to get onto this new assignment
> protocol
> > > like
> > > > > > > > >> enhancing
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> > JoinGroup request/response and SyncGroup and
> > > using
> > > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > > >> > > > > >> > API etc again on similar lines to streams (or
> > > there
> > > > > > might be
> > > > > > > > >> > > > > >> deviations). I
> > > > > > > > >> > > > > >> > would try to perform a detailed analysis of
> the
> > > same
> > > > > > and we
> > > > > > > > >> > can have
> > > > > > > > >> > > > > a
> > > > > > > > >> > > > > >> > separate discussion thread for that as that
> would
> > > > > > derail this
> > > > > > > > >> > > > > discussion
> > > > > > > > >> > > > > >> > thread. Let me know if that sounds good to
> you.
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> > Thanks!
> > > > > > > > >> > > > > >> > Sagar.
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> > wrote:
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`.
> > > Sure, I
> > > > > > can
> > > > > > > > >> > mention it.
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> > > 2) The idea is to transition C from his
> current
> > > > > > assignment
> > > > > > > > >> to
> > > > > > > > >> > his
> > > > > > > > >> > > > > >> > > target assignment when he can move to
> epoch 3.
> > > When
> > > > > > that
> > > > > > > > >> > happens,
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > member assignment is updated and persisted
> > > with all
> > > > > > its
> > > > > > > > >> > assigned
> > > > > > > > >> > > > > >> > > partitions even if they are not all revoked
> > > yet. In
> > > > > > other
> > > > > > > > >> > words, the
> > > > > > > > >> > > > > >> > > member assignment becomes the target
> > > assignment.
> > > > > > This is
> > > > > > > > >> > basically
> > > > > > > > >> > > > > an
> > > > > > > > >> > > > > >> > > optimization to avoid having to write all
> the
> > > > > > changes to
> > > > > > > > >> the
> > > > > > > > >> > log.
> > > > > > > > >> > > > > The
> > > > > > > > >> > > > > >> > > examples are based on the persisted state
> so I
> > > > > > understand
> > > > > > > > >> the
> > > > > > > > >> > > > > >> > > confusion. Let me see if I can improve
> this in
> > > the
> > > > > > > > >> > description.
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse the
> > > protocol
> > > > > > with a
> > > > > > > > >> > client side
> > > > > > > > >> > > > > >> > > assignor if it fits in the protocol. The
> > > assignment
> > > > > > is
> > > > > > > > >> about
> > > > > > > > >> > > > > >> > > topicid-partitions + metadata, could
> Connect
> > > fit
> > > > > > into this?
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> > > Best,
> > > > > > > > >> > > > > >> > > David
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > > >> > > > > >> wrote:
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor
> > > observations:
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > 1) In the Assignment Error section in
> Client
> > > Side
> > > > > > mode
> > > > > > > > >> > Assignment
> > > > > > > > >> > > > > >> > > process,
> > > > > > > > >> > > > > >> > > > you mentioned => `In this case, the
> client
> > > side
> > > > > > assignor
> > > > > > > > >> can
> > > > > > > > >> > > > > return
> > > > > > > > >> > > > > >> an
> > > > > > > > >> > > > > >> > > > error to the group coordinator`. In this
> > > case are
> > > > > > you
> > > > > > > > >> > referring to
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > Assignor returning an AssignmentError
> that's
> > > > > > listed down
> > > > > > > > >> > towards
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > end?
> > > > > > > > >> > > > > >> > > > If yes, do you think it would make sense
> to
> > > > > > mention this
> > > > > > > > >> > > > > explicitly
> > > > > > > > >> > > > > >> > here?
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I have a
> > > slight
> > > > > > > > >> confusion,
> > > > > > > > >> > not
> > > > > > > > >> > > > > sure
> > > > > > > > >> > > > > >> if
> > > > > > > > >> > > > > >> > > > others have the same. Consider this step:
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > When B heartbeats, the group coordinator
> > > > > > transitions him
> > > > > > > > >> to
> > > > > > > > >> > epoch
> > > > > > > > >> > > > > 3
> > > > > > > > >> > > > > >> > > because
> > > > > > > > >> > > > > >> > > > B has no partitions to revoke. It
> persists
> > > the
> > > > > > change and
> > > > > > > > >> > reply.
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > > >> > > > > >> > > >       - A
> > > > > > > > >> > > > > >> > > >       - B
> > > > > > > > >> > > > > >> > > >       - C
> > > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0,
> > > foo-1]
> > > > > > > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > > > > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > When C heartbeats, it transitions to
> epoch 3
> > > but
> > > > > > cannot
> > > > > > > > >> get
> > > > > > > > >> > foo-1
> > > > > > > > >> > > > > >> yet.
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > Here,it's mentioned that member C can't
> get
> > > the
> > > > > > foo-1
> > > > > > > > >> > partition
> > > > > > > > >> > > > > yet,
> > > > > > > > >> > > > > >> > but
> > > > > > > > >> > > > > >> > > > based on the description above, it seems
> it
> > > > > > already has
> > > > > > > > >> it.
> > > > > > > > >> > Do you
> > > > > > > > >> > > > > >> > think
> > > > > > > > >> > > > > >> > > it
> > > > > > > > >> > > > > >> > > > would be better to remove it and
> populate it
> > > only
> > > > > > when it
> > > > > > > > >> > actually
> > > > > > > > >> > > > > >> gets
> > > > > > > > >> > > > > >> > > it?
> > > > > > > > >> > > > > >> > > > I see this in a lot of other places, so
> have
> > > I
> > > > > > > > >> understood it
> > > > > > > > >> > > > > >> > incorrectly
> > > > > > > > >> > > > > >> > > ?
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > Regarding connect , it might be out of
> scope
> > > of
> > > > > > this
> > > > > > > > >> > discussion,
> > > > > > > > >> > > > > but
> > > > > > > > >> > > > > >> > from
> > > > > > > > >> > > > > >> > > > what I understood it would probably be
> > > running in
> > > > > > client
> > > > > > > > >> > side
> > > > > > > > >> > > > > >> assignor
> > > > > > > > >> > > > > >> > > mode
> > > > > > > > >> > > > > >> > > > even on the new rebalance protocol as it
> has
> > > its
> > > > > > own
> > > > > > > > >> Custom
> > > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David
> Jacot
> > > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > wrote:
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move
> forward
> > > with
> > > > > > > > >> > specialized API
> > > > > > > > >> > > > > >> > > > > instead of relying on one generic API.
> For
> > > > > > Connect, we
> > > > > > > > >> > can apply
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > exact same pattern and reuse/share the
> core
> > > > > > > > >> > implementation on
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > server side. For the schema registry, I
> > > think
> > > > > > that we
> > > > > > > > >> > should
> > > > > > > > >> > > > > >> consider
> > > > > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > > > > membership/leader
> > > > > > > > >> > election.
> > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > >> > > > > Best,
> > > > > > > > >> > > > > >> > > > > David
> > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael
> > > Juma <
> > > > > > > > >> > ismael@juma.me.uk
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > > >> > > wrote:
> > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern
> vs
> > > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > >> > > > > > document the differences in more
> detail
> > > before
> > > > > > > > >> deciding
> > > > > > > > >> > one
> > > > > > > > >> > > > > way
> > > > > > > > >> > > > > >> or
> > > > > > > > >> > > > > >> > > > > another.
> > > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > > java.util.regex.Pattern,
> > > > > > > > >> they
> > > > > > > > >> > expect
> > > > > > > > >> > > > > >> > their
> > > > > > > > >> > > > > >> > > > > > semantics to be honored. If we are
> doing
> > > > > > something
> > > > > > > > >> > different,
> > > > > > > > >> > > > > >> then
> > > > > > > > >> > > > > >> > we
> > > > > > > > >> > > > > >> > > > > > should consider adding an overload
> with
> > > our own
> > > > > > > > >> Pattern
> > > > > > > > >> > class
> > > > > > > > >> > > > > (I
> > > > > > > > >> > > > > >> > > don't
> > > > > > > > >> > > > > >> > > > > > think we'd want to expose re2j's at
> this
> > > > > > point).
> > > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new
> > > protocol
> > > > > > should
> > > > > > > > >> > > > > integrate
> > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > >> > > > > with
> > > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > > recreation case
> > > > > > > > >> > correctly.
> > > > > > > > >> > > > > That's
> > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > >> > > main
> > > > > > > > >> > > > > >> > > > > > part we need to handle. I agree with
> > > David
> > > > > > that we'd
> > > > > > > > >> > want to
> > > > > > > > >> > > > > add
> > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > >> > > > > ids
> > > > > > > > >> > > > > >> > > > > > to the relevant protocols that don't
> > > have it
> > > > > > yet and
> > > > > > > > >> > that we
> > > > > > > > >> > > > > can
> > > > > > > > >> > > > > >> > > probably
> > > > > > > > >> > > > > >> > > > > > focus on the internals versus adding
> new
> > > APIs
> > > > > > to the
> > > > > > > > >> > Java
> > > > > > > > >> > > > > >> Consumer
> > > > > > > > >> > > > > >> > > > > (unless
> > > > > > > > >> > > > > >> > > > > > we find that adding new APIs is
> required
> > > for
> > > > > > > > >> reasonable
> > > > > > > > >> > > > > >> semantics).
> > > > > > > > >> > > > > >> > > > > > 3. I am still not sure about the
> > > coordinator
> > > > > > storing
> > > > > > > > >> the
> > > > > > > > >> > > > > >> configs.
> > > > > > > > >> > > > > >> > > It's
> > > > > > > > >> > > > > >> > > > > > powerful for configs to be
> centralized
> > > in the
> > > > > > > > >> metadata
> > > > > > > > >> > log for
> > > > > > > > >> > > > > >> > > various
> > > > > > > > >> > > > > >> > > > > > reasons (auditability, visibility,
> > > consistency,
> > > > > > > > >> etc.).
> > > > > > > > >> > > > > >> Similarly, I
> > > > > > > > >> > > > > >> > > am
> > > > > > > > >> > > > > >> > > > > not
> > > > > > > > >> > > > > >> > > > > > sure about automatically deleting
> > > configs in a
> > > > > > way
> > > > > > > > >> that
> > > > > > > > >> > they
> > > > > > > > >> > > > > >> cannot
> > > > > > > > >> > > > > >> > > be
> > > > > > > > >> > > > > >> > > > > > recovered. A good property for modern
> > > systems
> > > > > > is to
> > > > > > > > >> > minimize
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > number
> > > > > > > > >> > > > > >> > > > > of
> > > > > > > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David
> > > Jacot
> > > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > > >> > > > > >> > > > > >
> > > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are
> below:
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > > the last
> > > > > > step
> > > > > > > > >> when
> > > > > > > > >> > > > > clients
> > > > > > > > >> > > > > >> > > flip the
> > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in
> which
> > > we
> > > > > > would
> > > > > > > > >> have a
> > > > > > > > >> > > > > window
> > > > > > > > >> > > > > >> > where
> > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> protocols /
> > > rpcs
> > > > > > are
> > > > > > > > >> used
> > > > > > > > >> > by
> > > > > > > > >> > > > > >> members
> > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > > "mimic"
> > > > > > the old
> > > > > > > > >> > behavior
> > > > > > > > >> > > > > >> while
> > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> need to
> > > > > > present
> > > > > > > > >> about.
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > Noted. I just published a new
> version
> > > of KIP
> > > > > > which
> > > > > > > > >> > includes
> > > > > > > > >> > > > > >> more
> > > > > > > > >> > > > > >> > > > > > > details about this. See the
> "Supporting
> > > > > > Online
> > > > > > > > >> > Consumer
> > > > > > > > >> > > > > Group
> > > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > > >> > > > > >> > > > > > > and the "Compatibility,
> Deprecation,
> > > and
> > > > > > Migration
> > > > > > > > >> > Plan". I
> > > > > > > > >> > > > > >> think
> > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > >> > > > > > > I have to think through a few cases
> > > now but
> > > > > > the
> > > > > > > > >> > overall idea
> > > > > > > > >> > > > > >> and
> > > > > > > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> far as
> > > > > > KIP-516 the
> > > > > > > > >> > topic ids
> > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > >> > > only
> > > > > > > > >> > > > > >> > > > > used
> > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client,
> but
> > > they
> > > > > > are not
> > > > > > > > >> > exposed
> > > > > > > > >> > > > > via
> > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > >> > > > > public
> > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > question is,
> > > > > > first
> > > > > > > > >> > should we
> > > > > > > > >> > > > > >> let
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> names
> > > -> ids
> > > > > > mapping
> > > > > > > > >> > itself
> > > > > > > > >> > > > > to
> > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> RPCs
> > > and the
> > > > > > new
> > > > > > > > >> RPCs
> > > > > > > > >> > with
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > > exposing
> > > > > > the
> > > > > > > > >> > topic ids
> > > > > > > > >> > > > > in
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > subscribe/assign, as
> > > > > > > > >> well
> > > > > > > > >> > as in
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > >> > deletion-and-recreation).
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > a) Assuming that we would include
> > > converting
> > > > > > all
> > > > > > > > >> the
> > > > > > > > >> > offsets
> > > > > > > > >> > > > > >> > > related
> > > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this
> KIP,
> > > the
> > > > > > consumer
> > > > > > > > >> > would be
> > > > > > > > >> > > > > >> able
> > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > >> > > > > > > fully operate with topic ids. That
> > > being
> > > > > > said, it
> > > > > > > > >> > still has
> > > > > > > > >> > > > > to
> > > > > > > > >> > > > > >> > > provide
> > > > > > > > >> > > > > >> > > > > > > the topics names in various APIs so
> > > having a
> > > > > > > > >> mapping
> > > > > > > > >> > in the
> > > > > > > > >> > > > > >> > > consumer
> > > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > > >> > > > > >> > > > > > > b) I don't have a strong opinion on
> > > this.
> > > > > > Here I
> > > > > > > > >> > wonder if
> > > > > > > > >> > > > > >> this
> > > > > > > > >> > > > > >> > > goes
> > > > > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I
> would
> > > rather
> > > > > > focus
> > > > > > > > >> on
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> > internals
> > > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > > separately if
> > > > > > we see
> > > > > > > > >> > value in
> > > > > > > > >> > > > > >> doing
> > > > > > > > >> > > > > >> > > it.
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's point about
> > > using
> > > > > > topic ids
> > > > > > > > >> > in the
> > > > > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I
> think
> > > that
> > > > > > there
> > > > > > > > >> is
> > > > > > > > >> > one
> > > > > > > > >> > > > > >> > advantage
> > > > > > > > >> > > > > >> > > in
> > > > > > > > >> > > > > >> > > > > > > favour of it. The consumer will
> have
> > > the
> > > > > > > > >> opportunity
> > > > > > > > >> > to
> > > > > > > > >> > > > > >> validate
> > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > >> > > > > > > the topics exists before passing
> them
> > > into
> > > > > > the
> > > > > > > > >> group
> > > > > > > > >> > > > > rebalance
> > > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the
> coordinator
> > > will
> > > > > > also
> > > > > > > > >> notice
> > > > > > > > >> > it but
> > > > > > > > >> > > > > >> it
> > > > > > > > >> > > > > >> > > does
> > > > > > > > >> > > > > >> > > > > > > not really have a way to reject an
> > > invalid
> > > > > > topic in
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> response.
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all
> other
> > > minor
> > > > > > > > >> questions
> > > > > > > > >> > > > > except
> > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > > personally
> > > > > > I think
> > > > > > > > >> > it's not
> > > > > > > > >> > > > > >> > > necessary
> > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > > Pattern,
> > > > > > but
> > > > > > > > >> > instead we
> > > > > > > > >> > > > > >> still
> > > > > > > > >> > > > > >> > > use
> > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > > subscription may
> > > > > > > > >> be
> > > > > > > > >> > > > > >> rejected by
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > > very rare
> > > > > > > > >> scenario
> > > > > > > > >> > I felt
> > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > >> > > an
> > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > subscription may
> > > > > > be
> > > > > > > > >> more
> > > > > > > > >> > > > > >> vulnerable
> > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > >> > > > > various
> > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > That could work. I have to look at
> the
> > > > > > differences
> > > > > > > > >> > between
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > two
> > > > > > > > >> > > > > >> > > > > > > engines to better understand the
> > > potential
> > > > > > issues.
> > > > > > > > >> My
> > > > > > > > >> > > > > >> > > understanding is
> > > > > > > > >> > > > > >> > > > > > > that would work for all the basic
> > > regular
> > > > > > > > >> > expressions. The
> > > > > > > > >> > > > > >> > > differences
> > > > > > > > >> > > > > >> > > > > > > between the two are mainly about
> the
> > > various
> > > > > > > > >> character
> > > > > > > > >> > > > > >> classes. I
> > > > > > > > >> > > > > >> > > > > > > wonder what other people think
> about
> > > this.
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > > >> > > > > >> > > > > > > David
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM
> > > Guozhang
> > > > > > Wang <
> > > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > > >> > > > > >> > > >
> > > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on the high
> > > level
> > > > > > there are
> > > > > > > > >> > two meta
> > > > > > > > >> > > > > >> > > points we
> > > > > > > > >> > > > > >> > > > > need
> > > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > > the last
> > > > > > step
> > > > > > > > >> when
> > > > > > > > >> > > > > clients
> > > > > > > > >> > > > > >> > > flip the
> > > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in
> which
> > > we
> > > > > > would
> > > > > > > > >> have a
> > > > > > > > >> > > > > window
> > > > > > > > >> > > > > >> > where
> > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > >> > > > > > > new
> > > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old
> protocols /
> > > rpcs
> > > > > > are
> > > > > > > > >> used
> > > > > > > > >> > by
> > > > > > > > >> > > > > >> members
> > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > > "mimic"
> > > > > > the old
> > > > > > > > >> > behavior
> > > > > > > > >> > > > > >> while
> > > > > > > > >> > > > > >> > > > > using the
> > > > > > > > >> > > > > >> > > > > > > > new protocol is something we
> need to
> > > > > > present
> > > > > > > > >> about.
> > > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So
> far as
> > > > > > KIP-516 the
> > > > > > > > >> > topic ids
> > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > >> > > only
> > > > > > > > >> > > > > >> > > > > used
> > > > > > > > >> > > > > >> > > > > > > as
> > > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client,
> but
> > > they
> > > > > > are not
> > > > > > > > >> > exposed
> > > > > > > > >> > > > > via
> > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > >> > > > > public
> > > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > > question is,
> > > > > > first
> > > > > > > > >> > should we
> > > > > > > > >> > > > > >> let
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > > >> > > > > >> > > > > > > > client to be maintaining the
> names
> > > -> ids
> > > > > > mapping
> > > > > > > > >> > itself
> > > > > > > > >> > > > > to
> > > > > > > > >> > > > > >> > fully
> > > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > > >> > > > > >> > > > > > > > on all the augmented existing
> RPCs
> > > and the
> > > > > > new
> > > > > > > > >> RPCs
> > > > > > > > >> > with
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > >> > > > > ids;
> > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > > exposing
> > > > > > the
> > > > > > > > >> > topic ids
> > > > > > > > >> > > > > in
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > consumer
> > > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > > subscribe/assign, as
> > > > > > > > >> well
> > > > > > > > >> > as in
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > > >> > deletion-and-recreation).
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all
> other
> > > minor
> > > > > > > > >> questions
> > > > > > > > >> > > > > except
> > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > > personally
> > > > > > I think
> > > > > > > > >> > it's not
> > > > > > > > >> > > > > >> > > necessary
> > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > > Pattern,
> > > > > > but
> > > > > > > > >> > instead we
> > > > > > > > >> > > > > >> still
> > > > > > > > >> > > > > >> > > use
> > > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > > subscription may
> > > > > > > > >> be
> > > > > > > > >> > > > > >> rejected by
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > > very rare
> > > > > > > > >> scenario
> > > > > > > > >> > I felt
> > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > >> > > an
> > > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > > subscription may
> > > > > > be
> > > > > > > > >> more
> > > > > > > > >> > > > > >> vulnerable
> > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > >> > > > > various
> > > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM
> > > David Jacot
> > > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let
> me
> > > answer
> > > > > > your
> > > > > > > > >> > questions
> > > > > > > > >> > > > > >> > inline.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's premature to
> > > talk about
> > > > > > > > >> target
> > > > > > > > >> > > > > versions
> > > > > > > > >> > > > > >> for
> > > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > >> > > > > > > > > > removal of the existing group
> > > protocol.
> > > > > > > > >> Unlike
> > > > > > > > >> > KRaft,
> > > > > > > > >> > > > > >> this
> > > > > > > > >> > > > > >> > > > > affects a
> > > > > > > > >> > > > > >> > > > > > > core
> > > > > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > > > > deprecation/removal
> > > > > > > > >> > will be
> > > > > > > > >> > > > > >> > heavily
> > > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > >> > > > > >> > > > > > > > > > how quickly applications
> migrate
> > > to
> > > > > > the new
> > > > > > > > >> > protocol.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will
> remove it.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to
> > > release
> > > > > > this in
> > > > > > > > >> > 4.x, but
> > > > > > > > >> > > > > it
> > > > > > > > >> > > > > >> > > wasn't
> > > > > > > > >> > > > > >> > > > > made
> > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > >> > > > > >> > > > > > > > > > why. If we added that as a
> way to
> > > > > > estimate
> > > > > > > > >> when
> > > > > > > > >> > we'd
> > > > > > > > >> > > > > >> > > deprecate
> > > > > > > > >> > > > > >> > > > > and
> > > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I also
> > > suggest
> > > > > > removing
> > > > > > > > >> > this part.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> > > > > > explained, I
> > > > > > > > >> plan
> > > > > > > > >> > to
> > > > > > > > >> > > > > >> rewrite
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > group
> > > > > > > > >> > > > > >> > > > > > > > > coordinator in Java while we
> > > implement
> > > > > > the new
> > > > > > > > >> > protocol.
> > > > > > > > >> > > > > >> This
> > > > > > > > >> > > > > >> > > means
> > > > > > > > >> > > > > >> > > > > > > > > that the internals will be
> slightly
> > > > > > different
> > > > > > > > >> > (e.g.
> > > > > > > > >> > > > > >> threading
> > > > > > > > >> > > > > >> > > > > model).
> > > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten
> the
> > > > > > switch from
> > > > > > > > >> > the old
> > > > > > > > >> > > > > >> group
> > > > > > > > >> > > > > >> > > > > > > > > coordinator to the new group
> > > coordinator
> > > > > > to a
> > > > > > > > >> > major
> > > > > > > > >> > > > > >> release.
> > > > > > > > >> > > > > >> > > The
> > > > > > > > >> > > > > >> > > > > > > > > alternative would be to use a
> flag
> > > to do
> > > > > > the
> > > > > > > > >> > switch
> > > > > > > > >> > > > > >> instead
> > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > >> > > > > relying
> > > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the
> > > details of
> > > > > > the
> > > > > > > > >> > migration
> > > > > > > > >> > > > > >> story.
> > > > > > > > >> > > > > >> > > It
> > > > > > > > >> > > > > >> > > > > sounds
> > > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > > >> > > > > >> > > > > > > > > > we're saying we will support
> > > online
> > > > > > > > >> migrations.
> > > > > > > > >> > Is
> > > > > > > > >> > > > > that
> > > > > > > > >> > > > > >> > > correct?
> > > > > > > > >> > > > > >> > > > > We
> > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > >> > > > > > > > > > explain this in detail. It
> could
> > > also
> > > > > > be done
> > > > > > > > >> > as a
> > > > > > > > >> > > > > >> separate
> > > > > > > > >> > > > > >> > > KIP,
> > > > > > > > >> > > > > >> > > > > if
> > > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Yes, we will support online
> > > migrations
> > > > > > for the
> > > > > > > > >> > group.
> > > > > > > > >> > > > > That
> > > > > > > > >> > > > > >> > > means
> > > > > > > > >> > > > > >> > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > a group using the old protocol
> > > will be
> > > > > > able to
> > > > > > > > >> > switch to
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain how that
> > > will work
> > > > > > > > >> though.
> > > > > > > > >> > It is
> > > > > > > > >> > > > > >> > > basically a
> > > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be
> upgraded or
> > > > > > rolled to a
> > > > > > > > >> > software
> > > > > > > > >> > > > > >> > > supporting
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > new group coordinator. Both the
> > > old and
> > > > > > the new
> > > > > > > > >> > > > > >> coordinator
> > > > > > > > >> > > > > >> > > will
> > > > > > > > >> > > > > >> > > > > > > > > support the old protocol and
> rely
> > > on the
> > > > > > same
> > > > > > > > >> > persisted
> > > > > > > > >> > > > > >> > > metadata so
> > > > > > > > >> > > > > >> > > > > > > > > they can work together. This
> point
> > > is an
> > > > > > > > >> offline
> > > > > > > > >> > > > > >> migration.
> > > > > > > > >> > > > > >> > We
> > > > > > > > >> > > > > >> > > > > cannot
> > > > > > > > >> > > > > >> > > > > > > > > do this one live because it
> would
> > > require
> > > > > > > > >> > shutting down
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > current
> > > > > > > > >> > > > > >> > > > > > > > > coordinator and starting up the
> > > new one
> > > > > > and
> > > > > > > > >> that
> > > > > > > > >> > would
> > > > > > > > >> > > > > >> cause
> > > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > > version/IBP
> > > > > > must be
> > > > > > > > >> > upgraded
> > > > > > > > >> > > > > to
> > > > > > > > >> > > > > >> X
> > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > >> > > > > order
> > > > > > > > >> > > > > >> > > > > > > > > to enable the new protocol.
> This
> > > cannot
> > > > > > be done
> > > > > > > > >> > before
> > > > > > > > >> > > > > 1)
> > > > > > > > >> > > > > >> is
> > > > > > > > >> > > > > >> > > > > > > > > terminated because the old
> > > coordinator
> > > > > > doesn't
> > > > > > > > >> > support
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > new
> > > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be
> upgraded
> > > to a
> > > > > > version
> > > > > > > > >> > > > > supporting
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > online
> > > > > > > > >> > > > > >> > > > > > > > > migration (must have KIP-792).
> If
> > > the
> > > > > > consumer
> > > > > > > > >> is
> > > > > > > > >> > > > > already
> > > > > > > > >> > > > > >> > > there.
> > > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at this
> point.
> > > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled
> > > with the
> > > > > > > > >> feature
> > > > > > > > >> > flag
> > > > > > > > >> > > > > >> turned
> > > > > > > > >> > > > > >> > > on.
> > > > > > > > >> > > > > >> > > > > The
> > > > > > > > >> > > > > >> > > > > > > > > consumer group is automatically
> > > > > > converted when
> > > > > > > > >> > the first
> > > > > > > > >> > > > > >> > > consumer
> > > > > > > > >> > > > > >> > > > > > > > > using the new protocol joins
> the
> > > group.
> > > > > > While
> > > > > > > > >> the
> > > > > > > > >> > > > > members
> > > > > > > > >> > > > > >> > > using the
> > > > > > > > >> > > > > >> > > > > > > > > old protocol are being
> upgraded,
> > > the old
> > > > > > > > >> protocol
> > > > > > > > >> > is
> > > > > > > > >> > > > > >> proxied
> > > > > > > > >> > > > > >> > > into
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of this in
> the
> > > KIP.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are
> > > pushing the
> > > > > > pattern
> > > > > > > > >> > > > > >> subscriptions
> > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > >> > > > > >> > > > > > > > > > but it seems like there
> could be
> > > some
> > > > > > tricky
> > > > > > > > >> > > > > >> compatibility
> > > > > > > > >> > > > > >> > > > > issues.
> > > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for users to
> > > detect
> > > > > > that
> > > > > > > > >> they
> > > > > > > > >> > need to
> > > > > > > > >> > > > > >> > update
> > > > > > > > >> > > > > >> > > > > their
> > > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > > >> > > > > >> > > > > > > > > > before switching to the new
> > > protocol?
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit more
> > > optimistic
> > > > > > than
> > > > > > > > >> you
> > > > > > > > >> > on this
> > > > > > > > >> > > > > >> > > point. I
> > > > > > > > >> > > > > >> > > > > > > > > believe that the majority of
> the
> > > cases
> > > > > > are
> > > > > > > > >> simple
> > > > > > > > >> > > > > regexes
> > > > > > > > >> > > > > >> > which
> > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > >> > > > > > > > > work with the new engine. The
> > > > > > coordinator will
> > > > > > > > >> > verify
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > regex
> > > > > > > > >> > > > > >> > > > > anyway
> > > > > > > > >> > > > > >> > > > > > > > > and reject the consumer if the
> > > regex is
> > > > > > not
> > > > > > > > >> valid.
> > > > > > > > >> > > > > Coming
> > > > > > > > >> > > > > >> > back
> > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > migration path, in the worst
> case,
> > > the
> > > > > > first
> > > > > > > > >> > upgraded
> > > > > > > > >> > > > > >> > consumer
> > > > > > > > >> > > > > >> > > > > joining
> > > > > > > > >> > > > > >> > > > > > > > > the group will be rejected.
> This
> > > should
> > > > > > be used
> > > > > > > > >> > as the
> > > > > > > > >> > > > > >> last
> > > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > One way for customers to
> validate
> > > their
> > > > > > regex
> > > > > > > > >> > before
> > > > > > > > >> > > > > >> > upgrading
> > > > > > > > >> > > > > >> > > > > their
> > > > > > > > >> > > > > >> > > > > > > > > prod would be to test them with
> > > another
> > > > > > group.
> > > > > > > > >> For
> > > > > > > > >> > > > > >> instance,
> > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod
> > > environment.
> > > > > > > > >> Another
> > > > > > > > >> > way
> > > > > > > > >> > > > > >> would be
> > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > >> > > > > > > > > extend the consumer-group tool
> to
> > > > > > provide a
> > > > > > > > >> regex
> > > > > > > > >> > > > > >> validation
> > > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be
> enough in
> > > your
> > > > > > > > >> opinion?
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last
> question,
> > > will
> > > > > > the
> > > > > > > > >> Java
> > > > > > > > >> > client
> > > > > > > > >> > > > > >> allow
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > > >> > > > > >> > > > > > > > > > stick with the current regex
> > > engine for
> > > > > > > > >> > compatibility
> > > > > > > > >> > > > > >> > > reasons?
> > > > > > > > >> > > > > >> > > > > For
> > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > >> > > > > >> > > > > > > > > > it may be handy to keep using
> > > client
> > > > > > based
> > > > > > > > >> > regex at
> > > > > > > > >> > > > > >> first
> > > > > > > > >> > > > > >> > to
> > > > > > > > >> > > > > >> > > keep
> > > > > > > > >> > > > > >> > > > > > > > > > migrations simple and then
> > > migrate to
> > > > > > server
> > > > > > > > >> > based
> > > > > > > > >> > > > > >> regexes
> > > > > > > > >> > > > > >> > > as a
> > > > > > > > >> > > > > >> > > > > > > second
> > > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > I understand your point but I
> am
> > > > > > concerned that
> > > > > > > > >> > this
> > > > > > > > >> > > > > would
> > > > > > > > >> > > > > >> > > allow
> > > > > > > > >> > > > > >> > > > > users
> > > > > > > > >> > > > > >> > > > > > > > > to actually stay in this mode.
> That
> > > > > > would go
> > > > > > > > >> > against our
> > > > > > > > >> > > > > >> goal
> > > > > > > > >> > > > > >> > > of
> > > > > > > > >> > > > > >> > > > > > > > > simplifying the client because
> we
> > > would
> > > > > > have to
> > > > > > > > >> > continue
> > > > > > > > >> > > > > >> > > monitoring
> > > > > > > > >> > > > > >> > > > > > > > > the metadata on the client
> side. I
> > > would
> > > > > > rather
> > > > > > > > >> > not do
> > > > > > > > >> > > > > >> this.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> > > > > > coordinator
> > > > > > > > >> will
> > > > > > > > >> > be
> > > > > > > > >> > > > > >> > > responsible for
> > > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > > >> > > > > >> > > > > > > > > > the configurations and that
> the
> > > > > > > > >> configurations
> > > > > > > > >> > will be
> > > > > > > > >> > > > > >> > > deleted
> > > > > > > > >> > > > > >> > > > > when
> > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a
> > > transition to
> > > > > > DEAD
> > > > > > > > >> > trigger
> > > > > > > > >> > > > > >> > deletion
> > > > > > > > >> > > > > >> > > of
> > > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > That's right. The
> configurations
> > > will be
> > > > > > > > >> deleted
> > > > > > > > >> > when
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > group is
> > > > > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to store
> the
> > > > > > configs in
> > > > > > > > >> the
> > > > > > > > >> > group
> > > > > > > > >> > > > > >> > > coordinator
> > > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > > >> > > > > >> > > > > > > > > > harder to list all cluster
> > > configs and
> > > > > > their
> > > > > > > > >> > values?
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > > > > configurations are
> > > > > > > > >> > overrides
> > > > > > > > >> > > > > >> of
> > > > > > > > >> > > > > >> > > cluster
> > > > > > > > >> > > > > >> > > > > > > > > configs. If you want to know
> all
> > > the
> > > > > > overrides
> > > > > > > > >> > though,
> > > > > > > > >> > > > > you
> > > > > > > > >> > > > > >> > > would
> > > > > > > > >> > > > > >> > > > > have
> > > > > > > > >> > > > > >> > > > > > > > > to ask all the group
> coordinators.
> > > You
> > > > > > cannot
> > > > > > > > >> > rely on
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > metadata
> > > > > > > > >> > > > > >> > > > > log
> > > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone
> configure a
> > > group
> > > > > > before
> > > > > > > > >> > starting
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > > >> > > > > >> > > > > > > > > > we considered allowing the
> > > explicit
> > > > > > creation
> > > > > > > > >> of
> > > > > > > > >> > > > > groups?
> > > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > configs could be decoupled
> from
> > > the
> > > > > > group
> > > > > > > > >> > lifecycle.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > > automatically
> > > > > > created in
> > > > > > > > >> > this
> > > > > > > > >> > > > > case.
> > > > > > > > >> > > > > >> > > However,
> > > > > > > > >> > > > > >> > > > > > > > > the configs will be lost after
> the
> > > > > > retention
> > > > > > > > >> > period of
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > group
> > > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 9. Will the
> Consumer.subscribe
> > > method
> > > > > > for the
> > > > > > > > >> > Java
> > > > > > > > >> > > > > >> client
> > > > > > > > >> > > > > >> > > still
> > > > > > > > >> > > > > >> > > > > take
> > > > > > > > >> > > > > >> > > > > > > a
> > > > > > > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of
> do
> > > we
> > > > > > have to
> > > > > > > > >> > introduce
> > > > > > > > >> > > > > an
> > > > > > > > >> > > > > >> > > overload?
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > That's a very group question. I
> > > forgot
> > > > > > about
> > > > > > > > >> that
> > > > > > > > >> > one.
> > > > > > > > >> > > > > As
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is
> not
> > > fully
> > > > > > > > >> compatible
> > > > > > > > >> > with
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > engine
> > > > > > > > >> > > > > >> > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might be
> better
> > > to
> > > > > > deprecate
> > > > > > > > >> > it and
> > > > > > > > >> > > > > >> use an
> > > > > > > > >> > > > > >> > > > > overload
> > > > > > > > >> > > > > >> > > > > > > > > which takes a string. We would
> > > rely on
> > > > > > the
> > > > > > > > >> server
> > > > > > > > >> > side
> > > > > > > > >> > > > > >> > > validation.
> > > > > > > > >> > > > > >> > > > > > > > > During the migration, I think
> that
> > > we
> > > > > > could
> > > > > > > > >> still
> > > > > > > > >> > try to
> > > > > > > > >> > > > > >> > > toString
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > regex and use it. That should
> > > work, I
> > > > > > think, in
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> majority
> > > > > > > > >> > > > > >> > > of the
> > > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine
> that we
> > > > > > should be
> > > > > > > > >> > clearer
> > > > > > > > >> > > > > about
> > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > >> > > > > reason
> > > > > > > > >> > > > > >> > > > > > > to
> > > > > > > > >> > > > > >> > > > > > > > > > switch to
> IBP/metadata.version
> > > from the
> > > > > > > > >> feature
> > > > > > > > >> > flag.
> > > > > > > > >> > > > > >> Maybe
> > > > > > > > >> > > > > >> > > we
> > > > > > > > >> > > > > >> > > > > mean
> > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > > >> > > > > >> > > > > > > > > > can switch the default for
> the
> > > feature
> > > > > > flag
> > > > > > > > >> to
> > > > > > > > >> > true
> > > > > > > > >> > > > > >> based
> > > > > > > > >> > > > > >> > on
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > > > > metadata.version once we
> want to
> > > make
> > > > > > it the
> > > > > > > > >> > default.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > My plan was to use that feature
> > > flag
> > > > > > mainly
> > > > > > > > >> > during the
> > > > > > > > >> > > > > >> > > development
> > > > > > > > >> > > > > >> > > > > > > > > phase. I should not have
> mentioned
> > > it, I
> > > > > > think,
> > > > > > > > >> > because
> > > > > > > > >> > > > > we
> > > > > > > > >> > > > > >> > > could
> > > > > > > > >> > > > > >> > > > > use
> > > > > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs
> > > don't
> > > > > > mention
> > > > > > > > >> the
> > > > > > > > >> > > > > required
> > > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > > >> > > > > >> > > > > > > would
> > > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > > >> > > > > >> > > > > > > > > > good to add that for
> consistency.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > > >> > > > > requires
> > > > > > > > >> > > > > >> > > "Read
> > > > > > > > >> > > > > >> > > > > Group"
> > > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > > >> > > > > >> > > > > > > > > > though it seems to do more
> than
> > > > > > reading.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the
> current
> > > > > > protocol works
> > > > > > > > >> > though.
> > > > > > > > >> > > > > We
> > > > > > > > >> > > > > >> > only
> > > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to join a
> > > group. We
> > > > > > could
> > > > > > > > >> > consider
> > > > > > > > >> > > > > >> > > changing
> > > > > > > > >> > > > > >> > > > > this
> > > > > > > > >> > > > > >> > > > > > > > > but I am not sure that it is
> worth
> > > it.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic recreation
> > > handled by
> > > > > > the
> > > > > > > > >> > consumer
> > > > > > > > >> > > > > with
> > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be good to
> > > have a
> > > > > > section
> > > > > > > > >> on
> > > > > > > > >> > this.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > > perspective, the
> > > > > > new
> > > > > > > > >> topic
> > > > > > > > >> > will
> > > > > > > > >> > > > > >> have a
> > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > >> > > > > > > > > topic id so it will treat it
> like a
> > > > > > topic with
> > > > > > > > >> a
> > > > > > > > >> > > > > different
> > > > > > > > >> > > > > >> > > name.
> > > > > > > > >> > > > > >> > > > > The
> > > > > > > > >> > > > > >> > > > > > > > > only issue is that the
> fetch/commit
> > > > > > offsets
> > > > > > > > >> APIs
> > > > > > > > >> > do not
> > > > > > > > >> > > > > >> > support
> > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would
> reuse the
> > > > > > offsets
> > > > > > > > >> based
> > > > > > > > >> > on the
> > > > > > > > >> > > > > >> > same.
> > > > > > > > >> > > > > >> > > I
> > > > > > > > >> > > > > >> > > > > think
> > > > > > > > >> > > > > >> > > > > > > > > that we should update those
> APIs
> > > as well
> > > > > > in
> > > > > > > > >> order
> > > > > > > > >> > to be
> > > > > > > > >> > > > > >> > > consistent
> > > > > > > > >> > > > > >> > > > > end
> > > > > > > > >> > > > > >> > > > > > > > > to end. That would strengthen
> the
> > > > > > semantics of
> > > > > > > > >> the
> > > > > > > > >> > > > > >> consumer.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will
> > > write the
> > > > > > new
> > > > > > > > >> > coordinator
> > > > > > > > >> > > > > >> in
> > > > > > > > >> > > > > >> > > Java.
> > > > > > > > >> > > > > >> > > > > Even
> > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > >> > > > > >> > > > > > > > > > this is an implementation
> > > detail, do
> > > > > > we plan
> > > > > > > > >> to
> > > > > > > > >> > have a
> > > > > > > > >> > > > > >> new
> > > > > > > > >> > > > > >> > > gradle
> > > > > > > > >> > > > > >> > > > > > > module
> > > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability
> > > goal when
> > > > > > it
> > > > > > > > >> comes
> > > > > > > > >> > to how
> > > > > > > > >> > > > > >> many
> > > > > > > > >> > > > > >> > > > > members
> > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at the
> > > moment. The
> > > > > > > > >> protocol
> > > > > > > > >> > should
> > > > > > > > >> > > > > >> > > support
> > > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > > >> > > > > >> > > > > > > > > of members per group. We will
> > > measure
> > > > > > this when
> > > > > > > > >> > we have
> > > > > > > > >> > > > > a
> > > > > > > > >> > > > > >> > first
> > > > > > > > >> > > > > >> > > > > > > > > implementation. Note that we
> might
> > > have
> > > > > > other
> > > > > > > > >> > > > > bottlenecks
> > > > > > > > >> > > > > >> > down
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> > > > > > SubscribedTopidIds
> > > > > > > > >> > instead
> > > > > > > > >> > > > > of
> > > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > > >> > > > > >> Is
> > > > > > > > >> > > > > >> > the
> > > > > > > > >> > > > > >> > > > > idea
> > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > > since we have to resolve the
> > > regex on
> > > > > > the
> > > > > > > > >> > server, we
> > > > > > > > >> > > > > >> can do
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > same
> > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > >> > > > > >> > > > > > > > > > the topic name? The
> difference
> > > is that
> > > > > > > > >> sending
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> regex is
> > > > > > > > >> > > > > >> > > more
> > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > >> > > > > >> > > > > > > > > > whereas sending the topic
> names
> > > is less
> > > > > > > > >> > efficient.
> > > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to
> handle
> > > if we
> > > > > > have
> > > > > > > > >> topic
> > > > > > > > >> > ids.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > The idea was to consolidate the
> > > metadata
> > > > > > lookup
> > > > > > > > >> > on the
> > > > > > > > >> > > > > >> server
> > > > > > > > >> > > > > >> > > for
> > > > > > > > >> > > > > >> > > > > both
> > > > > > > > >> > > > > >> > > > > > > > > paths but I do agree with your
> > > point. As
> > > > > > a
> > > > > > > > >> second
> > > > > > > > >> > > > > though,
> > > > > > > > >> > > > > >> > using
> > > > > > > > >> > > > > >> > > > > topic
> > > > > > > > >> > > > > >> > > > > > > > > ids may be better here for the
> > > delete and
> > > > > > > > >> > recreation
> > > > > > > > >> > > > > case.
> > > > > > > > >> > > > > >> > > Also, I
> > > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow
> users to
> > > > > > subscribe
> > > > > > > > >> with
> > > > > > > > >> > topic
> > > > > > > > >> > > > > >> ids
> > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > > > future because that is the only
> > > way to be
> > > > > > > > >> really
> > > > > > > > >> > robust
> > > > > > > > >> > > > > to
> > > > > > > > >> > > > > >> > > topic
> > > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM
> > > David
> > > > > > Jacot <
> > > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > Thanks for your comments.
> Please
> > > find
> > > > > > my
> > > > > > > > >> answers
> > > > > > > > >> > > > > below.
> > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol
> relies on
> > > > > > topic IDs
> > > > > > > > >> > with the
> > > > > > > > >> > > > > >> > > exception
> > > > > > > > >> > > > > >> > > > > of the
> > > > > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > > >> > > > > >> I
> > > > > > > > >> > > > > >> > am
> > > > > > > > >> > > > > >> > > not
> > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > >> > > > > >> > > > > > > > > > if using topic names is the
> > > right call
> > > > > > here.
> > > > > > > > >> I
> > > > > > > > >> > need to
> > > > > > > > >> > > > > >> > think
> > > > > > > > >> > > > > >> > > > > about it
> > > > > > > > >> > > > > >> > > > > > > > > > a little more. Obviously, the
> > > KIP does
> > > > > > not
> > > > > > > > >> > change the
> > > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic
> IDs.
> > > This
> > > > > > may be
> > > > > > > > >> > something
> > > > > > > > >> > > > > >> that
> > > > > > > > >> > > > > >> > we
> > > > > > > > >> > > > > >> > > > > should
> > > > > > > > >> > > > > >> > > > > > > > > > include though as it would
> give
> > > better
> > > > > > > > >> overall
> > > > > > > > >> > > > > >> guarantee in
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > > >> > > > > >> > > > > > > > > > - You're right. I think that
> I
> > > should
> > > > > > not
> > > > > > > > >> have
> > > > > > > > >> > > > > mentioned
> > > > > > > > >> > > > > >> > this
> > > > > > > > >> > > > > >> > > > > flag at
> > > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it. We can
> > > use an
> > > > > > internal
> > > > > > > > >> > > > > >> configuration
> > > > > > > > >> > > > > >> > > while
> > > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> > > > > > supported. The
> > > > > > > > >> > change is
> > > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > > >> > > > > >> > > > > > > > > > only requirement is that the
> > > cluster
> > > > > > uses
> > > > > > > > >> topic
> > > > > > > > >> > IDs.
> > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53
> PM
> > > > > > Guozhang
> > > > > > > > >> Wang <
> > > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the feedback.
> Here
> > > are
> > > > > > some
> > > > > > > > >> replies
> > > > > > > > >> > > > > inlined
> > > > > > > > >> > > > > >> > > below:
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at
> 2:53 AM
> > > > > > Ismael Juma
> > > > > > > > >> <
> > > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This
> has
> > > the
> > > > > > > > >> potential
> > > > > > > > >> > to be a
> > > > > > > > >> > > > > >> > great
> > > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > > >> > > > > >> > > > > > > > > > > > initial
> questions/comments:
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's
> premature to
> > > talk
> > > > > > about
> > > > > > > > >> > target
> > > > > > > > >> > > > > >> versions
> > > > > > > > >> > > > > >> > > for
> > > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > > >> > > > > >> > > > > > > > > > > > removal of the existing
> group
> > > > > > protocol.
> > > > > > > > >> > Unlike
> > > > > > > > >> > > > > >> KRaft,
> > > > > > > > >> > > > > >> > > this
> > > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > > > > > > >> > deprecation/removal will
> > > > > > > > >> > > > > >> be
> > > > > > > > >> > > > > >> > > heavily
> > > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > > >> > > > > >> > > > > > > > > > > > how quickly applications
> > > migrate
> > > > > > to the
> > > > > > > > >> new
> > > > > > > > >> > > > > >> protocol.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I
> think
> > > we can
> > > > > > > > >> remove
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> proposed
> > > > > > > > >> > > > > >> > > > > timeline
> > > > > > > > >> > > > > >> > > > > > > in
> > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility,
> Deprecation,
> > > and
> > > > > > Migration
> > > > > > > > >> > Plan` and
> > > > > > > > >> > > > > >> > > instead
> > > > > > > > >> > > > > >> > > > > just
> > > > > > > > >> > > > > >> > > > > > > state
> > > > > > > > >> > > > > >> > > > > > > > > > > that we will decide in the
> > > future
> > > > > > about
> > > > > > > > >> when
> > > > > > > > >> > we
> > > > > > > > >> > > > > would
> > > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we
> intend to
> > > > > > release this
> > > > > > > > >> > in 4.x,
> > > > > > > > >> > > > > >> but
> > > > > > > > >> > > > > >> > it
> > > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > > >> > > > > >> > > > > > > made
> > > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added that as
> a
> > > way to
> > > > > > > > >> estimate
> > > > > > > > >> > when
> > > > > > > > >> > > > > we'd
> > > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > > >> > > > > >> > > > > > > and
> > > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I
> also
> > > suggest
> > > > > > > > >> removing
> > > > > > > > >> > this
> > > > > > > > >> > > > > >> part.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > I think that's not
> specifically
> > > > > > related to
> > > > > > > > >> the
> > > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for
> client
> > > > > > upgrades.
> > > > > > > > >> I.e.
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> > > broker-side
> > > > > > > > >> > > > > >> > > > > > > > > > > implementation may be done
> > > first,
> > > > > > and then
> > > > > > > > >> the
> > > > > > > > >> > > > > client
> > > > > > > > >> > > > > >> > side,
> > > > > > > > >> > > > > >> > > > > and we
> > > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > > >> > > > > >> > > > > > > > > > > only mark it as "released"
> by
> > > the
> > > > > > time
> > > > > > > > >> clients
> > > > > > > > >> > > > > >> > > implementations
> > > > > > > > >> > > > > >> > > > > are
> > > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable the
> > > feature the
> > > > > > > > >> clients
> > > > > > > > >> > need to
> > > > > > > > >> > > > > >> > first
> > > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling
> bounce
> > > and
> > > > > > then set
> > > > > > > > >> > the flag
> > > > > > > > >> > > > > >> > with a
> > > > > > > > >> > > > > >> > > > > second
> > > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel
> it's
> > > > > > better to be
> > > > > > > > >> > released
> > > > > > > > >> > > > > >> in a
> > > > > > > > >> > > > > >> > > major
> > > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out
> the
> > > > > > details of
> > > > > > > > >> the
> > > > > > > > >> > > > > migration
> > > > > > > > >> > > > > >> > > story.
> > > > > > > > >> > > > > >> > > > > It
> > > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we will
> support
> > > online
> > > > > > > > >> > migrations. Is
> > > > > > > > >> > > > > >> that
> > > > > > > > >> > > > > >> > > > > correct?
> > > > > > > > >> > > > > >> > > > > > > We
> > > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > > >> > > > > >> > > > > > > > > > > > explain this in detail.
> It
> > > could
> > > > > > also be
> > > > > > > > >> > done as a
> > > > > > > > >> > > > > >> > > separate
> > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the
> part we
> > > can
> > > > > > be more
> > > > > > > > >> > concrete
> > > > > > > > >> > > > > >> about
> > > > > > > > >> > > > > >> > > for
> > > > > > > > >> > > > > >> > > > > sure
> > > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > > >> > > > > >> > > > > > > > > > > this is related to your
> > > question 2)
> > > > > > above).
> > > > > > > > >> > We will
> > > > > > > > >> > > > > >> work
> > > > > > > > >> > > > > >> > on
> > > > > > > > >> > > > > >> > > > > making
> > > > > > > > >> > > > > >> > > > > > > it
> > > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel as we
> > > solicit
> > > > > > more
> > > > > > > > >> > feedback.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are
> > > pushing
> > > > > > the
> > > > > > > > >> > pattern
> > > > > > > > >> > > > > >> > > subscriptions
> > > > > > > > >> > > > > >> > > > > to
> > > > > > > > >> > > > > >> > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like there
> > > could be
> > > > > > some
> > > > > > > > >> tricky
> > > > > > > > >> > > > > >> > > compatibility
> > > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for
> users to
> > > > > > detect that
> > > > > > > > >> > they
> > > > > > > > >> > > > > need
> > > > > > > > >> > > > > >> to
> > > > > > > > >> > > > > >> > > update
> > > > > > > > >> > > > > >> > > > > > > their
> > > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > > >> > > > > >> > > > > > > > > > > > before switching to the
> new
> > > > > > protocol?
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need some
> > > tooling for
> > > > > > > > >> non-java
> > > > > > > > >> > client
> > > > > > > > >> > > > > >> > users
> > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > >> > > > > sort
> > > > > > > > >> > > > > >> > > > > > > of
> > > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> > > > > > switching to
> > > > > > > > >> the
> > > > > > > > >> > new
> > > > > > > > >> > > > > >> > protocol.
> > > > > > > > >> > > > > >> > > I
> > > > > > > > >> > > > > >> > > > > do not
> > > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top of my
> head
> > > > > > though,
> > > > > > > > >> maybe
> > > > > > > > >> > others
> > > > > > > > >> > > > > >> like
> > > > > > > > >> > > > > >> > > @Matt
> > > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the last
> > > question,
> > > > > > will the
> > > > > > > > >> > Java
> > > > > > > > >> > > > > >> client
> > > > > > > > >> > > > > >> > > allow
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > > >> > > > > >> > > > > > > > > > > > stick with the current
> regex
> > > > > > engine for
> > > > > > > > >> > > > > >> compatibility
> > > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > > >> > > > > >> > > > > > > For
> > > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to keep
> using
> > > > > > client
> > > > > > > > >> based
> > > > > > > > >> > regex
> > > > > > > > >> > > > > at
> > > > > > > > >> > > > > >> > > first to
> > > > > > > > >> > > > > >> > > > > keep
> > > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple and
> then
> > > migrate
> > > > > > to
> > > > > > > > >> > server based
> > > > > > > > >> > > > > >> > > regexes
> > > > > > > > >> > > > > >> > > > > as a
> > > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not thought
> > > about
> > > > > > that for
> > > > > > > > >> > java
> > > > > > > > >> > > > > >> clients,
> > > > > > > > >> > > > > >> > > and
> > > > > > > > >> > > > > >> > > > > we can
> > > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > > compatibility
> > > > > > issues do
> > > > > > > > >> > you have
> > > > > > > > >> > > > > in
> > > > > > > > >> > > > > >> > > mind?
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that the
> group
> > > > > > coordinator
> > > > > > > > >> > will be
> > > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > > >> > > > > >> > > > > > > > > > > > the configurations and
> that
> > > the
> > > > > > > > >> > configurations
> > > > > > > > >> > > > > will
> > > > > > > > >> > > > > >> be
> > > > > > > > >> > > > > >> > > > > deleted
> > > > > > > > >> > > > > >> > > > > > > when
> > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> > > > > > transition to
> > > > > > > > >> DEAD
> > > > > > > > >> > > > > trigger
> > > > > > > > >> > > > > >> > > deletion
> > > > > > > > >> > > > > >> > > > > of
> > > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state
> is an
> > > > > > ending
> > > > > > > > >> state
> > > > > > > > >> > (we
> > > > > > > > >> > > > > would
> > > > > > > > >> > > > > >> > only
> > > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > > > state when the group is
> EMPTY
> > > and
> > > > > > also all
> > > > > > > > >> of
> > > > > > > > >> > its
> > > > > > > > >> > > > > >> > metadata
> > > > > > > > >> > > > > >> > > are
> > > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this
> > > group
> > > > > > would
> > > > > > > > >> never
> > > > > > > > >> > be
> > > > > > > > >> > > > > >> revived.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to
> store
> > > the
> > > > > > configs
> > > > > > > > >> in
> > > > > > > > >> > the
> > > > > > > > >> > > > > group
> > > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all
> cluster
> > > configs
> > > > > > and
> > > > > > > > >> their
> > > > > > > > >> > > > > values?
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > That's a good question,
> and our
> > > > > > thoughts
> > > > > > > > >> are
> > > > > > > > >> > that
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > so-called
> > > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > > >> > > > > >> > > > > > > > > > > configurations" are
> overrides
> > > of the
> > > > > > > > >> > cluster-level
> > > > > > > > >> > > > > >> > > > > configurations
> > > > > > > > >> > > > > >> > > > > > > > > > > customized per group so
> when an
> > > > > > admin list
> > > > > > > > >> > cluster
> > > > > > > > >> > > > > >> > configs
> > > > > > > > >> > > > > >> > > it's
> > > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> > > > > > "defaults", not
> > > > > > > > >> > showing
> > > > > > > > >> > > > > >> any
> > > > > > > > >> > > > > >> > > > > per-group
> > > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone
> > > configure a
> > > > > > group
> > > > > > > > >> > before
> > > > > > > > >> > > > > >> starting
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > > >> > > > > >> > > > > > > > > > > > we considered allowing
> the
> > > explicit
> > > > > > > > >> > creation of
> > > > > > > > >> > > > > >> groups?
> > > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > > >> > > > > >> > > > > > > > > > > > configs could be
> decoupled
> > > from
> > > > > > the group
> > > > > > > > >> > > > > lifecycle.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > The configs can be created
> > > before
> > > > > > the group
> > > > > > > > >> > itself
> > > > > > > > >> > > > > as
> > > > > > > > >> > > > > >> an
> > > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this
> requires
> > > the
> > > > > > > > >> corresponding
> > > > > > > > >> > > > > >> request to
> > > > > > > > >> > > > > >> > > be
> > > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > > right coordinator based on
> the
> > > group
> > > > > > id ---
> > > > > > > > >> > the only
> > > > > > > > >> > > > > >> > thing
> > > > > > > > >> > > > > >> > > that
> > > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > > >> > > > > >> > > > > > > > > > > when the group itself is
> gone
> > > we
> > > > > > also check
> > > > > > > > >> > if there
> > > > > > > > >> > > > > >> are
> > > > > > > > >> > > > > >> > > any
> > > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > > >> > > > > >> > > > > > > > > > > entities related to that
> group
> > > and
> > > > > > delete
> > > > > > > > >> as
> > > > > > > > >> > well.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed
> > > introduces an
> > > > > > > > >> > asymmetry on
> > > > > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > creation
> > > > > > > > >> > > > > >> > > > > /
> > > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the config
> > > entities,
> > > > > > and we
> > > > > > > > >> > would like
> > > > > > > > >> > > > > >> to
> > > > > > > > >> > > > > >> > > hear
> > > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we should
> aim
> > > for
> > > > > > symmetry
> > > > > > > > >> > i.e.
> > > > > > > > >> > > > > >> totally
> > > > > > > > >> > > > > >> > > > > decouple
> > > > > > > > >> > > > > >> > > > > > > group
> > > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not
> delete
> > > them at
> > > > > > all
> > > > > > > > >> when
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> group
> > > > > > > > >> > > > > >> > is
> > > > > > > > >> > > > > >> > > > > gone,
> > > > > > > > >> > > > > >> > > > > > > but
> > > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > > >> > > > > >> > > > > > > > > > > require explicit deletion
> > > operations
> > > > > > by
> > > > > > > > >> > themselves.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > > Consumer.subscribe
> > > > > > method for
> > > > > > > > >> > the Java
> > > > > > > > >> > > > > >> > client
> > > > > > > > >> > > > > >> > > > > still
> > > > > > > > >> > > > > >> > > > > > > take
> > > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> `java.util.regex.Pattern` of
> > > do we
> > > > > > have
> > > > > > > > >> to
> > > > > > > > >> > > > > >> introduce an
> > > > > > > > >> > > > > >> > > > > overload?
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > I think we do not need to
> > > introduce
> > > > > > an
> > > > > > > > >> > overload, but
> > > > > > > > >> > > > > >> I'm
> > > > > > > > >> > > > > >> > > all
> > > > > > > > >> > > > > >> > > > > ears
> > > > > > > > >> > > > > >> > > > > > > if
> > > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > > >> > > > > >> > > > > > > > > > > may be some compatibility
> > > issues
> > > > > > that we
> > > > > > > > >> may
> > > > > > > > >> > > > > overlook.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine
> > > that we
> > > > > > should
> > > > > > > > >> be
> > > > > > > > >> > clearer
> > > > > > > > >> > > > > >> > about
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > > IBP/metadata.version
> > > > > > from the
> > > > > > > > >> > feature
> > > > > > > > >> > > > > >> flag.
> > > > > > > > >> > > > > >> > > Maybe
> > > > > > > > >> > > > > >> > > > > we
> > > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > > >> > > > > >> > > > > > > > > > > > can switch the default
> for
> > > the
> > > > > > feature
> > > > > > > > >> flag
> > > > > > > > >> > to
> > > > > > > > >> > > > > true
> > > > > > > > >> > > > > >> > > based on
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version once we
> > > want to
> > > > > > make it
> > > > > > > > >> the
> > > > > > > > >> > > > > >> default.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol
> APIs
> > > don't
> > > > > > mention
> > > > > > > > >> > the
> > > > > > > > >> > > > > >> required
> > > > > > > > >> > > > > >> > > ACLs,
> > > > > > > > >> > > > > >> > > > > it
> > > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > > >> > > > > >> > > > > > > > > > > > good to add that for
> > > consistency.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do
> that.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > > >> > > > > >> > requires
> > > > > > > > >> > > > > >> > > > > "Read
> > > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to do
> more
> > > than
> > > > > > reading.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > I had that thought myself
> as
> > > well,
> > > > > > but in
> > > > > > > > >> the
> > > > > > > > >> > end we
> > > > > > > > >> > > > > >> > could
> > > > > > > > >> > > > > >> > > not
> > > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > > >> > > > > >> > > > > > > > > > > better alternative: adding
> > > Write
> > > > > > Group
> > > > > > > > >> seems
> > > > > > > > >> > an
> > > > > > > > >> > > > > >> overkill
> > > > > > > > >> > > > > >> > > here
> > > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we
> only
> > > have
> > > > > > Read /
> > > > > > > > >> > Delete
> > > > > > > > >> > > > > and
> > > > > > > > >> > > > > >> > > Describe
> > > > > > > > >> > > > > >> > > > > on
> > > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to hear
> others
> > > > > > thoughts.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic
> recreation
> > > > > > handled by
> > > > > > > > >> the
> > > > > > > > >> > > > > consumer
> > > > > > > > >> > > > > >> > with
> > > > > > > > >> > > > > >> > > the
> > > > > > > > >> > > > > >> > > > > new
> > > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be
> good
> > > to have
> > > > > > a
> > > > > > > > >> > section on
> > > > > > > > >> > > > > >> this.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > > subscription
> > > > > > right? Yes
> > > > > > > > >> > we can
> > > > > > > > >> > > > > >> add a
> > > > > > > > >> > > > > >> > > > > section
> > > > > > > > >> > > > > >> > > > > > > about
> > > > > > > > >> > > > > >> > > > > > > > > > > that, but basically the
> idea
> > > is that
> > > > > > > > >> consumer
> > > > > > > > >> > would
> > > > > > > > >> > > > > be
> > > > > > > > >> > > > > >> > > totally
> > > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as it's
> > > handled all
> > > > > > by the
> > > > > > > > >> > brokers.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we
> will
> > > write
> > > > > > the
> > > > > > > > >> new
> > > > > > > > >> > > > > >> coordinator
> > > > > > > > >> > > > > >> > in
> > > > > > > > >> > > > > >> > > > > Java.
> > > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > > >> > > > > >> > > > > > > > > > > > this is an implementation
> > > detail,
> > > > > > do we
> > > > > > > > >> > plan to
> > > > > > > > >> > > > > >> have a
> > > > > > > > >> > > > > >> > > new
> > > > > > > > >> > > > > >> > > > > gradle
> > > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > We have not thought about
> > > that. But
> > > > > > I think
> > > > > > > > >> > the
> > > > > > > > >> > > > > answer
> > > > > > > > >> > > > > >> > > should
> > > > > > > > >> > > > > >> > > > > be
> > > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a
> scalability
> > > goal
> > > > > > when it
> > > > > > > > >> > comes to
> > > > > > > > >> > > > > >> how
> > > > > > > > >> > > > > >> > > many
> > > > > > > > >> > > > > >> > > > > > > members
> > > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can
> support?
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I think we
> > > should
> > > > > > shoot for
> > > > > > > > >> > 1000s of
> > > > > > > > >> > > > > >> > > members.
> > > > > > > > >> > > > > >> > > > > But
> > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > > > scalability goals also
> depend
> > > on the
> > > > > > offset
> > > > > > > > >> > > > > management
> > > > > > > > >> > > > > >> > > (commit,
> > > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the
> coordinator
> > > > > > which we
> > > > > > > > >> did
> > > > > > > > >> > not
> > > > > > > > >> > > > > >> cover in
> > > > > > > > >> > > > > >> > > this
> > > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > > >> > > > > >> > > > > > > > > > > hard to give a number that
> > > applies
> > > > > > > > >> > universally.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider
> having
> > > > > > > > >> > SubscribedTopidIds
> > > > > > > > >> > > > > >> instead
> > > > > > > > >> > > > > >> > of
> > > > > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > > >> > > > > >> > > Is the
> > > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > > > > since we have to resolve
> the
> > > regex
> > > > > > on the
> > > > > > > > >> > server,
> > > > > > > > >> > > > > we
> > > > > > > > >> > > > > >> > can
> > > > > > > > >> > > > > >> > > do
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > same
> > > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > > difference is
> > > > > > that
> > > > > > > > >> > sending the
> > > > > > > > >> > > > > >> > regex
> > > > > > > > >> > > > > >> > > is
> > > > > > > > >> > > > > >> > > > > more
> > > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending the topic
> > > names is
> > > > > > less
> > > > > > > > >> > efficient.
> > > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > > >> > > > > >> > > > > > > > > > > > recreation is easier to
> > > handle if
> > > > > > we have
> > > > > > > > >> > topic
> > > > > > > > >> > > > > ids.
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > The main reason to still
> let
> > > the
> > > > > > clients
> > > > > > > > >> send
> > > > > > > > >> > names
> > > > > > > > >> > > > > >> is to
> > > > > > > > >> > > > > >> > > keep
> > > > > > > > >> > > > > >> > > > > the
> > > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids
> on
> > > the
> > > > > > broker /
> > > > > > > > >> > admin
> > > > > > > > >> > > > > client
> > > > > > > > >> > > > > >> > > only.
> > > > > > > > >> > > > > >> > > > > Note
> > > > > > > > >> > > > > >> > > > > > > that
> > > > > > > > >> > > > > >> > > > > > > > > > > although we added topic id
> in
> > > > > > KIP-516, we
> > > > > > > > >> > never
> > > > > > > > >> > > > > >> > > implemented the
> > > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers
> leveraging
> > > the
> > > > > > related
> > > > > > > > >> > newer
> > > > > > > > >> > > > > >> versioned
> > > > > > > > >> > > > > >> > > RPCs,
> > > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > > >> > > > > >> > > > > > > > > > > just set the topic id as
> empty
> > > UUID.
> > > > > > We
> > > > > > > > >> want
> > > > > > > > >> > to keep
> > > > > > > > >> > > > > >> the
> > > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only
> delegate
> > > the
> > > > > > reasoning
> > > > > > > > >> on
> > > > > > > > >> > broker
> > > > > > > > >> > > > > >> and
> > > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at
> 10:45
> > > AM
> > > > > > David
> > > > > > > > >> Jacot
> > > > > > > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> > > > > > discussion
> > > > > > > > >> thread
> > > > > > > > >> > on
> > > > > > > > >> > > > > >> KIP-848:
> > > > > > > > >> > > > > >> > > The
> > > > > > > > >> > > > > >> > > > > Next
> > > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the
> Consumer
> > > > > > Rebalance
> > > > > > > > >> > Protocol.
> > > > > > > > >> > > > > >> With
> > > > > > > > >> > > > > >> > > this
> > > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > > >> > > > > >> > > > > > > we
> > > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > > >> > > > > >> > > > > > > > > > > > > to make the rebalance
> > > protocol
> > > > > > (for
> > > > > > > > >> > consumers)
> > > > > > > > >> > > > > >> more
> > > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > > >> > > > > >> > > > > > > more
> > > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to
> > > implement for
> > > > > > > > >> > clients, and
> > > > > > > > >> > > > > >> easier
> > > > > > > > >> > > > > >> > > to
> > > > > > > > >> > > > > >> > > > > debug
> > > > > > > > >> > > > > >> > > > > > > for
> > > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > > >> > > > > >> > > > >
> > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a look and
> let
> > > me
> > > > > > know what
> > > > > > > > >> > you
> > > > > > > > >> > > > > think.
> > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from
> > > July
> > > > > > 18th to
> > > > > > > > >> > August 8th.
> > > > > > > > >> > > > > >> That
> > > > > > > > >> > > > > >> > > gives
> > > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read and
> digest
> > > this
> > > > > > long
> > > > > > > > >> KIP.
> > > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > >
> > > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > > >> > > > > >> > > > > > >
> > > > > > > > >> > > > > >> > > > >
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> > >
> > > > > > > > >> > > > > >> >
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > >> --
> > > > > > > > >> > > > > >> -- Guozhang
> > > > > > > > >> > > > > >>
> > > > > > > > >> > > > > >
> > > > > > > > >> > > > >
> > > > > > > > >> >
> > > > > > > > >>
> > > > > > > > >
> > > > > >
> > >
> >
> >
> > --
> > -- Guozhang
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Guozhang,

I thought that the assignor will always be consulted when the next
heartbeat request is constructed. In other words,
`PartitionAssignor#metadata` will be called for every heartbeat. This
gives the opportunity for the assignor to enforce a rebalance by
setting the reason to a non-zero value or by changing the bytes. Do
you think that this is not sufficient? Are you concerned by the delay?

Best,
David

On Fri, Sep 9, 2022 at 7:10 AM Guozhang Wang <wa...@gmail.com> wrote:
>
> Hello David,
>
> One of Jun's comments make me thinking:
>
> ```
> In this case, a new assignment is triggered by the client side
> assignor. When constructing the HB, the consumer will always consult
> the client side assignor and propagate the information to the group
> coordinator. In other words, we don't expect users to call
> Consumer#enforceRebalance anymore.
> ```
>
> As I looked at the current PartitionAssignor's interface, we actually do
> not have a way yet to instruct how to construct the next HB request, e.g.
> when the assignor wants to enforce a new rebalance with a new assignment,
> we'd need some customizable APIs inside the PartitionAssignor to indicate
> the next HB telling broker about so. WDYT about adding such an API on the
> PartitionAssignor?
>
>
> Guozhang
>
>
> On Tue, Sep 6, 2022 at 6:09 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Jun,
> >
> > I have updated the KIP to include your feedback. I have also tried to
> > clarify the parts which were not cleared.
> >
> > Best,
> > David
> >
> > On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io> wrote:
> > >
> > > Hi Jun,
> > >
> > > Thanks for your feedback. Let me start by answering your questions
> > > inline and I will update the KIP next week.
> > >
> > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to be
> > fewer
> > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > comments below.
> > >
> > > I would also add that the KIP removes the global sync barrier in the
> > > protocol which is essential to improve group stability and
> > > scalability, and the KIP also simplifies the client by moving most of
> > > the logic to the server side.
> > >
> > > > 30. ConsumerGroupHeartbeatRequest
> > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > changing
> > > > of the partition assignor in the consumers?
> > >
> > > Definitely. The group coordinator will use the assignor used by a
> > > majority of the members. This allows the group to move from one
> > > assignor to another by a roll. This is explained in the Assignor
> > > Selection chapter.
> > >
> > > > 30.2 For each field, could you explain whether it's required in every
> > > > request or the scenarios when it needs to be filled? For example, it's
> > not
> > > > clear to me when TopicPartitions needs to be filled.
> > >
> > > The client is expected to set those fields in case of a connection
> > > issue (e.g. timeout) or when the fields have changed since the last
> > > HB. The server populates those fields as long as the member is not
> > > fully reconciled - the member should acknowledge that it has the
> > > expected epoch and assignment. I will clarify this in the KIP.
> > >
> > > > 31. In the current consumer protocol, the rack affinity between the
> > client
> > > > and the broker is only considered during fetching, but not during
> > assigning
> > > > partitions to consumers. Sometimes, once the assignment is made, there
> > is
> > > > no opportunity for read affinity because no replicas of assigned
> > partitions
> > > > are close to the member. I am wondering if we should use this
> > opportunity
> > > > to address this by including rack in GroupMember.
> > >
> > > That's an interesting idea. I don't see any issue with adding the rack
> > > to the members. I will do so.
> > >
> > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > coordinator is. By moving the event loop model, it seems that we could
> > add
> > > > a metric that tracks the fraction of the time the event loop is doing
> > the
> > > > actual work.
> > >
> > > That's a great idea. I will add it. Thanks.
> > >
> > > > 33. Could we add a section on coordinator failover handling? For
> > example,
> > > > does it need to trigger the check if any group with the wildcard
> > > > subscription now has a new matching topic?
> > >
> > > Sure. When the new group coordinator takes over, it has to:
> > > * Setup the session timeouts.
> > > * Trigger a new assignment if a client side assignor is used. We don't
> > > store the information about the member selected to run the assignment
> > > so we have to start a new one.
> > > * Update the topics metadata, verify the wildcard subscriptions, and
> > > trigger a rebalance if needed.
> > >
> > > > 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > field
> > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > reflect
> > > > the latest group epoch? What about the one in
> > > > ConsumerGroupPartitionMetadataValue and
> > ConsumerGroupMemberMetadataValue?
> > >
> > > Sure. I will clarify that but it is always the latest group epoch.
> > > When the group state is updated, the group epoch is bumped so we use
> > > that one for all the change records related to the update.
> > >
> > > > 35. "the group coordinator will ensure that the following invariants
> > are
> > > > met: ... All members exists." It's possible for a member not to get any
> > > > assigned partitions, right?
> > >
> > > That's right. Here I meant that the members provided by the assignor
> > > in the assignment must exist in the group. The assignor can not make
> > > up new member ids.
> > >
> > > > 36. "He can rejoins the group with a member epoch equals to 0": When
> > would
> > > > a consumer rejoin and what member id would be used?
> > >
> > > A member is expected to abandon all its partitions and rejoins when it
> > > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > > coordinator will have removed the member from the group. The member
> > > can rejoin the group with the same member id but with 0 as epoch. Let
> > > me see if I can clarify this in the KIP.
> > >
> > > > 37. "Instead, power users will have the ability to trigger a
> > reassignment
> > > > by either providing a non-zero reason or by updating the assignor
> > > > metadata." Hmm, this seems to be conflicting with the deprecation of
> > > > Consumer#enforeRebalance.
> > >
> > > In this case, a new assignment is triggered by the client side
> > > assignor. When constructing the HB, the consumer will always consult
> > > the client side assignor and propagate the information to the group
> > > coordinator. In other words, we don't expect users to call
> > > Consumer#enforceRebalance anymore.
> > >
> > > > 38. The reassignment examples are nice. But the section seems to have
> > > > multiple typos.
> > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > epoch=3,
> > > > partitions=[foo-1] too early.
> > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > partitions=[foo-0].
> > >
> > > Sorry for that! I will revise them.
> > >
> > > > 39. Rolling upgrade of consumers: Do we support the upgrade from any
> > old
> > > > version to new one?
> > >
> > > We will support upgrading from the consumer protocol version 3,
> > > introduced in KIP-792. KIP-792 is not implemented yet so the earliest
> > > version is unknown at the moment. This is explained in the migration
> > > plan chapter.
> > >
> > > Thanks again for your feedback, Jun. I will update the KIP based on it
> > > next week.
> > >
> > > Best,
> > > David
> > >
> > > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <ju...@confluent.io.invalid> wrote:
> > > >
> > > > Hi, David,
> > > >
> > > > Thanks for the KIP. Overall, the main benefits of the KIP seem to be
> > fewer
> > > > RPCs during rebalance and more efficient support of wildcard. A few
> > > > comments below.
> > > >
> > > > 30. ConsumerGroupHeartbeatRequest
> > > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> > changing
> > > > of the partition assignor in the consumers?
> > > > 30.2 For each field, could you explain whether it's required in every
> > > > request or the scenarios when it needs to be filled? For example, it's
> > not
> > > > clear to me when TopicPartitions needs to be filled.
> > > >
> > > > 31. In the current consumer protocol, the rack affinity between the
> > client
> > > > and the broker is only considered during fetching, but not during
> > assigning
> > > > partitions to consumers. Sometimes, once the assignment is made, there
> > is
> > > > no opportunity for read affinity because no replicas of assigned
> > partitions
> > > > are close to the member. I am wondering if we should use this
> > opportunity
> > > > to address this by including rack in GroupMember.
> > > >
> > > > 32. On the metric side, often, it's useful to know how busy a group
> > > > coordinator is. By moving the event loop model, it seems that we could
> > add
> > > > a metric that tracks the fraction of the time the event loop is doing
> > the
> > > > actual work.
> > > >
> > > > 33. Could we add a section on coordinator failover handling? For
> > example,
> > > > does it need to trigger the check if any group with the wildcard
> > > > subscription now has a new matching topic?
> > > >
> > > > 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> > > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> > field
> > > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> > reflect
> > > > the latest group epoch? What about the one in
> > > > ConsumerGroupPartitionMetadataValue and
> > ConsumerGroupMemberMetadataValue?
> > > >
> > > > 35. "the group coordinator will ensure that the following invariants
> > are
> > > > met: ... All members exists." It's possible for a member not to get any
> > > > assigned partitions, right?
> > > >
> > > > 36. "He can rejoins the group with a member epoch equals to 0": When
> > would
> > > > a consumer rejoin and what member id would be used?
> > > >
> > > > 37. "Instead, power users will have the ability to trigger a
> > reassignment
> > > > by either providing a non-zero reason or by updating the assignor
> > > > metadata." Hmm, this seems to be conflicting with the deprecation of
> > > > Consumer#enforeRebalance.
> > > >
> > > > 38. The reassignment examples are nice. But the section seems to have
> > > > multiple typos.
> > > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > > 38.2 When the group transitions to epoch 3, C seems to get into
> > epoch=3,
> > > > partitions=[foo-1] too early.
> > > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > > partitions=[foo-0].
> > > >
> > > > 39. Rolling upgrade of consumers: Do we support the upgrade from any
> > old
> > > > version to new one?
> > > >
> > > > Thanks,
> > > >
> > > > Jun
> > > >
> > > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> > <dj...@confluent.io.invalid>
> > > > wrote:
> > > >
> > > > > Hi all,
> > > > >
> > > > > The KIP states that we will re-implement the coordinator in Java. I
> > > > > discussed this offline with a few folks and folks are concerned that
> > > > > we could introduce many regressions in the old protocol if we do so.
> > > > > Therefore, I am going to remove this statement from the KIP. It is an
> > > > > implementation detail after all so it does not have to be decided at
> > > > > this stage. We will likely start by trying to refactor the current
> > > > > implementation as a first step.
> > > > >
> > > > > Cheers,
> > > > > David
> > > > >
> > > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <dj...@confluent.io>
> > wrote:
> > > > > >
> > > > > > Hi Luke,
> > > > > >
> > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > reconciling,
> > > > > stable,
> > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > >
> > > > > > This sentence does not refer to those group states but rather to a
> > > > > > state machine replication (SMR). This refers to the entire state of
> > > > > > group coordinator which is replicated via the log layer. I will
> > > > > > clarify this in the KIP.
> > > > > >
> > > > > > > 1.2. What do you mean "each state machine is modelled as an event
> > > > > loop"?
> > > > > >
> > > > > > The idea is to follow a model similar to the new quorum
> > controller. We
> > > > > > will have N threads to process events. Each __consumer_offsets
> > > > > > partition is assigned to a unique thread and all the events (e.g.
> > > > > > requests, callbacks, etc.) are processed by this thread. This
> > simplify
> > > > > > concurrency and will enable us to do simulation testing for the
> > group
> > > > > > coordinator.
> > > > > >
> > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > partitions?
> > > > > > > Not a state machine "per consumer group" owned by a group
> > coordinator?
> > > > > For
> > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > both
> > > > > exist in
> > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it, or
> > 2?
> > > > > >
> > > > > > See 1.1. The confusion comes from there, I think.
> > > > > >
> > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number of
> > threads
> > > > > used
> > > > > > > to run the state machines. But I'm wondering if the purpose of
> > the
> > > > > threads
> > > > > > > is only to keep the state of each consumer group (or
> > > > > *__consumer_offsets*
> > > > > > > partitions?), and no heavy computation, why should we need
> > > > > multi-threads
> > > > > > > here?
> > > > > >
> > > > > > See 1.2. The idea is to have an ability to shard the processing as
> > the
> > > > > > computation could be heavy.
> > > > > >
> > > > > > > 2.1. The consumer session timeout, why does the default session
> > > > > timeout not
> > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > session
> > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > >
> > > > > > > group.consumer.session.timeout.ms int 30s The timeout to detect
> > client
> > > > > > > failures when using the consumer group protocol.
> > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > session
> > > > > timeout.
> > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > session
> > > > > timeout.
> > > > > >
> > > > > > This is indeed a mistake. The default session timeout should be 45s
> > > > > > (the current default).
> > > > > >
> > > > > > > 2.2. The default server side assignor are [range, uniform],
> > which means
> > > > > > > we'll default to "range" assignor. I'd like to know why not
> > uniform
> > > > > one? I
> > > > > > > thought usually users will choose uniform assignor (former sticky
> > > > > assinor)
> > > > > > > for better evenly distribution. Any other reason we choose range
> > > > > assignor
> > > > > > > as default?
> > > > > > > group.consumer.assignors List range, uniform The server side
> > assignors.
> > > > > >
> > > > > > The order on the server side has no influence because the client
> > must
> > > > > > chose the selector that he wants to use. There is no default in the
> > > > > > current proposal. If the assignor is not specified by the client,
> > the
> > > > > > request is rejected. The default client value for
> > > > > > `group.remote.assignor` is `uniform` though.
> > > > > >
> > > > > > Thanks for your very good comments, Luke. I hope that my answers
> > help
> > > > > > to clarify things. I will update the KIP as well based on your
> > > > > > feedback.
> > > > > >
> > > > > > Cheers,
> > > > > > David
> > > > > >
> > > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com>
> > wrote:
> > > > > > >
> > > > > > > Hi David,
> > > > > > >
> > > > > > > Thanks for the update.
> > > > > > >
> > > > > > > Some more questions:
> > > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > > The new group coordinator will have a state machine per
> > > > > > > *__consumer_offsets* partitions, where each state machine is
> > modelled
> > > > > as an
> > > > > > > event loop. Those state machines will be executed in
> > > > > > > *group.coordinator.threads* threads.
> > > > > > >
> > > > > > > 1.1. I think the state machine are: "Empty, assigning,
> > reconciling,
> > > > > stable,
> > > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > > 1.2. What do you mean "each state machine is modelled as an event
> > > > > loop"?
> > > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > > partitions?
> > > > > > > Not a state machine "per consumer group" owned by a group
> > coordinator?
> > > > > For
> > > > > > > example, if one group coordinator owns 2 consumer groups, and
> > both
> > > > > exist in
> > > > > > > *__consumer_offsets-0*, will we have 1 state machine for it, or
> > 2?
> > > > > > > 1.4. I know the "*group.coordinator.threads" *is the number of
> > threads
> > > > > used
> > > > > > > to run the state machines. But I'm wondering if the purpose of
> > the
> > > > > threads
> > > > > > > is only to keep the state of each consumer group (or
> > > > > *__consumer_offsets*
> > > > > > > partitions?), and no heavy computation, why should we need
> > > > > multi-threads
> > > > > > > here?
> > > > > > >
> > > > > > > 2. For the default value in the new configs:
> > > > > > > 2.1. The consumer session timeout, why does the default session
> > > > > timeout not
> > > > > > > locate between min (45s) and max(60s)? I thought the min/max
> > session
> > > > > > > timeout is to define lower/upper bound of it, no?
> > > > > > >
> > > > > > > group.consumer.session.timeout.ms int 30s The timeout to detect
> > client
> > > > > > > failures when using the consumer group protocol.
> > > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> > session
> > > > > timeout.
> > > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> > session
> > > > > timeout.
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > 2.2. The default server side assignor are [range, uniform],
> > which means
> > > > > > > we'll default to "range" assignor. I'd like to know why not
> > uniform
> > > > > one? I
> > > > > > > thought usually users will choose uniform assignor (former sticky
> > > > > assinor)
> > > > > > > for better evenly distribution. Any other reason we choose range
> > > > > assignor
> > > > > > > as default?
> > > > > > > group.consumer.assignors List range, uniform The server side
> > assignors.
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > Thank you.
> > > > > > > Luke
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com>
> > wrote:
> > > > > > >
> > > > > > > > Hi Sagar,
> > > > > > > >
> > > > > > > > I have some thoughts about Kafka Connect integrating with
> > KIP-848,
> > > > > but I
> > > > > > > > think we should have a separate discussion thread for the Kafka
> > > > > Connect
> > > > > > > > KIP: Integrating Kafka Connect With New Consumer Rebalance
> > Protocol
> > > > > [1],
> > > > > > > > and let this discussion thread focus on consumer rebalance
> > protocol,
> > > > > WDYT?
> > > > > > > >
> > > > > > > > [1]
> > > > > > > >
> > > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >
> > > > > > > > Thank you.
> > > > > > > > Luke
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> > sagarmeansocean@gmail.com>
> > > > > wrote:
> > > > > > > >
> > > > > > > >> Thank you Guozhang/David for the feedback. Looks like there's
> > > > > agreement on
> > > > > > > >> using separate APIs for Connect. I would revisit the doc and
> > see
> > > > > what
> > > > > > > >> changes are to be made.
> > > > > > > >>
> > > > > > > >> Thanks!
> > > > > > > >> Sagar.
> > > > > > > >>
> > > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > > <dj...@confluent.io.invalid>
> > > > > > > >> wrote:
> > > > > > > >>
> > > > > > > >> > Hi Sagar,
> > > > > > > >> >
> > > > > > > >> > Thanks for the feedback and the document. That's really
> > helpful. I
> > > > > > > >> > will take a look at it.
> > > > > > > >> >
> > > > > > > >> > Overall, it seems to me that both Connect and the Consumer
> > could
> > > > > share
> > > > > > > >> > the same underlying "engine". The main difference is that
> > the
> > > > > Consumer
> > > > > > > >> > assigns topic-partitions to members whereas Connect assigns
> > tasks
> > > > > to
> > > > > > > >> > workers. I see two ways to move forward:
> > > > > > > >> > 1) We extend the new proposed APIs to support different
> > resource
> > > > > types
> > > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > > >> > 2) We use new dedicated APIs for Connect. The dedicated APIs
> > > > > would be
> > > > > > > >> > similar to the new ones but different on the
> > content/resources and
> > > > > > > >> > they would rely on the same engine on the coordinator side.
> > > > > > > >> >
> > > > > > > >> > I personally lean towards 2) because I am not a fan of
> > > > > overcharging
> > > > > > > >> > APIs to serve different purposes. That being said, I am not
> > > > > opposed to
> > > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > > >> >
> > > > > > > >> > I think that we can continue to discuss it here for now in
> > order
> > > > > to
> > > > > > > >> > ensure that this KIP is compatible with what we will do for
> > > > > Connect in
> > > > > > > >> > the future.
> > > > > > > >> >
> > > > > > > >> > Best,
> > > > > > > >> > David
> > > > > > > >> >
> > > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> > djacot@confluent.io>
> > > > > wrote:
> > > > > > > >> > >
> > > > > > > >> > > Hi all,
> > > > > > > >> > >
> > > > > > > >> > > I am back from vacation. I will go through and address
> > your
> > > > > comments
> > > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > > >> > >
> > > > > > > >> > > Cheers,
> > > > > > > >> > > David
> > > > > > > >> > >
> > > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > > gharris1727@gmail.com
> > > > > > > >> >
> > > > > > > >> > wrote:
> > > > > > > >> > > >
> > > > > > > >> > > > Hey All!
> > > > > > > >> > > >
> > > > > > > >> > > > Thanks for the KIP, it's wonderful to see cooperative
> > > > > rebalancing
> > > > > > > >> > making it
> > > > > > > >> > > > down the stack!
> > > > > > > >> > > >
> > > > > > > >> > > > I had a few questions:
> > > > > > > >> > > >
> > > > > > > >> > > > 1. The 'Rejected Alternatives' section describes how
> > member
> > > > > epoch
> > > > > > > >> > should
> > > > > > > >> > > > advance in step with the group epoch and assignment
> > epoch
> > > > > values. I
> > > > > > > >> > think
> > > > > > > >> > > > that this is a good idea for the reasons described in
> > the
> > > > > KIP. When
> > > > > > > >> the
> > > > > > > >> > > > protocol is incrementally assigning partitions to a
> > worker,
> > > > > what
> > > > > > > >> member
> > > > > > > >> > > > epoch does each incremental assignment use? Are member
> > epochs
> > > > > > > >> re-used,
> > > > > > > >> > and
> > > > > > > >> > > > a single member epoch can correspond to multiple
> > different
> > > > > > > >> > (monotonically
> > > > > > > >> > > > larger) assignments?
> > > > > > > >> > > >
> > > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> > > > > > > >> coordinator? If
> > > > > > > >> > > > not, should custom client-side assignor implementations
> > > > > interact
> > > > > > > >> with
> > > > > > > >> > the
> > > > > > > >> > > > Reason field, and how is its common meaning agreed
> > upon? If
> > > > > so, what
> > > > > > > >> > is the
> > > > > > > >> > > > benefit of a distinct Reason field over including such
> > > > > functionality
> > > > > > > >> > in the
> > > > > > > >> > > > opaque metadata?
> > > > > > > >> > > >
> > > > > > > >> > > > 3. The following is included in the KIP: "Thanks to
> > this, the
> > > > > input
> > > > > > > >> of
> > > > > > > >> > the
> > > > > > > >> > > > client side assignor is entirely driven by the group
> > > > > coordinator.
> > > > > > > >> The
> > > > > > > >> > > > consumer is no longer responsible for maintaining any
> > state
> > > > > besides
> > > > > > > >> its
> > > > > > > >> > > > assigned partitions." Does this mean that the
> > client-side
> > > > > assignor
> > > > > > > >> MAY
> > > > > > > >> > > > incorporate additional non-Metadata state (such as
> > partition
> > > > > > > >> > throughput,
> > > > > > > >> > > > cpu/memory metrics, config topics, etc), or that
> > additional
> > > > > > > >> > non-Metadata
> > > > > > > >> > > > state SHOULD NOT be used?
> > > > > > > >> > > >
> > > > > > > >> > > > 4. I see that there are separate classes
> > > > > > > >> > > > for
> > org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > > >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor
> > that
> > > > > seem to
> > > > > > > >> > > > overlap significantly. Is it possible for these two
> > > > > implementations
> > > > > > > >> to
> > > > > > > >> > be
> > > > > > > >> > > > unified? This would serve to promote feature parity of
> > > > > server-side
> > > > > > > >> and
> > > > > > > >> > > > client-side assignors, and would also facilitate
> > operational
> > > > > > > >> > flexibility in
> > > > > > > >> > > > certain situations. For example, if a server-side
> > assignor
> > > > > has some
> > > > > > > >> > poor
> > > > > > > >> > > > behavior and needs a patch, deploying the patched
> > assignor to
> > > > > the
> > > > > > > >> > client
> > > > > > > >> > > > and switching one consumer group to a client-side
> > assignor
> > > > > may be
> > > > > > > >> > faster
> > > > > > > >> > > > and less risky than patching all of the brokers. With
> > the
> > > > > currently
> > > > > > > >> > > > proposed distinct APIs, a non-trivial reimplementation
> > would
> > > > > have
> > > > > > > >> to be
> > > > > > > >> > > > assembled, and if the two APIs have diverged
> > significantly,
> > > > > then it
> > > > > > > >> is
> > > > > > > >> > > > possible that a reimplementation would not be possible.
> > > > > > > >> > > >
> > > > > > > >> > > > --
> > > > > > > >> > > > Greg Harris
> > > > > > > >> > > > gharris1727@gmail.com
> > > > > > > >> > > > github.com/gharris1727
> > > > > > > >> > > >
> > > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > wrote:
> > > > > > > >> > > >
> > > > > > > >> > > > > Hi Guozhang/David,
> > > > > > > >> > > > >
> > > > > > > >> > > > > I created a confluence page to discuss how Connect
> > would
> > > > > need to
> > > > > > > >> > change
> > > > > > > >> > > > > based on the new rebalance protocol. Here's the page:
> > > > > > > >> > > > >
> > > > > > > >> > > > >
> > > > > > > >> >
> > > > > > > >>
> > > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > > >> > > > >
> > > > > > > >> > > > > It's also pretty longish and I have tried to keep a
> > format
> > > > > > > >> similar to
> > > > > > > >> > > > > KIP-848. Let me know what you think. Also, do you
> > think this
> > > > > > > >> should
> > > > > > > >> > be
> > > > > > > >> > > > > moved to a separate discussion thread or is this one
> > fine?
> > > > > > > >> > > > >
> > > > > > > >> > > > > Thanks!
> > > > > > > >> > > > > Sagar.
> > > > > > > >> > > > >
> > > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > > sagarmeansocean@gmail.com>
> > > > > > > >> > wrote:
> > > > > > > >> > > > >
> > > > > > > >> > > > > > Hello Guozhang,
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > Thank you so much for the doc on Kafka Streams.
> > Sure, I
> > > > > would do
> > > > > > > >> > the
> > > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > Thanks!
> > > > > > > >> > > > > > Sagar.
> > > > > > > >> > > > > >
> > > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > > > > > >> wangguoz@gmail.com>
> > > > > > > >> > > > > wrote:
> > > > > > > >> > > > > >
> > > > > > > >> > > > > >> Hello Sagar,
> > > > > > > >> > > > > >>
> > > > > > > >> > > > > >> It would be great if you could come back with some
> > > > > analysis on
> > > > > > > >> > how to
> > > > > > > >> > > > > >> implement the Connect side integration with the new
> > > > > protocol;
> > > > > > > >> so
> > > > > > > >> > far
> > > > > > > >> > > > > >> besides leveraging on the new "protocol type" we
> > did not
> > > > > yet
> > > > > > > >> think
> > > > > > > >> > > > > through
> > > > > > > >> > > > > >> the Connect side implementations. For Streams
> > here's a
> > > > > draft of
> > > > > > > >> > > > > >> integration
> > > > > > > >> > > > > >> plan:
> > > > > > > >> > > > > >>
> > > > > > > >> > > > > >>
> > > > > > > >> > > > >
> > > > > > > >> >
> > > > > > > >>
> > > > >
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > > >> > > > > >>
> > > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > > >> sagarmeansocean@gmail.com
> > > > > > > >> > >
> > > > > > > >> > > > > wrote:
> > > > > > > >> > > > > >>
> > > > > > > >> > > > > >> > Hi David,
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> > Thank you for your response. The reason I thought
> > > > > connect can
> > > > > > > >> > also fit
> > > > > > > >> > > > > >> into
> > > > > > > >> > > > > >> > this new scheme is that even today the connect
> > uses a
> > > > > > > >> > > > > WorkerCoordinator
> > > > > > > >> > > > > >> > extending from AbstractCoordinator to empower
> > > > > rebalances of
> > > > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> > > > > > > >> protocolType()
> > > > > > > >> > to
> > > > > > > >> > > > > >> connect
> > > > > > > >> > > > > >> > and uses the metadata() method by plumbing into
> > > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> > I think the changes to support connect would be
> > > > > similar at a
> > > > > > > >> > high
> > > > > > > >> > > > > level
> > > > > > > >> > > > > >> to
> > > > > > > >> > > > > >> > the changes in streams mainly because of the
> > Client
> > > > > side
> > > > > > > >> > assignors
> > > > > > > >> > > > > being
> > > > > > > >> > > > > >> > used in both. At an implementation level, we
> > might
> > > > > need to
> > > > > > > >> make
> > > > > > > >> > a lot
> > > > > > > >> > > > > of
> > > > > > > >> > > > > >> > changes to get onto this new assignment protocol
> > like
> > > > > > > >> enhancing
> > > > > > > >> > the
> > > > > > > >> > > > > >> > JoinGroup request/response and SyncGroup and
> > using
> > > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > > >> > > > > >> > API etc again on similar lines to streams (or
> > there
> > > > > might be
> > > > > > > >> > > > > >> deviations). I
> > > > > > > >> > > > > >> > would try to perform a detailed analysis of the
> > same
> > > > > and we
> > > > > > > >> > can have
> > > > > > > >> > > > > a
> > > > > > > >> > > > > >> > separate discussion thread for that as that would
> > > > > derail this
> > > > > > > >> > > > > discussion
> > > > > > > >> > > > > >> > thread. Let me know if that sounds good to you.
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> > Thanks!
> > > > > > > >> > > > > >> > Sagar.
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> > wrote:
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`.
> > Sure, I
> > > > > can
> > > > > > > >> > mention it.
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> > > 2) The idea is to transition C from his current
> > > > > assignment
> > > > > > > >> to
> > > > > > > >> > his
> > > > > > > >> > > > > >> > > target assignment when he can move to epoch 3.
> > When
> > > > > that
> > > > > > > >> > happens,
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > member assignment is updated and persisted
> > with all
> > > > > its
> > > > > > > >> > assigned
> > > > > > > >> > > > > >> > > partitions even if they are not all revoked
> > yet. In
> > > > > other
> > > > > > > >> > words, the
> > > > > > > >> > > > > >> > > member assignment becomes the target
> > assignment.
> > > > > This is
> > > > > > > >> > basically
> > > > > > > >> > > > > an
> > > > > > > >> > > > > >> > > optimization to avoid having to write all the
> > > > > changes to
> > > > > > > >> the
> > > > > > > >> > log.
> > > > > > > >> > > > > The
> > > > > > > >> > > > > >> > > examples are based on the persisted state so I
> > > > > understand
> > > > > > > >> the
> > > > > > > >> > > > > >> > > confusion. Let me see if I can improve this in
> > the
> > > > > > > >> > description.
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse the
> > protocol
> > > > > with a
> > > > > > > >> > client side
> > > > > > > >> > > > > >> > > assignor if it fits in the protocol. The
> > assignment
> > > > > is
> > > > > > > >> about
> > > > > > > >> > > > > >> > > topicid-partitions + metadata, could Connect
> > fit
> > > > > into this?
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> > > Best,
> > > > > > > >> > > > > >> > > David
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > > >> > > > > >> wrote:
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > Hi David,
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor
> > observations:
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > 1) In the Assignment Error section in Client
> > Side
> > > > > mode
> > > > > > > >> > Assignment
> > > > > > > >> > > > > >> > > process,
> > > > > > > >> > > > > >> > > > you mentioned => `In this case, the client
> > side
> > > > > assignor
> > > > > > > >> can
> > > > > > > >> > > > > return
> > > > > > > >> > > > > >> an
> > > > > > > >> > > > > >> > > > error to the group coordinator`. In this
> > case are
> > > > > you
> > > > > > > >> > referring to
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > Assignor returning an AssignmentError that's
> > > > > listed down
> > > > > > > >> > towards
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > end?
> > > > > > > >> > > > > >> > > > If yes, do you think it would make sense to
> > > > > mention this
> > > > > > > >> > > > > explicitly
> > > > > > > >> > > > > >> > here?
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > 2) In the Case Studies section, I have a
> > slight
> > > > > > > >> confusion,
> > > > > > > >> > not
> > > > > > > >> > > > > sure
> > > > > > > >> > > > > >> if
> > > > > > > >> > > > > >> > > > others have the same. Consider this step:
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > When B heartbeats, the group coordinator
> > > > > transitions him
> > > > > > > >> to
> > > > > > > >> > epoch
> > > > > > > >> > > > > 3
> > > > > > > >> > > > > >> > > because
> > > > > > > >> > > > > >> > > > B has no partitions to revoke. It persists
> > the
> > > > > change and
> > > > > > > >> > reply.
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > > >> > > > > >> > > >       - A
> > > > > > > >> > > > > >> > > >       - B
> > > > > > > >> > > > > >> > > >       - C
> > > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0,
> > foo-1]
> > > > > > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > > > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > When C heartbeats, it transitions to epoch 3
> > but
> > > > > cannot
> > > > > > > >> get
> > > > > > > >> > foo-1
> > > > > > > >> > > > > >> yet.
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > Here,it's mentioned that member C can't get
> > the
> > > > > foo-1
> > > > > > > >> > partition
> > > > > > > >> > > > > yet,
> > > > > > > >> > > > > >> > but
> > > > > > > >> > > > > >> > > > based on the description above, it seems it
> > > > > already has
> > > > > > > >> it.
> > > > > > > >> > Do you
> > > > > > > >> > > > > >> > think
> > > > > > > >> > > > > >> > > it
> > > > > > > >> > > > > >> > > > would be better to remove it and populate it
> > only
> > > > > when it
> > > > > > > >> > actually
> > > > > > > >> > > > > >> gets
> > > > > > > >> > > > > >> > > it?
> > > > > > > >> > > > > >> > > > I see this in a lot of other places, so have
> > I
> > > > > > > >> understood it
> > > > > > > >> > > > > >> > incorrectly
> > > > > > > >> > > > > >> > > ?
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > Regarding connect , it might be out of scope
> > of
> > > > > this
> > > > > > > >> > discussion,
> > > > > > > >> > > > > but
> > > > > > > >> > > > > >> > from
> > > > > > > >> > > > > >> > > > what I understood it would probably be
> > running in
> > > > > client
> > > > > > > >> > side
> > > > > > > >> > > > > >> assignor
> > > > > > > >> > > > > >> > > mode
> > > > > > > >> > > > > >> > > > even on the new rebalance protocol as it has
> > its
> > > > > own
> > > > > > > >> Custom
> > > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > Thanks!
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > Sagar.
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > wrote:
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move forward
> > with
> > > > > > > >> > specialized API
> > > > > > > >> > > > > >> > > > > instead of relying on one generic API. For
> > > > > Connect, we
> > > > > > > >> > can apply
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > exact same pattern and reuse/share the core
> > > > > > > >> > implementation on
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > server side. For the schema registry, I
> > think
> > > > > that we
> > > > > > > >> > should
> > > > > > > >> > > > > >> consider
> > > > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > > > membership/leader
> > > > > > > >> > election.
> > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > >> > > > > Best,
> > > > > > > >> > > > > >> > > > > David
> > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael
> > Juma <
> > > > > > > >> > ismael@juma.me.uk
> > > > > > > >> > > > > >
> > > > > > > >> > > > > >> > > wrote:
> > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > >> > > > > > document the differences in more detail
> > before
> > > > > > > >> deciding
> > > > > > > >> > one
> > > > > > > >> > > > > way
> > > > > > > >> > > > > >> or
> > > > > > > >> > > > > >> > > > > another.
> > > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > > java.util.regex.Pattern,
> > > > > > > >> they
> > > > > > > >> > expect
> > > > > > > >> > > > > >> > their
> > > > > > > >> > > > > >> > > > > > semantics to be honored. If we are doing
> > > > > something
> > > > > > > >> > different,
> > > > > > > >> > > > > >> then
> > > > > > > >> > > > > >> > we
> > > > > > > >> > > > > >> > > > > > should consider adding an overload with
> > our own
> > > > > > > >> Pattern
> > > > > > > >> > class
> > > > > > > >> > > > > (I
> > > > > > > >> > > > > >> > > don't
> > > > > > > >> > > > > >> > > > > > think we'd want to expose re2j's at this
> > > > > point).
> > > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new
> > protocol
> > > > > should
> > > > > > > >> > > > > integrate
> > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > >> > > > > with
> > > > > > > >> > > > > >> > > > > > it and should handle the topic
> > recreation case
> > > > > > > >> > correctly.
> > > > > > > >> > > > > That's
> > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > >> > > main
> > > > > > > >> > > > > >> > > > > > part we need to handle. I agree with
> > David
> > > > > that we'd
> > > > > > > >> > want to
> > > > > > > >> > > > > add
> > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > >> > > > > ids
> > > > > > > >> > > > > >> > > > > > to the relevant protocols that don't
> > have it
> > > > > yet and
> > > > > > > >> > that we
> > > > > > > >> > > > > can
> > > > > > > >> > > > > >> > > probably
> > > > > > > >> > > > > >> > > > > > focus on the internals versus adding new
> > APIs
> > > > > to the
> > > > > > > >> > Java
> > > > > > > >> > > > > >> Consumer
> > > > > > > >> > > > > >> > > > > (unless
> > > > > > > >> > > > > >> > > > > > we find that adding new APIs is required
> > for
> > > > > > > >> reasonable
> > > > > > > >> > > > > >> semantics).
> > > > > > > >> > > > > >> > > > > > 3. I am still not sure about the
> > coordinator
> > > > > storing
> > > > > > > >> the
> > > > > > > >> > > > > >> configs.
> > > > > > > >> > > > > >> > > It's
> > > > > > > >> > > > > >> > > > > > powerful for configs to be centralized
> > in the
> > > > > > > >> metadata
> > > > > > > >> > log for
> > > > > > > >> > > > > >> > > various
> > > > > > > >> > > > > >> > > > > > reasons (auditability, visibility,
> > consistency,
> > > > > > > >> etc.).
> > > > > > > >> > > > > >> Similarly, I
> > > > > > > >> > > > > >> > > am
> > > > > > > >> > > > > >> > > > > not
> > > > > > > >> > > > > >> > > > > > sure about automatically deleting
> > configs in a
> > > > > way
> > > > > > > >> that
> > > > > > > >> > they
> > > > > > > >> > > > > >> cannot
> > > > > > > >> > > > > >> > > be
> > > > > > > >> > > > > >> > > > > > recovered. A good property for modern
> > systems
> > > > > is to
> > > > > > > >> > minimize
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > number
> > > > > > > >> > > > > >> > > > > of
> > > > > > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > >> > > > > > Ismael
> > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David
> > Jacot
> > > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > >> > > > > > wrote:
> > > > > > > >> > > > > >> > > > > >
> > > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > the last
> > > > > step
> > > > > > > >> when
> > > > > > > >> > > > > clients
> > > > > > > >> > > > > >> > > flip the
> > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in which
> > we
> > > > > would
> > > > > > > >> have a
> > > > > > > >> > > > > window
> > > > > > > >> > > > > >> > where
> > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > >> > > > > > > new
> > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols /
> > rpcs
> > > > > are
> > > > > > > >> used
> > > > > > > >> > by
> > > > > > > >> > > > > >> members
> > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > "mimic"
> > > > > the old
> > > > > > > >> > behavior
> > > > > > > >> > > > > >> while
> > > > > > > >> > > > > >> > > > > using the
> > > > > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > > > > present
> > > > > > > >> about.
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > Noted. I just published a new version
> > of KIP
> > > > > which
> > > > > > > >> > includes
> > > > > > > >> > > > > >> more
> > > > > > > >> > > > > >> > > > > > > details about this. See the "Supporting
> > > > > Online
> > > > > > > >> > Consumer
> > > > > > > >> > > > > Group
> > > > > > > >> > > > > >> > > Upgrade"
> > > > > > > >> > > > > >> > > > > > > and the "Compatibility, Deprecation,
> > and
> > > > > Migration
> > > > > > > >> > Plan". I
> > > > > > > >> > > > > >> think
> > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > >> > > > > > > I have to think through a few cases
> > now but
> > > > > the
> > > > > > > >> > overall idea
> > > > > > > >> > > > > >> and
> > > > > > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > > > > KIP-516 the
> > > > > > > >> > topic ids
> > > > > > > >> > > > > >> are
> > > > > > > >> > > > > >> > > only
> > > > > > > >> > > > > >> > > > > used
> > > > > > > >> > > > > >> > > > > > > as
> > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but
> > they
> > > > > are not
> > > > > > > >> > exposed
> > > > > > > >> > > > > via
> > > > > > > >> > > > > >> any
> > > > > > > >> > > > > >> > > > > public
> > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > question is,
> > > > > first
> > > > > > > >> > should we
> > > > > > > >> > > > > >> let
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > >> > > > > >> > > > > > > > client to be maintaining the names
> > -> ids
> > > > > mapping
> > > > > > > >> > itself
> > > > > > > >> > > > > to
> > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs
> > and the
> > > > > new
> > > > > > > >> RPCs
> > > > > > > >> > with
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > >> > > > > ids;
> > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > exposing
> > > > > the
> > > > > > > >> > topic ids
> > > > > > > >> > > > > in
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > consumer
> > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > subscribe/assign, as
> > > > > > > >> well
> > > > > > > >> > as in
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > >> > deletion-and-recreation).
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > a) Assuming that we would include
> > converting
> > > > > all
> > > > > > > >> the
> > > > > > > >> > offsets
> > > > > > > >> > > > > >> > > related
> > > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this KIP,
> > the
> > > > > consumer
> > > > > > > >> > would be
> > > > > > > >> > > > > >> able
> > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > >> > > > > > > fully operate with topic ids. That
> > being
> > > > > said, it
> > > > > > > >> > still has
> > > > > > > >> > > > > to
> > > > > > > >> > > > > >> > > provide
> > > > > > > >> > > > > >> > > > > > > the topics names in various APIs so
> > having a
> > > > > > > >> mapping
> > > > > > > >> > in the
> > > > > > > >> > > > > >> > > consumer
> > > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > > >> > > > > >> > > > > > > b) I don't have a strong opinion on
> > this.
> > > > > Here I
> > > > > > > >> > wonder if
> > > > > > > >> > > > > >> this
> > > > > > > >> > > > > >> > > goes
> > > > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I would
> > rather
> > > > > focus
> > > > > > > >> on
> > > > > > > >> > the
> > > > > > > >> > > > > >> > internals
> > > > > > > >> > > > > >> > > > > > > here and we can consider this
> > separately if
> > > > > we see
> > > > > > > >> > value in
> > > > > > > >> > > > > >> doing
> > > > > > > >> > > > > >> > > it.
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > Coming back to Ismael's point about
> > using
> > > > > topic ids
> > > > > > > >> > in the
> > > > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think
> > that
> > > > > there
> > > > > > > >> is
> > > > > > > >> > one
> > > > > > > >> > > > > >> > advantage
> > > > > > > >> > > > > >> > > in
> > > > > > > >> > > > > >> > > > > > > favour of it. The consumer will have
> > the
> > > > > > > >> opportunity
> > > > > > > >> > to
> > > > > > > >> > > > > >> validate
> > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > >> > > > > > > the topics exists before passing them
> > into
> > > > > the
> > > > > > > >> group
> > > > > > > >> > > > > rebalance
> > > > > > > >> > > > > >> > > > > > > protocol. Obviously, the coordinator
> > will
> > > > > also
> > > > > > > >> notice
> > > > > > > >> > it but
> > > > > > > >> > > > > >> it
> > > > > > > >> > > > > >> > > does
> > > > > > > >> > > > > >> > > > > > > not really have a way to reject an
> > invalid
> > > > > topic in
> > > > > > > >> > the
> > > > > > > >> > > > > >> response.
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other
> > minor
> > > > > > > >> questions
> > > > > > > >> > > > > except
> > > > > > > >> > > > > >> for
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > personally
> > > > > I think
> > > > > > > >> > it's not
> > > > > > > >> > > > > >> > > necessary
> > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > Pattern,
> > > > > but
> > > > > > > >> > instead we
> > > > > > > >> > > > > >> still
> > > > > > > >> > > > > >> > > use
> > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > subscription may
> > > > > > > >> be
> > > > > > > >> > > > > >> rejected by
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > very rare
> > > > > > > >> scenario
> > > > > > > >> > I felt
> > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > >> > > an
> > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > subscription may
> > > > > be
> > > > > > > >> more
> > > > > > > >> > > > > >> vulnerable
> > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > >> > > > > various
> > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > That could work. I have to look at the
> > > > > differences
> > > > > > > >> > between
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > two
> > > > > > > >> > > > > >> > > > > > > engines to better understand the
> > potential
> > > > > issues.
> > > > > > > >> My
> > > > > > > >> > > > > >> > > understanding is
> > > > > > > >> > > > > >> > > > > > > that would work for all the basic
> > regular
> > > > > > > >> > expressions. The
> > > > > > > >> > > > > >> > > differences
> > > > > > > >> > > > > >> > > > > > > between the two are mainly about the
> > various
> > > > > > > >> character
> > > > > > > >> > > > > >> classes. I
> > > > > > > >> > > > > >> > > > > > > wonder what other people think about
> > this.
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > Best,
> > > > > > > >> > > > > >> > > > > > > David
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM
> > Guozhang
> > > > > Wang <
> > > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > > >> > > > > >> > > >
> > > > > > > >> > > > > >> > > > > wrote:
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > Thanks David! I think on the high
> > level
> > > > > there are
> > > > > > > >> > two meta
> > > > > > > >> > > > > >> > > points we
> > > > > > > >> > > > > >> > > > > need
> > > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> > the last
> > > > > step
> > > > > > > >> when
> > > > > > > >> > > > > clients
> > > > > > > >> > > > > >> > > flip the
> > > > > > > >> > > > > >> > > > > > > flag
> > > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in which
> > we
> > > > > would
> > > > > > > >> have a
> > > > > > > >> > > > > window
> > > > > > > >> > > > > >> > where
> > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > >> > > > > > > new
> > > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols /
> > rpcs
> > > > > are
> > > > > > > >> used
> > > > > > > >> > by
> > > > > > > >> > > > > >> members
> > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> > "mimic"
> > > > > the old
> > > > > > > >> > behavior
> > > > > > > >> > > > > >> while
> > > > > > > >> > > > > >> > > > > using the
> > > > > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > > > > present
> > > > > > > >> about.
> > > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > > > > KIP-516 the
> > > > > > > >> > topic ids
> > > > > > > >> > > > > >> are
> > > > > > > >> > > > > >> > > only
> > > > > > > >> > > > > >> > > > > used
> > > > > > > >> > > > > >> > > > > > > as
> > > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but
> > they
> > > > > are not
> > > > > > > >> > exposed
> > > > > > > >> > > > > via
> > > > > > > >> > > > > >> any
> > > > > > > >> > > > > >> > > > > public
> > > > > > > >> > > > > >> > > > > > > APIs
> > > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> > question is,
> > > > > first
> > > > > > > >> > should we
> > > > > > > >> > > > > >> let
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > consumer
> > > > > > > >> > > > > >> > > > > > > > client to be maintaining the names
> > -> ids
> > > > > mapping
> > > > > > > >> > itself
> > > > > > > >> > > > > to
> > > > > > > >> > > > > >> > fully
> > > > > > > >> > > > > >> > > > > > > leverage
> > > > > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs
> > and the
> > > > > new
> > > > > > > >> RPCs
> > > > > > > >> > with
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > >> > > > > ids;
> > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> > exposing
> > > > > the
> > > > > > > >> > topic ids
> > > > > > > >> > > > > in
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > consumer
> > > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > > subscribe/assign, as
> > > > > > > >> well
> > > > > > > >> > as in
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > rebalance
> > > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > > >> > deletion-and-recreation).
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other
> > minor
> > > > > > > >> questions
> > > > > > > >> > > > > except
> > > > > > > >> > > > > >> for
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> > personally
> > > > > I think
> > > > > > > >> > it's not
> > > > > > > >> > > > > >> > > necessary
> > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> > Pattern,
> > > > > but
> > > > > > > >> > instead we
> > > > > > > >> > > > > >> still
> > > > > > > >> > > > > >> > > use
> > > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > > subscription may
> > > > > > > >> be
> > > > > > > >> > > > > >> rejected by
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > server.
> > > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> > very rare
> > > > > > > >> scenario
> > > > > > > >> > I felt
> > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > >> > > an
> > > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> > subscription may
> > > > > be
> > > > > > > >> more
> > > > > > > >> > > > > >> vulnerable
> > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > >> > > > > various
> > > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM
> > David Jacot
> > > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let me
> > answer
> > > > > your
> > > > > > > >> > questions
> > > > > > > >> > > > > >> > inline.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 1. I think it's premature to
> > talk about
> > > > > > > >> target
> > > > > > > >> > > > > versions
> > > > > > > >> > > > > >> for
> > > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > >> > > > > >> > > > > > > > > > removal of the existing group
> > protocol.
> > > > > > > >> Unlike
> > > > > > > >> > KRaft,
> > > > > > > >> > > > > >> this
> > > > > > > >> > > > > >> > > > > affects a
> > > > > > > >> > > > > >> > > > > > > core
> > > > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > > > deprecation/removal
> > > > > > > >> > will be
> > > > > > > >> > > > > >> > heavily
> > > > > > > >> > > > > >> > > > > > > dependent
> > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > >> > > > > >> > > > > > > > > > how quickly applications migrate
> > to
> > > > > the new
> > > > > > > >> > protocol.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to
> > release
> > > > > this in
> > > > > > > >> > 4.x, but
> > > > > > > >> > > > > it
> > > > > > > >> > > > > >> > > wasn't
> > > > > > > >> > > > > >> > > > > made
> > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > >> > > > > >> > > > > > > > > > why. If we added that as a way to
> > > > > estimate
> > > > > > > >> when
> > > > > > > >> > we'd
> > > > > > > >> > > > > >> > > deprecate
> > > > > > > >> > > > > >> > > > > and
> > > > > > > >> > > > > >> > > > > > > remove
> > > > > > > >> > > > > >> > > > > > > > > > the group protocol, I also
> > suggest
> > > > > removing
> > > > > > > >> > this part.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> > > > > explained, I
> > > > > > > >> plan
> > > > > > > >> > to
> > > > > > > >> > > > > >> rewrite
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > group
> > > > > > > >> > > > > >> > > > > > > > > coordinator in Java while we
> > implement
> > > > > the new
> > > > > > > >> > protocol.
> > > > > > > >> > > > > >> This
> > > > > > > >> > > > > >> > > means
> > > > > > > >> > > > > >> > > > > > > > > that the internals will be slightly
> > > > > different
> > > > > > > >> > (e.g.
> > > > > > > >> > > > > >> threading
> > > > > > > >> > > > > >> > > > > model).
> > > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the
> > > > > switch from
> > > > > > > >> > the old
> > > > > > > >> > > > > >> group
> > > > > > > >> > > > > >> > > > > > > > > coordinator to the new group
> > coordinator
> > > > > to a
> > > > > > > >> > major
> > > > > > > >> > > > > >> release.
> > > > > > > >> > > > > >> > > The
> > > > > > > >> > > > > >> > > > > > > > > alternative would be to use a flag
> > to do
> > > > > the
> > > > > > > >> > switch
> > > > > > > >> > > > > >> instead
> > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > >> > > > > relying
> > > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the
> > details of
> > > > > the
> > > > > > > >> > migration
> > > > > > > >> > > > > >> story.
> > > > > > > >> > > > > >> > > It
> > > > > > > >> > > > > >> > > > > sounds
> > > > > > > >> > > > > >> > > > > > > > > like
> > > > > > > >> > > > > >> > > > > > > > > > we're saying we will support
> > online
> > > > > > > >> migrations.
> > > > > > > >> > Is
> > > > > > > >> > > > > that
> > > > > > > >> > > > > >> > > correct?
> > > > > > > >> > > > > >> > > > > We
> > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > >> > > > > >> > > > > > > > > > explain this in detail. It could
> > also
> > > > > be done
> > > > > > > >> > as a
> > > > > > > >> > > > > >> separate
> > > > > > > >> > > > > >> > > KIP,
> > > > > > > >> > > > > >> > > > > if
> > > > > > > >> > > > > >> > > > > > > it's
> > > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Yes, we will support online
> > migrations
> > > > > for the
> > > > > > > >> > group.
> > > > > > > >> > > > > That
> > > > > > > >> > > > > >> > > means
> > > > > > > >> > > > > >> > > > > that
> > > > > > > >> > > > > >> > > > > > > > > a group using the old protocol
> > will be
> > > > > able to
> > > > > > > >> > switch to
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Let me briefly explain how that
> > will work
> > > > > > > >> though.
> > > > > > > >> > It is
> > > > > > > >> > > > > >> > > basically a
> > > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or
> > > > > rolled to a
> > > > > > > >> > software
> > > > > > > >> > > > > >> > > supporting
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > new group coordinator. Both the
> > old and
> > > > > the new
> > > > > > > >> > > > > >> coordinator
> > > > > > > >> > > > > >> > > will
> > > > > > > >> > > > > >> > > > > > > > > support the old protocol and rely
> > on the
> > > > > same
> > > > > > > >> > persisted
> > > > > > > >> > > > > >> > > metadata so
> > > > > > > >> > > > > >> > > > > > > > > they can work together. This point
> > is an
> > > > > > > >> offline
> > > > > > > >> > > > > >> migration.
> > > > > > > >> > > > > >> > We
> > > > > > > >> > > > > >> > > > > cannot
> > > > > > > >> > > > > >> > > > > > > > > do this one live because it would
> > require
> > > > > > > >> > shutting down
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > current
> > > > > > > >> > > > > >> > > > > > > > > coordinator and starting up the
> > new one
> > > > > and
> > > > > > > >> that
> > > > > > > >> > would
> > > > > > > >> > > > > >> cause
> > > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> > version/IBP
> > > > > must be
> > > > > > > >> > upgraded
> > > > > > > >> > > > > to
> > > > > > > >> > > > > >> X
> > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > >> > > > > order
> > > > > > > >> > > > > >> > > > > > > > > to enable the new protocol. This
> > cannot
> > > > > be done
> > > > > > > >> > before
> > > > > > > >> > > > > 1)
> > > > > > > >> > > > > >> is
> > > > > > > >> > > > > >> > > > > > > > > terminated because the old
> > coordinator
> > > > > doesn't
> > > > > > > >> > support
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > new
> > > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be upgraded
> > to a
> > > > > version
> > > > > > > >> > > > > supporting
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > online
> > > > > > > >> > > > > >> > > > > > > > > migration (must have KIP-792). If
> > the
> > > > > consumer
> > > > > > > >> is
> > > > > > > >> > > > > already
> > > > > > > >> > > > > >> > > there.
> > > > > > > >> > > > > >> > > > > > > > > Nothing must be done at this point.
> > > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled
> > with the
> > > > > > > >> feature
> > > > > > > >> > flag
> > > > > > > >> > > > > >> turned
> > > > > > > >> > > > > >> > > on.
> > > > > > > >> > > > > >> > > > > The
> > > > > > > >> > > > > >> > > > > > > > > consumer group is automatically
> > > > > converted when
> > > > > > > >> > the first
> > > > > > > >> > > > > >> > > consumer
> > > > > > > >> > > > > >> > > > > > > > > using the new protocol joins the
> > group.
> > > > > While
> > > > > > > >> the
> > > > > > > >> > > > > members
> > > > > > > >> > > > > >> > > using the
> > > > > > > >> > > > > >> > > > > > > > > old protocol are being upgraded,
> > the old
> > > > > > > >> protocol
> > > > > > > >> > is
> > > > > > > >> > > > > >> proxied
> > > > > > > >> > > > > >> > > into
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Let me clarify all of this in the
> > KIP.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are
> > pushing the
> > > > > pattern
> > > > > > > >> > > > > >> subscriptions
> > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > >> > > > > >> > > > > > > > > > but it seems like there could be
> > some
> > > > > tricky
> > > > > > > >> > > > > >> compatibility
> > > > > > > >> > > > > >> > > > > issues.
> > > > > > > >> > > > > >> > > > > > > Will
> > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > >> > > > > >> > > > > > > > > > have a mechanism for users to
> > detect
> > > > > that
> > > > > > > >> they
> > > > > > > >> > need to
> > > > > > > >> > > > > >> > update
> > > > > > > >> > > > > >> > > > > their
> > > > > > > >> > > > > >> > > > > > > regex
> > > > > > > >> > > > > >> > > > > > > > > > before switching to the new
> > protocol?
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > I think that I am a bit more
> > optimistic
> > > > > than
> > > > > > > >> you
> > > > > > > >> > on this
> > > > > > > >> > > > > >> > > point. I
> > > > > > > >> > > > > >> > > > > > > > > believe that the majority of the
> > cases
> > > > > are
> > > > > > > >> simple
> > > > > > > >> > > > > regexes
> > > > > > > >> > > > > >> > which
> > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > >> > > > > > > > > work with the new engine. The
> > > > > coordinator will
> > > > > > > >> > verify
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > regex
> > > > > > > >> > > > > >> > > > > anyway
> > > > > > > >> > > > > >> > > > > > > > > and reject the consumer if the
> > regex is
> > > > > not
> > > > > > > >> valid.
> > > > > > > >> > > > > Coming
> > > > > > > >> > > > > >> > back
> > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > migration path, in the worst case,
> > the
> > > > > first
> > > > > > > >> > upgraded
> > > > > > > >> > > > > >> > consumer
> > > > > > > >> > > > > >> > > > > joining
> > > > > > > >> > > > > >> > > > > > > > > the group will be rejected. This
> > should
> > > > > be used
> > > > > > > >> > as the
> > > > > > > >> > > > > >> last
> > > > > > > >> > > > > >> > > > > defence, I
> > > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > One way for customers to validate
> > their
> > > > > regex
> > > > > > > >> > before
> > > > > > > >> > > > > >> > upgrading
> > > > > > > >> > > > > >> > > > > their
> > > > > > > >> > > > > >> > > > > > > > > prod would be to test them with
> > another
> > > > > group.
> > > > > > > >> For
> > > > > > > >> > > > > >> instance,
> > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod
> > environment.
> > > > > > > >> Another
> > > > > > > >> > way
> > > > > > > >> > > > > >> would be
> > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > >> > > > > > > > > extend the consumer-group tool to
> > > > > provide a
> > > > > > > >> regex
> > > > > > > >> > > > > >> validation
> > > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be enough in
> > your
> > > > > > > >> opinion?
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last question,
> > will
> > > > > the
> > > > > > > >> Java
> > > > > > > >> > client
> > > > > > > >> > > > > >> allow
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > users to
> > > > > > > >> > > > > >> > > > > > > > > > stick with the current regex
> > engine for
> > > > > > > >> > compatibility
> > > > > > > >> > > > > >> > > reasons?
> > > > > > > >> > > > > >> > > > > For
> > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > >> > > > > >> > > > > > > > > > it may be handy to keep using
> > client
> > > > > based
> > > > > > > >> > regex at
> > > > > > > >> > > > > >> first
> > > > > > > >> > > > > >> > to
> > > > > > > >> > > > > >> > > keep
> > > > > > > >> > > > > >> > > > > > > > > > migrations simple and then
> > migrate to
> > > > > server
> > > > > > > >> > based
> > > > > > > >> > > > > >> regexes
> > > > > > > >> > > > > >> > > as a
> > > > > > > >> > > > > >> > > > > > > second
> > > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > I understand your point but I am
> > > > > concerned that
> > > > > > > >> > this
> > > > > > > >> > > > > would
> > > > > > > >> > > > > >> > > allow
> > > > > > > >> > > > > >> > > > > users
> > > > > > > >> > > > > >> > > > > > > > > to actually stay in this mode. That
> > > > > would go
> > > > > > > >> > against our
> > > > > > > >> > > > > >> goal
> > > > > > > >> > > > > >> > > of
> > > > > > > >> > > > > >> > > > > > > > > simplifying the client because we
> > would
> > > > > have to
> > > > > > > >> > continue
> > > > > > > >> > > > > >> > > monitoring
> > > > > > > >> > > > > >> > > > > > > > > the metadata on the client side. I
> > would
> > > > > rather
> > > > > > > >> > not do
> > > > > > > >> > > > > >> this.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> > > > > coordinator
> > > > > > > >> will
> > > > > > > >> > be
> > > > > > > >> > > > > >> > > responsible for
> > > > > > > >> > > > > >> > > > > > > storing
> > > > > > > >> > > > > >> > > > > > > > > > the configurations and that the
> > > > > > > >> configurations
> > > > > > > >> > will be
> > > > > > > >> > > > > >> > > deleted
> > > > > > > >> > > > > >> > > > > when
> > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a
> > transition to
> > > > > DEAD
> > > > > > > >> > trigger
> > > > > > > >> > > > > >> > deletion
> > > > > > > >> > > > > >> > > of
> > > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > That's right. The configurations
> > will be
> > > > > > > >> deleted
> > > > > > > >> > when
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > group is
> > > > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to store the
> > > > > configs in
> > > > > > > >> the
> > > > > > > >> > group
> > > > > > > >> > > > > >> > > coordinator
> > > > > > > >> > > > > >> > > > > > > make it
> > > > > > > >> > > > > >> > > > > > > > > > harder to list all cluster
> > configs and
> > > > > their
> > > > > > > >> > values?
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > > > configurations are
> > > > > > > >> > overrides
> > > > > > > >> > > > > >> of
> > > > > > > >> > > > > >> > > cluster
> > > > > > > >> > > > > >> > > > > > > > > configs. If you want to know all
> > the
> > > > > overrides
> > > > > > > >> > though,
> > > > > > > >> > > > > you
> > > > > > > >> > > > > >> > > would
> > > > > > > >> > > > > >> > > > > have
> > > > > > > >> > > > > >> > > > > > > > > to ask all the group coordinators.
> > You
> > > > > cannot
> > > > > > > >> > rely on
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > metadata
> > > > > > > >> > > > > >> > > > > log
> > > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 8. How would someone configure a
> > group
> > > > > before
> > > > > > > >> > starting
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > consumers?
> > > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > > >> > > > > >> > > > > > > > > > we considered allowing the
> > explicit
> > > > > creation
> > > > > > > >> of
> > > > > > > >> > > > > groups?
> > > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > configs could be decoupled from
> > the
> > > > > group
> > > > > > > >> > lifecycle.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> > automatically
> > > > > created in
> > > > > > > >> > this
> > > > > > > >> > > > > case.
> > > > > > > >> > > > > >> > > However,
> > > > > > > >> > > > > >> > > > > > > > > the configs will be lost after the
> > > > > retention
> > > > > > > >> > period of
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > group
> > > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe
> > method
> > > > > for the
> > > > > > > >> > Java
> > > > > > > >> > > > > >> client
> > > > > > > >> > > > > >> > > still
> > > > > > > >> > > > > >> > > > > take
> > > > > > > >> > > > > >> > > > > > > a
> > > > > > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do
> > we
> > > > > have to
> > > > > > > >> > introduce
> > > > > > > >> > > > > an
> > > > > > > >> > > > > >> > > overload?
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > That's a very group question. I
> > forgot
> > > > > about
> > > > > > > >> that
> > > > > > > >> > one.
> > > > > > > >> > > > > As
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not
> > fully
> > > > > > > >> compatible
> > > > > > > >> > with
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > engine
> > > > > > > >> > > > > >> > > > > that
> > > > > > > >> > > > > >> > > > > > > > > we plan to use, it might be better
> > to
> > > > > deprecate
> > > > > > > >> > it and
> > > > > > > >> > > > > >> use an
> > > > > > > >> > > > > >> > > > > overload
> > > > > > > >> > > > > >> > > > > > > > > which takes a string. We would
> > rely on
> > > > > the
> > > > > > > >> server
> > > > > > > >> > side
> > > > > > > >> > > > > >> > > validation.
> > > > > > > >> > > > > >> > > > > > > > > During the migration, I think that
> > we
> > > > > could
> > > > > > > >> still
> > > > > > > >> > try to
> > > > > > > >> > > > > >> > > toString
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > regex and use it. That should
> > work, I
> > > > > think, in
> > > > > > > >> > the
> > > > > > > >> > > > > >> majority
> > > > > > > >> > > > > >> > > of the
> > > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine that we
> > > > > should be
> > > > > > > >> > clearer
> > > > > > > >> > > > > about
> > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > >> > > > > reason
> > > > > > > >> > > > > >> > > > > > > to
> > > > > > > >> > > > > >> > > > > > > > > > switch to IBP/metadata.version
> > from the
> > > > > > > >> feature
> > > > > > > >> > flag.
> > > > > > > >> > > > > >> Maybe
> > > > > > > >> > > > > >> > > we
> > > > > > > >> > > > > >> > > > > mean
> > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > >> > > > > > > > > we
> > > > > > > >> > > > > >> > > > > > > > > > can switch the default for the
> > feature
> > > > > flag
> > > > > > > >> to
> > > > > > > >> > true
> > > > > > > >> > > > > >> based
> > > > > > > >> > > > > >> > on
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > > > > metadata.version once we want to
> > make
> > > > > it the
> > > > > > > >> > default.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > My plan was to use that feature
> > flag
> > > > > mainly
> > > > > > > >> > during the
> > > > > > > >> > > > > >> > > development
> > > > > > > >> > > > > >> > > > > > > > > phase. I should not have mentioned
> > it, I
> > > > > think,
> > > > > > > >> > because
> > > > > > > >> > > > > we
> > > > > > > >> > > > > >> > > could
> > > > > > > >> > > > > >> > > > > use
> > > > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs
> > don't
> > > > > mention
> > > > > > > >> the
> > > > > > > >> > > > > required
> > > > > > > >> > > > > >> > > ACLs, it
> > > > > > > >> > > > > >> > > > > > > would
> > > > > > > >> > > > > >> > > > > > > > > be
> > > > > > > >> > > > > >> > > > > > > > > > good to add that for consistency.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > > > >> ConsumerGroupHeartbeat
> > > > > > > >> > > > > requires
> > > > > > > >> > > > > >> > > "Read
> > > > > > > >> > > > > >> > > > > Group"
> > > > > > > >> > > > > >> > > > > > > > > even
> > > > > > > >> > > > > >> > > > > > > > > > though it seems to do more than
> > > > > reading.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > I agree. This is how the current
> > > > > protocol works
> > > > > > > >> > though.
> > > > > > > >> > > > > We
> > > > > > > >> > > > > >> > only
> > > > > > > >> > > > > >> > > > > > > > > require "Read Group" to join a
> > group. We
> > > > > could
> > > > > > > >> > consider
> > > > > > > >> > > > > >> > > changing
> > > > > > > >> > > > > >> > > > > this
> > > > > > > >> > > > > >> > > > > > > > > but I am not sure that it is worth
> > it.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 13. How is topic recreation
> > handled by
> > > > > the
> > > > > > > >> > consumer
> > > > > > > >> > > > > with
> > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > >> > > > > > > group
> > > > > > > >> > > > > >> > > > > > > > > > protocol? It would be good to
> > have a
> > > > > section
> > > > > > > >> on
> > > > > > > >> > this.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> > perspective, the
> > > > > new
> > > > > > > >> topic
> > > > > > > >> > will
> > > > > > > >> > > > > >> have a
> > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > >> > > > > > > > > topic id so it will treat it like a
> > > > > topic with
> > > > > > > >> a
> > > > > > > >> > > > > different
> > > > > > > >> > > > > >> > > name.
> > > > > > > >> > > > > >> > > > > The
> > > > > > > >> > > > > >> > > > > > > > > only issue is that the fetch/commit
> > > > > offsets
> > > > > > > >> APIs
> > > > > > > >> > do not
> > > > > > > >> > > > > >> > support
> > > > > > > >> > > > > >> > > > > topic
> > > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would reuse the
> > > > > offsets
> > > > > > > >> based
> > > > > > > >> > on the
> > > > > > > >> > > > > >> > same.
> > > > > > > >> > > > > >> > > I
> > > > > > > >> > > > > >> > > > > think
> > > > > > > >> > > > > >> > > > > > > > > that we should update those APIs
> > as well
> > > > > in
> > > > > > > >> order
> > > > > > > >> > to be
> > > > > > > >> > > > > >> > > consistent
> > > > > > > >> > > > > >> > > > > end
> > > > > > > >> > > > > >> > > > > > > > > to end. That would strengthen the
> > > > > semantics of
> > > > > > > >> the
> > > > > > > >> > > > > >> consumer.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will
> > write the
> > > > > new
> > > > > > > >> > coordinator
> > > > > > > >> > > > > >> in
> > > > > > > >> > > > > >> > > Java.
> > > > > > > >> > > > > >> > > > > Even
> > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > >> > > > > >> > > > > > > > > > this is an implementation
> > detail, do
> > > > > we plan
> > > > > > > >> to
> > > > > > > >> > have a
> > > > > > > >> > > > > >> new
> > > > > > > >> > > > > >> > > gradle
> > > > > > > >> > > > > >> > > > > > > module
> > > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability
> > goal when
> > > > > it
> > > > > > > >> comes
> > > > > > > >> > to how
> > > > > > > >> > > > > >> many
> > > > > > > >> > > > > >> > > > > members
> > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > new
> > > > > > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > We don't have numbers at the
> > moment. The
> > > > > > > >> protocol
> > > > > > > >> > should
> > > > > > > >> > > > > >> > > support
> > > > > > > >> > > > > >> > > > > 1000s
> > > > > > > >> > > > > >> > > > > > > > > of members per group. We will
> > measure
> > > > > this when
> > > > > > > >> > we have
> > > > > > > >> > > > > a
> > > > > > > >> > > > > >> > first
> > > > > > > >> > > > > >> > > > > > > > > implementation. Note that we might
> > have
> > > > > other
> > > > > > > >> > > > > bottlenecks
> > > > > > > >> > > > > >> > down
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> > > > > SubscribedTopidIds
> > > > > > > >> > instead
> > > > > > > >> > > > > of
> > > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > > >> > > > > >> Is
> > > > > > > >> > > > > >> > the
> > > > > > > >> > > > > >> > > > > idea
> > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > >> > > > > > > > > > since we have to resolve the
> > regex on
> > > > > the
> > > > > > > >> > server, we
> > > > > > > >> > > > > >> can do
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > same
> > > > > > > >> > > > > >> > > > > > > for
> > > > > > > >> > > > > >> > > > > > > > > > the topic name? The difference
> > is that
> > > > > > > >> sending
> > > > > > > >> > the
> > > > > > > >> > > > > >> regex is
> > > > > > > >> > > > > >> > > more
> > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > >> > > > > >> > > > > > > > > > whereas sending the topic names
> > is less
> > > > > > > >> > efficient.
> > > > > > > >> > > > > >> > > Furthermore,
> > > > > > > >> > > > > >> > > > > > > delete
> > > > > > > >> > > > > >> > > > > > > > > and
> > > > > > > >> > > > > >> > > > > > > > > > recreation is easier to handle
> > if we
> > > > > have
> > > > > > > >> topic
> > > > > > > >> > ids.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > The idea was to consolidate the
> > metadata
> > > > > lookup
> > > > > > > >> > on the
> > > > > > > >> > > > > >> server
> > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > >> > > > > both
> > > > > > > >> > > > > >> > > > > > > > > paths but I do agree with your
> > point. As
> > > > > a
> > > > > > > >> second
> > > > > > > >> > > > > though,
> > > > > > > >> > > > > >> > using
> > > > > > > >> > > > > >> > > > > topic
> > > > > > > >> > > > > >> > > > > > > > > ids may be better here for the
> > delete and
> > > > > > > >> > recreation
> > > > > > > >> > > > > case.
> > > > > > > >> > > > > >> > > Also, I
> > > > > > > >> > > > > >> > > > > > > > > suppose that we may allow users to
> > > > > subscribe
> > > > > > > >> with
> > > > > > > >> > topic
> > > > > > > >> > > > > >> ids
> > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > > > future because that is the only
> > way to be
> > > > > > > >> really
> > > > > > > >> > robust
> > > > > > > >> > > > > to
> > > > > > > >> > > > > >> > > topic
> > > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > > >> > > > > >> > > > > > > > > David
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM
> > David
> > > > > Jacot <
> > > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > Thanks for your comments. Please
> > find
> > > > > my
> > > > > > > >> answers
> > > > > > > >> > > > > below.
> > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on
> > > > > topic IDs
> > > > > > > >> > with the
> > > > > > > >> > > > > >> > > exception
> > > > > > > >> > > > > >> > > > > of the
> > > > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > > >> > > > > >> I
> > > > > > > >> > > > > >> > am
> > > > > > > >> > > > > >> > > not
> > > > > > > >> > > > > >> > > > > sure
> > > > > > > >> > > > > >> > > > > > > > > > if using topic names is the
> > right call
> > > > > here.
> > > > > > > >> I
> > > > > > > >> > need to
> > > > > > > >> > > > > >> > think
> > > > > > > >> > > > > >> > > > > about it
> > > > > > > >> > > > > >> > > > > > > > > > a little more. Obviously, the
> > KIP does
> > > > > not
> > > > > > > >> > change the
> > > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs.
> > This
> > > > > may be
> > > > > > > >> > something
> > > > > > > >> > > > > >> that
> > > > > > > >> > > > > >> > we
> > > > > > > >> > > > > >> > > > > should
> > > > > > > >> > > > > >> > > > > > > > > > include though as it would give
> > better
> > > > > > > >> overall
> > > > > > > >> > > > > >> guarantee in
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > > >> > > > > >> > > > > > > > > > - You're right. I think that I
> > should
> > > > > not
> > > > > > > >> have
> > > > > > > >> > > > > mentioned
> > > > > > > >> > > > > >> > this
> > > > > > > >> > > > > >> > > > > flag at
> > > > > > > >> > > > > >> > > > > > > > > > all. I will remove it. We can
> > use an
> > > > > internal
> > > > > > > >> > > > > >> configuration
> > > > > > > >> > > > > >> > > while
> > > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> > > > > supported. The
> > > > > > > >> > change is
> > > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > > >> > > > > >> > > > > > > > > > only requirement is that the
> > cluster
> > > > > uses
> > > > > > > >> topic
> > > > > > > >> > IDs.
> > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > > >> > > > > >> > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM
> > > > > Guozhang
> > > > > > > >> Wang <
> > > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here
> > are
> > > > > some
> > > > > > > >> replies
> > > > > > > >> > > > > inlined
> > > > > > > >> > > > > >> > > below:
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM
> > > > > Ismael Juma
> > > > > > > >> <
> > > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has
> > the
> > > > > > > >> potential
> > > > > > > >> > to be a
> > > > > > > >> > > > > >> > great
> > > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > > >> > > > > >> > > > > > > > > > > > initial questions/comments:
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's premature to
> > talk
> > > > > about
> > > > > > > >> > target
> > > > > > > >> > > > > >> versions
> > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > > >> > > > > >> > > > > > > > > > > > removal of the existing group
> > > > > protocol.
> > > > > > > >> > Unlike
> > > > > > > >> > > > > >> KRaft,
> > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > >> > > > > > > affects a
> > > > > > > >> > > > > >> > > > > > > > > core
> > > > > > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > > > > > >> > deprecation/removal will
> > > > > > > >> > > > > >> be
> > > > > > > >> > > > > >> > > heavily
> > > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > > >> > > > > >> > > > > > > > > > > > how quickly applications
> > migrate
> > > > > to the
> > > > > > > >> new
> > > > > > > >> > > > > >> protocol.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think
> > we can
> > > > > > > >> remove
> > > > > > > >> > the
> > > > > > > >> > > > > >> proposed
> > > > > > > >> > > > > >> > > > > timeline
> > > > > > > >> > > > > >> > > > > > > in
> > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation,
> > and
> > > > > Migration
> > > > > > > >> > Plan` and
> > > > > > > >> > > > > >> > > instead
> > > > > > > >> > > > > >> > > > > just
> > > > > > > >> > > > > >> > > > > > > state
> > > > > > > >> > > > > >> > > > > > > > > > > that we will decide in the
> > future
> > > > > about
> > > > > > > >> when
> > > > > > > >> > we
> > > > > > > >> > > > > would
> > > > > > > >> > > > > >> > > > > deprecate old
> > > > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to
> > > > > release this
> > > > > > > >> > in 4.x,
> > > > > > > >> > > > > >> but
> > > > > > > >> > > > > >> > it
> > > > > > > >> > > > > >> > > > > wasn't
> > > > > > > >> > > > > >> > > > > > > made
> > > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > > >> > > > > >> > > > > > > > > > > > why. If we added that as a
> > way to
> > > > > > > >> estimate
> > > > > > > >> > when
> > > > > > > >> > > > > we'd
> > > > > > > >> > > > > >> > > > > deprecate
> > > > > > > >> > > > > >> > > > > > > and
> > > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I also
> > suggest
> > > > > > > >> removing
> > > > > > > >> > this
> > > > > > > >> > > > > >> part.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > I think that's not specifically
> > > > > related to
> > > > > > > >> the
> > > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for client
> > > > > upgrades.
> > > > > > > >> I.e.
> > > > > > > >> > the
> > > > > > > >> > > > > >> > > broker-side
> > > > > > > >> > > > > >> > > > > > > > > > > implementation may be done
> > first,
> > > > > and then
> > > > > > > >> the
> > > > > > > >> > > > > client
> > > > > > > >> > > > > >> > side,
> > > > > > > >> > > > > >> > > > > and we
> > > > > > > >> > > > > >> > > > > > > > > would
> > > > > > > >> > > > > >> > > > > > > > > > > only mark it as "released" by
> > the
> > > > > time
> > > > > > > >> clients
> > > > > > > >> > > > > >> > > implementations
> > > > > > > >> > > > > >> > > > > are
> > > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > > >> > > > > >> > > > > > > > > > > that time, to enable the
> > feature the
> > > > > > > >> clients
> > > > > > > >> > need to
> > > > > > > >> > > > > >> > first
> > > > > > > >> > > > > >> > > > > swap-in
> > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce
> > and
> > > > > then set
> > > > > > > >> > the flag
> > > > > > > >> > > > > >> > with a
> > > > > > > >> > > > > >> > > > > second
> > > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's
> > > > > better to be
> > > > > > > >> > released
> > > > > > > >> > > > > >> in a
> > > > > > > >> > > > > >> > > major
> > > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the
> > > > > details of
> > > > > > > >> the
> > > > > > > >> > > > > migration
> > > > > > > >> > > > > >> > > story.
> > > > > > > >> > > > > >> > > > > It
> > > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > > >> > > > > >> > > > > > > > > > > > we're saying we will support
> > online
> > > > > > > >> > migrations. Is
> > > > > > > >> > > > > >> that
> > > > > > > >> > > > > >> > > > > correct?
> > > > > > > >> > > > > >> > > > > > > We
> > > > > > > >> > > > > >> > > > > > > > > should
> > > > > > > >> > > > > >> > > > > > > > > > > > explain this in detail. It
> > could
> > > > > also be
> > > > > > > >> > done as a
> > > > > > > >> > > > > >> > > separate
> > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > >> > > > > >> > > > > > > if
> > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the part we
> > can
> > > > > be more
> > > > > > > >> > concrete
> > > > > > > >> > > > > >> about
> > > > > > > >> > > > > >> > > for
> > > > > > > >> > > > > >> > > > > sure
> > > > > > > >> > > > > >> > > > > > > (and
> > > > > > > >> > > > > >> > > > > > > > > > > this is related to your
> > question 2)
> > > > > above).
> > > > > > > >> > We will
> > > > > > > >> > > > > >> work
> > > > > > > >> > > > > >> > on
> > > > > > > >> > > > > >> > > > > making
> > > > > > > >> > > > > >> > > > > > > it
> > > > > > > >> > > > > >> > > > > > > > > more
> > > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel as we
> > solicit
> > > > > more
> > > > > > > >> > feedback.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are
> > pushing
> > > > > the
> > > > > > > >> > pattern
> > > > > > > >> > > > > >> > > subscriptions
> > > > > > > >> > > > > >> > > > > to
> > > > > > > >> > > > > >> > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > > >> > > > > >> > > > > > > > > > > > but it seems like there
> > could be
> > > > > some
> > > > > > > >> tricky
> > > > > > > >> > > > > >> > > compatibility
> > > > > > > >> > > > > >> > > > > > > issues.
> > > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for users to
> > > > > detect that
> > > > > > > >> > they
> > > > > > > >> > > > > need
> > > > > > > >> > > > > >> to
> > > > > > > >> > > > > >> > > update
> > > > > > > >> > > > > >> > > > > > > their
> > > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > > >> > > > > >> > > > > > > > > > > > before switching to the new
> > > > > protocol?
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need some
> > tooling for
> > > > > > > >> non-java
> > > > > > > >> > client
> > > > > > > >> > > > > >> > users
> > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > >> > > > > sort
> > > > > > > >> > > > > >> > > > > > > of
> > > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> > > > > switching to
> > > > > > > >> the
> > > > > > > >> > new
> > > > > > > >> > > > > >> > protocol.
> > > > > > > >> > > > > >> > > I
> > > > > > > >> > > > > >> > > > > do not
> > > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > > >> > > > > >> > > > > > > > > > > specific idea on top of my head
> > > > > though,
> > > > > > > >> maybe
> > > > > > > >> > others
> > > > > > > >> > > > > >> like
> > > > > > > >> > > > > >> > > @Matt
> > > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the last
> > question,
> > > > > will the
> > > > > > > >> > Java
> > > > > > > >> > > > > >> client
> > > > > > > >> > > > > >> > > allow
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > > >> > > > > >> > > > > > > > > > > > stick with the current regex
> > > > > engine for
> > > > > > > >> > > > > >> compatibility
> > > > > > > >> > > > > >> > > > > reasons?
> > > > > > > >> > > > > >> > > > > > > For
> > > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to keep using
> > > > > client
> > > > > > > >> based
> > > > > > > >> > regex
> > > > > > > >> > > > > at
> > > > > > > >> > > > > >> > > first to
> > > > > > > >> > > > > >> > > > > keep
> > > > > > > >> > > > > >> > > > > > > > > > > > migrations simple and then
> > migrate
> > > > > to
> > > > > > > >> > server based
> > > > > > > >> > > > > >> > > regexes
> > > > > > > >> > > > > >> > > > > as a
> > > > > > > >> > > > > >> > > > > > > > > second
> > > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not thought
> > about
> > > > > that for
> > > > > > > >> > java
> > > > > > > >> > > > > >> clients,
> > > > > > > >> > > > > >> > > and
> > > > > > > >> > > > > >> > > > > we can
> > > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> > compatibility
> > > > > issues do
> > > > > > > >> > you have
> > > > > > > >> > > > > in
> > > > > > > >> > > > > >> > > mind?
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that the group
> > > > > coordinator
> > > > > > > >> > will be
> > > > > > > >> > > > > >> > > > > responsible for
> > > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > > >> > > > > >> > > > > > > > > > > > the configurations and that
> > the
> > > > > > > >> > configurations
> > > > > > > >> > > > > will
> > > > > > > >> > > > > >> be
> > > > > > > >> > > > > >> > > > > deleted
> > > > > > > >> > > > > >> > > > > > > when
> > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> > > > > transition to
> > > > > > > >> DEAD
> > > > > > > >> > > > > trigger
> > > > > > > >> > > > > >> > > deletion
> > > > > > > >> > > > > >> > > > > of
> > > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an
> > > > > ending
> > > > > > > >> state
> > > > > > > >> > (we
> > > > > > > >> > > > > would
> > > > > > > >> > > > > >> > only
> > > > > > > >> > > > > >> > > > > > > transit to
> > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > >> > > > > >> > > > > > > > > > > state when the group is EMPTY
> > and
> > > > > also all
> > > > > > > >> of
> > > > > > > >> > its
> > > > > > > >> > > > > >> > metadata
> > > > > > > >> > > > > >> > > are
> > > > > > > >> > > > > >> > > > > > > gone),
> > > > > > > >> > > > > >> > > > > > > > > once
> > > > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this
> > group
> > > > > would
> > > > > > > >> never
> > > > > > > >> > be
> > > > > > > >> > > > > >> revived.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to store
> > the
> > > > > configs
> > > > > > > >> in
> > > > > > > >> > the
> > > > > > > >> > > > > group
> > > > > > > >> > > > > >> > > > > coordinator
> > > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > > >> > > > > >> > > > > > > > > > > > harder to list all cluster
> > configs
> > > > > and
> > > > > > > >> their
> > > > > > > >> > > > > values?
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > That's a good question, and our
> > > > > thoughts
> > > > > > > >> are
> > > > > > > >> > that
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > so-called
> > > > > > > >> > > > > >> > > > > > > "group
> > > > > > > >> > > > > >> > > > > > > > > > > configurations" are overrides
> > of the
> > > > > > > >> > cluster-level
> > > > > > > >> > > > > >> > > > > configurations
> > > > > > > >> > > > > >> > > > > > > > > > > customized per group so when an
> > > > > admin list
> > > > > > > >> > cluster
> > > > > > > >> > > > > >> > configs
> > > > > > > >> > > > > >> > > it's
> > > > > > > >> > > > > >> > > > > > > okay to
> > > > > > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> > > > > "defaults", not
> > > > > > > >> > showing
> > > > > > > >> > > > > >> any
> > > > > > > >> > > > > >> > > > > per-group
> > > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone
> > configure a
> > > > > group
> > > > > > > >> > before
> > > > > > > >> > > > > >> starting
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > > >> > > > > >> > > > > > > > > > > > we considered allowing the
> > explicit
> > > > > > > >> > creation of
> > > > > > > >> > > > > >> groups?
> > > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > > >> > > > > >> > > > > > > > > > > > configs could be decoupled
> > from
> > > > > the group
> > > > > > > >> > > > > lifecycle.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > The configs can be created
> > before
> > > > > the group
> > > > > > > >> > itself
> > > > > > > >> > > > > as
> > > > > > > >> > > > > >> an
> > > > > > > >> > > > > >> > > > > > > independent
> > > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > > >> > > > > >> > > > > > > > > > > --- of course, this requires
> > the
> > > > > > > >> corresponding
> > > > > > > >> > > > > >> request to
> > > > > > > >> > > > > >> > > be
> > > > > > > >> > > > > >> > > > > > > routed to
> > > > > > > >> > > > > >> > > > > > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > > right coordinator based on the
> > group
> > > > > id ---
> > > > > > > >> > the only
> > > > > > > >> > > > > >> > thing
> > > > > > > >> > > > > >> > > that
> > > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > > >> > > > > >> > > > > > > > > > > when the group itself is gone
> > we
> > > > > also check
> > > > > > > >> > if there
> > > > > > > >> > > > > >> are
> > > > > > > >> > > > > >> > > any
> > > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > > >> > > > > >> > > > > > > > > > > entities related to that group
> > and
> > > > > delete
> > > > > > > >> as
> > > > > > > >> > well.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed
> > introduces an
> > > > > > > >> > asymmetry on
> > > > > > > >> > > > > the
> > > > > > > >> > > > > >> > > creation
> > > > > > > >> > > > > >> > > > > /
> > > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the config
> > entities,
> > > > > and we
> > > > > > > >> > would like
> > > > > > > >> > > > > >> to
> > > > > > > >> > > > > >> > > hear
> > > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > > >> > > > > >> > > > > > > > > > > feelings whether we should aim
> > for
> > > > > symmetry
> > > > > > > >> > i.e.
> > > > > > > >> > > > > >> totally
> > > > > > > >> > > > > >> > > > > decouple
> > > > > > > >> > > > > >> > > > > > > group
> > > > > > > >> > > > > >> > > > > > > > > > > configs and hence not delete
> > them at
> > > > > all
> > > > > > > >> when
> > > > > > > >> > the
> > > > > > > >> > > > > >> group
> > > > > > > >> > > > > >> > is
> > > > > > > >> > > > > >> > > > > gone,
> > > > > > > >> > > > > >> > > > > > > but
> > > > > > > >> > > > > >> > > > > > > > > always
> > > > > > > >> > > > > >> > > > > > > > > > > require explicit deletion
> > operations
> > > > > by
> > > > > > > >> > themselves.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> > Consumer.subscribe
> > > > > method for
> > > > > > > >> > the Java
> > > > > > > >> > > > > >> > client
> > > > > > > >> > > > > >> > > > > still
> > > > > > > >> > > > > >> > > > > > > take
> > > > > > > >> > > > > >> > > > > > > > > a
> > > > > > > >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of
> > do we
> > > > > have
> > > > > > > >> to
> > > > > > > >> > > > > >> introduce an
> > > > > > > >> > > > > >> > > > > overload?
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > I think we do not need to
> > introduce
> > > > > an
> > > > > > > >> > overload, but
> > > > > > > >> > > > > >> I'm
> > > > > > > >> > > > > >> > > all
> > > > > > > >> > > > > >> > > > > ears
> > > > > > > >> > > > > >> > > > > > > if
> > > > > > > >> > > > > >> > > > > > > > > there
> > > > > > > >> > > > > >> > > > > > > > > > > may be some compatibility
> > issues
> > > > > that we
> > > > > > > >> may
> > > > > > > >> > > > > overlook.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine
> > that we
> > > > > should
> > > > > > > >> be
> > > > > > > >> > clearer
> > > > > > > >> > > > > >> > about
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > > > reason
> > > > > > > >> > > > > >> > > > > > > > > to
> > > > > > > >> > > > > >> > > > > > > > > > > > switch to
> > IBP/metadata.version
> > > > > from the
> > > > > > > >> > feature
> > > > > > > >> > > > > >> flag.
> > > > > > > >> > > > > >> > > Maybe
> > > > > > > >> > > > > >> > > > > we
> > > > > > > >> > > > > >> > > > > > > mean
> > > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > > >> > > > > >> > > > > > > > > > > > can switch the default for
> > the
> > > > > feature
> > > > > > > >> flag
> > > > > > > >> > to
> > > > > > > >> > > > > true
> > > > > > > >> > > > > >> > > based on
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > > > metadata.version once we
> > want to
> > > > > make it
> > > > > > > >> the
> > > > > > > >> > > > > >> default.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs
> > don't
> > > > > mention
> > > > > > > >> > the
> > > > > > > >> > > > > >> required
> > > > > > > >> > > > > >> > > ACLs,
> > > > > > > >> > > > > >> > > > > it
> > > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > > >> > > > > >> > > > > > > > > > > > good to add that for
> > consistency.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > > > > > >> > ConsumerGroupHeartbeat
> > > > > > > >> > > > > >> > requires
> > > > > > > >> > > > > >> > > > > "Read
> > > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > > >> > > > > >> > > > > > > > > > > > though it seems to do more
> > than
> > > > > reading.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > I had that thought myself as
> > well,
> > > > > but in
> > > > > > > >> the
> > > > > > > >> > end we
> > > > > > > >> > > > > >> > could
> > > > > > > >> > > > > >> > > not
> > > > > > > >> > > > > >> > > > > > > find a
> > > > > > > >> > > > > >> > > > > > > > > > > better alternative: adding
> > Write
> > > > > Group
> > > > > > > >> seems
> > > > > > > >> > an
> > > > > > > >> > > > > >> overkill
> > > > > > > >> > > > > >> > > here
> > > > > > > >> > > > > >> > > > > > > since we
> > > > > > > >> > > > > >> > > > > > > > > do
> > > > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we only
> > have
> > > > > Read /
> > > > > > > >> > Delete
> > > > > > > >> > > > > and
> > > > > > > >> > > > > >> > > Describe
> > > > > > > >> > > > > >> > > > > on
> > > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > > >> > > > > >> > > > > > > > > > > far). Would like to hear others
> > > > > thoughts.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic recreation
> > > > > handled by
> > > > > > > >> the
> > > > > > > >> > > > > consumer
> > > > > > > >> > > > > >> > with
> > > > > > > >> > > > > >> > > the
> > > > > > > >> > > > > >> > > > > new
> > > > > > > >> > > > > >> > > > > > > > > group
> > > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be good
> > to have
> > > > > a
> > > > > > > >> > section on
> > > > > > > >> > > > > >> this.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> > subscription
> > > > > right? Yes
> > > > > > > >> > we can
> > > > > > > >> > > > > >> add a
> > > > > > > >> > > > > >> > > > > section
> > > > > > > >> > > > > >> > > > > > > about
> > > > > > > >> > > > > >> > > > > > > > > > > that, but basically the idea
> > is that
> > > > > > > >> consumer
> > > > > > > >> > would
> > > > > > > >> > > > > be
> > > > > > > >> > > > > >> > > totally
> > > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > > >> > > > > >> > > > > > > > > > > the new protocol as it's
> > handled all
> > > > > by the
> > > > > > > >> > brokers.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will
> > write
> > > > > the
> > > > > > > >> new
> > > > > > > >> > > > > >> coordinator
> > > > > > > >> > > > > >> > in
> > > > > > > >> > > > > >> > > > > Java.
> > > > > > > >> > > > > >> > > > > > > Even
> > > > > > > >> > > > > >> > > > > > > > > though
> > > > > > > >> > > > > >> > > > > > > > > > > > this is an implementation
> > detail,
> > > > > do we
> > > > > > > >> > plan to
> > > > > > > >> > > > > >> have a
> > > > > > > >> > > > > >> > > new
> > > > > > > >> > > > > >> > > > > gradle
> > > > > > > >> > > > > >> > > > > > > > > module
> > > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > We have not thought about
> > that. But
> > > > > I think
> > > > > > > >> > the
> > > > > > > >> > > > > answer
> > > > > > > >> > > > > >> > > should
> > > > > > > >> > > > > >> > > > > be
> > > > > > > >> > > > > >> > > > > > > yes.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability
> > goal
> > > > > when it
> > > > > > > >> > comes to
> > > > > > > >> > > > > >> how
> > > > > > > >> > > > > >> > > many
> > > > > > > >> > > > > >> > > > > > > members
> > > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > > >> > > > > >> > > > > > > > > > > > group protocol can support?
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > Within a group, I think we
> > should
> > > > > shoot for
> > > > > > > >> > 1000s of
> > > > > > > >> > > > > >> > > members.
> > > > > > > >> > > > > >> > > > > But
> > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > >> > > > > > > > > > > scalability goals also depend
> > on the
> > > > > offset
> > > > > > > >> > > > > management
> > > > > > > >> > > > > >> > > (commit,
> > > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > > >> > > > > >> > > > > > > > > > > capabilities of the coordinator
> > > > > which we
> > > > > > > >> did
> > > > > > > >> > not
> > > > > > > >> > > > > >> cover in
> > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > > >> > > > > >> > > > > > > > > > > hard to give a number that
> > applies
> > > > > > > >> > universally.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> > > > > > > >> > SubscribedTopidIds
> > > > > > > >> > > > > >> instead
> > > > > > > >> > > > > >> > of
> > > > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > > >> > > > > >> > > Is the
> > > > > > > >> > > > > >> > > > > > > idea
> > > > > > > >> > > > > >> > > > > > > > > that
> > > > > > > >> > > > > >> > > > > > > > > > > > since we have to resolve the
> > regex
> > > > > on the
> > > > > > > >> > server,
> > > > > > > >> > > > > we
> > > > > > > >> > > > > >> > can
> > > > > > > >> > > > > >> > > do
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > same
> > > > > > > >> > > > > >> > > > > > > > > for
> > > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> > difference is
> > > > > that
> > > > > > > >> > sending the
> > > > > > > >> > > > > >> > regex
> > > > > > > >> > > > > >> > > is
> > > > > > > >> > > > > >> > > > > more
> > > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > > >> > > > > >> > > > > > > > > > > > whereas sending the topic
> > names is
> > > > > less
> > > > > > > >> > efficient.
> > > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > > >> > > > > >> > > > > > > > > > > > recreation is easier to
> > handle if
> > > > > we have
> > > > > > > >> > topic
> > > > > > > >> > > > > ids.
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > The main reason to still let
> > the
> > > > > clients
> > > > > > > >> send
> > > > > > > >> > names
> > > > > > > >> > > > > >> is to
> > > > > > > >> > > > > >> > > keep
> > > > > > > >> > > > > >> > > > > the
> > > > > > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids on
> > the
> > > > > broker /
> > > > > > > >> > admin
> > > > > > > >> > > > > client
> > > > > > > >> > > > > >> > > only.
> > > > > > > >> > > > > >> > > > > Note
> > > > > > > >> > > > > >> > > > > > > that
> > > > > > > >> > > > > >> > > > > > > > > > > although we added topic id in
> > > > > KIP-516, we
> > > > > > > >> > never
> > > > > > > >> > > > > >> > > implemented the
> > > > > > > >> > > > > >> > > > > > > logic
> > > > > > > >> > > > > >> > > > > > > > > on
> > > > > > > >> > > > > >> > > > > > > > > > > consumer/producers leveraging
> > the
> > > > > related
> > > > > > > >> > newer
> > > > > > > >> > > > > >> versioned
> > > > > > > >> > > > > >> > > RPCs,
> > > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > > >> > > > > >> > > > > > > > > > > just set the topic id as empty
> > UUID.
> > > > > We
> > > > > > > >> want
> > > > > > > >> > to keep
> > > > > > > >> > > > > >> the
> > > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > > >> > > > > >> > > > > > > > > > > to be thin and only delegate
> > the
> > > > > reasoning
> > > > > > > >> on
> > > > > > > >> > broker
> > > > > > > >> > > > > >> and
> > > > > > > >> > > > > >> > > > > > > potentially
> > > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45
> > AM
> > > > > David
> > > > > > > >> Jacot
> > > > > > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> > > > > discussion
> > > > > > > >> thread
> > > > > > > >> > on
> > > > > > > >> > > > > >> KIP-848:
> > > > > > > >> > > > > >> > > The
> > > > > > > >> > > > > >> > > > > Next
> > > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the Consumer
> > > > > Rebalance
> > > > > > > >> > Protocol.
> > > > > > > >> > > > > >> With
> > > > > > > >> > > > > >> > > this
> > > > > > > >> > > > > >> > > > > KIP,
> > > > > > > >> > > > > >> > > > > > > we
> > > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > > >> > > > > >> > > > > > > > > > > > > to make the rebalance
> > protocol
> > > > > (for
> > > > > > > >> > consumers)
> > > > > > > >> > > > > >> more
> > > > > > > >> > > > > >> > > > > reliable,
> > > > > > > >> > > > > >> > > > > > > more
> > > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to
> > implement for
> > > > > > > >> > clients, and
> > > > > > > >> > > > > >> easier
> > > > > > > >> > > > > >> > > to
> > > > > > > >> > > > > >> > > > > debug
> > > > > > > >> > > > > >> > > > > > > for
> > > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > > >> > > > > >> > > > >
> > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > > Please take a look and let
> > me
> > > > > know what
> > > > > > > >> > you
> > > > > > > >> > > > > think.
> > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from
> > July
> > > > > 18th to
> > > > > > > >> > August 8th.
> > > > > > > >> > > > > >> That
> > > > > > > >> > > > > >> > > gives
> > > > > > > >> > > > > >> > > > > > > you a
> > > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > > >> > > > > >> > > > > > > > > > > > > of time to read and digest
> > this
> > > > > long
> > > > > > > >> KIP.
> > > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > > >> > > > > >> > > > > > > > >
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > >
> > > > > > > >> > > > > >> > > > > > > > --
> > > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > > >> > > > > >> > > > > > >
> > > > > > > >> > > > > >> > > > >
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> > >
> > > > > > > >> > > > > >> >
> > > > > > > >> > > > > >>
> > > > > > > >> > > > > >>
> > > > > > > >> > > > > >> --
> > > > > > > >> > > > > >> -- Guozhang
> > > > > > > >> > > > > >>
> > > > > > > >> > > > > >
> > > > > > > >> > > > >
> > > > > > > >> >
> > > > > > > >>
> > > > > > > >
> > > > >
> >
>
>
> --
> -- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Guozhang Wang <wa...@gmail.com>]

Hello David,

One of Jun's comments make me thinking:

```
In this case, a new assignment is triggered by the client side
assignor. When constructing the HB, the consumer will always consult
the client side assignor and propagate the information to the group
coordinator. In other words, we don't expect users to call
Consumer#enforceRebalance anymore.
```

As I looked at the current PartitionAssignor's interface, we actually do
not have a way yet to instruct how to construct the next HB request, e.g.
when the assignor wants to enforce a new rebalance with a new assignment,
we'd need some customizable APIs inside the PartitionAssignor to indicate
the next HB telling broker about so. WDYT about adding such an API on the
PartitionAssignor?


Guozhang


On Tue, Sep 6, 2022 at 6:09 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Jun,
>
> I have updated the KIP to include your feedback. I have also tried to
> clarify the parts which were not cleared.
>
> Best,
> David
>
> On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io> wrote:
> >
> > Hi Jun,
> >
> > Thanks for your feedback. Let me start by answering your questions
> > inline and I will update the KIP next week.
> >
> > > Thanks for the KIP. Overall, the main benefits of the KIP seem to be
> fewer
> > > RPCs during rebalance and more efficient support of wildcard. A few
> > > comments below.
> >
> > I would also add that the KIP removes the global sync barrier in the
> > protocol which is essential to improve group stability and
> > scalability, and the KIP also simplifies the client by moving most of
> > the logic to the server side.
> >
> > > 30. ConsumerGroupHeartbeatRequest
> > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> changing
> > > of the partition assignor in the consumers?
> >
> > Definitely. The group coordinator will use the assignor used by a
> > majority of the members. This allows the group to move from one
> > assignor to another by a roll. This is explained in the Assignor
> > Selection chapter.
> >
> > > 30.2 For each field, could you explain whether it's required in every
> > > request or the scenarios when it needs to be filled? For example, it's
> not
> > > clear to me when TopicPartitions needs to be filled.
> >
> > The client is expected to set those fields in case of a connection
> > issue (e.g. timeout) or when the fields have changed since the last
> > HB. The server populates those fields as long as the member is not
> > fully reconciled - the member should acknowledge that it has the
> > expected epoch and assignment. I will clarify this in the KIP.
> >
> > > 31. In the current consumer protocol, the rack affinity between the
> client
> > > and the broker is only considered during fetching, but not during
> assigning
> > > partitions to consumers. Sometimes, once the assignment is made, there
> is
> > > no opportunity for read affinity because no replicas of assigned
> partitions
> > > are close to the member. I am wondering if we should use this
> opportunity
> > > to address this by including rack in GroupMember.
> >
> > That's an interesting idea. I don't see any issue with adding the rack
> > to the members. I will do so.
> >
> > > 32. On the metric side, often, it's useful to know how busy a group
> > > coordinator is. By moving the event loop model, it seems that we could
> add
> > > a metric that tracks the fraction of the time the event loop is doing
> the
> > > actual work.
> >
> > That's a great idea. I will add it. Thanks.
> >
> > > 33. Could we add a section on coordinator failover handling? For
> example,
> > > does it need to trigger the check if any group with the wildcard
> > > subscription now has a new matching topic?
> >
> > Sure. When the new group coordinator takes over, it has to:
> > * Setup the session timeouts.
> > * Trigger a new assignment if a client side assignor is used. We don't
> > store the information about the member selected to run the assignment
> > so we have to start a new one.
> > * Update the topics metadata, verify the wildcard subscriptions, and
> > trigger a rebalance if needed.
> >
> > > 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> field
> > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> reflect
> > > the latest group epoch? What about the one in
> > > ConsumerGroupPartitionMetadataValue and
> ConsumerGroupMemberMetadataValue?
> >
> > Sure. I will clarify that but it is always the latest group epoch.
> > When the group state is updated, the group epoch is bumped so we use
> > that one for all the change records related to the update.
> >
> > > 35. "the group coordinator will ensure that the following invariants
> are
> > > met: ... All members exists." It's possible for a member not to get any
> > > assigned partitions, right?
> >
> > That's right. Here I meant that the members provided by the assignor
> > in the assignment must exist in the group. The assignor can not make
> > up new member ids.
> >
> > > 36. "He can rejoins the group with a member epoch equals to 0": When
> would
> > > a consumer rejoin and what member id would be used?
> >
> > A member is expected to abandon all its partitions and rejoins when it
> > receives the FENCED_MEMBER_EPOCH error. In this case, the group
> > coordinator will have removed the member from the group. The member
> > can rejoin the group with the same member id but with 0 as epoch. Let
> > me see if I can clarify this in the KIP.
> >
> > > 37. "Instead, power users will have the ability to trigger a
> reassignment
> > > by either providing a non-zero reason or by updating the assignor
> > > metadata." Hmm, this seems to be conflicting with the deprecation of
> > > Consumer#enforeRebalance.
> >
> > In this case, a new assignment is triggered by the client side
> > assignor. When constructing the HB, the consumer will always consult
> > the client side assignor and propagate the information to the group
> > coordinator. In other words, we don't expect users to call
> > Consumer#enforceRebalance anymore.
> >
> > > 38. The reassignment examples are nice. But the section seems to have
> > > multiple typos.
> > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > 38.2 When the group transitions to epoch 3, C seems to get into
> epoch=3,
> > > partitions=[foo-1] too early.
> > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > partitions=[foo-0].
> >
> > Sorry for that! I will revise them.
> >
> > > 39. Rolling upgrade of consumers: Do we support the upgrade from any
> old
> > > version to new one?
> >
> > We will support upgrading from the consumer protocol version 3,
> > introduced in KIP-792. KIP-792 is not implemented yet so the earliest
> > version is unknown at the moment. This is explained in the migration
> > plan chapter.
> >
> > Thanks again for your feedback, Jun. I will update the KIP based on it
> > next week.
> >
> > Best,
> > David
> >
> > On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <ju...@confluent.io.invalid> wrote:
> > >
> > > Hi, David,
> > >
> > > Thanks for the KIP. Overall, the main benefits of the KIP seem to be
> fewer
> > > RPCs during rebalance and more efficient support of wildcard. A few
> > > comments below.
> > >
> > > 30. ConsumerGroupHeartbeatRequest
> > > 30.1 ServerAssignor is a singleton. Do we plan to support rolling
> changing
> > > of the partition assignor in the consumers?
> > > 30.2 For each field, could you explain whether it's required in every
> > > request or the scenarios when it needs to be filled? For example, it's
> not
> > > clear to me when TopicPartitions needs to be filled.
> > >
> > > 31. In the current consumer protocol, the rack affinity between the
> client
> > > and the broker is only considered during fetching, but not during
> assigning
> > > partitions to consumers. Sometimes, once the assignment is made, there
> is
> > > no opportunity for read affinity because no replicas of assigned
> partitions
> > > are close to the member. I am wondering if we should use this
> opportunity
> > > to address this by including rack in GroupMember.
> > >
> > > 32. On the metric side, often, it's useful to know how busy a group
> > > coordinator is. By moving the event loop model, it seems that we could
> add
> > > a metric that tracks the fraction of the time the event loop is doing
> the
> > > actual work.
> > >
> > > 33. Could we add a section on coordinator failover handling? For
> example,
> > > does it need to trigger the check if any group with the wildcard
> > > subscription now has a new matching topic?
> > >
> > > 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> > > ConsumerGroupMemberMetadataValue: Could we document what the epoch
> field
> > > reflects? For example, does the epoch in ConsumerGroupMetadataValue
> reflect
> > > the latest group epoch? What about the one in
> > > ConsumerGroupPartitionMetadataValue and
> ConsumerGroupMemberMetadataValue?
> > >
> > > 35. "the group coordinator will ensure that the following invariants
> are
> > > met: ... All members exists." It's possible for a member not to get any
> > > assigned partitions, right?
> > >
> > > 36. "He can rejoins the group with a member epoch equals to 0": When
> would
> > > a consumer rejoin and what member id would be used?
> > >
> > > 37. "Instead, power users will have the ability to trigger a
> reassignment
> > > by either providing a non-zero reason or by updating the assignor
> > > metadata." Hmm, this seems to be conflicting with the deprecation of
> > > Consumer#enforeRebalance.
> > >
> > > 38. The reassignment examples are nice. But the section seems to have
> > > multiple typos.
> > > 38.1 When the group transitions to epoch 2, B immediately gets into
> > > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > > 38.2 When the group transitions to epoch 3, C seems to get into
> epoch=3,
> > > partitions=[foo-1] too early.
> > > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > > partitions=[foo-0].
> > >
> > > 39. Rolling upgrade of consumers: Do we support the upgrade from any
> old
> > > version to new one?
> > >
> > > Thanks,
> > >
> > > Jun
> > >
> > > On Mon, Aug 29, 2022 at 9:20 AM David Jacot
> <dj...@confluent.io.invalid>
> > > wrote:
> > >
> > > > Hi all,
> > > >
> > > > The KIP states that we will re-implement the coordinator in Java. I
> > > > discussed this offline with a few folks and folks are concerned that
> > > > we could introduce many regressions in the old protocol if we do so.
> > > > Therefore, I am going to remove this statement from the KIP. It is an
> > > > implementation detail after all so it does not have to be decided at
> > > > this stage. We will likely start by trying to refactor the current
> > > > implementation as a first step.
> > > >
> > > > Cheers,
> > > > David
> > > >
> > > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <dj...@confluent.io>
> wrote:
> > > > >
> > > > > Hi Luke,
> > > > >
> > > > > > 1.1. I think the state machine are: "Empty, assigning,
> reconciling,
> > > > stable,
> > > > > > dead" mentioned in Consumer Group States section, right?
> > > > >
> > > > > This sentence does not refer to those group states but rather to a
> > > > > state machine replication (SMR). This refers to the entire state of
> > > > > group coordinator which is replicated via the log layer. I will
> > > > > clarify this in the KIP.
> > > > >
> > > > > > 1.2. What do you mean "each state machine is modelled as an event
> > > > loop"?
> > > > >
> > > > > The idea is to follow a model similar to the new quorum
> controller. We
> > > > > will have N threads to process events. Each __consumer_offsets
> > > > > partition is assigned to a unique thread and all the events (e.g.
> > > > > requests, callbacks, etc.) are processed by this thread. This
> simplify
> > > > > concurrency and will enable us to do simulation testing for the
> group
> > > > > coordinator.
> > > > >
> > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > partitions?
> > > > > > Not a state machine "per consumer group" owned by a group
> coordinator?
> > > > For
> > > > > > example, if one group coordinator owns 2 consumer groups, and
> both
> > > > exist in
> > > > > > *__consumer_offsets-0*, will we have 1 state machine for it, or
> 2?
> > > > >
> > > > > See 1.1. The confusion comes from there, I think.
> > > > >
> > > > > > 1.4. I know the "*group.coordinator.threads" *is the number of
> threads
> > > > used
> > > > > > to run the state machines. But I'm wondering if the purpose of
> the
> > > > threads
> > > > > > is only to keep the state of each consumer group (or
> > > > *__consumer_offsets*
> > > > > > partitions?), and no heavy computation, why should we need
> > > > multi-threads
> > > > > > here?
> > > > >
> > > > > See 1.2. The idea is to have an ability to shard the processing as
> the
> > > > > computation could be heavy.
> > > > >
> > > > > > 2.1. The consumer session timeout, why does the default session
> > > > timeout not
> > > > > > locate between min (45s) and max(60s)? I thought the min/max
> session
> > > > > > timeout is to define lower/upper bound of it, no?
> > > > > >
> > > > > > group.consumer.session.timeout.ms int 30s The timeout to detect
> client
> > > > > > failures when using the consumer group protocol.
> > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> session
> > > > timeout.
> > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> session
> > > > timeout.
> > > > >
> > > > > This is indeed a mistake. The default session timeout should be 45s
> > > > > (the current default).
> > > > >
> > > > > > 2.2. The default server side assignor are [range, uniform],
> which means
> > > > > > we'll default to "range" assignor. I'd like to know why not
> uniform
> > > > one? I
> > > > > > thought usually users will choose uniform assignor (former sticky
> > > > assinor)
> > > > > > for better evenly distribution. Any other reason we choose range
> > > > assignor
> > > > > > as default?
> > > > > > group.consumer.assignors List range, uniform The server side
> assignors.
> > > > >
> > > > > The order on the server side has no influence because the client
> must
> > > > > chose the selector that he wants to use. There is no default in the
> > > > > current proposal. If the assignor is not specified by the client,
> the
> > > > > request is rejected. The default client value for
> > > > > `group.remote.assignor` is `uniform` though.
> > > > >
> > > > > Thanks for your very good comments, Luke. I hope that my answers
> help
> > > > > to clarify things. I will update the KIP as well based on your
> > > > > feedback.
> > > > >
> > > > > Cheers,
> > > > > David
> > > > >
> > > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com>
> wrote:
> > > > > >
> > > > > > Hi David,
> > > > > >
> > > > > > Thanks for the update.
> > > > > >
> > > > > > Some more questions:
> > > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > > The new group coordinator will have a state machine per
> > > > > > *__consumer_offsets* partitions, where each state machine is
> modelled
> > > > as an
> > > > > > event loop. Those state machines will be executed in
> > > > > > *group.coordinator.threads* threads.
> > > > > >
> > > > > > 1.1. I think the state machine are: "Empty, assigning,
> reconciling,
> > > > stable,
> > > > > > dead" mentioned in Consumer Group States section, right?
> > > > > > 1.2. What do you mean "each state machine is modelled as an event
> > > > loop"?
> > > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > > partitions?
> > > > > > Not a state machine "per consumer group" owned by a group
> coordinator?
> > > > For
> > > > > > example, if one group coordinator owns 2 consumer groups, and
> both
> > > > exist in
> > > > > > *__consumer_offsets-0*, will we have 1 state machine for it, or
> 2?
> > > > > > 1.4. I know the "*group.coordinator.threads" *is the number of
> threads
> > > > used
> > > > > > to run the state machines. But I'm wondering if the purpose of
> the
> > > > threads
> > > > > > is only to keep the state of each consumer group (or
> > > > *__consumer_offsets*
> > > > > > partitions?), and no heavy computation, why should we need
> > > > multi-threads
> > > > > > here?
> > > > > >
> > > > > > 2. For the default value in the new configs:
> > > > > > 2.1. The consumer session timeout, why does the default session
> > > > timeout not
> > > > > > locate between min (45s) and max(60s)? I thought the min/max
> session
> > > > > > timeout is to define lower/upper bound of it, no?
> > > > > >
> > > > > > group.consumer.session.timeout.ms int 30s The timeout to detect
> client
> > > > > > failures when using the consumer group protocol.
> > > > > > group.consumer.min.session.timeout.ms int 45s The minimum
> session
> > > > timeout.
> > > > > > group.consumer.max.session.timeout.ms int 60s The maximum
> session
> > > > timeout.
> > > > > >
> > > > > >
> > > > > >
> > > > > > 2.2. The default server side assignor are [range, uniform],
> which means
> > > > > > we'll default to "range" assignor. I'd like to know why not
> uniform
> > > > one? I
> > > > > > thought usually users will choose uniform assignor (former sticky
> > > > assinor)
> > > > > > for better evenly distribution. Any other reason we choose range
> > > > assignor
> > > > > > as default?
> > > > > > group.consumer.assignors List range, uniform The server side
> assignors.
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > Thank you.
> > > > > > Luke
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com>
> wrote:
> > > > > >
> > > > > > > Hi Sagar,
> > > > > > >
> > > > > > > I have some thoughts about Kafka Connect integrating with
> KIP-848,
> > > > but I
> > > > > > > think we should have a separate discussion thread for the Kafka
> > > > Connect
> > > > > > > KIP: Integrating Kafka Connect With New Consumer Rebalance
> Protocol
> > > > [1],
> > > > > > > and let this discussion thread focus on consumer rebalance
> protocol,
> > > > WDYT?
> > > > > > >
> > > > > > > [1]
> > > > > > >
> > > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >
> > > > > > > Thank you.
> > > > > > > Luke
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <
> sagarmeansocean@gmail.com>
> > > > wrote:
> > > > > > >
> > > > > > >> Thank you Guozhang/David for the feedback. Looks like there's
> > > > agreement on
> > > > > > >> using separate APIs for Connect. I would revisit the doc and
> see
> > > > what
> > > > > > >> changes are to be made.
> > > > > > >>
> > > > > > >> Thanks!
> > > > > > >> Sagar.
> > > > > > >>
> > > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > > <dj...@confluent.io.invalid>
> > > > > > >> wrote:
> > > > > > >>
> > > > > > >> > Hi Sagar,
> > > > > > >> >
> > > > > > >> > Thanks for the feedback and the document. That's really
> helpful. I
> > > > > > >> > will take a look at it.
> > > > > > >> >
> > > > > > >> > Overall, it seems to me that both Connect and the Consumer
> could
> > > > share
> > > > > > >> > the same underlying "engine". The main difference is that
> the
> > > > Consumer
> > > > > > >> > assigns topic-partitions to members whereas Connect assigns
> tasks
> > > > to
> > > > > > >> > workers. I see two ways to move forward:
> > > > > > >> > 1) We extend the new proposed APIs to support different
> resource
> > > > types
> > > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > > >> > 2) We use new dedicated APIs for Connect. The dedicated APIs
> > > > would be
> > > > > > >> > similar to the new ones but different on the
> content/resources and
> > > > > > >> > they would rely on the same engine on the coordinator side.
> > > > > > >> >
> > > > > > >> > I personally lean towards 2) because I am not a fan of
> > > > overcharging
> > > > > > >> > APIs to serve different purposes. That being said, I am not
> > > > opposed to
> > > > > > >> > 1) if we can find an elegant way to do it.
> > > > > > >> >
> > > > > > >> > I think that we can continue to discuss it here for now in
> order
> > > > to
> > > > > > >> > ensure that this KIP is compatible with what we will do for
> > > > Connect in
> > > > > > >> > the future.
> > > > > > >> >
> > > > > > >> > Best,
> > > > > > >> > David
> > > > > > >> >
> > > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <
> djacot@confluent.io>
> > > > wrote:
> > > > > > >> > >
> > > > > > >> > > Hi all,
> > > > > > >> > >
> > > > > > >> > > I am back from vacation. I will go through and address
> your
> > > > comments
> > > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > > >> > >
> > > > > > >> > > Cheers,
> > > > > > >> > > David
> > > > > > >> > >
> > > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > > gharris1727@gmail.com
> > > > > > >> >
> > > > > > >> > wrote:
> > > > > > >> > > >
> > > > > > >> > > > Hey All!
> > > > > > >> > > >
> > > > > > >> > > > Thanks for the KIP, it's wonderful to see cooperative
> > > > rebalancing
> > > > > > >> > making it
> > > > > > >> > > > down the stack!
> > > > > > >> > > >
> > > > > > >> > > > I had a few questions:
> > > > > > >> > > >
> > > > > > >> > > > 1. The 'Rejected Alternatives' section describes how
> member
> > > > epoch
> > > > > > >> > should
> > > > > > >> > > > advance in step with the group epoch and assignment
> epoch
> > > > values. I
> > > > > > >> > think
> > > > > > >> > > > that this is a good idea for the reasons described in
> the
> > > > KIP. When
> > > > > > >> the
> > > > > > >> > > > protocol is incrementally assigning partitions to a
> worker,
> > > > what
> > > > > > >> member
> > > > > > >> > > > epoch does each incremental assignment use? Are member
> epochs
> > > > > > >> re-used,
> > > > > > >> > and
> > > > > > >> > > > a single member epoch can correspond to multiple
> different
> > > > > > >> > (monotonically
> > > > > > >> > > > larger) assignments?
> > > > > > >> > > >
> > > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> > > > > > >> coordinator? If
> > > > > > >> > > > not, should custom client-side assignor implementations
> > > > interact
> > > > > > >> with
> > > > > > >> > the
> > > > > > >> > > > Reason field, and how is its common meaning agreed
> upon? If
> > > > so, what
> > > > > > >> > is the
> > > > > > >> > > > benefit of a distinct Reason field over including such
> > > > functionality
> > > > > > >> > in the
> > > > > > >> > > > opaque metadata?
> > > > > > >> > > >
> > > > > > >> > > > 3. The following is included in the KIP: "Thanks to
> this, the
> > > > input
> > > > > > >> of
> > > > > > >> > the
> > > > > > >> > > > client side assignor is entirely driven by the group
> > > > coordinator.
> > > > > > >> The
> > > > > > >> > > > consumer is no longer responsible for maintaining any
> state
> > > > besides
> > > > > > >> its
> > > > > > >> > > > assigned partitions." Does this mean that the
> client-side
> > > > assignor
> > > > > > >> MAY
> > > > > > >> > > > incorporate additional non-Metadata state (such as
> partition
> > > > > > >> > throughput,
> > > > > > >> > > > cpu/memory metrics, config topics, etc), or that
> additional
> > > > > > >> > non-Metadata
> > > > > > >> > > > state SHOULD NOT be used?
> > > > > > >> > > >
> > > > > > >> > > > 4. I see that there are separate classes
> > > > > > >> > > > for
> org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > > >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor
> that
> > > > seem to
> > > > > > >> > > > overlap significantly. Is it possible for these two
> > > > implementations
> > > > > > >> to
> > > > > > >> > be
> > > > > > >> > > > unified? This would serve to promote feature parity of
> > > > server-side
> > > > > > >> and
> > > > > > >> > > > client-side assignors, and would also facilitate
> operational
> > > > > > >> > flexibility in
> > > > > > >> > > > certain situations. For example, if a server-side
> assignor
> > > > has some
> > > > > > >> > poor
> > > > > > >> > > > behavior and needs a patch, deploying the patched
> assignor to
> > > > the
> > > > > > >> > client
> > > > > > >> > > > and switching one consumer group to a client-side
> assignor
> > > > may be
> > > > > > >> > faster
> > > > > > >> > > > and less risky than patching all of the brokers. With
> the
> > > > currently
> > > > > > >> > > > proposed distinct APIs, a non-trivial reimplementation
> would
> > > > have
> > > > > > >> to be
> > > > > > >> > > > assembled, and if the two APIs have diverged
> significantly,
> > > > then it
> > > > > > >> is
> > > > > > >> > > > possible that a reimplementation would not be possible.
> > > > > > >> > > >
> > > > > > >> > > > --
> > > > > > >> > > > Greg Harris
> > > > > > >> > > > gharris1727@gmail.com
> > > > > > >> > > > github.com/gharris1727
> > > > > > >> > > >
> > > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > > sagarmeansocean@gmail.com>
> > > > > > >> > wrote:
> > > > > > >> > > >
> > > > > > >> > > > > Hi Guozhang/David,
> > > > > > >> > > > >
> > > > > > >> > > > > I created a confluence page to discuss how Connect
> would
> > > > need to
> > > > > > >> > change
> > > > > > >> > > > > based on the new rebalance protocol. Here's the page:
> > > > > > >> > > > >
> > > > > > >> > > > >
> > > > > > >> >
> > > > > > >>
> > > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > > >> > > > >
> > > > > > >> > > > > It's also pretty longish and I have tried to keep a
> format
> > > > > > >> similar to
> > > > > > >> > > > > KIP-848. Let me know what you think. Also, do you
> think this
> > > > > > >> should
> > > > > > >> > be
> > > > > > >> > > > > moved to a separate discussion thread or is this one
> fine?
> > > > > > >> > > > >
> > > > > > >> > > > > Thanks!
> > > > > > >> > > > > Sagar.
> > > > > > >> > > > >
> > > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > > sagarmeansocean@gmail.com>
> > > > > > >> > wrote:
> > > > > > >> > > > >
> > > > > > >> > > > > > Hello Guozhang,
> > > > > > >> > > > > >
> > > > > > >> > > > > > Thank you so much for the doc on Kafka Streams.
> Sure, I
> > > > would do
> > > > > > >> > the
> > > > > > >> > > > > > analysis and come up with such a document.
> > > > > > >> > > > > >
> > > > > > >> > > > > > Thanks!
> > > > > > >> > > > > > Sagar.
> > > > > > >> > > > > >
> > > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > > > > >> wangguoz@gmail.com>
> > > > > > >> > > > > wrote:
> > > > > > >> > > > > >
> > > > > > >> > > > > >> Hello Sagar,
> > > > > > >> > > > > >>
> > > > > > >> > > > > >> It would be great if you could come back with some
> > > > analysis on
> > > > > > >> > how to
> > > > > > >> > > > > >> implement the Connect side integration with the new
> > > > protocol;
> > > > > > >> so
> > > > > > >> > far
> > > > > > >> > > > > >> besides leveraging on the new "protocol type" we
> did not
> > > > yet
> > > > > > >> think
> > > > > > >> > > > > through
> > > > > > >> > > > > >> the Connect side implementations. For Streams
> here's a
> > > > draft of
> > > > > > >> > > > > >> integration
> > > > > > >> > > > > >> plan:
> > > > > > >> > > > > >>
> > > > > > >> > > > > >>
> > > > > > >> > > > >
> > > > > > >> >
> > > > > > >>
> > > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > > >> > > > > >>
> > > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > > >> sagarmeansocean@gmail.com
> > > > > > >> > >
> > > > > > >> > > > > wrote:
> > > > > > >> > > > > >>
> > > > > > >> > > > > >> > Hi David,
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> > Thank you for your response. The reason I thought
> > > > connect can
> > > > > > >> > also fit
> > > > > > >> > > > > >> into
> > > > > > >> > > > > >> > this new scheme is that even today the connect
> uses a
> > > > > > >> > > > > WorkerCoordinator
> > > > > > >> > > > > >> > extending from AbstractCoordinator to empower
> > > > rebalances of
> > > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> > > > > > >> protocolType()
> > > > > > >> > to
> > > > > > >> > > > > >> connect
> > > > > > >> > > > > >> > and uses the metadata() method by plumbing into
> > > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> > I think the changes to support connect would be
> > > > similar at a
> > > > > > >> > high
> > > > > > >> > > > > level
> > > > > > >> > > > > >> to
> > > > > > >> > > > > >> > the changes in streams mainly because of the
> Client
> > > > side
> > > > > > >> > assignors
> > > > > > >> > > > > being
> > > > > > >> > > > > >> > used in both. At an implementation level, we
> might
> > > > need to
> > > > > > >> make
> > > > > > >> > a lot
> > > > > > >> > > > > of
> > > > > > >> > > > > >> > changes to get onto this new assignment protocol
> like
> > > > > > >> enhancing
> > > > > > >> > the
> > > > > > >> > > > > >> > JoinGroup request/response and SyncGroup and
> using
> > > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > > >> > > > > >> > API etc again on similar lines to streams (or
> there
> > > > might be
> > > > > > >> > > > > >> deviations). I
> > > > > > >> > > > > >> > would try to perform a detailed analysis of the
> same
> > > > and we
> > > > > > >> > can have
> > > > > > >> > > > > a
> > > > > > >> > > > > >> > separate discussion thread for that as that would
> > > > derail this
> > > > > > >> > > > > discussion
> > > > > > >> > > > > >> > thread. Let me know if that sounds good to you.
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> > Thanks!
> > > > > > >> > > > > >> > Sagar.
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> > wrote:
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >> > > Hi Sagar,
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> > > Thanks for your comments.
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`.
> Sure, I
> > > > can
> > > > > > >> > mention it.
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> > > 2) The idea is to transition C from his current
> > > > assignment
> > > > > > >> to
> > > > > > >> > his
> > > > > > >> > > > > >> > > target assignment when he can move to epoch 3.
> When
> > > > that
> > > > > > >> > happens,
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > member assignment is updated and persisted
> with all
> > > > its
> > > > > > >> > assigned
> > > > > > >> > > > > >> > > partitions even if they are not all revoked
> yet. In
> > > > other
> > > > > > >> > words, the
> > > > > > >> > > > > >> > > member assignment becomes the target
> assignment.
> > > > This is
> > > > > > >> > basically
> > > > > > >> > > > > an
> > > > > > >> > > > > >> > > optimization to avoid having to write all the
> > > > changes to
> > > > > > >> the
> > > > > > >> > log.
> > > > > > >> > > > > The
> > > > > > >> > > > > >> > > examples are based on the persisted state so I
> > > > understand
> > > > > > >> the
> > > > > > >> > > > > >> > > confusion. Let me see if I can improve this in
> the
> > > > > > >> > description.
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse the
> protocol
> > > > with a
> > > > > > >> > client side
> > > > > > >> > > > > >> > > assignor if it fits in the protocol. The
> assignment
> > > > is
> > > > > > >> about
> > > > > > >> > > > > >> > > topicid-partitions + metadata, could Connect
> fit
> > > > into this?
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> > > Best,
> > > > > > >> > > > > >> > > David
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > > >> > sagarmeansocean@gmail.com>
> > > > > > >> > > > > >> wrote:
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > Hi David,
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor
> observations:
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > 1) In the Assignment Error section in Client
> Side
> > > > mode
> > > > > > >> > Assignment
> > > > > > >> > > > > >> > > process,
> > > > > > >> > > > > >> > > > you mentioned => `In this case, the client
> side
> > > > assignor
> > > > > > >> can
> > > > > > >> > > > > return
> > > > > > >> > > > > >> an
> > > > > > >> > > > > >> > > > error to the group coordinator`. In this
> case are
> > > > you
> > > > > > >> > referring to
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > Assignor returning an AssignmentError that's
> > > > listed down
> > > > > > >> > towards
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > end?
> > > > > > >> > > > > >> > > > If yes, do you think it would make sense to
> > > > mention this
> > > > > > >> > > > > explicitly
> > > > > > >> > > > > >> > here?
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > 2) In the Case Studies section, I have a
> slight
> > > > > > >> confusion,
> > > > > > >> > not
> > > > > > >> > > > > sure
> > > > > > >> > > > > >> if
> > > > > > >> > > > > >> > > > others have the same. Consider this step:
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > When B heartbeats, the group coordinator
> > > > transitions him
> > > > > > >> to
> > > > > > >> > epoch
> > > > > > >> > > > > 3
> > > > > > >> > > > > >> > > because
> > > > > > >> > > > > >> > > > B has no partitions to revoke. It persists
> the
> > > > change and
> > > > > > >> > reply.
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > > >> > > > > >> > > >       - A
> > > > > > >> > > > > >> > > >       - B
> > > > > > >> > > > > >> > > >       - C
> > > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > > >> > > > > >> > > >    - Member Assignment
> > > > > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0,
> foo-1]
> > > > > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > When C heartbeats, it transitions to epoch 3
> but
> > > > cannot
> > > > > > >> get
> > > > > > >> > foo-1
> > > > > > >> > > > > >> yet.
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > Here,it's mentioned that member C can't get
> the
> > > > foo-1
> > > > > > >> > partition
> > > > > > >> > > > > yet,
> > > > > > >> > > > > >> > but
> > > > > > >> > > > > >> > > > based on the description above, it seems it
> > > > already has
> > > > > > >> it.
> > > > > > >> > Do you
> > > > > > >> > > > > >> > think
> > > > > > >> > > > > >> > > it
> > > > > > >> > > > > >> > > > would be better to remove it and populate it
> only
> > > > when it
> > > > > > >> > actually
> > > > > > >> > > > > >> gets
> > > > > > >> > > > > >> > > it?
> > > > > > >> > > > > >> > > > I see this in a lot of other places, so have
> I
> > > > > > >> understood it
> > > > > > >> > > > > >> > incorrectly
> > > > > > >> > > > > >> > > ?
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > Regarding connect , it might be out of scope
> of
> > > > this
> > > > > > >> > discussion,
> > > > > > >> > > > > but
> > > > > > >> > > > > >> > from
> > > > > > >> > > > > >> > > > what I understood it would probably be
> running in
> > > > client
> > > > > > >> > side
> > > > > > >> > > > > >> assignor
> > > > > > >> > > > > >> > > mode
> > > > > > >> > > > > >> > > > even on the new rebalance protocol as it has
> its
> > > > own
> > > > > > >> Custom
> > > > > > >> > > > > >> > > Assignors(Eager
> > > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > Thanks!
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > Sagar.
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > wrote:
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move forward
> with
> > > > > > >> > specialized API
> > > > > > >> > > > > >> > > > > instead of relying on one generic API. For
> > > > Connect, we
> > > > > > >> > can apply
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > exact same pattern and reuse/share the core
> > > > > > >> > implementation on
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > server side. For the schema registry, I
> think
> > > > that we
> > > > > > >> > should
> > > > > > >> > > > > >> consider
> > > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > > membership/leader
> > > > > > >> > election.
> > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > >> > > > > Best,
> > > > > > >> > > > > >> > > > > David
> > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael
> Juma <
> > > > > > >> > ismael@juma.me.uk
> > > > > > >> > > > > >
> > > > > > >> > > > > >> > > wrote:
> > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > >> > > > > > document the differences in more detail
> before
> > > > > > >> deciding
> > > > > > >> > one
> > > > > > >> > > > > way
> > > > > > >> > > > > >> or
> > > > > > >> > > > > >> > > > > another.
> > > > > > >> > > > > >> > > > > > That said, if people pass
> > > > java.util.regex.Pattern,
> > > > > > >> they
> > > > > > >> > expect
> > > > > > >> > > > > >> > their
> > > > > > >> > > > > >> > > > > > semantics to be honored. If we are doing
> > > > something
> > > > > > >> > different,
> > > > > > >> > > > > >> then
> > > > > > >> > > > > >> > we
> > > > > > >> > > > > >> > > > > > should consider adding an overload with
> our own
> > > > > > >> Pattern
> > > > > > >> > class
> > > > > > >> > > > > (I
> > > > > > >> > > > > >> > > don't
> > > > > > >> > > > > >> > > > > > think we'd want to expose re2j's at this
> > > > point).
> > > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new
> protocol
> > > > should
> > > > > > >> > > > > integrate
> > > > > > >> > > > > >> > fully
> > > > > > >> > > > > >> > > > > with
> > > > > > >> > > > > >> > > > > > it and should handle the topic
> recreation case
> > > > > > >> > correctly.
> > > > > > >> > > > > That's
> > > > > > >> > > > > >> > the
> > > > > > >> > > > > >> > > main
> > > > > > >> > > > > >> > > > > > part we need to handle. I agree with
> David
> > > > that we'd
> > > > > > >> > want to
> > > > > > >> > > > > add
> > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > >> > > > > ids
> > > > > > >> > > > > >> > > > > > to the relevant protocols that don't
> have it
> > > > yet and
> > > > > > >> > that we
> > > > > > >> > > > > can
> > > > > > >> > > > > >> > > probably
> > > > > > >> > > > > >> > > > > > focus on the internals versus adding new
> APIs
> > > > to the
> > > > > > >> > Java
> > > > > > >> > > > > >> Consumer
> > > > > > >> > > > > >> > > > > (unless
> > > > > > >> > > > > >> > > > > > we find that adding new APIs is required
> for
> > > > > > >> reasonable
> > > > > > >> > > > > >> semantics).
> > > > > > >> > > > > >> > > > > > 3. I am still not sure about the
> coordinator
> > > > storing
> > > > > > >> the
> > > > > > >> > > > > >> configs.
> > > > > > >> > > > > >> > > It's
> > > > > > >> > > > > >> > > > > > powerful for configs to be centralized
> in the
> > > > > > >> metadata
> > > > > > >> > log for
> > > > > > >> > > > > >> > > various
> > > > > > >> > > > > >> > > > > > reasons (auditability, visibility,
> consistency,
> > > > > > >> etc.).
> > > > > > >> > > > > >> Similarly, I
> > > > > > >> > > > > >> > > am
> > > > > > >> > > > > >> > > > > not
> > > > > > >> > > > > >> > > > > > sure about automatically deleting
> configs in a
> > > > way
> > > > > > >> that
> > > > > > >> > they
> > > > > > >> > > > > >> cannot
> > > > > > >> > > > > >> > > be
> > > > > > >> > > > > >> > > > > > recovered. A good property for modern
> systems
> > > > is to
> > > > > > >> > minimize
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > number
> > > > > > >> > > > > >> > > > > of
> > > > > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > >> > > > > > Ismael
> > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David
> Jacot
> > > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > >> > > > > > wrote:
> > > > > > >> > > > > >> > > > > >
> > > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> the last
> > > > step
> > > > > > >> when
> > > > > > >> > > > > clients
> > > > > > >> > > > > >> > > flip the
> > > > > > >> > > > > >> > > > > > > flag
> > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in which
> we
> > > > would
> > > > > > >> have a
> > > > > > >> > > > > window
> > > > > > >> > > > > >> > where
> > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > >> > > > > > > new
> > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols /
> rpcs
> > > > are
> > > > > > >> used
> > > > > > >> > by
> > > > > > >> > > > > >> members
> > > > > > >> > > > > >> > of
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> "mimic"
> > > > the old
> > > > > > >> > behavior
> > > > > > >> > > > > >> while
> > > > > > >> > > > > >> > > > > using the
> > > > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > > > present
> > > > > > >> about.
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > Noted. I just published a new version
> of KIP
> > > > which
> > > > > > >> > includes
> > > > > > >> > > > > >> more
> > > > > > >> > > > > >> > > > > > > details about this. See the "Supporting
> > > > Online
> > > > > > >> > Consumer
> > > > > > >> > > > > Group
> > > > > > >> > > > > >> > > Upgrade"
> > > > > > >> > > > > >> > > > > > > and the "Compatibility, Deprecation,
> and
> > > > Migration
> > > > > > >> > Plan". I
> > > > > > >> > > > > >> think
> > > > > > >> > > > > >> > > that
> > > > > > >> > > > > >> > > > > > > I have to think through a few cases
> now but
> > > > the
> > > > > > >> > overall idea
> > > > > > >> > > > > >> and
> > > > > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > > > KIP-516 the
> > > > > > >> > topic ids
> > > > > > >> > > > > >> are
> > > > > > >> > > > > >> > > only
> > > > > > >> > > > > >> > > > > used
> > > > > > >> > > > > >> > > > > > > as
> > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but
> they
> > > > are not
> > > > > > >> > exposed
> > > > > > >> > > > > via
> > > > > > >> > > > > >> any
> > > > > > >> > > > > >> > > > > public
> > > > > > >> > > > > >> > > > > > > APIs
> > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> question is,
> > > > first
> > > > > > >> > should we
> > > > > > >> > > > > >> let
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > consumer
> > > > > > >> > > > > >> > > > > > > > client to be maintaining the names
> -> ids
> > > > mapping
> > > > > > >> > itself
> > > > > > >> > > > > to
> > > > > > >> > > > > >> > fully
> > > > > > >> > > > > >> > > > > > > leverage
> > > > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs
> and the
> > > > new
> > > > > > >> RPCs
> > > > > > >> > with
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > >> > > > > ids;
> > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> exposing
> > > > the
> > > > > > >> > topic ids
> > > > > > >> > > > > in
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > consumer
> > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > subscribe/assign, as
> > > > > > >> well
> > > > > > >> > as in
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > rebalance
> > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > >> > deletion-and-recreation).
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > a) Assuming that we would include
> converting
> > > > all
> > > > > > >> the
> > > > > > >> > offsets
> > > > > > >> > > > > >> > > related
> > > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this KIP,
> the
> > > > consumer
> > > > > > >> > would be
> > > > > > >> > > > > >> able
> > > > > > >> > > > > >> > to
> > > > > > >> > > > > >> > > > > > > fully operate with topic ids. That
> being
> > > > said, it
> > > > > > >> > still has
> > > > > > >> > > > > to
> > > > > > >> > > > > >> > > provide
> > > > > > >> > > > > >> > > > > > > the topics names in various APIs so
> having a
> > > > > > >> mapping
> > > > > > >> > in the
> > > > > > >> > > > > >> > > consumer
> > > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > > >> > > > > >> > > > > > > b) I don't have a strong opinion on
> this.
> > > > Here I
> > > > > > >> > wonder if
> > > > > > >> > > > > >> this
> > > > > > >> > > > > >> > > goes
> > > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I would
> rather
> > > > focus
> > > > > > >> on
> > > > > > >> > the
> > > > > > >> > > > > >> > internals
> > > > > > >> > > > > >> > > > > > > here and we can consider this
> separately if
> > > > we see
> > > > > > >> > value in
> > > > > > >> > > > > >> doing
> > > > > > >> > > > > >> > > it.
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > Coming back to Ismael's point about
> using
> > > > topic ids
> > > > > > >> > in the
> > > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think
> that
> > > > there
> > > > > > >> is
> > > > > > >> > one
> > > > > > >> > > > > >> > advantage
> > > > > > >> > > > > >> > > in
> > > > > > >> > > > > >> > > > > > > favour of it. The consumer will have
> the
> > > > > > >> opportunity
> > > > > > >> > to
> > > > > > >> > > > > >> validate
> > > > > > >> > > > > >> > > that
> > > > > > >> > > > > >> > > > > > > the topics exists before passing them
> into
> > > > the
> > > > > > >> group
> > > > > > >> > > > > rebalance
> > > > > > >> > > > > >> > > > > > > protocol. Obviously, the coordinator
> will
> > > > also
> > > > > > >> notice
> > > > > > >> > it but
> > > > > > >> > > > > >> it
> > > > > > >> > > > > >> > > does
> > > > > > >> > > > > >> > > > > > > not really have a way to reject an
> invalid
> > > > topic in
> > > > > > >> > the
> > > > > > >> > > > > >> response.
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other
> minor
> > > > > > >> questions
> > > > > > >> > > > > except
> > > > > > >> > > > > >> for
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> personally
> > > > I think
> > > > > > >> > it's not
> > > > > > >> > > > > >> > > necessary
> > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> Pattern,
> > > > but
> > > > > > >> > instead we
> > > > > > >> > > > > >> still
> > > > > > >> > > > > >> > > use
> > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > subscription may
> > > > > > >> be
> > > > > > >> > > > > >> rejected by
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > server.
> > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> very rare
> > > > > > >> scenario
> > > > > > >> > I felt
> > > > > > >> > > > > >> > using
> > > > > > >> > > > > >> > > an
> > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> subscription may
> > > > be
> > > > > > >> more
> > > > > > >> > > > > >> vulnerable
> > > > > > >> > > > > >> > to
> > > > > > >> > > > > >> > > > > various
> > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > That could work. I have to look at the
> > > > differences
> > > > > > >> > between
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > two
> > > > > > >> > > > > >> > > > > > > engines to better understand the
> potential
> > > > issues.
> > > > > > >> My
> > > > > > >> > > > > >> > > understanding is
> > > > > > >> > > > > >> > > > > > > that would work for all the basic
> regular
> > > > > > >> > expressions. The
> > > > > > >> > > > > >> > > differences
> > > > > > >> > > > > >> > > > > > > between the two are mainly about the
> various
> > > > > > >> character
> > > > > > >> > > > > >> classes. I
> > > > > > >> > > > > >> > > > > > > wonder what other people think about
> this.
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > Best,
> > > > > > >> > > > > >> > > > > > > David
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM
> Guozhang
> > > > Wang <
> > > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > > >> > > > > >> > > >
> > > > > > >> > > > > >> > > > > wrote:
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > Thanks David! I think on the high
> level
> > > > there are
> > > > > > >> > two meta
> > > > > > >> > > > > >> > > points we
> > > > > > >> > > > > >> > > > > need
> > > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > 1) the migration path, especially
> the last
> > > > step
> > > > > > >> when
> > > > > > >> > > > > clients
> > > > > > >> > > > > >> > > flip the
> > > > > > >> > > > > >> > > > > > > flag
> > > > > > >> > > > > >> > > > > > > > to enable the new protocol, in which
> we
> > > > would
> > > > > > >> have a
> > > > > > >> > > > > window
> > > > > > >> > > > > >> > where
> > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > >> > > > > > > new
> > > > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols /
> rpcs
> > > > are
> > > > > > >> used
> > > > > > >> > by
> > > > > > >> > > > > >> members
> > > > > > >> > > > > >> > of
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > >> > > > > > > > group. How the coordinator could
> "mimic"
> > > > the old
> > > > > > >> > behavior
> > > > > > >> > > > > >> while
> > > > > > >> > > > > >> > > > > using the
> > > > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > > > present
> > > > > > >> about.
> > > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > > > KIP-516 the
> > > > > > >> > topic ids
> > > > > > >> > > > > >> are
> > > > > > >> > > > > >> > > only
> > > > > > >> > > > > >> > > > > used
> > > > > > >> > > > > >> > > > > > > as
> > > > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but
> they
> > > > are not
> > > > > > >> > exposed
> > > > > > >> > > > > via
> > > > > > >> > > > > >> any
> > > > > > >> > > > > >> > > > > public
> > > > > > >> > > > > >> > > > > > > APIs
> > > > > > >> > > > > >> > > > > > > > to consumers yet. I think the
> question is,
> > > > first
> > > > > > >> > should we
> > > > > > >> > > > > >> let
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > consumer
> > > > > > >> > > > > >> > > > > > > > client to be maintaining the names
> -> ids
> > > > mapping
> > > > > > >> > itself
> > > > > > >> > > > > to
> > > > > > >> > > > > >> > fully
> > > > > > >> > > > > >> > > > > > > leverage
> > > > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs
> and the
> > > > new
> > > > > > >> RPCs
> > > > > > >> > with
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > >> > > > > ids;
> > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > >> > > > > > > > secondly, should we ever consider
> exposing
> > > > the
> > > > > > >> > topic ids
> > > > > > >> > > > > in
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > consumer
> > > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > > subscribe/assign, as
> > > > > > >> well
> > > > > > >> > as in
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > rebalance
> > > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > > >> > deletion-and-recreation).
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other
> minor
> > > > > > >> questions
> > > > > > >> > > > > except
> > > > > > >> > > > > >> for
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question:
> personally
> > > > I think
> > > > > > >> > it's not
> > > > > > >> > > > > >> > > necessary
> > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with
> Pattern,
> > > > but
> > > > > > >> > instead we
> > > > > > >> > > > > >> still
> > > > > > >> > > > > >> > > use
> > > > > > >> > > > > >> > > > > > > Pattern
> > > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > > subscription may
> > > > > > >> be
> > > > > > >> > > > > >> rejected by
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > server.
> > > > > > >> > > > > >> > > > > > > > Since the incompatible case is a
> very rare
> > > > > > >> scenario
> > > > > > >> > I felt
> > > > > > >> > > > > >> > using
> > > > > > >> > > > > >> > > an
> > > > > > >> > > > > >> > > > > > > > overloaded `String` based
> subscription may
> > > > be
> > > > > > >> more
> > > > > > >> > > > > >> vulnerable
> > > > > > >> > > > > >> > to
> > > > > > >> > > > > >> > > > > various
> > > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM
> David Jacot
> > > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > wrote:
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let me
> answer
> > > > your
> > > > > > >> > questions
> > > > > > >> > > > > >> > inline.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 1. I think it's premature to
> talk about
> > > > > > >> target
> > > > > > >> > > > > versions
> > > > > > >> > > > > >> for
> > > > > > >> > > > > >> > > > > > > deprecation
> > > > > > >> > > > > >> > > > > > > > > and
> > > > > > >> > > > > >> > > > > > > > > > removal of the existing group
> protocol.
> > > > > > >> Unlike
> > > > > > >> > KRaft,
> > > > > > >> > > > > >> this
> > > > > > >> > > > > >> > > > > affects a
> > > > > > >> > > > > >> > > > > > > core
> > > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > > deprecation/removal
> > > > > > >> > will be
> > > > > > >> > > > > >> > heavily
> > > > > > >> > > > > >> > > > > > > dependent
> > > > > > >> > > > > >> > > > > > > > > on
> > > > > > >> > > > > >> > > > > > > > > > how quickly applications migrate
> to
> > > > the new
> > > > > > >> > protocol.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to
> release
> > > > this in
> > > > > > >> > 4.x, but
> > > > > > >> > > > > it
> > > > > > >> > > > > >> > > wasn't
> > > > > > >> > > > > >> > > > > made
> > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > >> > > > > >> > > > > > > > > > why. If we added that as a way to
> > > > estimate
> > > > > > >> when
> > > > > > >> > we'd
> > > > > > >> > > > > >> > > deprecate
> > > > > > >> > > > > >> > > > > and
> > > > > > >> > > > > >> > > > > > > remove
> > > > > > >> > > > > >> > > > > > > > > > the group protocol, I also
> suggest
> > > > removing
> > > > > > >> > this part.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> > > > explained, I
> > > > > > >> plan
> > > > > > >> > to
> > > > > > >> > > > > >> rewrite
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > group
> > > > > > >> > > > > >> > > > > > > > > coordinator in Java while we
> implement
> > > > the new
> > > > > > >> > protocol.
> > > > > > >> > > > > >> This
> > > > > > >> > > > > >> > > means
> > > > > > >> > > > > >> > > > > > > > > that the internals will be slightly
> > > > different
> > > > > > >> > (e.g.
> > > > > > >> > > > > >> threading
> > > > > > >> > > > > >> > > > > model).
> > > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the
> > > > switch from
> > > > > > >> > the old
> > > > > > >> > > > > >> group
> > > > > > >> > > > > >> > > > > > > > > coordinator to the new group
> coordinator
> > > > to a
> > > > > > >> > major
> > > > > > >> > > > > >> release.
> > > > > > >> > > > > >> > > The
> > > > > > >> > > > > >> > > > > > > > > alternative would be to use a flag
> to do
> > > > the
> > > > > > >> > switch
> > > > > > >> > > > > >> instead
> > > > > > >> > > > > >> > of
> > > > > > >> > > > > >> > > > > relying
> > > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the
> details of
> > > > the
> > > > > > >> > migration
> > > > > > >> > > > > >> story.
> > > > > > >> > > > > >> > > It
> > > > > > >> > > > > >> > > > > sounds
> > > > > > >> > > > > >> > > > > > > > > like
> > > > > > >> > > > > >> > > > > > > > > > we're saying we will support
> online
> > > > > > >> migrations.
> > > > > > >> > Is
> > > > > > >> > > > > that
> > > > > > >> > > > > >> > > correct?
> > > > > > >> > > > > >> > > > > We
> > > > > > >> > > > > >> > > > > > > > > should
> > > > > > >> > > > > >> > > > > > > > > > explain this in detail. It could
> also
> > > > be done
> > > > > > >> > as a
> > > > > > >> > > > > >> separate
> > > > > > >> > > > > >> > > KIP,
> > > > > > >> > > > > >> > > > > if
> > > > > > >> > > > > >> > > > > > > it's
> > > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Yes, we will support online
> migrations
> > > > for the
> > > > > > >> > group.
> > > > > > >> > > > > That
> > > > > > >> > > > > >> > > means
> > > > > > >> > > > > >> > > > > that
> > > > > > >> > > > > >> > > > > > > > > a group using the old protocol
> will be
> > > > able to
> > > > > > >> > switch to
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > new
> > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Let me briefly explain how that
> will work
> > > > > > >> though.
> > > > > > >> > It is
> > > > > > >> > > > > >> > > basically a
> > > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or
> > > > rolled to a
> > > > > > >> > software
> > > > > > >> > > > > >> > > supporting
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > new group coordinator. Both the
> old and
> > > > the new
> > > > > > >> > > > > >> coordinator
> > > > > > >> > > > > >> > > will
> > > > > > >> > > > > >> > > > > > > > > support the old protocol and rely
> on the
> > > > same
> > > > > > >> > persisted
> > > > > > >> > > > > >> > > metadata so
> > > > > > >> > > > > >> > > > > > > > > they can work together. This point
> is an
> > > > > > >> offline
> > > > > > >> > > > > >> migration.
> > > > > > >> > > > > >> > We
> > > > > > >> > > > > >> > > > > cannot
> > > > > > >> > > > > >> > > > > > > > > do this one live because it would
> require
> > > > > > >> > shutting down
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > current
> > > > > > >> > > > > >> > > > > > > > > coordinator and starting up the
> new one
> > > > and
> > > > > > >> that
> > > > > > >> > would
> > > > > > >> > > > > >> cause
> > > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata
> version/IBP
> > > > must be
> > > > > > >> > upgraded
> > > > > > >> > > > > to
> > > > > > >> > > > > >> X
> > > > > > >> > > > > >> > in
> > > > > > >> > > > > >> > > > > order
> > > > > > >> > > > > >> > > > > > > > > to enable the new protocol. This
> cannot
> > > > be done
> > > > > > >> > before
> > > > > > >> > > > > 1)
> > > > > > >> > > > > >> is
> > > > > > >> > > > > >> > > > > > > > > terminated because the old
> coordinator
> > > > doesn't
> > > > > > >> > support
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > new
> > > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > > >> > > > > >> > > > > > > > > 3. The consumers must be upgraded
> to a
> > > > version
> > > > > > >> > > > > supporting
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > online
> > > > > > >> > > > > >> > > > > > > > > migration (must have KIP-792). If
> the
> > > > consumer
> > > > > > >> is
> > > > > > >> > > > > already
> > > > > > >> > > > > >> > > there.
> > > > > > >> > > > > >> > > > > > > > > Nothing must be done at this point.
> > > > > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled
> with the
> > > > > > >> feature
> > > > > > >> > flag
> > > > > > >> > > > > >> turned
> > > > > > >> > > > > >> > > on.
> > > > > > >> > > > > >> > > > > The
> > > > > > >> > > > > >> > > > > > > > > consumer group is automatically
> > > > converted when
> > > > > > >> > the first
> > > > > > >> > > > > >> > > consumer
> > > > > > >> > > > > >> > > > > > > > > using the new protocol joins the
> group.
> > > > While
> > > > > > >> the
> > > > > > >> > > > > members
> > > > > > >> > > > > >> > > using the
> > > > > > >> > > > > >> > > > > > > > > old protocol are being upgraded,
> the old
> > > > > > >> protocol
> > > > > > >> > is
> > > > > > >> > > > > >> proxied
> > > > > > >> > > > > >> > > into
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > new one.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Let me clarify all of this in the
> KIP.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are
> pushing the
> > > > pattern
> > > > > > >> > > > > >> subscriptions
> > > > > > >> > > > > >> > > to
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > >> > > > > >> > > > > > > > > > but it seems like there could be
> some
> > > > tricky
> > > > > > >> > > > > >> compatibility
> > > > > > >> > > > > >> > > > > issues.
> > > > > > >> > > > > >> > > > > > > Will
> > > > > > >> > > > > >> > > > > > > > > we
> > > > > > >> > > > > >> > > > > > > > > > have a mechanism for users to
> detect
> > > > that
> > > > > > >> they
> > > > > > >> > need to
> > > > > > >> > > > > >> > update
> > > > > > >> > > > > >> > > > > their
> > > > > > >> > > > > >> > > > > > > regex
> > > > > > >> > > > > >> > > > > > > > > > before switching to the new
> protocol?
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > I think that I am a bit more
> optimistic
> > > > than
> > > > > > >> you
> > > > > > >> > on this
> > > > > > >> > > > > >> > > point. I
> > > > > > >> > > > > >> > > > > > > > > believe that the majority of the
> cases
> > > > are
> > > > > > >> simple
> > > > > > >> > > > > regexes
> > > > > > >> > > > > >> > which
> > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > >> > > > > > > > > work with the new engine. The
> > > > coordinator will
> > > > > > >> > verify
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > regex
> > > > > > >> > > > > >> > > > > anyway
> > > > > > >> > > > > >> > > > > > > > > and reject the consumer if the
> regex is
> > > > not
> > > > > > >> valid.
> > > > > > >> > > > > Coming
> > > > > > >> > > > > >> > back
> > > > > > >> > > > > >> > > to
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > migration path, in the worst case,
> the
> > > > first
> > > > > > >> > upgraded
> > > > > > >> > > > > >> > consumer
> > > > > > >> > > > > >> > > > > joining
> > > > > > >> > > > > >> > > > > > > > > the group will be rejected. This
> should
> > > > be used
> > > > > > >> > as the
> > > > > > >> > > > > >> last
> > > > > > >> > > > > >> > > > > defence, I
> > > > > > >> > > > > >> > > > > > > > > would say.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > One way for customers to validate
> their
> > > > regex
> > > > > > >> > before
> > > > > > >> > > > > >> > upgrading
> > > > > > >> > > > > >> > > > > their
> > > > > > >> > > > > >> > > > > > > > > prod would be to test them with
> another
> > > > group.
> > > > > > >> For
> > > > > > >> > > > > >> instance,
> > > > > > >> > > > > >> > > that
> > > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod
> environment.
> > > > > > >> Another
> > > > > > >> > way
> > > > > > >> > > > > >> would be
> > > > > > >> > > > > >> > > to
> > > > > > >> > > > > >> > > > > > > > > extend the consumer-group tool to
> > > > provide a
> > > > > > >> regex
> > > > > > >> > > > > >> validation
> > > > > > >> > > > > >> > > > > > > > > mechanism. Would this be enough in
> your
> > > > > > >> opinion?
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 5. Related to the last question,
> will
> > > > the
> > > > > > >> Java
> > > > > > >> > client
> > > > > > >> > > > > >> allow
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > users to
> > > > > > >> > > > > >> > > > > > > > > > stick with the current regex
> engine for
> > > > > > >> > compatibility
> > > > > > >> > > > > >> > > reasons?
> > > > > > >> > > > > >> > > > > For
> > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > >> > > > > >> > > > > > > > > > it may be handy to keep using
> client
> > > > based
> > > > > > >> > regex at
> > > > > > >> > > > > >> first
> > > > > > >> > > > > >> > to
> > > > > > >> > > > > >> > > keep
> > > > > > >> > > > > >> > > > > > > > > > migrations simple and then
> migrate to
> > > > server
> > > > > > >> > based
> > > > > > >> > > > > >> regexes
> > > > > > >> > > > > >> > > as a
> > > > > > >> > > > > >> > > > > > > second
> > > > > > >> > > > > >> > > > > > > > > step.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > I understand your point but I am
> > > > concerned that
> > > > > > >> > this
> > > > > > >> > > > > would
> > > > > > >> > > > > >> > > allow
> > > > > > >> > > > > >> > > > > users
> > > > > > >> > > > > >> > > > > > > > > to actually stay in this mode. That
> > > > would go
> > > > > > >> > against our
> > > > > > >> > > > > >> goal
> > > > > > >> > > > > >> > > of
> > > > > > >> > > > > >> > > > > > > > > simplifying the client because we
> would
> > > > have to
> > > > > > >> > continue
> > > > > > >> > > > > >> > > monitoring
> > > > > > >> > > > > >> > > > > > > > > the metadata on the client side. I
> would
> > > > rather
> > > > > > >> > not do
> > > > > > >> > > > > >> this.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> > > > coordinator
> > > > > > >> will
> > > > > > >> > be
> > > > > > >> > > > > >> > > responsible for
> > > > > > >> > > > > >> > > > > > > storing
> > > > > > >> > > > > >> > > > > > > > > > the configurations and that the
> > > > > > >> configurations
> > > > > > >> > will be
> > > > > > >> > > > > >> > > deleted
> > > > > > >> > > > > >> > > > > when
> > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a
> transition to
> > > > DEAD
> > > > > > >> > trigger
> > > > > > >> > > > > >> > deletion
> > > > > > >> > > > > >> > > of
> > > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > That's right. The configurations
> will be
> > > > > > >> deleted
> > > > > > >> > when
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > group is
> > > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to store the
> > > > configs in
> > > > > > >> the
> > > > > > >> > group
> > > > > > >> > > > > >> > > coordinator
> > > > > > >> > > > > >> > > > > > > make it
> > > > > > >> > > > > >> > > > > > > > > > harder to list all cluster
> configs and
> > > > their
> > > > > > >> > values?
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > > configurations are
> > > > > > >> > overrides
> > > > > > >> > > > > >> of
> > > > > > >> > > > > >> > > cluster
> > > > > > >> > > > > >> > > > > > > > > configs. If you want to know all
> the
> > > > overrides
> > > > > > >> > though,
> > > > > > >> > > > > you
> > > > > > >> > > > > >> > > would
> > > > > > >> > > > > >> > > > > have
> > > > > > >> > > > > >> > > > > > > > > to ask all the group coordinators.
> You
> > > > cannot
> > > > > > >> > rely on
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > metadata
> > > > > > >> > > > > >> > > > > log
> > > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 8. How would someone configure a
> group
> > > > before
> > > > > > >> > starting
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > consumers?
> > > > > > >> > > > > >> > > > > > > > > Have
> > > > > > >> > > > > >> > > > > > > > > > we considered allowing the
> explicit
> > > > creation
> > > > > > >> of
> > > > > > >> > > > > groups?
> > > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > > configs could be decoupled from
> the
> > > > group
> > > > > > >> > lifecycle.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Yes. The group will be
> automatically
> > > > created in
> > > > > > >> > this
> > > > > > >> > > > > case.
> > > > > > >> > > > > >> > > However,
> > > > > > >> > > > > >> > > > > > > > > the configs will be lost after the
> > > > retention
> > > > > > >> > period of
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > group
> > > > > > >> > > > > >> > > > > > > > > passes.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe
> method
> > > > for the
> > > > > > >> > Java
> > > > > > >> > > > > >> client
> > > > > > >> > > > > >> > > still
> > > > > > >> > > > > >> > > > > take
> > > > > > >> > > > > >> > > > > > > a
> > > > > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do
> we
> > > > have to
> > > > > > >> > introduce
> > > > > > >> > > > > an
> > > > > > >> > > > > >> > > overload?
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > That's a very group question. I
> forgot
> > > > about
> > > > > > >> that
> > > > > > >> > one.
> > > > > > >> > > > > As
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not
> fully
> > > > > > >> compatible
> > > > > > >> > with
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > engine
> > > > > > >> > > > > >> > > > > that
> > > > > > >> > > > > >> > > > > > > > > we plan to use, it might be better
> to
> > > > deprecate
> > > > > > >> > it and
> > > > > > >> > > > > >> use an
> > > > > > >> > > > > >> > > > > overload
> > > > > > >> > > > > >> > > > > > > > > which takes a string. We would
> rely on
> > > > the
> > > > > > >> server
> > > > > > >> > side
> > > > > > >> > > > > >> > > validation.
> > > > > > >> > > > > >> > > > > > > > > During the migration, I think that
> we
> > > > could
> > > > > > >> still
> > > > > > >> > try to
> > > > > > >> > > > > >> > > toString
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > regex and use it. That should
> work, I
> > > > think, in
> > > > > > >> > the
> > > > > > >> > > > > >> majority
> > > > > > >> > > > > >> > > of the
> > > > > > >> > > > > >> > > > > > > > > cases.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine that we
> > > > should be
> > > > > > >> > clearer
> > > > > > >> > > > > about
> > > > > > >> > > > > >> > the
> > > > > > >> > > > > >> > > > > reason
> > > > > > >> > > > > >> > > > > > > to
> > > > > > >> > > > > >> > > > > > > > > > switch to IBP/metadata.version
> from the
> > > > > > >> feature
> > > > > > >> > flag.
> > > > > > >> > > > > >> Maybe
> > > > > > >> > > > > >> > > we
> > > > > > >> > > > > >> > > > > mean
> > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > >> > > > > > > > > we
> > > > > > >> > > > > >> > > > > > > > > > can switch the default for the
> feature
> > > > flag
> > > > > > >> to
> > > > > > >> > true
> > > > > > >> > > > > >> based
> > > > > > >> > > > > >> > on
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > > > > metadata.version once we want to
> make
> > > > it the
> > > > > > >> > default.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > My plan was to use that feature
> flag
> > > > mainly
> > > > > > >> > during the
> > > > > > >> > > > > >> > > development
> > > > > > >> > > > > >> > > > > > > > > phase. I should not have mentioned
> it, I
> > > > think,
> > > > > > >> > because
> > > > > > >> > > > > we
> > > > > > >> > > > > >> > > could
> > > > > > >> > > > > >> > > > > use
> > > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs
> don't
> > > > mention
> > > > > > >> the
> > > > > > >> > > > > required
> > > > > > >> > > > > >> > > ACLs, it
> > > > > > >> > > > > >> > > > > > > would
> > > > > > >> > > > > >> > > > > > > > > be
> > > > > > >> > > > > >> > > > > > > > > > good to add that for consistency.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > > >> ConsumerGroupHeartbeat
> > > > > > >> > > > > requires
> > > > > > >> > > > > >> > > "Read
> > > > > > >> > > > > >> > > > > Group"
> > > > > > >> > > > > >> > > > > > > > > even
> > > > > > >> > > > > >> > > > > > > > > > though it seems to do more than
> > > > reading.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > I agree. This is how the current
> > > > protocol works
> > > > > > >> > though.
> > > > > > >> > > > > We
> > > > > > >> > > > > >> > only
> > > > > > >> > > > > >> > > > > > > > > require "Read Group" to join a
> group. We
> > > > could
> > > > > > >> > consider
> > > > > > >> > > > > >> > > changing
> > > > > > >> > > > > >> > > > > this
> > > > > > >> > > > > >> > > > > > > > > but I am not sure that it is worth
> it.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 13. How is topic recreation
> handled by
> > > > the
> > > > > > >> > consumer
> > > > > > >> > > > > with
> > > > > > >> > > > > >> > the
> > > > > > >> > > > > >> > > new
> > > > > > >> > > > > >> > > > > > > group
> > > > > > >> > > > > >> > > > > > > > > > protocol? It would be good to
> have a
> > > > section
> > > > > > >> on
> > > > > > >> > this.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Noted. From a protocol
> perspective, the
> > > > new
> > > > > > >> topic
> > > > > > >> > will
> > > > > > >> > > > > >> have a
> > > > > > >> > > > > >> > > new
> > > > > > >> > > > > >> > > > > > > > > topic id so it will treat it like a
> > > > topic with
> > > > > > >> a
> > > > > > >> > > > > different
> > > > > > >> > > > > >> > > name.
> > > > > > >> > > > > >> > > > > The
> > > > > > >> > > > > >> > > > > > > > > only issue is that the fetch/commit
> > > > offsets
> > > > > > >> APIs
> > > > > > >> > do not
> > > > > > >> > > > > >> > support
> > > > > > >> > > > > >> > > > > topic
> > > > > > >> > > > > >> > > > > > > > > IDs so the consumer would reuse the
> > > > offsets
> > > > > > >> based
> > > > > > >> > on the
> > > > > > >> > > > > >> > same.
> > > > > > >> > > > > >> > > I
> > > > > > >> > > > > >> > > > > think
> > > > > > >> > > > > >> > > > > > > > > that we should update those APIs
> as well
> > > > in
> > > > > > >> order
> > > > > > >> > to be
> > > > > > >> > > > > >> > > consistent
> > > > > > >> > > > > >> > > > > end
> > > > > > >> > > > > >> > > > > > > > > to end. That would strengthen the
> > > > semantics of
> > > > > > >> the
> > > > > > >> > > > > >> consumer.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will
> write the
> > > > new
> > > > > > >> > coordinator
> > > > > > >> > > > > >> in
> > > > > > >> > > > > >> > > Java.
> > > > > > >> > > > > >> > > > > Even
> > > > > > >> > > > > >> > > > > > > > > though
> > > > > > >> > > > > >> > > > > > > > > > this is an implementation
> detail, do
> > > > we plan
> > > > > > >> to
> > > > > > >> > have a
> > > > > > >> > > > > >> new
> > > > > > >> > > > > >> > > gradle
> > > > > > >> > > > > >> > > > > > > module
> > > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability
> goal when
> > > > it
> > > > > > >> comes
> > > > > > >> > to how
> > > > > > >> > > > > >> many
> > > > > > >> > > > > >> > > > > members
> > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > new
> > > > > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > We don't have numbers at the
> moment. The
> > > > > > >> protocol
> > > > > > >> > should
> > > > > > >> > > > > >> > > support
> > > > > > >> > > > > >> > > > > 1000s
> > > > > > >> > > > > >> > > > > > > > > of members per group. We will
> measure
> > > > this when
> > > > > > >> > we have
> > > > > > >> > > > > a
> > > > > > >> > > > > >> > first
> > > > > > >> > > > > >> > > > > > > > > implementation. Note that we might
> have
> > > > other
> > > > > > >> > > > > bottlenecks
> > > > > > >> > > > > >> > down
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> > > > SubscribedTopidIds
> > > > > > >> > instead
> > > > > > >> > > > > of
> > > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > > >> > > > > >> Is
> > > > > > >> > > > > >> > the
> > > > > > >> > > > > >> > > > > idea
> > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > >> > > > > > > > > > since we have to resolve the
> regex on
> > > > the
> > > > > > >> > server, we
> > > > > > >> > > > > >> can do
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > same
> > > > > > >> > > > > >> > > > > > > for
> > > > > > >> > > > > >> > > > > > > > > > the topic name? The difference
> is that
> > > > > > >> sending
> > > > > > >> > the
> > > > > > >> > > > > >> regex is
> > > > > > >> > > > > >> > > more
> > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > >> > > > > >> > > > > > > > > > whereas sending the topic names
> is less
> > > > > > >> > efficient.
> > > > > > >> > > > > >> > > Furthermore,
> > > > > > >> > > > > >> > > > > > > delete
> > > > > > >> > > > > >> > > > > > > > > and
> > > > > > >> > > > > >> > > > > > > > > > recreation is easier to handle
> if we
> > > > have
> > > > > > >> topic
> > > > > > >> > ids.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > The idea was to consolidate the
> metadata
> > > > lookup
> > > > > > >> > on the
> > > > > > >> > > > > >> server
> > > > > > >> > > > > >> > > for
> > > > > > >> > > > > >> > > > > both
> > > > > > >> > > > > >> > > > > > > > > paths but I do agree with your
> point. As
> > > > a
> > > > > > >> second
> > > > > > >> > > > > though,
> > > > > > >> > > > > >> > using
> > > > > > >> > > > > >> > > > > topic
> > > > > > >> > > > > >> > > > > > > > > ids may be better here for the
> delete and
> > > > > > >> > recreation
> > > > > > >> > > > > case.
> > > > > > >> > > > > >> > > Also, I
> > > > > > >> > > > > >> > > > > > > > > suppose that we may allow users to
> > > > subscribe
> > > > > > >> with
> > > > > > >> > topic
> > > > > > >> > > > > >> ids
> > > > > > >> > > > > >> > in
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > > > future because that is the only
> way to be
> > > > > > >> really
> > > > > > >> > robust
> > > > > > >> > > > > to
> > > > > > >> > > > > >> > > topic
> > > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > Best,
> > > > > > >> > > > > >> > > > > > > > > David
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM
> David
> > > > Jacot <
> > > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > Thanks for your comments. Please
> find
> > > > my
> > > > > > >> answers
> > > > > > >> > > > > below.
> > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on
> > > > topic IDs
> > > > > > >> > with the
> > > > > > >> > > > > >> > > exception
> > > > > > >> > > > > >> > > > > of the
> > > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > > >> > > > > >> I
> > > > > > >> > > > > >> > am
> > > > > > >> > > > > >> > > not
> > > > > > >> > > > > >> > > > > sure
> > > > > > >> > > > > >> > > > > > > > > > if using topic names is the
> right call
> > > > here.
> > > > > > >> I
> > > > > > >> > need to
> > > > > > >> > > > > >> > think
> > > > > > >> > > > > >> > > > > about it
> > > > > > >> > > > > >> > > > > > > > > > a little more. Obviously, the
> KIP does
> > > > not
> > > > > > >> > change the
> > > > > > >> > > > > >> > > > > fetch/commit
> > > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs.
> This
> > > > may be
> > > > > > >> > something
> > > > > > >> > > > > >> that
> > > > > > >> > > > > >> > we
> > > > > > >> > > > > >> > > > > should
> > > > > > >> > > > > >> > > > > > > > > > include though as it would give
> better
> > > > > > >> overall
> > > > > > >> > > > > >> guarantee in
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > > >> > > > > >> > > > > > > > > > - You're right. I think that I
> should
> > > > not
> > > > > > >> have
> > > > > > >> > > > > mentioned
> > > > > > >> > > > > >> > this
> > > > > > >> > > > > >> > > > > flag at
> > > > > > >> > > > > >> > > > > > > > > > all. I will remove it. We can
> use an
> > > > internal
> > > > > > >> > > > > >> configuration
> > > > > > >> > > > > >> > > while
> > > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> > > > supported. The
> > > > > > >> > change is
> > > > > > >> > > > > >> > > > > orthogonal. The
> > > > > > >> > > > > >> > > > > > > > > > only requirement is that the
> cluster
> > > > uses
> > > > > > >> topic
> > > > > > >> > IDs.
> > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > > >> > > > > >> > > > > > > > > > David
> > > > > > >> > > > > >> > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM
> > > > Guozhang
> > > > > > >> Wang <
> > > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here
> are
> > > > some
> > > > > > >> replies
> > > > > > >> > > > > inlined
> > > > > > >> > > > > >> > > below:
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM
> > > > Ismael Juma
> > > > > > >> <
> > > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > > >> > > > > >> > > > > > > wrote:
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has
> the
> > > > > > >> potential
> > > > > > >> > to be a
> > > > > > >> > > > > >> > great
> > > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > > >> > > > > >> > > > > > > > > > > > initial questions/comments:
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's premature to
> talk
> > > > about
> > > > > > >> > target
> > > > > > >> > > > > >> versions
> > > > > > >> > > > > >> > > for
> > > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > > >> > > > > >> > > > > > > > > > > > removal of the existing group
> > > > protocol.
> > > > > > >> > Unlike
> > > > > > >> > > > > >> KRaft,
> > > > > > >> > > > > >> > > this
> > > > > > >> > > > > >> > > > > > > affects a
> > > > > > >> > > > > >> > > > > > > > > core
> > > > > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > > > > >> > deprecation/removal will
> > > > > > >> > > > > >> be
> > > > > > >> > > > > >> > > heavily
> > > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > > >> > > > > >> > > > > > > > > > > > how quickly applications
> migrate
> > > > to the
> > > > > > >> new
> > > > > > >> > > > > >> protocol.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think
> we can
> > > > > > >> remove
> > > > > > >> > the
> > > > > > >> > > > > >> proposed
> > > > > > >> > > > > >> > > > > timeline
> > > > > > >> > > > > >> > > > > > > in
> > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation,
> and
> > > > Migration
> > > > > > >> > Plan` and
> > > > > > >> > > > > >> > > instead
> > > > > > >> > > > > >> > > > > just
> > > > > > >> > > > > >> > > > > > > state
> > > > > > >> > > > > >> > > > > > > > > > > that we will decide in the
> future
> > > > about
> > > > > > >> when
> > > > > > >> > we
> > > > > > >> > > > > would
> > > > > > >> > > > > >> > > > > deprecate old
> > > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to
> > > > release this
> > > > > > >> > in 4.x,
> > > > > > >> > > > > >> but
> > > > > > >> > > > > >> > it
> > > > > > >> > > > > >> > > > > wasn't
> > > > > > >> > > > > >> > > > > > > made
> > > > > > >> > > > > >> > > > > > > > > clear
> > > > > > >> > > > > >> > > > > > > > > > > > why. If we added that as a
> way to
> > > > > > >> estimate
> > > > > > >> > when
> > > > > > >> > > > > we'd
> > > > > > >> > > > > >> > > > > deprecate
> > > > > > >> > > > > >> > > > > > > and
> > > > > > >> > > > > >> > > > > > > > > remove
> > > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I also
> suggest
> > > > > > >> removing
> > > > > > >> > this
> > > > > > >> > > > > >> part.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > I think that's not specifically
> > > > related to
> > > > > > >> the
> > > > > > >> > > > > >> > > > > deprecation/removal
> > > > > > >> > > > > >> > > > > > > > > timeline
> > > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for client
> > > > upgrades.
> > > > > > >> I.e.
> > > > > > >> > the
> > > > > > >> > > > > >> > > broker-side
> > > > > > >> > > > > >> > > > > > > > > > > implementation may be done
> first,
> > > > and then
> > > > > > >> the
> > > > > > >> > > > > client
> > > > > > >> > > > > >> > side,
> > > > > > >> > > > > >> > > > > and we
> > > > > > >> > > > > >> > > > > > > > > would
> > > > > > >> > > > > >> > > > > > > > > > > only mark it as "released" by
> the
> > > > time
> > > > > > >> clients
> > > > > > >> > > > > >> > > implementations
> > > > > > >> > > > > >> > > > > are
> > > > > > >> > > > > >> > > > > > > > > done. At
> > > > > > >> > > > > >> > > > > > > > > > > that time, to enable the
> feature the
> > > > > > >> clients
> > > > > > >> > need to
> > > > > > >> > > > > >> > first
> > > > > > >> > > > > >> > > > > swap-in
> > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce
> and
> > > > then set
> > > > > > >> > the flag
> > > > > > >> > > > > >> > with a
> > > > > > >> > > > > >> > > > > second
> > > > > > >> > > > > >> > > > > > > > > rolling
> > > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's
> > > > better to be
> > > > > > >> > released
> > > > > > >> > > > > >> in a
> > > > > > >> > > > > >> > > major
> > > > > > >> > > > > >> > > > > > > > > version.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the
> > > > details of
> > > > > > >> the
> > > > > > >> > > > > migration
> > > > > > >> > > > > >> > > story.
> > > > > > >> > > > > >> > > > > It
> > > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > > >> > > > > >> > > > > > > > > > > > we're saying we will support
> online
> > > > > > >> > migrations. Is
> > > > > > >> > > > > >> that
> > > > > > >> > > > > >> > > > > correct?
> > > > > > >> > > > > >> > > > > > > We
> > > > > > >> > > > > >> > > > > > > > > should
> > > > > > >> > > > > >> > > > > > > > > > > > explain this in detail. It
> could
> > > > also be
> > > > > > >> > done as a
> > > > > > >> > > > > >> > > separate
> > > > > > >> > > > > >> > > > > KIP,
> > > > > > >> > > > > >> > > > > > > if
> > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the part we
> can
> > > > be more
> > > > > > >> > concrete
> > > > > > >> > > > > >> about
> > > > > > >> > > > > >> > > for
> > > > > > >> > > > > >> > > > > sure
> > > > > > >> > > > > >> > > > > > > (and
> > > > > > >> > > > > >> > > > > > > > > > > this is related to your
> question 2)
> > > > above).
> > > > > > >> > We will
> > > > > > >> > > > > >> work
> > > > > > >> > > > > >> > on
> > > > > > >> > > > > >> > > > > making
> > > > > > >> > > > > >> > > > > > > it
> > > > > > >> > > > > >> > > > > > > > > more
> > > > > > >> > > > > >> > > > > > > > > > > explicit in parallel as we
> solicit
> > > > more
> > > > > > >> > feedback.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are
> pushing
> > > > the
> > > > > > >> > pattern
> > > > > > >> > > > > >> > > subscriptions
> > > > > > >> > > > > >> > > > > to
> > > > > > >> > > > > >> > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > server,
> > > > > > >> > > > > >> > > > > > > > > > > > but it seems like there
> could be
> > > > some
> > > > > > >> tricky
> > > > > > >> > > > > >> > > compatibility
> > > > > > >> > > > > >> > > > > > > issues.
> > > > > > >> > > > > >> > > > > > > > > Will we
> > > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for users to
> > > > detect that
> > > > > > >> > they
> > > > > > >> > > > > need
> > > > > > >> > > > > >> to
> > > > > > >> > > > > >> > > update
> > > > > > >> > > > > >> > > > > > > their
> > > > > > >> > > > > >> > > > > > > > > regex
> > > > > > >> > > > > >> > > > > > > > > > > > before switching to the new
> > > > protocol?
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Yes I think we need some
> tooling for
> > > > > > >> non-java
> > > > > > >> > client
> > > > > > >> > > > > >> > users
> > > > > > >> > > > > >> > > to
> > > > > > >> > > > > >> > > > > sort
> > > > > > >> > > > > >> > > > > > > of
> > > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> > > > switching to
> > > > > > >> the
> > > > > > >> > new
> > > > > > >> > > > > >> > protocol.
> > > > > > >> > > > > >> > > I
> > > > > > >> > > > > >> > > > > do not
> > > > > > >> > > > > >> > > > > > > > > have a
> > > > > > >> > > > > >> > > > > > > > > > > specific idea on top of my head
> > > > though,
> > > > > > >> maybe
> > > > > > >> > others
> > > > > > >> > > > > >> like
> > > > > > >> > > > > >> > > @Matt
> > > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the last
> question,
> > > > will the
> > > > > > >> > Java
> > > > > > >> > > > > >> client
> > > > > > >> > > > > >> > > allow
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > users to
> > > > > > >> > > > > >> > > > > > > > > > > > stick with the current regex
> > > > engine for
> > > > > > >> > > > > >> compatibility
> > > > > > >> > > > > >> > > > > reasons?
> > > > > > >> > > > > >> > > > > > > For
> > > > > > >> > > > > >> > > > > > > > > example,
> > > > > > >> > > > > >> > > > > > > > > > > > it may be handy to keep using
> > > > client
> > > > > > >> based
> > > > > > >> > regex
> > > > > > >> > > > > at
> > > > > > >> > > > > >> > > first to
> > > > > > >> > > > > >> > > > > keep
> > > > > > >> > > > > >> > > > > > > > > > > > migrations simple and then
> migrate
> > > > to
> > > > > > >> > server based
> > > > > > >> > > > > >> > > regexes
> > > > > > >> > > > > >> > > > > as a
> > > > > > >> > > > > >> > > > > > > > > second
> > > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Honestly I have not thought
> about
> > > > that for
> > > > > > >> > java
> > > > > > >> > > > > >> clients,
> > > > > > >> > > > > >> > > and
> > > > > > >> > > > > >> > > > > we can
> > > > > > >> > > > > >> > > > > > > > > discuss
> > > > > > >> > > > > >> > > > > > > > > > > that. What kind of
> compatibility
> > > > issues do
> > > > > > >> > you have
> > > > > > >> > > > > in
> > > > > > >> > > > > >> > > mind?
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that the group
> > > > coordinator
> > > > > > >> > will be
> > > > > > >> > > > > >> > > > > responsible for
> > > > > > >> > > > > >> > > > > > > > > storing
> > > > > > >> > > > > >> > > > > > > > > > > > the configurations and that
> the
> > > > > > >> > configurations
> > > > > > >> > > > > will
> > > > > > >> > > > > >> be
> > > > > > >> > > > > >> > > > > deleted
> > > > > > >> > > > > >> > > > > > > when
> > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> > > > transition to
> > > > > > >> DEAD
> > > > > > >> > > > > trigger
> > > > > > >> > > > > >> > > deletion
> > > > > > >> > > > > >> > > > > of
> > > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an
> > > > ending
> > > > > > >> state
> > > > > > >> > (we
> > > > > > >> > > > > would
> > > > > > >> > > > > >> > only
> > > > > > >> > > > > >> > > > > > > transit to
> > > > > > >> > > > > >> > > > > > > > > that
> > > > > > >> > > > > >> > > > > > > > > > > state when the group is EMPTY
> and
> > > > also all
> > > > > > >> of
> > > > > > >> > its
> > > > > > >> > > > > >> > metadata
> > > > > > >> > > > > >> > > are
> > > > > > >> > > > > >> > > > > > > gone),
> > > > > > >> > > > > >> > > > > > > > > once
> > > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this
> group
> > > > would
> > > > > > >> never
> > > > > > >> > be
> > > > > > >> > > > > >> revived.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to store
> the
> > > > configs
> > > > > > >> in
> > > > > > >> > the
> > > > > > >> > > > > group
> > > > > > >> > > > > >> > > > > coordinator
> > > > > > >> > > > > >> > > > > > > > > make it
> > > > > > >> > > > > >> > > > > > > > > > > > harder to list all cluster
> configs
> > > > and
> > > > > > >> their
> > > > > > >> > > > > values?
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > That's a good question, and our
> > > > thoughts
> > > > > > >> are
> > > > > > >> > that
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > so-called
> > > > > > >> > > > > >> > > > > > > "group
> > > > > > >> > > > > >> > > > > > > > > > > configurations" are overrides
> of the
> > > > > > >> > cluster-level
> > > > > > >> > > > > >> > > > > configurations
> > > > > > >> > > > > >> > > > > > > > > > > customized per group so when an
> > > > admin list
> > > > > > >> > cluster
> > > > > > >> > > > > >> > configs
> > > > > > >> > > > > >> > > it's
> > > > > > >> > > > > >> > > > > > > okay to
> > > > > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> > > > "defaults", not
> > > > > > >> > showing
> > > > > > >> > > > > >> any
> > > > > > >> > > > > >> > > > > per-group
> > > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone
> configure a
> > > > group
> > > > > > >> > before
> > > > > > >> > > > > >> starting
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > > >> > > > > >> > > > > > > > > > > > we considered allowing the
> explicit
> > > > > > >> > creation of
> > > > > > >> > > > > >> groups?
> > > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > > >> > > > > >> > > > > > > > > > > > configs could be decoupled
> from
> > > > the group
> > > > > > >> > > > > lifecycle.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > The configs can be created
> before
> > > > the group
> > > > > > >> > itself
> > > > > > >> > > > > as
> > > > > > >> > > > > >> an
> > > > > > >> > > > > >> > > > > > > independent
> > > > > > >> > > > > >> > > > > > > > > entity
> > > > > > >> > > > > >> > > > > > > > > > > --- of course, this requires
> the
> > > > > > >> corresponding
> > > > > > >> > > > > >> request to
> > > > > > >> > > > > >> > > be
> > > > > > >> > > > > >> > > > > > > routed to
> > > > > > >> > > > > >> > > > > > > > > the
> > > > > > >> > > > > >> > > > > > > > > > > right coordinator based on the
> group
> > > > id ---
> > > > > > >> > the only
> > > > > > >> > > > > >> > thing
> > > > > > >> > > > > >> > > that
> > > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > > >> > > > > >> > > > > > > > > > > when the group itself is gone
> we
> > > > also check
> > > > > > >> > if there
> > > > > > >> > > > > >> are
> > > > > > >> > > > > >> > > any
> > > > > > >> > > > > >> > > > > > > > > configuration
> > > > > > >> > > > > >> > > > > > > > > > > entities related to that group
> and
> > > > delete
> > > > > > >> as
> > > > > > >> > well.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed
> introduces an
> > > > > > >> > asymmetry on
> > > > > > >> > > > > the
> > > > > > >> > > > > >> > > creation
> > > > > > >> > > > > >> > > > > /
> > > > > > >> > > > > >> > > > > > > > > deletion
> > > > > > >> > > > > >> > > > > > > > > > > lifecycles of the config
> entities,
> > > > and we
> > > > > > >> > would like
> > > > > > >> > > > > >> to
> > > > > > >> > > > > >> > > hear
> > > > > > >> > > > > >> > > > > > > everyone's
> > > > > > >> > > > > >> > > > > > > > > > > feelings whether we should aim
> for
> > > > symmetry
> > > > > > >> > i.e.
> > > > > > >> > > > > >> totally
> > > > > > >> > > > > >> > > > > decouple
> > > > > > >> > > > > >> > > > > > > group
> > > > > > >> > > > > >> > > > > > > > > > > configs and hence not delete
> them at
> > > > all
> > > > > > >> when
> > > > > > >> > the
> > > > > > >> > > > > >> group
> > > > > > >> > > > > >> > is
> > > > > > >> > > > > >> > > > > gone,
> > > > > > >> > > > > >> > > > > > > but
> > > > > > >> > > > > >> > > > > > > > > always
> > > > > > >> > > > > >> > > > > > > > > > > require explicit deletion
> operations
> > > > by
> > > > > > >> > themselves.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 9. Will the
> Consumer.subscribe
> > > > method for
> > > > > > >> > the Java
> > > > > > >> > > > > >> > client
> > > > > > >> > > > > >> > > > > still
> > > > > > >> > > > > >> > > > > > > take
> > > > > > >> > > > > >> > > > > > > > > a
> > > > > > >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of
> do we
> > > > have
> > > > > > >> to
> > > > > > >> > > > > >> introduce an
> > > > > > >> > > > > >> > > > > overload?
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > I think we do not need to
> introduce
> > > > an
> > > > > > >> > overload, but
> > > > > > >> > > > > >> I'm
> > > > > > >> > > > > >> > > all
> > > > > > >> > > > > >> > > > > ears
> > > > > > >> > > > > >> > > > > > > if
> > > > > > >> > > > > >> > > > > > > > > there
> > > > > > >> > > > > >> > > > > > > > > > > may be some compatibility
> issues
> > > > that we
> > > > > > >> may
> > > > > > >> > > > > overlook.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine
> that we
> > > > should
> > > > > > >> be
> > > > > > >> > clearer
> > > > > > >> > > > > >> > about
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > > > reason
> > > > > > >> > > > > >> > > > > > > > > to
> > > > > > >> > > > > >> > > > > > > > > > > > switch to
> IBP/metadata.version
> > > > from the
> > > > > > >> > feature
> > > > > > >> > > > > >> flag.
> > > > > > >> > > > > >> > > Maybe
> > > > > > >> > > > > >> > > > > we
> > > > > > >> > > > > >> > > > > > > mean
> > > > > > >> > > > > >> > > > > > > > > that we
> > > > > > >> > > > > >> > > > > > > > > > > > can switch the default for
> the
> > > > feature
> > > > > > >> flag
> > > > > > >> > to
> > > > > > >> > > > > true
> > > > > > >> > > > > >> > > based on
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > > > > metadata.version once we
> want to
> > > > make it
> > > > > > >> the
> > > > > > >> > > > > >> default.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs
> don't
> > > > mention
> > > > > > >> > the
> > > > > > >> > > > > >> required
> > > > > > >> > > > > >> > > ACLs,
> > > > > > >> > > > > >> > > > > it
> > > > > > >> > > > > >> > > > > > > > > would be
> > > > > > >> > > > > >> > > > > > > > > > > > good to add that for
> consistency.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > > > > >> > ConsumerGroupHeartbeat
> > > > > > >> > > > > >> > requires
> > > > > > >> > > > > >> > > > > "Read
> > > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > > >> > > > > >> > > > > > > > > > > > though it seems to do more
> than
> > > > reading.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > I had that thought myself as
> well,
> > > > but in
> > > > > > >> the
> > > > > > >> > end we
> > > > > > >> > > > > >> > could
> > > > > > >> > > > > >> > > not
> > > > > > >> > > > > >> > > > > > > find a
> > > > > > >> > > > > >> > > > > > > > > > > better alternative: adding
> Write
> > > > Group
> > > > > > >> seems
> > > > > > >> > an
> > > > > > >> > > > > >> overkill
> > > > > > >> > > > > >> > > here
> > > > > > >> > > > > >> > > > > > > since we
> > > > > > >> > > > > >> > > > > > > > > do
> > > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we only
> have
> > > > Read /
> > > > > > >> > Delete
> > > > > > >> > > > > and
> > > > > > >> > > > > >> > > Describe
> > > > > > >> > > > > >> > > > > on
> > > > > > >> > > > > >> > > > > > > > > groups so
> > > > > > >> > > > > >> > > > > > > > > > > far). Would like to hear others
> > > > thoughts.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic recreation
> > > > handled by
> > > > > > >> the
> > > > > > >> > > > > consumer
> > > > > > >> > > > > >> > with
> > > > > > >> > > > > >> > > the
> > > > > > >> > > > > >> > > > > new
> > > > > > >> > > > > >> > > > > > > > > group
> > > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be good
> to have
> > > > a
> > > > > > >> > section on
> > > > > > >> > > > > >> this.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > You mean with regex
> subscription
> > > > right? Yes
> > > > > > >> > we can
> > > > > > >> > > > > >> add a
> > > > > > >> > > > > >> > > > > section
> > > > > > >> > > > > >> > > > > > > about
> > > > > > >> > > > > >> > > > > > > > > > > that, but basically the idea
> is that
> > > > > > >> consumer
> > > > > > >> > would
> > > > > > >> > > > > be
> > > > > > >> > > > > >> > > totally
> > > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > > >> > > > > >> > > > > > > > > > > the new protocol as it's
> handled all
> > > > by the
> > > > > > >> > brokers.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will
> write
> > > > the
> > > > > > >> new
> > > > > > >> > > > > >> coordinator
> > > > > > >> > > > > >> > in
> > > > > > >> > > > > >> > > > > Java.
> > > > > > >> > > > > >> > > > > > > Even
> > > > > > >> > > > > >> > > > > > > > > though
> > > > > > >> > > > > >> > > > > > > > > > > > this is an implementation
> detail,
> > > > do we
> > > > > > >> > plan to
> > > > > > >> > > > > >> have a
> > > > > > >> > > > > >> > > new
> > > > > > >> > > > > >> > > > > gradle
> > > > > > >> > > > > >> > > > > > > > > module
> > > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > We have not thought about
> that. But
> > > > I think
> > > > > > >> > the
> > > > > > >> > > > > answer
> > > > > > >> > > > > >> > > should
> > > > > > >> > > > > >> > > > > be
> > > > > > >> > > > > >> > > > > > > yes.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability
> goal
> > > > when it
> > > > > > >> > comes to
> > > > > > >> > > > > >> how
> > > > > > >> > > > > >> > > many
> > > > > > >> > > > > >> > > > > > > members
> > > > > > >> > > > > >> > > > > > > > > the new
> > > > > > >> > > > > >> > > > > > > > > > > > group protocol can support?
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > Within a group, I think we
> should
> > > > shoot for
> > > > > > >> > 1000s of
> > > > > > >> > > > > >> > > members.
> > > > > > >> > > > > >> > > > > But
> > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > >> > > > > > > > > > > scalability goals also depend
> on the
> > > > offset
> > > > > > >> > > > > management
> > > > > > >> > > > > >> > > (commit,
> > > > > > >> > > > > >> > > > > > > fetch)
> > > > > > >> > > > > >> > > > > > > > > > > capabilities of the coordinator
> > > > which we
> > > > > > >> did
> > > > > > >> > not
> > > > > > >> > > > > >> cover in
> > > > > > >> > > > > >> > > this
> > > > > > >> > > > > >> > > > > > > KIP, so
> > > > > > >> > > > > >> > > > > > > > > it's
> > > > > > >> > > > > >> > > > > > > > > > > hard to give a number that
> applies
> > > > > > >> > universally.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> > > > > > >> > SubscribedTopidIds
> > > > > > >> > > > > >> instead
> > > > > > >> > > > > >> > of
> > > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > > >> > > > > >> > > Is the
> > > > > > >> > > > > >> > > > > > > idea
> > > > > > >> > > > > >> > > > > > > > > that
> > > > > > >> > > > > >> > > > > > > > > > > > since we have to resolve the
> regex
> > > > on the
> > > > > > >> > server,
> > > > > > >> > > > > we
> > > > > > >> > > > > >> > can
> > > > > > >> > > > > >> > > do
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > same
> > > > > > >> > > > > >> > > > > > > > > for
> > > > > > >> > > > > >> > > > > > > > > > > > the topic name? The
> difference is
> > > > that
> > > > > > >> > sending the
> > > > > > >> > > > > >> > regex
> > > > > > >> > > > > >> > > is
> > > > > > >> > > > > >> > > > > more
> > > > > > >> > > > > >> > > > > > > > > efficient
> > > > > > >> > > > > >> > > > > > > > > > > > whereas sending the topic
> names is
> > > > less
> > > > > > >> > efficient.
> > > > > > >> > > > > >> > > > > Furthermore,
> > > > > > >> > > > > >> > > > > > > > > delete and
> > > > > > >> > > > > >> > > > > > > > > > > > recreation is easier to
> handle if
> > > > we have
> > > > > > >> > topic
> > > > > > >> > > > > ids.
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > The main reason to still let
> the
> > > > clients
> > > > > > >> send
> > > > > > >> > names
> > > > > > >> > > > > >> is to
> > > > > > >> > > > > >> > > keep
> > > > > > >> > > > > >> > > > > the
> > > > > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids on
> the
> > > > broker /
> > > > > > >> > admin
> > > > > > >> > > > > client
> > > > > > >> > > > > >> > > only.
> > > > > > >> > > > > >> > > > > Note
> > > > > > >> > > > > >> > > > > > > that
> > > > > > >> > > > > >> > > > > > > > > > > although we added topic id in
> > > > KIP-516, we
> > > > > > >> > never
> > > > > > >> > > > > >> > > implemented the
> > > > > > >> > > > > >> > > > > > > logic
> > > > > > >> > > > > >> > > > > > > > > on
> > > > > > >> > > > > >> > > > > > > > > > > consumer/producers leveraging
> the
> > > > related
> > > > > > >> > newer
> > > > > > >> > > > > >> versioned
> > > > > > >> > > > > >> > > RPCs,
> > > > > > >> > > > > >> > > > > > > > > instead we
> > > > > > >> > > > > >> > > > > > > > > > > just set the topic id as empty
> UUID.
> > > > We
> > > > > > >> want
> > > > > > >> > to keep
> > > > > > >> > > > > >> the
> > > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > > >> > > > > >> > > > > > > > > > > to be thin and only delegate
> the
> > > > reasoning
> > > > > > >> on
> > > > > > >> > broker
> > > > > > >> > > > > >> and
> > > > > > >> > > > > >> > > > > > > potentially
> > > > > > >> > > > > >> > > > > > > > > admin
> > > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45
> AM
> > > > David
> > > > > > >> Jacot
> > > > > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> > > > discussion
> > > > > > >> thread
> > > > > > >> > on
> > > > > > >> > > > > >> KIP-848:
> > > > > > >> > > > > >> > > The
> > > > > > >> > > > > >> > > > > Next
> > > > > > >> > > > > >> > > > > > > > > > > > > Generation of the Consumer
> > > > Rebalance
> > > > > > >> > Protocol.
> > > > > > >> > > > > >> With
> > > > > > >> > > > > >> > > this
> > > > > > >> > > > > >> > > > > KIP,
> > > > > > >> > > > > >> > > > > > > we
> > > > > > >> > > > > >> > > > > > > > > aim
> > > > > > >> > > > > >> > > > > > > > > > > > > to make the rebalance
> protocol
> > > > (for
> > > > > > >> > consumers)
> > > > > > >> > > > > >> more
> > > > > > >> > > > > >> > > > > reliable,
> > > > > > >> > > > > >> > > > > > > more
> > > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to
> implement for
> > > > > > >> > clients, and
> > > > > > >> > > > > >> easier
> > > > > > >> > > > > >> > > to
> > > > > > >> > > > > >> > > > > debug
> > > > > > >> > > > > >> > > > > > > for
> > > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > > >> > > > > >> > > > >
> https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > > Please take a look and let
> me
> > > > know what
> > > > > > >> > you
> > > > > > >> > > > > think.
> > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from
> July
> > > > 18th to
> > > > > > >> > August 8th.
> > > > > > >> > > > > >> That
> > > > > > >> > > > > >> > > gives
> > > > > > >> > > > > >> > > > > > > you a
> > > > > > >> > > > > >> > > > > > > > > bit
> > > > > > >> > > > > >> > > > > > > > > > > > > of time to read and digest
> this
> > > > long
> > > > > > >> KIP.
> > > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > >
> > > > > > >> > > > > >> > > > > > > > > > > --
> > > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > > >> > > > > >> > > > > > > > >
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > >
> > > > > > >> > > > > >> > > > > > > > --
> > > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > > >> > > > > >> > > > > > >
> > > > > > >> > > > > >> > > > >
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> > >
> > > > > > >> > > > > >> >
> > > > > > >> > > > > >>
> > > > > > >> > > > > >>
> > > > > > >> > > > > >> --
> > > > > > >> > > > > >> -- Guozhang
> > > > > > >> > > > > >>
> > > > > > >> > > > > >
> > > > > > >> > > > >
> > > > > > >> >
> > > > > > >>
> > > > > > >
> > > >
>


-- 
-- Guozhang

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Jun,

I have updated the KIP to include your feedback. I have also tried to
clarify the parts which were not cleared.

Best,
David

On Fri, Sep 2, 2022 at 4:18 PM David Jacot <dj...@confluent.io> wrote:
>
> Hi Jun,
>
> Thanks for your feedback. Let me start by answering your questions
> inline and I will update the KIP next week.
>
> > Thanks for the KIP. Overall, the main benefits of the KIP seem to be fewer
> > RPCs during rebalance and more efficient support of wildcard. A few
> > comments below.
>
> I would also add that the KIP removes the global sync barrier in the
> protocol which is essential to improve group stability and
> scalability, and the KIP also simplifies the client by moving most of
> the logic to the server side.
>
> > 30. ConsumerGroupHeartbeatRequest
> > 30.1 ServerAssignor is a singleton. Do we plan to support rolling changing
> > of the partition assignor in the consumers?
>
> Definitely. The group coordinator will use the assignor used by a
> majority of the members. This allows the group to move from one
> assignor to another by a roll. This is explained in the Assignor
> Selection chapter.
>
> > 30.2 For each field, could you explain whether it's required in every
> > request or the scenarios when it needs to be filled? For example, it's not
> > clear to me when TopicPartitions needs to be filled.
>
> The client is expected to set those fields in case of a connection
> issue (e.g. timeout) or when the fields have changed since the last
> HB. The server populates those fields as long as the member is not
> fully reconciled - the member should acknowledge that it has the
> expected epoch and assignment. I will clarify this in the KIP.
>
> > 31. In the current consumer protocol, the rack affinity between the client
> > and the broker is only considered during fetching, but not during assigning
> > partitions to consumers. Sometimes, once the assignment is made, there is
> > no opportunity for read affinity because no replicas of assigned partitions
> > are close to the member. I am wondering if we should use this opportunity
> > to address this by including rack in GroupMember.
>
> That's an interesting idea. I don't see any issue with adding the rack
> to the members. I will do so.
>
> > 32. On the metric side, often, it's useful to know how busy a group
> > coordinator is. By moving the event loop model, it seems that we could add
> > a metric that tracks the fraction of the time the event loop is doing the
> > actual work.
>
> That's a great idea. I will add it. Thanks.
>
> > 33. Could we add a section on coordinator failover handling? For example,
> > does it need to trigger the check if any group with the wildcard
> > subscription now has a new matching topic?
>
> Sure. When the new group coordinator takes over, it has to:
> * Setup the session timeouts.
> * Trigger a new assignment if a client side assignor is used. We don't
> store the information about the member selected to run the assignment
> so we have to start a new one.
> * Update the topics metadata, verify the wildcard subscriptions, and
> trigger a rebalance if needed.
>
> > 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> > ConsumerGroupMemberMetadataValue: Could we document what the epoch field
> > reflects? For example, does the epoch in ConsumerGroupMetadataValue reflect
> > the latest group epoch? What about the one in
> > ConsumerGroupPartitionMetadataValue and ConsumerGroupMemberMetadataValue?
>
> Sure. I will clarify that but it is always the latest group epoch.
> When the group state is updated, the group epoch is bumped so we use
> that one for all the change records related to the update.
>
> > 35. "the group coordinator will ensure that the following invariants are
> > met: ... All members exists." It's possible for a member not to get any
> > assigned partitions, right?
>
> That's right. Here I meant that the members provided by the assignor
> in the assignment must exist in the group. The assignor can not make
> up new member ids.
>
> > 36. "He can rejoins the group with a member epoch equals to 0": When would
> > a consumer rejoin and what member id would be used?
>
> A member is expected to abandon all its partitions and rejoins when it
> receives the FENCED_MEMBER_EPOCH error. In this case, the group
> coordinator will have removed the member from the group. The member
> can rejoin the group with the same member id but with 0 as epoch. Let
> me see if I can clarify this in the KIP.
>
> > 37. "Instead, power users will have the ability to trigger a reassignment
> > by either providing a non-zero reason or by updating the assignor
> > metadata." Hmm, this seems to be conflicting with the deprecation of
> > Consumer#enforeRebalance.
>
> In this case, a new assignment is triggered by the client side
> assignor. When constructing the HB, the consumer will always consult
> the client side assignor and propagate the information to the group
> coordinator. In other words, we don't expect users to call
> Consumer#enforceRebalance anymore.
>
> > 38. The reassignment examples are nice. But the section seems to have
> > multiple typos.
> > 38.1 When the group transitions to epoch 2, B immediately gets into
> > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > 38.2 When the group transitions to epoch 3, C seems to get into epoch=3,
> > partitions=[foo-1] too early.
> > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > partitions=[foo-0].
>
> Sorry for that! I will revise them.
>
> > 39. Rolling upgrade of consumers: Do we support the upgrade from any old
> > version to new one?
>
> We will support upgrading from the consumer protocol version 3,
> introduced in KIP-792. KIP-792 is not implemented yet so the earliest
> version is unknown at the moment. This is explained in the migration
> plan chapter.
>
> Thanks again for your feedback, Jun. I will update the KIP based on it
> next week.
>
> Best,
> David
>
> On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <ju...@confluent.io.invalid> wrote:
> >
> > Hi, David,
> >
> > Thanks for the KIP. Overall, the main benefits of the KIP seem to be fewer
> > RPCs during rebalance and more efficient support of wildcard. A few
> > comments below.
> >
> > 30. ConsumerGroupHeartbeatRequest
> > 30.1 ServerAssignor is a singleton. Do we plan to support rolling changing
> > of the partition assignor in the consumers?
> > 30.2 For each field, could you explain whether it's required in every
> > request or the scenarios when it needs to be filled? For example, it's not
> > clear to me when TopicPartitions needs to be filled.
> >
> > 31. In the current consumer protocol, the rack affinity between the client
> > and the broker is only considered during fetching, but not during assigning
> > partitions to consumers. Sometimes, once the assignment is made, there is
> > no opportunity for read affinity because no replicas of assigned partitions
> > are close to the member. I am wondering if we should use this opportunity
> > to address this by including rack in GroupMember.
> >
> > 32. On the metric side, often, it's useful to know how busy a group
> > coordinator is. By moving the event loop model, it seems that we could add
> > a metric that tracks the fraction of the time the event loop is doing the
> > actual work.
> >
> > 33. Could we add a section on coordinator failover handling? For example,
> > does it need to trigger the check if any group with the wildcard
> > subscription now has a new matching topic?
> >
> > 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> > ConsumerGroupMemberMetadataValue: Could we document what the epoch field
> > reflects? For example, does the epoch in ConsumerGroupMetadataValue reflect
> > the latest group epoch? What about the one in
> > ConsumerGroupPartitionMetadataValue and ConsumerGroupMemberMetadataValue?
> >
> > 35. "the group coordinator will ensure that the following invariants are
> > met: ... All members exists." It's possible for a member not to get any
> > assigned partitions, right?
> >
> > 36. "He can rejoins the group with a member epoch equals to 0": When would
> > a consumer rejoin and what member id would be used?
> >
> > 37. "Instead, power users will have the ability to trigger a reassignment
> > by either providing a non-zero reason or by updating the assignor
> > metadata." Hmm, this seems to be conflicting with the deprecation of
> > Consumer#enforeRebalance.
> >
> > 38. The reassignment examples are nice. But the section seems to have
> > multiple typos.
> > 38.1 When the group transitions to epoch 2, B immediately gets into
> > "epoch=1, partitions=[foo-2]", which seems incorrect.
> > 38.2 When the group transitions to epoch 3, C seems to get into epoch=3,
> > partitions=[foo-1] too early.
> > 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> > partitions=[foo-0].
> >
> > 39. Rolling upgrade of consumers: Do we support the upgrade from any old
> > version to new one?
> >
> > Thanks,
> >
> > Jun
> >
> > On Mon, Aug 29, 2022 at 9:20 AM David Jacot <dj...@confluent.io.invalid>
> > wrote:
> >
> > > Hi all,
> > >
> > > The KIP states that we will re-implement the coordinator in Java. I
> > > discussed this offline with a few folks and folks are concerned that
> > > we could introduce many regressions in the old protocol if we do so.
> > > Therefore, I am going to remove this statement from the KIP. It is an
> > > implementation detail after all so it does not have to be decided at
> > > this stage. We will likely start by trying to refactor the current
> > > implementation as a first step.
> > >
> > > Cheers,
> > > David
> > >
> > > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <dj...@confluent.io> wrote:
> > > >
> > > > Hi Luke,
> > > >
> > > > > 1.1. I think the state machine are: "Empty, assigning, reconciling,
> > > stable,
> > > > > dead" mentioned in Consumer Group States section, right?
> > > >
> > > > This sentence does not refer to those group states but rather to a
> > > > state machine replication (SMR). This refers to the entire state of
> > > > group coordinator which is replicated via the log layer. I will
> > > > clarify this in the KIP.
> > > >
> > > > > 1.2. What do you mean "each state machine is modelled as an event
> > > loop"?
> > > >
> > > > The idea is to follow a model similar to the new quorum controller. We
> > > > will have N threads to process events. Each __consumer_offsets
> > > > partition is assigned to a unique thread and all the events (e.g.
> > > > requests, callbacks, etc.) are processed by this thread. This simplify
> > > > concurrency and will enable us to do simulation testing for the group
> > > > coordinator.
> > > >
> > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > partitions?
> > > > > Not a state machine "per consumer group" owned by a group coordinator?
> > > For
> > > > > example, if one group coordinator owns 2 consumer groups, and both
> > > exist in
> > > > > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> > > >
> > > > See 1.1. The confusion comes from there, I think.
> > > >
> > > > > 1.4. I know the "*group.coordinator.threads" *is the number of threads
> > > used
> > > > > to run the state machines. But I'm wondering if the purpose of the
> > > threads
> > > > > is only to keep the state of each consumer group (or
> > > *__consumer_offsets*
> > > > > partitions?), and no heavy computation, why should we need
> > > multi-threads
> > > > > here?
> > > >
> > > > See 1.2. The idea is to have an ability to shard the processing as the
> > > > computation could be heavy.
> > > >
> > > > > 2.1. The consumer session timeout, why does the default session
> > > timeout not
> > > > > locate between min (45s) and max(60s)? I thought the min/max session
> > > > > timeout is to define lower/upper bound of it, no?
> > > > >
> > > > > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > > > > failures when using the consumer group protocol.
> > > > > group.consumer.min.session.timeout.ms int 45s The minimum session
> > > timeout.
> > > > > group.consumer.max.session.timeout.ms int 60s The maximum session
> > > timeout.
> > > >
> > > > This is indeed a mistake. The default session timeout should be 45s
> > > > (the current default).
> > > >
> > > > > 2.2. The default server side assignor are [range, uniform], which means
> > > > > we'll default to "range" assignor. I'd like to know why not uniform
> > > one? I
> > > > > thought usually users will choose uniform assignor (former sticky
> > > assinor)
> > > > > for better evenly distribution. Any other reason we choose range
> > > assignor
> > > > > as default?
> > > > > group.consumer.assignors List range, uniform The server side assignors.
> > > >
> > > > The order on the server side has no influence because the client must
> > > > chose the selector that he wants to use. There is no default in the
> > > > current proposal. If the assignor is not specified by the client, the
> > > > request is rejected. The default client value for
> > > > `group.remote.assignor` is `uniform` though.
> > > >
> > > > Thanks for your very good comments, Luke. I hope that my answers help
> > > > to clarify things. I will update the KIP as well based on your
> > > > feedback.
> > > >
> > > > Cheers,
> > > > David
> > > >
> > > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com> wrote:
> > > > >
> > > > > Hi David,
> > > > >
> > > > > Thanks for the update.
> > > > >
> > > > > Some more questions:
> > > > > 1. In Group Coordinator section, you mentioned:
> > > > > > The new group coordinator will have a state machine per
> > > > > *__consumer_offsets* partitions, where each state machine is modelled
> > > as an
> > > > > event loop. Those state machines will be executed in
> > > > > *group.coordinator.threads* threads.
> > > > >
> > > > > 1.1. I think the state machine are: "Empty, assigning, reconciling,
> > > stable,
> > > > > dead" mentioned in Consumer Group States section, right?
> > > > > 1.2. What do you mean "each state machine is modelled as an event
> > > loop"?
> > > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > > partitions?
> > > > > Not a state machine "per consumer group" owned by a group coordinator?
> > > For
> > > > > example, if one group coordinator owns 2 consumer groups, and both
> > > exist in
> > > > > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> > > > > 1.4. I know the "*group.coordinator.threads" *is the number of threads
> > > used
> > > > > to run the state machines. But I'm wondering if the purpose of the
> > > threads
> > > > > is only to keep the state of each consumer group (or
> > > *__consumer_offsets*
> > > > > partitions?), and no heavy computation, why should we need
> > > multi-threads
> > > > > here?
> > > > >
> > > > > 2. For the default value in the new configs:
> > > > > 2.1. The consumer session timeout, why does the default session
> > > timeout not
> > > > > locate between min (45s) and max(60s)? I thought the min/max session
> > > > > timeout is to define lower/upper bound of it, no?
> > > > >
> > > > > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > > > > failures when using the consumer group protocol.
> > > > > group.consumer.min.session.timeout.ms int 45s The minimum session
> > > timeout.
> > > > > group.consumer.max.session.timeout.ms int 60s The maximum session
> > > timeout.
> > > > >
> > > > >
> > > > >
> > > > > 2.2. The default server side assignor are [range, uniform], which means
> > > > > we'll default to "range" assignor. I'd like to know why not uniform
> > > one? I
> > > > > thought usually users will choose uniform assignor (former sticky
> > > assinor)
> > > > > for better evenly distribution. Any other reason we choose range
> > > assignor
> > > > > as default?
> > > > > group.consumer.assignors List range, uniform The server side assignors.
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > Thank you.
> > > > > Luke
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com> wrote:
> > > > >
> > > > > > Hi Sagar,
> > > > > >
> > > > > > I have some thoughts about Kafka Connect integrating with KIP-848,
> > > but I
> > > > > > think we should have a separate discussion thread for the Kafka
> > > Connect
> > > > > > KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol
> > > [1],
> > > > > > and let this discussion thread focus on consumer rebalance protocol,
> > > WDYT?
> > > > > >
> > > > > > [1]
> > > > > >
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > >
> > > > > > Thank you.
> > > > > > Luke
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com>
> > > wrote:
> > > > > >
> > > > > >> Thank you Guozhang/David for the feedback. Looks like there's
> > > agreement on
> > > > > >> using separate APIs for Connect. I would revisit the doc and see
> > > what
> > > > > >> changes are to be made.
> > > > > >>
> > > > > >> Thanks!
> > > > > >> Sagar.
> > > > > >>
> > > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > > <dj...@confluent.io.invalid>
> > > > > >> wrote:
> > > > > >>
> > > > > >> > Hi Sagar,
> > > > > >> >
> > > > > >> > Thanks for the feedback and the document. That's really helpful. I
> > > > > >> > will take a look at it.
> > > > > >> >
> > > > > >> > Overall, it seems to me that both Connect and the Consumer could
> > > share
> > > > > >> > the same underlying "engine". The main difference is that the
> > > Consumer
> > > > > >> > assigns topic-partitions to members whereas Connect assigns tasks
> > > to
> > > > > >> > workers. I see two ways to move forward:
> > > > > >> > 1) We extend the new proposed APIs to support different resource
> > > types
> > > > > >> > (e.g. partitions, tasks, etc.); or
> > > > > >> > 2) We use new dedicated APIs for Connect. The dedicated APIs
> > > would be
> > > > > >> > similar to the new ones but different on the content/resources and
> > > > > >> > they would rely on the same engine on the coordinator side.
> > > > > >> >
> > > > > >> > I personally lean towards 2) because I am not a fan of
> > > overcharging
> > > > > >> > APIs to serve different purposes. That being said, I am not
> > > opposed to
> > > > > >> > 1) if we can find an elegant way to do it.
> > > > > >> >
> > > > > >> > I think that we can continue to discuss it here for now in order
> > > to
> > > > > >> > ensure that this KIP is compatible with what we will do for
> > > Connect in
> > > > > >> > the future.
> > > > > >> >
> > > > > >> > Best,
> > > > > >> > David
> > > > > >> >
> > > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io>
> > > wrote:
> > > > > >> > >
> > > > > >> > > Hi all,
> > > > > >> > >
> > > > > >> > > I am back from vacation. I will go through and address your
> > > comments
> > > > > >> > > in the coming days. Thanks for your feedback.
> > > > > >> > >
> > > > > >> > > Cheers,
> > > > > >> > > David
> > > > > >> > >
> > > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > > gharris1727@gmail.com
> > > > > >> >
> > > > > >> > wrote:
> > > > > >> > > >
> > > > > >> > > > Hey All!
> > > > > >> > > >
> > > > > >> > > > Thanks for the KIP, it's wonderful to see cooperative
> > > rebalancing
> > > > > >> > making it
> > > > > >> > > > down the stack!
> > > > > >> > > >
> > > > > >> > > > I had a few questions:
> > > > > >> > > >
> > > > > >> > > > 1. The 'Rejected Alternatives' section describes how member
> > > epoch
> > > > > >> > should
> > > > > >> > > > advance in step with the group epoch and assignment epoch
> > > values. I
> > > > > >> > think
> > > > > >> > > > that this is a good idea for the reasons described in the
> > > KIP. When
> > > > > >> the
> > > > > >> > > > protocol is incrementally assigning partitions to a worker,
> > > what
> > > > > >> member
> > > > > >> > > > epoch does each incremental assignment use? Are member epochs
> > > > > >> re-used,
> > > > > >> > and
> > > > > >> > > > a single member epoch can correspond to multiple different
> > > > > >> > (monotonically
> > > > > >> > > > larger) assignments?
> > > > > >> > > >
> > > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> > > > > >> coordinator? If
> > > > > >> > > > not, should custom client-side assignor implementations
> > > interact
> > > > > >> with
> > > > > >> > the
> > > > > >> > > > Reason field, and how is its common meaning agreed upon? If
> > > so, what
> > > > > >> > is the
> > > > > >> > > > benefit of a distinct Reason field over including such
> > > functionality
> > > > > >> > in the
> > > > > >> > > > opaque metadata?
> > > > > >> > > >
> > > > > >> > > > 3. The following is included in the KIP: "Thanks to this, the
> > > input
> > > > > >> of
> > > > > >> > the
> > > > > >> > > > client side assignor is entirely driven by the group
> > > coordinator.
> > > > > >> The
> > > > > >> > > > consumer is no longer responsible for maintaining any state
> > > besides
> > > > > >> its
> > > > > >> > > > assigned partitions." Does this mean that the client-side
> > > assignor
> > > > > >> MAY
> > > > > >> > > > incorporate additional non-Metadata state (such as partition
> > > > > >> > throughput,
> > > > > >> > > > cpu/memory metrics, config topics, etc), or that additional
> > > > > >> > non-Metadata
> > > > > >> > > > state SHOULD NOT be used?
> > > > > >> > > >
> > > > > >> > > > 4. I see that there are separate classes
> > > > > >> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > > >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that
> > > seem to
> > > > > >> > > > overlap significantly. Is it possible for these two
> > > implementations
> > > > > >> to
> > > > > >> > be
> > > > > >> > > > unified? This would serve to promote feature parity of
> > > server-side
> > > > > >> and
> > > > > >> > > > client-side assignors, and would also facilitate operational
> > > > > >> > flexibility in
> > > > > >> > > > certain situations. For example, if a server-side assignor
> > > has some
> > > > > >> > poor
> > > > > >> > > > behavior and needs a patch, deploying the patched assignor to
> > > the
> > > > > >> > client
> > > > > >> > > > and switching one consumer group to a client-side assignor
> > > may be
> > > > > >> > faster
> > > > > >> > > > and less risky than patching all of the brokers. With the
> > > currently
> > > > > >> > > > proposed distinct APIs, a non-trivial reimplementation would
> > > have
> > > > > >> to be
> > > > > >> > > > assembled, and if the two APIs have diverged significantly,
> > > then it
> > > > > >> is
> > > > > >> > > > possible that a reimplementation would not be possible.
> > > > > >> > > >
> > > > > >> > > > --
> > > > > >> > > > Greg Harris
> > > > > >> > > > gharris1727@gmail.com
> > > > > >> > > > github.com/gharris1727
> > > > > >> > > >
> > > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > > sagarmeansocean@gmail.com>
> > > > > >> > wrote:
> > > > > >> > > >
> > > > > >> > > > > Hi Guozhang/David,
> > > > > >> > > > >
> > > > > >> > > > > I created a confluence page to discuss how Connect would
> > > need to
> > > > > >> > change
> > > > > >> > > > > based on the new rebalance protocol. Here's the page:
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> >
> > > > > >>
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > > >> > > > >
> > > > > >> > > > > It's also pretty longish and I have tried to keep a format
> > > > > >> similar to
> > > > > >> > > > > KIP-848. Let me know what you think. Also, do you think this
> > > > > >> should
> > > > > >> > be
> > > > > >> > > > > moved to a separate discussion thread or is this one fine?
> > > > > >> > > > >
> > > > > >> > > > > Thanks!
> > > > > >> > > > > Sagar.
> > > > > >> > > > >
> > > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > > sagarmeansocean@gmail.com>
> > > > > >> > wrote:
> > > > > >> > > > >
> > > > > >> > > > > > Hello Guozhang,
> > > > > >> > > > > >
> > > > > >> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I
> > > would do
> > > > > >> > the
> > > > > >> > > > > > analysis and come up with such a document.
> > > > > >> > > > > >
> > > > > >> > > > > > Thanks!
> > > > > >> > > > > > Sagar.
> > > > > >> > > > > >
> > > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > > > >> wangguoz@gmail.com>
> > > > > >> > > > > wrote:
> > > > > >> > > > > >
> > > > > >> > > > > >> Hello Sagar,
> > > > > >> > > > > >>
> > > > > >> > > > > >> It would be great if you could come back with some
> > > analysis on
> > > > > >> > how to
> > > > > >> > > > > >> implement the Connect side integration with the new
> > > protocol;
> > > > > >> so
> > > > > >> > far
> > > > > >> > > > > >> besides leveraging on the new "protocol type" we did not
> > > yet
> > > > > >> think
> > > > > >> > > > > through
> > > > > >> > > > > >> the Connect side implementations. For Streams here's a
> > > draft of
> > > > > >> > > > > >> integration
> > > > > >> > > > > >> plan:
> > > > > >> > > > > >>
> > > > > >> > > > > >>
> > > > > >> > > > >
> > > > > >> >
> > > > > >>
> > > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > > >> > > > > >>
> > > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > > >> sagarmeansocean@gmail.com
> > > > > >> > >
> > > > > >> > > > > wrote:
> > > > > >> > > > > >>
> > > > > >> > > > > >> > Hi David,
> > > > > >> > > > > >> >
> > > > > >> > > > > >> > Thank you for your response. The reason I thought
> > > connect can
> > > > > >> > also fit
> > > > > >> > > > > >> into
> > > > > >> > > > > >> > this new scheme is that even today the connect uses a
> > > > > >> > > > > WorkerCoordinator
> > > > > >> > > > > >> > extending from AbstractCoordinator to empower
> > > rebalances of
> > > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> > > > > >> protocolType()
> > > > > >> > to
> > > > > >> > > > > >> connect
> > > > > >> > > > > >> > and uses the metadata() method by plumbing into
> > > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > > >> > > > > >> >
> > > > > >> > > > > >> > I think the changes to support connect would be
> > > similar at a
> > > > > >> > high
> > > > > >> > > > > level
> > > > > >> > > > > >> to
> > > > > >> > > > > >> > the changes in streams mainly because of the Client
> > > side
> > > > > >> > assignors
> > > > > >> > > > > being
> > > > > >> > > > > >> > used in both. At an implementation level, we might
> > > need to
> > > > > >> make
> > > > > >> > a lot
> > > > > >> > > > > of
> > > > > >> > > > > >> > changes to get onto this new assignment protocol like
> > > > > >> enhancing
> > > > > >> > the
> > > > > >> > > > > >> > JoinGroup request/response and SyncGroup and using
> > > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > > >> > > > > >> > API etc again on similar lines to streams (or there
> > > might be
> > > > > >> > > > > >> deviations). I
> > > > > >> > > > > >> > would try to perform a detailed analysis of the same
> > > and we
> > > > > >> > can have
> > > > > >> > > > > a
> > > > > >> > > > > >> > separate discussion thread for that as that would
> > > derail this
> > > > > >> > > > > discussion
> > > > > >> > > > > >> > thread. Let me know if that sounds good to you.
> > > > > >> > > > > >> >
> > > > > >> > > > > >> > Thanks!
> > > > > >> > > > > >> > Sagar.
> > > > > >> > > > > >> >
> > > > > >> > > > > >> >
> > > > > >> > > > > >> >
> > > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > >> > > > > >> >
> > > > > >> > > > > >> > wrote:
> > > > > >> > > > > >> >
> > > > > >> > > > > >> > > Hi Sagar,
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> > > Thanks for your comments.
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I
> > > can
> > > > > >> > mention it.
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> > > 2) The idea is to transition C from his current
> > > assignment
> > > > > >> to
> > > > > >> > his
> > > > > >> > > > > >> > > target assignment when he can move to epoch 3. When
> > > that
> > > > > >> > happens,
> > > > > >> > > > > the
> > > > > >> > > > > >> > > member assignment is updated and persisted with all
> > > its
> > > > > >> > assigned
> > > > > >> > > > > >> > > partitions even if they are not all revoked yet. In
> > > other
> > > > > >> > words, the
> > > > > >> > > > > >> > > member assignment becomes the target assignment.
> > > This is
> > > > > >> > basically
> > > > > >> > > > > an
> > > > > >> > > > > >> > > optimization to avoid having to write all the
> > > changes to
> > > > > >> the
> > > > > >> > log.
> > > > > >> > > > > The
> > > > > >> > > > > >> > > examples are based on the persisted state so I
> > > understand
> > > > > >> the
> > > > > >> > > > > >> > > confusion. Let me see if I can improve this in the
> > > > > >> > description.
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> > > 3) Regarding Connect, it could reuse the protocol
> > > with a
> > > > > >> > client side
> > > > > >> > > > > >> > > assignor if it fits in the protocol. The assignment
> > > is
> > > > > >> about
> > > > > >> > > > > >> > > topicid-partitions + metadata, could Connect fit
> > > into this?
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> > > Best,
> > > > > >> > > > > >> > > David
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > > >> > sagarmeansocean@gmail.com>
> > > > > >> > > > > >> wrote:
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > Hi David,
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > Thanks for the KIP. I just had minor observations:
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > 1) In the Assignment Error section in Client Side
> > > mode
> > > > > >> > Assignment
> > > > > >> > > > > >> > > process,
> > > > > >> > > > > >> > > > you mentioned => `In this case, the client side
> > > assignor
> > > > > >> can
> > > > > >> > > > > return
> > > > > >> > > > > >> an
> > > > > >> > > > > >> > > > error to the group coordinator`. In this case are
> > > you
> > > > > >> > referring to
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > Assignor returning an AssignmentError that's
> > > listed down
> > > > > >> > towards
> > > > > >> > > > > the
> > > > > >> > > > > >> > end?
> > > > > >> > > > > >> > > > If yes, do you think it would make sense to
> > > mention this
> > > > > >> > > > > explicitly
> > > > > >> > > > > >> > here?
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > 2) In the Case Studies section, I have a slight
> > > > > >> confusion,
> > > > > >> > not
> > > > > >> > > > > sure
> > > > > >> > > > > >> if
> > > > > >> > > > > >> > > > others have the same. Consider this step:
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > When B heartbeats, the group coordinator
> > > transitions him
> > > > > >> to
> > > > > >> > epoch
> > > > > >> > > > > 3
> > > > > >> > > > > >> > > because
> > > > > >> > > > > >> > > > B has no partitions to revoke. It persists the
> > > change and
> > > > > >> > reply.
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > > >> > > > > >> > > >       - A
> > > > > >> > > > > >> > > >       - B
> > > > > >> > > > > >> > > >       - C
> > > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > >> > > > > >> > > >    - Member Assignment
> > > > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but
> > > cannot
> > > > > >> get
> > > > > >> > foo-1
> > > > > >> > > > > >> yet.
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > Here,it's mentioned that member C can't get the
> > > foo-1
> > > > > >> > partition
> > > > > >> > > > > yet,
> > > > > >> > > > > >> > but
> > > > > >> > > > > >> > > > based on the description above, it seems it
> > > already has
> > > > > >> it.
> > > > > >> > Do you
> > > > > >> > > > > >> > think
> > > > > >> > > > > >> > > it
> > > > > >> > > > > >> > > > would be better to remove it and populate it only
> > > when it
> > > > > >> > actually
> > > > > >> > > > > >> gets
> > > > > >> > > > > >> > > it?
> > > > > >> > > > > >> > > > I see this in a lot of other places, so have I
> > > > > >> understood it
> > > > > >> > > > > >> > incorrectly
> > > > > >> > > > > >> > > ?
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > Regarding connect , it might be out of scope of
> > > this
> > > > > >> > discussion,
> > > > > >> > > > > but
> > > > > >> > > > > >> > from
> > > > > >> > > > > >> > > > what I understood it would probably be running in
> > > client
> > > > > >> > side
> > > > > >> > > > > >> assignor
> > > > > >> > > > > >> > > mode
> > > > > >> > > > > >> > > > even on the new rebalance protocol as it has its
> > > own
> > > > > >> Custom
> > > > > >> > > > > >> > > Assignors(Eager
> > > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > Thanks!
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > Sagar.
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > wrote:
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
> > > > > >> > specialized API
> > > > > >> > > > > >> > > > > instead of relying on one generic API. For
> > > Connect, we
> > > > > >> > can apply
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > exact same pattern and reuse/share the core
> > > > > >> > implementation on
> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > server side. For the schema registry, I think
> > > that we
> > > > > >> > should
> > > > > >> > > > > >> consider
> > > > > >> > > > > >> > > > > having a tailored API to do simple
> > > membership/leader
> > > > > >> > election.
> > > > > >> > > > > >> > > > >
> > > > > >> > > > > >> > > > > Best,
> > > > > >> > > > > >> > > > > David
> > > > > >> > > > > >> > > > >
> > > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> > > > > >> > ismael@juma.me.uk
> > > > > >> > > > > >
> > > > > >> > > > > >> > > wrote:
> > > > > >> > > > > >> > > > > >
> > > > > >> > > > > >> > > > > > Three quick comments:
> > > > > >> > > > > >> > > > > >
> > > > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > > >> > > > > >> > > > > should
> > > > > >> > > > > >> > > > > > document the differences in more detail before
> > > > > >> deciding
> > > > > >> > one
> > > > > >> > > > > way
> > > > > >> > > > > >> or
> > > > > >> > > > > >> > > > > another.
> > > > > >> > > > > >> > > > > > That said, if people pass
> > > java.util.regex.Pattern,
> > > > > >> they
> > > > > >> > expect
> > > > > >> > > > > >> > their
> > > > > >> > > > > >> > > > > > semantics to be honored. If we are doing
> > > something
> > > > > >> > different,
> > > > > >> > > > > >> then
> > > > > >> > > > > >> > we
> > > > > >> > > > > >> > > > > > should consider adding an overload with our own
> > > > > >> Pattern
> > > > > >> > class
> > > > > >> > > > > (I
> > > > > >> > > > > >> > > don't
> > > > > >> > > > > >> > > > > > think we'd want to expose re2j's at this
> > > point).
> > > > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol
> > > should
> > > > > >> > > > > integrate
> > > > > >> > > > > >> > fully
> > > > > >> > > > > >> > > > > with
> > > > > >> > > > > >> > > > > > it and should handle the topic recreation case
> > > > > >> > correctly.
> > > > > >> > > > > That's
> > > > > >> > > > > >> > the
> > > > > >> > > > > >> > > main
> > > > > >> > > > > >> > > > > > part we need to handle. I agree with David
> > > that we'd
> > > > > >> > want to
> > > > > >> > > > > add
> > > > > >> > > > > >> > > topic
> > > > > >> > > > > >> > > > > ids
> > > > > >> > > > > >> > > > > > to the relevant protocols that don't have it
> > > yet and
> > > > > >> > that we
> > > > > >> > > > > can
> > > > > >> > > > > >> > > probably
> > > > > >> > > > > >> > > > > > focus on the internals versus adding new APIs
> > > to the
> > > > > >> > Java
> > > > > >> > > > > >> Consumer
> > > > > >> > > > > >> > > > > (unless
> > > > > >> > > > > >> > > > > > we find that adding new APIs is required for
> > > > > >> reasonable
> > > > > >> > > > > >> semantics).
> > > > > >> > > > > >> > > > > > 3. I am still not sure about the coordinator
> > > storing
> > > > > >> the
> > > > > >> > > > > >> configs.
> > > > > >> > > > > >> > > It's
> > > > > >> > > > > >> > > > > > powerful for configs to be centralized in the
> > > > > >> metadata
> > > > > >> > log for
> > > > > >> > > > > >> > > various
> > > > > >> > > > > >> > > > > > reasons (auditability, visibility, consistency,
> > > > > >> etc.).
> > > > > >> > > > > >> Similarly, I
> > > > > >> > > > > >> > > am
> > > > > >> > > > > >> > > > > not
> > > > > >> > > > > >> > > > > > sure about automatically deleting configs in a
> > > way
> > > > > >> that
> > > > > >> > they
> > > > > >> > > > > >> cannot
> > > > > >> > > > > >> > > be
> > > > > >> > > > > >> > > > > > recovered. A good property for modern systems
> > > is to
> > > > > >> > minimize
> > > > > >> > > > > the
> > > > > >> > > > > >> > > number
> > > > > >> > > > > >> > > > > of
> > > > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > > >> > > > > >> > > > > >
> > > > > >> > > > > >> > > > > > Ismael
> > > > > >> > > > > >> > > > > >
> > > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > > >> > > > > >> > > > > >
> > > > > >> > > > > >> > > > > > wrote:
> > > > > >> > > > > >> > > > > >
> > > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > > 1) the migration path, especially the last
> > > step
> > > > > >> when
> > > > > >> > > > > clients
> > > > > >> > > > > >> > > flip the
> > > > > >> > > > > >> > > > > > > flag
> > > > > >> > > > > >> > > > > > > > to enable the new protocol, in which we
> > > would
> > > > > >> have a
> > > > > >> > > > > window
> > > > > >> > > > > >> > where
> > > > > >> > > > > >> > > > > both
> > > > > >> > > > > >> > > > > > > new
> > > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs
> > > are
> > > > > >> used
> > > > > >> > by
> > > > > >> > > > > >> members
> > > > > >> > > > > >> > of
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > same
> > > > > >> > > > > >> > > > > > > > group. How the coordinator could "mimic"
> > > the old
> > > > > >> > behavior
> > > > > >> > > > > >> while
> > > > > >> > > > > >> > > > > using the
> > > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > > present
> > > > > >> about.
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > Noted. I just published a new version of KIP
> > > which
> > > > > >> > includes
> > > > > >> > > > > >> more
> > > > > >> > > > > >> > > > > > > details about this. See the "Supporting
> > > Online
> > > > > >> > Consumer
> > > > > >> > > > > Group
> > > > > >> > > > > >> > > Upgrade"
> > > > > >> > > > > >> > > > > > > and the "Compatibility, Deprecation, and
> > > Migration
> > > > > >> > Plan". I
> > > > > >> > > > > >> think
> > > > > >> > > > > >> > > that
> > > > > >> > > > > >> > > > > > > I have to think through a few cases now but
> > > the
> > > > > >> > overall idea
> > > > > >> > > > > >> and
> > > > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > > KIP-516 the
> > > > > >> > topic ids
> > > > > >> > > > > >> are
> > > > > >> > > > > >> > > only
> > > > > >> > > > > >> > > > > used
> > > > > >> > > > > >> > > > > > > as
> > > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but they
> > > are not
> > > > > >> > exposed
> > > > > >> > > > > via
> > > > > >> > > > > >> any
> > > > > >> > > > > >> > > > > public
> > > > > >> > > > > >> > > > > > > APIs
> > > > > >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> > > first
> > > > > >> > should we
> > > > > >> > > > > >> let
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > consumer
> > > > > >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> > > mapping
> > > > > >> > itself
> > > > > >> > > > > to
> > > > > >> > > > > >> > fully
> > > > > >> > > > > >> > > > > > > leverage
> > > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the
> > > new
> > > > > >> RPCs
> > > > > >> > with
> > > > > >> > > > > the
> > > > > >> > > > > >> > > topic
> > > > > >> > > > > >> > > > > ids;
> > > > > >> > > > > >> > > > > > > and
> > > > > >> > > > > >> > > > > > > > secondly, should we ever consider exposing
> > > the
> > > > > >> > topic ids
> > > > > >> > > > > in
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > consumer
> > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > subscribe/assign, as
> > > > > >> well
> > > > > >> > as in
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > rebalance
> > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > >> > deletion-and-recreation).
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > a) Assuming that we would include converting
> > > all
> > > > > >> the
> > > > > >> > offsets
> > > > > >> > > > > >> > > related
> > > > > >> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the
> > > consumer
> > > > > >> > would be
> > > > > >> > > > > >> able
> > > > > >> > > > > >> > to
> > > > > >> > > > > >> > > > > > > fully operate with topic ids. That being
> > > said, it
> > > > > >> > still has
> > > > > >> > > > > to
> > > > > >> > > > > >> > > provide
> > > > > >> > > > > >> > > > > > > the topics names in various APIs so having a
> > > > > >> mapping
> > > > > >> > in the
> > > > > >> > > > > >> > > consumer
> > > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > > >> > > > > >> > > > > > > b) I don't have a strong opinion on this.
> > > Here I
> > > > > >> > wonder if
> > > > > >> > > > > >> this
> > > > > >> > > > > >> > > goes
> > > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I would rather
> > > focus
> > > > > >> on
> > > > > >> > the
> > > > > >> > > > > >> > internals
> > > > > >> > > > > >> > > > > > > here and we can consider this separately if
> > > we see
> > > > > >> > value in
> > > > > >> > > > > >> doing
> > > > > >> > > > > >> > > it.
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > Coming back to Ismael's point about using
> > > topic ids
> > > > > >> > in the
> > > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that
> > > there
> > > > > >> is
> > > > > >> > one
> > > > > >> > > > > >> > advantage
> > > > > >> > > > > >> > > in
> > > > > >> > > > > >> > > > > > > favour of it. The consumer will have the
> > > > > >> opportunity
> > > > > >> > to
> > > > > >> > > > > >> validate
> > > > > >> > > > > >> > > that
> > > > > >> > > > > >> > > > > > > the topics exists before passing them into
> > > the
> > > > > >> group
> > > > > >> > > > > rebalance
> > > > > >> > > > > >> > > > > > > protocol. Obviously, the coordinator will
> > > also
> > > > > >> notice
> > > > > >> > it but
> > > > > >> > > > > >> it
> > > > > >> > > > > >> > > does
> > > > > >> > > > > >> > > > > > > not really have a way to reject an invalid
> > > topic in
> > > > > >> > the
> > > > > >> > > > > >> response.
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > > > > >> questions
> > > > > >> > > > > except
> > > > > >> > > > > >> for
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally
> > > I think
> > > > > >> > it's not
> > > > > >> > > > > >> > > necessary
> > > > > >> > > > > >> > > > > to
> > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern,
> > > but
> > > > > >> > instead we
> > > > > >> > > > > >> still
> > > > > >> > > > > >> > > use
> > > > > >> > > > > >> > > > > > > Pattern
> > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > subscription may
> > > > > >> be
> > > > > >> > > > > >> rejected by
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > server.
> > > > > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > > > > >> scenario
> > > > > >> > I felt
> > > > > >> > > > > >> > using
> > > > > >> > > > > >> > > an
> > > > > >> > > > > >> > > > > > > > overloaded `String` based subscription may
> > > be
> > > > > >> more
> > > > > >> > > > > >> vulnerable
> > > > > >> > > > > >> > to
> > > > > >> > > > > >> > > > > various
> > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > That could work. I have to look at the
> > > differences
> > > > > >> > between
> > > > > >> > > > > the
> > > > > >> > > > > >> > two
> > > > > >> > > > > >> > > > > > > engines to better understand the potential
> > > issues.
> > > > > >> My
> > > > > >> > > > > >> > > understanding is
> > > > > >> > > > > >> > > > > > > that would work for all the basic regular
> > > > > >> > expressions. The
> > > > > >> > > > > >> > > differences
> > > > > >> > > > > >> > > > > > > between the two are mainly about the various
> > > > > >> character
> > > > > >> > > > > >> classes. I
> > > > > >> > > > > >> > > > > > > wonder what other people think about this.
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > Best,
> > > > > >> > > > > >> > > > > > > David
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang
> > > Wang <
> > > > > >> > > > > >> > wangguoz@gmail.com
> > > > > >> > > > > >> > > >
> > > > > >> > > > > >> > > > > wrote:
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > Thanks David! I think on the high level
> > > there are
> > > > > >> > two meta
> > > > > >> > > > > >> > > points we
> > > > > >> > > > > >> > > > > need
> > > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > 1) the migration path, especially the last
> > > step
> > > > > >> when
> > > > > >> > > > > clients
> > > > > >> > > > > >> > > flip the
> > > > > >> > > > > >> > > > > > > flag
> > > > > >> > > > > >> > > > > > > > to enable the new protocol, in which we
> > > would
> > > > > >> have a
> > > > > >> > > > > window
> > > > > >> > > > > >> > where
> > > > > >> > > > > >> > > > > both
> > > > > >> > > > > >> > > > > > > new
> > > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs
> > > are
> > > > > >> used
> > > > > >> > by
> > > > > >> > > > > >> members
> > > > > >> > > > > >> > of
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > same
> > > > > >> > > > > >> > > > > > > > group. How the coordinator could "mimic"
> > > the old
> > > > > >> > behavior
> > > > > >> > > > > >> while
> > > > > >> > > > > >> > > > > using the
> > > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > > present
> > > > > >> about.
> > > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > > KIP-516 the
> > > > > >> > topic ids
> > > > > >> > > > > >> are
> > > > > >> > > > > >> > > only
> > > > > >> > > > > >> > > > > used
> > > > > >> > > > > >> > > > > > > as
> > > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but they
> > > are not
> > > > > >> > exposed
> > > > > >> > > > > via
> > > > > >> > > > > >> any
> > > > > >> > > > > >> > > > > public
> > > > > >> > > > > >> > > > > > > APIs
> > > > > >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> > > first
> > > > > >> > should we
> > > > > >> > > > > >> let
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > consumer
> > > > > >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> > > mapping
> > > > > >> > itself
> > > > > >> > > > > to
> > > > > >> > > > > >> > fully
> > > > > >> > > > > >> > > > > > > leverage
> > > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the
> > > new
> > > > > >> RPCs
> > > > > >> > with
> > > > > >> > > > > the
> > > > > >> > > > > >> > > topic
> > > > > >> > > > > >> > > > > ids;
> > > > > >> > > > > >> > > > > > > and
> > > > > >> > > > > >> > > > > > > > secondly, should we ever consider exposing
> > > the
> > > > > >> > topic ids
> > > > > >> > > > > in
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > consumer
> > > > > >> > > > > >> > > > > > > > public APIs as well (both
> > > subscribe/assign, as
> > > > > >> well
> > > > > >> > as in
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > rebalance
> > > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > > >> > deletion-and-recreation).
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > > > > >> questions
> > > > > >> > > > > except
> > > > > >> > > > > >> for
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally
> > > I think
> > > > > >> > it's not
> > > > > >> > > > > >> > > necessary
> > > > > >> > > > > >> > > > > to
> > > > > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern,
> > > but
> > > > > >> > instead we
> > > > > >> > > > > >> still
> > > > > >> > > > > >> > > use
> > > > > >> > > > > >> > > > > > > Pattern
> > > > > >> > > > > >> > > > > > > > while just documenting that our
> > > subscription may
> > > > > >> be
> > > > > >> > > > > >> rejected by
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > server.
> > > > > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > > > > >> scenario
> > > > > >> > I felt
> > > > > >> > > > > >> > using
> > > > > >> > > > > >> > > an
> > > > > >> > > > > >> > > > > > > > overloaded `String` based subscription may
> > > be
> > > > > >> more
> > > > > >> > > > > >> vulnerable
> > > > > >> > > > > >> > to
> > > > > >> > > > > >> > > > > various
> > > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > Guozhang
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > wrote:
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer
> > > your
> > > > > >> > questions
> > > > > >> > > > > >> > inline.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 1. I think it's premature to talk about
> > > > > >> target
> > > > > >> > > > > versions
> > > > > >> > > > > >> for
> > > > > >> > > > > >> > > > > > > deprecation
> > > > > >> > > > > >> > > > > > > > > and
> > > > > >> > > > > >> > > > > > > > > > removal of the existing group protocol.
> > > > > >> Unlike
> > > > > >> > KRaft,
> > > > > >> > > > > >> this
> > > > > >> > > > > >> > > > > affects a
> > > > > >> > > > > >> > > > > > > core
> > > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > > deprecation/removal
> > > > > >> > will be
> > > > > >> > > > > >> > heavily
> > > > > >> > > > > >> > > > > > > dependent
> > > > > >> > > > > >> > > > > > > > > on
> > > > > >> > > > > >> > > > > > > > > > how quickly applications migrate to
> > > the new
> > > > > >> > protocol.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to release
> > > this in
> > > > > >> > 4.x, but
> > > > > >> > > > > it
> > > > > >> > > > > >> > > wasn't
> > > > > >> > > > > >> > > > > made
> > > > > >> > > > > >> > > > > > > > > clear
> > > > > >> > > > > >> > > > > > > > > > why. If we added that as a way to
> > > estimate
> > > > > >> when
> > > > > >> > we'd
> > > > > >> > > > > >> > > deprecate
> > > > > >> > > > > >> > > > > and
> > > > > >> > > > > >> > > > > > > remove
> > > > > >> > > > > >> > > > > > > > > > the group protocol, I also suggest
> > > removing
> > > > > >> > this part.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> > > explained, I
> > > > > >> plan
> > > > > >> > to
> > > > > >> > > > > >> rewrite
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > group
> > > > > >> > > > > >> > > > > > > > > coordinator in Java while we implement
> > > the new
> > > > > >> > protocol.
> > > > > >> > > > > >> This
> > > > > >> > > > > >> > > means
> > > > > >> > > > > >> > > > > > > > > that the internals will be slightly
> > > different
> > > > > >> > (e.g.
> > > > > >> > > > > >> threading
> > > > > >> > > > > >> > > > > model).
> > > > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the
> > > switch from
> > > > > >> > the old
> > > > > >> > > > > >> group
> > > > > >> > > > > >> > > > > > > > > coordinator to the new group coordinator
> > > to a
> > > > > >> > major
> > > > > >> > > > > >> release.
> > > > > >> > > > > >> > > The
> > > > > >> > > > > >> > > > > > > > > alternative would be to use a flag to do
> > > the
> > > > > >> > switch
> > > > > >> > > > > >> instead
> > > > > >> > > > > >> > of
> > > > > >> > > > > >> > > > > relying
> > > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the details of
> > > the
> > > > > >> > migration
> > > > > >> > > > > >> story.
> > > > > >> > > > > >> > > It
> > > > > >> > > > > >> > > > > sounds
> > > > > >> > > > > >> > > > > > > > > like
> > > > > >> > > > > >> > > > > > > > > > we're saying we will support online
> > > > > >> migrations.
> > > > > >> > Is
> > > > > >> > > > > that
> > > > > >> > > > > >> > > correct?
> > > > > >> > > > > >> > > > > We
> > > > > >> > > > > >> > > > > > > > > should
> > > > > >> > > > > >> > > > > > > > > > explain this in detail. It could also
> > > be done
> > > > > >> > as a
> > > > > >> > > > > >> separate
> > > > > >> > > > > >> > > KIP,
> > > > > >> > > > > >> > > > > if
> > > > > >> > > > > >> > > > > > > it's
> > > > > >> > > > > >> > > > > > > > > > easier.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Yes, we will support online migrations
> > > for the
> > > > > >> > group.
> > > > > >> > > > > That
> > > > > >> > > > > >> > > means
> > > > > >> > > > > >> > > > > that
> > > > > >> > > > > >> > > > > > > > > a group using the old protocol will be
> > > able to
> > > > > >> > switch to
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > new
> > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Let me briefly explain how that will work
> > > > > >> though.
> > > > > >> > It is
> > > > > >> > > > > >> > > basically a
> > > > > >> > > > > >> > > > > > > > > four step process:
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or
> > > rolled to a
> > > > > >> > software
> > > > > >> > > > > >> > > supporting
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > new group coordinator. Both the old and
> > > the new
> > > > > >> > > > > >> coordinator
> > > > > >> > > > > >> > > will
> > > > > >> > > > > >> > > > > > > > > support the old protocol and rely on the
> > > same
> > > > > >> > persisted
> > > > > >> > > > > >> > > metadata so
> > > > > >> > > > > >> > > > > > > > > they can work together. This point is an
> > > > > >> offline
> > > > > >> > > > > >> migration.
> > > > > >> > > > > >> > We
> > > > > >> > > > > >> > > > > cannot
> > > > > >> > > > > >> > > > > > > > > do this one live because it would require
> > > > > >> > shutting down
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > current
> > > > > >> > > > > >> > > > > > > > > coordinator and starting up the new one
> > > and
> > > > > >> that
> > > > > >> > would
> > > > > >> > > > > >> cause
> > > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP
> > > must be
> > > > > >> > upgraded
> > > > > >> > > > > to
> > > > > >> > > > > >> X
> > > > > >> > > > > >> > in
> > > > > >> > > > > >> > > > > order
> > > > > >> > > > > >> > > > > > > > > to enable the new protocol. This cannot
> > > be done
> > > > > >> > before
> > > > > >> > > > > 1)
> > > > > >> > > > > >> is
> > > > > >> > > > > >> > > > > > > > > terminated because the old coordinator
> > > doesn't
> > > > > >> > support
> > > > > >> > > > > the
> > > > > >> > > > > >> > new
> > > > > >> > > > > >> > > > > > > > > protocol.
> > > > > >> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a
> > > version
> > > > > >> > > > > supporting
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > online
> > > > > >> > > > > >> > > > > > > > > migration (must have KIP-792). If the
> > > consumer
> > > > > >> is
> > > > > >> > > > > already
> > > > > >> > > > > >> > > there.
> > > > > >> > > > > >> > > > > > > > > Nothing must be done at this point.
> > > > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled with the
> > > > > >> feature
> > > > > >> > flag
> > > > > >> > > > > >> turned
> > > > > >> > > > > >> > > on.
> > > > > >> > > > > >> > > > > The
> > > > > >> > > > > >> > > > > > > > > consumer group is automatically
> > > converted when
> > > > > >> > the first
> > > > > >> > > > > >> > > consumer
> > > > > >> > > > > >> > > > > > > > > using the new protocol joins the group.
> > > While
> > > > > >> the
> > > > > >> > > > > members
> > > > > >> > > > > >> > > using the
> > > > > >> > > > > >> > > > > > > > > old protocol are being upgraded, the old
> > > > > >> protocol
> > > > > >> > is
> > > > > >> > > > > >> proxied
> > > > > >> > > > > >> > > into
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > new one.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the
> > > pattern
> > > > > >> > > > > >> subscriptions
> > > > > >> > > > > >> > > to
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > server,
> > > > > >> > > > > >> > > > > > > > > > but it seems like there could be some
> > > tricky
> > > > > >> > > > > >> compatibility
> > > > > >> > > > > >> > > > > issues.
> > > > > >> > > > > >> > > > > > > Will
> > > > > >> > > > > >> > > > > > > > > we
> > > > > >> > > > > >> > > > > > > > > > have a mechanism for users to detect
> > > that
> > > > > >> they
> > > > > >> > need to
> > > > > >> > > > > >> > update
> > > > > >> > > > > >> > > > > their
> > > > > >> > > > > >> > > > > > > regex
> > > > > >> > > > > >> > > > > > > > > > before switching to the new protocol?
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > I think that I am a bit more optimistic
> > > than
> > > > > >> you
> > > > > >> > on this
> > > > > >> > > > > >> > > point. I
> > > > > >> > > > > >> > > > > > > > > believe that the majority of the cases
> > > are
> > > > > >> simple
> > > > > >> > > > > regexes
> > > > > >> > > > > >> > which
> > > > > >> > > > > >> > > > > should
> > > > > >> > > > > >> > > > > > > > > work with the new engine. The
> > > coordinator will
> > > > > >> > verify
> > > > > >> > > > > the
> > > > > >> > > > > >> > regex
> > > > > >> > > > > >> > > > > anyway
> > > > > >> > > > > >> > > > > > > > > and reject the consumer if the regex is
> > > not
> > > > > >> valid.
> > > > > >> > > > > Coming
> > > > > >> > > > > >> > back
> > > > > >> > > > > >> > > to
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > migration path, in the worst case, the
> > > first
> > > > > >> > upgraded
> > > > > >> > > > > >> > consumer
> > > > > >> > > > > >> > > > > joining
> > > > > >> > > > > >> > > > > > > > > the group will be rejected. This should
> > > be used
> > > > > >> > as the
> > > > > >> > > > > >> last
> > > > > >> > > > > >> > > > > defence, I
> > > > > >> > > > > >> > > > > > > > > would say.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > One way for customers to validate their
> > > regex
> > > > > >> > before
> > > > > >> > > > > >> > upgrading
> > > > > >> > > > > >> > > > > their
> > > > > >> > > > > >> > > > > > > > > prod would be to test them with another
> > > group.
> > > > > >> For
> > > > > >> > > > > >> instance,
> > > > > >> > > > > >> > > that
> > > > > >> > > > > >> > > > > > > > > could be done in a pre-prod environment.
> > > > > >> Another
> > > > > >> > way
> > > > > >> > > > > >> would be
> > > > > >> > > > > >> > > to
> > > > > >> > > > > >> > > > > > > > > extend the consumer-group tool to
> > > provide a
> > > > > >> regex
> > > > > >> > > > > >> validation
> > > > > >> > > > > >> > > > > > > > > mechanism. Would this be enough in your
> > > > > >> opinion?
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 5. Related to the last question, will
> > > the
> > > > > >> Java
> > > > > >> > client
> > > > > >> > > > > >> allow
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > users to
> > > > > >> > > > > >> > > > > > > > > > stick with the current regex engine for
> > > > > >> > compatibility
> > > > > >> > > > > >> > > reasons?
> > > > > >> > > > > >> > > > > For
> > > > > >> > > > > >> > > > > > > > > example,
> > > > > >> > > > > >> > > > > > > > > > it may be handy to keep using client
> > > based
> > > > > >> > regex at
> > > > > >> > > > > >> first
> > > > > >> > > > > >> > to
> > > > > >> > > > > >> > > keep
> > > > > >> > > > > >> > > > > > > > > > migrations simple and then migrate to
> > > server
> > > > > >> > based
> > > > > >> > > > > >> regexes
> > > > > >> > > > > >> > > as a
> > > > > >> > > > > >> > > > > > > second
> > > > > >> > > > > >> > > > > > > > > step.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > I understand your point but I am
> > > concerned that
> > > > > >> > this
> > > > > >> > > > > would
> > > > > >> > > > > >> > > allow
> > > > > >> > > > > >> > > > > users
> > > > > >> > > > > >> > > > > > > > > to actually stay in this mode. That
> > > would go
> > > > > >> > against our
> > > > > >> > > > > >> goal
> > > > > >> > > > > >> > > of
> > > > > >> > > > > >> > > > > > > > > simplifying the client because we would
> > > have to
> > > > > >> > continue
> > > > > >> > > > > >> > > monitoring
> > > > > >> > > > > >> > > > > > > > > the metadata on the client side. I would
> > > rather
> > > > > >> > not do
> > > > > >> > > > > >> this.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> > > coordinator
> > > > > >> will
> > > > > >> > be
> > > > > >> > > > > >> > > responsible for
> > > > > >> > > > > >> > > > > > > storing
> > > > > >> > > > > >> > > > > > > > > > the configurations and that the
> > > > > >> configurations
> > > > > >> > will be
> > > > > >> > > > > >> > > deleted
> > > > > >> > > > > >> > > > > when
> > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > >> > > > > > > > > > group is deleted. Will a transition to
> > > DEAD
> > > > > >> > trigger
> > > > > >> > > > > >> > deletion
> > > > > >> > > > > >> > > of
> > > > > >> > > > > >> > > > > > > > > > configurations?
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > That's right. The configurations will be
> > > > > >> deleted
> > > > > >> > when
> > > > > >> > > > > the
> > > > > >> > > > > >> > > group is
> > > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 7. Will the choice to store the
> > > configs in
> > > > > >> the
> > > > > >> > group
> > > > > >> > > > > >> > > coordinator
> > > > > >> > > > > >> > > > > > > make it
> > > > > >> > > > > >> > > > > > > > > > harder to list all cluster configs and
> > > their
> > > > > >> > values?
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > > configurations are
> > > > > >> > overrides
> > > > > >> > > > > >> of
> > > > > >> > > > > >> > > cluster
> > > > > >> > > > > >> > > > > > > > > configs. If you want to know all the
> > > overrides
> > > > > >> > though,
> > > > > >> > > > > you
> > > > > >> > > > > >> > > would
> > > > > >> > > > > >> > > > > have
> > > > > >> > > > > >> > > > > > > > > to ask all the group coordinators. You
> > > cannot
> > > > > >> > rely on
> > > > > >> > > > > the
> > > > > >> > > > > >> > > metadata
> > > > > >> > > > > >> > > > > log
> > > > > >> > > > > >> > > > > > > > > for instance.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 8. How would someone configure a group
> > > before
> > > > > >> > starting
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > consumers?
> > > > > >> > > > > >> > > > > > > > > Have
> > > > > >> > > > > >> > > > > > > > > > we considered allowing the explicit
> > > creation
> > > > > >> of
> > > > > >> > > > > groups?
> > > > > >> > > > > >> > > > > > > Alternatively,
> > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > >> > > > > > > > > > configs could be decoupled from the
> > > group
> > > > > >> > lifecycle.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Yes. The group will be automatically
> > > created in
> > > > > >> > this
> > > > > >> > > > > case.
> > > > > >> > > > > >> > > However,
> > > > > >> > > > > >> > > > > > > > > the configs will be lost after the
> > > retention
> > > > > >> > period of
> > > > > >> > > > > the
> > > > > >> > > > > >> > > group
> > > > > >> > > > > >> > > > > > > > > passes.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method
> > > for the
> > > > > >> > Java
> > > > > >> > > > > >> client
> > > > > >> > > > > >> > > still
> > > > > >> > > > > >> > > > > take
> > > > > >> > > > > >> > > > > > > a
> > > > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we
> > > have to
> > > > > >> > introduce
> > > > > >> > > > > an
> > > > > >> > > > > >> > > overload?
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > That's a very group question. I forgot
> > > about
> > > > > >> that
> > > > > >> > one.
> > > > > >> > > > > As
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
> > > > > >> compatible
> > > > > >> > with
> > > > > >> > > > > the
> > > > > >> > > > > >> > > engine
> > > > > >> > > > > >> > > > > that
> > > > > >> > > > > >> > > > > > > > > we plan to use, it might be better to
> > > deprecate
> > > > > >> > it and
> > > > > >> > > > > >> use an
> > > > > >> > > > > >> > > > > overload
> > > > > >> > > > > >> > > > > > > > > which takes a string. We would rely on
> > > the
> > > > > >> server
> > > > > >> > side
> > > > > >> > > > > >> > > validation.
> > > > > >> > > > > >> > > > > > > > > During the migration, I think that we
> > > could
> > > > > >> still
> > > > > >> > try to
> > > > > >> > > > > >> > > toString
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > regex and use it. That should work, I
> > > think, in
> > > > > >> > the
> > > > > >> > > > > >> majority
> > > > > >> > > > > >> > > of the
> > > > > >> > > > > >> > > > > > > > > cases.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine that we
> > > should be
> > > > > >> > clearer
> > > > > >> > > > > about
> > > > > >> > > > > >> > the
> > > > > >> > > > > >> > > > > reason
> > > > > >> > > > > >> > > > > > > to
> > > > > >> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
> > > > > >> feature
> > > > > >> > flag.
> > > > > >> > > > > >> Maybe
> > > > > >> > > > > >> > > we
> > > > > >> > > > > >> > > > > mean
> > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > >> > > > > > > > > we
> > > > > >> > > > > >> > > > > > > > > > can switch the default for the feature
> > > flag
> > > > > >> to
> > > > > >> > true
> > > > > >> > > > > >> based
> > > > > >> > > > > >> > on
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > > > > metadata.version once we want to make
> > > it the
> > > > > >> > default.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > My plan was to use that feature flag
> > > mainly
> > > > > >> > during the
> > > > > >> > > > > >> > > development
> > > > > >> > > > > >> > > > > > > > > phase. I should not have mentioned it, I
> > > think,
> > > > > >> > because
> > > > > >> > > > > we
> > > > > >> > > > > >> > > could
> > > > > >> > > > > >> > > > > use
> > > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't
> > > mention
> > > > > >> the
> > > > > >> > > > > required
> > > > > >> > > > > >> > > ACLs, it
> > > > > >> > > > > >> > > > > > > would
> > > > > >> > > > > >> > > > > > > > > be
> > > > > >> > > > > >> > > > > > > > > > good to add that for consistency.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Noted.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > > >> ConsumerGroupHeartbeat
> > > > > >> > > > > requires
> > > > > >> > > > > >> > > "Read
> > > > > >> > > > > >> > > > > Group"
> > > > > >> > > > > >> > > > > > > > > even
> > > > > >> > > > > >> > > > > > > > > > though it seems to do more than
> > > reading.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > I agree. This is how the current
> > > protocol works
> > > > > >> > though.
> > > > > >> > > > > We
> > > > > >> > > > > >> > only
> > > > > >> > > > > >> > > > > > > > > require "Read Group" to join a group. We
> > > could
> > > > > >> > consider
> > > > > >> > > > > >> > > changing
> > > > > >> > > > > >> > > > > this
> > > > > >> > > > > >> > > > > > > > > but I am not sure that it is worth it.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 13. How is topic recreation handled by
> > > the
> > > > > >> > consumer
> > > > > >> > > > > with
> > > > > >> > > > > >> > the
> > > > > >> > > > > >> > > new
> > > > > >> > > > > >> > > > > > > group
> > > > > >> > > > > >> > > > > > > > > > protocol? It would be good to have a
> > > section
> > > > > >> on
> > > > > >> > this.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Noted. From a protocol perspective, the
> > > new
> > > > > >> topic
> > > > > >> > will
> > > > > >> > > > > >> have a
> > > > > >> > > > > >> > > new
> > > > > >> > > > > >> > > > > > > > > topic id so it will treat it like a
> > > topic with
> > > > > >> a
> > > > > >> > > > > different
> > > > > >> > > > > >> > > name.
> > > > > >> > > > > >> > > > > The
> > > > > >> > > > > >> > > > > > > > > only issue is that the fetch/commit
> > > offsets
> > > > > >> APIs
> > > > > >> > do not
> > > > > >> > > > > >> > support
> > > > > >> > > > > >> > > > > topic
> > > > > >> > > > > >> > > > > > > > > IDs so the consumer would reuse the
> > > offsets
> > > > > >> based
> > > > > >> > on the
> > > > > >> > > > > >> > same.
> > > > > >> > > > > >> > > I
> > > > > >> > > > > >> > > > > think
> > > > > >> > > > > >> > > > > > > > > that we should update those APIs as well
> > > in
> > > > > >> order
> > > > > >> > to be
> > > > > >> > > > > >> > > consistent
> > > > > >> > > > > >> > > > > end
> > > > > >> > > > > >> > > > > > > > > to end. That would strengthen the
> > > semantics of
> > > > > >> the
> > > > > >> > > > > >> consumer.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the
> > > new
> > > > > >> > coordinator
> > > > > >> > > > > >> in
> > > > > >> > > > > >> > > Java.
> > > > > >> > > > > >> > > > > Even
> > > > > >> > > > > >> > > > > > > > > though
> > > > > >> > > > > >> > > > > > > > > > this is an implementation detail, do
> > > we plan
> > > > > >> to
> > > > > >> > have a
> > > > > >> > > > > >> new
> > > > > >> > > > > >> > > gradle
> > > > > >> > > > > >> > > > > > > module
> > > > > >> > > > > >> > > > > > > > > > for it?
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Yes.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when
> > > it
> > > > > >> comes
> > > > > >> > to how
> > > > > >> > > > > >> many
> > > > > >> > > > > >> > > > > members
> > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > >> > > > > > > > > new
> > > > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > We don't have numbers at the moment. The
> > > > > >> protocol
> > > > > >> > should
> > > > > >> > > > > >> > > support
> > > > > >> > > > > >> > > > > 1000s
> > > > > >> > > > > >> > > > > > > > > of members per group. We will measure
> > > this when
> > > > > >> > we have
> > > > > >> > > > > a
> > > > > >> > > > > >> > first
> > > > > >> > > > > >> > > > > > > > > implementation. Note that we might have
> > > other
> > > > > >> > > > > bottlenecks
> > > > > >> > > > > >> > down
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> > > SubscribedTopidIds
> > > > > >> > instead
> > > > > >> > > > > of
> > > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > > >> > ConsumerGroupHeartbeatRequest?
> > > > > >> > > > > >> Is
> > > > > >> > > > > >> > the
> > > > > >> > > > > >> > > > > idea
> > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > >> > > > > > > > > > since we have to resolve the regex on
> > > the
> > > > > >> > server, we
> > > > > >> > > > > >> can do
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > same
> > > > > >> > > > > >> > > > > > > for
> > > > > >> > > > > >> > > > > > > > > > the topic name? The difference is that
> > > > > >> sending
> > > > > >> > the
> > > > > >> > > > > >> regex is
> > > > > >> > > > > >> > > more
> > > > > >> > > > > >> > > > > > > > > efficient
> > > > > >> > > > > >> > > > > > > > > > whereas sending the topic names is less
> > > > > >> > efficient.
> > > > > >> > > > > >> > > Furthermore,
> > > > > >> > > > > >> > > > > > > delete
> > > > > >> > > > > >> > > > > > > > > and
> > > > > >> > > > > >> > > > > > > > > > recreation is easier to handle if we
> > > have
> > > > > >> topic
> > > > > >> > ids.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > The idea was to consolidate the metadata
> > > lookup
> > > > > >> > on the
> > > > > >> > > > > >> server
> > > > > >> > > > > >> > > for
> > > > > >> > > > > >> > > > > both
> > > > > >> > > > > >> > > > > > > > > paths but I do agree with your point. As
> > > a
> > > > > >> second
> > > > > >> > > > > though,
> > > > > >> > > > > >> > using
> > > > > >> > > > > >> > > > > topic
> > > > > >> > > > > >> > > > > > > > > ids may be better here for the delete and
> > > > > >> > recreation
> > > > > >> > > > > case.
> > > > > >> > > > > >> > > Also, I
> > > > > >> > > > > >> > > > > > > > > suppose that we may allow users to
> > > subscribe
> > > > > >> with
> > > > > >> > topic
> > > > > >> > > > > >> ids
> > > > > >> > > > > >> > in
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > > > future because that is the only way to be
> > > > > >> really
> > > > > >> > robust
> > > > > >> > > > > to
> > > > > >> > > > > >> > > topic
> > > > > >> > > > > >> > > > > > > > > re-creation.
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > Best,
> > > > > >> > > > > >> > > > > > > > > David
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David
> > > Jacot <
> > > > > >> > > > > >> > > djacot@confluent.io>
> > > > > >> > > > > >> > > > > > > wrote:
> > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > Thanks for your comments. Please find
> > > my
> > > > > >> answers
> > > > > >> > > > > below.
> > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on
> > > topic IDs
> > > > > >> > with the
> > > > > >> > > > > >> > > exception
> > > > > >> > > > > >> > > > > of the
> > > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > > >> > > > > >> I
> > > > > >> > > > > >> > am
> > > > > >> > > > > >> > > not
> > > > > >> > > > > >> > > > > sure
> > > > > >> > > > > >> > > > > > > > > > if using topic names is the right call
> > > here.
> > > > > >> I
> > > > > >> > need to
> > > > > >> > > > > >> > think
> > > > > >> > > > > >> > > > > about it
> > > > > >> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does
> > > not
> > > > > >> > change the
> > > > > >> > > > > >> > > > > fetch/commit
> > > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This
> > > may be
> > > > > >> > something
> > > > > >> > > > > >> that
> > > > > >> > > > > >> > we
> > > > > >> > > > > >> > > > > should
> > > > > >> > > > > >> > > > > > > > > > include though as it would give better
> > > > > >> overall
> > > > > >> > > > > >> guarantee in
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > > > > producer.
> > > > > >> > > > > >> > > > > > > > > > - You're right. I think that I should
> > > not
> > > > > >> have
> > > > > >> > > > > mentioned
> > > > > >> > > > > >> > this
> > > > > >> > > > > >> > > > > flag at
> > > > > >> > > > > >> > > > > > > > > > all. I will remove it. We can use an
> > > internal
> > > > > >> > > > > >> configuration
> > > > > >> > > > > >> > > while
> > > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> > > supported. The
> > > > > >> > change is
> > > > > >> > > > > >> > > > > orthogonal. The
> > > > > >> > > > > >> > > > > > > > > > only requirement is that the cluster
> > > uses
> > > > > >> topic
> > > > > >> > IDs.
> > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > Best,
> > > > > >> > > > > >> > > > > > > > > > David
> > > > > >> > > > > >> > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM
> > > Guozhang
> > > > > >> Wang <
> > > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > > >> > > > > >> > > > > > > > > wrote:
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are
> > > some
> > > > > >> replies
> > > > > >> > > > > inlined
> > > > > >> > > > > >> > > below:
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM
> > > Ismael Juma
> > > > > >> <
> > > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > > >> > > > > >> > > > > > > wrote:
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the
> > > > > >> potential
> > > > > >> > to be a
> > > > > >> > > > > >> > great
> > > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > > >> > > > > >> > > > > > > > > > > > initial questions/comments:
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk
> > > about
> > > > > >> > target
> > > > > >> > > > > >> versions
> > > > > >> > > > > >> > > for
> > > > > >> > > > > >> > > > > > > > > deprecation and
> > > > > >> > > > > >> > > > > > > > > > > > removal of the existing group
> > > protocol.
> > > > > >> > Unlike
> > > > > >> > > > > >> KRaft,
> > > > > >> > > > > >> > > this
> > > > > >> > > > > >> > > > > > > affects a
> > > > > >> > > > > >> > > > > > > > > core
> > > > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > > > >> > deprecation/removal will
> > > > > >> > > > > >> be
> > > > > >> > > > > >> > > heavily
> > > > > >> > > > > >> > > > > > > > > dependent on
> > > > > >> > > > > >> > > > > > > > > > > > how quickly applications migrate
> > > to the
> > > > > >> new
> > > > > >> > > > > >> protocol.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can
> > > > > >> remove
> > > > > >> > the
> > > > > >> > > > > >> proposed
> > > > > >> > > > > >> > > > > timeline
> > > > > >> > > > > >> > > > > > > in
> > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and
> > > Migration
> > > > > >> > Plan` and
> > > > > >> > > > > >> > > instead
> > > > > >> > > > > >> > > > > just
> > > > > >> > > > > >> > > > > > > state
> > > > > >> > > > > >> > > > > > > > > > > that we will decide in the future
> > > about
> > > > > >> when
> > > > > >> > we
> > > > > >> > > > > would
> > > > > >> > > > > >> > > > > deprecate old
> > > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to
> > > release this
> > > > > >> > in 4.x,
> > > > > >> > > > > >> but
> > > > > >> > > > > >> > it
> > > > > >> > > > > >> > > > > wasn't
> > > > > >> > > > > >> > > > > > > made
> > > > > >> > > > > >> > > > > > > > > clear
> > > > > >> > > > > >> > > > > > > > > > > > why. If we added that as a way to
> > > > > >> estimate
> > > > > >> > when
> > > > > >> > > > > we'd
> > > > > >> > > > > >> > > > > deprecate
> > > > > >> > > > > >> > > > > > > and
> > > > > >> > > > > >> > > > > > > > > remove
> > > > > >> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
> > > > > >> removing
> > > > > >> > this
> > > > > >> > > > > >> part.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > I think that's not specifically
> > > related to
> > > > > >> the
> > > > > >> > > > > >> > > > > deprecation/removal
> > > > > >> > > > > >> > > > > > > > > timeline
> > > > > >> > > > > >> > > > > > > > > > > plan, but it's more for client
> > > upgrades.
> > > > > >> I.e.
> > > > > >> > the
> > > > > >> > > > > >> > > broker-side
> > > > > >> > > > > >> > > > > > > > > > > implementation may be done first,
> > > and then
> > > > > >> the
> > > > > >> > > > > client
> > > > > >> > > > > >> > side,
> > > > > >> > > > > >> > > > > and we
> > > > > >> > > > > >> > > > > > > > > would
> > > > > >> > > > > >> > > > > > > > > > > only mark it as "released" by the
> > > time
> > > > > >> clients
> > > > > >> > > > > >> > > implementations
> > > > > >> > > > > >> > > > > are
> > > > > >> > > > > >> > > > > > > > > done. At
> > > > > >> > > > > >> > > > > > > > > > > that time, to enable the feature the
> > > > > >> clients
> > > > > >> > need to
> > > > > >> > > > > >> > first
> > > > > >> > > > > >> > > > > swap-in
> > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and
> > > then set
> > > > > >> > the flag
> > > > > >> > > > > >> > with a
> > > > > >> > > > > >> > > > > second
> > > > > >> > > > > >> > > > > > > > > rolling
> > > > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's
> > > better to be
> > > > > >> > released
> > > > > >> > > > > >> in a
> > > > > >> > > > > >> > > major
> > > > > >> > > > > >> > > > > > > > > version.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the
> > > details of
> > > > > >> the
> > > > > >> > > > > migration
> > > > > >> > > > > >> > > story.
> > > > > >> > > > > >> > > > > It
> > > > > >> > > > > >> > > > > > > > > sounds like
> > > > > >> > > > > >> > > > > > > > > > > > we're saying we will support online
> > > > > >> > migrations. Is
> > > > > >> > > > > >> that
> > > > > >> > > > > >> > > > > correct?
> > > > > >> > > > > >> > > > > > > We
> > > > > >> > > > > >> > > > > > > > > should
> > > > > >> > > > > >> > > > > > > > > > > > explain this in detail. It could
> > > also be
> > > > > >> > done as a
> > > > > >> > > > > >> > > separate
> > > > > >> > > > > >> > > > > KIP,
> > > > > >> > > > > >> > > > > > > if
> > > > > >> > > > > >> > > > > > > > > it's
> > > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Yes I think that's the part we can
> > > be more
> > > > > >> > concrete
> > > > > >> > > > > >> about
> > > > > >> > > > > >> > > for
> > > > > >> > > > > >> > > > > sure
> > > > > >> > > > > >> > > > > > > (and
> > > > > >> > > > > >> > > > > > > > > > > this is related to your question 2)
> > > above).
> > > > > >> > We will
> > > > > >> > > > > >> work
> > > > > >> > > > > >> > on
> > > > > >> > > > > >> > > > > making
> > > > > >> > > > > >> > > > > > > it
> > > > > >> > > > > >> > > > > > > > > more
> > > > > >> > > > > >> > > > > > > > > > > explicit in parallel as we solicit
> > > more
> > > > > >> > feedback.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing
> > > the
> > > > > >> > pattern
> > > > > >> > > > > >> > > subscriptions
> > > > > >> > > > > >> > > > > to
> > > > > >> > > > > >> > > > > > > the
> > > > > >> > > > > >> > > > > > > > > server,
> > > > > >> > > > > >> > > > > > > > > > > > but it seems like there could be
> > > some
> > > > > >> tricky
> > > > > >> > > > > >> > > compatibility
> > > > > >> > > > > >> > > > > > > issues.
> > > > > >> > > > > >> > > > > > > > > Will we
> > > > > >> > > > > >> > > > > > > > > > > > have a mechanism for users to
> > > detect that
> > > > > >> > they
> > > > > >> > > > > need
> > > > > >> > > > > >> to
> > > > > >> > > > > >> > > update
> > > > > >> > > > > >> > > > > > > their
> > > > > >> > > > > >> > > > > > > > > regex
> > > > > >> > > > > >> > > > > > > > > > > > before switching to the new
> > > protocol?
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
> > > > > >> non-java
> > > > > >> > client
> > > > > >> > > > > >> > users
> > > > > >> > > > > >> > > to
> > > > > >> > > > > >> > > > > sort
> > > > > >> > > > > >> > > > > > > of
> > > > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> > > switching to
> > > > > >> the
> > > > > >> > new
> > > > > >> > > > > >> > protocol.
> > > > > >> > > > > >> > > I
> > > > > >> > > > > >> > > > > do not
> > > > > >> > > > > >> > > > > > > > > have a
> > > > > >> > > > > >> > > > > > > > > > > specific idea on top of my head
> > > though,
> > > > > >> maybe
> > > > > >> > others
> > > > > >> > > > > >> like
> > > > > >> > > > > >> > > @Matt
> > > > > >> > > > > >> > > > > > > > > Howlett can
> > > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 5. Related to the last question,
> > > will the
> > > > > >> > Java
> > > > > >> > > > > >> client
> > > > > >> > > > > >> > > allow
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > users to
> > > > > >> > > > > >> > > > > > > > > > > > stick with the current regex
> > > engine for
> > > > > >> > > > > >> compatibility
> > > > > >> > > > > >> > > > > reasons?
> > > > > >> > > > > >> > > > > > > For
> > > > > >> > > > > >> > > > > > > > > example,
> > > > > >> > > > > >> > > > > > > > > > > > it may be handy to keep using
> > > client
> > > > > >> based
> > > > > >> > regex
> > > > > >> > > > > at
> > > > > >> > > > > >> > > first to
> > > > > >> > > > > >> > > > > keep
> > > > > >> > > > > >> > > > > > > > > > > > migrations simple and then migrate
> > > to
> > > > > >> > server based
> > > > > >> > > > > >> > > regexes
> > > > > >> > > > > >> > > > > as a
> > > > > >> > > > > >> > > > > > > > > second
> > > > > >> > > > > >> > > > > > > > > > > > step.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Honestly I have not thought about
> > > that for
> > > > > >> > java
> > > > > >> > > > > >> clients,
> > > > > >> > > > > >> > > and
> > > > > >> > > > > >> > > > > we can
> > > > > >> > > > > >> > > > > > > > > discuss
> > > > > >> > > > > >> > > > > > > > > > > that. What kind of compatibility
> > > issues do
> > > > > >> > you have
> > > > > >> > > > > in
> > > > > >> > > > > >> > > mind?
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 6. When we say that the group
> > > coordinator
> > > > > >> > will be
> > > > > >> > > > > >> > > > > responsible for
> > > > > >> > > > > >> > > > > > > > > storing
> > > > > >> > > > > >> > > > > > > > > > > > the configurations and that the
> > > > > >> > configurations
> > > > > >> > > > > will
> > > > > >> > > > > >> be
> > > > > >> > > > > >> > > > > deleted
> > > > > >> > > > > >> > > > > > > when
> > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> > > transition to
> > > > > >> DEAD
> > > > > >> > > > > trigger
> > > > > >> > > > > >> > > deletion
> > > > > >> > > > > >> > > > > of
> > > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an
> > > ending
> > > > > >> state
> > > > > >> > (we
> > > > > >> > > > > would
> > > > > >> > > > > >> > only
> > > > > >> > > > > >> > > > > > > transit to
> > > > > >> > > > > >> > > > > > > > > that
> > > > > >> > > > > >> > > > > > > > > > > state when the group is EMPTY and
> > > also all
> > > > > >> of
> > > > > >> > its
> > > > > >> > > > > >> > metadata
> > > > > >> > > > > >> > > are
> > > > > >> > > > > >> > > > > > > gone),
> > > > > >> > > > > >> > > > > > > > > once
> > > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this group
> > > would
> > > > > >> never
> > > > > >> > be
> > > > > >> > > > > >> revived.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to store the
> > > configs
> > > > > >> in
> > > > > >> > the
> > > > > >> > > > > group
> > > > > >> > > > > >> > > > > coordinator
> > > > > >> > > > > >> > > > > > > > > make it
> > > > > >> > > > > >> > > > > > > > > > > > harder to list all cluster configs
> > > and
> > > > > >> their
> > > > > >> > > > > values?
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > That's a good question, and our
> > > thoughts
> > > > > >> are
> > > > > >> > that
> > > > > >> > > > > the
> > > > > >> > > > > >> > > so-called
> > > > > >> > > > > >> > > > > > > "group
> > > > > >> > > > > >> > > > > > > > > > > configurations" are overrides of the
> > > > > >> > cluster-level
> > > > > >> > > > > >> > > > > configurations
> > > > > >> > > > > >> > > > > > > > > > > customized per group so when an
> > > admin list
> > > > > >> > cluster
> > > > > >> > > > > >> > configs
> > > > > >> > > > > >> > > it's
> > > > > >> > > > > >> > > > > > > okay to
> > > > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> > > "defaults", not
> > > > > >> > showing
> > > > > >> > > > > >> any
> > > > > >> > > > > >> > > > > per-group
> > > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 8. How would someone configure a
> > > group
> > > > > >> > before
> > > > > >> > > > > >> starting
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > > >> > > > > >> > > > > > > > > > > > we considered allowing the explicit
> > > > > >> > creation of
> > > > > >> > > > > >> groups?
> > > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > > >> > > > > >> > > > > > > > > > > > configs could be decoupled from
> > > the group
> > > > > >> > > > > lifecycle.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > The configs can be created before
> > > the group
> > > > > >> > itself
> > > > > >> > > > > as
> > > > > >> > > > > >> an
> > > > > >> > > > > >> > > > > > > independent
> > > > > >> > > > > >> > > > > > > > > entity
> > > > > >> > > > > >> > > > > > > > > > > --- of course, this requires the
> > > > > >> corresponding
> > > > > >> > > > > >> request to
> > > > > >> > > > > >> > > be
> > > > > >> > > > > >> > > > > > > routed to
> > > > > >> > > > > >> > > > > > > > > the
> > > > > >> > > > > >> > > > > > > > > > > right coordinator based on the group
> > > id ---
> > > > > >> > the only
> > > > > >> > > > > >> > thing
> > > > > >> > > > > >> > > that
> > > > > >> > > > > >> > > > > > > > > differs is,
> > > > > >> > > > > >> > > > > > > > > > > when the group itself is gone we
> > > also check
> > > > > >> > if there
> > > > > >> > > > > >> are
> > > > > >> > > > > >> > > any
> > > > > >> > > > > >> > > > > > > > > configuration
> > > > > >> > > > > >> > > > > > > > > > > entities related to that group and
> > > delete
> > > > > >> as
> > > > > >> > well.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> > > > > >> > asymmetry on
> > > > > >> > > > > the
> > > > > >> > > > > >> > > creation
> > > > > >> > > > > >> > > > > /
> > > > > >> > > > > >> > > > > > > > > deletion
> > > > > >> > > > > >> > > > > > > > > > > lifecycles of the config entities,
> > > and we
> > > > > >> > would like
> > > > > >> > > > > >> to
> > > > > >> > > > > >> > > hear
> > > > > >> > > > > >> > > > > > > everyone's
> > > > > >> > > > > >> > > > > > > > > > > feelings whether we should aim for
> > > symmetry
> > > > > >> > i.e.
> > > > > >> > > > > >> totally
> > > > > >> > > > > >> > > > > decouple
> > > > > >> > > > > >> > > > > > > group
> > > > > >> > > > > >> > > > > > > > > > > configs and hence not delete them at
> > > all
> > > > > >> when
> > > > > >> > the
> > > > > >> > > > > >> group
> > > > > >> > > > > >> > is
> > > > > >> > > > > >> > > > > gone,
> > > > > >> > > > > >> > > > > > > but
> > > > > >> > > > > >> > > > > > > > > always
> > > > > >> > > > > >> > > > > > > > > > > require explicit deletion operations
> > > by
> > > > > >> > themselves.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe
> > > method for
> > > > > >> > the Java
> > > > > >> > > > > >> > client
> > > > > >> > > > > >> > > > > still
> > > > > >> > > > > >> > > > > > > take
> > > > > >> > > > > >> > > > > > > > > a
> > > > > >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we
> > > have
> > > > > >> to
> > > > > >> > > > > >> introduce an
> > > > > >> > > > > >> > > > > overload?
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > I think we do not need to introduce
> > > an
> > > > > >> > overload, but
> > > > > >> > > > > >> I'm
> > > > > >> > > > > >> > > all
> > > > > >> > > > > >> > > > > ears
> > > > > >> > > > > >> > > > > > > if
> > > > > >> > > > > >> > > > > > > > > there
> > > > > >> > > > > >> > > > > > > > > > > may be some compatibility issues
> > > that we
> > > > > >> may
> > > > > >> > > > > overlook.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we
> > > should
> > > > > >> be
> > > > > >> > clearer
> > > > > >> > > > > >> > about
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > > > reason
> > > > > >> > > > > >> > > > > > > > > to
> > > > > >> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version
> > > from the
> > > > > >> > feature
> > > > > >> > > > > >> flag.
> > > > > >> > > > > >> > > Maybe
> > > > > >> > > > > >> > > > > we
> > > > > >> > > > > >> > > > > > > mean
> > > > > >> > > > > >> > > > > > > > > that we
> > > > > >> > > > > >> > > > > > > > > > > > can switch the default for the
> > > feature
> > > > > >> flag
> > > > > >> > to
> > > > > >> > > > > true
> > > > > >> > > > > >> > > based on
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > > > > metadata.version once we want to
> > > make it
> > > > > >> the
> > > > > >> > > > > >> default.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't
> > > mention
> > > > > >> > the
> > > > > >> > > > > >> required
> > > > > >> > > > > >> > > ACLs,
> > > > > >> > > > > >> > > > > it
> > > > > >> > > > > >> > > > > > > > > would be
> > > > > >> > > > > >> > > > > > > > > > > > good to add that for consistency.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > > > >> > ConsumerGroupHeartbeat
> > > > > >> > > > > >> > requires
> > > > > >> > > > > >> > > > > "Read
> > > > > >> > > > > >> > > > > > > > > Group" even
> > > > > >> > > > > >> > > > > > > > > > > > though it seems to do more than
> > > reading.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > I had that thought myself as well,
> > > but in
> > > > > >> the
> > > > > >> > end we
> > > > > >> > > > > >> > could
> > > > > >> > > > > >> > > not
> > > > > >> > > > > >> > > > > > > find a
> > > > > >> > > > > >> > > > > > > > > > > better alternative: adding Write
> > > Group
> > > > > >> seems
> > > > > >> > an
> > > > > >> > > > > >> overkill
> > > > > >> > > > > >> > > here
> > > > > >> > > > > >> > > > > > > since we
> > > > > >> > > > > >> > > > > > > > > do
> > > > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we only have
> > > Read /
> > > > > >> > Delete
> > > > > >> > > > > and
> > > > > >> > > > > >> > > Describe
> > > > > >> > > > > >> > > > > on
> > > > > >> > > > > >> > > > > > > > > groups so
> > > > > >> > > > > >> > > > > > > > > > > far). Would like to hear others
> > > thoughts.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 13. How is topic recreation
> > > handled by
> > > > > >> the
> > > > > >> > > > > consumer
> > > > > >> > > > > >> > with
> > > > > >> > > > > >> > > the
> > > > > >> > > > > >> > > > > new
> > > > > >> > > > > >> > > > > > > > > group
> > > > > >> > > > > >> > > > > > > > > > > > protocol? It would be good to have
> > > a
> > > > > >> > section on
> > > > > >> > > > > >> this.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > You mean with regex subscription
> > > right? Yes
> > > > > >> > we can
> > > > > >> > > > > >> add a
> > > > > >> > > > > >> > > > > section
> > > > > >> > > > > >> > > > > > > about
> > > > > >> > > > > >> > > > > > > > > > > that, but basically the idea is that
> > > > > >> consumer
> > > > > >> > would
> > > > > >> > > > > be
> > > > > >> > > > > >> > > totally
> > > > > >> > > > > >> > > > > > > > > agnostic in
> > > > > >> > > > > >> > > > > > > > > > > the new protocol as it's handled all
> > > by the
> > > > > >> > brokers.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write
> > > the
> > > > > >> new
> > > > > >> > > > > >> coordinator
> > > > > >> > > > > >> > in
> > > > > >> > > > > >> > > > > Java.
> > > > > >> > > > > >> > > > > > > Even
> > > > > >> > > > > >> > > > > > > > > though
> > > > > >> > > > > >> > > > > > > > > > > > this is an implementation detail,
> > > do we
> > > > > >> > plan to
> > > > > >> > > > > >> have a
> > > > > >> > > > > >> > > new
> > > > > >> > > > > >> > > > > gradle
> > > > > >> > > > > >> > > > > > > > > module
> > > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > We have not thought about that. But
> > > I think
> > > > > >> > the
> > > > > >> > > > > answer
> > > > > >> > > > > >> > > should
> > > > > >> > > > > >> > > > > be
> > > > > >> > > > > >> > > > > > > yes.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal
> > > when it
> > > > > >> > comes to
> > > > > >> > > > > >> how
> > > > > >> > > > > >> > > many
> > > > > >> > > > > >> > > > > > > members
> > > > > >> > > > > >> > > > > > > > > the new
> > > > > >> > > > > >> > > > > > > > > > > > group protocol can support?
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > Within a group, I think we should
> > > shoot for
> > > > > >> > 1000s of
> > > > > >> > > > > >> > > members.
> > > > > >> > > > > >> > > > > But
> > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > >> > > > > > > > > > > scalability goals also depend on the
> > > offset
> > > > > >> > > > > management
> > > > > >> > > > > >> > > (commit,
> > > > > >> > > > > >> > > > > > > fetch)
> > > > > >> > > > > >> > > > > > > > > > > capabilities of the coordinator
> > > which we
> > > > > >> did
> > > > > >> > not
> > > > > >> > > > > >> cover in
> > > > > >> > > > > >> > > this
> > > > > >> > > > > >> > > > > > > KIP, so
> > > > > >> > > > > >> > > > > > > > > it's
> > > > > >> > > > > >> > > > > > > > > > > hard to give a number that applies
> > > > > >> > universally.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> > > > > >> > SubscribedTopidIds
> > > > > >> > > > > >> instead
> > > > > >> > > > > >> > of
> > > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > >> > > > > >> > > Is the
> > > > > >> > > > > >> > > > > > > idea
> > > > > >> > > > > >> > > > > > > > > that
> > > > > >> > > > > >> > > > > > > > > > > > since we have to resolve the regex
> > > on the
> > > > > >> > server,
> > > > > >> > > > > we
> > > > > >> > > > > >> > can
> > > > > >> > > > > >> > > do
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > same
> > > > > >> > > > > >> > > > > > > > > for
> > > > > >> > > > > >> > > > > > > > > > > > the topic name? The difference is
> > > that
> > > > > >> > sending the
> > > > > >> > > > > >> > regex
> > > > > >> > > > > >> > > is
> > > > > >> > > > > >> > > > > more
> > > > > >> > > > > >> > > > > > > > > efficient
> > > > > >> > > > > >> > > > > > > > > > > > whereas sending the topic names is
> > > less
> > > > > >> > efficient.
> > > > > >> > > > > >> > > > > Furthermore,
> > > > > >> > > > > >> > > > > > > > > delete and
> > > > > >> > > > > >> > > > > > > > > > > > recreation is easier to handle if
> > > we have
> > > > > >> > topic
> > > > > >> > > > > ids.
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > The main reason to still let the
> > > clients
> > > > > >> send
> > > > > >> > names
> > > > > >> > > > > >> is to
> > > > > >> > > > > >> > > keep
> > > > > >> > > > > >> > > > > the
> > > > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids on the
> > > broker /
> > > > > >> > admin
> > > > > >> > > > > client
> > > > > >> > > > > >> > > only.
> > > > > >> > > > > >> > > > > Note
> > > > > >> > > > > >> > > > > > > that
> > > > > >> > > > > >> > > > > > > > > > > although we added topic id in
> > > KIP-516, we
> > > > > >> > never
> > > > > >> > > > > >> > > implemented the
> > > > > >> > > > > >> > > > > > > logic
> > > > > >> > > > > >> > > > > > > > > on
> > > > > >> > > > > >> > > > > > > > > > > consumer/producers leveraging the
> > > related
> > > > > >> > newer
> > > > > >> > > > > >> versioned
> > > > > >> > > > > >> > > RPCs,
> > > > > >> > > > > >> > > > > > > > > instead we
> > > > > >> > > > > >> > > > > > > > > > > just set the topic id as empty UUID.
> > > We
> > > > > >> want
> > > > > >> > to keep
> > > > > >> > > > > >> the
> > > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > > >> > > > > >> > > > > > > > > > > to be thin and only delegate the
> > > reasoning
> > > > > >> on
> > > > > >> > broker
> > > > > >> > > > > >> and
> > > > > >> > > > > >> > > > > > > potentially
> > > > > >> > > > > >> > > > > > > > > admin
> > > > > >> > > > > >> > > > > > > > > > > clients.
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM
> > > David
> > > > > >> Jacot
> > > > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> > > discussion
> > > > > >> thread
> > > > > >> > on
> > > > > >> > > > > >> KIP-848:
> > > > > >> > > > > >> > > The
> > > > > >> > > > > >> > > > > Next
> > > > > >> > > > > >> > > > > > > > > > > > > Generation of the Consumer
> > > Rebalance
> > > > > >> > Protocol.
> > > > > >> > > > > >> With
> > > > > >> > > > > >> > > this
> > > > > >> > > > > >> > > > > KIP,
> > > > > >> > > > > >> > > > > > > we
> > > > > >> > > > > >> > > > > > > > > aim
> > > > > >> > > > > >> > > > > > > > > > > > > to make the rebalance protocol
> > > (for
> > > > > >> > consumers)
> > > > > >> > > > > >> more
> > > > > >> > > > > >> > > > > reliable,
> > > > > >> > > > > >> > > > > > > more
> > > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
> > > > > >> > clients, and
> > > > > >> > > > > >> easier
> > > > > >> > > > > >> > > to
> > > > > >> > > > > >> > > > > debug
> > > > > >> > > > > >> > > > > > > for
> > > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > >> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > > Please take a look and let me
> > > know what
> > > > > >> > you
> > > > > >> > > > > think.
> > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > > >> > > > > >> > > > > > > > > > > > > David
> > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from July
> > > 18th to
> > > > > >> > August 8th.
> > > > > >> > > > > >> That
> > > > > >> > > > > >> > > gives
> > > > > >> > > > > >> > > > > > > you a
> > > > > >> > > > > >> > > > > > > > > bit
> > > > > >> > > > > >> > > > > > > > > > > > > of time to read and digest this
> > > long
> > > > > >> KIP.
> > > > > >> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > >
> > > > > >> > > > > >> > > > > > > > > > > --
> > > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > >> > > > > >> > > > > > > > >
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > >
> > > > > >> > > > > >> > > > > > > > --
> > > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > > >> > > > > >> > > > > > >
> > > > > >> > > > > >> > > > >
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> > >
> > > > > >> > > > > >> >
> > > > > >> > > > > >>
> > > > > >> > > > > >>
> > > > > >> > > > > >> --
> > > > > >> > > > > >> -- Guozhang
> > > > > >> > > > > >>
> > > > > >> > > > > >
> > > > > >> > > > >
> > > > > >> >
> > > > > >>
> > > > > >
> > >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Jun,

Thanks for your feedback. Let me start by answering your questions
inline and I will update the KIP next week.

> Thanks for the KIP. Overall, the main benefits of the KIP seem to be fewer
> RPCs during rebalance and more efficient support of wildcard. A few
> comments below.

I would also add that the KIP removes the global sync barrier in the
protocol which is essential to improve group stability and
scalability, and the KIP also simplifies the client by moving most of
the logic to the server side.

> 30. ConsumerGroupHeartbeatRequest
> 30.1 ServerAssignor is a singleton. Do we plan to support rolling changing
> of the partition assignor in the consumers?

Definitely. The group coordinator will use the assignor used by a
majority of the members. This allows the group to move from one
assignor to another by a roll. This is explained in the Assignor
Selection chapter.

> 30.2 For each field, could you explain whether it's required in every
> request or the scenarios when it needs to be filled? For example, it's not
> clear to me when TopicPartitions needs to be filled.

The client is expected to set those fields in case of a connection
issue (e.g. timeout) or when the fields have changed since the last
HB. The server populates those fields as long as the member is not
fully reconciled - the member should acknowledge that it has the
expected epoch and assignment. I will clarify this in the KIP.

> 31. In the current consumer protocol, the rack affinity between the client
> and the broker is only considered during fetching, but not during assigning
> partitions to consumers. Sometimes, once the assignment is made, there is
> no opportunity for read affinity because no replicas of assigned partitions
> are close to the member. I am wondering if we should use this opportunity
> to address this by including rack in GroupMember.

That's an interesting idea. I don't see any issue with adding the rack
to the members. I will do so.

> 32. On the metric side, often, it's useful to know how busy a group
> coordinator is. By moving the event loop model, it seems that we could add
> a metric that tracks the fraction of the time the event loop is doing the
> actual work.

That's a great idea. I will add it. Thanks.

> 33. Could we add a section on coordinator failover handling? For example,
> does it need to trigger the check if any group with the wildcard
> subscription now has a new matching topic?

Sure. When the new group coordinator takes over, it has to:
* Setup the session timeouts.
* Trigger a new assignment if a client side assignor is used. We don't
store the information about the member selected to run the assignment
so we have to start a new one.
* Update the topics metadata, verify the wildcard subscriptions, and
trigger a rebalance if needed.

> 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> ConsumerGroupMemberMetadataValue: Could we document what the epoch field
> reflects? For example, does the epoch in ConsumerGroupMetadataValue reflect
> the latest group epoch? What about the one in
> ConsumerGroupPartitionMetadataValue and ConsumerGroupMemberMetadataValue?

Sure. I will clarify that but it is always the latest group epoch.
When the group state is updated, the group epoch is bumped so we use
that one for all the change records related to the update.

> 35. "the group coordinator will ensure that the following invariants are
> met: ... All members exists." It's possible for a member not to get any
> assigned partitions, right?

That's right. Here I meant that the members provided by the assignor
in the assignment must exist in the group. The assignor can not make
up new member ids.

> 36. "He can rejoins the group with a member epoch equals to 0": When would
> a consumer rejoin and what member id would be used?

A member is expected to abandon all its partitions and rejoins when it
receives the FENCED_MEMBER_EPOCH error. In this case, the group
coordinator will have removed the member from the group. The member
can rejoin the group with the same member id but with 0 as epoch. Let
me see if I can clarify this in the KIP.

> 37. "Instead, power users will have the ability to trigger a reassignment
> by either providing a non-zero reason or by updating the assignor
> metadata." Hmm, this seems to be conflicting with the deprecation of
> Consumer#enforeRebalance.

In this case, a new assignment is triggered by the client side
assignor. When constructing the HB, the consumer will always consult
the client side assignor and propagate the information to the group
coordinator. In other words, we don't expect users to call
Consumer#enforceRebalance anymore.

> 38. The reassignment examples are nice. But the section seems to have
> multiple typos.
> 38.1 When the group transitions to epoch 2, B immediately gets into
> "epoch=1, partitions=[foo-2]", which seems incorrect.
> 38.2 When the group transitions to epoch 3, C seems to get into epoch=3,
> partitions=[foo-1] too early.
> 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> partitions=[foo-0].

Sorry for that! I will revise them.

> 39. Rolling upgrade of consumers: Do we support the upgrade from any old
> version to new one?

We will support upgrading from the consumer protocol version 3,
introduced in KIP-792. KIP-792 is not implemented yet so the earliest
version is unknown at the moment. This is explained in the migration
plan chapter.

Thanks again for your feedback, Jun. I will update the KIP based on it
next week.

Best,
David

On Thu, Sep 1, 2022 at 9:07 PM Jun Rao <ju...@confluent.io.invalid> wrote:
>
> Hi, David,
>
> Thanks for the KIP. Overall, the main benefits of the KIP seem to be fewer
> RPCs during rebalance and more efficient support of wildcard. A few
> comments below.
>
> 30. ConsumerGroupHeartbeatRequest
> 30.1 ServerAssignor is a singleton. Do we plan to support rolling changing
> of the partition assignor in the consumers?
> 30.2 For each field, could you explain whether it's required in every
> request or the scenarios when it needs to be filled? For example, it's not
> clear to me when TopicPartitions needs to be filled.
>
> 31. In the current consumer protocol, the rack affinity between the client
> and the broker is only considered during fetching, but not during assigning
> partitions to consumers. Sometimes, once the assignment is made, there is
> no opportunity for read affinity because no replicas of assigned partitions
> are close to the member. I am wondering if we should use this opportunity
> to address this by including rack in GroupMember.
>
> 32. On the metric side, often, it's useful to know how busy a group
> coordinator is. By moving the event loop model, it seems that we could add
> a metric that tracks the fraction of the time the event loop is doing the
> actual work.
>
> 33. Could we add a section on coordinator failover handling? For example,
> does it need to trigger the check if any group with the wildcard
> subscription now has a new matching topic?
>
> 34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
> ConsumerGroupMemberMetadataValue: Could we document what the epoch field
> reflects? For example, does the epoch in ConsumerGroupMetadataValue reflect
> the latest group epoch? What about the one in
> ConsumerGroupPartitionMetadataValue and ConsumerGroupMemberMetadataValue?
>
> 35. "the group coordinator will ensure that the following invariants are
> met: ... All members exists." It's possible for a member not to get any
> assigned partitions, right?
>
> 36. "He can rejoins the group with a member epoch equals to 0": When would
> a consumer rejoin and what member id would be used?
>
> 37. "Instead, power users will have the ability to trigger a reassignment
> by either providing a non-zero reason or by updating the assignor
> metadata." Hmm, this seems to be conflicting with the deprecation of
> Consumer#enforeRebalance.
>
> 38. The reassignment examples are nice. But the section seems to have
> multiple typos.
> 38.1 When the group transitions to epoch 2, B immediately gets into
> "epoch=1, partitions=[foo-2]", which seems incorrect.
> 38.2 When the group transitions to epoch 3, C seems to get into epoch=3,
> partitions=[foo-1] too early.
> 38.3 After A transitions to epoch 3, C still has A - epoch=2,
> partitions=[foo-0].
>
> 39. Rolling upgrade of consumers: Do we support the upgrade from any old
> version to new one?
>
> Thanks,
>
> Jun
>
> On Mon, Aug 29, 2022 at 9:20 AM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi all,
> >
> > The KIP states that we will re-implement the coordinator in Java. I
> > discussed this offline with a few folks and folks are concerned that
> > we could introduce many regressions in the old protocol if we do so.
> > Therefore, I am going to remove this statement from the KIP. It is an
> > implementation detail after all so it does not have to be decided at
> > this stage. We will likely start by trying to refactor the current
> > implementation as a first step.
> >
> > Cheers,
> > David
> >
> > On Mon, Aug 29, 2022 at 3:52 PM David Jacot <dj...@confluent.io> wrote:
> > >
> > > Hi Luke,
> > >
> > > > 1.1. I think the state machine are: "Empty, assigning, reconciling,
> > stable,
> > > > dead" mentioned in Consumer Group States section, right?
> > >
> > > This sentence does not refer to those group states but rather to a
> > > state machine replication (SMR). This refers to the entire state of
> > > group coordinator which is replicated via the log layer. I will
> > > clarify this in the KIP.
> > >
> > > > 1.2. What do you mean "each state machine is modelled as an event
> > loop"?
> > >
> > > The idea is to follow a model similar to the new quorum controller. We
> > > will have N threads to process events. Each __consumer_offsets
> > > partition is assigned to a unique thread and all the events (e.g.
> > > requests, callbacks, etc.) are processed by this thread. This simplify
> > > concurrency and will enable us to do simulation testing for the group
> > > coordinator.
> > >
> > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > partitions?
> > > > Not a state machine "per consumer group" owned by a group coordinator?
> > For
> > > > example, if one group coordinator owns 2 consumer groups, and both
> > exist in
> > > > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> > >
> > > See 1.1. The confusion comes from there, I think.
> > >
> > > > 1.4. I know the "*group.coordinator.threads" *is the number of threads
> > used
> > > > to run the state machines. But I'm wondering if the purpose of the
> > threads
> > > > is only to keep the state of each consumer group (or
> > *__consumer_offsets*
> > > > partitions?), and no heavy computation, why should we need
> > multi-threads
> > > > here?
> > >
> > > See 1.2. The idea is to have an ability to shard the processing as the
> > > computation could be heavy.
> > >
> > > > 2.1. The consumer session timeout, why does the default session
> > timeout not
> > > > locate between min (45s) and max(60s)? I thought the min/max session
> > > > timeout is to define lower/upper bound of it, no?
> > > >
> > > > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > > > failures when using the consumer group protocol.
> > > > group.consumer.min.session.timeout.ms int 45s The minimum session
> > timeout.
> > > > group.consumer.max.session.timeout.ms int 60s The maximum session
> > timeout.
> > >
> > > This is indeed a mistake. The default session timeout should be 45s
> > > (the current default).
> > >
> > > > 2.2. The default server side assignor are [range, uniform], which means
> > > > we'll default to "range" assignor. I'd like to know why not uniform
> > one? I
> > > > thought usually users will choose uniform assignor (former sticky
> > assinor)
> > > > for better evenly distribution. Any other reason we choose range
> > assignor
> > > > as default?
> > > > group.consumer.assignors List range, uniform The server side assignors.
> > >
> > > The order on the server side has no influence because the client must
> > > chose the selector that he wants to use. There is no default in the
> > > current proposal. If the assignor is not specified by the client, the
> > > request is rejected. The default client value for
> > > `group.remote.assignor` is `uniform` though.
> > >
> > > Thanks for your very good comments, Luke. I hope that my answers help
> > > to clarify things. I will update the KIP as well based on your
> > > feedback.
> > >
> > > Cheers,
> > > David
> > >
> > > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com> wrote:
> > > >
> > > > Hi David,
> > > >
> > > > Thanks for the update.
> > > >
> > > > Some more questions:
> > > > 1. In Group Coordinator section, you mentioned:
> > > > > The new group coordinator will have a state machine per
> > > > *__consumer_offsets* partitions, where each state machine is modelled
> > as an
> > > > event loop. Those state machines will be executed in
> > > > *group.coordinator.threads* threads.
> > > >
> > > > 1.1. I think the state machine are: "Empty, assigning, reconciling,
> > stable,
> > > > dead" mentioned in Consumer Group States section, right?
> > > > 1.2. What do you mean "each state machine is modelled as an event
> > loop"?
> > > > 1.3. Why do we need a state machine per *__consumer_offsets*
> > partitions?
> > > > Not a state machine "per consumer group" owned by a group coordinator?
> > For
> > > > example, if one group coordinator owns 2 consumer groups, and both
> > exist in
> > > > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> > > > 1.4. I know the "*group.coordinator.threads" *is the number of threads
> > used
> > > > to run the state machines. But I'm wondering if the purpose of the
> > threads
> > > > is only to keep the state of each consumer group (or
> > *__consumer_offsets*
> > > > partitions?), and no heavy computation, why should we need
> > multi-threads
> > > > here?
> > > >
> > > > 2. For the default value in the new configs:
> > > > 2.1. The consumer session timeout, why does the default session
> > timeout not
> > > > locate between min (45s) and max(60s)? I thought the min/max session
> > > > timeout is to define lower/upper bound of it, no?
> > > >
> > > > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > > > failures when using the consumer group protocol.
> > > > group.consumer.min.session.timeout.ms int 45s The minimum session
> > timeout.
> > > > group.consumer.max.session.timeout.ms int 60s The maximum session
> > timeout.
> > > >
> > > >
> > > >
> > > > 2.2. The default server side assignor are [range, uniform], which means
> > > > we'll default to "range" assignor. I'd like to know why not uniform
> > one? I
> > > > thought usually users will choose uniform assignor (former sticky
> > assinor)
> > > > for better evenly distribution. Any other reason we choose range
> > assignor
> > > > as default?
> > > > group.consumer.assignors List range, uniform The server side assignors.
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > Thank you.
> > > > Luke
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com> wrote:
> > > >
> > > > > Hi Sagar,
> > > > >
> > > > > I have some thoughts about Kafka Connect integrating with KIP-848,
> > but I
> > > > > think we should have a separate discussion thread for the Kafka
> > Connect
> > > > > KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol
> > [1],
> > > > > and let this discussion thread focus on consumer rebalance protocol,
> > WDYT?
> > > > >
> > > > > [1]
> > > > >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >
> > > > > Thank you.
> > > > > Luke
> > > > >
> > > > >
> > > > >
> > > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com>
> > wrote:
> > > > >
> > > > >> Thank you Guozhang/David for the feedback. Looks like there's
> > agreement on
> > > > >> using separate APIs for Connect. I would revisit the doc and see
> > what
> > > > >> changes are to be made.
> > > > >>
> > > > >> Thanks!
> > > > >> Sagar.
> > > > >>
> > > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> > <dj...@confluent.io.invalid>
> > > > >> wrote:
> > > > >>
> > > > >> > Hi Sagar,
> > > > >> >
> > > > >> > Thanks for the feedback and the document. That's really helpful. I
> > > > >> > will take a look at it.
> > > > >> >
> > > > >> > Overall, it seems to me that both Connect and the Consumer could
> > share
> > > > >> > the same underlying "engine". The main difference is that the
> > Consumer
> > > > >> > assigns topic-partitions to members whereas Connect assigns tasks
> > to
> > > > >> > workers. I see two ways to move forward:
> > > > >> > 1) We extend the new proposed APIs to support different resource
> > types
> > > > >> > (e.g. partitions, tasks, etc.); or
> > > > >> > 2) We use new dedicated APIs for Connect. The dedicated APIs
> > would be
> > > > >> > similar to the new ones but different on the content/resources and
> > > > >> > they would rely on the same engine on the coordinator side.
> > > > >> >
> > > > >> > I personally lean towards 2) because I am not a fan of
> > overcharging
> > > > >> > APIs to serve different purposes. That being said, I am not
> > opposed to
> > > > >> > 1) if we can find an elegant way to do it.
> > > > >> >
> > > > >> > I think that we can continue to discuss it here for now in order
> > to
> > > > >> > ensure that this KIP is compatible with what we will do for
> > Connect in
> > > > >> > the future.
> > > > >> >
> > > > >> > Best,
> > > > >> > David
> > > > >> >
> > > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io>
> > wrote:
> > > > >> > >
> > > > >> > > Hi all,
> > > > >> > >
> > > > >> > > I am back from vacation. I will go through and address your
> > comments
> > > > >> > > in the coming days. Thanks for your feedback.
> > > > >> > >
> > > > >> > > Cheers,
> > > > >> > > David
> > > > >> > >
> > > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> > gharris1727@gmail.com
> > > > >> >
> > > > >> > wrote:
> > > > >> > > >
> > > > >> > > > Hey All!
> > > > >> > > >
> > > > >> > > > Thanks for the KIP, it's wonderful to see cooperative
> > rebalancing
> > > > >> > making it
> > > > >> > > > down the stack!
> > > > >> > > >
> > > > >> > > > I had a few questions:
> > > > >> > > >
> > > > >> > > > 1. The 'Rejected Alternatives' section describes how member
> > epoch
> > > > >> > should
> > > > >> > > > advance in step with the group epoch and assignment epoch
> > values. I
> > > > >> > think
> > > > >> > > > that this is a good idea for the reasons described in the
> > KIP. When
> > > > >> the
> > > > >> > > > protocol is incrementally assigning partitions to a worker,
> > what
> > > > >> member
> > > > >> > > > epoch does each incremental assignment use? Are member epochs
> > > > >> re-used,
> > > > >> > and
> > > > >> > > > a single member epoch can correspond to multiple different
> > > > >> > (monotonically
> > > > >> > > > larger) assignments?
> > > > >> > > >
> > > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> > > > >> coordinator? If
> > > > >> > > > not, should custom client-side assignor implementations
> > interact
> > > > >> with
> > > > >> > the
> > > > >> > > > Reason field, and how is its common meaning agreed upon? If
> > so, what
> > > > >> > is the
> > > > >> > > > benefit of a distinct Reason field over including such
> > functionality
> > > > >> > in the
> > > > >> > > > opaque metadata?
> > > > >> > > >
> > > > >> > > > 3. The following is included in the KIP: "Thanks to this, the
> > input
> > > > >> of
> > > > >> > the
> > > > >> > > > client side assignor is entirely driven by the group
> > coordinator.
> > > > >> The
> > > > >> > > > consumer is no longer responsible for maintaining any state
> > besides
> > > > >> its
> > > > >> > > > assigned partitions." Does this mean that the client-side
> > assignor
> > > > >> MAY
> > > > >> > > > incorporate additional non-Metadata state (such as partition
> > > > >> > throughput,
> > > > >> > > > cpu/memory metrics, config topics, etc), or that additional
> > > > >> > non-Metadata
> > > > >> > > > state SHOULD NOT be used?
> > > > >> > > >
> > > > >> > > > 4. I see that there are separate classes
> > > > >> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that
> > seem to
> > > > >> > > > overlap significantly. Is it possible for these two
> > implementations
> > > > >> to
> > > > >> > be
> > > > >> > > > unified? This would serve to promote feature parity of
> > server-side
> > > > >> and
> > > > >> > > > client-side assignors, and would also facilitate operational
> > > > >> > flexibility in
> > > > >> > > > certain situations. For example, if a server-side assignor
> > has some
> > > > >> > poor
> > > > >> > > > behavior and needs a patch, deploying the patched assignor to
> > the
> > > > >> > client
> > > > >> > > > and switching one consumer group to a client-side assignor
> > may be
> > > > >> > faster
> > > > >> > > > and less risky than patching all of the brokers. With the
> > currently
> > > > >> > > > proposed distinct APIs, a non-trivial reimplementation would
> > have
> > > > >> to be
> > > > >> > > > assembled, and if the two APIs have diverged significantly,
> > then it
> > > > >> is
> > > > >> > > > possible that a reimplementation would not be possible.
> > > > >> > > >
> > > > >> > > > --
> > > > >> > > > Greg Harris
> > > > >> > > > gharris1727@gmail.com
> > > > >> > > > github.com/gharris1727
> > > > >> > > >
> > > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> > sagarmeansocean@gmail.com>
> > > > >> > wrote:
> > > > >> > > >
> > > > >> > > > > Hi Guozhang/David,
> > > > >> > > > >
> > > > >> > > > > I created a confluence page to discuss how Connect would
> > need to
> > > > >> > change
> > > > >> > > > > based on the new rebalance protocol. Here's the page:
> > > > >> > > > >
> > > > >> > > > >
> > > > >> >
> > > > >>
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >> > > > >
> > > > >> > > > > It's also pretty longish and I have tried to keep a format
> > > > >> similar to
> > > > >> > > > > KIP-848. Let me know what you think. Also, do you think this
> > > > >> should
> > > > >> > be
> > > > >> > > > > moved to a separate discussion thread or is this one fine?
> > > > >> > > > >
> > > > >> > > > > Thanks!
> > > > >> > > > > Sagar.
> > > > >> > > > >
> > > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> > sagarmeansocean@gmail.com>
> > > > >> > wrote:
> > > > >> > > > >
> > > > >> > > > > > Hello Guozhang,
> > > > >> > > > > >
> > > > >> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I
> > would do
> > > > >> > the
> > > > >> > > > > > analysis and come up with such a document.
> > > > >> > > > > >
> > > > >> > > > > > Thanks!
> > > > >> > > > > > Sagar.
> > > > >> > > > > >
> > > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > > >> wangguoz@gmail.com>
> > > > >> > > > > wrote:
> > > > >> > > > > >
> > > > >> > > > > >> Hello Sagar,
> > > > >> > > > > >>
> > > > >> > > > > >> It would be great if you could come back with some
> > analysis on
> > > > >> > how to
> > > > >> > > > > >> implement the Connect side integration with the new
> > protocol;
> > > > >> so
> > > > >> > far
> > > > >> > > > > >> besides leveraging on the new "protocol type" we did not
> > yet
> > > > >> think
> > > > >> > > > > through
> > > > >> > > > > >> the Connect side implementations. For Streams here's a
> > draft of
> > > > >> > > > > >> integration
> > > > >> > > > > >> plan:
> > > > >> > > > > >>
> > > > >> > > > > >>
> > > > >> > > > >
> > > > >> >
> > > > >>
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > >> > > > > >> just FYI for your analysis on Connect.
> > > > >> > > > > >>
> > > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > > >> sagarmeansocean@gmail.com
> > > > >> > >
> > > > >> > > > > wrote:
> > > > >> > > > > >>
> > > > >> > > > > >> > Hi David,
> > > > >> > > > > >> >
> > > > >> > > > > >> > Thank you for your response. The reason I thought
> > connect can
> > > > >> > also fit
> > > > >> > > > > >> into
> > > > >> > > > > >> > this new scheme is that even today the connect uses a
> > > > >> > > > > WorkerCoordinator
> > > > >> > > > > >> > extending from AbstractCoordinator to empower
> > rebalances of
> > > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> > > > >> protocolType()
> > > > >> > to
> > > > >> > > > > >> connect
> > > > >> > > > > >> > and uses the metadata() method by plumbing into
> > > > >> > > > > >> JoinGroupRequestProtocol.
> > > > >> > > > > >> >
> > > > >> > > > > >> > I think the changes to support connect would be
> > similar at a
> > > > >> > high
> > > > >> > > > > level
> > > > >> > > > > >> to
> > > > >> > > > > >> > the changes in streams mainly because of the Client
> > side
> > > > >> > assignors
> > > > >> > > > > being
> > > > >> > > > > >> > used in both. At an implementation level, we might
> > need to
> > > > >> make
> > > > >> > a lot
> > > > >> > > > > of
> > > > >> > > > > >> > changes to get onto this new assignment protocol like
> > > > >> enhancing
> > > > >> > the
> > > > >> > > > > >> > JoinGroup request/response and SyncGroup and using
> > > > >> > > > > >> ConsumerGroupHeartbeat
> > > > >> > > > > >> > API etc again on similar lines to streams (or there
> > might be
> > > > >> > > > > >> deviations). I
> > > > >> > > > > >> > would try to perform a detailed analysis of the same
> > and we
> > > > >> > can have
> > > > >> > > > > a
> > > > >> > > > > >> > separate discussion thread for that as that would
> > derail this
> > > > >> > > > > discussion
> > > > >> > > > > >> > thread. Let me know if that sounds good to you.
> > > > >> > > > > >> >
> > > > >> > > > > >> > Thanks!
> > > > >> > > > > >> > Sagar.
> > > > >> > > > > >> >
> > > > >> > > > > >> >
> > > > >> > > > > >> >
> > > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > >> > > > > <djacot@confluent.io.invalid
> > > > >> > > > > >> >
> > > > >> > > > > >> > wrote:
> > > > >> > > > > >> >
> > > > >> > > > > >> > > Hi Sagar,
> > > > >> > > > > >> > >
> > > > >> > > > > >> > > Thanks for your comments.
> > > > >> > > > > >> > >
> > > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I
> > can
> > > > >> > mention it.
> > > > >> > > > > >> > >
> > > > >> > > > > >> > > 2) The idea is to transition C from his current
> > assignment
> > > > >> to
> > > > >> > his
> > > > >> > > > > >> > > target assignment when he can move to epoch 3. When
> > that
> > > > >> > happens,
> > > > >> > > > > the
> > > > >> > > > > >> > > member assignment is updated and persisted with all
> > its
> > > > >> > assigned
> > > > >> > > > > >> > > partitions even if they are not all revoked yet. In
> > other
> > > > >> > words, the
> > > > >> > > > > >> > > member assignment becomes the target assignment.
> > This is
> > > > >> > basically
> > > > >> > > > > an
> > > > >> > > > > >> > > optimization to avoid having to write all the
> > changes to
> > > > >> the
> > > > >> > log.
> > > > >> > > > > The
> > > > >> > > > > >> > > examples are based on the persisted state so I
> > understand
> > > > >> the
> > > > >> > > > > >> > > confusion. Let me see if I can improve this in the
> > > > >> > description.
> > > > >> > > > > >> > >
> > > > >> > > > > >> > > 3) Regarding Connect, it could reuse the protocol
> > with a
> > > > >> > client side
> > > > >> > > > > >> > > assignor if it fits in the protocol. The assignment
> > is
> > > > >> about
> > > > >> > > > > >> > > topicid-partitions + metadata, could Connect fit
> > into this?
> > > > >> > > > > >> > >
> > > > >> > > > > >> > > Best,
> > > > >> > > > > >> > > David
> > > > >> > > > > >> > >
> > > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > > >> > sagarmeansocean@gmail.com>
> > > > >> > > > > >> wrote:
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > Hi David,
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > Thanks for the KIP. I just had minor observations:
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > 1) In the Assignment Error section in Client Side
> > mode
> > > > >> > Assignment
> > > > >> > > > > >> > > process,
> > > > >> > > > > >> > > > you mentioned => `In this case, the client side
> > assignor
> > > > >> can
> > > > >> > > > > return
> > > > >> > > > > >> an
> > > > >> > > > > >> > > > error to the group coordinator`. In this case are
> > you
> > > > >> > referring to
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > Assignor returning an AssignmentError that's
> > listed down
> > > > >> > towards
> > > > >> > > > > the
> > > > >> > > > > >> > end?
> > > > >> > > > > >> > > > If yes, do you think it would make sense to
> > mention this
> > > > >> > > > > explicitly
> > > > >> > > > > >> > here?
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > 2) In the Case Studies section, I have a slight
> > > > >> confusion,
> > > > >> > not
> > > > >> > > > > sure
> > > > >> > > > > >> if
> > > > >> > > > > >> > > > others have the same. Consider this step:
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > When B heartbeats, the group coordinator
> > transitions him
> > > > >> to
> > > > >> > epoch
> > > > >> > > > > 3
> > > > >> > > > > >> > > because
> > > > >> > > > > >> > > > B has no partitions to revoke. It persists the
> > change and
> > > > >> > reply.
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > >    - Group (epoch=3)
> > > > >> > > > > >> > > >       - A
> > > > >> > > > > >> > > >       - B
> > > > >> > > > > >> > > >       - C
> > > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > > >> > > > > >> > > >    - Member Assignment
> > > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but
> > cannot
> > > > >> get
> > > > >> > foo-1
> > > > >> > > > > >> yet.
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > Here,it's mentioned that member C can't get the
> > foo-1
> > > > >> > partition
> > > > >> > > > > yet,
> > > > >> > > > > >> > but
> > > > >> > > > > >> > > > based on the description above, it seems it
> > already has
> > > > >> it.
> > > > >> > Do you
> > > > >> > > > > >> > think
> > > > >> > > > > >> > > it
> > > > >> > > > > >> > > > would be better to remove it and populate it only
> > when it
> > > > >> > actually
> > > > >> > > > > >> gets
> > > > >> > > > > >> > > it?
> > > > >> > > > > >> > > > I see this in a lot of other places, so have I
> > > > >> understood it
> > > > >> > > > > >> > incorrectly
> > > > >> > > > > >> > > ?
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > Regarding connect , it might be out of scope of
> > this
> > > > >> > discussion,
> > > > >> > > > > but
> > > > >> > > > > >> > from
> > > > >> > > > > >> > > > what I understood it would probably be running in
> > client
> > > > >> > side
> > > > >> > > > > >> assignor
> > > > >> > > > > >> > > mode
> > > > >> > > > > >> > > > even on the new rebalance protocol as it has its
> > own
> > > > >> Custom
> > > > >> > > > > >> > > Assignors(Eager
> > > > >> > > > > >> > > > and IncrementalCooperative).
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > Thanks!
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > Sagar.
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > > >> > > > > >> > <djacot@confluent.io.invalid
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > wrote:
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
> > > > >> > specialized API
> > > > >> > > > > >> > > > > instead of relying on one generic API. For
> > Connect, we
> > > > >> > can apply
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > exact same pattern and reuse/share the core
> > > > >> > implementation on
> > > > >> > > > > the
> > > > >> > > > > >> > > > > server side. For the schema registry, I think
> > that we
> > > > >> > should
> > > > >> > > > > >> consider
> > > > >> > > > > >> > > > > having a tailored API to do simple
> > membership/leader
> > > > >> > election.
> > > > >> > > > > >> > > > >
> > > > >> > > > > >> > > > > Best,
> > > > >> > > > > >> > > > > David
> > > > >> > > > > >> > > > >
> > > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> > > > >> > ismael@juma.me.uk
> > > > >> > > > > >
> > > > >> > > > > >> > > wrote:
> > > > >> > > > > >> > > > > >
> > > > >> > > > > >> > > > > > Three quick comments:
> > > > >> > > > > >> > > > > >
> > > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > > >> > > > > >> com.google.re2j.Pattern, we
> > > > >> > > > > >> > > > > should
> > > > >> > > > > >> > > > > > document the differences in more detail before
> > > > >> deciding
> > > > >> > one
> > > > >> > > > > way
> > > > >> > > > > >> or
> > > > >> > > > > >> > > > > another.
> > > > >> > > > > >> > > > > > That said, if people pass
> > java.util.regex.Pattern,
> > > > >> they
> > > > >> > expect
> > > > >> > > > > >> > their
> > > > >> > > > > >> > > > > > semantics to be honored. If we are doing
> > something
> > > > >> > different,
> > > > >> > > > > >> then
> > > > >> > > > > >> > we
> > > > >> > > > > >> > > > > > should consider adding an overload with our own
> > > > >> Pattern
> > > > >> > class
> > > > >> > > > > (I
> > > > >> > > > > >> > > don't
> > > > >> > > > > >> > > > > > think we'd want to expose re2j's at this
> > point).
> > > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol
> > should
> > > > >> > > > > integrate
> > > > >> > > > > >> > fully
> > > > >> > > > > >> > > > > with
> > > > >> > > > > >> > > > > > it and should handle the topic recreation case
> > > > >> > correctly.
> > > > >> > > > > That's
> > > > >> > > > > >> > the
> > > > >> > > > > >> > > main
> > > > >> > > > > >> > > > > > part we need to handle. I agree with David
> > that we'd
> > > > >> > want to
> > > > >> > > > > add
> > > > >> > > > > >> > > topic
> > > > >> > > > > >> > > > > ids
> > > > >> > > > > >> > > > > > to the relevant protocols that don't have it
> > yet and
> > > > >> > that we
> > > > >> > > > > can
> > > > >> > > > > >> > > probably
> > > > >> > > > > >> > > > > > focus on the internals versus adding new APIs
> > to the
> > > > >> > Java
> > > > >> > > > > >> Consumer
> > > > >> > > > > >> > > > > (unless
> > > > >> > > > > >> > > > > > we find that adding new APIs is required for
> > > > >> reasonable
> > > > >> > > > > >> semantics).
> > > > >> > > > > >> > > > > > 3. I am still not sure about the coordinator
> > storing
> > > > >> the
> > > > >> > > > > >> configs.
> > > > >> > > > > >> > > It's
> > > > >> > > > > >> > > > > > powerful for configs to be centralized in the
> > > > >> metadata
> > > > >> > log for
> > > > >> > > > > >> > > various
> > > > >> > > > > >> > > > > > reasons (auditability, visibility, consistency,
> > > > >> etc.).
> > > > >> > > > > >> Similarly, I
> > > > >> > > > > >> > > am
> > > > >> > > > > >> > > > > not
> > > > >> > > > > >> > > > > > sure about automatically deleting configs in a
> > way
> > > > >> that
> > > > >> > they
> > > > >> > > > > >> cannot
> > > > >> > > > > >> > > be
> > > > >> > > > > >> > > > > > recovered. A good property for modern systems
> > is to
> > > > >> > minimize
> > > > >> > > > > the
> > > > >> > > > > >> > > number
> > > > >> > > > > >> > > > > of
> > > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > >> > > > > >> > > > > >
> > > > >> > > > > >> > > > > > Ismael
> > > > >> > > > > >> > > > > >
> > > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > > >> > > > > >> > > > > >
> > > > >> > > > > >> > > > > > wrote:
> > > > >> > > > > >> > > > > >
> > > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > > 1) the migration path, especially the last
> > step
> > > > >> when
> > > > >> > > > > clients
> > > > >> > > > > >> > > flip the
> > > > >> > > > > >> > > > > > > flag
> > > > >> > > > > >> > > > > > > > to enable the new protocol, in which we
> > would
> > > > >> have a
> > > > >> > > > > window
> > > > >> > > > > >> > where
> > > > >> > > > > >> > > > > both
> > > > >> > > > > >> > > > > > > new
> > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs
> > are
> > > > >> used
> > > > >> > by
> > > > >> > > > > >> members
> > > > >> > > > > >> > of
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > same
> > > > >> > > > > >> > > > > > > > group. How the coordinator could "mimic"
> > the old
> > > > >> > behavior
> > > > >> > > > > >> while
> > > > >> > > > > >> > > > > using the
> > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > present
> > > > >> about.
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > Noted. I just published a new version of KIP
> > which
> > > > >> > includes
> > > > >> > > > > >> more
> > > > >> > > > > >> > > > > > > details about this. See the "Supporting
> > Online
> > > > >> > Consumer
> > > > >> > > > > Group
> > > > >> > > > > >> > > Upgrade"
> > > > >> > > > > >> > > > > > > and the "Compatibility, Deprecation, and
> > Migration
> > > > >> > Plan". I
> > > > >> > > > > >> think
> > > > >> > > > > >> > > that
> > > > >> > > > > >> > > > > > > I have to think through a few cases now but
> > the
> > > > >> > overall idea
> > > > >> > > > > >> and
> > > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > KIP-516 the
> > > > >> > topic ids
> > > > >> > > > > >> are
> > > > >> > > > > >> > > only
> > > > >> > > > > >> > > > > used
> > > > >> > > > > >> > > > > > > as
> > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but they
> > are not
> > > > >> > exposed
> > > > >> > > > > via
> > > > >> > > > > >> any
> > > > >> > > > > >> > > > > public
> > > > >> > > > > >> > > > > > > APIs
> > > > >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> > first
> > > > >> > should we
> > > > >> > > > > >> let
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > consumer
> > > > >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> > mapping
> > > > >> > itself
> > > > >> > > > > to
> > > > >> > > > > >> > fully
> > > > >> > > > > >> > > > > > > leverage
> > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the
> > new
> > > > >> RPCs
> > > > >> > with
> > > > >> > > > > the
> > > > >> > > > > >> > > topic
> > > > >> > > > > >> > > > > ids;
> > > > >> > > > > >> > > > > > > and
> > > > >> > > > > >> > > > > > > > secondly, should we ever consider exposing
> > the
> > > > >> > topic ids
> > > > >> > > > > in
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > consumer
> > > > >> > > > > >> > > > > > > > public APIs as well (both
> > subscribe/assign, as
> > > > >> well
> > > > >> > as in
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > rebalance
> > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > >> > deletion-and-recreation).
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > a) Assuming that we would include converting
> > all
> > > > >> the
> > > > >> > offsets
> > > > >> > > > > >> > > related
> > > > >> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the
> > consumer
> > > > >> > would be
> > > > >> > > > > >> able
> > > > >> > > > > >> > to
> > > > >> > > > > >> > > > > > > fully operate with topic ids. That being
> > said, it
> > > > >> > still has
> > > > >> > > > > to
> > > > >> > > > > >> > > provide
> > > > >> > > > > >> > > > > > > the topics names in various APIs so having a
> > > > >> mapping
> > > > >> > in the
> > > > >> > > > > >> > > consumer
> > > > >> > > > > >> > > > > > > seems inevitable to me.
> > > > >> > > > > >> > > > > > > b) I don't have a strong opinion on this.
> > Here I
> > > > >> > wonder if
> > > > >> > > > > >> this
> > > > >> > > > > >> > > goes
> > > > >> > > > > >> > > > > > > beyond the scope of this KIP. I would rather
> > focus
> > > > >> on
> > > > >> > the
> > > > >> > > > > >> > internals
> > > > >> > > > > >> > > > > > > here and we can consider this separately if
> > we see
> > > > >> > value in
> > > > >> > > > > >> doing
> > > > >> > > > > >> > > it.
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > Coming back to Ismael's point about using
> > topic ids
> > > > >> > in the
> > > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that
> > there
> > > > >> is
> > > > >> > one
> > > > >> > > > > >> > advantage
> > > > >> > > > > >> > > in
> > > > >> > > > > >> > > > > > > favour of it. The consumer will have the
> > > > >> opportunity
> > > > >> > to
> > > > >> > > > > >> validate
> > > > >> > > > > >> > > that
> > > > >> > > > > >> > > > > > > the topics exists before passing them into
> > the
> > > > >> group
> > > > >> > > > > rebalance
> > > > >> > > > > >> > > > > > > protocol. Obviously, the coordinator will
> > also
> > > > >> notice
> > > > >> > it but
> > > > >> > > > > >> it
> > > > >> > > > > >> > > does
> > > > >> > > > > >> > > > > > > not really have a way to reject an invalid
> > topic in
> > > > >> > the
> > > > >> > > > > >> response.
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > > > >> questions
> > > > >> > > > > except
> > > > >> > > > > >> for
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally
> > I think
> > > > >> > it's not
> > > > >> > > > > >> > > necessary
> > > > >> > > > > >> > > > > to
> > > > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern,
> > but
> > > > >> > instead we
> > > > >> > > > > >> still
> > > > >> > > > > >> > > use
> > > > >> > > > > >> > > > > > > Pattern
> > > > >> > > > > >> > > > > > > > while just documenting that our
> > subscription may
> > > > >> be
> > > > >> > > > > >> rejected by
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > server.
> > > > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > > > >> scenario
> > > > >> > I felt
> > > > >> > > > > >> > using
> > > > >> > > > > >> > > an
> > > > >> > > > > >> > > > > > > > overloaded `String` based subscription may
> > be
> > > > >> more
> > > > >> > > > > >> vulnerable
> > > > >> > > > > >> > to
> > > > >> > > > > >> > > > > various
> > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > That could work. I have to look at the
> > differences
> > > > >> > between
> > > > >> > > > > the
> > > > >> > > > > >> > two
> > > > >> > > > > >> > > > > > > engines to better understand the potential
> > issues.
> > > > >> My
> > > > >> > > > > >> > > understanding is
> > > > >> > > > > >> > > > > > > that would work for all the basic regular
> > > > >> > expressions. The
> > > > >> > > > > >> > > differences
> > > > >> > > > > >> > > > > > > between the two are mainly about the various
> > > > >> character
> > > > >> > > > > >> classes. I
> > > > >> > > > > >> > > > > > > wonder what other people think about this.
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > Best,
> > > > >> > > > > >> > > > > > > David
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang
> > Wang <
> > > > >> > > > > >> > wangguoz@gmail.com
> > > > >> > > > > >> > > >
> > > > >> > > > > >> > > > > wrote:
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > Thanks David! I think on the high level
> > there are
> > > > >> > two meta
> > > > >> > > > > >> > > points we
> > > > >> > > > > >> > > > > need
> > > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > 1) the migration path, especially the last
> > step
> > > > >> when
> > > > >> > > > > clients
> > > > >> > > > > >> > > flip the
> > > > >> > > > > >> > > > > > > flag
> > > > >> > > > > >> > > > > > > > to enable the new protocol, in which we
> > would
> > > > >> have a
> > > > >> > > > > window
> > > > >> > > > > >> > where
> > > > >> > > > > >> > > > > both
> > > > >> > > > > >> > > > > > > new
> > > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs
> > are
> > > > >> used
> > > > >> > by
> > > > >> > > > > >> members
> > > > >> > > > > >> > of
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > same
> > > > >> > > > > >> > > > > > > > group. How the coordinator could "mimic"
> > the old
> > > > >> > behavior
> > > > >> > > > > >> while
> > > > >> > > > > >> > > > > using the
> > > > >> > > > > >> > > > > > > > new protocol is something we need to
> > present
> > > > >> about.
> > > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> > KIP-516 the
> > > > >> > topic ids
> > > > >> > > > > >> are
> > > > >> > > > > >> > > only
> > > > >> > > > > >> > > > > used
> > > > >> > > > > >> > > > > > > as
> > > > >> > > > > >> > > > > > > > part of RPCs and admin client, but they
> > are not
> > > > >> > exposed
> > > > >> > > > > via
> > > > >> > > > > >> any
> > > > >> > > > > >> > > > > public
> > > > >> > > > > >> > > > > > > APIs
> > > > >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> > first
> > > > >> > should we
> > > > >> > > > > >> let
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > consumer
> > > > >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> > mapping
> > > > >> > itself
> > > > >> > > > > to
> > > > >> > > > > >> > fully
> > > > >> > > > > >> > > > > > > leverage
> > > > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the
> > new
> > > > >> RPCs
> > > > >> > with
> > > > >> > > > > the
> > > > >> > > > > >> > > topic
> > > > >> > > > > >> > > > > ids;
> > > > >> > > > > >> > > > > > > and
> > > > >> > > > > >> > > > > > > > secondly, should we ever consider exposing
> > the
> > > > >> > topic ids
> > > > >> > > > > in
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > consumer
> > > > >> > > > > >> > > > > > > > public APIs as well (both
> > subscribe/assign, as
> > > > >> well
> > > > >> > as in
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > rebalance
> > > > >> > > > > >> > > > > > > > listener for cases like topic
> > > > >> > deletion-and-recreation).
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > > > >> questions
> > > > >> > > > > except
> > > > >> > > > > >> for
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally
> > I think
> > > > >> > it's not
> > > > >> > > > > >> > > necessary
> > > > >> > > > > >> > > > > to
> > > > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern,
> > but
> > > > >> > instead we
> > > > >> > > > > >> still
> > > > >> > > > > >> > > use
> > > > >> > > > > >> > > > > > > Pattern
> > > > >> > > > > >> > > > > > > > while just documenting that our
> > subscription may
> > > > >> be
> > > > >> > > > > >> rejected by
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > server.
> > > > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > > > >> scenario
> > > > >> > I felt
> > > > >> > > > > >> > using
> > > > >> > > > > >> > > an
> > > > >> > > > > >> > > > > > > > overloaded `String` based subscription may
> > be
> > > > >> more
> > > > >> > > > > >> vulnerable
> > > > >> > > > > >> > to
> > > > >> > > > > >> > > > > various
> > > > >> > > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > Guozhang
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > wrote:
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer
> > your
> > > > >> > questions
> > > > >> > > > > >> > inline.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 1. I think it's premature to talk about
> > > > >> target
> > > > >> > > > > versions
> > > > >> > > > > >> for
> > > > >> > > > > >> > > > > > > deprecation
> > > > >> > > > > >> > > > > > > > > and
> > > > >> > > > > >> > > > > > > > > > removal of the existing group protocol.
> > > > >> Unlike
> > > > >> > KRaft,
> > > > >> > > > > >> this
> > > > >> > > > > >> > > > > affects a
> > > > >> > > > > >> > > > > > > core
> > > > >> > > > > >> > > > > > > > > > client protocol and hence
> > deprecation/removal
> > > > >> > will be
> > > > >> > > > > >> > heavily
> > > > >> > > > > >> > > > > > > dependent
> > > > >> > > > > >> > > > > > > > > on
> > > > >> > > > > >> > > > > > > > > > how quickly applications migrate to
> > the new
> > > > >> > protocol.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to release
> > this in
> > > > >> > 4.x, but
> > > > >> > > > > it
> > > > >> > > > > >> > > wasn't
> > > > >> > > > > >> > > > > made
> > > > >> > > > > >> > > > > > > > > clear
> > > > >> > > > > >> > > > > > > > > > why. If we added that as a way to
> > estimate
> > > > >> when
> > > > >> > we'd
> > > > >> > > > > >> > > deprecate
> > > > >> > > > > >> > > > > and
> > > > >> > > > > >> > > > > > > remove
> > > > >> > > > > >> > > > > > > > > > the group protocol, I also suggest
> > removing
> > > > >> > this part.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> > explained, I
> > > > >> plan
> > > > >> > to
> > > > >> > > > > >> rewrite
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > group
> > > > >> > > > > >> > > > > > > > > coordinator in Java while we implement
> > the new
> > > > >> > protocol.
> > > > >> > > > > >> This
> > > > >> > > > > >> > > means
> > > > >> > > > > >> > > > > > > > > that the internals will be slightly
> > different
> > > > >> > (e.g.
> > > > >> > > > > >> threading
> > > > >> > > > > >> > > > > model).
> > > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the
> > switch from
> > > > >> > the old
> > > > >> > > > > >> group
> > > > >> > > > > >> > > > > > > > > coordinator to the new group coordinator
> > to a
> > > > >> > major
> > > > >> > > > > >> release.
> > > > >> > > > > >> > > The
> > > > >> > > > > >> > > > > > > > > alternative would be to use a flag to do
> > the
> > > > >> > switch
> > > > >> > > > > >> instead
> > > > >> > > > > >> > of
> > > > >> > > > > >> > > > > relying
> > > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the details of
> > the
> > > > >> > migration
> > > > >> > > > > >> story.
> > > > >> > > > > >> > > It
> > > > >> > > > > >> > > > > sounds
> > > > >> > > > > >> > > > > > > > > like
> > > > >> > > > > >> > > > > > > > > > we're saying we will support online
> > > > >> migrations.
> > > > >> > Is
> > > > >> > > > > that
> > > > >> > > > > >> > > correct?
> > > > >> > > > > >> > > > > We
> > > > >> > > > > >> > > > > > > > > should
> > > > >> > > > > >> > > > > > > > > > explain this in detail. It could also
> > be done
> > > > >> > as a
> > > > >> > > > > >> separate
> > > > >> > > > > >> > > KIP,
> > > > >> > > > > >> > > > > if
> > > > >> > > > > >> > > > > > > it's
> > > > >> > > > > >> > > > > > > > > > easier.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Yes, we will support online migrations
> > for the
> > > > >> > group.
> > > > >> > > > > That
> > > > >> > > > > >> > > means
> > > > >> > > > > >> > > > > that
> > > > >> > > > > >> > > > > > > > > a group using the old protocol will be
> > able to
> > > > >> > switch to
> > > > >> > > > > >> the
> > > > >> > > > > >> > > new
> > > > >> > > > > >> > > > > > > > > protocol.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Let me briefly explain how that will work
> > > > >> though.
> > > > >> > It is
> > > > >> > > > > >> > > basically a
> > > > >> > > > > >> > > > > > > > > four step process:
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or
> > rolled to a
> > > > >> > software
> > > > >> > > > > >> > > supporting
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > new group coordinator. Both the old and
> > the new
> > > > >> > > > > >> coordinator
> > > > >> > > > > >> > > will
> > > > >> > > > > >> > > > > > > > > support the old protocol and rely on the
> > same
> > > > >> > persisted
> > > > >> > > > > >> > > metadata so
> > > > >> > > > > >> > > > > > > > > they can work together. This point is an
> > > > >> offline
> > > > >> > > > > >> migration.
> > > > >> > > > > >> > We
> > > > >> > > > > >> > > > > cannot
> > > > >> > > > > >> > > > > > > > > do this one live because it would require
> > > > >> > shutting down
> > > > >> > > > > >> the
> > > > >> > > > > >> > > current
> > > > >> > > > > >> > > > > > > > > coordinator and starting up the new one
> > and
> > > > >> that
> > > > >> > would
> > > > >> > > > > >> cause
> > > > >> > > > > >> > > > > > > > > unavailabilities.
> > > > >> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP
> > must be
> > > > >> > upgraded
> > > > >> > > > > to
> > > > >> > > > > >> X
> > > > >> > > > > >> > in
> > > > >> > > > > >> > > > > order
> > > > >> > > > > >> > > > > > > > > to enable the new protocol. This cannot
> > be done
> > > > >> > before
> > > > >> > > > > 1)
> > > > >> > > > > >> is
> > > > >> > > > > >> > > > > > > > > terminated because the old coordinator
> > doesn't
> > > > >> > support
> > > > >> > > > > the
> > > > >> > > > > >> > new
> > > > >> > > > > >> > > > > > > > > protocol.
> > > > >> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a
> > version
> > > > >> > > > > supporting
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > online
> > > > >> > > > > >> > > > > > > > > migration (must have KIP-792). If the
> > consumer
> > > > >> is
> > > > >> > > > > already
> > > > >> > > > > >> > > there.
> > > > >> > > > > >> > > > > > > > > Nothing must be done at this point.
> > > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled with the
> > > > >> feature
> > > > >> > flag
> > > > >> > > > > >> turned
> > > > >> > > > > >> > > on.
> > > > >> > > > > >> > > > > The
> > > > >> > > > > >> > > > > > > > > consumer group is automatically
> > converted when
> > > > >> > the first
> > > > >> > > > > >> > > consumer
> > > > >> > > > > >> > > > > > > > > using the new protocol joins the group.
> > While
> > > > >> the
> > > > >> > > > > members
> > > > >> > > > > >> > > using the
> > > > >> > > > > >> > > > > > > > > old protocol are being upgraded, the old
> > > > >> protocol
> > > > >> > is
> > > > >> > > > > >> proxied
> > > > >> > > > > >> > > into
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > new one.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the
> > pattern
> > > > >> > > > > >> subscriptions
> > > > >> > > > > >> > > to
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > server,
> > > > >> > > > > >> > > > > > > > > > but it seems like there could be some
> > tricky
> > > > >> > > > > >> compatibility
> > > > >> > > > > >> > > > > issues.
> > > > >> > > > > >> > > > > > > Will
> > > > >> > > > > >> > > > > > > > > we
> > > > >> > > > > >> > > > > > > > > > have a mechanism for users to detect
> > that
> > > > >> they
> > > > >> > need to
> > > > >> > > > > >> > update
> > > > >> > > > > >> > > > > their
> > > > >> > > > > >> > > > > > > regex
> > > > >> > > > > >> > > > > > > > > > before switching to the new protocol?
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > I think that I am a bit more optimistic
> > than
> > > > >> you
> > > > >> > on this
> > > > >> > > > > >> > > point. I
> > > > >> > > > > >> > > > > > > > > believe that the majority of the cases
> > are
> > > > >> simple
> > > > >> > > > > regexes
> > > > >> > > > > >> > which
> > > > >> > > > > >> > > > > should
> > > > >> > > > > >> > > > > > > > > work with the new engine. The
> > coordinator will
> > > > >> > verify
> > > > >> > > > > the
> > > > >> > > > > >> > regex
> > > > >> > > > > >> > > > > anyway
> > > > >> > > > > >> > > > > > > > > and reject the consumer if the regex is
> > not
> > > > >> valid.
> > > > >> > > > > Coming
> > > > >> > > > > >> > back
> > > > >> > > > > >> > > to
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > migration path, in the worst case, the
> > first
> > > > >> > upgraded
> > > > >> > > > > >> > consumer
> > > > >> > > > > >> > > > > joining
> > > > >> > > > > >> > > > > > > > > the group will be rejected. This should
> > be used
> > > > >> > as the
> > > > >> > > > > >> last
> > > > >> > > > > >> > > > > defence, I
> > > > >> > > > > >> > > > > > > > > would say.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > One way for customers to validate their
> > regex
> > > > >> > before
> > > > >> > > > > >> > upgrading
> > > > >> > > > > >> > > > > their
> > > > >> > > > > >> > > > > > > > > prod would be to test them with another
> > group.
> > > > >> For
> > > > >> > > > > >> instance,
> > > > >> > > > > >> > > that
> > > > >> > > > > >> > > > > > > > > could be done in a pre-prod environment.
> > > > >> Another
> > > > >> > way
> > > > >> > > > > >> would be
> > > > >> > > > > >> > > to
> > > > >> > > > > >> > > > > > > > > extend the consumer-group tool to
> > provide a
> > > > >> regex
> > > > >> > > > > >> validation
> > > > >> > > > > >> > > > > > > > > mechanism. Would this be enough in your
> > > > >> opinion?
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 5. Related to the last question, will
> > the
> > > > >> Java
> > > > >> > client
> > > > >> > > > > >> allow
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > users to
> > > > >> > > > > >> > > > > > > > > > stick with the current regex engine for
> > > > >> > compatibility
> > > > >> > > > > >> > > reasons?
> > > > >> > > > > >> > > > > For
> > > > >> > > > > >> > > > > > > > > example,
> > > > >> > > > > >> > > > > > > > > > it may be handy to keep using client
> > based
> > > > >> > regex at
> > > > >> > > > > >> first
> > > > >> > > > > >> > to
> > > > >> > > > > >> > > keep
> > > > >> > > > > >> > > > > > > > > > migrations simple and then migrate to
> > server
> > > > >> > based
> > > > >> > > > > >> regexes
> > > > >> > > > > >> > > as a
> > > > >> > > > > >> > > > > > > second
> > > > >> > > > > >> > > > > > > > > step.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > I understand your point but I am
> > concerned that
> > > > >> > this
> > > > >> > > > > would
> > > > >> > > > > >> > > allow
> > > > >> > > > > >> > > > > users
> > > > >> > > > > >> > > > > > > > > to actually stay in this mode. That
> > would go
> > > > >> > against our
> > > > >> > > > > >> goal
> > > > >> > > > > >> > > of
> > > > >> > > > > >> > > > > > > > > simplifying the client because we would
> > have to
> > > > >> > continue
> > > > >> > > > > >> > > monitoring
> > > > >> > > > > >> > > > > > > > > the metadata on the client side. I would
> > rather
> > > > >> > not do
> > > > >> > > > > >> this.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> > coordinator
> > > > >> will
> > > > >> > be
> > > > >> > > > > >> > > responsible for
> > > > >> > > > > >> > > > > > > storing
> > > > >> > > > > >> > > > > > > > > > the configurations and that the
> > > > >> configurations
> > > > >> > will be
> > > > >> > > > > >> > > deleted
> > > > >> > > > > >> > > > > when
> > > > >> > > > > >> > > > > > > the
> > > > >> > > > > >> > > > > > > > > > group is deleted. Will a transition to
> > DEAD
> > > > >> > trigger
> > > > >> > > > > >> > deletion
> > > > >> > > > > >> > > of
> > > > >> > > > > >> > > > > > > > > > configurations?
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > That's right. The configurations will be
> > > > >> deleted
> > > > >> > when
> > > > >> > > > > the
> > > > >> > > > > >> > > group is
> > > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 7. Will the choice to store the
> > configs in
> > > > >> the
> > > > >> > group
> > > > >> > > > > >> > > coordinator
> > > > >> > > > > >> > > > > > > make it
> > > > >> > > > > >> > > > > > > > > > harder to list all cluster configs and
> > their
> > > > >> > values?
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > I don't think so. The group
> > configurations are
> > > > >> > overrides
> > > > >> > > > > >> of
> > > > >> > > > > >> > > cluster
> > > > >> > > > > >> > > > > > > > > configs. If you want to know all the
> > overrides
> > > > >> > though,
> > > > >> > > > > you
> > > > >> > > > > >> > > would
> > > > >> > > > > >> > > > > have
> > > > >> > > > > >> > > > > > > > > to ask all the group coordinators. You
> > cannot
> > > > >> > rely on
> > > > >> > > > > the
> > > > >> > > > > >> > > metadata
> > > > >> > > > > >> > > > > log
> > > > >> > > > > >> > > > > > > > > for instance.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 8. How would someone configure a group
> > before
> > > > >> > starting
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > consumers?
> > > > >> > > > > >> > > > > > > > > Have
> > > > >> > > > > >> > > > > > > > > > we considered allowing the explicit
> > creation
> > > > >> of
> > > > >> > > > > groups?
> > > > >> > > > > >> > > > > > > Alternatively,
> > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > >> > > > > > > > > > configs could be decoupled from the
> > group
> > > > >> > lifecycle.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Yes. The group will be automatically
> > created in
> > > > >> > this
> > > > >> > > > > case.
> > > > >> > > > > >> > > However,
> > > > >> > > > > >> > > > > > > > > the configs will be lost after the
> > retention
> > > > >> > period of
> > > > >> > > > > the
> > > > >> > > > > >> > > group
> > > > >> > > > > >> > > > > > > > > passes.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method
> > for the
> > > > >> > Java
> > > > >> > > > > >> client
> > > > >> > > > > >> > > still
> > > > >> > > > > >> > > > > take
> > > > >> > > > > >> > > > > > > a
> > > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we
> > have to
> > > > >> > introduce
> > > > >> > > > > an
> > > > >> > > > > >> > > overload?
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > That's a very group question. I forgot
> > about
> > > > >> that
> > > > >> > one.
> > > > >> > > > > As
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
> > > > >> compatible
> > > > >> > with
> > > > >> > > > > the
> > > > >> > > > > >> > > engine
> > > > >> > > > > >> > > > > that
> > > > >> > > > > >> > > > > > > > > we plan to use, it might be better to
> > deprecate
> > > > >> > it and
> > > > >> > > > > >> use an
> > > > >> > > > > >> > > > > overload
> > > > >> > > > > >> > > > > > > > > which takes a string. We would rely on
> > the
> > > > >> server
> > > > >> > side
> > > > >> > > > > >> > > validation.
> > > > >> > > > > >> > > > > > > > > During the migration, I think that we
> > could
> > > > >> still
> > > > >> > try to
> > > > >> > > > > >> > > toString
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > regex and use it. That should work, I
> > think, in
> > > > >> > the
> > > > >> > > > > >> majority
> > > > >> > > > > >> > > of the
> > > > >> > > > > >> > > > > > > > > cases.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 10. I agree with Justine that we
> > should be
> > > > >> > clearer
> > > > >> > > > > about
> > > > >> > > > > >> > the
> > > > >> > > > > >> > > > > reason
> > > > >> > > > > >> > > > > > > to
> > > > >> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
> > > > >> feature
> > > > >> > flag.
> > > > >> > > > > >> Maybe
> > > > >> > > > > >> > > we
> > > > >> > > > > >> > > > > mean
> > > > >> > > > > >> > > > > > > that
> > > > >> > > > > >> > > > > > > > > we
> > > > >> > > > > >> > > > > > > > > > can switch the default for the feature
> > flag
> > > > >> to
> > > > >> > true
> > > > >> > > > > >> based
> > > > >> > > > > >> > on
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > > > > metadata.version once we want to make
> > it the
> > > > >> > default.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > My plan was to use that feature flag
> > mainly
> > > > >> > during the
> > > > >> > > > > >> > > development
> > > > >> > > > > >> > > > > > > > > phase. I should not have mentioned it, I
> > think,
> > > > >> > because
> > > > >> > > > > we
> > > > >> > > > > >> > > could
> > > > >> > > > > >> > > > > use
> > > > >> > > > > >> > > > > > > > > an internal config for it.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't
> > mention
> > > > >> the
> > > > >> > > > > required
> > > > >> > > > > >> > > ACLs, it
> > > > >> > > > > >> > > > > > > would
> > > > >> > > > > >> > > > > > > > > be
> > > > >> > > > > >> > > > > > > > > > good to add that for consistency.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Noted.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > > >> ConsumerGroupHeartbeat
> > > > >> > > > > requires
> > > > >> > > > > >> > > "Read
> > > > >> > > > > >> > > > > Group"
> > > > >> > > > > >> > > > > > > > > even
> > > > >> > > > > >> > > > > > > > > > though it seems to do more than
> > reading.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > I agree. This is how the current
> > protocol works
> > > > >> > though.
> > > > >> > > > > We
> > > > >> > > > > >> > only
> > > > >> > > > > >> > > > > > > > > require "Read Group" to join a group. We
> > could
> > > > >> > consider
> > > > >> > > > > >> > > changing
> > > > >> > > > > >> > > > > this
> > > > >> > > > > >> > > > > > > > > but I am not sure that it is worth it.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 13. How is topic recreation handled by
> > the
> > > > >> > consumer
> > > > >> > > > > with
> > > > >> > > > > >> > the
> > > > >> > > > > >> > > new
> > > > >> > > > > >> > > > > > > group
> > > > >> > > > > >> > > > > > > > > > protocol? It would be good to have a
> > section
> > > > >> on
> > > > >> > this.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Noted. From a protocol perspective, the
> > new
> > > > >> topic
> > > > >> > will
> > > > >> > > > > >> have a
> > > > >> > > > > >> > > new
> > > > >> > > > > >> > > > > > > > > topic id so it will treat it like a
> > topic with
> > > > >> a
> > > > >> > > > > different
> > > > >> > > > > >> > > name.
> > > > >> > > > > >> > > > > The
> > > > >> > > > > >> > > > > > > > > only issue is that the fetch/commit
> > offsets
> > > > >> APIs
> > > > >> > do not
> > > > >> > > > > >> > support
> > > > >> > > > > >> > > > > topic
> > > > >> > > > > >> > > > > > > > > IDs so the consumer would reuse the
> > offsets
> > > > >> based
> > > > >> > on the
> > > > >> > > > > >> > same.
> > > > >> > > > > >> > > I
> > > > >> > > > > >> > > > > think
> > > > >> > > > > >> > > > > > > > > that we should update those APIs as well
> > in
> > > > >> order
> > > > >> > to be
> > > > >> > > > > >> > > consistent
> > > > >> > > > > >> > > > > end
> > > > >> > > > > >> > > > > > > > > to end. That would strengthen the
> > semantics of
> > > > >> the
> > > > >> > > > > >> consumer.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the
> > new
> > > > >> > coordinator
> > > > >> > > > > >> in
> > > > >> > > > > >> > > Java.
> > > > >> > > > > >> > > > > Even
> > > > >> > > > > >> > > > > > > > > though
> > > > >> > > > > >> > > > > > > > > > this is an implementation detail, do
> > we plan
> > > > >> to
> > > > >> > have a
> > > > >> > > > > >> new
> > > > >> > > > > >> > > gradle
> > > > >> > > > > >> > > > > > > module
> > > > >> > > > > >> > > > > > > > > > for it?
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Yes.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when
> > it
> > > > >> comes
> > > > >> > to how
> > > > >> > > > > >> many
> > > > >> > > > > >> > > > > members
> > > > >> > > > > >> > > > > > > the
> > > > >> > > > > >> > > > > > > > > new
> > > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > We don't have numbers at the moment. The
> > > > >> protocol
> > > > >> > should
> > > > >> > > > > >> > > support
> > > > >> > > > > >> > > > > 1000s
> > > > >> > > > > >> > > > > > > > > of members per group. We will measure
> > this when
> > > > >> > we have
> > > > >> > > > > a
> > > > >> > > > > >> > first
> > > > >> > > > > >> > > > > > > > > implementation. Note that we might have
> > other
> > > > >> > > > > bottlenecks
> > > > >> > > > > >> > down
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> > SubscribedTopidIds
> > > > >> > instead
> > > > >> > > > > of
> > > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > > >> > ConsumerGroupHeartbeatRequest?
> > > > >> > > > > >> Is
> > > > >> > > > > >> > the
> > > > >> > > > > >> > > > > idea
> > > > >> > > > > >> > > > > > > that
> > > > >> > > > > >> > > > > > > > > > since we have to resolve the regex on
> > the
> > > > >> > server, we
> > > > >> > > > > >> can do
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > same
> > > > >> > > > > >> > > > > > > for
> > > > >> > > > > >> > > > > > > > > > the topic name? The difference is that
> > > > >> sending
> > > > >> > the
> > > > >> > > > > >> regex is
> > > > >> > > > > >> > > more
> > > > >> > > > > >> > > > > > > > > efficient
> > > > >> > > > > >> > > > > > > > > > whereas sending the topic names is less
> > > > >> > efficient.
> > > > >> > > > > >> > > Furthermore,
> > > > >> > > > > >> > > > > > > delete
> > > > >> > > > > >> > > > > > > > > and
> > > > >> > > > > >> > > > > > > > > > recreation is easier to handle if we
> > have
> > > > >> topic
> > > > >> > ids.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > The idea was to consolidate the metadata
> > lookup
> > > > >> > on the
> > > > >> > > > > >> server
> > > > >> > > > > >> > > for
> > > > >> > > > > >> > > > > both
> > > > >> > > > > >> > > > > > > > > paths but I do agree with your point. As
> > a
> > > > >> second
> > > > >> > > > > though,
> > > > >> > > > > >> > using
> > > > >> > > > > >> > > > > topic
> > > > >> > > > > >> > > > > > > > > ids may be better here for the delete and
> > > > >> > recreation
> > > > >> > > > > case.
> > > > >> > > > > >> > > Also, I
> > > > >> > > > > >> > > > > > > > > suppose that we may allow users to
> > subscribe
> > > > >> with
> > > > >> > topic
> > > > >> > > > > >> ids
> > > > >> > > > > >> > in
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > > > future because that is the only way to be
> > > > >> really
> > > > >> > robust
> > > > >> > > > > to
> > > > >> > > > > >> > > topic
> > > > >> > > > > >> > > > > > > > > re-creation.
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > Best,
> > > > >> > > > > >> > > > > > > > > David
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David
> > Jacot <
> > > > >> > > > > >> > > djacot@confluent.io>
> > > > >> > > > > >> > > > > > > wrote:
> > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > Thanks for your comments. Please find
> > my
> > > > >> answers
> > > > >> > > > > below.
> > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on
> > topic IDs
> > > > >> > with the
> > > > >> > > > > >> > > exception
> > > > >> > > > > >> > > > > of the
> > > > >> > > > > >> > > > > > > > > > topic names based in the
> > > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > > >> > > > > >> I
> > > > >> > > > > >> > am
> > > > >> > > > > >> > > not
> > > > >> > > > > >> > > > > sure
> > > > >> > > > > >> > > > > > > > > > if using topic names is the right call
> > here.
> > > > >> I
> > > > >> > need to
> > > > >> > > > > >> > think
> > > > >> > > > > >> > > > > about it
> > > > >> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does
> > not
> > > > >> > change the
> > > > >> > > > > >> > > > > fetch/commit
> > > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This
> > may be
> > > > >> > something
> > > > >> > > > > >> that
> > > > >> > > > > >> > we
> > > > >> > > > > >> > > > > should
> > > > >> > > > > >> > > > > > > > > > include though as it would give better
> > > > >> overall
> > > > >> > > > > >> guarantee in
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > > > > producer.
> > > > >> > > > > >> > > > > > > > > > - You're right. I think that I should
> > not
> > > > >> have
> > > > >> > > > > mentioned
> > > > >> > > > > >> > this
> > > > >> > > > > >> > > > > flag at
> > > > >> > > > > >> > > > > > > > > > all. I will remove it. We can use an
> > internal
> > > > >> > > > > >> configuration
> > > > >> > > > > >> > > while
> > > > >> > > > > >> > > > > > > > > > developing the feature.
> > > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> > supported. The
> > > > >> > change is
> > > > >> > > > > >> > > > > orthogonal. The
> > > > >> > > > > >> > > > > > > > > > only requirement is that the cluster
> > uses
> > > > >> topic
> > > > >> > IDs.
> > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > Best,
> > > > >> > > > > >> > > > > > > > > > David
> > > > >> > > > > >> > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM
> > Guozhang
> > > > >> Wang <
> > > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > > >> > > > > >> > > > > > > > > wrote:
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are
> > some
> > > > >> replies
> > > > >> > > > > inlined
> > > > >> > > > > >> > > below:
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM
> > Ismael Juma
> > > > >> <
> > > > >> > > > > >> > > ismael@juma.me.uk>
> > > > >> > > > > >> > > > > > > wrote:
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the
> > > > >> potential
> > > > >> > to be a
> > > > >> > > > > >> > great
> > > > >> > > > > >> > > > > > > > > improvement. A few
> > > > >> > > > > >> > > > > > > > > > > > initial questions/comments:
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk
> > about
> > > > >> > target
> > > > >> > > > > >> versions
> > > > >> > > > > >> > > for
> > > > >> > > > > >> > > > > > > > > deprecation and
> > > > >> > > > > >> > > > > > > > > > > > removal of the existing group
> > protocol.
> > > > >> > Unlike
> > > > >> > > > > >> KRaft,
> > > > >> > > > > >> > > this
> > > > >> > > > > >> > > > > > > affects a
> > > > >> > > > > >> > > > > > > > > core
> > > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > > >> > deprecation/removal will
> > > > >> > > > > >> be
> > > > >> > > > > >> > > heavily
> > > > >> > > > > >> > > > > > > > > dependent on
> > > > >> > > > > >> > > > > > > > > > > > how quickly applications migrate
> > to the
> > > > >> new
> > > > >> > > > > >> protocol.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can
> > > > >> remove
> > > > >> > the
> > > > >> > > > > >> proposed
> > > > >> > > > > >> > > > > timeline
> > > > >> > > > > >> > > > > > > in
> > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and
> > Migration
> > > > >> > Plan` and
> > > > >> > > > > >> > > instead
> > > > >> > > > > >> > > > > just
> > > > >> > > > > >> > > > > > > state
> > > > >> > > > > >> > > > > > > > > > > that we will decide in the future
> > about
> > > > >> when
> > > > >> > we
> > > > >> > > > > would
> > > > >> > > > > >> > > > > deprecate old
> > > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to
> > release this
> > > > >> > in 4.x,
> > > > >> > > > > >> but
> > > > >> > > > > >> > it
> > > > >> > > > > >> > > > > wasn't
> > > > >> > > > > >> > > > > > > made
> > > > >> > > > > >> > > > > > > > > clear
> > > > >> > > > > >> > > > > > > > > > > > why. If we added that as a way to
> > > > >> estimate
> > > > >> > when
> > > > >> > > > > we'd
> > > > >> > > > > >> > > > > deprecate
> > > > >> > > > > >> > > > > > > and
> > > > >> > > > > >> > > > > > > > > remove
> > > > >> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
> > > > >> removing
> > > > >> > this
> > > > >> > > > > >> part.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > I think that's not specifically
> > related to
> > > > >> the
> > > > >> > > > > >> > > > > deprecation/removal
> > > > >> > > > > >> > > > > > > > > timeline
> > > > >> > > > > >> > > > > > > > > > > plan, but it's more for client
> > upgrades.
> > > > >> I.e.
> > > > >> > the
> > > > >> > > > > >> > > broker-side
> > > > >> > > > > >> > > > > > > > > > > implementation may be done first,
> > and then
> > > > >> the
> > > > >> > > > > client
> > > > >> > > > > >> > side,
> > > > >> > > > > >> > > > > and we
> > > > >> > > > > >> > > > > > > > > would
> > > > >> > > > > >> > > > > > > > > > > only mark it as "released" by the
> > time
> > > > >> clients
> > > > >> > > > > >> > > implementations
> > > > >> > > > > >> > > > > are
> > > > >> > > > > >> > > > > > > > > done. At
> > > > >> > > > > >> > > > > > > > > > > that time, to enable the feature the
> > > > >> clients
> > > > >> > need to
> > > > >> > > > > >> > first
> > > > >> > > > > >> > > > > swap-in
> > > > >> > > > > >> > > > > > > the
> > > > >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and
> > then set
> > > > >> > the flag
> > > > >> > > > > >> > with a
> > > > >> > > > > >> > > > > second
> > > > >> > > > > >> > > > > > > > > rolling
> > > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's
> > better to be
> > > > >> > released
> > > > >> > > > > >> in a
> > > > >> > > > > >> > > major
> > > > >> > > > > >> > > > > > > > > version.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the
> > details of
> > > > >> the
> > > > >> > > > > migration
> > > > >> > > > > >> > > story.
> > > > >> > > > > >> > > > > It
> > > > >> > > > > >> > > > > > > > > sounds like
> > > > >> > > > > >> > > > > > > > > > > > we're saying we will support online
> > > > >> > migrations. Is
> > > > >> > > > > >> that
> > > > >> > > > > >> > > > > correct?
> > > > >> > > > > >> > > > > > > We
> > > > >> > > > > >> > > > > > > > > should
> > > > >> > > > > >> > > > > > > > > > > > explain this in detail. It could
> > also be
> > > > >> > done as a
> > > > >> > > > > >> > > separate
> > > > >> > > > > >> > > > > KIP,
> > > > >> > > > > >> > > > > > > if
> > > > >> > > > > >> > > > > > > > > it's
> > > > >> > > > > >> > > > > > > > > > > > easier.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Yes I think that's the part we can
> > be more
> > > > >> > concrete
> > > > >> > > > > >> about
> > > > >> > > > > >> > > for
> > > > >> > > > > >> > > > > sure
> > > > >> > > > > >> > > > > > > (and
> > > > >> > > > > >> > > > > > > > > > > this is related to your question 2)
> > above).
> > > > >> > We will
> > > > >> > > > > >> work
> > > > >> > > > > >> > on
> > > > >> > > > > >> > > > > making
> > > > >> > > > > >> > > > > > > it
> > > > >> > > > > >> > > > > > > > > more
> > > > >> > > > > >> > > > > > > > > > > explicit in parallel as we solicit
> > more
> > > > >> > feedback.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing
> > the
> > > > >> > pattern
> > > > >> > > > > >> > > subscriptions
> > > > >> > > > > >> > > > > to
> > > > >> > > > > >> > > > > > > the
> > > > >> > > > > >> > > > > > > > > server,
> > > > >> > > > > >> > > > > > > > > > > > but it seems like there could be
> > some
> > > > >> tricky
> > > > >> > > > > >> > > compatibility
> > > > >> > > > > >> > > > > > > issues.
> > > > >> > > > > >> > > > > > > > > Will we
> > > > >> > > > > >> > > > > > > > > > > > have a mechanism for users to
> > detect that
> > > > >> > they
> > > > >> > > > > need
> > > > >> > > > > >> to
> > > > >> > > > > >> > > update
> > > > >> > > > > >> > > > > > > their
> > > > >> > > > > >> > > > > > > > > regex
> > > > >> > > > > >> > > > > > > > > > > > before switching to the new
> > protocol?
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
> > > > >> non-java
> > > > >> > client
> > > > >> > > > > >> > users
> > > > >> > > > > >> > > to
> > > > >> > > > > >> > > > > sort
> > > > >> > > > > >> > > > > > > of
> > > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> > switching to
> > > > >> the
> > > > >> > new
> > > > >> > > > > >> > protocol.
> > > > >> > > > > >> > > I
> > > > >> > > > > >> > > > > do not
> > > > >> > > > > >> > > > > > > > > have a
> > > > >> > > > > >> > > > > > > > > > > specific idea on top of my head
> > though,
> > > > >> maybe
> > > > >> > others
> > > > >> > > > > >> like
> > > > >> > > > > >> > > @Matt
> > > > >> > > > > >> > > > > > > > > Howlett can
> > > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 5. Related to the last question,
> > will the
> > > > >> > Java
> > > > >> > > > > >> client
> > > > >> > > > > >> > > allow
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > users to
> > > > >> > > > > >> > > > > > > > > > > > stick with the current regex
> > engine for
> > > > >> > > > > >> compatibility
> > > > >> > > > > >> > > > > reasons?
> > > > >> > > > > >> > > > > > > For
> > > > >> > > > > >> > > > > > > > > example,
> > > > >> > > > > >> > > > > > > > > > > > it may be handy to keep using
> > client
> > > > >> based
> > > > >> > regex
> > > > >> > > > > at
> > > > >> > > > > >> > > first to
> > > > >> > > > > >> > > > > keep
> > > > >> > > > > >> > > > > > > > > > > > migrations simple and then migrate
> > to
> > > > >> > server based
> > > > >> > > > > >> > > regexes
> > > > >> > > > > >> > > > > as a
> > > > >> > > > > >> > > > > > > > > second
> > > > >> > > > > >> > > > > > > > > > > > step.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Honestly I have not thought about
> > that for
> > > > >> > java
> > > > >> > > > > >> clients,
> > > > >> > > > > >> > > and
> > > > >> > > > > >> > > > > we can
> > > > >> > > > > >> > > > > > > > > discuss
> > > > >> > > > > >> > > > > > > > > > > that. What kind of compatibility
> > issues do
> > > > >> > you have
> > > > >> > > > > in
> > > > >> > > > > >> > > mind?
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 6. When we say that the group
> > coordinator
> > > > >> > will be
> > > > >> > > > > >> > > > > responsible for
> > > > >> > > > > >> > > > > > > > > storing
> > > > >> > > > > >> > > > > > > > > > > > the configurations and that the
> > > > >> > configurations
> > > > >> > > > > will
> > > > >> > > > > >> be
> > > > >> > > > > >> > > > > deleted
> > > > >> > > > > >> > > > > > > when
> > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> > transition to
> > > > >> DEAD
> > > > >> > > > > trigger
> > > > >> > > > > >> > > deletion
> > > > >> > > > > >> > > > > of
> > > > >> > > > > >> > > > > > > > > > > > configurations?
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an
> > ending
> > > > >> state
> > > > >> > (we
> > > > >> > > > > would
> > > > >> > > > > >> > only
> > > > >> > > > > >> > > > > > > transit to
> > > > >> > > > > >> > > > > > > > > that
> > > > >> > > > > >> > > > > > > > > > > state when the group is EMPTY and
> > also all
> > > > >> of
> > > > >> > its
> > > > >> > > > > >> > metadata
> > > > >> > > > > >> > > are
> > > > >> > > > > >> > > > > > > gone),
> > > > >> > > > > >> > > > > > > > > once
> > > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this group
> > would
> > > > >> never
> > > > >> > be
> > > > >> > > > > >> revived.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to store the
> > configs
> > > > >> in
> > > > >> > the
> > > > >> > > > > group
> > > > >> > > > > >> > > > > coordinator
> > > > >> > > > > >> > > > > > > > > make it
> > > > >> > > > > >> > > > > > > > > > > > harder to list all cluster configs
> > and
> > > > >> their
> > > > >> > > > > values?
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > That's a good question, and our
> > thoughts
> > > > >> are
> > > > >> > that
> > > > >> > > > > the
> > > > >> > > > > >> > > so-called
> > > > >> > > > > >> > > > > > > "group
> > > > >> > > > > >> > > > > > > > > > > configurations" are overrides of the
> > > > >> > cluster-level
> > > > >> > > > > >> > > > > configurations
> > > > >> > > > > >> > > > > > > > > > > customized per group so when an
> > admin list
> > > > >> > cluster
> > > > >> > > > > >> > configs
> > > > >> > > > > >> > > it's
> > > > >> > > > > >> > > > > > > okay to
> > > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> > "defaults", not
> > > > >> > showing
> > > > >> > > > > >> any
> > > > >> > > > > >> > > > > per-group
> > > > >> > > > > >> > > > > > > > > > > customizations.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 8. How would someone configure a
> > group
> > > > >> > before
> > > > >> > > > > >> starting
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > > > consumers? Have
> > > > >> > > > > >> > > > > > > > > > > > we considered allowing the explicit
> > > > >> > creation of
> > > > >> > > > > >> groups?
> > > > >> > > > > >> > > > > > > > > Alternatively, the
> > > > >> > > > > >> > > > > > > > > > > > configs could be decoupled from
> > the group
> > > > >> > > > > lifecycle.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > The configs can be created before
> > the group
> > > > >> > itself
> > > > >> > > > > as
> > > > >> > > > > >> an
> > > > >> > > > > >> > > > > > > independent
> > > > >> > > > > >> > > > > > > > > entity
> > > > >> > > > > >> > > > > > > > > > > --- of course, this requires the
> > > > >> corresponding
> > > > >> > > > > >> request to
> > > > >> > > > > >> > > be
> > > > >> > > > > >> > > > > > > routed to
> > > > >> > > > > >> > > > > > > > > the
> > > > >> > > > > >> > > > > > > > > > > right coordinator based on the group
> > id ---
> > > > >> > the only
> > > > >> > > > > >> > thing
> > > > >> > > > > >> > > that
> > > > >> > > > > >> > > > > > > > > differs is,
> > > > >> > > > > >> > > > > > > > > > > when the group itself is gone we
> > also check
> > > > >> > if there
> > > > >> > > > > >> are
> > > > >> > > > > >> > > any
> > > > >> > > > > >> > > > > > > > > configuration
> > > > >> > > > > >> > > > > > > > > > > entities related to that group and
> > delete
> > > > >> as
> > > > >> > well.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> > > > >> > asymmetry on
> > > > >> > > > > the
> > > > >> > > > > >> > > creation
> > > > >> > > > > >> > > > > /
> > > > >> > > > > >> > > > > > > > > deletion
> > > > >> > > > > >> > > > > > > > > > > lifecycles of the config entities,
> > and we
> > > > >> > would like
> > > > >> > > > > >> to
> > > > >> > > > > >> > > hear
> > > > >> > > > > >> > > > > > > everyone's
> > > > >> > > > > >> > > > > > > > > > > feelings whether we should aim for
> > symmetry
> > > > >> > i.e.
> > > > >> > > > > >> totally
> > > > >> > > > > >> > > > > decouple
> > > > >> > > > > >> > > > > > > group
> > > > >> > > > > >> > > > > > > > > > > configs and hence not delete them at
> > all
> > > > >> when
> > > > >> > the
> > > > >> > > > > >> group
> > > > >> > > > > >> > is
> > > > >> > > > > >> > > > > gone,
> > > > >> > > > > >> > > > > > > but
> > > > >> > > > > >> > > > > > > > > always
> > > > >> > > > > >> > > > > > > > > > > require explicit deletion operations
> > by
> > > > >> > themselves.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe
> > method for
> > > > >> > the Java
> > > > >> > > > > >> > client
> > > > >> > > > > >> > > > > still
> > > > >> > > > > >> > > > > > > take
> > > > >> > > > > >> > > > > > > > > a
> > > > >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we
> > have
> > > > >> to
> > > > >> > > > > >> introduce an
> > > > >> > > > > >> > > > > overload?
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > I think we do not need to introduce
> > an
> > > > >> > overload, but
> > > > >> > > > > >> I'm
> > > > >> > > > > >> > > all
> > > > >> > > > > >> > > > > ears
> > > > >> > > > > >> > > > > > > if
> > > > >> > > > > >> > > > > > > > > there
> > > > >> > > > > >> > > > > > > > > > > may be some compatibility issues
> > that we
> > > > >> may
> > > > >> > > > > overlook.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we
> > should
> > > > >> be
> > > > >> > clearer
> > > > >> > > > > >> > about
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > > > reason
> > > > >> > > > > >> > > > > > > > > to
> > > > >> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version
> > from the
> > > > >> > feature
> > > > >> > > > > >> flag.
> > > > >> > > > > >> > > Maybe
> > > > >> > > > > >> > > > > we
> > > > >> > > > > >> > > > > > > mean
> > > > >> > > > > >> > > > > > > > > that we
> > > > >> > > > > >> > > > > > > > > > > > can switch the default for the
> > feature
> > > > >> flag
> > > > >> > to
> > > > >> > > > > true
> > > > >> > > > > >> > > based on
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > > > > metadata.version once we want to
> > make it
> > > > >> the
> > > > >> > > > > >> default.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't
> > mention
> > > > >> > the
> > > > >> > > > > >> required
> > > > >> > > > > >> > > ACLs,
> > > > >> > > > > >> > > > > it
> > > > >> > > > > >> > > > > > > > > would be
> > > > >> > > > > >> > > > > > > > > > > > good to add that for consistency.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > > >> > ConsumerGroupHeartbeat
> > > > >> > > > > >> > requires
> > > > >> > > > > >> > > > > "Read
> > > > >> > > > > >> > > > > > > > > Group" even
> > > > >> > > > > >> > > > > > > > > > > > though it seems to do more than
> > reading.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > I had that thought myself as well,
> > but in
> > > > >> the
> > > > >> > end we
> > > > >> > > > > >> > could
> > > > >> > > > > >> > > not
> > > > >> > > > > >> > > > > > > find a
> > > > >> > > > > >> > > > > > > > > > > better alternative: adding Write
> > Group
> > > > >> seems
> > > > >> > an
> > > > >> > > > > >> overkill
> > > > >> > > > > >> > > here
> > > > >> > > > > >> > > > > > > since we
> > > > >> > > > > >> > > > > > > > > do
> > > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we only have
> > Read /
> > > > >> > Delete
> > > > >> > > > > and
> > > > >> > > > > >> > > Describe
> > > > >> > > > > >> > > > > on
> > > > >> > > > > >> > > > > > > > > groups so
> > > > >> > > > > >> > > > > > > > > > > far). Would like to hear others
> > thoughts.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 13. How is topic recreation
> > handled by
> > > > >> the
> > > > >> > > > > consumer
> > > > >> > > > > >> > with
> > > > >> > > > > >> > > the
> > > > >> > > > > >> > > > > new
> > > > >> > > > > >> > > > > > > > > group
> > > > >> > > > > >> > > > > > > > > > > > protocol? It would be good to have
> > a
> > > > >> > section on
> > > > >> > > > > >> this.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > You mean with regex subscription
> > right? Yes
> > > > >> > we can
> > > > >> > > > > >> add a
> > > > >> > > > > >> > > > > section
> > > > >> > > > > >> > > > > > > about
> > > > >> > > > > >> > > > > > > > > > > that, but basically the idea is that
> > > > >> consumer
> > > > >> > would
> > > > >> > > > > be
> > > > >> > > > > >> > > totally
> > > > >> > > > > >> > > > > > > > > agnostic in
> > > > >> > > > > >> > > > > > > > > > > the new protocol as it's handled all
> > by the
> > > > >> > brokers.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write
> > the
> > > > >> new
> > > > >> > > > > >> coordinator
> > > > >> > > > > >> > in
> > > > >> > > > > >> > > > > Java.
> > > > >> > > > > >> > > > > > > Even
> > > > >> > > > > >> > > > > > > > > though
> > > > >> > > > > >> > > > > > > > > > > > this is an implementation detail,
> > do we
> > > > >> > plan to
> > > > >> > > > > >> have a
> > > > >> > > > > >> > > new
> > > > >> > > > > >> > > > > gradle
> > > > >> > > > > >> > > > > > > > > module
> > > > >> > > > > >> > > > > > > > > > > > for it?
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > We have not thought about that. But
> > I think
> > > > >> > the
> > > > >> > > > > answer
> > > > >> > > > > >> > > should
> > > > >> > > > > >> > > > > be
> > > > >> > > > > >> > > > > > > yes.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal
> > when it
> > > > >> > comes to
> > > > >> > > > > >> how
> > > > >> > > > > >> > > many
> > > > >> > > > > >> > > > > > > members
> > > > >> > > > > >> > > > > > > > > the new
> > > > >> > > > > >> > > > > > > > > > > > group protocol can support?
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > Within a group, I think we should
> > shoot for
> > > > >> > 1000s of
> > > > >> > > > > >> > > members.
> > > > >> > > > > >> > > > > But
> > > > >> > > > > >> > > > > > > that
> > > > >> > > > > >> > > > > > > > > > > scalability goals also depend on the
> > offset
> > > > >> > > > > management
> > > > >> > > > > >> > > (commit,
> > > > >> > > > > >> > > > > > > fetch)
> > > > >> > > > > >> > > > > > > > > > > capabilities of the coordinator
> > which we
> > > > >> did
> > > > >> > not
> > > > >> > > > > >> cover in
> > > > >> > > > > >> > > this
> > > > >> > > > > >> > > > > > > KIP, so
> > > > >> > > > > >> > > > > > > > > it's
> > > > >> > > > > >> > > > > > > > > > > hard to give a number that applies
> > > > >> > universally.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> > > > >> > SubscribedTopidIds
> > > > >> > > > > >> instead
> > > > >> > > > > >> > of
> > > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > >> > > > > >> > > Is the
> > > > >> > > > > >> > > > > > > idea
> > > > >> > > > > >> > > > > > > > > that
> > > > >> > > > > >> > > > > > > > > > > > since we have to resolve the regex
> > on the
> > > > >> > server,
> > > > >> > > > > we
> > > > >> > > > > >> > can
> > > > >> > > > > >> > > do
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > same
> > > > >> > > > > >> > > > > > > > > for
> > > > >> > > > > >> > > > > > > > > > > > the topic name? The difference is
> > that
> > > > >> > sending the
> > > > >> > > > > >> > regex
> > > > >> > > > > >> > > is
> > > > >> > > > > >> > > > > more
> > > > >> > > > > >> > > > > > > > > efficient
> > > > >> > > > > >> > > > > > > > > > > > whereas sending the topic names is
> > less
> > > > >> > efficient.
> > > > >> > > > > >> > > > > Furthermore,
> > > > >> > > > > >> > > > > > > > > delete and
> > > > >> > > > > >> > > > > > > > > > > > recreation is easier to handle if
> > we have
> > > > >> > topic
> > > > >> > > > > ids.
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > The main reason to still let the
> > clients
> > > > >> send
> > > > >> > names
> > > > >> > > > > >> is to
> > > > >> > > > > >> > > keep
> > > > >> > > > > >> > > > > the
> > > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids on the
> > broker /
> > > > >> > admin
> > > > >> > > > > client
> > > > >> > > > > >> > > only.
> > > > >> > > > > >> > > > > Note
> > > > >> > > > > >> > > > > > > that
> > > > >> > > > > >> > > > > > > > > > > although we added topic id in
> > KIP-516, we
> > > > >> > never
> > > > >> > > > > >> > > implemented the
> > > > >> > > > > >> > > > > > > logic
> > > > >> > > > > >> > > > > > > > > on
> > > > >> > > > > >> > > > > > > > > > > consumer/producers leveraging the
> > related
> > > > >> > newer
> > > > >> > > > > >> versioned
> > > > >> > > > > >> > > RPCs,
> > > > >> > > > > >> > > > > > > > > instead we
> > > > >> > > > > >> > > > > > > > > > > just set the topic id as empty UUID.
> > We
> > > > >> want
> > > > >> > to keep
> > > > >> > > > > >> the
> > > > >> > > > > >> > > > > > > > > consumer/producer
> > > > >> > > > > >> > > > > > > > > > > to be thin and only delegate the
> > reasoning
> > > > >> on
> > > > >> > broker
> > > > >> > > > > >> and
> > > > >> > > > > >> > > > > > > potentially
> > > > >> > > > > >> > > > > > > > > admin
> > > > >> > > > > >> > > > > > > > > > > clients.
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > > >> > > > > >> > > > > > > > > > > > Ismael
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM
> > David
> > > > >> Jacot
> > > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > >> > > > > > > > > > > > wrote:
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> > discussion
> > > > >> thread
> > > > >> > on
> > > > >> > > > > >> KIP-848:
> > > > >> > > > > >> > > The
> > > > >> > > > > >> > > > > Next
> > > > >> > > > > >> > > > > > > > > > > > > Generation of the Consumer
> > Rebalance
> > > > >> > Protocol.
> > > > >> > > > > >> With
> > > > >> > > > > >> > > this
> > > > >> > > > > >> > > > > KIP,
> > > > >> > > > > >> > > > > > > we
> > > > >> > > > > >> > > > > > > > > aim
> > > > >> > > > > >> > > > > > > > > > > > > to make the rebalance protocol
> > (for
> > > > >> > consumers)
> > > > >> > > > > >> more
> > > > >> > > > > >> > > > > reliable,
> > > > >> > > > > >> > > > > > > more
> > > > >> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
> > > > >> > clients, and
> > > > >> > > > > >> easier
> > > > >> > > > > >> > > to
> > > > >> > > > > >> > > > > debug
> > > > >> > > > > >> > > > > > > for
> > > > >> > > > > >> > > > > > > > > > > > > operators.
> > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > >> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > > Please take a look and let me
> > know what
> > > > >> > you
> > > > >> > > > > think.
> > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > > Best,
> > > > >> > > > > >> > > > > > > > > > > > > David
> > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from July
> > 18th to
> > > > >> > August 8th.
> > > > >> > > > > >> That
> > > > >> > > > > >> > > gives
> > > > >> > > > > >> > > > > > > you a
> > > > >> > > > > >> > > > > > > > > bit
> > > > >> > > > > >> > > > > > > > > > > > > of time to read and digest this
> > long
> > > > >> KIP.
> > > > >> > > > > >> > > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > >
> > > > >> > > > > >> > > > > > > > > > > --
> > > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > > >> > > > > >> > > > > > > > >
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > >
> > > > >> > > > > >> > > > > > > > --
> > > > >> > > > > >> > > > > > > > -- Guozhang
> > > > >> > > > > >> > > > > > >
> > > > >> > > > > >> > > > >
> > > > >> > > > > >> > >
> > > > >> > > > > >> > >
> > > > >> > > > > >> >
> > > > >> > > > > >>
> > > > >> > > > > >>
> > > > >> > > > > >> --
> > > > >> > > > > >> -- Guozhang
> > > > >> > > > > >>
> > > > >> > > > > >
> > > > >> > > > >
> > > > >> >
> > > > >>
> > > > >
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Jun Rao <ju...@confluent.io.INVALID>]

Hi, David,

Thanks for the KIP. Overall, the main benefits of the KIP seem to be fewer
RPCs during rebalance and more efficient support of wildcard. A few
comments below.

30. ConsumerGroupHeartbeatRequest
30.1 ServerAssignor is a singleton. Do we plan to support rolling changing
of the partition assignor in the consumers?
30.2 For each field, could you explain whether it's required in every
request or the scenarios when it needs to be filled? For example, it's not
clear to me when TopicPartitions needs to be filled.

31. In the current consumer protocol, the rack affinity between the client
and the broker is only considered during fetching, but not during assigning
partitions to consumers. Sometimes, once the assignment is made, there is
no opportunity for read affinity because no replicas of assigned partitions
are close to the member. I am wondering if we should use this opportunity
to address this by including rack in GroupMember.

32. On the metric side, often, it's useful to know how busy a group
coordinator is. By moving the event loop model, it seems that we could add
a metric that tracks the fraction of the time the event loop is doing the
actual work.

33. Could we add a section on coordinator failover handling? For example,
does it need to trigger the check if any group with the wildcard
subscription now has a new matching topic?

34. ConsumerGroupMetadataValue, ConsumerGroupPartitionMetadataValue,
ConsumerGroupMemberMetadataValue: Could we document what the epoch field
reflects? For example, does the epoch in ConsumerGroupMetadataValue reflect
the latest group epoch? What about the one in
ConsumerGroupPartitionMetadataValue and ConsumerGroupMemberMetadataValue?

35. "the group coordinator will ensure that the following invariants are
met: ... All members exists." It's possible for a member not to get any
assigned partitions, right?

36. "He can rejoins the group with a member epoch equals to 0": When would
a consumer rejoin and what member id would be used?

37. "Instead, power users will have the ability to trigger a reassignment
by either providing a non-zero reason or by updating the assignor
metadata." Hmm, this seems to be conflicting with the deprecation of
Consumer#enforeRebalance.

38. The reassignment examples are nice. But the section seems to have
multiple typos.
38.1 When the group transitions to epoch 2, B immediately gets into
"epoch=1, partitions=[foo-2]", which seems incorrect.
38.2 When the group transitions to epoch 3, C seems to get into epoch=3,
partitions=[foo-1] too early.
38.3 After A transitions to epoch 3, C still has A - epoch=2,
partitions=[foo-0].

39. Rolling upgrade of consumers: Do we support the upgrade from any old
version to new one?

Thanks,

Jun

On Mon, Aug 29, 2022 at 9:20 AM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi all,
>
> The KIP states that we will re-implement the coordinator in Java. I
> discussed this offline with a few folks and folks are concerned that
> we could introduce many regressions in the old protocol if we do so.
> Therefore, I am going to remove this statement from the KIP. It is an
> implementation detail after all so it does not have to be decided at
> this stage. We will likely start by trying to refactor the current
> implementation as a first step.
>
> Cheers,
> David
>
> On Mon, Aug 29, 2022 at 3:52 PM David Jacot <dj...@confluent.io> wrote:
> >
> > Hi Luke,
> >
> > > 1.1. I think the state machine are: "Empty, assigning, reconciling,
> stable,
> > > dead" mentioned in Consumer Group States section, right?
> >
> > This sentence does not refer to those group states but rather to a
> > state machine replication (SMR). This refers to the entire state of
> > group coordinator which is replicated via the log layer. I will
> > clarify this in the KIP.
> >
> > > 1.2. What do you mean "each state machine is modelled as an event
> loop"?
> >
> > The idea is to follow a model similar to the new quorum controller. We
> > will have N threads to process events. Each __consumer_offsets
> > partition is assigned to a unique thread and all the events (e.g.
> > requests, callbacks, etc.) are processed by this thread. This simplify
> > concurrency and will enable us to do simulation testing for the group
> > coordinator.
> >
> > > 1.3. Why do we need a state machine per *__consumer_offsets*
> partitions?
> > > Not a state machine "per consumer group" owned by a group coordinator?
> For
> > > example, if one group coordinator owns 2 consumer groups, and both
> exist in
> > > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> >
> > See 1.1. The confusion comes from there, I think.
> >
> > > 1.4. I know the "*group.coordinator.threads" *is the number of threads
> used
> > > to run the state machines. But I'm wondering if the purpose of the
> threads
> > > is only to keep the state of each consumer group (or
> *__consumer_offsets*
> > > partitions?), and no heavy computation, why should we need
> multi-threads
> > > here?
> >
> > See 1.2. The idea is to have an ability to shard the processing as the
> > computation could be heavy.
> >
> > > 2.1. The consumer session timeout, why does the default session
> timeout not
> > > locate between min (45s) and max(60s)? I thought the min/max session
> > > timeout is to define lower/upper bound of it, no?
> > >
> > > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > > failures when using the consumer group protocol.
> > > group.consumer.min.session.timeout.ms int 45s The minimum session
> timeout.
> > > group.consumer.max.session.timeout.ms int 60s The maximum session
> timeout.
> >
> > This is indeed a mistake. The default session timeout should be 45s
> > (the current default).
> >
> > > 2.2. The default server side assignor are [range, uniform], which means
> > > we'll default to "range" assignor. I'd like to know why not uniform
> one? I
> > > thought usually users will choose uniform assignor (former sticky
> assinor)
> > > for better evenly distribution. Any other reason we choose range
> assignor
> > > as default?
> > > group.consumer.assignors List range, uniform The server side assignors.
> >
> > The order on the server side has no influence because the client must
> > chose the selector that he wants to use. There is no default in the
> > current proposal. If the assignor is not specified by the client, the
> > request is rejected. The default client value for
> > `group.remote.assignor` is `uniform` though.
> >
> > Thanks for your very good comments, Luke. I hope that my answers help
> > to clarify things. I will update the KIP as well based on your
> > feedback.
> >
> > Cheers,
> > David
> >
> > On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com> wrote:
> > >
> > > Hi David,
> > >
> > > Thanks for the update.
> > >
> > > Some more questions:
> > > 1. In Group Coordinator section, you mentioned:
> > > > The new group coordinator will have a state machine per
> > > *__consumer_offsets* partitions, where each state machine is modelled
> as an
> > > event loop. Those state machines will be executed in
> > > *group.coordinator.threads* threads.
> > >
> > > 1.1. I think the state machine are: "Empty, assigning, reconciling,
> stable,
> > > dead" mentioned in Consumer Group States section, right?
> > > 1.2. What do you mean "each state machine is modelled as an event
> loop"?
> > > 1.3. Why do we need a state machine per *__consumer_offsets*
> partitions?
> > > Not a state machine "per consumer group" owned by a group coordinator?
> For
> > > example, if one group coordinator owns 2 consumer groups, and both
> exist in
> > > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> > > 1.4. I know the "*group.coordinator.threads" *is the number of threads
> used
> > > to run the state machines. But I'm wondering if the purpose of the
> threads
> > > is only to keep the state of each consumer group (or
> *__consumer_offsets*
> > > partitions?), and no heavy computation, why should we need
> multi-threads
> > > here?
> > >
> > > 2. For the default value in the new configs:
> > > 2.1. The consumer session timeout, why does the default session
> timeout not
> > > locate between min (45s) and max(60s)? I thought the min/max session
> > > timeout is to define lower/upper bound of it, no?
> > >
> > > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > > failures when using the consumer group protocol.
> > > group.consumer.min.session.timeout.ms int 45s The minimum session
> timeout.
> > > group.consumer.max.session.timeout.ms int 60s The maximum session
> timeout.
> > >
> > >
> > >
> > > 2.2. The default server side assignor are [range, uniform], which means
> > > we'll default to "range" assignor. I'd like to know why not uniform
> one? I
> > > thought usually users will choose uniform assignor (former sticky
> assinor)
> > > for better evenly distribution. Any other reason we choose range
> assignor
> > > as default?
> > > group.consumer.assignors List range, uniform The server side assignors.
> > >
> > >
> > >
> > >
> > >
> > >
> > > Thank you.
> > > Luke
> > >
> > >
> > >
> > >
> > >
> > >
> > > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com> wrote:
> > >
> > > > Hi Sagar,
> > > >
> > > > I have some thoughts about Kafka Connect integrating with KIP-848,
> but I
> > > > think we should have a separate discussion thread for the Kafka
> Connect
> > > > KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol
> [1],
> > > > and let this discussion thread focus on consumer rebalance protocol,
> WDYT?
> > > >
> > > > [1]
> > > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >
> > > > Thank you.
> > > > Luke
> > > >
> > > >
> > > >
> > > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com>
> wrote:
> > > >
> > > >> Thank you Guozhang/David for the feedback. Looks like there's
> agreement on
> > > >> using separate APIs for Connect. I would revisit the doc and see
> what
> > > >> changes are to be made.
> > > >>
> > > >> Thanks!
> > > >> Sagar.
> > > >>
> > > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot
> <dj...@confluent.io.invalid>
> > > >> wrote:
> > > >>
> > > >> > Hi Sagar,
> > > >> >
> > > >> > Thanks for the feedback and the document. That's really helpful. I
> > > >> > will take a look at it.
> > > >> >
> > > >> > Overall, it seems to me that both Connect and the Consumer could
> share
> > > >> > the same underlying "engine". The main difference is that the
> Consumer
> > > >> > assigns topic-partitions to members whereas Connect assigns tasks
> to
> > > >> > workers. I see two ways to move forward:
> > > >> > 1) We extend the new proposed APIs to support different resource
> types
> > > >> > (e.g. partitions, tasks, etc.); or
> > > >> > 2) We use new dedicated APIs for Connect. The dedicated APIs
> would be
> > > >> > similar to the new ones but different on the content/resources and
> > > >> > they would rely on the same engine on the coordinator side.
> > > >> >
> > > >> > I personally lean towards 2) because I am not a fan of
> overcharging
> > > >> > APIs to serve different purposes. That being said, I am not
> opposed to
> > > >> > 1) if we can find an elegant way to do it.
> > > >> >
> > > >> > I think that we can continue to discuss it here for now in order
> to
> > > >> > ensure that this KIP is compatible with what we will do for
> Connect in
> > > >> > the future.
> > > >> >
> > > >> > Best,
> > > >> > David
> > > >> >
> > > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io>
> wrote:
> > > >> > >
> > > >> > > Hi all,
> > > >> > >
> > > >> > > I am back from vacation. I will go through and address your
> comments
> > > >> > > in the coming days. Thanks for your feedback.
> > > >> > >
> > > >> > > Cheers,
> > > >> > > David
> > > >> > >
> > > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> gharris1727@gmail.com
> > > >> >
> > > >> > wrote:
> > > >> > > >
> > > >> > > > Hey All!
> > > >> > > >
> > > >> > > > Thanks for the KIP, it's wonderful to see cooperative
> rebalancing
> > > >> > making it
> > > >> > > > down the stack!
> > > >> > > >
> > > >> > > > I had a few questions:
> > > >> > > >
> > > >> > > > 1. The 'Rejected Alternatives' section describes how member
> epoch
> > > >> > should
> > > >> > > > advance in step with the group epoch and assignment epoch
> values. I
> > > >> > think
> > > >> > > > that this is a good idea for the reasons described in the
> KIP. When
> > > >> the
> > > >> > > > protocol is incrementally assigning partitions to a worker,
> what
> > > >> member
> > > >> > > > epoch does each incremental assignment use? Are member epochs
> > > >> re-used,
> > > >> > and
> > > >> > > > a single member epoch can correspond to multiple different
> > > >> > (monotonically
> > > >> > > > larger) assignments?
> > > >> > > >
> > > >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> > > >> coordinator? If
> > > >> > > > not, should custom client-side assignor implementations
> interact
> > > >> with
> > > >> > the
> > > >> > > > Reason field, and how is its common meaning agreed upon? If
> so, what
> > > >> > is the
> > > >> > > > benefit of a distinct Reason field over including such
> functionality
> > > >> > in the
> > > >> > > > opaque metadata?
> > > >> > > >
> > > >> > > > 3. The following is included in the KIP: "Thanks to this, the
> input
> > > >> of
> > > >> > the
> > > >> > > > client side assignor is entirely driven by the group
> coordinator.
> > > >> The
> > > >> > > > consumer is no longer responsible for maintaining any state
> besides
> > > >> its
> > > >> > > > assigned partitions." Does this mean that the client-side
> assignor
> > > >> MAY
> > > >> > > > incorporate additional non-Metadata state (such as partition
> > > >> > throughput,
> > > >> > > > cpu/memory metrics, config topics, etc), or that additional
> > > >> > non-Metadata
> > > >> > > > state SHOULD NOT be used?
> > > >> > > >
> > > >> > > > 4. I see that there are separate classes
> > > >> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > > >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that
> seem to
> > > >> > > > overlap significantly. Is it possible for these two
> implementations
> > > >> to
> > > >> > be
> > > >> > > > unified? This would serve to promote feature parity of
> server-side
> > > >> and
> > > >> > > > client-side assignors, and would also facilitate operational
> > > >> > flexibility in
> > > >> > > > certain situations. For example, if a server-side assignor
> has some
> > > >> > poor
> > > >> > > > behavior and needs a patch, deploying the patched assignor to
> the
> > > >> > client
> > > >> > > > and switching one consumer group to a client-side assignor
> may be
> > > >> > faster
> > > >> > > > and less risky than patching all of the brokers. With the
> currently
> > > >> > > > proposed distinct APIs, a non-trivial reimplementation would
> have
> > > >> to be
> > > >> > > > assembled, and if the two APIs have diverged significantly,
> then it
> > > >> is
> > > >> > > > possible that a reimplementation would not be possible.
> > > >> > > >
> > > >> > > > --
> > > >> > > > Greg Harris
> > > >> > > > gharris1727@gmail.com
> > > >> > > > github.com/gharris1727
> > > >> > > >
> > > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <
> sagarmeansocean@gmail.com>
> > > >> > wrote:
> > > >> > > >
> > > >> > > > > Hi Guozhang/David,
> > > >> > > > >
> > > >> > > > > I created a confluence page to discuss how Connect would
> need to
> > > >> > change
> > > >> > > > > based on the new rebalance protocol. Here's the page:
> > > >> > > > >
> > > >> > > > >
> > > >> >
> > > >>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >> > > > >
> > > >> > > > > It's also pretty longish and I have tried to keep a format
> > > >> similar to
> > > >> > > > > KIP-848. Let me know what you think. Also, do you think this
> > > >> should
> > > >> > be
> > > >> > > > > moved to a separate discussion thread or is this one fine?
> > > >> > > > >
> > > >> > > > > Thanks!
> > > >> > > > > Sagar.
> > > >> > > > >
> > > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> sagarmeansocean@gmail.com>
> > > >> > wrote:
> > > >> > > > >
> > > >> > > > > > Hello Guozhang,
> > > >> > > > > >
> > > >> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I
> would do
> > > >> > the
> > > >> > > > > > analysis and come up with such a document.
> > > >> > > > > >
> > > >> > > > > > Thanks!
> > > >> > > > > > Sagar.
> > > >> > > > > >
> > > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > > >> wangguoz@gmail.com>
> > > >> > > > > wrote:
> > > >> > > > > >
> > > >> > > > > >> Hello Sagar,
> > > >> > > > > >>
> > > >> > > > > >> It would be great if you could come back with some
> analysis on
> > > >> > how to
> > > >> > > > > >> implement the Connect side integration with the new
> protocol;
> > > >> so
> > > >> > far
> > > >> > > > > >> besides leveraging on the new "protocol type" we did not
> yet
> > > >> think
> > > >> > > > > through
> > > >> > > > > >> the Connect side implementations. For Streams here's a
> draft of
> > > >> > > > > >> integration
> > > >> > > > > >> plan:
> > > >> > > > > >>
> > > >> > > > > >>
> > > >> > > > >
> > > >> >
> > > >>
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > >> > > > > >> just FYI for your analysis on Connect.
> > > >> > > > > >>
> > > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > > >> sagarmeansocean@gmail.com
> > > >> > >
> > > >> > > > > wrote:
> > > >> > > > > >>
> > > >> > > > > >> > Hi David,
> > > >> > > > > >> >
> > > >> > > > > >> > Thank you for your response. The reason I thought
> connect can
> > > >> > also fit
> > > >> > > > > >> into
> > > >> > > > > >> > this new scheme is that even today the connect uses a
> > > >> > > > > WorkerCoordinator
> > > >> > > > > >> > extending from AbstractCoordinator to empower
> rebalances of
> > > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> > > >> protocolType()
> > > >> > to
> > > >> > > > > >> connect
> > > >> > > > > >> > and uses the metadata() method by plumbing into
> > > >> > > > > >> JoinGroupRequestProtocol.
> > > >> > > > > >> >
> > > >> > > > > >> > I think the changes to support connect would be
> similar at a
> > > >> > high
> > > >> > > > > level
> > > >> > > > > >> to
> > > >> > > > > >> > the changes in streams mainly because of the Client
> side
> > > >> > assignors
> > > >> > > > > being
> > > >> > > > > >> > used in both. At an implementation level, we might
> need to
> > > >> make
> > > >> > a lot
> > > >> > > > > of
> > > >> > > > > >> > changes to get onto this new assignment protocol like
> > > >> enhancing
> > > >> > the
> > > >> > > > > >> > JoinGroup request/response and SyncGroup and using
> > > >> > > > > >> ConsumerGroupHeartbeat
> > > >> > > > > >> > API etc again on similar lines to streams (or there
> might be
> > > >> > > > > >> deviations). I
> > > >> > > > > >> > would try to perform a detailed analysis of the same
> and we
> > > >> > can have
> > > >> > > > > a
> > > >> > > > > >> > separate discussion thread for that as that would
> derail this
> > > >> > > > > discussion
> > > >> > > > > >> > thread. Let me know if that sounds good to you.
> > > >> > > > > >> >
> > > >> > > > > >> > Thanks!
> > > >> > > > > >> > Sagar.
> > > >> > > > > >> >
> > > >> > > > > >> >
> > > >> > > > > >> >
> > > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > >> > > > > <djacot@confluent.io.invalid
> > > >> > > > > >> >
> > > >> > > > > >> > wrote:
> > > >> > > > > >> >
> > > >> > > > > >> > > Hi Sagar,
> > > >> > > > > >> > >
> > > >> > > > > >> > > Thanks for your comments.
> > > >> > > > > >> > >
> > > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I
> can
> > > >> > mention it.
> > > >> > > > > >> > >
> > > >> > > > > >> > > 2) The idea is to transition C from his current
> assignment
> > > >> to
> > > >> > his
> > > >> > > > > >> > > target assignment when he can move to epoch 3. When
> that
> > > >> > happens,
> > > >> > > > > the
> > > >> > > > > >> > > member assignment is updated and persisted with all
> its
> > > >> > assigned
> > > >> > > > > >> > > partitions even if they are not all revoked yet. In
> other
> > > >> > words, the
> > > >> > > > > >> > > member assignment becomes the target assignment.
> This is
> > > >> > basically
> > > >> > > > > an
> > > >> > > > > >> > > optimization to avoid having to write all the
> changes to
> > > >> the
> > > >> > log.
> > > >> > > > > The
> > > >> > > > > >> > > examples are based on the persisted state so I
> understand
> > > >> the
> > > >> > > > > >> > > confusion. Let me see if I can improve this in the
> > > >> > description.
> > > >> > > > > >> > >
> > > >> > > > > >> > > 3) Regarding Connect, it could reuse the protocol
> with a
> > > >> > client side
> > > >> > > > > >> > > assignor if it fits in the protocol. The assignment
> is
> > > >> about
> > > >> > > > > >> > > topicid-partitions + metadata, could Connect fit
> into this?
> > > >> > > > > >> > >
> > > >> > > > > >> > > Best,
> > > >> > > > > >> > > David
> > > >> > > > > >> > >
> > > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > > >> > sagarmeansocean@gmail.com>
> > > >> > > > > >> wrote:
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > Hi David,
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > Thanks for the KIP. I just had minor observations:
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > 1) In the Assignment Error section in Client Side
> mode
> > > >> > Assignment
> > > >> > > > > >> > > process,
> > > >> > > > > >> > > > you mentioned => `In this case, the client side
> assignor
> > > >> can
> > > >> > > > > return
> > > >> > > > > >> an
> > > >> > > > > >> > > > error to the group coordinator`. In this case are
> you
> > > >> > referring to
> > > >> > > > > >> the
> > > >> > > > > >> > > > Assignor returning an AssignmentError that's
> listed down
> > > >> > towards
> > > >> > > > > the
> > > >> > > > > >> > end?
> > > >> > > > > >> > > > If yes, do you think it would make sense to
> mention this
> > > >> > > > > explicitly
> > > >> > > > > >> > here?
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > 2) In the Case Studies section, I have a slight
> > > >> confusion,
> > > >> > not
> > > >> > > > > sure
> > > >> > > > > >> if
> > > >> > > > > >> > > > others have the same. Consider this step:
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > When B heartbeats, the group coordinator
> transitions him
> > > >> to
> > > >> > epoch
> > > >> > > > > 3
> > > >> > > > > >> > > because
> > > >> > > > > >> > > > B has no partitions to revoke. It persists the
> change and
> > > >> > reply.
> > > >> > > > > >> > > >
> > > >> > > > > >> > > >    - Group (epoch=3)
> > > >> > > > > >> > > >       - A
> > > >> > > > > >> > > >       - B
> > > >> > > > > >> > > >       - C
> > > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > > >> > > > > >> > > >       - A - partitions=[foo-0]
> > > >> > > > > >> > > >       - B - partitions=[foo-2]
> > > >> > > > > >> > > >       - C - partitions=[foo-1]
> > > >> > > > > >> > > >    - Member Assignment
> > > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but
> cannot
> > > >> get
> > > >> > foo-1
> > > >> > > > > >> yet.
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > Here,it's mentioned that member C can't get the
> foo-1
> > > >> > partition
> > > >> > > > > yet,
> > > >> > > > > >> > but
> > > >> > > > > >> > > > based on the description above, it seems it
> already has
> > > >> it.
> > > >> > Do you
> > > >> > > > > >> > think
> > > >> > > > > >> > > it
> > > >> > > > > >> > > > would be better to remove it and populate it only
> when it
> > > >> > actually
> > > >> > > > > >> gets
> > > >> > > > > >> > > it?
> > > >> > > > > >> > > > I see this in a lot of other places, so have I
> > > >> understood it
> > > >> > > > > >> > incorrectly
> > > >> > > > > >> > > ?
> > > >> > > > > >> > > >
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > Regarding connect , it might be out of scope of
> this
> > > >> > discussion,
> > > >> > > > > but
> > > >> > > > > >> > from
> > > >> > > > > >> > > > what I understood it would probably be running in
> client
> > > >> > side
> > > >> > > > > >> assignor
> > > >> > > > > >> > > mode
> > > >> > > > > >> > > > even on the new rebalance protocol as it has its
> own
> > > >> Custom
> > > >> > > > > >> > > Assignors(Eager
> > > >> > > > > >> > > > and IncrementalCooperative).
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > Thanks!
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > Sagar.
> > > >> > > > > >> > > >
> > > >> > > > > >> > > >
> > > >> > > > > >> > > >
> > > >> > > > > >> > > >
> > > >> > > > > >> > > >
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > >> > > > > >> > <djacot@confluent.io.invalid
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > wrote:
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
> > > >> > specialized API
> > > >> > > > > >> > > > > instead of relying on one generic API. For
> Connect, we
> > > >> > can apply
> > > >> > > > > >> the
> > > >> > > > > >> > > > > exact same pattern and reuse/share the core
> > > >> > implementation on
> > > >> > > > > the
> > > >> > > > > >> > > > > server side. For the schema registry, I think
> that we
> > > >> > should
> > > >> > > > > >> consider
> > > >> > > > > >> > > > > having a tailored API to do simple
> membership/leader
> > > >> > election.
> > > >> > > > > >> > > > >
> > > >> > > > > >> > > > > Best,
> > > >> > > > > >> > > > > David
> > > >> > > > > >> > > > >
> > > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> > > >> > ismael@juma.me.uk
> > > >> > > > > >
> > > >> > > > > >> > > wrote:
> > > >> > > > > >> > > > > >
> > > >> > > > > >> > > > > > Three quick comments:
> > > >> > > > > >> > > > > >
> > > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > >> > > > > >> com.google.re2j.Pattern, we
> > > >> > > > > >> > > > > should
> > > >> > > > > >> > > > > > document the differences in more detail before
> > > >> deciding
> > > >> > one
> > > >> > > > > way
> > > >> > > > > >> or
> > > >> > > > > >> > > > > another.
> > > >> > > > > >> > > > > > That said, if people pass
> java.util.regex.Pattern,
> > > >> they
> > > >> > expect
> > > >> > > > > >> > their
> > > >> > > > > >> > > > > > semantics to be honored. If we are doing
> something
> > > >> > different,
> > > >> > > > > >> then
> > > >> > > > > >> > we
> > > >> > > > > >> > > > > > should consider adding an overload with our own
> > > >> Pattern
> > > >> > class
> > > >> > > > > (I
> > > >> > > > > >> > > don't
> > > >> > > > > >> > > > > > think we'd want to expose re2j's at this
> point).
> > > >> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol
> should
> > > >> > > > > integrate
> > > >> > > > > >> > fully
> > > >> > > > > >> > > > > with
> > > >> > > > > >> > > > > > it and should handle the topic recreation case
> > > >> > correctly.
> > > >> > > > > That's
> > > >> > > > > >> > the
> > > >> > > > > >> > > main
> > > >> > > > > >> > > > > > part we need to handle. I agree with David
> that we'd
> > > >> > want to
> > > >> > > > > add
> > > >> > > > > >> > > topic
> > > >> > > > > >> > > > > ids
> > > >> > > > > >> > > > > > to the relevant protocols that don't have it
> yet and
> > > >> > that we
> > > >> > > > > can
> > > >> > > > > >> > > probably
> > > >> > > > > >> > > > > > focus on the internals versus adding new APIs
> to the
> > > >> > Java
> > > >> > > > > >> Consumer
> > > >> > > > > >> > > > > (unless
> > > >> > > > > >> > > > > > we find that adding new APIs is required for
> > > >> reasonable
> > > >> > > > > >> semantics).
> > > >> > > > > >> > > > > > 3. I am still not sure about the coordinator
> storing
> > > >> the
> > > >> > > > > >> configs.
> > > >> > > > > >> > > It's
> > > >> > > > > >> > > > > > powerful for configs to be centralized in the
> > > >> metadata
> > > >> > log for
> > > >> > > > > >> > > various
> > > >> > > > > >> > > > > > reasons (auditability, visibility, consistency,
> > > >> etc.).
> > > >> > > > > >> Similarly, I
> > > >> > > > > >> > > am
> > > >> > > > > >> > > > > not
> > > >> > > > > >> > > > > > sure about automatically deleting configs in a
> way
> > > >> that
> > > >> > they
> > > >> > > > > >> cannot
> > > >> > > > > >> > > be
> > > >> > > > > >> > > > > > recovered. A good property for modern systems
> is to
> > > >> > minimize
> > > >> > > > > the
> > > >> > > > > >> > > number
> > > >> > > > > >> > > > > of
> > > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > >> > > > > >> > > > > >
> > > >> > > > > >> > > > > > Ismael
> > > >> > > > > >> > > > > >
> > > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > >> > > > > >> > > <djacot@confluent.io.invalid
> > > >> > > > > >> > > > > >
> > > >> > > > > >> > > > > > wrote:
> > > >> > > > > >> > > > > >
> > > >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > > 1) the migration path, especially the last
> step
> > > >> when
> > > >> > > > > clients
> > > >> > > > > >> > > flip the
> > > >> > > > > >> > > > > > > flag
> > > >> > > > > >> > > > > > > > to enable the new protocol, in which we
> would
> > > >> have a
> > > >> > > > > window
> > > >> > > > > >> > where
> > > >> > > > > >> > > > > both
> > > >> > > > > >> > > > > > > new
> > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs
> are
> > > >> used
> > > >> > by
> > > >> > > > > >> members
> > > >> > > > > >> > of
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > same
> > > >> > > > > >> > > > > > > > group. How the coordinator could "mimic"
> the old
> > > >> > behavior
> > > >> > > > > >> while
> > > >> > > > > >> > > > > using the
> > > >> > > > > >> > > > > > > > new protocol is something we need to
> present
> > > >> about.
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > Noted. I just published a new version of KIP
> which
> > > >> > includes
> > > >> > > > > >> more
> > > >> > > > > >> > > > > > > details about this. See the "Supporting
> Online
> > > >> > Consumer
> > > >> > > > > Group
> > > >> > > > > >> > > Upgrade"
> > > >> > > > > >> > > > > > > and the "Compatibility, Deprecation, and
> Migration
> > > >> > Plan". I
> > > >> > > > > >> think
> > > >> > > > > >> > > that
> > > >> > > > > >> > > > > > > I have to think through a few cases now but
> the
> > > >> > overall idea
> > > >> > > > > >> and
> > > >> > > > > >> > > > > > > mechanism should be understandable.
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> KIP-516 the
> > > >> > topic ids
> > > >> > > > > >> are
> > > >> > > > > >> > > only
> > > >> > > > > >> > > > > used
> > > >> > > > > >> > > > > > > as
> > > >> > > > > >> > > > > > > > part of RPCs and admin client, but they
> are not
> > > >> > exposed
> > > >> > > > > via
> > > >> > > > > >> any
> > > >> > > > > >> > > > > public
> > > >> > > > > >> > > > > > > APIs
> > > >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> first
> > > >> > should we
> > > >> > > > > >> let
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > consumer
> > > >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> mapping
> > > >> > itself
> > > >> > > > > to
> > > >> > > > > >> > fully
> > > >> > > > > >> > > > > > > leverage
> > > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the
> new
> > > >> RPCs
> > > >> > with
> > > >> > > > > the
> > > >> > > > > >> > > topic
> > > >> > > > > >> > > > > ids;
> > > >> > > > > >> > > > > > > and
> > > >> > > > > >> > > > > > > > secondly, should we ever consider exposing
> the
> > > >> > topic ids
> > > >> > > > > in
> > > >> > > > > >> the
> > > >> > > > > >> > > > > consumer
> > > >> > > > > >> > > > > > > > public APIs as well (both
> subscribe/assign, as
> > > >> well
> > > >> > as in
> > > >> > > > > >> the
> > > >> > > > > >> > > > > rebalance
> > > >> > > > > >> > > > > > > > listener for cases like topic
> > > >> > deletion-and-recreation).
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > a) Assuming that we would include converting
> all
> > > >> the
> > > >> > offsets
> > > >> > > > > >> > > related
> > > >> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the
> consumer
> > > >> > would be
> > > >> > > > > >> able
> > > >> > > > > >> > to
> > > >> > > > > >> > > > > > > fully operate with topic ids. That being
> said, it
> > > >> > still has
> > > >> > > > > to
> > > >> > > > > >> > > provide
> > > >> > > > > >> > > > > > > the topics names in various APIs so having a
> > > >> mapping
> > > >> > in the
> > > >> > > > > >> > > consumer
> > > >> > > > > >> > > > > > > seems inevitable to me.
> > > >> > > > > >> > > > > > > b) I don't have a strong opinion on this.
> Here I
> > > >> > wonder if
> > > >> > > > > >> this
> > > >> > > > > >> > > goes
> > > >> > > > > >> > > > > > > beyond the scope of this KIP. I would rather
> focus
> > > >> on
> > > >> > the
> > > >> > > > > >> > internals
> > > >> > > > > >> > > > > > > here and we can consider this separately if
> we see
> > > >> > value in
> > > >> > > > > >> doing
> > > >> > > > > >> > > it.
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > Coming back to Ismael's point about using
> topic ids
> > > >> > in the
> > > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that
> there
> > > >> is
> > > >> > one
> > > >> > > > > >> > advantage
> > > >> > > > > >> > > in
> > > >> > > > > >> > > > > > > favour of it. The consumer will have the
> > > >> opportunity
> > > >> > to
> > > >> > > > > >> validate
> > > >> > > > > >> > > that
> > > >> > > > > >> > > > > > > the topics exists before passing them into
> the
> > > >> group
> > > >> > > > > rebalance
> > > >> > > > > >> > > > > > > protocol. Obviously, the coordinator will
> also
> > > >> notice
> > > >> > it but
> > > >> > > > > >> it
> > > >> > > > > >> > > does
> > > >> > > > > >> > > > > > > not really have a way to reject an invalid
> topic in
> > > >> > the
> > > >> > > > > >> response.
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > > >> questions
> > > >> > > > > except
> > > >> > > > > >> for
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally
> I think
> > > >> > it's not
> > > >> > > > > >> > > necessary
> > > >> > > > > >> > > > > to
> > > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern,
> but
> > > >> > instead we
> > > >> > > > > >> still
> > > >> > > > > >> > > use
> > > >> > > > > >> > > > > > > Pattern
> > > >> > > > > >> > > > > > > > while just documenting that our
> subscription may
> > > >> be
> > > >> > > > > >> rejected by
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > server.
> > > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > > >> scenario
> > > >> > I felt
> > > >> > > > > >> > using
> > > >> > > > > >> > > an
> > > >> > > > > >> > > > > > > > overloaded `String` based subscription may
> be
> > > >> more
> > > >> > > > > >> vulnerable
> > > >> > > > > >> > to
> > > >> > > > > >> > > > > various
> > > >> > > > > >> > > > > > > > invalid regexes.
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > That could work. I have to look at the
> differences
> > > >> > between
> > > >> > > > > the
> > > >> > > > > >> > two
> > > >> > > > > >> > > > > > > engines to better understand the potential
> issues.
> > > >> My
> > > >> > > > > >> > > understanding is
> > > >> > > > > >> > > > > > > that would work for all the basic regular
> > > >> > expressions. The
> > > >> > > > > >> > > differences
> > > >> > > > > >> > > > > > > between the two are mainly about the various
> > > >> character
> > > >> > > > > >> classes. I
> > > >> > > > > >> > > > > > > wonder what other people think about this.
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > Best,
> > > >> > > > > >> > > > > > > David
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang
> Wang <
> > > >> > > > > >> > wangguoz@gmail.com
> > > >> > > > > >> > > >
> > > >> > > > > >> > > > > wrote:
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > Thanks David! I think on the high level
> there are
> > > >> > two meta
> > > >> > > > > >> > > points we
> > > >> > > > > >> > > > > need
> > > >> > > > > >> > > > > > > > to concretize a bit more:
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > 1) the migration path, especially the last
> step
> > > >> when
> > > >> > > > > clients
> > > >> > > > > >> > > flip the
> > > >> > > > > >> > > > > > > flag
> > > >> > > > > >> > > > > > > > to enable the new protocol, in which we
> would
> > > >> have a
> > > >> > > > > window
> > > >> > > > > >> > where
> > > >> > > > > >> > > > > both
> > > >> > > > > >> > > > > > > new
> > > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs
> are
> > > >> used
> > > >> > by
> > > >> > > > > >> members
> > > >> > > > > >> > of
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > same
> > > >> > > > > >> > > > > > > > group. How the coordinator could "mimic"
> the old
> > > >> > behavior
> > > >> > > > > >> while
> > > >> > > > > >> > > > > using the
> > > >> > > > > >> > > > > > > > new protocol is something we need to
> present
> > > >> about.
> > > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as
> KIP-516 the
> > > >> > topic ids
> > > >> > > > > >> are
> > > >> > > > > >> > > only
> > > >> > > > > >> > > > > used
> > > >> > > > > >> > > > > > > as
> > > >> > > > > >> > > > > > > > part of RPCs and admin client, but they
> are not
> > > >> > exposed
> > > >> > > > > via
> > > >> > > > > >> any
> > > >> > > > > >> > > > > public
> > > >> > > > > >> > > > > > > APIs
> > > >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> first
> > > >> > should we
> > > >> > > > > >> let
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > consumer
> > > >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> mapping
> > > >> > itself
> > > >> > > > > to
> > > >> > > > > >> > fully
> > > >> > > > > >> > > > > > > leverage
> > > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the
> new
> > > >> RPCs
> > > >> > with
> > > >> > > > > the
> > > >> > > > > >> > > topic
> > > >> > > > > >> > > > > ids;
> > > >> > > > > >> > > > > > > and
> > > >> > > > > >> > > > > > > > secondly, should we ever consider exposing
> the
> > > >> > topic ids
> > > >> > > > > in
> > > >> > > > > >> the
> > > >> > > > > >> > > > > consumer
> > > >> > > > > >> > > > > > > > public APIs as well (both
> subscribe/assign, as
> > > >> well
> > > >> > as in
> > > >> > > > > >> the
> > > >> > > > > >> > > > > rebalance
> > > >> > > > > >> > > > > > > > listener for cases like topic
> > > >> > deletion-and-recreation).
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > > >> questions
> > > >> > > > > except
> > > >> > > > > >> for
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally
> I think
> > > >> > it's not
> > > >> > > > > >> > > necessary
> > > >> > > > > >> > > > > to
> > > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern,
> but
> > > >> > instead we
> > > >> > > > > >> still
> > > >> > > > > >> > > use
> > > >> > > > > >> > > > > > > Pattern
> > > >> > > > > >> > > > > > > > while just documenting that our
> subscription may
> > > >> be
> > > >> > > > > >> rejected by
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > server.
> > > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > > >> scenario
> > > >> > I felt
> > > >> > > > > >> > using
> > > >> > > > > >> > > an
> > > >> > > > > >> > > > > > > > overloaded `String` based subscription may
> be
> > > >> more
> > > >> > > > > >> vulnerable
> > > >> > > > > >> > to
> > > >> > > > > >> > > > > various
> > > >> > > > > >> > > > > > > > invalid regexes.
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > Guozhang
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > wrote:
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > > Hi Ismael,
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer
> your
> > > >> > questions
> > > >> > > > > >> > inline.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 1. I think it's premature to talk about
> > > >> target
> > > >> > > > > versions
> > > >> > > > > >> for
> > > >> > > > > >> > > > > > > deprecation
> > > >> > > > > >> > > > > > > > > and
> > > >> > > > > >> > > > > > > > > > removal of the existing group protocol.
> > > >> Unlike
> > > >> > KRaft,
> > > >> > > > > >> this
> > > >> > > > > >> > > > > affects a
> > > >> > > > > >> > > > > > > core
> > > >> > > > > >> > > > > > > > > > client protocol and hence
> deprecation/removal
> > > >> > will be
> > > >> > > > > >> > heavily
> > > >> > > > > >> > > > > > > dependent
> > > >> > > > > >> > > > > > > > > on
> > > >> > > > > >> > > > > > > > > > how quickly applications migrate to
> the new
> > > >> > protocol.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to release
> this in
> > > >> > 4.x, but
> > > >> > > > > it
> > > >> > > > > >> > > wasn't
> > > >> > > > > >> > > > > made
> > > >> > > > > >> > > > > > > > > clear
> > > >> > > > > >> > > > > > > > > > why. If we added that as a way to
> estimate
> > > >> when
> > > >> > we'd
> > > >> > > > > >> > > deprecate
> > > >> > > > > >> > > > > and
> > > >> > > > > >> > > > > > > remove
> > > >> > > > > >> > > > > > > > > > the group protocol, I also suggest
> removing
> > > >> > this part.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Let me explain my reasoning. As
> explained, I
> > > >> plan
> > > >> > to
> > > >> > > > > >> rewrite
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > group
> > > >> > > > > >> > > > > > > > > coordinator in Java while we implement
> the new
> > > >> > protocol.
> > > >> > > > > >> This
> > > >> > > > > >> > > means
> > > >> > > > > >> > > > > > > > > that the internals will be slightly
> different
> > > >> > (e.g.
> > > >> > > > > >> threading
> > > >> > > > > >> > > > > model).
> > > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the
> switch from
> > > >> > the old
> > > >> > > > > >> group
> > > >> > > > > >> > > > > > > > > coordinator to the new group coordinator
> to a
> > > >> > major
> > > >> > > > > >> release.
> > > >> > > > > >> > > The
> > > >> > > > > >> > > > > > > > > alternative would be to use a flag to do
> the
> > > >> > switch
> > > >> > > > > >> instead
> > > >> > > > > >> > of
> > > >> > > > > >> > > > > relying
> > > >> > > > > >> > > > > > > > > on the software upgrade.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 3. We need to flesh out the details of
> the
> > > >> > migration
> > > >> > > > > >> story.
> > > >> > > > > >> > > It
> > > >> > > > > >> > > > > sounds
> > > >> > > > > >> > > > > > > > > like
> > > >> > > > > >> > > > > > > > > > we're saying we will support online
> > > >> migrations.
> > > >> > Is
> > > >> > > > > that
> > > >> > > > > >> > > correct?
> > > >> > > > > >> > > > > We
> > > >> > > > > >> > > > > > > > > should
> > > >> > > > > >> > > > > > > > > > explain this in detail. It could also
> be done
> > > >> > as a
> > > >> > > > > >> separate
> > > >> > > > > >> > > KIP,
> > > >> > > > > >> > > > > if
> > > >> > > > > >> > > > > > > it's
> > > >> > > > > >> > > > > > > > > > easier.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Yes, we will support online migrations
> for the
> > > >> > group.
> > > >> > > > > That
> > > >> > > > > >> > > means
> > > >> > > > > >> > > > > that
> > > >> > > > > >> > > > > > > > > a group using the old protocol will be
> able to
> > > >> > switch to
> > > >> > > > > >> the
> > > >> > > > > >> > > new
> > > >> > > > > >> > > > > > > > > protocol.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Let me briefly explain how that will work
> > > >> though.
> > > >> > It is
> > > >> > > > > >> > > basically a
> > > >> > > > > >> > > > > > > > > four step process:
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or
> rolled to a
> > > >> > software
> > > >> > > > > >> > > supporting
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > new group coordinator. Both the old and
> the new
> > > >> > > > > >> coordinator
> > > >> > > > > >> > > will
> > > >> > > > > >> > > > > > > > > support the old protocol and rely on the
> same
> > > >> > persisted
> > > >> > > > > >> > > metadata so
> > > >> > > > > >> > > > > > > > > they can work together. This point is an
> > > >> offline
> > > >> > > > > >> migration.
> > > >> > > > > >> > We
> > > >> > > > > >> > > > > cannot
> > > >> > > > > >> > > > > > > > > do this one live because it would require
> > > >> > shutting down
> > > >> > > > > >> the
> > > >> > > > > >> > > current
> > > >> > > > > >> > > > > > > > > coordinator and starting up the new one
> and
> > > >> that
> > > >> > would
> > > >> > > > > >> cause
> > > >> > > > > >> > > > > > > > > unavailabilities.
> > > >> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP
> must be
> > > >> > upgraded
> > > >> > > > > to
> > > >> > > > > >> X
> > > >> > > > > >> > in
> > > >> > > > > >> > > > > order
> > > >> > > > > >> > > > > > > > > to enable the new protocol. This cannot
> be done
> > > >> > before
> > > >> > > > > 1)
> > > >> > > > > >> is
> > > >> > > > > >> > > > > > > > > terminated because the old coordinator
> doesn't
> > > >> > support
> > > >> > > > > the
> > > >> > > > > >> > new
> > > >> > > > > >> > > > > > > > > protocol.
> > > >> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a
> version
> > > >> > > > > supporting
> > > >> > > > > >> the
> > > >> > > > > >> > > > > online
> > > >> > > > > >> > > > > > > > > migration (must have KIP-792). If the
> consumer
> > > >> is
> > > >> > > > > already
> > > >> > > > > >> > > there.
> > > >> > > > > >> > > > > > > > > Nothing must be done at this point.
> > > >> > > > > >> > > > > > > > > 4. The consumers must be rolled with the
> > > >> feature
> > > >> > flag
> > > >> > > > > >> turned
> > > >> > > > > >> > > on.
> > > >> > > > > >> > > > > The
> > > >> > > > > >> > > > > > > > > consumer group is automatically
> converted when
> > > >> > the first
> > > >> > > > > >> > > consumer
> > > >> > > > > >> > > > > > > > > using the new protocol joins the group.
> While
> > > >> the
> > > >> > > > > members
> > > >> > > > > >> > > using the
> > > >> > > > > >> > > > > > > > > old protocol are being upgraded, the old
> > > >> protocol
> > > >> > is
> > > >> > > > > >> proxied
> > > >> > > > > >> > > into
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > new one.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the
> pattern
> > > >> > > > > >> subscriptions
> > > >> > > > > >> > > to
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > server,
> > > >> > > > > >> > > > > > > > > > but it seems like there could be some
> tricky
> > > >> > > > > >> compatibility
> > > >> > > > > >> > > > > issues.
> > > >> > > > > >> > > > > > > Will
> > > >> > > > > >> > > > > > > > > we
> > > >> > > > > >> > > > > > > > > > have a mechanism for users to detect
> that
> > > >> they
> > > >> > need to
> > > >> > > > > >> > update
> > > >> > > > > >> > > > > their
> > > >> > > > > >> > > > > > > regex
> > > >> > > > > >> > > > > > > > > > before switching to the new protocol?
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > I think that I am a bit more optimistic
> than
> > > >> you
> > > >> > on this
> > > >> > > > > >> > > point. I
> > > >> > > > > >> > > > > > > > > believe that the majority of the cases
> are
> > > >> simple
> > > >> > > > > regexes
> > > >> > > > > >> > which
> > > >> > > > > >> > > > > should
> > > >> > > > > >> > > > > > > > > work with the new engine. The
> coordinator will
> > > >> > verify
> > > >> > > > > the
> > > >> > > > > >> > regex
> > > >> > > > > >> > > > > anyway
> > > >> > > > > >> > > > > > > > > and reject the consumer if the regex is
> not
> > > >> valid.
> > > >> > > > > Coming
> > > >> > > > > >> > back
> > > >> > > > > >> > > to
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > migration path, in the worst case, the
> first
> > > >> > upgraded
> > > >> > > > > >> > consumer
> > > >> > > > > >> > > > > joining
> > > >> > > > > >> > > > > > > > > the group will be rejected. This should
> be used
> > > >> > as the
> > > >> > > > > >> last
> > > >> > > > > >> > > > > defence, I
> > > >> > > > > >> > > > > > > > > would say.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > One way for customers to validate their
> regex
> > > >> > before
> > > >> > > > > >> > upgrading
> > > >> > > > > >> > > > > their
> > > >> > > > > >> > > > > > > > > prod would be to test them with another
> group.
> > > >> For
> > > >> > > > > >> instance,
> > > >> > > > > >> > > that
> > > >> > > > > >> > > > > > > > > could be done in a pre-prod environment.
> > > >> Another
> > > >> > way
> > > >> > > > > >> would be
> > > >> > > > > >> > > to
> > > >> > > > > >> > > > > > > > > extend the consumer-group tool to
> provide a
> > > >> regex
> > > >> > > > > >> validation
> > > >> > > > > >> > > > > > > > > mechanism. Would this be enough in your
> > > >> opinion?
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 5. Related to the last question, will
> the
> > > >> Java
> > > >> > client
> > > >> > > > > >> allow
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > users to
> > > >> > > > > >> > > > > > > > > > stick with the current regex engine for
> > > >> > compatibility
> > > >> > > > > >> > > reasons?
> > > >> > > > > >> > > > > For
> > > >> > > > > >> > > > > > > > > example,
> > > >> > > > > >> > > > > > > > > > it may be handy to keep using client
> based
> > > >> > regex at
> > > >> > > > > >> first
> > > >> > > > > >> > to
> > > >> > > > > >> > > keep
> > > >> > > > > >> > > > > > > > > > migrations simple and then migrate to
> server
> > > >> > based
> > > >> > > > > >> regexes
> > > >> > > > > >> > > as a
> > > >> > > > > >> > > > > > > second
> > > >> > > > > >> > > > > > > > > step.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > I understand your point but I am
> concerned that
> > > >> > this
> > > >> > > > > would
> > > >> > > > > >> > > allow
> > > >> > > > > >> > > > > users
> > > >> > > > > >> > > > > > > > > to actually stay in this mode. That
> would go
> > > >> > against our
> > > >> > > > > >> goal
> > > >> > > > > >> > > of
> > > >> > > > > >> > > > > > > > > simplifying the client because we would
> have to
> > > >> > continue
> > > >> > > > > >> > > monitoring
> > > >> > > > > >> > > > > > > > > the metadata on the client side. I would
> rather
> > > >> > not do
> > > >> > > > > >> this.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 6. When we say that the group
> coordinator
> > > >> will
> > > >> > be
> > > >> > > > > >> > > responsible for
> > > >> > > > > >> > > > > > > storing
> > > >> > > > > >> > > > > > > > > > the configurations and that the
> > > >> configurations
> > > >> > will be
> > > >> > > > > >> > > deleted
> > > >> > > > > >> > > > > when
> > > >> > > > > >> > > > > > > the
> > > >> > > > > >> > > > > > > > > > group is deleted. Will a transition to
> DEAD
> > > >> > trigger
> > > >> > > > > >> > deletion
> > > >> > > > > >> > > of
> > > >> > > > > >> > > > > > > > > > configurations?
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > That's right. The configurations will be
> > > >> deleted
> > > >> > when
> > > >> > > > > the
> > > >> > > > > >> > > group is
> > > >> > > > > >> > > > > > > > > deleted. They go together.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 7. Will the choice to store the
> configs in
> > > >> the
> > > >> > group
> > > >> > > > > >> > > coordinator
> > > >> > > > > >> > > > > > > make it
> > > >> > > > > >> > > > > > > > > > harder to list all cluster configs and
> their
> > > >> > values?
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > I don't think so. The group
> configurations are
> > > >> > overrides
> > > >> > > > > >> of
> > > >> > > > > >> > > cluster
> > > >> > > > > >> > > > > > > > > configs. If you want to know all the
> overrides
> > > >> > though,
> > > >> > > > > you
> > > >> > > > > >> > > would
> > > >> > > > > >> > > > > have
> > > >> > > > > >> > > > > > > > > to ask all the group coordinators. You
> cannot
> > > >> > rely on
> > > >> > > > > the
> > > >> > > > > >> > > metadata
> > > >> > > > > >> > > > > log
> > > >> > > > > >> > > > > > > > > for instance.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 8. How would someone configure a group
> before
> > > >> > starting
> > > >> > > > > >> the
> > > >> > > > > >> > > > > consumers?
> > > >> > > > > >> > > > > > > > > Have
> > > >> > > > > >> > > > > > > > > > we considered allowing the explicit
> creation
> > > >> of
> > > >> > > > > groups?
> > > >> > > > > >> > > > > > > Alternatively,
> > > >> > > > > >> > > > > > > > > the
> > > >> > > > > >> > > > > > > > > > configs could be decoupled from the
> group
> > > >> > lifecycle.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Yes. The group will be automatically
> created in
> > > >> > this
> > > >> > > > > case.
> > > >> > > > > >> > > However,
> > > >> > > > > >> > > > > > > > > the configs will be lost after the
> retention
> > > >> > period of
> > > >> > > > > the
> > > >> > > > > >> > > group
> > > >> > > > > >> > > > > > > > > passes.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method
> for the
> > > >> > Java
> > > >> > > > > >> client
> > > >> > > > > >> > > still
> > > >> > > > > >> > > > > take
> > > >> > > > > >> > > > > > > a
> > > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we
> have to
> > > >> > introduce
> > > >> > > > > an
> > > >> > > > > >> > > overload?
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > That's a very group question. I forgot
> about
> > > >> that
> > > >> > one.
> > > >> > > > > As
> > > >> > > > > >> the
> > > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
> > > >> compatible
> > > >> > with
> > > >> > > > > the
> > > >> > > > > >> > > engine
> > > >> > > > > >> > > > > that
> > > >> > > > > >> > > > > > > > > we plan to use, it might be better to
> deprecate
> > > >> > it and
> > > >> > > > > >> use an
> > > >> > > > > >> > > > > overload
> > > >> > > > > >> > > > > > > > > which takes a string. We would rely on
> the
> > > >> server
> > > >> > side
> > > >> > > > > >> > > validation.
> > > >> > > > > >> > > > > > > > > During the migration, I think that we
> could
> > > >> still
> > > >> > try to
> > > >> > > > > >> > > toString
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > regex and use it. That should work, I
> think, in
> > > >> > the
> > > >> > > > > >> majority
> > > >> > > > > >> > > of the
> > > >> > > > > >> > > > > > > > > cases.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 10. I agree with Justine that we
> should be
> > > >> > clearer
> > > >> > > > > about
> > > >> > > > > >> > the
> > > >> > > > > >> > > > > reason
> > > >> > > > > >> > > > > > > to
> > > >> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
> > > >> feature
> > > >> > flag.
> > > >> > > > > >> Maybe
> > > >> > > > > >> > > we
> > > >> > > > > >> > > > > mean
> > > >> > > > > >> > > > > > > that
> > > >> > > > > >> > > > > > > > > we
> > > >> > > > > >> > > > > > > > > > can switch the default for the feature
> flag
> > > >> to
> > > >> > true
> > > >> > > > > >> based
> > > >> > > > > >> > on
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > > > > metadata.version once we want to make
> it the
> > > >> > default.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > My plan was to use that feature flag
> mainly
> > > >> > during the
> > > >> > > > > >> > > development
> > > >> > > > > >> > > > > > > > > phase. I should not have mentioned it, I
> think,
> > > >> > because
> > > >> > > > > we
> > > >> > > > > >> > > could
> > > >> > > > > >> > > > > use
> > > >> > > > > >> > > > > > > > > an internal config for it.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't
> mention
> > > >> the
> > > >> > > > > required
> > > >> > > > > >> > > ACLs, it
> > > >> > > > > >> > > > > > > would
> > > >> > > > > >> > > > > > > > > be
> > > >> > > > > >> > > > > > > > > > good to add that for consistency.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Noted.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > > >> ConsumerGroupHeartbeat
> > > >> > > > > requires
> > > >> > > > > >> > > "Read
> > > >> > > > > >> > > > > Group"
> > > >> > > > > >> > > > > > > > > even
> > > >> > > > > >> > > > > > > > > > though it seems to do more than
> reading.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > I agree. This is how the current
> protocol works
> > > >> > though.
> > > >> > > > > We
> > > >> > > > > >> > only
> > > >> > > > > >> > > > > > > > > require "Read Group" to join a group. We
> could
> > > >> > consider
> > > >> > > > > >> > > changing
> > > >> > > > > >> > > > > this
> > > >> > > > > >> > > > > > > > > but I am not sure that it is worth it.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 13. How is topic recreation handled by
> the
> > > >> > consumer
> > > >> > > > > with
> > > >> > > > > >> > the
> > > >> > > > > >> > > new
> > > >> > > > > >> > > > > > > group
> > > >> > > > > >> > > > > > > > > > protocol? It would be good to have a
> section
> > > >> on
> > > >> > this.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Noted. From a protocol perspective, the
> new
> > > >> topic
> > > >> > will
> > > >> > > > > >> have a
> > > >> > > > > >> > > new
> > > >> > > > > >> > > > > > > > > topic id so it will treat it like a
> topic with
> > > >> a
> > > >> > > > > different
> > > >> > > > > >> > > name.
> > > >> > > > > >> > > > > The
> > > >> > > > > >> > > > > > > > > only issue is that the fetch/commit
> offsets
> > > >> APIs
> > > >> > do not
> > > >> > > > > >> > support
> > > >> > > > > >> > > > > topic
> > > >> > > > > >> > > > > > > > > IDs so the consumer would reuse the
> offsets
> > > >> based
> > > >> > on the
> > > >> > > > > >> > same.
> > > >> > > > > >> > > I
> > > >> > > > > >> > > > > think
> > > >> > > > > >> > > > > > > > > that we should update those APIs as well
> in
> > > >> order
> > > >> > to be
> > > >> > > > > >> > > consistent
> > > >> > > > > >> > > > > end
> > > >> > > > > >> > > > > > > > > to end. That would strengthen the
> semantics of
> > > >> the
> > > >> > > > > >> consumer.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the
> new
> > > >> > coordinator
> > > >> > > > > >> in
> > > >> > > > > >> > > Java.
> > > >> > > > > >> > > > > Even
> > > >> > > > > >> > > > > > > > > though
> > > >> > > > > >> > > > > > > > > > this is an implementation detail, do
> we plan
> > > >> to
> > > >> > have a
> > > >> > > > > >> new
> > > >> > > > > >> > > gradle
> > > >> > > > > >> > > > > > > module
> > > >> > > > > >> > > > > > > > > > for it?
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Yes.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when
> it
> > > >> comes
> > > >> > to how
> > > >> > > > > >> many
> > > >> > > > > >> > > > > members
> > > >> > > > > >> > > > > > > the
> > > >> > > > > >> > > > > > > > > new
> > > >> > > > > >> > > > > > > > > > group protocol can support?
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > We don't have numbers at the moment. The
> > > >> protocol
> > > >> > should
> > > >> > > > > >> > > support
> > > >> > > > > >> > > > > 1000s
> > > >> > > > > >> > > > > > > > > of members per group. We will measure
> this when
> > > >> > we have
> > > >> > > > > a
> > > >> > > > > >> > first
> > > >> > > > > >> > > > > > > > > implementation. Note that we might have
> other
> > > >> > > > > bottlenecks
> > > >> > > > > >> > down
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > > 16. Did we consider having
> SubscribedTopidIds
> > > >> > instead
> > > >> > > > > of
> > > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > > >> > ConsumerGroupHeartbeatRequest?
> > > >> > > > > >> Is
> > > >> > > > > >> > the
> > > >> > > > > >> > > > > idea
> > > >> > > > > >> > > > > > > that
> > > >> > > > > >> > > > > > > > > > since we have to resolve the regex on
> the
> > > >> > server, we
> > > >> > > > > >> can do
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > same
> > > >> > > > > >> > > > > > > for
> > > >> > > > > >> > > > > > > > > > the topic name? The difference is that
> > > >> sending
> > > >> > the
> > > >> > > > > >> regex is
> > > >> > > > > >> > > more
> > > >> > > > > >> > > > > > > > > efficient
> > > >> > > > > >> > > > > > > > > > whereas sending the topic names is less
> > > >> > efficient.
> > > >> > > > > >> > > Furthermore,
> > > >> > > > > >> > > > > > > delete
> > > >> > > > > >> > > > > > > > > and
> > > >> > > > > >> > > > > > > > > > recreation is easier to handle if we
> have
> > > >> topic
> > > >> > ids.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > The idea was to consolidate the metadata
> lookup
> > > >> > on the
> > > >> > > > > >> server
> > > >> > > > > >> > > for
> > > >> > > > > >> > > > > both
> > > >> > > > > >> > > > > > > > > paths but I do agree with your point. As
> a
> > > >> second
> > > >> > > > > though,
> > > >> > > > > >> > using
> > > >> > > > > >> > > > > topic
> > > >> > > > > >> > > > > > > > > ids may be better here for the delete and
> > > >> > recreation
> > > >> > > > > case.
> > > >> > > > > >> > > Also, I
> > > >> > > > > >> > > > > > > > > suppose that we may allow users to
> subscribe
> > > >> with
> > > >> > topic
> > > >> > > > > >> ids
> > > >> > > > > >> > in
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > > > future because that is the only way to be
> > > >> really
> > > >> > robust
> > > >> > > > > to
> > > >> > > > > >> > > topic
> > > >> > > > > >> > > > > > > > > re-creation.
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > Best,
> > > >> > > > > >> > > > > > > > > David
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David
> Jacot <
> > > >> > > > > >> > > djacot@confluent.io>
> > > >> > > > > >> > > > > > > wrote:
> > > >> > > > > >> > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > Hi Justine,
> > > >> > > > > >> > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > Thanks for your comments. Please find
> my
> > > >> answers
> > > >> > > > > below.
> > > >> > > > > >> > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on
> topic IDs
> > > >> > with the
> > > >> > > > > >> > > exception
> > > >> > > > > >> > > > > of the
> > > >> > > > > >> > > > > > > > > > topic names based in the
> > > >> > > > > ConsumerGroupHeartbeatRequest.
> > > >> > > > > >> I
> > > >> > > > > >> > am
> > > >> > > > > >> > > not
> > > >> > > > > >> > > > > sure
> > > >> > > > > >> > > > > > > > > > if using topic names is the right call
> here.
> > > >> I
> > > >> > need to
> > > >> > > > > >> > think
> > > >> > > > > >> > > > > about it
> > > >> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does
> not
> > > >> > change the
> > > >> > > > > >> > > > > fetch/commit
> > > >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This
> may be
> > > >> > something
> > > >> > > > > >> that
> > > >> > > > > >> > we
> > > >> > > > > >> > > > > should
> > > >> > > > > >> > > > > > > > > > include though as it would give better
> > > >> overall
> > > >> > > > > >> guarantee in
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > > > > producer.
> > > >> > > > > >> > > > > > > > > > - You're right. I think that I should
> not
> > > >> have
> > > >> > > > > mentioned
> > > >> > > > > >> > this
> > > >> > > > > >> > > > > flag at
> > > >> > > > > >> > > > > > > > > > all. I will remove it. We can use an
> internal
> > > >> > > > > >> configuration
> > > >> > > > > >> > > while
> > > >> > > > > >> > > > > > > > > > developing the feature.
> > > >> > > > > >> > > > > > > > > > - Both cluster types will be
> supported. The
> > > >> > change is
> > > >> > > > > >> > > > > orthogonal. The
> > > >> > > > > >> > > > > > > > > > only requirement is that the cluster
> uses
> > > >> topic
> > > >> > IDs.
> > > >> > > > > >> > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > Best,
> > > >> > > > > >> > > > > > > > > > David
> > > >> > > > > >> > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM
> Guozhang
> > > >> Wang <
> > > >> > > > > >> > > > > wangguoz@gmail.com>
> > > >> > > > > >> > > > > > > > > wrote:
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are
> some
> > > >> replies
> > > >> > > > > inlined
> > > >> > > > > >> > > below:
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM
> Ismael Juma
> > > >> <
> > > >> > > > > >> > > ismael@juma.me.uk>
> > > >> > > > > >> > > > > > > wrote:
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the
> > > >> potential
> > > >> > to be a
> > > >> > > > > >> > great
> > > >> > > > > >> > > > > > > > > improvement. A few
> > > >> > > > > >> > > > > > > > > > > > initial questions/comments:
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk
> about
> > > >> > target
> > > >> > > > > >> versions
> > > >> > > > > >> > > for
> > > >> > > > > >> > > > > > > > > deprecation and
> > > >> > > > > >> > > > > > > > > > > > removal of the existing group
> protocol.
> > > >> > Unlike
> > > >> > > > > >> KRaft,
> > > >> > > > > >> > > this
> > > >> > > > > >> > > > > > > affects a
> > > >> > > > > >> > > > > > > > > core
> > > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > > >> > deprecation/removal will
> > > >> > > > > >> be
> > > >> > > > > >> > > heavily
> > > >> > > > > >> > > > > > > > > dependent on
> > > >> > > > > >> > > > > > > > > > > > how quickly applications migrate
> to the
> > > >> new
> > > >> > > > > >> protocol.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can
> > > >> remove
> > > >> > the
> > > >> > > > > >> proposed
> > > >> > > > > >> > > > > timeline
> > > >> > > > > >> > > > > > > in
> > > >> > > > > >> > > > > > > > > the
> > > >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and
> Migration
> > > >> > Plan` and
> > > >> > > > > >> > > instead
> > > >> > > > > >> > > > > just
> > > >> > > > > >> > > > > > > state
> > > >> > > > > >> > > > > > > > > > > that we will decide in the future
> about
> > > >> when
> > > >> > we
> > > >> > > > > would
> > > >> > > > > >> > > > > deprecate old
> > > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to
> release this
> > > >> > in 4.x,
> > > >> > > > > >> but
> > > >> > > > > >> > it
> > > >> > > > > >> > > > > wasn't
> > > >> > > > > >> > > > > > > made
> > > >> > > > > >> > > > > > > > > clear
> > > >> > > > > >> > > > > > > > > > > > why. If we added that as a way to
> > > >> estimate
> > > >> > when
> > > >> > > > > we'd
> > > >> > > > > >> > > > > deprecate
> > > >> > > > > >> > > > > > > and
> > > >> > > > > >> > > > > > > > > remove
> > > >> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
> > > >> removing
> > > >> > this
> > > >> > > > > >> part.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > I think that's not specifically
> related to
> > > >> the
> > > >> > > > > >> > > > > deprecation/removal
> > > >> > > > > >> > > > > > > > > timeline
> > > >> > > > > >> > > > > > > > > > > plan, but it's more for client
> upgrades.
> > > >> I.e.
> > > >> > the
> > > >> > > > > >> > > broker-side
> > > >> > > > > >> > > > > > > > > > > implementation may be done first,
> and then
> > > >> the
> > > >> > > > > client
> > > >> > > > > >> > side,
> > > >> > > > > >> > > > > and we
> > > >> > > > > >> > > > > > > > > would
> > > >> > > > > >> > > > > > > > > > > only mark it as "released" by the
> time
> > > >> clients
> > > >> > > > > >> > > implementations
> > > >> > > > > >> > > > > are
> > > >> > > > > >> > > > > > > > > done. At
> > > >> > > > > >> > > > > > > > > > > that time, to enable the feature the
> > > >> clients
> > > >> > need to
> > > >> > > > > >> > first
> > > >> > > > > >> > > > > swap-in
> > > >> > > > > >> > > > > > > the
> > > >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and
> then set
> > > >> > the flag
> > > >> > > > > >> > with a
> > > >> > > > > >> > > > > second
> > > >> > > > > >> > > > > > > > > rolling
> > > >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's
> better to be
> > > >> > released
> > > >> > > > > >> in a
> > > >> > > > > >> > > major
> > > >> > > > > >> > > > > > > > > version.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the
> details of
> > > >> the
> > > >> > > > > migration
> > > >> > > > > >> > > story.
> > > >> > > > > >> > > > > It
> > > >> > > > > >> > > > > > > > > sounds like
> > > >> > > > > >> > > > > > > > > > > > we're saying we will support online
> > > >> > migrations. Is
> > > >> > > > > >> that
> > > >> > > > > >> > > > > correct?
> > > >> > > > > >> > > > > > > We
> > > >> > > > > >> > > > > > > > > should
> > > >> > > > > >> > > > > > > > > > > > explain this in detail. It could
> also be
> > > >> > done as a
> > > >> > > > > >> > > separate
> > > >> > > > > >> > > > > KIP,
> > > >> > > > > >> > > > > > > if
> > > >> > > > > >> > > > > > > > > it's
> > > >> > > > > >> > > > > > > > > > > > easier.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Yes I think that's the part we can
> be more
> > > >> > concrete
> > > >> > > > > >> about
> > > >> > > > > >> > > for
> > > >> > > > > >> > > > > sure
> > > >> > > > > >> > > > > > > (and
> > > >> > > > > >> > > > > > > > > > > this is related to your question 2)
> above).
> > > >> > We will
> > > >> > > > > >> work
> > > >> > > > > >> > on
> > > >> > > > > >> > > > > making
> > > >> > > > > >> > > > > > > it
> > > >> > > > > >> > > > > > > > > more
> > > >> > > > > >> > > > > > > > > > > explicit in parallel as we solicit
> more
> > > >> > feedback.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing
> the
> > > >> > pattern
> > > >> > > > > >> > > subscriptions
> > > >> > > > > >> > > > > to
> > > >> > > > > >> > > > > > > the
> > > >> > > > > >> > > > > > > > > server,
> > > >> > > > > >> > > > > > > > > > > > but it seems like there could be
> some
> > > >> tricky
> > > >> > > > > >> > > compatibility
> > > >> > > > > >> > > > > > > issues.
> > > >> > > > > >> > > > > > > > > Will we
> > > >> > > > > >> > > > > > > > > > > > have a mechanism for users to
> detect that
> > > >> > they
> > > >> > > > > need
> > > >> > > > > >> to
> > > >> > > > > >> > > update
> > > >> > > > > >> > > > > > > their
> > > >> > > > > >> > > > > > > > > regex
> > > >> > > > > >> > > > > > > > > > > > before switching to the new
> protocol?
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
> > > >> non-java
> > > >> > client
> > > >> > > > > >> > users
> > > >> > > > > >> > > to
> > > >> > > > > >> > > > > sort
> > > >> > > > > >> > > > > > > of
> > > >> > > > > >> > > > > > > > > > > "dry-run" the client before
> switching to
> > > >> the
> > > >> > new
> > > >> > > > > >> > protocol.
> > > >> > > > > >> > > I
> > > >> > > > > >> > > > > do not
> > > >> > > > > >> > > > > > > > > have a
> > > >> > > > > >> > > > > > > > > > > specific idea on top of my head
> though,
> > > >> maybe
> > > >> > others
> > > >> > > > > >> like
> > > >> > > > > >> > > @Matt
> > > >> > > > > >> > > > > > > > > Howlett can
> > > >> > > > > >> > > > > > > > > > > chime-in here?
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 5. Related to the last question,
> will the
> > > >> > Java
> > > >> > > > > >> client
> > > >> > > > > >> > > allow
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > users to
> > > >> > > > > >> > > > > > > > > > > > stick with the current regex
> engine for
> > > >> > > > > >> compatibility
> > > >> > > > > >> > > > > reasons?
> > > >> > > > > >> > > > > > > For
> > > >> > > > > >> > > > > > > > > example,
> > > >> > > > > >> > > > > > > > > > > > it may be handy to keep using
> client
> > > >> based
> > > >> > regex
> > > >> > > > > at
> > > >> > > > > >> > > first to
> > > >> > > > > >> > > > > keep
> > > >> > > > > >> > > > > > > > > > > > migrations simple and then migrate
> to
> > > >> > server based
> > > >> > > > > >> > > regexes
> > > >> > > > > >> > > > > as a
> > > >> > > > > >> > > > > > > > > second
> > > >> > > > > >> > > > > > > > > > > > step.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Honestly I have not thought about
> that for
> > > >> > java
> > > >> > > > > >> clients,
> > > >> > > > > >> > > and
> > > >> > > > > >> > > > > we can
> > > >> > > > > >> > > > > > > > > discuss
> > > >> > > > > >> > > > > > > > > > > that. What kind of compatibility
> issues do
> > > >> > you have
> > > >> > > > > in
> > > >> > > > > >> > > mind?
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 6. When we say that the group
> coordinator
> > > >> > will be
> > > >> > > > > >> > > > > responsible for
> > > >> > > > > >> > > > > > > > > storing
> > > >> > > > > >> > > > > > > > > > > > the configurations and that the
> > > >> > configurations
> > > >> > > > > will
> > > >> > > > > >> be
> > > >> > > > > >> > > > > deleted
> > > >> > > > > >> > > > > > > when
> > > >> > > > > >> > > > > > > > > the
> > > >> > > > > >> > > > > > > > > > > > group is deleted. Will a
> transition to
> > > >> DEAD
> > > >> > > > > trigger
> > > >> > > > > >> > > deletion
> > > >> > > > > >> > > > > of
> > > >> > > > > >> > > > > > > > > > > > configurations?
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an
> ending
> > > >> state
> > > >> > (we
> > > >> > > > > would
> > > >> > > > > >> > only
> > > >> > > > > >> > > > > > > transit to
> > > >> > > > > >> > > > > > > > > that
> > > >> > > > > >> > > > > > > > > > > state when the group is EMPTY and
> also all
> > > >> of
> > > >> > its
> > > >> > > > > >> > metadata
> > > >> > > > > >> > > are
> > > >> > > > > >> > > > > > > gone),
> > > >> > > > > >> > > > > > > > > once
> > > >> > > > > >> > > > > > > > > > > it's transited to DEAD this group
> would
> > > >> never
> > > >> > be
> > > >> > > > > >> revived.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 7. Will the choice to store the
> configs
> > > >> in
> > > >> > the
> > > >> > > > > group
> > > >> > > > > >> > > > > coordinator
> > > >> > > > > >> > > > > > > > > make it
> > > >> > > > > >> > > > > > > > > > > > harder to list all cluster configs
> and
> > > >> their
> > > >> > > > > values?
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > That's a good question, and our
> thoughts
> > > >> are
> > > >> > that
> > > >> > > > > the
> > > >> > > > > >> > > so-called
> > > >> > > > > >> > > > > > > "group
> > > >> > > > > >> > > > > > > > > > > configurations" are overrides of the
> > > >> > cluster-level
> > > >> > > > > >> > > > > configurations
> > > >> > > > > >> > > > > > > > > > > customized per group so when an
> admin list
> > > >> > cluster
> > > >> > > > > >> > configs
> > > >> > > > > >> > > it's
> > > >> > > > > >> > > > > > > okay to
> > > >> > > > > >> > > > > > > > > > > list just the cluster-level
> "defaults", not
> > > >> > showing
> > > >> > > > > >> any
> > > >> > > > > >> > > > > per-group
> > > >> > > > > >> > > > > > > > > > > customizations.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 8. How would someone configure a
> group
> > > >> > before
> > > >> > > > > >> starting
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > > > consumers? Have
> > > >> > > > > >> > > > > > > > > > > > we considered allowing the explicit
> > > >> > creation of
> > > >> > > > > >> groups?
> > > >> > > > > >> > > > > > > > > Alternatively, the
> > > >> > > > > >> > > > > > > > > > > > configs could be decoupled from
> the group
> > > >> > > > > lifecycle.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > The configs can be created before
> the group
> > > >> > itself
> > > >> > > > > as
> > > >> > > > > >> an
> > > >> > > > > >> > > > > > > independent
> > > >> > > > > >> > > > > > > > > entity
> > > >> > > > > >> > > > > > > > > > > --- of course, this requires the
> > > >> corresponding
> > > >> > > > > >> request to
> > > >> > > > > >> > > be
> > > >> > > > > >> > > > > > > routed to
> > > >> > > > > >> > > > > > > > > the
> > > >> > > > > >> > > > > > > > > > > right coordinator based on the group
> id ---
> > > >> > the only
> > > >> > > > > >> > thing
> > > >> > > > > >> > > that
> > > >> > > > > >> > > > > > > > > differs is,
> > > >> > > > > >> > > > > > > > > > > when the group itself is gone we
> also check
> > > >> > if there
> > > >> > > > > >> are
> > > >> > > > > >> > > any
> > > >> > > > > >> > > > > > > > > configuration
> > > >> > > > > >> > > > > > > > > > > entities related to that group and
> delete
> > > >> as
> > > >> > well.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> > > >> > asymmetry on
> > > >> > > > > the
> > > >> > > > > >> > > creation
> > > >> > > > > >> > > > > /
> > > >> > > > > >> > > > > > > > > deletion
> > > >> > > > > >> > > > > > > > > > > lifecycles of the config entities,
> and we
> > > >> > would like
> > > >> > > > > >> to
> > > >> > > > > >> > > hear
> > > >> > > > > >> > > > > > > everyone's
> > > >> > > > > >> > > > > > > > > > > feelings whether we should aim for
> symmetry
> > > >> > i.e.
> > > >> > > > > >> totally
> > > >> > > > > >> > > > > decouple
> > > >> > > > > >> > > > > > > group
> > > >> > > > > >> > > > > > > > > > > configs and hence not delete them at
> all
> > > >> when
> > > >> > the
> > > >> > > > > >> group
> > > >> > > > > >> > is
> > > >> > > > > >> > > > > gone,
> > > >> > > > > >> > > > > > > but
> > > >> > > > > >> > > > > > > > > always
> > > >> > > > > >> > > > > > > > > > > require explicit deletion operations
> by
> > > >> > themselves.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe
> method for
> > > >> > the Java
> > > >> > > > > >> > client
> > > >> > > > > >> > > > > still
> > > >> > > > > >> > > > > > > take
> > > >> > > > > >> > > > > > > > > a
> > > >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we
> have
> > > >> to
> > > >> > > > > >> introduce an
> > > >> > > > > >> > > > > overload?
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > I think we do not need to introduce
> an
> > > >> > overload, but
> > > >> > > > > >> I'm
> > > >> > > > > >> > > all
> > > >> > > > > >> > > > > ears
> > > >> > > > > >> > > > > > > if
> > > >> > > > > >> > > > > > > > > there
> > > >> > > > > >> > > > > > > > > > > may be some compatibility issues
> that we
> > > >> may
> > > >> > > > > overlook.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we
> should
> > > >> be
> > > >> > clearer
> > > >> > > > > >> > about
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > > > reason
> > > >> > > > > >> > > > > > > > > to
> > > >> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version
> from the
> > > >> > feature
> > > >> > > > > >> flag.
> > > >> > > > > >> > > Maybe
> > > >> > > > > >> > > > > we
> > > >> > > > > >> > > > > > > mean
> > > >> > > > > >> > > > > > > > > that we
> > > >> > > > > >> > > > > > > > > > > > can switch the default for the
> feature
> > > >> flag
> > > >> > to
> > > >> > > > > true
> > > >> > > > > >> > > based on
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > > > > metadata.version once we want to
> make it
> > > >> the
> > > >> > > > > >> default.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't
> mention
> > > >> > the
> > > >> > > > > >> required
> > > >> > > > > >> > > ACLs,
> > > >> > > > > >> > > > > it
> > > >> > > > > >> > > > > > > > > would be
> > > >> > > > > >> > > > > > > > > > > > good to add that for consistency.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > > >> > ConsumerGroupHeartbeat
> > > >> > > > > >> > requires
> > > >> > > > > >> > > > > "Read
> > > >> > > > > >> > > > > > > > > Group" even
> > > >> > > > > >> > > > > > > > > > > > though it seems to do more than
> reading.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > I had that thought myself as well,
> but in
> > > >> the
> > > >> > end we
> > > >> > > > > >> > could
> > > >> > > > > >> > > not
> > > >> > > > > >> > > > > > > find a
> > > >> > > > > >> > > > > > > > > > > better alternative: adding Write
> Group
> > > >> seems
> > > >> > an
> > > >> > > > > >> overkill
> > > >> > > > > >> > > here
> > > >> > > > > >> > > > > > > since we
> > > >> > > > > >> > > > > > > > > do
> > > >> > > > > >> > > > > > > > > > > not have it elsewhere (we only have
> Read /
> > > >> > Delete
> > > >> > > > > and
> > > >> > > > > >> > > Describe
> > > >> > > > > >> > > > > on
> > > >> > > > > >> > > > > > > > > groups so
> > > >> > > > > >> > > > > > > > > > > far). Would like to hear others
> thoughts.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 13. How is topic recreation
> handled by
> > > >> the
> > > >> > > > > consumer
> > > >> > > > > >> > with
> > > >> > > > > >> > > the
> > > >> > > > > >> > > > > new
> > > >> > > > > >> > > > > > > > > group
> > > >> > > > > >> > > > > > > > > > > > protocol? It would be good to have
> a
> > > >> > section on
> > > >> > > > > >> this.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > You mean with regex subscription
> right? Yes
> > > >> > we can
> > > >> > > > > >> add a
> > > >> > > > > >> > > > > section
> > > >> > > > > >> > > > > > > about
> > > >> > > > > >> > > > > > > > > > > that, but basically the idea is that
> > > >> consumer
> > > >> > would
> > > >> > > > > be
> > > >> > > > > >> > > totally
> > > >> > > > > >> > > > > > > > > agnostic in
> > > >> > > > > >> > > > > > > > > > > the new protocol as it's handled all
> by the
> > > >> > brokers.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write
> the
> > > >> new
> > > >> > > > > >> coordinator
> > > >> > > > > >> > in
> > > >> > > > > >> > > > > Java.
> > > >> > > > > >> > > > > > > Even
> > > >> > > > > >> > > > > > > > > though
> > > >> > > > > >> > > > > > > > > > > > this is an implementation detail,
> do we
> > > >> > plan to
> > > >> > > > > >> have a
> > > >> > > > > >> > > new
> > > >> > > > > >> > > > > gradle
> > > >> > > > > >> > > > > > > > > module
> > > >> > > > > >> > > > > > > > > > > > for it?
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > We have not thought about that. But
> I think
> > > >> > the
> > > >> > > > > answer
> > > >> > > > > >> > > should
> > > >> > > > > >> > > > > be
> > > >> > > > > >> > > > > > > yes.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal
> when it
> > > >> > comes to
> > > >> > > > > >> how
> > > >> > > > > >> > > many
> > > >> > > > > >> > > > > > > members
> > > >> > > > > >> > > > > > > > > the new
> > > >> > > > > >> > > > > > > > > > > > group protocol can support?
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > Within a group, I think we should
> shoot for
> > > >> > 1000s of
> > > >> > > > > >> > > members.
> > > >> > > > > >> > > > > But
> > > >> > > > > >> > > > > > > that
> > > >> > > > > >> > > > > > > > > > > scalability goals also depend on the
> offset
> > > >> > > > > management
> > > >> > > > > >> > > (commit,
> > > >> > > > > >> > > > > > > fetch)
> > > >> > > > > >> > > > > > > > > > > capabilities of the coordinator
> which we
> > > >> did
> > > >> > not
> > > >> > > > > >> cover in
> > > >> > > > > >> > > this
> > > >> > > > > >> > > > > > > KIP, so
> > > >> > > > > >> > > > > > > > > it's
> > > >> > > > > >> > > > > > > > > > > hard to give a number that applies
> > > >> > universally.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> > > >> > SubscribedTopidIds
> > > >> > > > > >> instead
> > > >> > > > > >> > of
> > > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > > >> > > > > >> > > Is the
> > > >> > > > > >> > > > > > > idea
> > > >> > > > > >> > > > > > > > > that
> > > >> > > > > >> > > > > > > > > > > > since we have to resolve the regex
> on the
> > > >> > server,
> > > >> > > > > we
> > > >> > > > > >> > can
> > > >> > > > > >> > > do
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > same
> > > >> > > > > >> > > > > > > > > for
> > > >> > > > > >> > > > > > > > > > > > the topic name? The difference is
> that
> > > >> > sending the
> > > >> > > > > >> > regex
> > > >> > > > > >> > > is
> > > >> > > > > >> > > > > more
> > > >> > > > > >> > > > > > > > > efficient
> > > >> > > > > >> > > > > > > > > > > > whereas sending the topic names is
> less
> > > >> > efficient.
> > > >> > > > > >> > > > > Furthermore,
> > > >> > > > > >> > > > > > > > > delete and
> > > >> > > > > >> > > > > > > > > > > > recreation is easier to handle if
> we have
> > > >> > topic
> > > >> > > > > ids.
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > The main reason to still let the
> clients
> > > >> send
> > > >> > names
> > > >> > > > > >> is to
> > > >> > > > > >> > > keep
> > > >> > > > > >> > > > > the
> > > >> > > > > >> > > > > > > > > > > reasoning of names -> ids on the
> broker /
> > > >> > admin
> > > >> > > > > client
> > > >> > > > > >> > > only.
> > > >> > > > > >> > > > > Note
> > > >> > > > > >> > > > > > > that
> > > >> > > > > >> > > > > > > > > > > although we added topic id in
> KIP-516, we
> > > >> > never
> > > >> > > > > >> > > implemented the
> > > >> > > > > >> > > > > > > logic
> > > >> > > > > >> > > > > > > > > on
> > > >> > > > > >> > > > > > > > > > > consumer/producers leveraging the
> related
> > > >> > newer
> > > >> > > > > >> versioned
> > > >> > > > > >> > > RPCs,
> > > >> > > > > >> > > > > > > > > instead we
> > > >> > > > > >> > > > > > > > > > > just set the topic id as empty UUID.
> We
> > > >> want
> > > >> > to keep
> > > >> > > > > >> the
> > > >> > > > > >> > > > > > > > > consumer/producer
> > > >> > > > > >> > > > > > > > > > > to be thin and only delegate the
> reasoning
> > > >> on
> > > >> > broker
> > > >> > > > > >> and
> > > >> > > > > >> > > > > > > potentially
> > > >> > > > > >> > > > > > > > > admin
> > > >> > > > > >> > > > > > > > > > > clients.
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > Thanks,
> > > >> > > > > >> > > > > > > > > > > > Ismael
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM
> David
> > > >> Jacot
> > > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > >> > > > > >> > > > > > > > > > > > wrote:
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > > Hi all,
> > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > > I would like to start a
> discussion
> > > >> thread
> > > >> > on
> > > >> > > > > >> KIP-848:
> > > >> > > > > >> > > The
> > > >> > > > > >> > > > > Next
> > > >> > > > > >> > > > > > > > > > > > > Generation of the Consumer
> Rebalance
> > > >> > Protocol.
> > > >> > > > > >> With
> > > >> > > > > >> > > this
> > > >> > > > > >> > > > > KIP,
> > > >> > > > > >> > > > > > > we
> > > >> > > > > >> > > > > > > > > aim
> > > >> > > > > >> > > > > > > > > > > > > to make the rebalance protocol
> (for
> > > >> > consumers)
> > > >> > > > > >> more
> > > >> > > > > >> > > > > reliable,
> > > >> > > > > >> > > > > > > more
> > > >> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
> > > >> > clients, and
> > > >> > > > > >> easier
> > > >> > > > > >> > > to
> > > >> > > > > >> > > > > debug
> > > >> > > > > >> > > > > > > for
> > > >> > > > > >> > > > > > > > > > > > > operators.
> > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > >> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > > Please take a look and let me
> know what
> > > >> > you
> > > >> > > > > think.
> > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > > Best,
> > > >> > > > > >> > > > > > > > > > > > > David
> > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > > > PS: I will be away from July
> 18th to
> > > >> > August 8th.
> > > >> > > > > >> That
> > > >> > > > > >> > > gives
> > > >> > > > > >> > > > > > > you a
> > > >> > > > > >> > > > > > > > > bit
> > > >> > > > > >> > > > > > > > > > > > > of time to read and digest this
> long
> > > >> KIP.
> > > >> > > > > >> > > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > >
> > > >> > > > > >> > > > > > > > > > > --
> > > >> > > > > >> > > > > > > > > > > -- Guozhang
> > > >> > > > > >> > > > > > > > >
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > >
> > > >> > > > > >> > > > > > > > --
> > > >> > > > > >> > > > > > > > -- Guozhang
> > > >> > > > > >> > > > > > >
> > > >> > > > > >> > > > >
> > > >> > > > > >> > >
> > > >> > > > > >> > >
> > > >> > > > > >> >
> > > >> > > > > >>
> > > >> > > > > >>
> > > >> > > > > >> --
> > > >> > > > > >> -- Guozhang
> > > >> > > > > >>
> > > >> > > > > >
> > > >> > > > >
> > > >> >
> > > >>
> > > >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi all,

The KIP states that we will re-implement the coordinator in Java. I
discussed this offline with a few folks and folks are concerned that
we could introduce many regressions in the old protocol if we do so.
Therefore, I am going to remove this statement from the KIP. It is an
implementation detail after all so it does not have to be decided at
this stage. We will likely start by trying to refactor the current
implementation as a first step.

Cheers,
David

On Mon, Aug 29, 2022 at 3:52 PM David Jacot <dj...@confluent.io> wrote:
>
> Hi Luke,
>
> > 1.1. I think the state machine are: "Empty, assigning, reconciling, stable,
> > dead" mentioned in Consumer Group States section, right?
>
> This sentence does not refer to those group states but rather to a
> state machine replication (SMR). This refers to the entire state of
> group coordinator which is replicated via the log layer. I will
> clarify this in the KIP.
>
> > 1.2. What do you mean "each state machine is modelled as an event loop"?
>
> The idea is to follow a model similar to the new quorum controller. We
> will have N threads to process events. Each __consumer_offsets
> partition is assigned to a unique thread and all the events (e.g.
> requests, callbacks, etc.) are processed by this thread. This simplify
> concurrency and will enable us to do simulation testing for the group
> coordinator.
>
> > 1.3. Why do we need a state machine per *__consumer_offsets* partitions?
> > Not a state machine "per consumer group" owned by a group coordinator? For
> > example, if one group coordinator owns 2 consumer groups, and both exist in
> > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
>
> See 1.1. The confusion comes from there, I think.
>
> > 1.4. I know the "*group.coordinator.threads" *is the number of threads used
> > to run the state machines. But I'm wondering if the purpose of the threads
> > is only to keep the state of each consumer group (or *__consumer_offsets*
> > partitions?), and no heavy computation, why should we need multi-threads
> > here?
>
> See 1.2. The idea is to have an ability to shard the processing as the
> computation could be heavy.
>
> > 2.1. The consumer session timeout, why does the default session timeout not
> > locate between min (45s) and max(60s)? I thought the min/max session
> > timeout is to define lower/upper bound of it, no?
> >
> > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > failures when using the consumer group protocol.
> > group.consumer.min.session.timeout.ms int 45s The minimum session timeout.
> > group.consumer.max.session.timeout.ms int 60s The maximum session timeout.
>
> This is indeed a mistake. The default session timeout should be 45s
> (the current default).
>
> > 2.2. The default server side assignor are [range, uniform], which means
> > we'll default to "range" assignor. I'd like to know why not uniform one? I
> > thought usually users will choose uniform assignor (former sticky assinor)
> > for better evenly distribution. Any other reason we choose range assignor
> > as default?
> > group.consumer.assignors List range, uniform The server side assignors.
>
> The order on the server side has no influence because the client must
> chose the selector that he wants to use. There is no default in the
> current proposal. If the assignor is not specified by the client, the
> request is rejected. The default client value for
> `group.remote.assignor` is `uniform` though.
>
> Thanks for your very good comments, Luke. I hope that my answers help
> to clarify things. I will update the KIP as well based on your
> feedback.
>
> Cheers,
> David
>
> On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com> wrote:
> >
> > Hi David,
> >
> > Thanks for the update.
> >
> > Some more questions:
> > 1. In Group Coordinator section, you mentioned:
> > > The new group coordinator will have a state machine per
> > *__consumer_offsets* partitions, where each state machine is modelled as an
> > event loop. Those state machines will be executed in
> > *group.coordinator.threads* threads.
> >
> > 1.1. I think the state machine are: "Empty, assigning, reconciling, stable,
> > dead" mentioned in Consumer Group States section, right?
> > 1.2. What do you mean "each state machine is modelled as an event loop"?
> > 1.3. Why do we need a state machine per *__consumer_offsets* partitions?
> > Not a state machine "per consumer group" owned by a group coordinator? For
> > example, if one group coordinator owns 2 consumer groups, and both exist in
> > *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> > 1.4. I know the "*group.coordinator.threads" *is the number of threads used
> > to run the state machines. But I'm wondering if the purpose of the threads
> > is only to keep the state of each consumer group (or *__consumer_offsets*
> > partitions?), and no heavy computation, why should we need multi-threads
> > here?
> >
> > 2. For the default value in the new configs:
> > 2.1. The consumer session timeout, why does the default session timeout not
> > locate between min (45s) and max(60s)? I thought the min/max session
> > timeout is to define lower/upper bound of it, no?
> >
> > group.consumer.session.timeout.ms int 30s The timeout to detect client
> > failures when using the consumer group protocol.
> > group.consumer.min.session.timeout.ms int 45s The minimum session timeout.
> > group.consumer.max.session.timeout.ms int 60s The maximum session timeout.
> >
> >
> >
> > 2.2. The default server side assignor are [range, uniform], which means
> > we'll default to "range" assignor. I'd like to know why not uniform one? I
> > thought usually users will choose uniform assignor (former sticky assinor)
> > for better evenly distribution. Any other reason we choose range assignor
> > as default?
> > group.consumer.assignors List range, uniform The server side assignors.
> >
> >
> >
> >
> >
> >
> > Thank you.
> > Luke
> >
> >
> >
> >
> >
> >
> > On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com> wrote:
> >
> > > Hi Sagar,
> > >
> > > I have some thoughts about Kafka Connect integrating with KIP-848, but I
> > > think we should have a separate discussion thread for the Kafka Connect
> > > KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol [1],
> > > and let this discussion thread focus on consumer rebalance protocol, WDYT?
> > >
> > > [1]
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >
> > > Thank you.
> > > Luke
> > >
> > >
> > >
> > > On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com> wrote:
> > >
> > >> Thank you Guozhang/David for the feedback. Looks like there's agreement on
> > >> using separate APIs for Connect. I would revisit the doc and see what
> > >> changes are to be made.
> > >>
> > >> Thanks!
> > >> Sagar.
> > >>
> > >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot <dj...@confluent.io.invalid>
> > >> wrote:
> > >>
> > >> > Hi Sagar,
> > >> >
> > >> > Thanks for the feedback and the document. That's really helpful. I
> > >> > will take a look at it.
> > >> >
> > >> > Overall, it seems to me that both Connect and the Consumer could share
> > >> > the same underlying "engine". The main difference is that the Consumer
> > >> > assigns topic-partitions to members whereas Connect assigns tasks to
> > >> > workers. I see two ways to move forward:
> > >> > 1) We extend the new proposed APIs to support different resource types
> > >> > (e.g. partitions, tasks, etc.); or
> > >> > 2) We use new dedicated APIs for Connect. The dedicated APIs would be
> > >> > similar to the new ones but different on the content/resources and
> > >> > they would rely on the same engine on the coordinator side.
> > >> >
> > >> > I personally lean towards 2) because I am not a fan of overcharging
> > >> > APIs to serve different purposes. That being said, I am not opposed to
> > >> > 1) if we can find an elegant way to do it.
> > >> >
> > >> > I think that we can continue to discuss it here for now in order to
> > >> > ensure that this KIP is compatible with what we will do for Connect in
> > >> > the future.
> > >> >
> > >> > Best,
> > >> > David
> > >> >
> > >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io> wrote:
> > >> > >
> > >> > > Hi all,
> > >> > >
> > >> > > I am back from vacation. I will go through and address your comments
> > >> > > in the coming days. Thanks for your feedback.
> > >> > >
> > >> > > Cheers,
> > >> > > David
> > >> > >
> > >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gharris1727@gmail.com
> > >> >
> > >> > wrote:
> > >> > > >
> > >> > > > Hey All!
> > >> > > >
> > >> > > > Thanks for the KIP, it's wonderful to see cooperative rebalancing
> > >> > making it
> > >> > > > down the stack!
> > >> > > >
> > >> > > > I had a few questions:
> > >> > > >
> > >> > > > 1. The 'Rejected Alternatives' section describes how member epoch
> > >> > should
> > >> > > > advance in step with the group epoch and assignment epoch values. I
> > >> > think
> > >> > > > that this is a good idea for the reasons described in the KIP. When
> > >> the
> > >> > > > protocol is incrementally assigning partitions to a worker, what
> > >> member
> > >> > > > epoch does each incremental assignment use? Are member epochs
> > >> re-used,
> > >> > and
> > >> > > > a single member epoch can correspond to multiple different
> > >> > (monotonically
> > >> > > > larger) assignments?
> > >> > > >
> > >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> > >> coordinator? If
> > >> > > > not, should custom client-side assignor implementations interact
> > >> with
> > >> > the
> > >> > > > Reason field, and how is its common meaning agreed upon? If so, what
> > >> > is the
> > >> > > > benefit of a distinct Reason field over including such functionality
> > >> > in the
> > >> > > > opaque metadata?
> > >> > > >
> > >> > > > 3. The following is included in the KIP: "Thanks to this, the input
> > >> of
> > >> > the
> > >> > > > client side assignor is entirely driven by the group coordinator.
> > >> The
> > >> > > > consumer is no longer responsible for maintaining any state besides
> > >> its
> > >> > > > assigned partitions." Does this mean that the client-side assignor
> > >> MAY
> > >> > > > incorporate additional non-Metadata state (such as partition
> > >> > throughput,
> > >> > > > cpu/memory metrics, config topics, etc), or that additional
> > >> > non-Metadata
> > >> > > > state SHOULD NOT be used?
> > >> > > >
> > >> > > > 4. I see that there are separate classes
> > >> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> > >> > > > overlap significantly. Is it possible for these two implementations
> > >> to
> > >> > be
> > >> > > > unified? This would serve to promote feature parity of server-side
> > >> and
> > >> > > > client-side assignors, and would also facilitate operational
> > >> > flexibility in
> > >> > > > certain situations. For example, if a server-side assignor has some
> > >> > poor
> > >> > > > behavior and needs a patch, deploying the patched assignor to the
> > >> > client
> > >> > > > and switching one consumer group to a client-side assignor may be
> > >> > faster
> > >> > > > and less risky than patching all of the brokers. With the currently
> > >> > > > proposed distinct APIs, a non-trivial reimplementation would have
> > >> to be
> > >> > > > assembled, and if the two APIs have diverged significantly, then it
> > >> is
> > >> > > > possible that a reimplementation would not be possible.
> > >> > > >
> > >> > > > --
> > >> > > > Greg Harris
> > >> > > > gharris1727@gmail.com
> > >> > > > github.com/gharris1727
> > >> > > >
> > >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com>
> > >> > wrote:
> > >> > > >
> > >> > > > > Hi Guozhang/David,
> > >> > > > >
> > >> > > > > I created a confluence page to discuss how Connect would need to
> > >> > change
> > >> > > > > based on the new rebalance protocol. Here's the page:
> > >> > > > >
> > >> > > > >
> > >> >
> > >> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >> > > > >
> > >> > > > > It's also pretty longish and I have tried to keep a format
> > >> similar to
> > >> > > > > KIP-848. Let me know what you think. Also, do you think this
> > >> should
> > >> > be
> > >> > > > > moved to a separate discussion thread or is this one fine?
> > >> > > > >
> > >> > > > > Thanks!
> > >> > > > > Sagar.
> > >> > > > >
> > >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com>
> > >> > wrote:
> > >> > > > >
> > >> > > > > > Hello Guozhang,
> > >> > > > > >
> > >> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I would do
> > >> > the
> > >> > > > > > analysis and come up with such a document.
> > >> > > > > >
> > >> > > > > > Thanks!
> > >> > > > > > Sagar.
> > >> > > > > >
> > >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> > >> wangguoz@gmail.com>
> > >> > > > > wrote:
> > >> > > > > >
> > >> > > > > >> Hello Sagar,
> > >> > > > > >>
> > >> > > > > >> It would be great if you could come back with some analysis on
> > >> > how to
> > >> > > > > >> implement the Connect side integration with the new protocol;
> > >> so
> > >> > far
> > >> > > > > >> besides leveraging on the new "protocol type" we did not yet
> > >> think
> > >> > > > > through
> > >> > > > > >> the Connect side implementations. For Streams here's a draft of
> > >> > > > > >> integration
> > >> > > > > >> plan:
> > >> > > > > >>
> > >> > > > > >>
> > >> > > > >
> > >> >
> > >> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > >> > > > > >> just FYI for your analysis on Connect.
> > >> > > > > >>
> > >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> > >> sagarmeansocean@gmail.com
> > >> > >
> > >> > > > > wrote:
> > >> > > > > >>
> > >> > > > > >> > Hi David,
> > >> > > > > >> >
> > >> > > > > >> > Thank you for your response. The reason I thought connect can
> > >> > also fit
> > >> > > > > >> into
> > >> > > > > >> > this new scheme is that even today the connect uses a
> > >> > > > > WorkerCoordinator
> > >> > > > > >> > extending from AbstractCoordinator to empower rebalances of
> > >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> > >> protocolType()
> > >> > to
> > >> > > > > >> connect
> > >> > > > > >> > and uses the metadata() method by plumbing into
> > >> > > > > >> JoinGroupRequestProtocol.
> > >> > > > > >> >
> > >> > > > > >> > I think the changes to support connect would be similar at a
> > >> > high
> > >> > > > > level
> > >> > > > > >> to
> > >> > > > > >> > the changes in streams mainly because of the Client side
> > >> > assignors
> > >> > > > > being
> > >> > > > > >> > used in both. At an implementation level, we might need to
> > >> make
> > >> > a lot
> > >> > > > > of
> > >> > > > > >> > changes to get onto this new assignment protocol like
> > >> enhancing
> > >> > the
> > >> > > > > >> > JoinGroup request/response and SyncGroup and using
> > >> > > > > >> ConsumerGroupHeartbeat
> > >> > > > > >> > API etc again on similar lines to streams (or there might be
> > >> > > > > >> deviations). I
> > >> > > > > >> > would try to perform a detailed analysis of the same  and we
> > >> > can have
> > >> > > > > a
> > >> > > > > >> > separate discussion thread for that as that would derail this
> > >> > > > > discussion
> > >> > > > > >> > thread. Let me know if that sounds good to you.
> > >> > > > > >> >
> > >> > > > > >> > Thanks!
> > >> > > > > >> > Sagar.
> > >> > > > > >> >
> > >> > > > > >> >
> > >> > > > > >> >
> > >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > >> > > > > <djacot@confluent.io.invalid
> > >> > > > > >> >
> > >> > > > > >> > wrote:
> > >> > > > > >> >
> > >> > > > > >> > > Hi Sagar,
> > >> > > > > >> > >
> > >> > > > > >> > > Thanks for your comments.
> > >> > > > > >> > >
> > >> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can
> > >> > mention it.
> > >> > > > > >> > >
> > >> > > > > >> > > 2) The idea is to transition C from his current assignment
> > >> to
> > >> > his
> > >> > > > > >> > > target assignment when he can move to epoch 3. When that
> > >> > happens,
> > >> > > > > the
> > >> > > > > >> > > member assignment is updated and persisted with all its
> > >> > assigned
> > >> > > > > >> > > partitions even if they are not all revoked yet. In other
> > >> > words, the
> > >> > > > > >> > > member assignment becomes the target assignment. This is
> > >> > basically
> > >> > > > > an
> > >> > > > > >> > > optimization to avoid having to write all the changes to
> > >> the
> > >> > log.
> > >> > > > > The
> > >> > > > > >> > > examples are based on the persisted state so I understand
> > >> the
> > >> > > > > >> > > confusion. Let me see if I can improve this in the
> > >> > description.
> > >> > > > > >> > >
> > >> > > > > >> > > 3) Regarding Connect, it could reuse the protocol with a
> > >> > client side
> > >> > > > > >> > > assignor if it fits in the protocol. The assignment is
> > >> about
> > >> > > > > >> > > topicid-partitions + metadata, could Connect fit into this?
> > >> > > > > >> > >
> > >> > > > > >> > > Best,
> > >> > > > > >> > > David
> > >> > > > > >> > >
> > >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > >> > sagarmeansocean@gmail.com>
> > >> > > > > >> wrote:
> > >> > > > > >> > > >
> > >> > > > > >> > > > Hi David,
> > >> > > > > >> > > >
> > >> > > > > >> > > > Thanks for the KIP. I just had minor observations:
> > >> > > > > >> > > >
> > >> > > > > >> > > > 1) In the Assignment Error section in Client Side mode
> > >> > Assignment
> > >> > > > > >> > > process,
> > >> > > > > >> > > > you mentioned => `In this case, the client side assignor
> > >> can
> > >> > > > > return
> > >> > > > > >> an
> > >> > > > > >> > > > error to the group coordinator`. In this case are you
> > >> > referring to
> > >> > > > > >> the
> > >> > > > > >> > > > Assignor returning an AssignmentError that's listed down
> > >> > towards
> > >> > > > > the
> > >> > > > > >> > end?
> > >> > > > > >> > > > If yes, do you think it would make sense to mention this
> > >> > > > > explicitly
> > >> > > > > >> > here?
> > >> > > > > >> > > >
> > >> > > > > >> > > > 2) In the Case Studies section, I have a slight
> > >> confusion,
> > >> > not
> > >> > > > > sure
> > >> > > > > >> if
> > >> > > > > >> > > > others have the same. Consider this step:
> > >> > > > > >> > > >
> > >> > > > > >> > > > When B heartbeats, the group coordinator transitions him
> > >> to
> > >> > epoch
> > >> > > > > 3
> > >> > > > > >> > > because
> > >> > > > > >> > > > B has no partitions to revoke. It persists the change and
> > >> > reply.
> > >> > > > > >> > > >
> > >> > > > > >> > > >    - Group (epoch=3)
> > >> > > > > >> > > >       - A
> > >> > > > > >> > > >       - B
> > >> > > > > >> > > >       - C
> > >> > > > > >> > > >    - Target Assignment (epoch=3)
> > >> > > > > >> > > >       - A - partitions=[foo-0]
> > >> > > > > >> > > >       - B - partitions=[foo-2]
> > >> > > > > >> > > >       - C - partitions=[foo-1]
> > >> > > > > >> > > >    - Member Assignment
> > >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > >> > > > > >> > > >
> > >> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot
> > >> get
> > >> > foo-1
> > >> > > > > >> yet.
> > >> > > > > >> > > >
> > >> > > > > >> > > > Here,it's mentioned that member C can't get the foo-1
> > >> > partition
> > >> > > > > yet,
> > >> > > > > >> > but
> > >> > > > > >> > > > based on the description above, it seems it already has
> > >> it.
> > >> > Do you
> > >> > > > > >> > think
> > >> > > > > >> > > it
> > >> > > > > >> > > > would be better to remove it and populate it only when it
> > >> > actually
> > >> > > > > >> gets
> > >> > > > > >> > > it?
> > >> > > > > >> > > > I see this in a lot of other places, so have I
> > >> understood it
> > >> > > > > >> > incorrectly
> > >> > > > > >> > > ?
> > >> > > > > >> > > >
> > >> > > > > >> > > >
> > >> > > > > >> > > > Regarding connect , it might be out of scope of this
> > >> > discussion,
> > >> > > > > but
> > >> > > > > >> > from
> > >> > > > > >> > > > what I understood it would probably be running in client
> > >> > side
> > >> > > > > >> assignor
> > >> > > > > >> > > mode
> > >> > > > > >> > > > even on the new rebalance protocol as it has its own
> > >> Custom
> > >> > > > > >> > > Assignors(Eager
> > >> > > > > >> > > > and IncrementalCooperative).
> > >> > > > > >> > > >
> > >> > > > > >> > > > Thanks!
> > >> > > > > >> > > >
> > >> > > > > >> > > > Sagar.
> > >> > > > > >> > > >
> > >> > > > > >> > > >
> > >> > > > > >> > > >
> > >> > > > > >> > > >
> > >> > > > > >> > > >
> > >> > > > > >> > > >
> > >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > >> > > > > >> > <djacot@confluent.io.invalid
> > >> > > > > >> > > >
> > >> > > > > >> > > > wrote:
> > >> > > > > >> > > >
> > >> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
> > >> > specialized API
> > >> > > > > >> > > > > instead of relying on one generic API. For Connect, we
> > >> > can apply
> > >> > > > > >> the
> > >> > > > > >> > > > > exact same pattern and reuse/share the core
> > >> > implementation on
> > >> > > > > the
> > >> > > > > >> > > > > server side. For the schema registry, I think that we
> > >> > should
> > >> > > > > >> consider
> > >> > > > > >> > > > > having a tailored API to do simple membership/leader
> > >> > election.
> > >> > > > > >> > > > >
> > >> > > > > >> > > > > Best,
> > >> > > > > >> > > > > David
> > >> > > > > >> > > > >
> > >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> > >> > ismael@juma.me.uk
> > >> > > > > >
> > >> > > > > >> > > wrote:
> > >> > > > > >> > > > > >
> > >> > > > > >> > > > > > Three quick comments:
> > >> > > > > >> > > > > >
> > >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > >> > > > > >> com.google.re2j.Pattern, we
> > >> > > > > >> > > > > should
> > >> > > > > >> > > > > > document the differences in more detail before
> > >> deciding
> > >> > one
> > >> > > > > way
> > >> > > > > >> or
> > >> > > > > >> > > > > another.
> > >> > > > > >> > > > > > That said, if people pass java.util.regex.Pattern,
> > >> they
> > >> > expect
> > >> > > > > >> > their
> > >> > > > > >> > > > > > semantics to be honored. If we are doing something
> > >> > different,
> > >> > > > > >> then
> > >> > > > > >> > we
> > >> > > > > >> > > > > > should consider adding an overload with our own
> > >> Pattern
> > >> > class
> > >> > > > > (I
> > >> > > > > >> > > don't
> > >> > > > > >> > > > > > think we'd want to expose re2j's at this point).
> > >> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > >> > > > > integrate
> > >> > > > > >> > fully
> > >> > > > > >> > > > > with
> > >> > > > > >> > > > > > it and should handle the topic recreation case
> > >> > correctly.
> > >> > > > > That's
> > >> > > > > >> > the
> > >> > > > > >> > > main
> > >> > > > > >> > > > > > part we need to handle. I agree with David that we'd
> > >> > want to
> > >> > > > > add
> > >> > > > > >> > > topic
> > >> > > > > >> > > > > ids
> > >> > > > > >> > > > > > to the relevant protocols that don't have it yet and
> > >> > that we
> > >> > > > > can
> > >> > > > > >> > > probably
> > >> > > > > >> > > > > > focus on the internals versus adding new APIs to the
> > >> > Java
> > >> > > > > >> Consumer
> > >> > > > > >> > > > > (unless
> > >> > > > > >> > > > > > we find that adding new APIs is required for
> > >> reasonable
> > >> > > > > >> semantics).
> > >> > > > > >> > > > > > 3. I am still not sure about the coordinator storing
> > >> the
> > >> > > > > >> configs.
> > >> > > > > >> > > It's
> > >> > > > > >> > > > > > powerful for configs to be centralized in the
> > >> metadata
> > >> > log for
> > >> > > > > >> > > various
> > >> > > > > >> > > > > > reasons (auditability, visibility, consistency,
> > >> etc.).
> > >> > > > > >> Similarly, I
> > >> > > > > >> > > am
> > >> > > > > >> > > > > not
> > >> > > > > >> > > > > > sure about automatically deleting configs in a way
> > >> that
> > >> > they
> > >> > > > > >> cannot
> > >> > > > > >> > > be
> > >> > > > > >> > > > > > recovered. A good property for modern systems is to
> > >> > minimize
> > >> > > > > the
> > >> > > > > >> > > number
> > >> > > > > >> > > > > of
> > >> > > > > >> > > > > > unrecoverable data loss scenarios.
> > >> > > > > >> > > > > >
> > >> > > > > >> > > > > > Ismael
> > >> > > > > >> > > > > >
> > >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > >> > > > > >> > > <djacot@confluent.io.invalid
> > >> > > > > >> > > > > >
> > >> > > > > >> > > > > > wrote:
> > >> > > > > >> > > > > >
> > >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > > 1) the migration path, especially the last step
> > >> when
> > >> > > > > clients
> > >> > > > > >> > > flip the
> > >> > > > > >> > > > > > > flag
> > >> > > > > >> > > > > > > > to enable the new protocol, in which we would
> > >> have a
> > >> > > > > window
> > >> > > > > >> > where
> > >> > > > > >> > > > > both
> > >> > > > > >> > > > > > > new
> > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
> > >> used
> > >> > by
> > >> > > > > >> members
> > >> > > > > >> > of
> > >> > > > > >> > > the
> > >> > > > > >> > > > > same
> > >> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> > >> > behavior
> > >> > > > > >> while
> > >> > > > > >> > > > > using the
> > >> > > > > >> > > > > > > > new protocol is something we need to present
> > >> about.
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > Noted. I just published a new version of KIP which
> > >> > includes
> > >> > > > > >> more
> > >> > > > > >> > > > > > > details about this. See the "Supporting Online
> > >> > Consumer
> > >> > > > > Group
> > >> > > > > >> > > Upgrade"
> > >> > > > > >> > > > > > > and the "Compatibility, Deprecation, and Migration
> > >> > Plan". I
> > >> > > > > >> think
> > >> > > > > >> > > that
> > >> > > > > >> > > > > > > I have to think through a few cases now but the
> > >> > overall idea
> > >> > > > > >> and
> > >> > > > > >> > > > > > > mechanism should be understandable.
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> > >> > topic ids
> > >> > > > > >> are
> > >> > > > > >> > > only
> > >> > > > > >> > > > > used
> > >> > > > > >> > > > > > > as
> > >> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> > >> > exposed
> > >> > > > > via
> > >> > > > > >> any
> > >> > > > > >> > > > > public
> > >> > > > > >> > > > > > > APIs
> > >> > > > > >> > > > > > > > to consumers yet. I think the question is, first
> > >> > should we
> > >> > > > > >> let
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > consumer
> > >> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> > >> > itself
> > >> > > > > to
> > >> > > > > >> > fully
> > >> > > > > >> > > > > > > leverage
> > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> > >> RPCs
> > >> > with
> > >> > > > > the
> > >> > > > > >> > > topic
> > >> > > > > >> > > > > ids;
> > >> > > > > >> > > > > > > and
> > >> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> > >> > topic ids
> > >> > > > > in
> > >> > > > > >> the
> > >> > > > > >> > > > > consumer
> > >> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> > >> well
> > >> > as in
> > >> > > > > >> the
> > >> > > > > >> > > > > rebalance
> > >> > > > > >> > > > > > > > listener for cases like topic
> > >> > deletion-and-recreation).
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > a) Assuming that we would include converting all
> > >> the
> > >> > offsets
> > >> > > > > >> > > related
> > >> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer
> > >> > would be
> > >> > > > > >> able
> > >> > > > > >> > to
> > >> > > > > >> > > > > > > fully operate with topic ids. That being said, it
> > >> > still has
> > >> > > > > to
> > >> > > > > >> > > provide
> > >> > > > > >> > > > > > > the topics names in various APIs so having a
> > >> mapping
> > >> > in the
> > >> > > > > >> > > consumer
> > >> > > > > >> > > > > > > seems inevitable to me.
> > >> > > > > >> > > > > > > b) I don't have a strong opinion on this. Here I
> > >> > wonder if
> > >> > > > > >> this
> > >> > > > > >> > > goes
> > >> > > > > >> > > > > > > beyond the scope of this KIP. I would rather focus
> > >> on
> > >> > the
> > >> > > > > >> > internals
> > >> > > > > >> > > > > > > here and we can consider this separately if we see
> > >> > value in
> > >> > > > > >> doing
> > >> > > > > >> > > it.
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > Coming back to Ismael's point about using topic ids
> > >> > in the
> > >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there
> > >> is
> > >> > one
> > >> > > > > >> > advantage
> > >> > > > > >> > > in
> > >> > > > > >> > > > > > > favour of it. The consumer will have the
> > >> opportunity
> > >> > to
> > >> > > > > >> validate
> > >> > > > > >> > > that
> > >> > > > > >> > > > > > > the topics exists before passing them into the
> > >> group
> > >> > > > > rebalance
> > >> > > > > >> > > > > > > protocol. Obviously, the coordinator will also
> > >> notice
> > >> > it but
> > >> > > > > >> it
> > >> > > > > >> > > does
> > >> > > > > >> > > > > > > not really have a way to reject an invalid topic in
> > >> > the
> > >> > > > > >> response.
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > >> questions
> > >> > > > > except
> > >> > > > > >> for
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> > >> > it's not
> > >> > > > > >> > > necessary
> > >> > > > > >> > > > > to
> > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> > >> > instead we
> > >> > > > > >> still
> > >> > > > > >> > > use
> > >> > > > > >> > > > > > > Pattern
> > >> > > > > >> > > > > > > > while just documenting that our subscription may
> > >> be
> > >> > > > > >> rejected by
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > server.
> > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > >> scenario
> > >> > I felt
> > >> > > > > >> > using
> > >> > > > > >> > > an
> > >> > > > > >> > > > > > > > overloaded `String` based subscription may be
> > >> more
> > >> > > > > >> vulnerable
> > >> > > > > >> > to
> > >> > > > > >> > > > > various
> > >> > > > > >> > > > > > > > invalid regexes.
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > That could work. I have to look at the differences
> > >> > between
> > >> > > > > the
> > >> > > > > >> > two
> > >> > > > > >> > > > > > > engines to better understand the potential issues.
> > >> My
> > >> > > > > >> > > understanding is
> > >> > > > > >> > > > > > > that would work for all the basic regular
> > >> > expressions. The
> > >> > > > > >> > > differences
> > >> > > > > >> > > > > > > between the two are mainly about the various
> > >> character
> > >> > > > > >> classes. I
> > >> > > > > >> > > > > > > wonder what other people think about this.
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > Best,
> > >> > > > > >> > > > > > > David
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > >> > > > > >> > wangguoz@gmail.com
> > >> > > > > >> > > >
> > >> > > > > >> > > > > wrote:
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > Thanks David! I think on the high level there are
> > >> > two meta
> > >> > > > > >> > > points we
> > >> > > > > >> > > > > need
> > >> > > > > >> > > > > > > > to concretize a bit more:
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > 1) the migration path, especially the last step
> > >> when
> > >> > > > > clients
> > >> > > > > >> > > flip the
> > >> > > > > >> > > > > > > flag
> > >> > > > > >> > > > > > > > to enable the new protocol, in which we would
> > >> have a
> > >> > > > > window
> > >> > > > > >> > where
> > >> > > > > >> > > > > both
> > >> > > > > >> > > > > > > new
> > >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
> > >> used
> > >> > by
> > >> > > > > >> members
> > >> > > > > >> > of
> > >> > > > > >> > > the
> > >> > > > > >> > > > > same
> > >> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> > >> > behavior
> > >> > > > > >> while
> > >> > > > > >> > > > > using the
> > >> > > > > >> > > > > > > > new protocol is something we need to present
> > >> about.
> > >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> > >> > topic ids
> > >> > > > > >> are
> > >> > > > > >> > > only
> > >> > > > > >> > > > > used
> > >> > > > > >> > > > > > > as
> > >> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> > >> > exposed
> > >> > > > > via
> > >> > > > > >> any
> > >> > > > > >> > > > > public
> > >> > > > > >> > > > > > > APIs
> > >> > > > > >> > > > > > > > to consumers yet. I think the question is, first
> > >> > should we
> > >> > > > > >> let
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > consumer
> > >> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> > >> > itself
> > >> > > > > to
> > >> > > > > >> > fully
> > >> > > > > >> > > > > > > leverage
> > >> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> > >> RPCs
> > >> > with
> > >> > > > > the
> > >> > > > > >> > > topic
> > >> > > > > >> > > > > ids;
> > >> > > > > >> > > > > > > and
> > >> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> > >> > topic ids
> > >> > > > > in
> > >> > > > > >> the
> > >> > > > > >> > > > > consumer
> > >> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> > >> well
> > >> > as in
> > >> > > > > >> the
> > >> > > > > >> > > > > rebalance
> > >> > > > > >> > > > > > > > listener for cases like topic
> > >> > deletion-and-recreation).
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> > >> questions
> > >> > > > > except
> > >> > > > > >> for
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> > >> > it's not
> > >> > > > > >> > > necessary
> > >> > > > > >> > > > > to
> > >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> > >> > instead we
> > >> > > > > >> still
> > >> > > > > >> > > use
> > >> > > > > >> > > > > > > Pattern
> > >> > > > > >> > > > > > > > while just documenting that our subscription may
> > >> be
> > >> > > > > >> rejected by
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > server.
> > >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> > >> scenario
> > >> > I felt
> > >> > > > > >> > using
> > >> > > > > >> > > an
> > >> > > > > >> > > > > > > > overloaded `String` based subscription may be
> > >> more
> > >> > > > > >> vulnerable
> > >> > > > > >> > to
> > >> > > > > >> > > > > various
> > >> > > > > >> > > > > > > > invalid regexes.
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > Guozhang
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > >> > > > > >> > > > > <djacot@confluent.io.invalid
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > wrote:
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > > Hi Ismael,
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer your
> > >> > questions
> > >> > > > > >> > inline.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 1. I think it's premature to talk about
> > >> target
> > >> > > > > versions
> > >> > > > > >> for
> > >> > > > > >> > > > > > > deprecation
> > >> > > > > >> > > > > > > > > and
> > >> > > > > >> > > > > > > > > > removal of the existing group protocol.
> > >> Unlike
> > >> > KRaft,
> > >> > > > > >> this
> > >> > > > > >> > > > > affects a
> > >> > > > > >> > > > > > > core
> > >> > > > > >> > > > > > > > > > client protocol and hence deprecation/removal
> > >> > will be
> > >> > > > > >> > heavily
> > >> > > > > >> > > > > > > dependent
> > >> > > > > >> > > > > > > > > on
> > >> > > > > >> > > > > > > > > > how quickly applications migrate to the new
> > >> > protocol.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 2. The KIP says we intend to release this in
> > >> > 4.x, but
> > >> > > > > it
> > >> > > > > >> > > wasn't
> > >> > > > > >> > > > > made
> > >> > > > > >> > > > > > > > > clear
> > >> > > > > >> > > > > > > > > > why. If we added that as a way to estimate
> > >> when
> > >> > we'd
> > >> > > > > >> > > deprecate
> > >> > > > > >> > > > > and
> > >> > > > > >> > > > > > > remove
> > >> > > > > >> > > > > > > > > > the group protocol, I also suggest removing
> > >> > this part.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Let me explain my reasoning. As explained, I
> > >> plan
> > >> > to
> > >> > > > > >> rewrite
> > >> > > > > >> > > the
> > >> > > > > >> > > > > group
> > >> > > > > >> > > > > > > > > coordinator in Java while we implement the new
> > >> > protocol.
> > >> > > > > >> This
> > >> > > > > >> > > means
> > >> > > > > >> > > > > > > > > that the internals will be slightly different
> > >> > (e.g.
> > >> > > > > >> threading
> > >> > > > > >> > > > > model).
> > >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the switch from
> > >> > the old
> > >> > > > > >> group
> > >> > > > > >> > > > > > > > > coordinator to the new group coordinator to a
> > >> > major
> > >> > > > > >> release.
> > >> > > > > >> > > The
> > >> > > > > >> > > > > > > > > alternative would be to use a flag to do the
> > >> > switch
> > >> > > > > >> instead
> > >> > > > > >> > of
> > >> > > > > >> > > > > relying
> > >> > > > > >> > > > > > > > > on the software upgrade.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 3. We need to flesh out the details of the
> > >> > migration
> > >> > > > > >> story.
> > >> > > > > >> > > It
> > >> > > > > >> > > > > sounds
> > >> > > > > >> > > > > > > > > like
> > >> > > > > >> > > > > > > > > > we're saying we will support online
> > >> migrations.
> > >> > Is
> > >> > > > > that
> > >> > > > > >> > > correct?
> > >> > > > > >> > > > > We
> > >> > > > > >> > > > > > > > > should
> > >> > > > > >> > > > > > > > > > explain this in detail. It could also be done
> > >> > as a
> > >> > > > > >> separate
> > >> > > > > >> > > KIP,
> > >> > > > > >> > > > > if
> > >> > > > > >> > > > > > > it's
> > >> > > > > >> > > > > > > > > > easier.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Yes, we will support online migrations for the
> > >> > group.
> > >> > > > > That
> > >> > > > > >> > > means
> > >> > > > > >> > > > > that
> > >> > > > > >> > > > > > > > > a group using the old protocol will be able to
> > >> > switch to
> > >> > > > > >> the
> > >> > > > > >> > > new
> > >> > > > > >> > > > > > > > > protocol.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Let me briefly explain how that will work
> > >> though.
> > >> > It is
> > >> > > > > >> > > basically a
> > >> > > > > >> > > > > > > > > four step process:
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a
> > >> > software
> > >> > > > > >> > > supporting
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > new group coordinator. Both the old and the new
> > >> > > > > >> coordinator
> > >> > > > > >> > > will
> > >> > > > > >> > > > > > > > > support the old protocol and rely on the same
> > >> > persisted
> > >> > > > > >> > > metadata so
> > >> > > > > >> > > > > > > > > they can work together. This point is an
> > >> offline
> > >> > > > > >> migration.
> > >> > > > > >> > We
> > >> > > > > >> > > > > cannot
> > >> > > > > >> > > > > > > > > do this one live because it would require
> > >> > shutting down
> > >> > > > > >> the
> > >> > > > > >> > > current
> > >> > > > > >> > > > > > > > > coordinator and starting up the new one and
> > >> that
> > >> > would
> > >> > > > > >> cause
> > >> > > > > >> > > > > > > > > unavailabilities.
> > >> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be
> > >> > upgraded
> > >> > > > > to
> > >> > > > > >> X
> > >> > > > > >> > in
> > >> > > > > >> > > > > order
> > >> > > > > >> > > > > > > > > to enable the new protocol. This cannot be done
> > >> > before
> > >> > > > > 1)
> > >> > > > > >> is
> > >> > > > > >> > > > > > > > > terminated because the old coordinator doesn't
> > >> > support
> > >> > > > > the
> > >> > > > > >> > new
> > >> > > > > >> > > > > > > > > protocol.
> > >> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > >> > > > > supporting
> > >> > > > > >> the
> > >> > > > > >> > > > > online
> > >> > > > > >> > > > > > > > > migration (must have KIP-792). If the consumer
> > >> is
> > >> > > > > already
> > >> > > > > >> > > there.
> > >> > > > > >> > > > > > > > > Nothing must be done at this point.
> > >> > > > > >> > > > > > > > > 4. The consumers must be rolled with the
> > >> feature
> > >> > flag
> > >> > > > > >> turned
> > >> > > > > >> > > on.
> > >> > > > > >> > > > > The
> > >> > > > > >> > > > > > > > > consumer group is automatically converted when
> > >> > the first
> > >> > > > > >> > > consumer
> > >> > > > > >> > > > > > > > > using the new protocol joins the group. While
> > >> the
> > >> > > > > members
> > >> > > > > >> > > using the
> > >> > > > > >> > > > > > > > > old protocol are being upgraded, the old
> > >> protocol
> > >> > is
> > >> > > > > >> proxied
> > >> > > > > >> > > into
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > new one.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > >> > > > > >> subscriptions
> > >> > > > > >> > > to
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > server,
> > >> > > > > >> > > > > > > > > > but it seems like there could be some tricky
> > >> > > > > >> compatibility
> > >> > > > > >> > > > > issues.
> > >> > > > > >> > > > > > > Will
> > >> > > > > >> > > > > > > > > we
> > >> > > > > >> > > > > > > > > > have a mechanism for users to detect that
> > >> they
> > >> > need to
> > >> > > > > >> > update
> > >> > > > > >> > > > > their
> > >> > > > > >> > > > > > > regex
> > >> > > > > >> > > > > > > > > > before switching to the new protocol?
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > I think that I am a bit more optimistic than
> > >> you
> > >> > on this
> > >> > > > > >> > > point. I
> > >> > > > > >> > > > > > > > > believe that the majority of the cases are
> > >> simple
> > >> > > > > regexes
> > >> > > > > >> > which
> > >> > > > > >> > > > > should
> > >> > > > > >> > > > > > > > > work with the new engine. The coordinator will
> > >> > verify
> > >> > > > > the
> > >> > > > > >> > regex
> > >> > > > > >> > > > > anyway
> > >> > > > > >> > > > > > > > > and reject the consumer if the regex is not
> > >> valid.
> > >> > > > > Coming
> > >> > > > > >> > back
> > >> > > > > >> > > to
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > migration path, in the worst case, the first
> > >> > upgraded
> > >> > > > > >> > consumer
> > >> > > > > >> > > > > joining
> > >> > > > > >> > > > > > > > > the group will be rejected. This should be used
> > >> > as the
> > >> > > > > >> last
> > >> > > > > >> > > > > defence, I
> > >> > > > > >> > > > > > > > > would say.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > One way for customers to validate their regex
> > >> > before
> > >> > > > > >> > upgrading
> > >> > > > > >> > > > > their
> > >> > > > > >> > > > > > > > > prod would be to test them with another group.
> > >> For
> > >> > > > > >> instance,
> > >> > > > > >> > > that
> > >> > > > > >> > > > > > > > > could be done in a pre-prod environment.
> > >> Another
> > >> > way
> > >> > > > > >> would be
> > >> > > > > >> > > to
> > >> > > > > >> > > > > > > > > extend the consumer-group tool to provide a
> > >> regex
> > >> > > > > >> validation
> > >> > > > > >> > > > > > > > > mechanism. Would this be enough in your
> > >> opinion?
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 5. Related to the last question, will the
> > >> Java
> > >> > client
> > >> > > > > >> allow
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > users to
> > >> > > > > >> > > > > > > > > > stick with the current regex engine for
> > >> > compatibility
> > >> > > > > >> > > reasons?
> > >> > > > > >> > > > > For
> > >> > > > > >> > > > > > > > > example,
> > >> > > > > >> > > > > > > > > > it may be handy to keep using client based
> > >> > regex at
> > >> > > > > >> first
> > >> > > > > >> > to
> > >> > > > > >> > > keep
> > >> > > > > >> > > > > > > > > > migrations simple and then migrate to server
> > >> > based
> > >> > > > > >> regexes
> > >> > > > > >> > > as a
> > >> > > > > >> > > > > > > second
> > >> > > > > >> > > > > > > > > step.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > I understand your point but I am concerned that
> > >> > this
> > >> > > > > would
> > >> > > > > >> > > allow
> > >> > > > > >> > > > > users
> > >> > > > > >> > > > > > > > > to actually stay in this mode. That would go
> > >> > against our
> > >> > > > > >> goal
> > >> > > > > >> > > of
> > >> > > > > >> > > > > > > > > simplifying the client because we would have to
> > >> > continue
> > >> > > > > >> > > monitoring
> > >> > > > > >> > > > > > > > > the metadata on the client side. I would rather
> > >> > not do
> > >> > > > > >> this.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 6. When we say that the group coordinator
> > >> will
> > >> > be
> > >> > > > > >> > > responsible for
> > >> > > > > >> > > > > > > storing
> > >> > > > > >> > > > > > > > > > the configurations and that the
> > >> configurations
> > >> > will be
> > >> > > > > >> > > deleted
> > >> > > > > >> > > > > when
> > >> > > > > >> > > > > > > the
> > >> > > > > >> > > > > > > > > > group is deleted. Will a transition to DEAD
> > >> > trigger
> > >> > > > > >> > deletion
> > >> > > > > >> > > of
> > >> > > > > >> > > > > > > > > > configurations?
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > That's right. The configurations will be
> > >> deleted
> > >> > when
> > >> > > > > the
> > >> > > > > >> > > group is
> > >> > > > > >> > > > > > > > > deleted. They go together.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 7. Will the choice to store the configs in
> > >> the
> > >> > group
> > >> > > > > >> > > coordinator
> > >> > > > > >> > > > > > > make it
> > >> > > > > >> > > > > > > > > > harder to list all cluster configs and their
> > >> > values?
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > I don't think so. The group configurations are
> > >> > overrides
> > >> > > > > >> of
> > >> > > > > >> > > cluster
> > >> > > > > >> > > > > > > > > configs. If you want to know all the overrides
> > >> > though,
> > >> > > > > you
> > >> > > > > >> > > would
> > >> > > > > >> > > > > have
> > >> > > > > >> > > > > > > > > to ask all the group coordinators. You cannot
> > >> > rely on
> > >> > > > > the
> > >> > > > > >> > > metadata
> > >> > > > > >> > > > > log
> > >> > > > > >> > > > > > > > > for instance.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 8. How would someone configure a group before
> > >> > starting
> > >> > > > > >> the
> > >> > > > > >> > > > > consumers?
> > >> > > > > >> > > > > > > > > Have
> > >> > > > > >> > > > > > > > > > we considered allowing the explicit creation
> > >> of
> > >> > > > > groups?
> > >> > > > > >> > > > > > > Alternatively,
> > >> > > > > >> > > > > > > > > the
> > >> > > > > >> > > > > > > > > > configs could be decoupled from the group
> > >> > lifecycle.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Yes. The group will be automatically created in
> > >> > this
> > >> > > > > case.
> > >> > > > > >> > > However,
> > >> > > > > >> > > > > > > > > the configs will be lost after the retention
> > >> > period of
> > >> > > > > the
> > >> > > > > >> > > group
> > >> > > > > >> > > > > > > > > passes.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the
> > >> > Java
> > >> > > > > >> client
> > >> > > > > >> > > still
> > >> > > > > >> > > > > take
> > >> > > > > >> > > > > > > a
> > >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > >> > introduce
> > >> > > > > an
> > >> > > > > >> > > overload?
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > That's a very group question. I forgot about
> > >> that
> > >> > one.
> > >> > > > > As
> > >> > > > > >> the
> > >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
> > >> compatible
> > >> > with
> > >> > > > > the
> > >> > > > > >> > > engine
> > >> > > > > >> > > > > that
> > >> > > > > >> > > > > > > > > we plan to use, it might be better to deprecate
> > >> > it and
> > >> > > > > >> use an
> > >> > > > > >> > > > > overload
> > >> > > > > >> > > > > > > > > which takes a string. We would rely on the
> > >> server
> > >> > side
> > >> > > > > >> > > validation.
> > >> > > > > >> > > > > > > > > During the migration, I think that we could
> > >> still
> > >> > try to
> > >> > > > > >> > > toString
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > regex and use it. That should work, I think, in
> > >> > the
> > >> > > > > >> majority
> > >> > > > > >> > > of the
> > >> > > > > >> > > > > > > > > cases.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 10. I agree with Justine that we should be
> > >> > clearer
> > >> > > > > about
> > >> > > > > >> > the
> > >> > > > > >> > > > > reason
> > >> > > > > >> > > > > > > to
> > >> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
> > >> feature
> > >> > flag.
> > >> > > > > >> Maybe
> > >> > > > > >> > > we
> > >> > > > > >> > > > > mean
> > >> > > > > >> > > > > > > that
> > >> > > > > >> > > > > > > > > we
> > >> > > > > >> > > > > > > > > > can switch the default for the feature flag
> > >> to
> > >> > true
> > >> > > > > >> based
> > >> > > > > >> > on
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > > > > metadata.version once we want to make it the
> > >> > default.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > My plan was to use that feature flag mainly
> > >> > during the
> > >> > > > > >> > > development
> > >> > > > > >> > > > > > > > > phase. I should not have mentioned it, I think,
> > >> > because
> > >> > > > > we
> > >> > > > > >> > > could
> > >> > > > > >> > > > > use
> > >> > > > > >> > > > > > > > > an internal config for it.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention
> > >> the
> > >> > > > > required
> > >> > > > > >> > > ACLs, it
> > >> > > > > >> > > > > > > would
> > >> > > > > >> > > > > > > > > be
> > >> > > > > >> > > > > > > > > > good to add that for consistency.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Noted.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> > >> ConsumerGroupHeartbeat
> > >> > > > > requires
> > >> > > > > >> > > "Read
> > >> > > > > >> > > > > Group"
> > >> > > > > >> > > > > > > > > even
> > >> > > > > >> > > > > > > > > > though it seems to do more than reading.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > I agree. This is how the current protocol works
> > >> > though.
> > >> > > > > We
> > >> > > > > >> > only
> > >> > > > > >> > > > > > > > > require "Read Group" to join a group. We could
> > >> > consider
> > >> > > > > >> > > changing
> > >> > > > > >> > > > > this
> > >> > > > > >> > > > > > > > > but I am not sure that it is worth it.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 13. How is topic recreation handled by the
> > >> > consumer
> > >> > > > > with
> > >> > > > > >> > the
> > >> > > > > >> > > new
> > >> > > > > >> > > > > > > group
> > >> > > > > >> > > > > > > > > > protocol? It would be good to have a section
> > >> on
> > >> > this.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Noted. From a protocol perspective, the new
> > >> topic
> > >> > will
> > >> > > > > >> have a
> > >> > > > > >> > > new
> > >> > > > > >> > > > > > > > > topic id so it will treat it like a topic with
> > >> a
> > >> > > > > different
> > >> > > > > >> > > name.
> > >> > > > > >> > > > > The
> > >> > > > > >> > > > > > > > > only issue is that the fetch/commit offsets
> > >> APIs
> > >> > do not
> > >> > > > > >> > support
> > >> > > > > >> > > > > topic
> > >> > > > > >> > > > > > > > > IDs so the consumer would reuse the offsets
> > >> based
> > >> > on the
> > >> > > > > >> > same.
> > >> > > > > >> > > I
> > >> > > > > >> > > > > think
> > >> > > > > >> > > > > > > > > that we should update those APIs as well in
> > >> order
> > >> > to be
> > >> > > > > >> > > consistent
> > >> > > > > >> > > > > end
> > >> > > > > >> > > > > > > > > to end. That would strengthen the semantics of
> > >> the
> > >> > > > > >> consumer.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the new
> > >> > coordinator
> > >> > > > > >> in
> > >> > > > > >> > > Java.
> > >> > > > > >> > > > > Even
> > >> > > > > >> > > > > > > > > though
> > >> > > > > >> > > > > > > > > > this is an implementation detail, do we plan
> > >> to
> > >> > have a
> > >> > > > > >> new
> > >> > > > > >> > > gradle
> > >> > > > > >> > > > > > > module
> > >> > > > > >> > > > > > > > > > for it?
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Yes.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when it
> > >> comes
> > >> > to how
> > >> > > > > >> many
> > >> > > > > >> > > > > members
> > >> > > > > >> > > > > > > the
> > >> > > > > >> > > > > > > > > new
> > >> > > > > >> > > > > > > > > > group protocol can support?
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > We don't have numbers at the moment. The
> > >> protocol
> > >> > should
> > >> > > > > >> > > support
> > >> > > > > >> > > > > 1000s
> > >> > > > > >> > > > > > > > > of members per group. We will measure this when
> > >> > we have
> > >> > > > > a
> > >> > > > > >> > first
> > >> > > > > >> > > > > > > > > implementation. Note that we might have other
> > >> > > > > bottlenecks
> > >> > > > > >> > down
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > > > road (e.g. offset commits).
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > >> > instead
> > >> > > > > of
> > >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > >> > ConsumerGroupHeartbeatRequest?
> > >> > > > > >> Is
> > >> > > > > >> > the
> > >> > > > > >> > > > > idea
> > >> > > > > >> > > > > > > that
> > >> > > > > >> > > > > > > > > > since we have to resolve the regex on the
> > >> > server, we
> > >> > > > > >> can do
> > >> > > > > >> > > the
> > >> > > > > >> > > > > same
> > >> > > > > >> > > > > > > for
> > >> > > > > >> > > > > > > > > > the topic name? The difference is that
> > >> sending
> > >> > the
> > >> > > > > >> regex is
> > >> > > > > >> > > more
> > >> > > > > >> > > > > > > > > efficient
> > >> > > > > >> > > > > > > > > > whereas sending the topic names is less
> > >> > efficient.
> > >> > > > > >> > > Furthermore,
> > >> > > > > >> > > > > > > delete
> > >> > > > > >> > > > > > > > > and
> > >> > > > > >> > > > > > > > > > recreation is easier to handle if we have
> > >> topic
> > >> > ids.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > The idea was to consolidate the metadata lookup
> > >> > on the
> > >> > > > > >> server
> > >> > > > > >> > > for
> > >> > > > > >> > > > > both
> > >> > > > > >> > > > > > > > > paths but I do agree with your point. As a
> > >> second
> > >> > > > > though,
> > >> > > > > >> > using
> > >> > > > > >> > > > > topic
> > >> > > > > >> > > > > > > > > ids may be better here for the delete and
> > >> > recreation
> > >> > > > > case.
> > >> > > > > >> > > Also, I
> > >> > > > > >> > > > > > > > > suppose that we may allow users to subscribe
> > >> with
> > >> > topic
> > >> > > > > >> ids
> > >> > > > > >> > in
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > > > future because that is the only way to be
> > >> really
> > >> > robust
> > >> > > > > to
> > >> > > > > >> > > topic
> > >> > > > > >> > > > > > > > > re-creation.
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > Best,
> > >> > > > > >> > > > > > > > > David
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > >> > > > > >> > > djacot@confluent.io>
> > >> > > > > >> > > > > > > wrote:
> > >> > > > > >> > > > > > > > > >
> > >> > > > > >> > > > > > > > > > Hi Justine,
> > >> > > > > >> > > > > > > > > >
> > >> > > > > >> > > > > > > > > > Thanks for your comments. Please find my
> > >> answers
> > >> > > > > below.
> > >> > > > > >> > > > > > > > > >
> > >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs
> > >> > with the
> > >> > > > > >> > > exception
> > >> > > > > >> > > > > of the
> > >> > > > > >> > > > > > > > > > topic names based in the
> > >> > > > > ConsumerGroupHeartbeatRequest.
> > >> > > > > >> I
> > >> > > > > >> > am
> > >> > > > > >> > > not
> > >> > > > > >> > > > > sure
> > >> > > > > >> > > > > > > > > > if using topic names is the right call here.
> > >> I
> > >> > need to
> > >> > > > > >> > think
> > >> > > > > >> > > > > about it
> > >> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does not
> > >> > change the
> > >> > > > > >> > > > > fetch/commit
> > >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be
> > >> > something
> > >> > > > > >> that
> > >> > > > > >> > we
> > >> > > > > >> > > > > should
> > >> > > > > >> > > > > > > > > > include though as it would give better
> > >> overall
> > >> > > > > >> guarantee in
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > > > > producer.
> > >> > > > > >> > > > > > > > > > - You're right. I think that I should not
> > >> have
> > >> > > > > mentioned
> > >> > > > > >> > this
> > >> > > > > >> > > > > flag at
> > >> > > > > >> > > > > > > > > > all. I will remove it. We can use an internal
> > >> > > > > >> configuration
> > >> > > > > >> > > while
> > >> > > > > >> > > > > > > > > > developing the feature.
> > >> > > > > >> > > > > > > > > > - Both cluster types will be supported. The
> > >> > change is
> > >> > > > > >> > > > > orthogonal. The
> > >> > > > > >> > > > > > > > > > only requirement is that the cluster uses
> > >> topic
> > >> > IDs.
> > >> > > > > >> > > > > > > > > >
> > >> > > > > >> > > > > > > > > > Best,
> > >> > > > > >> > > > > > > > > > David
> > >> > > > > >> > > > > > > > > >
> > >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang
> > >> Wang <
> > >> > > > > >> > > > > wangguoz@gmail.com>
> > >> > > > > >> > > > > > > > > wrote:
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Hi Ismael,
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are some
> > >> replies
> > >> > > > > inlined
> > >> > > > > >> > > below:
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma
> > >> <
> > >> > > > > >> > > ismael@juma.me.uk>
> > >> > > > > >> > > > > > > wrote:
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the
> > >> potential
> > >> > to be a
> > >> > > > > >> > great
> > >> > > > > >> > > > > > > > > improvement. A few
> > >> > > > > >> > > > > > > > > > > > initial questions/comments:
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk about
> > >> > target
> > >> > > > > >> versions
> > >> > > > > >> > > for
> > >> > > > > >> > > > > > > > > deprecation and
> > >> > > > > >> > > > > > > > > > > > removal of the existing group protocol.
> > >> > Unlike
> > >> > > > > >> KRaft,
> > >> > > > > >> > > this
> > >> > > > > >> > > > > > > affects a
> > >> > > > > >> > > > > > > > > core
> > >> > > > > >> > > > > > > > > > > > client protocol and hence
> > >> > deprecation/removal will
> > >> > > > > >> be
> > >> > > > > >> > > heavily
> > >> > > > > >> > > > > > > > > dependent on
> > >> > > > > >> > > > > > > > > > > > how quickly applications migrate to the
> > >> new
> > >> > > > > >> protocol.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can
> > >> remove
> > >> > the
> > >> > > > > >> proposed
> > >> > > > > >> > > > > timeline
> > >> > > > > >> > > > > > > in
> > >> > > > > >> > > > > > > > > the
> > >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration
> > >> > Plan` and
> > >> > > > > >> > > instead
> > >> > > > > >> > > > > just
> > >> > > > > >> > > > > > > state
> > >> > > > > >> > > > > > > > > > > that we will decide in the future about
> > >> when
> > >> > we
> > >> > > > > would
> > >> > > > > >> > > > > deprecate old
> > >> > > > > >> > > > > > > > > > > protocol and behaviors.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to release this
> > >> > in 4.x,
> > >> > > > > >> but
> > >> > > > > >> > it
> > >> > > > > >> > > > > wasn't
> > >> > > > > >> > > > > > > made
> > >> > > > > >> > > > > > > > > clear
> > >> > > > > >> > > > > > > > > > > > why. If we added that as a way to
> > >> estimate
> > >> > when
> > >> > > > > we'd
> > >> > > > > >> > > > > deprecate
> > >> > > > > >> > > > > > > and
> > >> > > > > >> > > > > > > > > remove
> > >> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
> > >> removing
> > >> > this
> > >> > > > > >> part.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > I think that's not specifically related to
> > >> the
> > >> > > > > >> > > > > deprecation/removal
> > >> > > > > >> > > > > > > > > timeline
> > >> > > > > >> > > > > > > > > > > plan, but it's more for client upgrades.
> > >> I.e.
> > >> > the
> > >> > > > > >> > > broker-side
> > >> > > > > >> > > > > > > > > > > implementation may be done first, and then
> > >> the
> > >> > > > > client
> > >> > > > > >> > side,
> > >> > > > > >> > > > > and we
> > >> > > > > >> > > > > > > > > would
> > >> > > > > >> > > > > > > > > > > only mark it as "released" by the time
> > >> clients
> > >> > > > > >> > > implementations
> > >> > > > > >> > > > > are
> > >> > > > > >> > > > > > > > > done. At
> > >> > > > > >> > > > > > > > > > > that time, to enable the feature the
> > >> clients
> > >> > need to
> > >> > > > > >> > first
> > >> > > > > >> > > > > swap-in
> > >> > > > > >> > > > > > > the
> > >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and then set
> > >> > the flag
> > >> > > > > >> > with a
> > >> > > > > >> > > > > second
> > >> > > > > >> > > > > > > > > rolling
> > >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's better to be
> > >> > released
> > >> > > > > >> in a
> > >> > > > > >> > > major
> > >> > > > > >> > > > > > > > > version.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the details of
> > >> the
> > >> > > > > migration
> > >> > > > > >> > > story.
> > >> > > > > >> > > > > It
> > >> > > > > >> > > > > > > > > sounds like
> > >> > > > > >> > > > > > > > > > > > we're saying we will support online
> > >> > migrations. Is
> > >> > > > > >> that
> > >> > > > > >> > > > > correct?
> > >> > > > > >> > > > > > > We
> > >> > > > > >> > > > > > > > > should
> > >> > > > > >> > > > > > > > > > > > explain this in detail. It could also be
> > >> > done as a
> > >> > > > > >> > > separate
> > >> > > > > >> > > > > KIP,
> > >> > > > > >> > > > > > > if
> > >> > > > > >> > > > > > > > > it's
> > >> > > > > >> > > > > > > > > > > > easier.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Yes I think that's the part we can be more
> > >> > concrete
> > >> > > > > >> about
> > >> > > > > >> > > for
> > >> > > > > >> > > > > sure
> > >> > > > > >> > > > > > > (and
> > >> > > > > >> > > > > > > > > > > this is related to your question 2) above).
> > >> > We will
> > >> > > > > >> work
> > >> > > > > >> > on
> > >> > > > > >> > > > > making
> > >> > > > > >> > > > > > > it
> > >> > > > > >> > > > > > > > > more
> > >> > > > > >> > > > > > > > > > > explicit in parallel as we solicit more
> > >> > feedback.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing the
> > >> > pattern
> > >> > > > > >> > > subscriptions
> > >> > > > > >> > > > > to
> > >> > > > > >> > > > > > > the
> > >> > > > > >> > > > > > > > > server,
> > >> > > > > >> > > > > > > > > > > > but it seems like there could be some
> > >> tricky
> > >> > > > > >> > > compatibility
> > >> > > > > >> > > > > > > issues.
> > >> > > > > >> > > > > > > > > Will we
> > >> > > > > >> > > > > > > > > > > > have a mechanism for users to detect that
> > >> > they
> > >> > > > > need
> > >> > > > > >> to
> > >> > > > > >> > > update
> > >> > > > > >> > > > > > > their
> > >> > > > > >> > > > > > > > > regex
> > >> > > > > >> > > > > > > > > > > > before switching to the new protocol?
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
> > >> non-java
> > >> > client
> > >> > > > > >> > users
> > >> > > > > >> > > to
> > >> > > > > >> > > > > sort
> > >> > > > > >> > > > > > > of
> > >> > > > > >> > > > > > > > > > > "dry-run" the client before switching to
> > >> the
> > >> > new
> > >> > > > > >> > protocol.
> > >> > > > > >> > > I
> > >> > > > > >> > > > > do not
> > >> > > > > >> > > > > > > > > have a
> > >> > > > > >> > > > > > > > > > > specific idea on top of my head though,
> > >> maybe
> > >> > others
> > >> > > > > >> like
> > >> > > > > >> > > @Matt
> > >> > > > > >> > > > > > > > > Howlett can
> > >> > > > > >> > > > > > > > > > > chime-in here?
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 5. Related to the last question, will the
> > >> > Java
> > >> > > > > >> client
> > >> > > > > >> > > allow
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > users to
> > >> > > > > >> > > > > > > > > > > > stick with the current regex engine for
> > >> > > > > >> compatibility
> > >> > > > > >> > > > > reasons?
> > >> > > > > >> > > > > > > For
> > >> > > > > >> > > > > > > > > example,
> > >> > > > > >> > > > > > > > > > > > it may be handy to keep using client
> > >> based
> > >> > regex
> > >> > > > > at
> > >> > > > > >> > > first to
> > >> > > > > >> > > > > keep
> > >> > > > > >> > > > > > > > > > > > migrations simple and then migrate to
> > >> > server based
> > >> > > > > >> > > regexes
> > >> > > > > >> > > > > as a
> > >> > > > > >> > > > > > > > > second
> > >> > > > > >> > > > > > > > > > > > step.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Honestly I have not thought about that for
> > >> > java
> > >> > > > > >> clients,
> > >> > > > > >> > > and
> > >> > > > > >> > > > > we can
> > >> > > > > >> > > > > > > > > discuss
> > >> > > > > >> > > > > > > > > > > that. What kind of compatibility issues do
> > >> > you have
> > >> > > > > in
> > >> > > > > >> > > mind?
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 6. When we say that the group coordinator
> > >> > will be
> > >> > > > > >> > > > > responsible for
> > >> > > > > >> > > > > > > > > storing
> > >> > > > > >> > > > > > > > > > > > the configurations and that the
> > >> > configurations
> > >> > > > > will
> > >> > > > > >> be
> > >> > > > > >> > > > > deleted
> > >> > > > > >> > > > > > > when
> > >> > > > > >> > > > > > > > > the
> > >> > > > > >> > > > > > > > > > > > group is deleted. Will a transition to
> > >> DEAD
> > >> > > > > trigger
> > >> > > > > >> > > deletion
> > >> > > > > >> > > > > of
> > >> > > > > >> > > > > > > > > > > > configurations?
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an ending
> > >> state
> > >> > (we
> > >> > > > > would
> > >> > > > > >> > only
> > >> > > > > >> > > > > > > transit to
> > >> > > > > >> > > > > > > > > that
> > >> > > > > >> > > > > > > > > > > state when the group is EMPTY and also all
> > >> of
> > >> > its
> > >> > > > > >> > metadata
> > >> > > > > >> > > are
> > >> > > > > >> > > > > > > gone),
> > >> > > > > >> > > > > > > > > once
> > >> > > > > >> > > > > > > > > > > it's transited to DEAD this group would
> > >> never
> > >> > be
> > >> > > > > >> revived.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 7. Will the choice to store the configs
> > >> in
> > >> > the
> > >> > > > > group
> > >> > > > > >> > > > > coordinator
> > >> > > > > >> > > > > > > > > make it
> > >> > > > > >> > > > > > > > > > > > harder to list all cluster configs and
> > >> their
> > >> > > > > values?
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > That's a good question, and our thoughts
> > >> are
> > >> > that
> > >> > > > > the
> > >> > > > > >> > > so-called
> > >> > > > > >> > > > > > > "group
> > >> > > > > >> > > > > > > > > > > configurations" are overrides of the
> > >> > cluster-level
> > >> > > > > >> > > > > configurations
> > >> > > > > >> > > > > > > > > > > customized per group so when an admin list
> > >> > cluster
> > >> > > > > >> > configs
> > >> > > > > >> > > it's
> > >> > > > > >> > > > > > > okay to
> > >> > > > > >> > > > > > > > > > > list just the cluster-level "defaults", not
> > >> > showing
> > >> > > > > >> any
> > >> > > > > >> > > > > per-group
> > >> > > > > >> > > > > > > > > > > customizations.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 8. How would someone configure a group
> > >> > before
> > >> > > > > >> starting
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > > > consumers? Have
> > >> > > > > >> > > > > > > > > > > > we considered allowing the explicit
> > >> > creation of
> > >> > > > > >> groups?
> > >> > > > > >> > > > > > > > > Alternatively, the
> > >> > > > > >> > > > > > > > > > > > configs could be decoupled from the group
> > >> > > > > lifecycle.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > The configs can be created before the group
> > >> > itself
> > >> > > > > as
> > >> > > > > >> an
> > >> > > > > >> > > > > > > independent
> > >> > > > > >> > > > > > > > > entity
> > >> > > > > >> > > > > > > > > > > --- of course, this requires the
> > >> corresponding
> > >> > > > > >> request to
> > >> > > > > >> > > be
> > >> > > > > >> > > > > > > routed to
> > >> > > > > >> > > > > > > > > the
> > >> > > > > >> > > > > > > > > > > right coordinator based on the group id ---
> > >> > the only
> > >> > > > > >> > thing
> > >> > > > > >> > > that
> > >> > > > > >> > > > > > > > > differs is,
> > >> > > > > >> > > > > > > > > > > when the group itself is gone we also check
> > >> > if there
> > >> > > > > >> are
> > >> > > > > >> > > any
> > >> > > > > >> > > > > > > > > configuration
> > >> > > > > >> > > > > > > > > > > entities related to that group and delete
> > >> as
> > >> > well.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> > >> > asymmetry on
> > >> > > > > the
> > >> > > > > >> > > creation
> > >> > > > > >> > > > > /
> > >> > > > > >> > > > > > > > > deletion
> > >> > > > > >> > > > > > > > > > > lifecycles of the config entities, and we
> > >> > would like
> > >> > > > > >> to
> > >> > > > > >> > > hear
> > >> > > > > >> > > > > > > everyone's
> > >> > > > > >> > > > > > > > > > > feelings whether we should aim for symmetry
> > >> > i.e.
> > >> > > > > >> totally
> > >> > > > > >> > > > > decouple
> > >> > > > > >> > > > > > > group
> > >> > > > > >> > > > > > > > > > > configs and hence not delete them at all
> > >> when
> > >> > the
> > >> > > > > >> group
> > >> > > > > >> > is
> > >> > > > > >> > > > > gone,
> > >> > > > > >> > > > > > > but
> > >> > > > > >> > > > > > > > > always
> > >> > > > > >> > > > > > > > > > > require explicit deletion operations by
> > >> > themselves.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for
> > >> > the Java
> > >> > > > > >> > client
> > >> > > > > >> > > > > still
> > >> > > > > >> > > > > > > take
> > >> > > > > >> > > > > > > > > a
> > >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have
> > >> to
> > >> > > > > >> introduce an
> > >> > > > > >> > > > > overload?
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > I think we do not need to introduce an
> > >> > overload, but
> > >> > > > > >> I'm
> > >> > > > > >> > > all
> > >> > > > > >> > > > > ears
> > >> > > > > >> > > > > > > if
> > >> > > > > >> > > > > > > > > there
> > >> > > > > >> > > > > > > > > > > may be some compatibility issues that we
> > >> may
> > >> > > > > overlook.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we should
> > >> be
> > >> > clearer
> > >> > > > > >> > about
> > >> > > > > >> > > the
> > >> > > > > >> > > > > > > reason
> > >> > > > > >> > > > > > > > > to
> > >> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version from the
> > >> > feature
> > >> > > > > >> flag.
> > >> > > > > >> > > Maybe
> > >> > > > > >> > > > > we
> > >> > > > > >> > > > > > > mean
> > >> > > > > >> > > > > > > > > that we
> > >> > > > > >> > > > > > > > > > > > can switch the default for the feature
> > >> flag
> > >> > to
> > >> > > > > true
> > >> > > > > >> > > based on
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > > > > metadata.version once we want to make it
> > >> the
> > >> > > > > >> default.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention
> > >> > the
> > >> > > > > >> required
> > >> > > > > >> > > ACLs,
> > >> > > > > >> > > > > it
> > >> > > > > >> > > > > > > > > would be
> > >> > > > > >> > > > > > > > > > > > good to add that for consistency.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > >> > ConsumerGroupHeartbeat
> > >> > > > > >> > requires
> > >> > > > > >> > > > > "Read
> > >> > > > > >> > > > > > > > > Group" even
> > >> > > > > >> > > > > > > > > > > > though it seems to do more than reading.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > I had that thought myself as well, but in
> > >> the
> > >> > end we
> > >> > > > > >> > could
> > >> > > > > >> > > not
> > >> > > > > >> > > > > > > find a
> > >> > > > > >> > > > > > > > > > > better alternative: adding Write Group
> > >> seems
> > >> > an
> > >> > > > > >> overkill
> > >> > > > > >> > > here
> > >> > > > > >> > > > > > > since we
> > >> > > > > >> > > > > > > > > do
> > >> > > > > >> > > > > > > > > > > not have it elsewhere (we only have Read /
> > >> > Delete
> > >> > > > > and
> > >> > > > > >> > > Describe
> > >> > > > > >> > > > > on
> > >> > > > > >> > > > > > > > > groups so
> > >> > > > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 13. How is topic recreation handled by
> > >> the
> > >> > > > > consumer
> > >> > > > > >> > with
> > >> > > > > >> > > the
> > >> > > > > >> > > > > new
> > >> > > > > >> > > > > > > > > group
> > >> > > > > >> > > > > > > > > > > > protocol? It would be good to have a
> > >> > section on
> > >> > > > > >> this.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > You mean with regex subscription right? Yes
> > >> > we can
> > >> > > > > >> add a
> > >> > > > > >> > > > > section
> > >> > > > > >> > > > > > > about
> > >> > > > > >> > > > > > > > > > > that, but basically the idea is that
> > >> consumer
> > >> > would
> > >> > > > > be
> > >> > > > > >> > > totally
> > >> > > > > >> > > > > > > > > agnostic in
> > >> > > > > >> > > > > > > > > > > the new protocol as it's handled all by the
> > >> > brokers.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write the
> > >> new
> > >> > > > > >> coordinator
> > >> > > > > >> > in
> > >> > > > > >> > > > > Java.
> > >> > > > > >> > > > > > > Even
> > >> > > > > >> > > > > > > > > though
> > >> > > > > >> > > > > > > > > > > > this is an implementation detail, do we
> > >> > plan to
> > >> > > > > >> have a
> > >> > > > > >> > > new
> > >> > > > > >> > > > > gradle
> > >> > > > > >> > > > > > > > > module
> > >> > > > > >> > > > > > > > > > > > for it?
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > We have not thought about that. But I think
> > >> > the
> > >> > > > > answer
> > >> > > > > >> > > should
> > >> > > > > >> > > > > be
> > >> > > > > >> > > > > > > yes.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it
> > >> > comes to
> > >> > > > > >> how
> > >> > > > > >> > > many
> > >> > > > > >> > > > > > > members
> > >> > > > > >> > > > > > > > > the new
> > >> > > > > >> > > > > > > > > > > > group protocol can support?
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > Within a group, I think we should shoot for
> > >> > 1000s of
> > >> > > > > >> > > members.
> > >> > > > > >> > > > > But
> > >> > > > > >> > > > > > > that
> > >> > > > > >> > > > > > > > > > > scalability goals also depend on the offset
> > >> > > > > management
> > >> > > > > >> > > (commit,
> > >> > > > > >> > > > > > > fetch)
> > >> > > > > >> > > > > > > > > > > capabilities of the coordinator which we
> > >> did
> > >> > not
> > >> > > > > >> cover in
> > >> > > > > >> > > this
> > >> > > > > >> > > > > > > KIP, so
> > >> > > > > >> > > > > > > > > it's
> > >> > > > > >> > > > > > > > > > > hard to give a number that applies
> > >> > universally.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> > >> > SubscribedTopidIds
> > >> > > > > >> instead
> > >> > > > > >> > of
> > >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > >> > > > > >> ConsumerGroupHeartbeatRequest?
> > >> > > > > >> > > Is the
> > >> > > > > >> > > > > > > idea
> > >> > > > > >> > > > > > > > > that
> > >> > > > > >> > > > > > > > > > > > since we have to resolve the regex on the
> > >> > server,
> > >> > > > > we
> > >> > > > > >> > can
> > >> > > > > >> > > do
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > same
> > >> > > > > >> > > > > > > > > for
> > >> > > > > >> > > > > > > > > > > > the topic name? The difference is that
> > >> > sending the
> > >> > > > > >> > regex
> > >> > > > > >> > > is
> > >> > > > > >> > > > > more
> > >> > > > > >> > > > > > > > > efficient
> > >> > > > > >> > > > > > > > > > > > whereas sending the topic names is less
> > >> > efficient.
> > >> > > > > >> > > > > Furthermore,
> > >> > > > > >> > > > > > > > > delete and
> > >> > > > > >> > > > > > > > > > > > recreation is easier to handle if we have
> > >> > topic
> > >> > > > > ids.
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > The main reason to still let the clients
> > >> send
> > >> > names
> > >> > > > > >> is to
> > >> > > > > >> > > keep
> > >> > > > > >> > > > > the
> > >> > > > > >> > > > > > > > > > > reasoning of names -> ids on the broker /
> > >> > admin
> > >> > > > > client
> > >> > > > > >> > > only.
> > >> > > > > >> > > > > Note
> > >> > > > > >> > > > > > > that
> > >> > > > > >> > > > > > > > > > > although we added topic id in KIP-516, we
> > >> > never
> > >> > > > > >> > > implemented the
> > >> > > > > >> > > > > > > logic
> > >> > > > > >> > > > > > > > > on
> > >> > > > > >> > > > > > > > > > > consumer/producers leveraging the related
> > >> > newer
> > >> > > > > >> versioned
> > >> > > > > >> > > RPCs,
> > >> > > > > >> > > > > > > > > instead we
> > >> > > > > >> > > > > > > > > > > just set the topic id as empty UUID. We
> > >> want
> > >> > to keep
> > >> > > > > >> the
> > >> > > > > >> > > > > > > > > consumer/producer
> > >> > > > > >> > > > > > > > > > > to be thin and only delegate the reasoning
> > >> on
> > >> > broker
> > >> > > > > >> and
> > >> > > > > >> > > > > > > potentially
> > >> > > > > >> > > > > > > > > admin
> > >> > > > > >> > > > > > > > > > > clients.
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > Thanks,
> > >> > > > > >> > > > > > > > > > > > Ismael
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David
> > >> Jacot
> > >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > >> > > > > >> > > > > > > > > > > > wrote:
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > > Hi all,
> > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > > I would like to start a discussion
> > >> thread
> > >> > on
> > >> > > > > >> KIP-848:
> > >> > > > > >> > > The
> > >> > > > > >> > > > > Next
> > >> > > > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance
> > >> > Protocol.
> > >> > > > > >> With
> > >> > > > > >> > > this
> > >> > > > > >> > > > > KIP,
> > >> > > > > >> > > > > > > we
> > >> > > > > >> > > > > > > > > aim
> > >> > > > > >> > > > > > > > > > > > > to make the rebalance protocol (for
> > >> > consumers)
> > >> > > > > >> more
> > >> > > > > >> > > > > reliable,
> > >> > > > > >> > > > > > > more
> > >> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
> > >> > clients, and
> > >> > > > > >> easier
> > >> > > > > >> > > to
> > >> > > > > >> > > > > debug
> > >> > > > > >> > > > > > > for
> > >> > > > > >> > > > > > > > > > > > > operators.
> > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > > The KIP is here:
> > >> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > > Please take a look and let me know what
> > >> > you
> > >> > > > > think.
> > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > > Best,
> > >> > > > > >> > > > > > > > > > > > > David
> > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > > > PS: I will be away from July 18th to
> > >> > August 8th.
> > >> > > > > >> That
> > >> > > > > >> > > gives
> > >> > > > > >> > > > > > > you a
> > >> > > > > >> > > > > > > > > bit
> > >> > > > > >> > > > > > > > > > > > > of time to read and digest this long
> > >> KIP.
> > >> > > > > >> > > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > >
> > >> > > > > >> > > > > > > > > > > --
> > >> > > > > >> > > > > > > > > > > -- Guozhang
> > >> > > > > >> > > > > > > > >
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > >
> > >> > > > > >> > > > > > > > --
> > >> > > > > >> > > > > > > > -- Guozhang
> > >> > > > > >> > > > > > >
> > >> > > > > >> > > > >
> > >> > > > > >> > >
> > >> > > > > >> > >
> > >> > > > > >> >
> > >> > > > > >>
> > >> > > > > >>
> > >> > > > > >> --
> > >> > > > > >> -- Guozhang
> > >> > > > > >>
> > >> > > > > >
> > >> > > > >
> > >> >
> > >>
> > >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Luke,

> 1.1. I think the state machine are: "Empty, assigning, reconciling, stable,
> dead" mentioned in Consumer Group States section, right?

This sentence does not refer to those group states but rather to a
state machine replication (SMR). This refers to the entire state of
group coordinator which is replicated via the log layer. I will
clarify this in the KIP.

> 1.2. What do you mean "each state machine is modelled as an event loop"?

The idea is to follow a model similar to the new quorum controller. We
will have N threads to process events. Each __consumer_offsets
partition is assigned to a unique thread and all the events (e.g.
requests, callbacks, etc.) are processed by this thread. This simplify
concurrency and will enable us to do simulation testing for the group
coordinator.

> 1.3. Why do we need a state machine per *__consumer_offsets* partitions?
> Not a state machine "per consumer group" owned by a group coordinator? For
> example, if one group coordinator owns 2 consumer groups, and both exist in
> *__consumer_offsets-0*, will we have 1 state machine for it, or 2?

See 1.1. The confusion comes from there, I think.

> 1.4. I know the "*group.coordinator.threads" *is the number of threads used
> to run the state machines. But I'm wondering if the purpose of the threads
> is only to keep the state of each consumer group (or *__consumer_offsets*
> partitions?), and no heavy computation, why should we need multi-threads
> here?

See 1.2. The idea is to have an ability to shard the processing as the
computation could be heavy.

> 2.1. The consumer session timeout, why does the default session timeout not
> locate between min (45s) and max(60s)? I thought the min/max session
> timeout is to define lower/upper bound of it, no?
>
> group.consumer.session.timeout.ms int 30s The timeout to detect client
> failures when using the consumer group protocol.
> group.consumer.min.session.timeout.ms int 45s The minimum session timeout.
> group.consumer.max.session.timeout.ms int 60s The maximum session timeout.

This is indeed a mistake. The default session timeout should be 45s
(the current default).

> 2.2. The default server side assignor are [range, uniform], which means
> we'll default to "range" assignor. I'd like to know why not uniform one? I
> thought usually users will choose uniform assignor (former sticky assinor)
> for better evenly distribution. Any other reason we choose range assignor
> as default?
> group.consumer.assignors List range, uniform The server side assignors.

The order on the server side has no influence because the client must
chose the selector that he wants to use. There is no default in the
current proposal. If the assignor is not specified by the client, the
request is rejected. The default client value for
`group.remote.assignor` is `uniform` though.

Thanks for your very good comments, Luke. I hope that my answers help
to clarify things. I will update the KIP as well based on your
feedback.

Cheers,
David

On Mon, Aug 22, 2022 at 9:29 AM Luke Chen <sh...@gmail.com> wrote:
>
> Hi David,
>
> Thanks for the update.
>
> Some more questions:
> 1. In Group Coordinator section, you mentioned:
> > The new group coordinator will have a state machine per
> *__consumer_offsets* partitions, where each state machine is modelled as an
> event loop. Those state machines will be executed in
> *group.coordinator.threads* threads.
>
> 1.1. I think the state machine are: "Empty, assigning, reconciling, stable,
> dead" mentioned in Consumer Group States section, right?
> 1.2. What do you mean "each state machine is modelled as an event loop"?
> 1.3. Why do we need a state machine per *__consumer_offsets* partitions?
> Not a state machine "per consumer group" owned by a group coordinator? For
> example, if one group coordinator owns 2 consumer groups, and both exist in
> *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> 1.4. I know the "*group.coordinator.threads" *is the number of threads used
> to run the state machines. But I'm wondering if the purpose of the threads
> is only to keep the state of each consumer group (or *__consumer_offsets*
> partitions?), and no heavy computation, why should we need multi-threads
> here?
>
> 2. For the default value in the new configs:
> 2.1. The consumer session timeout, why does the default session timeout not
> locate between min (45s) and max(60s)? I thought the min/max session
> timeout is to define lower/upper bound of it, no?
>
> group.consumer.session.timeout.ms int 30s The timeout to detect client
> failures when using the consumer group protocol.
> group.consumer.min.session.timeout.ms int 45s The minimum session timeout.
> group.consumer.max.session.timeout.ms int 60s The maximum session timeout.
>
>
>
> 2.2. The default server side assignor are [range, uniform], which means
> we'll default to "range" assignor. I'd like to know why not uniform one? I
> thought usually users will choose uniform assignor (former sticky assinor)
> for better evenly distribution. Any other reason we choose range assignor
> as default?
> group.consumer.assignors List range, uniform The server side assignors.
>
>
>
>
>
>
> Thank you.
> Luke
>
>
>
>
>
>
> On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com> wrote:
>
> > Hi Sagar,
> >
> > I have some thoughts about Kafka Connect integrating with KIP-848, but I
> > think we should have a separate discussion thread for the Kafka Connect
> > KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol [1],
> > and let this discussion thread focus on consumer rebalance protocol, WDYT?
> >
> > [1]
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >
> > Thank you.
> > Luke
> >
> >
> >
> > On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com> wrote:
> >
> >> Thank you Guozhang/David for the feedback. Looks like there's agreement on
> >> using separate APIs for Connect. I would revisit the doc and see what
> >> changes are to be made.
> >>
> >> Thanks!
> >> Sagar.
> >>
> >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot <dj...@confluent.io.invalid>
> >> wrote:
> >>
> >> > Hi Sagar,
> >> >
> >> > Thanks for the feedback and the document. That's really helpful. I
> >> > will take a look at it.
> >> >
> >> > Overall, it seems to me that both Connect and the Consumer could share
> >> > the same underlying "engine". The main difference is that the Consumer
> >> > assigns topic-partitions to members whereas Connect assigns tasks to
> >> > workers. I see two ways to move forward:
> >> > 1) We extend the new proposed APIs to support different resource types
> >> > (e.g. partitions, tasks, etc.); or
> >> > 2) We use new dedicated APIs for Connect. The dedicated APIs would be
> >> > similar to the new ones but different on the content/resources and
> >> > they would rely on the same engine on the coordinator side.
> >> >
> >> > I personally lean towards 2) because I am not a fan of overcharging
> >> > APIs to serve different purposes. That being said, I am not opposed to
> >> > 1) if we can find an elegant way to do it.
> >> >
> >> > I think that we can continue to discuss it here for now in order to
> >> > ensure that this KIP is compatible with what we will do for Connect in
> >> > the future.
> >> >
> >> > Best,
> >> > David
> >> >
> >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io> wrote:
> >> > >
> >> > > Hi all,
> >> > >
> >> > > I am back from vacation. I will go through and address your comments
> >> > > in the coming days. Thanks for your feedback.
> >> > >
> >> > > Cheers,
> >> > > David
> >> > >
> >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gharris1727@gmail.com
> >> >
> >> > wrote:
> >> > > >
> >> > > > Hey All!
> >> > > >
> >> > > > Thanks for the KIP, it's wonderful to see cooperative rebalancing
> >> > making it
> >> > > > down the stack!
> >> > > >
> >> > > > I had a few questions:
> >> > > >
> >> > > > 1. The 'Rejected Alternatives' section describes how member epoch
> >> > should
> >> > > > advance in step with the group epoch and assignment epoch values. I
> >> > think
> >> > > > that this is a good idea for the reasons described in the KIP. When
> >> the
> >> > > > protocol is incrementally assigning partitions to a worker, what
> >> member
> >> > > > epoch does each incremental assignment use? Are member epochs
> >> re-used,
> >> > and
> >> > > > a single member epoch can correspond to multiple different
> >> > (monotonically
> >> > > > larger) assignments?
> >> > > >
> >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> >> coordinator? If
> >> > > > not, should custom client-side assignor implementations interact
> >> with
> >> > the
> >> > > > Reason field, and how is its common meaning agreed upon? If so, what
> >> > is the
> >> > > > benefit of a distinct Reason field over including such functionality
> >> > in the
> >> > > > opaque metadata?
> >> > > >
> >> > > > 3. The following is included in the KIP: "Thanks to this, the input
> >> of
> >> > the
> >> > > > client side assignor is entirely driven by the group coordinator.
> >> The
> >> > > > consumer is no longer responsible for maintaining any state besides
> >> its
> >> > > > assigned partitions." Does this mean that the client-side assignor
> >> MAY
> >> > > > incorporate additional non-Metadata state (such as partition
> >> > throughput,
> >> > > > cpu/memory metrics, config topics, etc), or that additional
> >> > non-Metadata
> >> > > > state SHOULD NOT be used?
> >> > > >
> >> > > > 4. I see that there are separate classes
> >> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> >> > > > overlap significantly. Is it possible for these two implementations
> >> to
> >> > be
> >> > > > unified? This would serve to promote feature parity of server-side
> >> and
> >> > > > client-side assignors, and would also facilitate operational
> >> > flexibility in
> >> > > > certain situations. For example, if a server-side assignor has some
> >> > poor
> >> > > > behavior and needs a patch, deploying the patched assignor to the
> >> > client
> >> > > > and switching one consumer group to a client-side assignor may be
> >> > faster
> >> > > > and less risky than patching all of the brokers. With the currently
> >> > > > proposed distinct APIs, a non-trivial reimplementation would have
> >> to be
> >> > > > assembled, and if the two APIs have diverged significantly, then it
> >> is
> >> > > > possible that a reimplementation would not be possible.
> >> > > >
> >> > > > --
> >> > > > Greg Harris
> >> > > > gharris1727@gmail.com
> >> > > > github.com/gharris1727
> >> > > >
> >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com>
> >> > wrote:
> >> > > >
> >> > > > > Hi Guozhang/David,
> >> > > > >
> >> > > > > I created a confluence page to discuss how Connect would need to
> >> > change
> >> > > > > based on the new rebalance protocol. Here's the page:
> >> > > > >
> >> > > > >
> >> >
> >> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >> > > > >
> >> > > > > It's also pretty longish and I have tried to keep a format
> >> similar to
> >> > > > > KIP-848. Let me know what you think. Also, do you think this
> >> should
> >> > be
> >> > > > > moved to a separate discussion thread or is this one fine?
> >> > > > >
> >> > > > > Thanks!
> >> > > > > Sagar.
> >> > > > >
> >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com>
> >> > wrote:
> >> > > > >
> >> > > > > > Hello Guozhang,
> >> > > > > >
> >> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I would do
> >> > the
> >> > > > > > analysis and come up with such a document.
> >> > > > > >
> >> > > > > > Thanks!
> >> > > > > > Sagar.
> >> > > > > >
> >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> >> wangguoz@gmail.com>
> >> > > > > wrote:
> >> > > > > >
> >> > > > > >> Hello Sagar,
> >> > > > > >>
> >> > > > > >> It would be great if you could come back with some analysis on
> >> > how to
> >> > > > > >> implement the Connect side integration with the new protocol;
> >> so
> >> > far
> >> > > > > >> besides leveraging on the new "protocol type" we did not yet
> >> think
> >> > > > > through
> >> > > > > >> the Connect side implementations. For Streams here's a draft of
> >> > > > > >> integration
> >> > > > > >> plan:
> >> > > > > >>
> >> > > > > >>
> >> > > > >
> >> >
> >> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >> > > > > >> just FYI for your analysis on Connect.
> >> > > > > >>
> >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> >> sagarmeansocean@gmail.com
> >> > >
> >> > > > > wrote:
> >> > > > > >>
> >> > > > > >> > Hi David,
> >> > > > > >> >
> >> > > > > >> > Thank you for your response. The reason I thought connect can
> >> > also fit
> >> > > > > >> into
> >> > > > > >> > this new scheme is that even today the connect uses a
> >> > > > > WorkerCoordinator
> >> > > > > >> > extending from AbstractCoordinator to empower rebalances of
> >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> >> protocolType()
> >> > to
> >> > > > > >> connect
> >> > > > > >> > and uses the metadata() method by plumbing into
> >> > > > > >> JoinGroupRequestProtocol.
> >> > > > > >> >
> >> > > > > >> > I think the changes to support connect would be similar at a
> >> > high
> >> > > > > level
> >> > > > > >> to
> >> > > > > >> > the changes in streams mainly because of the Client side
> >> > assignors
> >> > > > > being
> >> > > > > >> > used in both. At an implementation level, we might need to
> >> make
> >> > a lot
> >> > > > > of
> >> > > > > >> > changes to get onto this new assignment protocol like
> >> enhancing
> >> > the
> >> > > > > >> > JoinGroup request/response and SyncGroup and using
> >> > > > > >> ConsumerGroupHeartbeat
> >> > > > > >> > API etc again on similar lines to streams (or there might be
> >> > > > > >> deviations). I
> >> > > > > >> > would try to perform a detailed analysis of the same  and we
> >> > can have
> >> > > > > a
> >> > > > > >> > separate discussion thread for that as that would derail this
> >> > > > > discussion
> >> > > > > >> > thread. Let me know if that sounds good to you.
> >> > > > > >> >
> >> > > > > >> > Thanks!
> >> > > > > >> > Sagar.
> >> > > > > >> >
> >> > > > > >> >
> >> > > > > >> >
> >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> >> > > > > <djacot@confluent.io.invalid
> >> > > > > >> >
> >> > > > > >> > wrote:
> >> > > > > >> >
> >> > > > > >> > > Hi Sagar,
> >> > > > > >> > >
> >> > > > > >> > > Thanks for your comments.
> >> > > > > >> > >
> >> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can
> >> > mention it.
> >> > > > > >> > >
> >> > > > > >> > > 2) The idea is to transition C from his current assignment
> >> to
> >> > his
> >> > > > > >> > > target assignment when he can move to epoch 3. When that
> >> > happens,
> >> > > > > the
> >> > > > > >> > > member assignment is updated and persisted with all its
> >> > assigned
> >> > > > > >> > > partitions even if they are not all revoked yet. In other
> >> > words, the
> >> > > > > >> > > member assignment becomes the target assignment. This is
> >> > basically
> >> > > > > an
> >> > > > > >> > > optimization to avoid having to write all the changes to
> >> the
> >> > log.
> >> > > > > The
> >> > > > > >> > > examples are based on the persisted state so I understand
> >> the
> >> > > > > >> > > confusion. Let me see if I can improve this in the
> >> > description.
> >> > > > > >> > >
> >> > > > > >> > > 3) Regarding Connect, it could reuse the protocol with a
> >> > client side
> >> > > > > >> > > assignor if it fits in the protocol. The assignment is
> >> about
> >> > > > > >> > > topicid-partitions + metadata, could Connect fit into this?
> >> > > > > >> > >
> >> > > > > >> > > Best,
> >> > > > > >> > > David
> >> > > > > >> > >
> >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> >> > sagarmeansocean@gmail.com>
> >> > > > > >> wrote:
> >> > > > > >> > > >
> >> > > > > >> > > > Hi David,
> >> > > > > >> > > >
> >> > > > > >> > > > Thanks for the KIP. I just had minor observations:
> >> > > > > >> > > >
> >> > > > > >> > > > 1) In the Assignment Error section in Client Side mode
> >> > Assignment
> >> > > > > >> > > process,
> >> > > > > >> > > > you mentioned => `In this case, the client side assignor
> >> can
> >> > > > > return
> >> > > > > >> an
> >> > > > > >> > > > error to the group coordinator`. In this case are you
> >> > referring to
> >> > > > > >> the
> >> > > > > >> > > > Assignor returning an AssignmentError that's listed down
> >> > towards
> >> > > > > the
> >> > > > > >> > end?
> >> > > > > >> > > > If yes, do you think it would make sense to mention this
> >> > > > > explicitly
> >> > > > > >> > here?
> >> > > > > >> > > >
> >> > > > > >> > > > 2) In the Case Studies section, I have a slight
> >> confusion,
> >> > not
> >> > > > > sure
> >> > > > > >> if
> >> > > > > >> > > > others have the same. Consider this step:
> >> > > > > >> > > >
> >> > > > > >> > > > When B heartbeats, the group coordinator transitions him
> >> to
> >> > epoch
> >> > > > > 3
> >> > > > > >> > > because
> >> > > > > >> > > > B has no partitions to revoke. It persists the change and
> >> > reply.
> >> > > > > >> > > >
> >> > > > > >> > > >    - Group (epoch=3)
> >> > > > > >> > > >       - A
> >> > > > > >> > > >       - B
> >> > > > > >> > > >       - C
> >> > > > > >> > > >    - Target Assignment (epoch=3)
> >> > > > > >> > > >       - A - partitions=[foo-0]
> >> > > > > >> > > >       - B - partitions=[foo-2]
> >> > > > > >> > > >       - C - partitions=[foo-1]
> >> > > > > >> > > >    - Member Assignment
> >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> >> > > > > >> > > >
> >> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot
> >> get
> >> > foo-1
> >> > > > > >> yet.
> >> > > > > >> > > >
> >> > > > > >> > > > Here,it's mentioned that member C can't get the foo-1
> >> > partition
> >> > > > > yet,
> >> > > > > >> > but
> >> > > > > >> > > > based on the description above, it seems it already has
> >> it.
> >> > Do you
> >> > > > > >> > think
> >> > > > > >> > > it
> >> > > > > >> > > > would be better to remove it and populate it only when it
> >> > actually
> >> > > > > >> gets
> >> > > > > >> > > it?
> >> > > > > >> > > > I see this in a lot of other places, so have I
> >> understood it
> >> > > > > >> > incorrectly
> >> > > > > >> > > ?
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > > Regarding connect , it might be out of scope of this
> >> > discussion,
> >> > > > > but
> >> > > > > >> > from
> >> > > > > >> > > > what I understood it would probably be running in client
> >> > side
> >> > > > > >> assignor
> >> > > > > >> > > mode
> >> > > > > >> > > > even on the new rebalance protocol as it has its own
> >> Custom
> >> > > > > >> > > Assignors(Eager
> >> > > > > >> > > > and IncrementalCooperative).
> >> > > > > >> > > >
> >> > > > > >> > > > Thanks!
> >> > > > > >> > > >
> >> > > > > >> > > > Sagar.
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> >> > > > > >> > <djacot@confluent.io.invalid
> >> > > > > >> > > >
> >> > > > > >> > > > wrote:
> >> > > > > >> > > >
> >> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
> >> > specialized API
> >> > > > > >> > > > > instead of relying on one generic API. For Connect, we
> >> > can apply
> >> > > > > >> the
> >> > > > > >> > > > > exact same pattern and reuse/share the core
> >> > implementation on
> >> > > > > the
> >> > > > > >> > > > > server side. For the schema registry, I think that we
> >> > should
> >> > > > > >> consider
> >> > > > > >> > > > > having a tailored API to do simple membership/leader
> >> > election.
> >> > > > > >> > > > >
> >> > > > > >> > > > > Best,
> >> > > > > >> > > > > David
> >> > > > > >> > > > >
> >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> >> > ismael@juma.me.uk
> >> > > > > >
> >> > > > > >> > > wrote:
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > Three quick comments:
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> >> > > > > >> com.google.re2j.Pattern, we
> >> > > > > >> > > > > should
> >> > > > > >> > > > > > document the differences in more detail before
> >> deciding
> >> > one
> >> > > > > way
> >> > > > > >> or
> >> > > > > >> > > > > another.
> >> > > > > >> > > > > > That said, if people pass java.util.regex.Pattern,
> >> they
> >> > expect
> >> > > > > >> > their
> >> > > > > >> > > > > > semantics to be honored. If we are doing something
> >> > different,
> >> > > > > >> then
> >> > > > > >> > we
> >> > > > > >> > > > > > should consider adding an overload with our own
> >> Pattern
> >> > class
> >> > > > > (I
> >> > > > > >> > > don't
> >> > > > > >> > > > > > think we'd want to expose re2j's at this point).
> >> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol should
> >> > > > > integrate
> >> > > > > >> > fully
> >> > > > > >> > > > > with
> >> > > > > >> > > > > > it and should handle the topic recreation case
> >> > correctly.
> >> > > > > That's
> >> > > > > >> > the
> >> > > > > >> > > main
> >> > > > > >> > > > > > part we need to handle. I agree with David that we'd
> >> > want to
> >> > > > > add
> >> > > > > >> > > topic
> >> > > > > >> > > > > ids
> >> > > > > >> > > > > > to the relevant protocols that don't have it yet and
> >> > that we
> >> > > > > can
> >> > > > > >> > > probably
> >> > > > > >> > > > > > focus on the internals versus adding new APIs to the
> >> > Java
> >> > > > > >> Consumer
> >> > > > > >> > > > > (unless
> >> > > > > >> > > > > > we find that adding new APIs is required for
> >> reasonable
> >> > > > > >> semantics).
> >> > > > > >> > > > > > 3. I am still not sure about the coordinator storing
> >> the
> >> > > > > >> configs.
> >> > > > > >> > > It's
> >> > > > > >> > > > > > powerful for configs to be centralized in the
> >> metadata
> >> > log for
> >> > > > > >> > > various
> >> > > > > >> > > > > > reasons (auditability, visibility, consistency,
> >> etc.).
> >> > > > > >> Similarly, I
> >> > > > > >> > > am
> >> > > > > >> > > > > not
> >> > > > > >> > > > > > sure about automatically deleting configs in a way
> >> that
> >> > they
> >> > > > > >> cannot
> >> > > > > >> > > be
> >> > > > > >> > > > > > recovered. A good property for modern systems is to
> >> > minimize
> >> > > > > the
> >> > > > > >> > > number
> >> > > > > >> > > > > of
> >> > > > > >> > > > > > unrecoverable data loss scenarios.
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > Ismael
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> >> > > > > >> > > <djacot@confluent.io.invalid
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > wrote:
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > > 1) the migration path, especially the last step
> >> when
> >> > > > > clients
> >> > > > > >> > > flip the
> >> > > > > >> > > > > > > flag
> >> > > > > >> > > > > > > > to enable the new protocol, in which we would
> >> have a
> >> > > > > window
> >> > > > > >> > where
> >> > > > > >> > > > > both
> >> > > > > >> > > > > > > new
> >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
> >> used
> >> > by
> >> > > > > >> members
> >> > > > > >> > of
> >> > > > > >> > > the
> >> > > > > >> > > > > same
> >> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> >> > behavior
> >> > > > > >> while
> >> > > > > >> > > > > using the
> >> > > > > >> > > > > > > > new protocol is something we need to present
> >> about.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > Noted. I just published a new version of KIP which
> >> > includes
> >> > > > > >> more
> >> > > > > >> > > > > > > details about this. See the "Supporting Online
> >> > Consumer
> >> > > > > Group
> >> > > > > >> > > Upgrade"
> >> > > > > >> > > > > > > and the "Compatibility, Deprecation, and Migration
> >> > Plan". I
> >> > > > > >> think
> >> > > > > >> > > that
> >> > > > > >> > > > > > > I have to think through a few cases now but the
> >> > overall idea
> >> > > > > >> and
> >> > > > > >> > > > > > > mechanism should be understandable.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> >> > topic ids
> >> > > > > >> are
> >> > > > > >> > > only
> >> > > > > >> > > > > used
> >> > > > > >> > > > > > > as
> >> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> >> > exposed
> >> > > > > via
> >> > > > > >> any
> >> > > > > >> > > > > public
> >> > > > > >> > > > > > > APIs
> >> > > > > >> > > > > > > > to consumers yet. I think the question is, first
> >> > should we
> >> > > > > >> let
> >> > > > > >> > > the
> >> > > > > >> > > > > > > consumer
> >> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> >> > itself
> >> > > > > to
> >> > > > > >> > fully
> >> > > > > >> > > > > > > leverage
> >> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> >> RPCs
> >> > with
> >> > > > > the
> >> > > > > >> > > topic
> >> > > > > >> > > > > ids;
> >> > > > > >> > > > > > > and
> >> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> >> > topic ids
> >> > > > > in
> >> > > > > >> the
> >> > > > > >> > > > > consumer
> >> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> >> well
> >> > as in
> >> > > > > >> the
> >> > > > > >> > > > > rebalance
> >> > > > > >> > > > > > > > listener for cases like topic
> >> > deletion-and-recreation).
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > a) Assuming that we would include converting all
> >> the
> >> > offsets
> >> > > > > >> > > related
> >> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer
> >> > would be
> >> > > > > >> able
> >> > > > > >> > to
> >> > > > > >> > > > > > > fully operate with topic ids. That being said, it
> >> > still has
> >> > > > > to
> >> > > > > >> > > provide
> >> > > > > >> > > > > > > the topics names in various APIs so having a
> >> mapping
> >> > in the
> >> > > > > >> > > consumer
> >> > > > > >> > > > > > > seems inevitable to me.
> >> > > > > >> > > > > > > b) I don't have a strong opinion on this. Here I
> >> > wonder if
> >> > > > > >> this
> >> > > > > >> > > goes
> >> > > > > >> > > > > > > beyond the scope of this KIP. I would rather focus
> >> on
> >> > the
> >> > > > > >> > internals
> >> > > > > >> > > > > > > here and we can consider this separately if we see
> >> > value in
> >> > > > > >> doing
> >> > > > > >> > > it.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > Coming back to Ismael's point about using topic ids
> >> > in the
> >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there
> >> is
> >> > one
> >> > > > > >> > advantage
> >> > > > > >> > > in
> >> > > > > >> > > > > > > favour of it. The consumer will have the
> >> opportunity
> >> > to
> >> > > > > >> validate
> >> > > > > >> > > that
> >> > > > > >> > > > > > > the topics exists before passing them into the
> >> group
> >> > > > > rebalance
> >> > > > > >> > > > > > > protocol. Obviously, the coordinator will also
> >> notice
> >> > it but
> >> > > > > >> it
> >> > > > > >> > > does
> >> > > > > >> > > > > > > not really have a way to reject an invalid topic in
> >> > the
> >> > > > > >> response.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> >> questions
> >> > > > > except
> >> > > > > >> for
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> >> > it's not
> >> > > > > >> > > necessary
> >> > > > > >> > > > > to
> >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> >> > instead we
> >> > > > > >> still
> >> > > > > >> > > use
> >> > > > > >> > > > > > > Pattern
> >> > > > > >> > > > > > > > while just documenting that our subscription may
> >> be
> >> > > > > >> rejected by
> >> > > > > >> > > the
> >> > > > > >> > > > > > > server.
> >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> >> scenario
> >> > I felt
> >> > > > > >> > using
> >> > > > > >> > > an
> >> > > > > >> > > > > > > > overloaded `String` based subscription may be
> >> more
> >> > > > > >> vulnerable
> >> > > > > >> > to
> >> > > > > >> > > > > various
> >> > > > > >> > > > > > > > invalid regexes.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > That could work. I have to look at the differences
> >> > between
> >> > > > > the
> >> > > > > >> > two
> >> > > > > >> > > > > > > engines to better understand the potential issues.
> >> My
> >> > > > > >> > > understanding is
> >> > > > > >> > > > > > > that would work for all the basic regular
> >> > expressions. The
> >> > > > > >> > > differences
> >> > > > > >> > > > > > > between the two are mainly about the various
> >> character
> >> > > > > >> classes. I
> >> > > > > >> > > > > > > wonder what other people think about this.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > Best,
> >> > > > > >> > > > > > > David
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> >> > > > > >> > wangguoz@gmail.com
> >> > > > > >> > > >
> >> > > > > >> > > > > wrote:
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > Thanks David! I think on the high level there are
> >> > two meta
> >> > > > > >> > > points we
> >> > > > > >> > > > > need
> >> > > > > >> > > > > > > > to concretize a bit more:
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > 1) the migration path, especially the last step
> >> when
> >> > > > > clients
> >> > > > > >> > > flip the
> >> > > > > >> > > > > > > flag
> >> > > > > >> > > > > > > > to enable the new protocol, in which we would
> >> have a
> >> > > > > window
> >> > > > > >> > where
> >> > > > > >> > > > > both
> >> > > > > >> > > > > > > new
> >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
> >> used
> >> > by
> >> > > > > >> members
> >> > > > > >> > of
> >> > > > > >> > > the
> >> > > > > >> > > > > same
> >> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> >> > behavior
> >> > > > > >> while
> >> > > > > >> > > > > using the
> >> > > > > >> > > > > > > > new protocol is something we need to present
> >> about.
> >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> >> > topic ids
> >> > > > > >> are
> >> > > > > >> > > only
> >> > > > > >> > > > > used
> >> > > > > >> > > > > > > as
> >> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> >> > exposed
> >> > > > > via
> >> > > > > >> any
> >> > > > > >> > > > > public
> >> > > > > >> > > > > > > APIs
> >> > > > > >> > > > > > > > to consumers yet. I think the question is, first
> >> > should we
> >> > > > > >> let
> >> > > > > >> > > the
> >> > > > > >> > > > > > > consumer
> >> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> >> > itself
> >> > > > > to
> >> > > > > >> > fully
> >> > > > > >> > > > > > > leverage
> >> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> >> RPCs
> >> > with
> >> > > > > the
> >> > > > > >> > > topic
> >> > > > > >> > > > > ids;
> >> > > > > >> > > > > > > and
> >> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> >> > topic ids
> >> > > > > in
> >> > > > > >> the
> >> > > > > >> > > > > consumer
> >> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> >> well
> >> > as in
> >> > > > > >> the
> >> > > > > >> > > > > rebalance
> >> > > > > >> > > > > > > > listener for cases like topic
> >> > deletion-and-recreation).
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> >> questions
> >> > > > > except
> >> > > > > >> for
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> >> > it's not
> >> > > > > >> > > necessary
> >> > > > > >> > > > > to
> >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> >> > instead we
> >> > > > > >> still
> >> > > > > >> > > use
> >> > > > > >> > > > > > > Pattern
> >> > > > > >> > > > > > > > while just documenting that our subscription may
> >> be
> >> > > > > >> rejected by
> >> > > > > >> > > the
> >> > > > > >> > > > > > > server.
> >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> >> scenario
> >> > I felt
> >> > > > > >> > using
> >> > > > > >> > > an
> >> > > > > >> > > > > > > > overloaded `String` based subscription may be
> >> more
> >> > > > > >> vulnerable
> >> > > > > >> > to
> >> > > > > >> > > > > various
> >> > > > > >> > > > > > > > invalid regexes.
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > Guozhang
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> >> > > > > >> > > > > <djacot@confluent.io.invalid
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > wrote:
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > > Hi Ismael,
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer your
> >> > questions
> >> > > > > >> > inline.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 1. I think it's premature to talk about
> >> target
> >> > > > > versions
> >> > > > > >> for
> >> > > > > >> > > > > > > deprecation
> >> > > > > >> > > > > > > > > and
> >> > > > > >> > > > > > > > > > removal of the existing group protocol.
> >> Unlike
> >> > KRaft,
> >> > > > > >> this
> >> > > > > >> > > > > affects a
> >> > > > > >> > > > > > > core
> >> > > > > >> > > > > > > > > > client protocol and hence deprecation/removal
> >> > will be
> >> > > > > >> > heavily
> >> > > > > >> > > > > > > dependent
> >> > > > > >> > > > > > > > > on
> >> > > > > >> > > > > > > > > > how quickly applications migrate to the new
> >> > protocol.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 2. The KIP says we intend to release this in
> >> > 4.x, but
> >> > > > > it
> >> > > > > >> > > wasn't
> >> > > > > >> > > > > made
> >> > > > > >> > > > > > > > > clear
> >> > > > > >> > > > > > > > > > why. If we added that as a way to estimate
> >> when
> >> > we'd
> >> > > > > >> > > deprecate
> >> > > > > >> > > > > and
> >> > > > > >> > > > > > > remove
> >> > > > > >> > > > > > > > > > the group protocol, I also suggest removing
> >> > this part.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Let me explain my reasoning. As explained, I
> >> plan
> >> > to
> >> > > > > >> rewrite
> >> > > > > >> > > the
> >> > > > > >> > > > > group
> >> > > > > >> > > > > > > > > coordinator in Java while we implement the new
> >> > protocol.
> >> > > > > >> This
> >> > > > > >> > > means
> >> > > > > >> > > > > > > > > that the internals will be slightly different
> >> > (e.g.
> >> > > > > >> threading
> >> > > > > >> > > > > model).
> >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the switch from
> >> > the old
> >> > > > > >> group
> >> > > > > >> > > > > > > > > coordinator to the new group coordinator to a
> >> > major
> >> > > > > >> release.
> >> > > > > >> > > The
> >> > > > > >> > > > > > > > > alternative would be to use a flag to do the
> >> > switch
> >> > > > > >> instead
> >> > > > > >> > of
> >> > > > > >> > > > > relying
> >> > > > > >> > > > > > > > > on the software upgrade.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 3. We need to flesh out the details of the
> >> > migration
> >> > > > > >> story.
> >> > > > > >> > > It
> >> > > > > >> > > > > sounds
> >> > > > > >> > > > > > > > > like
> >> > > > > >> > > > > > > > > > we're saying we will support online
> >> migrations.
> >> > Is
> >> > > > > that
> >> > > > > >> > > correct?
> >> > > > > >> > > > > We
> >> > > > > >> > > > > > > > > should
> >> > > > > >> > > > > > > > > > explain this in detail. It could also be done
> >> > as a
> >> > > > > >> separate
> >> > > > > >> > > KIP,
> >> > > > > >> > > > > if
> >> > > > > >> > > > > > > it's
> >> > > > > >> > > > > > > > > > easier.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Yes, we will support online migrations for the
> >> > group.
> >> > > > > That
> >> > > > > >> > > means
> >> > > > > >> > > > > that
> >> > > > > >> > > > > > > > > a group using the old protocol will be able to
> >> > switch to
> >> > > > > >> the
> >> > > > > >> > > new
> >> > > > > >> > > > > > > > > protocol.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Let me briefly explain how that will work
> >> though.
> >> > It is
> >> > > > > >> > > basically a
> >> > > > > >> > > > > > > > > four step process:
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a
> >> > software
> >> > > > > >> > > supporting
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > new group coordinator. Both the old and the new
> >> > > > > >> coordinator
> >> > > > > >> > > will
> >> > > > > >> > > > > > > > > support the old protocol and rely on the same
> >> > persisted
> >> > > > > >> > > metadata so
> >> > > > > >> > > > > > > > > they can work together. This point is an
> >> offline
> >> > > > > >> migration.
> >> > > > > >> > We
> >> > > > > >> > > > > cannot
> >> > > > > >> > > > > > > > > do this one live because it would require
> >> > shutting down
> >> > > > > >> the
> >> > > > > >> > > current
> >> > > > > >> > > > > > > > > coordinator and starting up the new one and
> >> that
> >> > would
> >> > > > > >> cause
> >> > > > > >> > > > > > > > > unavailabilities.
> >> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be
> >> > upgraded
> >> > > > > to
> >> > > > > >> X
> >> > > > > >> > in
> >> > > > > >> > > > > order
> >> > > > > >> > > > > > > > > to enable the new protocol. This cannot be done
> >> > before
> >> > > > > 1)
> >> > > > > >> is
> >> > > > > >> > > > > > > > > terminated because the old coordinator doesn't
> >> > support
> >> > > > > the
> >> > > > > >> > new
> >> > > > > >> > > > > > > > > protocol.
> >> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a version
> >> > > > > supporting
> >> > > > > >> the
> >> > > > > >> > > > > online
> >> > > > > >> > > > > > > > > migration (must have KIP-792). If the consumer
> >> is
> >> > > > > already
> >> > > > > >> > > there.
> >> > > > > >> > > > > > > > > Nothing must be done at this point.
> >> > > > > >> > > > > > > > > 4. The consumers must be rolled with the
> >> feature
> >> > flag
> >> > > > > >> turned
> >> > > > > >> > > on.
> >> > > > > >> > > > > The
> >> > > > > >> > > > > > > > > consumer group is automatically converted when
> >> > the first
> >> > > > > >> > > consumer
> >> > > > > >> > > > > > > > > using the new protocol joins the group. While
> >> the
> >> > > > > members
> >> > > > > >> > > using the
> >> > > > > >> > > > > > > > > old protocol are being upgraded, the old
> >> protocol
> >> > is
> >> > > > > >> proxied
> >> > > > > >> > > into
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > new one.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> >> > > > > >> subscriptions
> >> > > > > >> > > to
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > server,
> >> > > > > >> > > > > > > > > > but it seems like there could be some tricky
> >> > > > > >> compatibility
> >> > > > > >> > > > > issues.
> >> > > > > >> > > > > > > Will
> >> > > > > >> > > > > > > > > we
> >> > > > > >> > > > > > > > > > have a mechanism for users to detect that
> >> they
> >> > need to
> >> > > > > >> > update
> >> > > > > >> > > > > their
> >> > > > > >> > > > > > > regex
> >> > > > > >> > > > > > > > > > before switching to the new protocol?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I think that I am a bit more optimistic than
> >> you
> >> > on this
> >> > > > > >> > > point. I
> >> > > > > >> > > > > > > > > believe that the majority of the cases are
> >> simple
> >> > > > > regexes
> >> > > > > >> > which
> >> > > > > >> > > > > should
> >> > > > > >> > > > > > > > > work with the new engine. The coordinator will
> >> > verify
> >> > > > > the
> >> > > > > >> > regex
> >> > > > > >> > > > > anyway
> >> > > > > >> > > > > > > > > and reject the consumer if the regex is not
> >> valid.
> >> > > > > Coming
> >> > > > > >> > back
> >> > > > > >> > > to
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > migration path, in the worst case, the first
> >> > upgraded
> >> > > > > >> > consumer
> >> > > > > >> > > > > joining
> >> > > > > >> > > > > > > > > the group will be rejected. This should be used
> >> > as the
> >> > > > > >> last
> >> > > > > >> > > > > defence, I
> >> > > > > >> > > > > > > > > would say.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > One way for customers to validate their regex
> >> > before
> >> > > > > >> > upgrading
> >> > > > > >> > > > > their
> >> > > > > >> > > > > > > > > prod would be to test them with another group.
> >> For
> >> > > > > >> instance,
> >> > > > > >> > > that
> >> > > > > >> > > > > > > > > could be done in a pre-prod environment.
> >> Another
> >> > way
> >> > > > > >> would be
> >> > > > > >> > > to
> >> > > > > >> > > > > > > > > extend the consumer-group tool to provide a
> >> regex
> >> > > > > >> validation
> >> > > > > >> > > > > > > > > mechanism. Would this be enough in your
> >> opinion?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 5. Related to the last question, will the
> >> Java
> >> > client
> >> > > > > >> allow
> >> > > > > >> > > the
> >> > > > > >> > > > > > > users to
> >> > > > > >> > > > > > > > > > stick with the current regex engine for
> >> > compatibility
> >> > > > > >> > > reasons?
> >> > > > > >> > > > > For
> >> > > > > >> > > > > > > > > example,
> >> > > > > >> > > > > > > > > > it may be handy to keep using client based
> >> > regex at
> >> > > > > >> first
> >> > > > > >> > to
> >> > > > > >> > > keep
> >> > > > > >> > > > > > > > > > migrations simple and then migrate to server
> >> > based
> >> > > > > >> regexes
> >> > > > > >> > > as a
> >> > > > > >> > > > > > > second
> >> > > > > >> > > > > > > > > step.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I understand your point but I am concerned that
> >> > this
> >> > > > > would
> >> > > > > >> > > allow
> >> > > > > >> > > > > users
> >> > > > > >> > > > > > > > > to actually stay in this mode. That would go
> >> > against our
> >> > > > > >> goal
> >> > > > > >> > > of
> >> > > > > >> > > > > > > > > simplifying the client because we would have to
> >> > continue
> >> > > > > >> > > monitoring
> >> > > > > >> > > > > > > > > the metadata on the client side. I would rather
> >> > not do
> >> > > > > >> this.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 6. When we say that the group coordinator
> >> will
> >> > be
> >> > > > > >> > > responsible for
> >> > > > > >> > > > > > > storing
> >> > > > > >> > > > > > > > > > the configurations and that the
> >> configurations
> >> > will be
> >> > > > > >> > > deleted
> >> > > > > >> > > > > when
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > > group is deleted. Will a transition to DEAD
> >> > trigger
> >> > > > > >> > deletion
> >> > > > > >> > > of
> >> > > > > >> > > > > > > > > > configurations?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > That's right. The configurations will be
> >> deleted
> >> > when
> >> > > > > the
> >> > > > > >> > > group is
> >> > > > > >> > > > > > > > > deleted. They go together.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 7. Will the choice to store the configs in
> >> the
> >> > group
> >> > > > > >> > > coordinator
> >> > > > > >> > > > > > > make it
> >> > > > > >> > > > > > > > > > harder to list all cluster configs and their
> >> > values?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I don't think so. The group configurations are
> >> > overrides
> >> > > > > >> of
> >> > > > > >> > > cluster
> >> > > > > >> > > > > > > > > configs. If you want to know all the overrides
> >> > though,
> >> > > > > you
> >> > > > > >> > > would
> >> > > > > >> > > > > have
> >> > > > > >> > > > > > > > > to ask all the group coordinators. You cannot
> >> > rely on
> >> > > > > the
> >> > > > > >> > > metadata
> >> > > > > >> > > > > log
> >> > > > > >> > > > > > > > > for instance.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 8. How would someone configure a group before
> >> > starting
> >> > > > > >> the
> >> > > > > >> > > > > consumers?
> >> > > > > >> > > > > > > > > Have
> >> > > > > >> > > > > > > > > > we considered allowing the explicit creation
> >> of
> >> > > > > groups?
> >> > > > > >> > > > > > > Alternatively,
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > configs could be decoupled from the group
> >> > lifecycle.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Yes. The group will be automatically created in
> >> > this
> >> > > > > case.
> >> > > > > >> > > However,
> >> > > > > >> > > > > > > > > the configs will be lost after the retention
> >> > period of
> >> > > > > the
> >> > > > > >> > > group
> >> > > > > >> > > > > > > > > passes.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the
> >> > Java
> >> > > > > >> client
> >> > > > > >> > > still
> >> > > > > >> > > > > take
> >> > > > > >> > > > > > > a
> >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to
> >> > introduce
> >> > > > > an
> >> > > > > >> > > overload?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > That's a very group question. I forgot about
> >> that
> >> > one.
> >> > > > > As
> >> > > > > >> the
> >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
> >> compatible
> >> > with
> >> > > > > the
> >> > > > > >> > > engine
> >> > > > > >> > > > > that
> >> > > > > >> > > > > > > > > we plan to use, it might be better to deprecate
> >> > it and
> >> > > > > >> use an
> >> > > > > >> > > > > overload
> >> > > > > >> > > > > > > > > which takes a string. We would rely on the
> >> server
> >> > side
> >> > > > > >> > > validation.
> >> > > > > >> > > > > > > > > During the migration, I think that we could
> >> still
> >> > try to
> >> > > > > >> > > toString
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > regex and use it. That should work, I think, in
> >> > the
> >> > > > > >> majority
> >> > > > > >> > > of the
> >> > > > > >> > > > > > > > > cases.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 10. I agree with Justine that we should be
> >> > clearer
> >> > > > > about
> >> > > > > >> > the
> >> > > > > >> > > > > reason
> >> > > > > >> > > > > > > to
> >> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
> >> feature
> >> > flag.
> >> > > > > >> Maybe
> >> > > > > >> > > we
> >> > > > > >> > > > > mean
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > we
> >> > > > > >> > > > > > > > > > can switch the default for the feature flag
> >> to
> >> > true
> >> > > > > >> based
> >> > > > > >> > on
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > > metadata.version once we want to make it the
> >> > default.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > My plan was to use that feature flag mainly
> >> > during the
> >> > > > > >> > > development
> >> > > > > >> > > > > > > > > phase. I should not have mentioned it, I think,
> >> > because
> >> > > > > we
> >> > > > > >> > > could
> >> > > > > >> > > > > use
> >> > > > > >> > > > > > > > > an internal config for it.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention
> >> the
> >> > > > > required
> >> > > > > >> > > ACLs, it
> >> > > > > >> > > > > > > would
> >> > > > > >> > > > > > > > > be
> >> > > > > >> > > > > > > > > > good to add that for consistency.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Noted.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> >> ConsumerGroupHeartbeat
> >> > > > > requires
> >> > > > > >> > > "Read
> >> > > > > >> > > > > Group"
> >> > > > > >> > > > > > > > > even
> >> > > > > >> > > > > > > > > > though it seems to do more than reading.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I agree. This is how the current protocol works
> >> > though.
> >> > > > > We
> >> > > > > >> > only
> >> > > > > >> > > > > > > > > require "Read Group" to join a group. We could
> >> > consider
> >> > > > > >> > > changing
> >> > > > > >> > > > > this
> >> > > > > >> > > > > > > > > but I am not sure that it is worth it.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 13. How is topic recreation handled by the
> >> > consumer
> >> > > > > with
> >> > > > > >> > the
> >> > > > > >> > > new
> >> > > > > >> > > > > > > group
> >> > > > > >> > > > > > > > > > protocol? It would be good to have a section
> >> on
> >> > this.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Noted. From a protocol perspective, the new
> >> topic
> >> > will
> >> > > > > >> have a
> >> > > > > >> > > new
> >> > > > > >> > > > > > > > > topic id so it will treat it like a topic with
> >> a
> >> > > > > different
> >> > > > > >> > > name.
> >> > > > > >> > > > > The
> >> > > > > >> > > > > > > > > only issue is that the fetch/commit offsets
> >> APIs
> >> > do not
> >> > > > > >> > support
> >> > > > > >> > > > > topic
> >> > > > > >> > > > > > > > > IDs so the consumer would reuse the offsets
> >> based
> >> > on the
> >> > > > > >> > same.
> >> > > > > >> > > I
> >> > > > > >> > > > > think
> >> > > > > >> > > > > > > > > that we should update those APIs as well in
> >> order
> >> > to be
> >> > > > > >> > > consistent
> >> > > > > >> > > > > end
> >> > > > > >> > > > > > > > > to end. That would strengthen the semantics of
> >> the
> >> > > > > >> consumer.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the new
> >> > coordinator
> >> > > > > >> in
> >> > > > > >> > > Java.
> >> > > > > >> > > > > Even
> >> > > > > >> > > > > > > > > though
> >> > > > > >> > > > > > > > > > this is an implementation detail, do we plan
> >> to
> >> > have a
> >> > > > > >> new
> >> > > > > >> > > gradle
> >> > > > > >> > > > > > > module
> >> > > > > >> > > > > > > > > > for it?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Yes.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when it
> >> comes
> >> > to how
> >> > > > > >> many
> >> > > > > >> > > > > members
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > new
> >> > > > > >> > > > > > > > > > group protocol can support?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > We don't have numbers at the moment. The
> >> protocol
> >> > should
> >> > > > > >> > > support
> >> > > > > >> > > > > 1000s
> >> > > > > >> > > > > > > > > of members per group. We will measure this when
> >> > we have
> >> > > > > a
> >> > > > > >> > first
> >> > > > > >> > > > > > > > > implementation. Note that we might have other
> >> > > > > bottlenecks
> >> > > > > >> > down
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > road (e.g. offset commits).
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> >> > instead
> >> > > > > of
> >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> >> > ConsumerGroupHeartbeatRequest?
> >> > > > > >> Is
> >> > > > > >> > the
> >> > > > > >> > > > > idea
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > > since we have to resolve the regex on the
> >> > server, we
> >> > > > > >> can do
> >> > > > > >> > > the
> >> > > > > >> > > > > same
> >> > > > > >> > > > > > > for
> >> > > > > >> > > > > > > > > > the topic name? The difference is that
> >> sending
> >> > the
> >> > > > > >> regex is
> >> > > > > >> > > more
> >> > > > > >> > > > > > > > > efficient
> >> > > > > >> > > > > > > > > > whereas sending the topic names is less
> >> > efficient.
> >> > > > > >> > > Furthermore,
> >> > > > > >> > > > > > > delete
> >> > > > > >> > > > > > > > > and
> >> > > > > >> > > > > > > > > > recreation is easier to handle if we have
> >> topic
> >> > ids.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > The idea was to consolidate the metadata lookup
> >> > on the
> >> > > > > >> server
> >> > > > > >> > > for
> >> > > > > >> > > > > both
> >> > > > > >> > > > > > > > > paths but I do agree with your point. As a
> >> second
> >> > > > > though,
> >> > > > > >> > using
> >> > > > > >> > > > > topic
> >> > > > > >> > > > > > > > > ids may be better here for the delete and
> >> > recreation
> >> > > > > case.
> >> > > > > >> > > Also, I
> >> > > > > >> > > > > > > > > suppose that we may allow users to subscribe
> >> with
> >> > topic
> >> > > > > >> ids
> >> > > > > >> > in
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > future because that is the only way to be
> >> really
> >> > robust
> >> > > > > to
> >> > > > > >> > > topic
> >> > > > > >> > > > > > > > > re-creation.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Best,
> >> > > > > >> > > > > > > > > David
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> >> > > > > >> > > djacot@confluent.io>
> >> > > > > >> > > > > > > wrote:
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > Hi Justine,
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > Thanks for your comments. Please find my
> >> answers
> >> > > > > below.
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs
> >> > with the
> >> > > > > >> > > exception
> >> > > > > >> > > > > of the
> >> > > > > >> > > > > > > > > > topic names based in the
> >> > > > > ConsumerGroupHeartbeatRequest.
> >> > > > > >> I
> >> > > > > >> > am
> >> > > > > >> > > not
> >> > > > > >> > > > > sure
> >> > > > > >> > > > > > > > > > if using topic names is the right call here.
> >> I
> >> > need to
> >> > > > > >> > think
> >> > > > > >> > > > > about it
> >> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does not
> >> > change the
> >> > > > > >> > > > > fetch/commit
> >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be
> >> > something
> >> > > > > >> that
> >> > > > > >> > we
> >> > > > > >> > > > > should
> >> > > > > >> > > > > > > > > > include though as it would give better
> >> overall
> >> > > > > >> guarantee in
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > > producer.
> >> > > > > >> > > > > > > > > > - You're right. I think that I should not
> >> have
> >> > > > > mentioned
> >> > > > > >> > this
> >> > > > > >> > > > > flag at
> >> > > > > >> > > > > > > > > > all. I will remove it. We can use an internal
> >> > > > > >> configuration
> >> > > > > >> > > while
> >> > > > > >> > > > > > > > > > developing the feature.
> >> > > > > >> > > > > > > > > > - Both cluster types will be supported. The
> >> > change is
> >> > > > > >> > > > > orthogonal. The
> >> > > > > >> > > > > > > > > > only requirement is that the cluster uses
> >> topic
> >> > IDs.
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > Best,
> >> > > > > >> > > > > > > > > > David
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang
> >> Wang <
> >> > > > > >> > > > > wangguoz@gmail.com>
> >> > > > > >> > > > > > > > > wrote:
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Hi Ismael,
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are some
> >> replies
> >> > > > > inlined
> >> > > > > >> > > below:
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma
> >> <
> >> > > > > >> > > ismael@juma.me.uk>
> >> > > > > >> > > > > > > wrote:
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the
> >> potential
> >> > to be a
> >> > > > > >> > great
> >> > > > > >> > > > > > > > > improvement. A few
> >> > > > > >> > > > > > > > > > > > initial questions/comments:
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk about
> >> > target
> >> > > > > >> versions
> >> > > > > >> > > for
> >> > > > > >> > > > > > > > > deprecation and
> >> > > > > >> > > > > > > > > > > > removal of the existing group protocol.
> >> > Unlike
> >> > > > > >> KRaft,
> >> > > > > >> > > this
> >> > > > > >> > > > > > > affects a
> >> > > > > >> > > > > > > > > core
> >> > > > > >> > > > > > > > > > > > client protocol and hence
> >> > deprecation/removal will
> >> > > > > >> be
> >> > > > > >> > > heavily
> >> > > > > >> > > > > > > > > dependent on
> >> > > > > >> > > > > > > > > > > > how quickly applications migrate to the
> >> new
> >> > > > > >> protocol.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can
> >> remove
> >> > the
> >> > > > > >> proposed
> >> > > > > >> > > > > timeline
> >> > > > > >> > > > > > > in
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration
> >> > Plan` and
> >> > > > > >> > > instead
> >> > > > > >> > > > > just
> >> > > > > >> > > > > > > state
> >> > > > > >> > > > > > > > > > > that we will decide in the future about
> >> when
> >> > we
> >> > > > > would
> >> > > > > >> > > > > deprecate old
> >> > > > > >> > > > > > > > > > > protocol and behaviors.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to release this
> >> > in 4.x,
> >> > > > > >> but
> >> > > > > >> > it
> >> > > > > >> > > > > wasn't
> >> > > > > >> > > > > > > made
> >> > > > > >> > > > > > > > > clear
> >> > > > > >> > > > > > > > > > > > why. If we added that as a way to
> >> estimate
> >> > when
> >> > > > > we'd
> >> > > > > >> > > > > deprecate
> >> > > > > >> > > > > > > and
> >> > > > > >> > > > > > > > > remove
> >> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
> >> removing
> >> > this
> >> > > > > >> part.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > I think that's not specifically related to
> >> the
> >> > > > > >> > > > > deprecation/removal
> >> > > > > >> > > > > > > > > timeline
> >> > > > > >> > > > > > > > > > > plan, but it's more for client upgrades.
> >> I.e.
> >> > the
> >> > > > > >> > > broker-side
> >> > > > > >> > > > > > > > > > > implementation may be done first, and then
> >> the
> >> > > > > client
> >> > > > > >> > side,
> >> > > > > >> > > > > and we
> >> > > > > >> > > > > > > > > would
> >> > > > > >> > > > > > > > > > > only mark it as "released" by the time
> >> clients
> >> > > > > >> > > implementations
> >> > > > > >> > > > > are
> >> > > > > >> > > > > > > > > done. At
> >> > > > > >> > > > > > > > > > > that time, to enable the feature the
> >> clients
> >> > need to
> >> > > > > >> > first
> >> > > > > >> > > > > swap-in
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and then set
> >> > the flag
> >> > > > > >> > with a
> >> > > > > >> > > > > second
> >> > > > > >> > > > > > > > > rolling
> >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's better to be
> >> > released
> >> > > > > >> in a
> >> > > > > >> > > major
> >> > > > > >> > > > > > > > > version.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the details of
> >> the
> >> > > > > migration
> >> > > > > >> > > story.
> >> > > > > >> > > > > It
> >> > > > > >> > > > > > > > > sounds like
> >> > > > > >> > > > > > > > > > > > we're saying we will support online
> >> > migrations. Is
> >> > > > > >> that
> >> > > > > >> > > > > correct?
> >> > > > > >> > > > > > > We
> >> > > > > >> > > > > > > > > should
> >> > > > > >> > > > > > > > > > > > explain this in detail. It could also be
> >> > done as a
> >> > > > > >> > > separate
> >> > > > > >> > > > > KIP,
> >> > > > > >> > > > > > > if
> >> > > > > >> > > > > > > > > it's
> >> > > > > >> > > > > > > > > > > > easier.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yes I think that's the part we can be more
> >> > concrete
> >> > > > > >> about
> >> > > > > >> > > for
> >> > > > > >> > > > > sure
> >> > > > > >> > > > > > > (and
> >> > > > > >> > > > > > > > > > > this is related to your question 2) above).
> >> > We will
> >> > > > > >> work
> >> > > > > >> > on
> >> > > > > >> > > > > making
> >> > > > > >> > > > > > > it
> >> > > > > >> > > > > > > > > more
> >> > > > > >> > > > > > > > > > > explicit in parallel as we solicit more
> >> > feedback.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing the
> >> > pattern
> >> > > > > >> > > subscriptions
> >> > > > > >> > > > > to
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > server,
> >> > > > > >> > > > > > > > > > > > but it seems like there could be some
> >> tricky
> >> > > > > >> > > compatibility
> >> > > > > >> > > > > > > issues.
> >> > > > > >> > > > > > > > > Will we
> >> > > > > >> > > > > > > > > > > > have a mechanism for users to detect that
> >> > they
> >> > > > > need
> >> > > > > >> to
> >> > > > > >> > > update
> >> > > > > >> > > > > > > their
> >> > > > > >> > > > > > > > > regex
> >> > > > > >> > > > > > > > > > > > before switching to the new protocol?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
> >> non-java
> >> > client
> >> > > > > >> > users
> >> > > > > >> > > to
> >> > > > > >> > > > > sort
> >> > > > > >> > > > > > > of
> >> > > > > >> > > > > > > > > > > "dry-run" the client before switching to
> >> the
> >> > new
> >> > > > > >> > protocol.
> >> > > > > >> > > I
> >> > > > > >> > > > > do not
> >> > > > > >> > > > > > > > > have a
> >> > > > > >> > > > > > > > > > > specific idea on top of my head though,
> >> maybe
> >> > others
> >> > > > > >> like
> >> > > > > >> > > @Matt
> >> > > > > >> > > > > > > > > Howlett can
> >> > > > > >> > > > > > > > > > > chime-in here?
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 5. Related to the last question, will the
> >> > Java
> >> > > > > >> client
> >> > > > > >> > > allow
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > users to
> >> > > > > >> > > > > > > > > > > > stick with the current regex engine for
> >> > > > > >> compatibility
> >> > > > > >> > > > > reasons?
> >> > > > > >> > > > > > > For
> >> > > > > >> > > > > > > > > example,
> >> > > > > >> > > > > > > > > > > > it may be handy to keep using client
> >> based
> >> > regex
> >> > > > > at
> >> > > > > >> > > first to
> >> > > > > >> > > > > keep
> >> > > > > >> > > > > > > > > > > > migrations simple and then migrate to
> >> > server based
> >> > > > > >> > > regexes
> >> > > > > >> > > > > as a
> >> > > > > >> > > > > > > > > second
> >> > > > > >> > > > > > > > > > > > step.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Honestly I have not thought about that for
> >> > java
> >> > > > > >> clients,
> >> > > > > >> > > and
> >> > > > > >> > > > > we can
> >> > > > > >> > > > > > > > > discuss
> >> > > > > >> > > > > > > > > > > that. What kind of compatibility issues do
> >> > you have
> >> > > > > in
> >> > > > > >> > > mind?
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 6. When we say that the group coordinator
> >> > will be
> >> > > > > >> > > > > responsible for
> >> > > > > >> > > > > > > > > storing
> >> > > > > >> > > > > > > > > > > > the configurations and that the
> >> > configurations
> >> > > > > will
> >> > > > > >> be
> >> > > > > >> > > > > deleted
> >> > > > > >> > > > > > > when
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > > > group is deleted. Will a transition to
> >> DEAD
> >> > > > > trigger
> >> > > > > >> > > deletion
> >> > > > > >> > > > > of
> >> > > > > >> > > > > > > > > > > > configurations?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an ending
> >> state
> >> > (we
> >> > > > > would
> >> > > > > >> > only
> >> > > > > >> > > > > > > transit to
> >> > > > > >> > > > > > > > > that
> >> > > > > >> > > > > > > > > > > state when the group is EMPTY and also all
> >> of
> >> > its
> >> > > > > >> > metadata
> >> > > > > >> > > are
> >> > > > > >> > > > > > > gone),
> >> > > > > >> > > > > > > > > once
> >> > > > > >> > > > > > > > > > > it's transited to DEAD this group would
> >> never
> >> > be
> >> > > > > >> revived.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 7. Will the choice to store the configs
> >> in
> >> > the
> >> > > > > group
> >> > > > > >> > > > > coordinator
> >> > > > > >> > > > > > > > > make it
> >> > > > > >> > > > > > > > > > > > harder to list all cluster configs and
> >> their
> >> > > > > values?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > That's a good question, and our thoughts
> >> are
> >> > that
> >> > > > > the
> >> > > > > >> > > so-called
> >> > > > > >> > > > > > > "group
> >> > > > > >> > > > > > > > > > > configurations" are overrides of the
> >> > cluster-level
> >> > > > > >> > > > > configurations
> >> > > > > >> > > > > > > > > > > customized per group so when an admin list
> >> > cluster
> >> > > > > >> > configs
> >> > > > > >> > > it's
> >> > > > > >> > > > > > > okay to
> >> > > > > >> > > > > > > > > > > list just the cluster-level "defaults", not
> >> > showing
> >> > > > > >> any
> >> > > > > >> > > > > per-group
> >> > > > > >> > > > > > > > > > > customizations.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 8. How would someone configure a group
> >> > before
> >> > > > > >> starting
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > consumers? Have
> >> > > > > >> > > > > > > > > > > > we considered allowing the explicit
> >> > creation of
> >> > > > > >> groups?
> >> > > > > >> > > > > > > > > Alternatively, the
> >> > > > > >> > > > > > > > > > > > configs could be decoupled from the group
> >> > > > > lifecycle.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > The configs can be created before the group
> >> > itself
> >> > > > > as
> >> > > > > >> an
> >> > > > > >> > > > > > > independent
> >> > > > > >> > > > > > > > > entity
> >> > > > > >> > > > > > > > > > > --- of course, this requires the
> >> corresponding
> >> > > > > >> request to
> >> > > > > >> > > be
> >> > > > > >> > > > > > > routed to
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > > right coordinator based on the group id ---
> >> > the only
> >> > > > > >> > thing
> >> > > > > >> > > that
> >> > > > > >> > > > > > > > > differs is,
> >> > > > > >> > > > > > > > > > > when the group itself is gone we also check
> >> > if there
> >> > > > > >> are
> >> > > > > >> > > any
> >> > > > > >> > > > > > > > > configuration
> >> > > > > >> > > > > > > > > > > entities related to that group and delete
> >> as
> >> > well.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> >> > asymmetry on
> >> > > > > the
> >> > > > > >> > > creation
> >> > > > > >> > > > > /
> >> > > > > >> > > > > > > > > deletion
> >> > > > > >> > > > > > > > > > > lifecycles of the config entities, and we
> >> > would like
> >> > > > > >> to
> >> > > > > >> > > hear
> >> > > > > >> > > > > > > everyone's
> >> > > > > >> > > > > > > > > > > feelings whether we should aim for symmetry
> >> > i.e.
> >> > > > > >> totally
> >> > > > > >> > > > > decouple
> >> > > > > >> > > > > > > group
> >> > > > > >> > > > > > > > > > > configs and hence not delete them at all
> >> when
> >> > the
> >> > > > > >> group
> >> > > > > >> > is
> >> > > > > >> > > > > gone,
> >> > > > > >> > > > > > > but
> >> > > > > >> > > > > > > > > always
> >> > > > > >> > > > > > > > > > > require explicit deletion operations by
> >> > themselves.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for
> >> > the Java
> >> > > > > >> > client
> >> > > > > >> > > > > still
> >> > > > > >> > > > > > > take
> >> > > > > >> > > > > > > > > a
> >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have
> >> to
> >> > > > > >> introduce an
> >> > > > > >> > > > > overload?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > I think we do not need to introduce an
> >> > overload, but
> >> > > > > >> I'm
> >> > > > > >> > > all
> >> > > > > >> > > > > ears
> >> > > > > >> > > > > > > if
> >> > > > > >> > > > > > > > > there
> >> > > > > >> > > > > > > > > > > may be some compatibility issues that we
> >> may
> >> > > > > overlook.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we should
> >> be
> >> > clearer
> >> > > > > >> > about
> >> > > > > >> > > the
> >> > > > > >> > > > > > > reason
> >> > > > > >> > > > > > > > > to
> >> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version from the
> >> > feature
> >> > > > > >> flag.
> >> > > > > >> > > Maybe
> >> > > > > >> > > > > we
> >> > > > > >> > > > > > > mean
> >> > > > > >> > > > > > > > > that we
> >> > > > > >> > > > > > > > > > > > can switch the default for the feature
> >> flag
> >> > to
> >> > > > > true
> >> > > > > >> > > based on
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > > > > metadata.version once we want to make it
> >> the
> >> > > > > >> default.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention
> >> > the
> >> > > > > >> required
> >> > > > > >> > > ACLs,
> >> > > > > >> > > > > it
> >> > > > > >> > > > > > > > > would be
> >> > > > > >> > > > > > > > > > > > good to add that for consistency.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> >> > ConsumerGroupHeartbeat
> >> > > > > >> > requires
> >> > > > > >> > > > > "Read
> >> > > > > >> > > > > > > > > Group" even
> >> > > > > >> > > > > > > > > > > > though it seems to do more than reading.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > I had that thought myself as well, but in
> >> the
> >> > end we
> >> > > > > >> > could
> >> > > > > >> > > not
> >> > > > > >> > > > > > > find a
> >> > > > > >> > > > > > > > > > > better alternative: adding Write Group
> >> seems
> >> > an
> >> > > > > >> overkill
> >> > > > > >> > > here
> >> > > > > >> > > > > > > since we
> >> > > > > >> > > > > > > > > do
> >> > > > > >> > > > > > > > > > > not have it elsewhere (we only have Read /
> >> > Delete
> >> > > > > and
> >> > > > > >> > > Describe
> >> > > > > >> > > > > on
> >> > > > > >> > > > > > > > > groups so
> >> > > > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 13. How is topic recreation handled by
> >> the
> >> > > > > consumer
> >> > > > > >> > with
> >> > > > > >> > > the
> >> > > > > >> > > > > new
> >> > > > > >> > > > > > > > > group
> >> > > > > >> > > > > > > > > > > > protocol? It would be good to have a
> >> > section on
> >> > > > > >> this.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > You mean with regex subscription right? Yes
> >> > we can
> >> > > > > >> add a
> >> > > > > >> > > > > section
> >> > > > > >> > > > > > > about
> >> > > > > >> > > > > > > > > > > that, but basically the idea is that
> >> consumer
> >> > would
> >> > > > > be
> >> > > > > >> > > totally
> >> > > > > >> > > > > > > > > agnostic in
> >> > > > > >> > > > > > > > > > > the new protocol as it's handled all by the
> >> > brokers.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write the
> >> new
> >> > > > > >> coordinator
> >> > > > > >> > in
> >> > > > > >> > > > > Java.
> >> > > > > >> > > > > > > Even
> >> > > > > >> > > > > > > > > though
> >> > > > > >> > > > > > > > > > > > this is an implementation detail, do we
> >> > plan to
> >> > > > > >> have a
> >> > > > > >> > > new
> >> > > > > >> > > > > gradle
> >> > > > > >> > > > > > > > > module
> >> > > > > >> > > > > > > > > > > > for it?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > We have not thought about that. But I think
> >> > the
> >> > > > > answer
> >> > > > > >> > > should
> >> > > > > >> > > > > be
> >> > > > > >> > > > > > > yes.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it
> >> > comes to
> >> > > > > >> how
> >> > > > > >> > > many
> >> > > > > >> > > > > > > members
> >> > > > > >> > > > > > > > > the new
> >> > > > > >> > > > > > > > > > > > group protocol can support?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Within a group, I think we should shoot for
> >> > 1000s of
> >> > > > > >> > > members.
> >> > > > > >> > > > > But
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > > > scalability goals also depend on the offset
> >> > > > > management
> >> > > > > >> > > (commit,
> >> > > > > >> > > > > > > fetch)
> >> > > > > >> > > > > > > > > > > capabilities of the coordinator which we
> >> did
> >> > not
> >> > > > > >> cover in
> >> > > > > >> > > this
> >> > > > > >> > > > > > > KIP, so
> >> > > > > >> > > > > > > > > it's
> >> > > > > >> > > > > > > > > > > hard to give a number that applies
> >> > universally.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> >> > SubscribedTopidIds
> >> > > > > >> instead
> >> > > > > >> > of
> >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> >> > > > > >> ConsumerGroupHeartbeatRequest?
> >> > > > > >> > > Is the
> >> > > > > >> > > > > > > idea
> >> > > > > >> > > > > > > > > that
> >> > > > > >> > > > > > > > > > > > since we have to resolve the regex on the
> >> > server,
> >> > > > > we
> >> > > > > >> > can
> >> > > > > >> > > do
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > same
> >> > > > > >> > > > > > > > > for
> >> > > > > >> > > > > > > > > > > > the topic name? The difference is that
> >> > sending the
> >> > > > > >> > regex
> >> > > > > >> > > is
> >> > > > > >> > > > > more
> >> > > > > >> > > > > > > > > efficient
> >> > > > > >> > > > > > > > > > > > whereas sending the topic names is less
> >> > efficient.
> >> > > > > >> > > > > Furthermore,
> >> > > > > >> > > > > > > > > delete and
> >> > > > > >> > > > > > > > > > > > recreation is easier to handle if we have
> >> > topic
> >> > > > > ids.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > The main reason to still let the clients
> >> send
> >> > names
> >> > > > > >> is to
> >> > > > > >> > > keep
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > > > reasoning of names -> ids on the broker /
> >> > admin
> >> > > > > client
> >> > > > > >> > > only.
> >> > > > > >> > > > > Note
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > > > although we added topic id in KIP-516, we
> >> > never
> >> > > > > >> > > implemented the
> >> > > > > >> > > > > > > logic
> >> > > > > >> > > > > > > > > on
> >> > > > > >> > > > > > > > > > > consumer/producers leveraging the related
> >> > newer
> >> > > > > >> versioned
> >> > > > > >> > > RPCs,
> >> > > > > >> > > > > > > > > instead we
> >> > > > > >> > > > > > > > > > > just set the topic id as empty UUID. We
> >> want
> >> > to keep
> >> > > > > >> the
> >> > > > > >> > > > > > > > > consumer/producer
> >> > > > > >> > > > > > > > > > > to be thin and only delegate the reasoning
> >> on
> >> > broker
> >> > > > > >> and
> >> > > > > >> > > > > > > potentially
> >> > > > > >> > > > > > > > > admin
> >> > > > > >> > > > > > > > > > > clients.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > Thanks,
> >> > > > > >> > > > > > > > > > > > Ismael
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David
> >> Jacot
> >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> >> > > > > >> > > > > > > > > > > > wrote:
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > Hi all,
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > I would like to start a discussion
> >> thread
> >> > on
> >> > > > > >> KIP-848:
> >> > > > > >> > > The
> >> > > > > >> > > > > Next
> >> > > > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance
> >> > Protocol.
> >> > > > > >> With
> >> > > > > >> > > this
> >> > > > > >> > > > > KIP,
> >> > > > > >> > > > > > > we
> >> > > > > >> > > > > > > > > aim
> >> > > > > >> > > > > > > > > > > > > to make the rebalance protocol (for
> >> > consumers)
> >> > > > > >> more
> >> > > > > >> > > > > reliable,
> >> > > > > >> > > > > > > more
> >> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
> >> > clients, and
> >> > > > > >> easier
> >> > > > > >> > > to
> >> > > > > >> > > > > debug
> >> > > > > >> > > > > > > for
> >> > > > > >> > > > > > > > > > > > > operators.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > The KIP is here:
> >> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > Please take a look and let me know what
> >> > you
> >> > > > > think.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > Best,
> >> > > > > >> > > > > > > > > > > > > David
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > PS: I will be away from July 18th to
> >> > August 8th.
> >> > > > > >> That
> >> > > > > >> > > gives
> >> > > > > >> > > > > > > you a
> >> > > > > >> > > > > > > > > bit
> >> > > > > >> > > > > > > > > > > > > of time to read and digest this long
> >> KIP.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > --
> >> > > > > >> > > > > > > > > > > -- Guozhang
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > --
> >> > > > > >> > > > > > > > -- Guozhang
> >> > > > > >> > > > > > >
> >> > > > > >> > > > >
> >> > > > > >> > >
> >> > > > > >> > >
> >> > > > > >> >
> >> > > > > >>
> >> > > > > >>
> >> > > > > >> --
> >> > > > > >> -- Guozhang
> >> > > > > >>
> >> > > > > >
> >> > > > >
> >> >
> >>
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Sagar <sa...@gmail.com>]

Hi All,

As per the suggestions from David/Guozhang above, I updated the page for
Connect to have it's own set of APIs and not extend the ones from consumer.
Plz review again.

@Luke,

Thank you. I am actually looking for review comments/thoughts/suggestions
as Connect changes are still in draft stage but since this is a wholesome
change, I am sure there might be many cases missed by me. Also, on this
discussion thread David suggested to continue discussing on this very
thread to ensure connect is compatible with whatever is done in KIP-848. I
am ok either way.

@Guozhang,

Regarding the migration plan and the link to KIP-415 that you had shared, I
had taken a look at it. From what I understood, that migration/downgrading
is from switching the assignment protocol from eager -> incremental or vice
versa. In this case, the migration is to the new rebalance protocol which
would also depend on the group coordinator and whether they support the new
rebalance protocol or not. I have tried to incorporate this aspect in the
draft page.

Thanks!
Sagar.




On Mon, Aug 22, 2022 at 1:00 PM Luke Chen <sh...@gmail.com> wrote:

> Hi David,
>
> Thanks for the update.
>
> Some more questions:
> 1. In Group Coordinator section, you mentioned:
> > The new group coordinator will have a state machine per
> *__consumer_offsets* partitions, where each state machine is modelled as an
> event loop. Those state machines will be executed in
> *group.coordinator.threads* threads.
>
> 1.1. I think the state machine are: "Empty, assigning, reconciling, stable,
> dead" mentioned in Consumer Group States section, right?
> 1.2. What do you mean "each state machine is modelled as an event loop"?
> 1.3. Why do we need a state machine per *__consumer_offsets* partitions?
> Not a state machine "per consumer group" owned by a group coordinator? For
> example, if one group coordinator owns 2 consumer groups, and both exist in
> *__consumer_offsets-0*, will we have 1 state machine for it, or 2?
> 1.4. I know the "*group.coordinator.threads" *is the number of threads used
> to run the state machines. But I'm wondering if the purpose of the threads
> is only to keep the state of each consumer group (or *__consumer_offsets*
> partitions?), and no heavy computation, why should we need multi-threads
> here?
>
> 2. For the default value in the new configs:
> 2.1. The consumer session timeout, why does the default session timeout not
> locate between min (45s) and max(60s)? I thought the min/max session
> timeout is to define lower/upper bound of it, no?
>
> group.consumer.session.timeout.ms int 30s The timeout to detect client
> failures when using the consumer group protocol.
> group.consumer.min.session.timeout.ms int 45s The minimum session timeout.
> group.consumer.max.session.timeout.ms int 60s The maximum session timeout.
>
>
>
> 2.2. The default server side assignor are [range, uniform], which means
> we'll default to "range" assignor. I'd like to know why not uniform one? I
> thought usually users will choose uniform assignor (former sticky assinor)
> for better evenly distribution. Any other reason we choose range assignor
> as default?
> group.consumer.assignors List range, uniform The server side assignors.
>
>
>
>
>
>
> Thank you.
> Luke
>
>
>
>
>
>
> On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com> wrote:
>
> > Hi Sagar,
> >
> > I have some thoughts about Kafka Connect integrating with KIP-848, but I
> > think we should have a separate discussion thread for the Kafka Connect
> > KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol [1],
> > and let this discussion thread focus on consumer rebalance protocol,
> WDYT?
> >
> > [1]
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >
> > Thank you.
> > Luke
> >
> >
> >
> > On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com> wrote:
> >
> >> Thank you Guozhang/David for the feedback. Looks like there's agreement
> on
> >> using separate APIs for Connect. I would revisit the doc and see what
> >> changes are to be made.
> >>
> >> Thanks!
> >> Sagar.
> >>
> >> On Tue, Aug 9, 2022 at 7:11 PM David Jacot <djacot@confluent.io.invalid
> >
> >> wrote:
> >>
> >> > Hi Sagar,
> >> >
> >> > Thanks for the feedback and the document. That's really helpful. I
> >> > will take a look at it.
> >> >
> >> > Overall, it seems to me that both Connect and the Consumer could share
> >> > the same underlying "engine". The main difference is that the Consumer
> >> > assigns topic-partitions to members whereas Connect assigns tasks to
> >> > workers. I see two ways to move forward:
> >> > 1) We extend the new proposed APIs to support different resource types
> >> > (e.g. partitions, tasks, etc.); or
> >> > 2) We use new dedicated APIs for Connect. The dedicated APIs would be
> >> > similar to the new ones but different on the content/resources and
> >> > they would rely on the same engine on the coordinator side.
> >> >
> >> > I personally lean towards 2) because I am not a fan of overcharging
> >> > APIs to serve different purposes. That being said, I am not opposed to
> >> > 1) if we can find an elegant way to do it.
> >> >
> >> > I think that we can continue to discuss it here for now in order to
> >> > ensure that this KIP is compatible with what we will do for Connect in
> >> > the future.
> >> >
> >> > Best,
> >> > David
> >> >
> >> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io>
> wrote:
> >> > >
> >> > > Hi all,
> >> > >
> >> > > I am back from vacation. I will go through and address your comments
> >> > > in the coming days. Thanks for your feedback.
> >> > >
> >> > > Cheers,
> >> > > David
> >> > >
> >> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <
> gharris1727@gmail.com
> >> >
> >> > wrote:
> >> > > >
> >> > > > Hey All!
> >> > > >
> >> > > > Thanks for the KIP, it's wonderful to see cooperative rebalancing
> >> > making it
> >> > > > down the stack!
> >> > > >
> >> > > > I had a few questions:
> >> > > >
> >> > > > 1. The 'Rejected Alternatives' section describes how member epoch
> >> > should
> >> > > > advance in step with the group epoch and assignment epoch values.
> I
> >> > think
> >> > > > that this is a good idea for the reasons described in the KIP.
> When
> >> the
> >> > > > protocol is incrementally assigning partitions to a worker, what
> >> member
> >> > > > epoch does each incremental assignment use? Are member epochs
> >> re-used,
> >> > and
> >> > > > a single member epoch can correspond to multiple different
> >> > (monotonically
> >> > > > larger) assignments?
> >> > > >
> >> > > > 2. Is the Assignor's 'Reason' field opaque to the group
> >> coordinator? If
> >> > > > not, should custom client-side assignor implementations interact
> >> with
> >> > the
> >> > > > Reason field, and how is its common meaning agreed upon? If so,
> what
> >> > is the
> >> > > > benefit of a distinct Reason field over including such
> functionality
> >> > in the
> >> > > > opaque metadata?
> >> > > >
> >> > > > 3. The following is included in the KIP: "Thanks to this, the
> input
> >> of
> >> > the
> >> > > > client side assignor is entirely driven by the group coordinator.
> >> The
> >> > > > consumer is no longer responsible for maintaining any state
> besides
> >> its
> >> > > > assigned partitions." Does this mean that the client-side assignor
> >> MAY
> >> > > > incorporate additional non-Metadata state (such as partition
> >> > throughput,
> >> > > > cpu/memory metrics, config topics, etc), or that additional
> >> > non-Metadata
> >> > > > state SHOULD NOT be used?
> >> > > >
> >> > > > 4. I see that there are separate classes
> >> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> >> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that seem
> to
> >> > > > overlap significantly. Is it possible for these two
> implementations
> >> to
> >> > be
> >> > > > unified? This would serve to promote feature parity of server-side
> >> and
> >> > > > client-side assignors, and would also facilitate operational
> >> > flexibility in
> >> > > > certain situations. For example, if a server-side assignor has
> some
> >> > poor
> >> > > > behavior and needs a patch, deploying the patched assignor to the
> >> > client
> >> > > > and switching one consumer group to a client-side assignor may be
> >> > faster
> >> > > > and less risky than patching all of the brokers. With the
> currently
> >> > > > proposed distinct APIs, a non-trivial reimplementation would have
> >> to be
> >> > > > assembled, and if the two APIs have diverged significantly, then
> it
> >> is
> >> > > > possible that a reimplementation would not be possible.
> >> > > >
> >> > > > --
> >> > > > Greg Harris
> >> > > > gharris1727@gmail.com
> >> > > > github.com/gharris1727
> >> > > >
> >> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com>
> >> > wrote:
> >> > > >
> >> > > > > Hi Guozhang/David,
> >> > > > >
> >> > > > > I created a confluence page to discuss how Connect would need to
> >> > change
> >> > > > > based on the new rebalance protocol. Here's the page:
> >> > > > >
> >> > > > >
> >> >
> >>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >> > > > >
> >> > > > > It's also pretty longish and I have tried to keep a format
> >> similar to
> >> > > > > KIP-848. Let me know what you think. Also, do you think this
> >> should
> >> > be
> >> > > > > moved to a separate discussion thread or is this one fine?
> >> > > > >
> >> > > > > Thanks!
> >> > > > > Sagar.
> >> > > > >
> >> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <
> sagarmeansocean@gmail.com>
> >> > wrote:
> >> > > > >
> >> > > > > > Hello Guozhang,
> >> > > > > >
> >> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I would
> do
> >> > the
> >> > > > > > analysis and come up with such a document.
> >> > > > > >
> >> > > > > > Thanks!
> >> > > > > > Sagar.
> >> > > > > >
> >> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> >> wangguoz@gmail.com>
> >> > > > > wrote:
> >> > > > > >
> >> > > > > >> Hello Sagar,
> >> > > > > >>
> >> > > > > >> It would be great if you could come back with some analysis
> on
> >> > how to
> >> > > > > >> implement the Connect side integration with the new protocol;
> >> so
> >> > far
> >> > > > > >> besides leveraging on the new "protocol type" we did not yet
> >> think
> >> > > > > through
> >> > > > > >> the Connect side implementations. For Streams here's a draft
> of
> >> > > > > >> integration
> >> > > > > >> plan:
> >> > > > > >>
> >> > > > > >>
> >> > > > >
> >> >
> >>
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >> > > > > >> just FYI for your analysis on Connect.
> >> > > > > >>
> >> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> >> sagarmeansocean@gmail.com
> >> > >
> >> > > > > wrote:
> >> > > > > >>
> >> > > > > >> > Hi David,
> >> > > > > >> >
> >> > > > > >> > Thank you for your response. The reason I thought connect
> can
> >> > also fit
> >> > > > > >> into
> >> > > > > >> > this new scheme is that even today the connect uses a
> >> > > > > WorkerCoordinator
> >> > > > > >> > extending from AbstractCoordinator to empower rebalances of
> >> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> >> protocolType()
> >> > to
> >> > > > > >> connect
> >> > > > > >> > and uses the metadata() method by plumbing into
> >> > > > > >> JoinGroupRequestProtocol.
> >> > > > > >> >
> >> > > > > >> > I think the changes to support connect would be similar at
> a
> >> > high
> >> > > > > level
> >> > > > > >> to
> >> > > > > >> > the changes in streams mainly because of the Client side
> >> > assignors
> >> > > > > being
> >> > > > > >> > used in both. At an implementation level, we might need to
> >> make
> >> > a lot
> >> > > > > of
> >> > > > > >> > changes to get onto this new assignment protocol like
> >> enhancing
> >> > the
> >> > > > > >> > JoinGroup request/response and SyncGroup and using
> >> > > > > >> ConsumerGroupHeartbeat
> >> > > > > >> > API etc again on similar lines to streams (or there might
> be
> >> > > > > >> deviations). I
> >> > > > > >> > would try to perform a detailed analysis of the same  and
> we
> >> > can have
> >> > > > > a
> >> > > > > >> > separate discussion thread for that as that would derail
> this
> >> > > > > discussion
> >> > > > > >> > thread. Let me know if that sounds good to you.
> >> > > > > >> >
> >> > > > > >> > Thanks!
> >> > > > > >> > Sagar.
> >> > > > > >> >
> >> > > > > >> >
> >> > > > > >> >
> >> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> >> > > > > <djacot@confluent.io.invalid
> >> > > > > >> >
> >> > > > > >> > wrote:
> >> > > > > >> >
> >> > > > > >> > > Hi Sagar,
> >> > > > > >> > >
> >> > > > > >> > > Thanks for your comments.
> >> > > > > >> > >
> >> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can
> >> > mention it.
> >> > > > > >> > >
> >> > > > > >> > > 2) The idea is to transition C from his current
> assignment
> >> to
> >> > his
> >> > > > > >> > > target assignment when he can move to epoch 3. When that
> >> > happens,
> >> > > > > the
> >> > > > > >> > > member assignment is updated and persisted with all its
> >> > assigned
> >> > > > > >> > > partitions even if they are not all revoked yet. In other
> >> > words, the
> >> > > > > >> > > member assignment becomes the target assignment. This is
> >> > basically
> >> > > > > an
> >> > > > > >> > > optimization to avoid having to write all the changes to
> >> the
> >> > log.
> >> > > > > The
> >> > > > > >> > > examples are based on the persisted state so I understand
> >> the
> >> > > > > >> > > confusion. Let me see if I can improve this in the
> >> > description.
> >> > > > > >> > >
> >> > > > > >> > > 3) Regarding Connect, it could reuse the protocol with a
> >> > client side
> >> > > > > >> > > assignor if it fits in the protocol. The assignment is
> >> about
> >> > > > > >> > > topicid-partitions + metadata, could Connect fit into
> this?
> >> > > > > >> > >
> >> > > > > >> > > Best,
> >> > > > > >> > > David
> >> > > > > >> > >
> >> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> >> > sagarmeansocean@gmail.com>
> >> > > > > >> wrote:
> >> > > > > >> > > >
> >> > > > > >> > > > Hi David,
> >> > > > > >> > > >
> >> > > > > >> > > > Thanks for the KIP. I just had minor observations:
> >> > > > > >> > > >
> >> > > > > >> > > > 1) In the Assignment Error section in Client Side mode
> >> > Assignment
> >> > > > > >> > > process,
> >> > > > > >> > > > you mentioned => `In this case, the client side
> assignor
> >> can
> >> > > > > return
> >> > > > > >> an
> >> > > > > >> > > > error to the group coordinator`. In this case are you
> >> > referring to
> >> > > > > >> the
> >> > > > > >> > > > Assignor returning an AssignmentError that's listed
> down
> >> > towards
> >> > > > > the
> >> > > > > >> > end?
> >> > > > > >> > > > If yes, do you think it would make sense to mention
> this
> >> > > > > explicitly
> >> > > > > >> > here?
> >> > > > > >> > > >
> >> > > > > >> > > > 2) In the Case Studies section, I have a slight
> >> confusion,
> >> > not
> >> > > > > sure
> >> > > > > >> if
> >> > > > > >> > > > others have the same. Consider this step:
> >> > > > > >> > > >
> >> > > > > >> > > > When B heartbeats, the group coordinator transitions
> him
> >> to
> >> > epoch
> >> > > > > 3
> >> > > > > >> > > because
> >> > > > > >> > > > B has no partitions to revoke. It persists the change
> and
> >> > reply.
> >> > > > > >> > > >
> >> > > > > >> > > >    - Group (epoch=3)
> >> > > > > >> > > >       - A
> >> > > > > >> > > >       - B
> >> > > > > >> > > >       - C
> >> > > > > >> > > >    - Target Assignment (epoch=3)
> >> > > > > >> > > >       - A - partitions=[foo-0]
> >> > > > > >> > > >       - B - partitions=[foo-2]
> >> > > > > >> > > >       - C - partitions=[foo-1]
> >> > > > > >> > > >    - Member Assignment
> >> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> >> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> >> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> >> > > > > >> > > >
> >> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot
> >> get
> >> > foo-1
> >> > > > > >> yet.
> >> > > > > >> > > >
> >> > > > > >> > > > Here,it's mentioned that member C can't get the foo-1
> >> > partition
> >> > > > > yet,
> >> > > > > >> > but
> >> > > > > >> > > > based on the description above, it seems it already has
> >> it.
> >> > Do you
> >> > > > > >> > think
> >> > > > > >> > > it
> >> > > > > >> > > > would be better to remove it and populate it only when
> it
> >> > actually
> >> > > > > >> gets
> >> > > > > >> > > it?
> >> > > > > >> > > > I see this in a lot of other places, so have I
> >> understood it
> >> > > > > >> > incorrectly
> >> > > > > >> > > ?
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > > Regarding connect , it might be out of scope of this
> >> > discussion,
> >> > > > > but
> >> > > > > >> > from
> >> > > > > >> > > > what I understood it would probably be running in
> client
> >> > side
> >> > > > > >> assignor
> >> > > > > >> > > mode
> >> > > > > >> > > > even on the new rebalance protocol as it has its own
> >> Custom
> >> > > > > >> > > Assignors(Eager
> >> > > > > >> > > > and IncrementalCooperative).
> >> > > > > >> > > >
> >> > > > > >> > > > Thanks!
> >> > > > > >> > > >
> >> > > > > >> > > > Sagar.
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > >
> >> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> >> > > > > >> > <djacot@confluent.io.invalid
> >> > > > > >> > > >
> >> > > > > >> > > > wrote:
> >> > > > > >> > > >
> >> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
> >> > specialized API
> >> > > > > >> > > > > instead of relying on one generic API. For Connect,
> we
> >> > can apply
> >> > > > > >> the
> >> > > > > >> > > > > exact same pattern and reuse/share the core
> >> > implementation on
> >> > > > > the
> >> > > > > >> > > > > server side. For the schema registry, I think that we
> >> > should
> >> > > > > >> consider
> >> > > > > >> > > > > having a tailored API to do simple membership/leader
> >> > election.
> >> > > > > >> > > > >
> >> > > > > >> > > > > Best,
> >> > > > > >> > > > > David
> >> > > > > >> > > > >
> >> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> >> > ismael@juma.me.uk
> >> > > > > >
> >> > > > > >> > > wrote:
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > Three quick comments:
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> >> > > > > >> com.google.re2j.Pattern, we
> >> > > > > >> > > > > should
> >> > > > > >> > > > > > document the differences in more detail before
> >> deciding
> >> > one
> >> > > > > way
> >> > > > > >> or
> >> > > > > >> > > > > another.
> >> > > > > >> > > > > > That said, if people pass java.util.regex.Pattern,
> >> they
> >> > expect
> >> > > > > >> > their
> >> > > > > >> > > > > > semantics to be honored. If we are doing something
> >> > different,
> >> > > > > >> then
> >> > > > > >> > we
> >> > > > > >> > > > > > should consider adding an overload with our own
> >> Pattern
> >> > class
> >> > > > > (I
> >> > > > > >> > > don't
> >> > > > > >> > > > > > think we'd want to expose re2j's at this point).
> >> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol
> should
> >> > > > > integrate
> >> > > > > >> > fully
> >> > > > > >> > > > > with
> >> > > > > >> > > > > > it and should handle the topic recreation case
> >> > correctly.
> >> > > > > That's
> >> > > > > >> > the
> >> > > > > >> > > main
> >> > > > > >> > > > > > part we need to handle. I agree with David that
> we'd
> >> > want to
> >> > > > > add
> >> > > > > >> > > topic
> >> > > > > >> > > > > ids
> >> > > > > >> > > > > > to the relevant protocols that don't have it yet
> and
> >> > that we
> >> > > > > can
> >> > > > > >> > > probably
> >> > > > > >> > > > > > focus on the internals versus adding new APIs to
> the
> >> > Java
> >> > > > > >> Consumer
> >> > > > > >> > > > > (unless
> >> > > > > >> > > > > > we find that adding new APIs is required for
> >> reasonable
> >> > > > > >> semantics).
> >> > > > > >> > > > > > 3. I am still not sure about the coordinator
> storing
> >> the
> >> > > > > >> configs.
> >> > > > > >> > > It's
> >> > > > > >> > > > > > powerful for configs to be centralized in the
> >> metadata
> >> > log for
> >> > > > > >> > > various
> >> > > > > >> > > > > > reasons (auditability, visibility, consistency,
> >> etc.).
> >> > > > > >> Similarly, I
> >> > > > > >> > > am
> >> > > > > >> > > > > not
> >> > > > > >> > > > > > sure about automatically deleting configs in a way
> >> that
> >> > they
> >> > > > > >> cannot
> >> > > > > >> > > be
> >> > > > > >> > > > > > recovered. A good property for modern systems is to
> >> > minimize
> >> > > > > the
> >> > > > > >> > > number
> >> > > > > >> > > > > of
> >> > > > > >> > > > > > unrecoverable data loss scenarios.
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > Ismael
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> >> > > > > >> > > <djacot@confluent.io.invalid
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > wrote:
> >> > > > > >> > > > > >
> >> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > > 1) the migration path, especially the last step
> >> when
> >> > > > > clients
> >> > > > > >> > > flip the
> >> > > > > >> > > > > > > flag
> >> > > > > >> > > > > > > > to enable the new protocol, in which we would
> >> have a
> >> > > > > window
> >> > > > > >> > where
> >> > > > > >> > > > > both
> >> > > > > >> > > > > > > new
> >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
> >> used
> >> > by
> >> > > > > >> members
> >> > > > > >> > of
> >> > > > > >> > > the
> >> > > > > >> > > > > same
> >> > > > > >> > > > > > > > group. How the coordinator could "mimic" the
> old
> >> > behavior
> >> > > > > >> while
> >> > > > > >> > > > > using the
> >> > > > > >> > > > > > > > new protocol is something we need to present
> >> about.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > Noted. I just published a new version of KIP
> which
> >> > includes
> >> > > > > >> more
> >> > > > > >> > > > > > > details about this. See the "Supporting Online
> >> > Consumer
> >> > > > > Group
> >> > > > > >> > > Upgrade"
> >> > > > > >> > > > > > > and the "Compatibility, Deprecation, and
> Migration
> >> > Plan". I
> >> > > > > >> think
> >> > > > > >> > > that
> >> > > > > >> > > > > > > I have to think through a few cases now but the
> >> > overall idea
> >> > > > > >> and
> >> > > > > >> > > > > > > mechanism should be understandable.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516
> the
> >> > topic ids
> >> > > > > >> are
> >> > > > > >> > > only
> >> > > > > >> > > > > used
> >> > > > > >> > > > > > > as
> >> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> >> > exposed
> >> > > > > via
> >> > > > > >> any
> >> > > > > >> > > > > public
> >> > > > > >> > > > > > > APIs
> >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> first
> >> > should we
> >> > > > > >> let
> >> > > > > >> > > the
> >> > > > > >> > > > > > > consumer
> >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> mapping
> >> > itself
> >> > > > > to
> >> > > > > >> > fully
> >> > > > > >> > > > > > > leverage
> >> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> >> RPCs
> >> > with
> >> > > > > the
> >> > > > > >> > > topic
> >> > > > > >> > > > > ids;
> >> > > > > >> > > > > > > and
> >> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> >> > topic ids
> >> > > > > in
> >> > > > > >> the
> >> > > > > >> > > > > consumer
> >> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> >> well
> >> > as in
> >> > > > > >> the
> >> > > > > >> > > > > rebalance
> >> > > > > >> > > > > > > > listener for cases like topic
> >> > deletion-and-recreation).
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > a) Assuming that we would include converting all
> >> the
> >> > offsets
> >> > > > > >> > > related
> >> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer
> >> > would be
> >> > > > > >> able
> >> > > > > >> > to
> >> > > > > >> > > > > > > fully operate with topic ids. That being said, it
> >> > still has
> >> > > > > to
> >> > > > > >> > > provide
> >> > > > > >> > > > > > > the topics names in various APIs so having a
> >> mapping
> >> > in the
> >> > > > > >> > > consumer
> >> > > > > >> > > > > > > seems inevitable to me.
> >> > > > > >> > > > > > > b) I don't have a strong opinion on this. Here I
> >> > wonder if
> >> > > > > >> this
> >> > > > > >> > > goes
> >> > > > > >> > > > > > > beyond the scope of this KIP. I would rather
> focus
> >> on
> >> > the
> >> > > > > >> > internals
> >> > > > > >> > > > > > > here and we can consider this separately if we
> see
> >> > value in
> >> > > > > >> doing
> >> > > > > >> > > it.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > Coming back to Ismael's point about using topic
> ids
> >> > in the
> >> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there
> >> is
> >> > one
> >> > > > > >> > advantage
> >> > > > > >> > > in
> >> > > > > >> > > > > > > favour of it. The consumer will have the
> >> opportunity
> >> > to
> >> > > > > >> validate
> >> > > > > >> > > that
> >> > > > > >> > > > > > > the topics exists before passing them into the
> >> group
> >> > > > > rebalance
> >> > > > > >> > > > > > > protocol. Obviously, the coordinator will also
> >> notice
> >> > it but
> >> > > > > >> it
> >> > > > > >> > > does
> >> > > > > >> > > > > > > not really have a way to reject an invalid topic
> in
> >> > the
> >> > > > > >> response.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> >> questions
> >> > > > > except
> >> > > > > >> for
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I
> think
> >> > it's not
> >> > > > > >> > > necessary
> >> > > > > >> > > > > to
> >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> >> > instead we
> >> > > > > >> still
> >> > > > > >> > > use
> >> > > > > >> > > > > > > Pattern
> >> > > > > >> > > > > > > > while just documenting that our subscription
> may
> >> be
> >> > > > > >> rejected by
> >> > > > > >> > > the
> >> > > > > >> > > > > > > server.
> >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> >> scenario
> >> > I felt
> >> > > > > >> > using
> >> > > > > >> > > an
> >> > > > > >> > > > > > > > overloaded `String` based subscription may be
> >> more
> >> > > > > >> vulnerable
> >> > > > > >> > to
> >> > > > > >> > > > > various
> >> > > > > >> > > > > > > > invalid regexes.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > That could work. I have to look at the
> differences
> >> > between
> >> > > > > the
> >> > > > > >> > two
> >> > > > > >> > > > > > > engines to better understand the potential
> issues.
> >> My
> >> > > > > >> > > understanding is
> >> > > > > >> > > > > > > that would work for all the basic regular
> >> > expressions. The
> >> > > > > >> > > differences
> >> > > > > >> > > > > > > between the two are mainly about the various
> >> character
> >> > > > > >> classes. I
> >> > > > > >> > > > > > > wonder what other people think about this.
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > Best,
> >> > > > > >> > > > > > > David
> >> > > > > >> > > > > > >
> >> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> >> > > > > >> > wangguoz@gmail.com
> >> > > > > >> > > >
> >> > > > > >> > > > > wrote:
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > Thanks David! I think on the high level there
> are
> >> > two meta
> >> > > > > >> > > points we
> >> > > > > >> > > > > need
> >> > > > > >> > > > > > > > to concretize a bit more:
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > 1) the migration path, especially the last step
> >> when
> >> > > > > clients
> >> > > > > >> > > flip the
> >> > > > > >> > > > > > > flag
> >> > > > > >> > > > > > > > to enable the new protocol, in which we would
> >> have a
> >> > > > > window
> >> > > > > >> > where
> >> > > > > >> > > > > both
> >> > > > > >> > > > > > > new
> >> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
> >> used
> >> > by
> >> > > > > >> members
> >> > > > > >> > of
> >> > > > > >> > > the
> >> > > > > >> > > > > same
> >> > > > > >> > > > > > > > group. How the coordinator could "mimic" the
> old
> >> > behavior
> >> > > > > >> while
> >> > > > > >> > > > > using the
> >> > > > > >> > > > > > > > new protocol is something we need to present
> >> about.
> >> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516
> the
> >> > topic ids
> >> > > > > >> are
> >> > > > > >> > > only
> >> > > > > >> > > > > used
> >> > > > > >> > > > > > > as
> >> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> >> > exposed
> >> > > > > via
> >> > > > > >> any
> >> > > > > >> > > > > public
> >> > > > > >> > > > > > > APIs
> >> > > > > >> > > > > > > > to consumers yet. I think the question is,
> first
> >> > should we
> >> > > > > >> let
> >> > > > > >> > > the
> >> > > > > >> > > > > > > consumer
> >> > > > > >> > > > > > > > client to be maintaining the names -> ids
> mapping
> >> > itself
> >> > > > > to
> >> > > > > >> > fully
> >> > > > > >> > > > > > > leverage
> >> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> >> RPCs
> >> > with
> >> > > > > the
> >> > > > > >> > > topic
> >> > > > > >> > > > > ids;
> >> > > > > >> > > > > > > and
> >> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> >> > topic ids
> >> > > > > in
> >> > > > > >> the
> >> > > > > >> > > > > consumer
> >> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> >> well
> >> > as in
> >> > > > > >> the
> >> > > > > >> > > > > rebalance
> >> > > > > >> > > > > > > > listener for cases like topic
> >> > deletion-and-recreation).
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> >> questions
> >> > > > > except
> >> > > > > >> for
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I
> think
> >> > it's not
> >> > > > > >> > > necessary
> >> > > > > >> > > > > to
> >> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> >> > instead we
> >> > > > > >> still
> >> > > > > >> > > use
> >> > > > > >> > > > > > > Pattern
> >> > > > > >> > > > > > > > while just documenting that our subscription
> may
> >> be
> >> > > > > >> rejected by
> >> > > > > >> > > the
> >> > > > > >> > > > > > > server.
> >> > > > > >> > > > > > > > Since the incompatible case is a very rare
> >> scenario
> >> > I felt
> >> > > > > >> > using
> >> > > > > >> > > an
> >> > > > > >> > > > > > > > overloaded `String` based subscription may be
> >> more
> >> > > > > >> vulnerable
> >> > > > > >> > to
> >> > > > > >> > > > > various
> >> > > > > >> > > > > > > > invalid regexes.
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > Guozhang
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> >> > > > > >> > > > > <djacot@confluent.io.invalid
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > wrote:
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > > Hi Ismael,
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer your
> >> > questions
> >> > > > > >> > inline.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 1. I think it's premature to talk about
> >> target
> >> > > > > versions
> >> > > > > >> for
> >> > > > > >> > > > > > > deprecation
> >> > > > > >> > > > > > > > > and
> >> > > > > >> > > > > > > > > > removal of the existing group protocol.
> >> Unlike
> >> > KRaft,
> >> > > > > >> this
> >> > > > > >> > > > > affects a
> >> > > > > >> > > > > > > core
> >> > > > > >> > > > > > > > > > client protocol and hence
> deprecation/removal
> >> > will be
> >> > > > > >> > heavily
> >> > > > > >> > > > > > > dependent
> >> > > > > >> > > > > > > > > on
> >> > > > > >> > > > > > > > > > how quickly applications migrate to the new
> >> > protocol.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > That makes sense. I will remove it.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 2. The KIP says we intend to release this
> in
> >> > 4.x, but
> >> > > > > it
> >> > > > > >> > > wasn't
> >> > > > > >> > > > > made
> >> > > > > >> > > > > > > > > clear
> >> > > > > >> > > > > > > > > > why. If we added that as a way to estimate
> >> when
> >> > we'd
> >> > > > > >> > > deprecate
> >> > > > > >> > > > > and
> >> > > > > >> > > > > > > remove
> >> > > > > >> > > > > > > > > > the group protocol, I also suggest removing
> >> > this part.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Let me explain my reasoning. As explained, I
> >> plan
> >> > to
> >> > > > > >> rewrite
> >> > > > > >> > > the
> >> > > > > >> > > > > group
> >> > > > > >> > > > > > > > > coordinator in Java while we implement the
> new
> >> > protocol.
> >> > > > > >> This
> >> > > > > >> > > means
> >> > > > > >> > > > > > > > > that the internals will be slightly different
> >> > (e.g.
> >> > > > > >> threading
> >> > > > > >> > > > > model).
> >> > > > > >> > > > > > > > > Therefore, I wanted to tighten the switch
> from
> >> > the old
> >> > > > > >> group
> >> > > > > >> > > > > > > > > coordinator to the new group coordinator to a
> >> > major
> >> > > > > >> release.
> >> > > > > >> > > The
> >> > > > > >> > > > > > > > > alternative would be to use a flag to do the
> >> > switch
> >> > > > > >> instead
> >> > > > > >> > of
> >> > > > > >> > > > > relying
> >> > > > > >> > > > > > > > > on the software upgrade.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 3. We need to flesh out the details of the
> >> > migration
> >> > > > > >> story.
> >> > > > > >> > > It
> >> > > > > >> > > > > sounds
> >> > > > > >> > > > > > > > > like
> >> > > > > >> > > > > > > > > > we're saying we will support online
> >> migrations.
> >> > Is
> >> > > > > that
> >> > > > > >> > > correct?
> >> > > > > >> > > > > We
> >> > > > > >> > > > > > > > > should
> >> > > > > >> > > > > > > > > > explain this in detail. It could also be
> done
> >> > as a
> >> > > > > >> separate
> >> > > > > >> > > KIP,
> >> > > > > >> > > > > if
> >> > > > > >> > > > > > > it's
> >> > > > > >> > > > > > > > > > easier.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Yes, we will support online migrations for
> the
> >> > group.
> >> > > > > That
> >> > > > > >> > > means
> >> > > > > >> > > > > that
> >> > > > > >> > > > > > > > > a group using the old protocol will be able
> to
> >> > switch to
> >> > > > > >> the
> >> > > > > >> > > new
> >> > > > > >> > > > > > > > > protocol.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Let me briefly explain how that will work
> >> though.
> >> > It is
> >> > > > > >> > > basically a
> >> > > > > >> > > > > > > > > four step process:
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to
> a
> >> > software
> >> > > > > >> > > supporting
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > new group coordinator. Both the old and the
> new
> >> > > > > >> coordinator
> >> > > > > >> > > will
> >> > > > > >> > > > > > > > > support the old protocol and rely on the same
> >> > persisted
> >> > > > > >> > > metadata so
> >> > > > > >> > > > > > > > > they can work together. This point is an
> >> offline
> >> > > > > >> migration.
> >> > > > > >> > We
> >> > > > > >> > > > > cannot
> >> > > > > >> > > > > > > > > do this one live because it would require
> >> > shutting down
> >> > > > > >> the
> >> > > > > >> > > current
> >> > > > > >> > > > > > > > > coordinator and starting up the new one and
> >> that
> >> > would
> >> > > > > >> cause
> >> > > > > >> > > > > > > > > unavailabilities.
> >> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be
> >> > upgraded
> >> > > > > to
> >> > > > > >> X
> >> > > > > >> > in
> >> > > > > >> > > > > order
> >> > > > > >> > > > > > > > > to enable the new protocol. This cannot be
> done
> >> > before
> >> > > > > 1)
> >> > > > > >> is
> >> > > > > >> > > > > > > > > terminated because the old coordinator
> doesn't
> >> > support
> >> > > > > the
> >> > > > > >> > new
> >> > > > > >> > > > > > > > > protocol.
> >> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a
> version
> >> > > > > supporting
> >> > > > > >> the
> >> > > > > >> > > > > online
> >> > > > > >> > > > > > > > > migration (must have KIP-792). If the
> consumer
> >> is
> >> > > > > already
> >> > > > > >> > > there.
> >> > > > > >> > > > > > > > > Nothing must be done at this point.
> >> > > > > >> > > > > > > > > 4. The consumers must be rolled with the
> >> feature
> >> > flag
> >> > > > > >> turned
> >> > > > > >> > > on.
> >> > > > > >> > > > > The
> >> > > > > >> > > > > > > > > consumer group is automatically converted
> when
> >> > the first
> >> > > > > >> > > consumer
> >> > > > > >> > > > > > > > > using the new protocol joins the group. While
> >> the
> >> > > > > members
> >> > > > > >> > > using the
> >> > > > > >> > > > > > > > > old protocol are being upgraded, the old
> >> protocol
> >> > is
> >> > > > > >> proxied
> >> > > > > >> > > into
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > new one.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the
> pattern
> >> > > > > >> subscriptions
> >> > > > > >> > > to
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > server,
> >> > > > > >> > > > > > > > > > but it seems like there could be some
> tricky
> >> > > > > >> compatibility
> >> > > > > >> > > > > issues.
> >> > > > > >> > > > > > > Will
> >> > > > > >> > > > > > > > > we
> >> > > > > >> > > > > > > > > > have a mechanism for users to detect that
> >> they
> >> > need to
> >> > > > > >> > update
> >> > > > > >> > > > > their
> >> > > > > >> > > > > > > regex
> >> > > > > >> > > > > > > > > > before switching to the new protocol?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I think that I am a bit more optimistic than
> >> you
> >> > on this
> >> > > > > >> > > point. I
> >> > > > > >> > > > > > > > > believe that the majority of the cases are
> >> simple
> >> > > > > regexes
> >> > > > > >> > which
> >> > > > > >> > > > > should
> >> > > > > >> > > > > > > > > work with the new engine. The coordinator
> will
> >> > verify
> >> > > > > the
> >> > > > > >> > regex
> >> > > > > >> > > > > anyway
> >> > > > > >> > > > > > > > > and reject the consumer if the regex is not
> >> valid.
> >> > > > > Coming
> >> > > > > >> > back
> >> > > > > >> > > to
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > migration path, in the worst case, the first
> >> > upgraded
> >> > > > > >> > consumer
> >> > > > > >> > > > > joining
> >> > > > > >> > > > > > > > > the group will be rejected. This should be
> used
> >> > as the
> >> > > > > >> last
> >> > > > > >> > > > > defence, I
> >> > > > > >> > > > > > > > > would say.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > One way for customers to validate their regex
> >> > before
> >> > > > > >> > upgrading
> >> > > > > >> > > > > their
> >> > > > > >> > > > > > > > > prod would be to test them with another
> group.
> >> For
> >> > > > > >> instance,
> >> > > > > >> > > that
> >> > > > > >> > > > > > > > > could be done in a pre-prod environment.
> >> Another
> >> > way
> >> > > > > >> would be
> >> > > > > >> > > to
> >> > > > > >> > > > > > > > > extend the consumer-group tool to provide a
> >> regex
> >> > > > > >> validation
> >> > > > > >> > > > > > > > > mechanism. Would this be enough in your
> >> opinion?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 5. Related to the last question, will the
> >> Java
> >> > client
> >> > > > > >> allow
> >> > > > > >> > > the
> >> > > > > >> > > > > > > users to
> >> > > > > >> > > > > > > > > > stick with the current regex engine for
> >> > compatibility
> >> > > > > >> > > reasons?
> >> > > > > >> > > > > For
> >> > > > > >> > > > > > > > > example,
> >> > > > > >> > > > > > > > > > it may be handy to keep using client based
> >> > regex at
> >> > > > > >> first
> >> > > > > >> > to
> >> > > > > >> > > keep
> >> > > > > >> > > > > > > > > > migrations simple and then migrate to
> server
> >> > based
> >> > > > > >> regexes
> >> > > > > >> > > as a
> >> > > > > >> > > > > > > second
> >> > > > > >> > > > > > > > > step.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I understand your point but I am concerned
> that
> >> > this
> >> > > > > would
> >> > > > > >> > > allow
> >> > > > > >> > > > > users
> >> > > > > >> > > > > > > > > to actually stay in this mode. That would go
> >> > against our
> >> > > > > >> goal
> >> > > > > >> > > of
> >> > > > > >> > > > > > > > > simplifying the client because we would have
> to
> >> > continue
> >> > > > > >> > > monitoring
> >> > > > > >> > > > > > > > > the metadata on the client side. I would
> rather
> >> > not do
> >> > > > > >> this.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 6. When we say that the group coordinator
> >> will
> >> > be
> >> > > > > >> > > responsible for
> >> > > > > >> > > > > > > storing
> >> > > > > >> > > > > > > > > > the configurations and that the
> >> configurations
> >> > will be
> >> > > > > >> > > deleted
> >> > > > > >> > > > > when
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > > group is deleted. Will a transition to DEAD
> >> > trigger
> >> > > > > >> > deletion
> >> > > > > >> > > of
> >> > > > > >> > > > > > > > > > configurations?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > That's right. The configurations will be
> >> deleted
> >> > when
> >> > > > > the
> >> > > > > >> > > group is
> >> > > > > >> > > > > > > > > deleted. They go together.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 7. Will the choice to store the configs in
> >> the
> >> > group
> >> > > > > >> > > coordinator
> >> > > > > >> > > > > > > make it
> >> > > > > >> > > > > > > > > > harder to list all cluster configs and
> their
> >> > values?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I don't think so. The group configurations
> are
> >> > overrides
> >> > > > > >> of
> >> > > > > >> > > cluster
> >> > > > > >> > > > > > > > > configs. If you want to know all the
> overrides
> >> > though,
> >> > > > > you
> >> > > > > >> > > would
> >> > > > > >> > > > > have
> >> > > > > >> > > > > > > > > to ask all the group coordinators. You cannot
> >> > rely on
> >> > > > > the
> >> > > > > >> > > metadata
> >> > > > > >> > > > > log
> >> > > > > >> > > > > > > > > for instance.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 8. How would someone configure a group
> before
> >> > starting
> >> > > > > >> the
> >> > > > > >> > > > > consumers?
> >> > > > > >> > > > > > > > > Have
> >> > > > > >> > > > > > > > > > we considered allowing the explicit
> creation
> >> of
> >> > > > > groups?
> >> > > > > >> > > > > > > Alternatively,
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > configs could be decoupled from the group
> >> > lifecycle.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Yes. The group will be automatically created
> in
> >> > this
> >> > > > > case.
> >> > > > > >> > > However,
> >> > > > > >> > > > > > > > > the configs will be lost after the retention
> >> > period of
> >> > > > > the
> >> > > > > >> > > group
> >> > > > > >> > > > > > > > > passes.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for
> the
> >> > Java
> >> > > > > >> client
> >> > > > > >> > > still
> >> > > > > >> > > > > take
> >> > > > > >> > > > > > > a
> >> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to
> >> > introduce
> >> > > > > an
> >> > > > > >> > > overload?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > That's a very group question. I forgot about
> >> that
> >> > one.
> >> > > > > As
> >> > > > > >> the
> >> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
> >> compatible
> >> > with
> >> > > > > the
> >> > > > > >> > > engine
> >> > > > > >> > > > > that
> >> > > > > >> > > > > > > > > we plan to use, it might be better to
> deprecate
> >> > it and
> >> > > > > >> use an
> >> > > > > >> > > > > overload
> >> > > > > >> > > > > > > > > which takes a string. We would rely on the
> >> server
> >> > side
> >> > > > > >> > > validation.
> >> > > > > >> > > > > > > > > During the migration, I think that we could
> >> still
> >> > try to
> >> > > > > >> > > toString
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > regex and use it. That should work, I think,
> in
> >> > the
> >> > > > > >> majority
> >> > > > > >> > > of the
> >> > > > > >> > > > > > > > > cases.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 10. I agree with Justine that we should be
> >> > clearer
> >> > > > > about
> >> > > > > >> > the
> >> > > > > >> > > > > reason
> >> > > > > >> > > > > > > to
> >> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
> >> feature
> >> > flag.
> >> > > > > >> Maybe
> >> > > > > >> > > we
> >> > > > > >> > > > > mean
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > we
> >> > > > > >> > > > > > > > > > can switch the default for the feature flag
> >> to
> >> > true
> >> > > > > >> based
> >> > > > > >> > on
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > > metadata.version once we want to make it
> the
> >> > default.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > My plan was to use that feature flag mainly
> >> > during the
> >> > > > > >> > > development
> >> > > > > >> > > > > > > > > phase. I should not have mentioned it, I
> think,
> >> > because
> >> > > > > we
> >> > > > > >> > > could
> >> > > > > >> > > > > use
> >> > > > > >> > > > > > > > > an internal config for it.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention
> >> the
> >> > > > > required
> >> > > > > >> > > ACLs, it
> >> > > > > >> > > > > > > would
> >> > > > > >> > > > > > > > > be
> >> > > > > >> > > > > > > > > > good to add that for consistency.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Noted.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 12. It is a bit odd that
> >> ConsumerGroupHeartbeat
> >> > > > > requires
> >> > > > > >> > > "Read
> >> > > > > >> > > > > Group"
> >> > > > > >> > > > > > > > > even
> >> > > > > >> > > > > > > > > > though it seems to do more than reading.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > I agree. This is how the current protocol
> works
> >> > though.
> >> > > > > We
> >> > > > > >> > only
> >> > > > > >> > > > > > > > > require "Read Group" to join a group. We
> could
> >> > consider
> >> > > > > >> > > changing
> >> > > > > >> > > > > this
> >> > > > > >> > > > > > > > > but I am not sure that it is worth it.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 13. How is topic recreation handled by the
> >> > consumer
> >> > > > > with
> >> > > > > >> > the
> >> > > > > >> > > new
> >> > > > > >> > > > > > > group
> >> > > > > >> > > > > > > > > > protocol? It would be good to have a
> section
> >> on
> >> > this.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Noted. From a protocol perspective, the new
> >> topic
> >> > will
> >> > > > > >> have a
> >> > > > > >> > > new
> >> > > > > >> > > > > > > > > topic id so it will treat it like a topic
> with
> >> a
> >> > > > > different
> >> > > > > >> > > name.
> >> > > > > >> > > > > The
> >> > > > > >> > > > > > > > > only issue is that the fetch/commit offsets
> >> APIs
> >> > do not
> >> > > > > >> > support
> >> > > > > >> > > > > topic
> >> > > > > >> > > > > > > > > IDs so the consumer would reuse the offsets
> >> based
> >> > on the
> >> > > > > >> > same.
> >> > > > > >> > > I
> >> > > > > >> > > > > think
> >> > > > > >> > > > > > > > > that we should update those APIs as well in
> >> order
> >> > to be
> >> > > > > >> > > consistent
> >> > > > > >> > > > > end
> >> > > > > >> > > > > > > > > to end. That would strengthen the semantics
> of
> >> the
> >> > > > > >> consumer.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the new
> >> > coordinator
> >> > > > > >> in
> >> > > > > >> > > Java.
> >> > > > > >> > > > > Even
> >> > > > > >> > > > > > > > > though
> >> > > > > >> > > > > > > > > > this is an implementation detail, do we
> plan
> >> to
> >> > have a
> >> > > > > >> new
> >> > > > > >> > > gradle
> >> > > > > >> > > > > > > module
> >> > > > > >> > > > > > > > > > for it?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Yes.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when it
> >> comes
> >> > to how
> >> > > > > >> many
> >> > > > > >> > > > > members
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > new
> >> > > > > >> > > > > > > > > > group protocol can support?
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > We don't have numbers at the moment. The
> >> protocol
> >> > should
> >> > > > > >> > > support
> >> > > > > >> > > > > 1000s
> >> > > > > >> > > > > > > > > of members per group. We will measure this
> when
> >> > we have
> >> > > > > a
> >> > > > > >> > first
> >> > > > > >> > > > > > > > > implementation. Note that we might have other
> >> > > > > bottlenecks
> >> > > > > >> > down
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > road (e.g. offset commits).
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > > 16. Did we consider having
> SubscribedTopidIds
> >> > instead
> >> > > > > of
> >> > > > > >> > > > > > > > > > SubscribedTopicNames in
> >> > ConsumerGroupHeartbeatRequest?
> >> > > > > >> Is
> >> > > > > >> > the
> >> > > > > >> > > > > idea
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > > since we have to resolve the regex on the
> >> > server, we
> >> > > > > >> can do
> >> > > > > >> > > the
> >> > > > > >> > > > > same
> >> > > > > >> > > > > > > for
> >> > > > > >> > > > > > > > > > the topic name? The difference is that
> >> sending
> >> > the
> >> > > > > >> regex is
> >> > > > > >> > > more
> >> > > > > >> > > > > > > > > efficient
> >> > > > > >> > > > > > > > > > whereas sending the topic names is less
> >> > efficient.
> >> > > > > >> > > Furthermore,
> >> > > > > >> > > > > > > delete
> >> > > > > >> > > > > > > > > and
> >> > > > > >> > > > > > > > > > recreation is easier to handle if we have
> >> topic
> >> > ids.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > The idea was to consolidate the metadata
> lookup
> >> > on the
> >> > > > > >> server
> >> > > > > >> > > for
> >> > > > > >> > > > > both
> >> > > > > >> > > > > > > > > paths but I do agree with your point. As a
> >> second
> >> > > > > though,
> >> > > > > >> > using
> >> > > > > >> > > > > topic
> >> > > > > >> > > > > > > > > ids may be better here for the delete and
> >> > recreation
> >> > > > > case.
> >> > > > > >> > > Also, I
> >> > > > > >> > > > > > > > > suppose that we may allow users to subscribe
> >> with
> >> > topic
> >> > > > > >> ids
> >> > > > > >> > in
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > future because that is the only way to be
> >> really
> >> > robust
> >> > > > > to
> >> > > > > >> > > topic
> >> > > > > >> > > > > > > > > re-creation.
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > Best,
> >> > > > > >> > > > > > > > > David
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> >> > > > > >> > > djacot@confluent.io>
> >> > > > > >> > > > > > > wrote:
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > Hi Justine,
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > Thanks for your comments. Please find my
> >> answers
> >> > > > > below.
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs
> >> > with the
> >> > > > > >> > > exception
> >> > > > > >> > > > > of the
> >> > > > > >> > > > > > > > > > topic names based in the
> >> > > > > ConsumerGroupHeartbeatRequest.
> >> > > > > >> I
> >> > > > > >> > am
> >> > > > > >> > > not
> >> > > > > >> > > > > sure
> >> > > > > >> > > > > > > > > > if using topic names is the right call
> here.
> >> I
> >> > need to
> >> > > > > >> > think
> >> > > > > >> > > > > about it
> >> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does not
> >> > change the
> >> > > > > >> > > > > fetch/commit
> >> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be
> >> > something
> >> > > > > >> that
> >> > > > > >> > we
> >> > > > > >> > > > > should
> >> > > > > >> > > > > > > > > > include though as it would give better
> >> overall
> >> > > > > >> guarantee in
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > > producer.
> >> > > > > >> > > > > > > > > > - You're right. I think that I should not
> >> have
> >> > > > > mentioned
> >> > > > > >> > this
> >> > > > > >> > > > > flag at
> >> > > > > >> > > > > > > > > > all. I will remove it. We can use an
> internal
> >> > > > > >> configuration
> >> > > > > >> > > while
> >> > > > > >> > > > > > > > > > developing the feature.
> >> > > > > >> > > > > > > > > > - Both cluster types will be supported. The
> >> > change is
> >> > > > > >> > > > > orthogonal. The
> >> > > > > >> > > > > > > > > > only requirement is that the cluster uses
> >> topic
> >> > IDs.
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > Best,
> >> > > > > >> > > > > > > > > > David
> >> > > > > >> > > > > > > > > >
> >> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang
> >> Wang <
> >> > > > > >> > > > > wangguoz@gmail.com>
> >> > > > > >> > > > > > > > > wrote:
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Hi Ismael,
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are some
> >> replies
> >> > > > > inlined
> >> > > > > >> > > below:
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael
> Juma
> >> <
> >> > > > > >> > > ismael@juma.me.uk>
> >> > > > > >> > > > > > > wrote:
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the
> >> potential
> >> > to be a
> >> > > > > >> > great
> >> > > > > >> > > > > > > > > improvement. A few
> >> > > > > >> > > > > > > > > > > > initial questions/comments:
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk about
> >> > target
> >> > > > > >> versions
> >> > > > > >> > > for
> >> > > > > >> > > > > > > > > deprecation and
> >> > > > > >> > > > > > > > > > > > removal of the existing group protocol.
> >> > Unlike
> >> > > > > >> KRaft,
> >> > > > > >> > > this
> >> > > > > >> > > > > > > affects a
> >> > > > > >> > > > > > > > > core
> >> > > > > >> > > > > > > > > > > > client protocol and hence
> >> > deprecation/removal will
> >> > > > > >> be
> >> > > > > >> > > heavily
> >> > > > > >> > > > > > > > > dependent on
> >> > > > > >> > > > > > > > > > > > how quickly applications migrate to the
> >> new
> >> > > > > >> protocol.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can
> >> remove
> >> > the
> >> > > > > >> proposed
> >> > > > > >> > > > > timeline
> >> > > > > >> > > > > > > in
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and
> Migration
> >> > Plan` and
> >> > > > > >> > > instead
> >> > > > > >> > > > > just
> >> > > > > >> > > > > > > state
> >> > > > > >> > > > > > > > > > > that we will decide in the future about
> >> when
> >> > we
> >> > > > > would
> >> > > > > >> > > > > deprecate old
> >> > > > > >> > > > > > > > > > > protocol and behaviors.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to release
> this
> >> > in 4.x,
> >> > > > > >> but
> >> > > > > >> > it
> >> > > > > >> > > > > wasn't
> >> > > > > >> > > > > > > made
> >> > > > > >> > > > > > > > > clear
> >> > > > > >> > > > > > > > > > > > why. If we added that as a way to
> >> estimate
> >> > when
> >> > > > > we'd
> >> > > > > >> > > > > deprecate
> >> > > > > >> > > > > > > and
> >> > > > > >> > > > > > > > > remove
> >> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
> >> removing
> >> > this
> >> > > > > >> part.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > I think that's not specifically related
> to
> >> the
> >> > > > > >> > > > > deprecation/removal
> >> > > > > >> > > > > > > > > timeline
> >> > > > > >> > > > > > > > > > > plan, but it's more for client upgrades.
> >> I.e.
> >> > the
> >> > > > > >> > > broker-side
> >> > > > > >> > > > > > > > > > > implementation may be done first, and
> then
> >> the
> >> > > > > client
> >> > > > > >> > side,
> >> > > > > >> > > > > and we
> >> > > > > >> > > > > > > > > would
> >> > > > > >> > > > > > > > > > > only mark it as "released" by the time
> >> clients
> >> > > > > >> > > implementations
> >> > > > > >> > > > > are
> >> > > > > >> > > > > > > > > done. At
> >> > > > > >> > > > > > > > > > > that time, to enable the feature the
> >> clients
> >> > need to
> >> > > > > >> > first
> >> > > > > >> > > > > swap-in
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and then
> set
> >> > the flag
> >> > > > > >> > with a
> >> > > > > >> > > > > second
> >> > > > > >> > > > > > > > > rolling
> >> > > > > >> > > > > > > > > > > bounce, and hence we feel it's better to
> be
> >> > released
> >> > > > > >> in a
> >> > > > > >> > > major
> >> > > > > >> > > > > > > > > version.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 3. We need to flesh out the details of
> >> the
> >> > > > > migration
> >> > > > > >> > > story.
> >> > > > > >> > > > > It
> >> > > > > >> > > > > > > > > sounds like
> >> > > > > >> > > > > > > > > > > > we're saying we will support online
> >> > migrations. Is
> >> > > > > >> that
> >> > > > > >> > > > > correct?
> >> > > > > >> > > > > > > We
> >> > > > > >> > > > > > > > > should
> >> > > > > >> > > > > > > > > > > > explain this in detail. It could also
> be
> >> > done as a
> >> > > > > >> > > separate
> >> > > > > >> > > > > KIP,
> >> > > > > >> > > > > > > if
> >> > > > > >> > > > > > > > > it's
> >> > > > > >> > > > > > > > > > > > easier.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yes I think that's the part we can be
> more
> >> > concrete
> >> > > > > >> about
> >> > > > > >> > > for
> >> > > > > >> > > > > sure
> >> > > > > >> > > > > > > (and
> >> > > > > >> > > > > > > > > > > this is related to your question 2)
> above).
> >> > We will
> >> > > > > >> work
> >> > > > > >> > on
> >> > > > > >> > > > > making
> >> > > > > >> > > > > > > it
> >> > > > > >> > > > > > > > > more
> >> > > > > >> > > > > > > > > > > explicit in parallel as we solicit more
> >> > feedback.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing the
> >> > pattern
> >> > > > > >> > > subscriptions
> >> > > > > >> > > > > to
> >> > > > > >> > > > > > > the
> >> > > > > >> > > > > > > > > server,
> >> > > > > >> > > > > > > > > > > > but it seems like there could be some
> >> tricky
> >> > > > > >> > > compatibility
> >> > > > > >> > > > > > > issues.
> >> > > > > >> > > > > > > > > Will we
> >> > > > > >> > > > > > > > > > > > have a mechanism for users to detect
> that
> >> > they
> >> > > > > need
> >> > > > > >> to
> >> > > > > >> > > update
> >> > > > > >> > > > > > > their
> >> > > > > >> > > > > > > > > regex
> >> > > > > >> > > > > > > > > > > > before switching to the new protocol?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
> >> non-java
> >> > client
> >> > > > > >> > users
> >> > > > > >> > > to
> >> > > > > >> > > > > sort
> >> > > > > >> > > > > > > of
> >> > > > > >> > > > > > > > > > > "dry-run" the client before switching to
> >> the
> >> > new
> >> > > > > >> > protocol.
> >> > > > > >> > > I
> >> > > > > >> > > > > do not
> >> > > > > >> > > > > > > > > have a
> >> > > > > >> > > > > > > > > > > specific idea on top of my head though,
> >> maybe
> >> > others
> >> > > > > >> like
> >> > > > > >> > > @Matt
> >> > > > > >> > > > > > > > > Howlett can
> >> > > > > >> > > > > > > > > > > chime-in here?
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 5. Related to the last question, will
> the
> >> > Java
> >> > > > > >> client
> >> > > > > >> > > allow
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > users to
> >> > > > > >> > > > > > > > > > > > stick with the current regex engine for
> >> > > > > >> compatibility
> >> > > > > >> > > > > reasons?
> >> > > > > >> > > > > > > For
> >> > > > > >> > > > > > > > > example,
> >> > > > > >> > > > > > > > > > > > it may be handy to keep using client
> >> based
> >> > regex
> >> > > > > at
> >> > > > > >> > > first to
> >> > > > > >> > > > > keep
> >> > > > > >> > > > > > > > > > > > migrations simple and then migrate to
> >> > server based
> >> > > > > >> > > regexes
> >> > > > > >> > > > > as a
> >> > > > > >> > > > > > > > > second
> >> > > > > >> > > > > > > > > > > > step.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Honestly I have not thought about that
> for
> >> > java
> >> > > > > >> clients,
> >> > > > > >> > > and
> >> > > > > >> > > > > we can
> >> > > > > >> > > > > > > > > discuss
> >> > > > > >> > > > > > > > > > > that. What kind of compatibility issues
> do
> >> > you have
> >> > > > > in
> >> > > > > >> > > mind?
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 6. When we say that the group
> coordinator
> >> > will be
> >> > > > > >> > > > > responsible for
> >> > > > > >> > > > > > > > > storing
> >> > > > > >> > > > > > > > > > > > the configurations and that the
> >> > configurations
> >> > > > > will
> >> > > > > >> be
> >> > > > > >> > > > > deleted
> >> > > > > >> > > > > > > when
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > > > group is deleted. Will a transition to
> >> DEAD
> >> > > > > trigger
> >> > > > > >> > > deletion
> >> > > > > >> > > > > of
> >> > > > > >> > > > > > > > > > > > configurations?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an ending
> >> state
> >> > (we
> >> > > > > would
> >> > > > > >> > only
> >> > > > > >> > > > > > > transit to
> >> > > > > >> > > > > > > > > that
> >> > > > > >> > > > > > > > > > > state when the group is EMPTY and also
> all
> >> of
> >> > its
> >> > > > > >> > metadata
> >> > > > > >> > > are
> >> > > > > >> > > > > > > gone),
> >> > > > > >> > > > > > > > > once
> >> > > > > >> > > > > > > > > > > it's transited to DEAD this group would
> >> never
> >> > be
> >> > > > > >> revived.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 7. Will the choice to store the configs
> >> in
> >> > the
> >> > > > > group
> >> > > > > >> > > > > coordinator
> >> > > > > >> > > > > > > > > make it
> >> > > > > >> > > > > > > > > > > > harder to list all cluster configs and
> >> their
> >> > > > > values?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > That's a good question, and our thoughts
> >> are
> >> > that
> >> > > > > the
> >> > > > > >> > > so-called
> >> > > > > >> > > > > > > "group
> >> > > > > >> > > > > > > > > > > configurations" are overrides of the
> >> > cluster-level
> >> > > > > >> > > > > configurations
> >> > > > > >> > > > > > > > > > > customized per group so when an admin
> list
> >> > cluster
> >> > > > > >> > configs
> >> > > > > >> > > it's
> >> > > > > >> > > > > > > okay to
> >> > > > > >> > > > > > > > > > > list just the cluster-level "defaults",
> not
> >> > showing
> >> > > > > >> any
> >> > > > > >> > > > > per-group
> >> > > > > >> > > > > > > > > > > customizations.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 8. How would someone configure a group
> >> > before
> >> > > > > >> starting
> >> > > > > >> > > the
> >> > > > > >> > > > > > > > > consumers? Have
> >> > > > > >> > > > > > > > > > > > we considered allowing the explicit
> >> > creation of
> >> > > > > >> groups?
> >> > > > > >> > > > > > > > > Alternatively, the
> >> > > > > >> > > > > > > > > > > > configs could be decoupled from the
> group
> >> > > > > lifecycle.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > The configs can be created before the
> group
> >> > itself
> >> > > > > as
> >> > > > > >> an
> >> > > > > >> > > > > > > independent
> >> > > > > >> > > > > > > > > entity
> >> > > > > >> > > > > > > > > > > --- of course, this requires the
> >> corresponding
> >> > > > > >> request to
> >> > > > > >> > > be
> >> > > > > >> > > > > > > routed to
> >> > > > > >> > > > > > > > > the
> >> > > > > >> > > > > > > > > > > right coordinator based on the group id
> ---
> >> > the only
> >> > > > > >> > thing
> >> > > > > >> > > that
> >> > > > > >> > > > > > > > > differs is,
> >> > > > > >> > > > > > > > > > > when the group itself is gone we also
> check
> >> > if there
> >> > > > > >> are
> >> > > > > >> > > any
> >> > > > > >> > > > > > > > > configuration
> >> > > > > >> > > > > > > > > > > entities related to that group and delete
> >> as
> >> > well.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> >> > asymmetry on
> >> > > > > the
> >> > > > > >> > > creation
> >> > > > > >> > > > > /
> >> > > > > >> > > > > > > > > deletion
> >> > > > > >> > > > > > > > > > > lifecycles of the config entities, and we
> >> > would like
> >> > > > > >> to
> >> > > > > >> > > hear
> >> > > > > >> > > > > > > everyone's
> >> > > > > >> > > > > > > > > > > feelings whether we should aim for
> symmetry
> >> > i.e.
> >> > > > > >> totally
> >> > > > > >> > > > > decouple
> >> > > > > >> > > > > > > group
> >> > > > > >> > > > > > > > > > > configs and hence not delete them at all
> >> when
> >> > the
> >> > > > > >> group
> >> > > > > >> > is
> >> > > > > >> > > > > gone,
> >> > > > > >> > > > > > > but
> >> > > > > >> > > > > > > > > always
> >> > > > > >> > > > > > > > > > > require explicit deletion operations by
> >> > themselves.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method
> for
> >> > the Java
> >> > > > > >> > client
> >> > > > > >> > > > > still
> >> > > > > >> > > > > > > take
> >> > > > > >> > > > > > > > > a
> >> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have
> >> to
> >> > > > > >> introduce an
> >> > > > > >> > > > > overload?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > I think we do not need to introduce an
> >> > overload, but
> >> > > > > >> I'm
> >> > > > > >> > > all
> >> > > > > >> > > > > ears
> >> > > > > >> > > > > > > if
> >> > > > > >> > > > > > > > > there
> >> > > > > >> > > > > > > > > > > may be some compatibility issues that we
> >> may
> >> > > > > overlook.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we should
> >> be
> >> > clearer
> >> > > > > >> > about
> >> > > > > >> > > the
> >> > > > > >> > > > > > > reason
> >> > > > > >> > > > > > > > > to
> >> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version from the
> >> > feature
> >> > > > > >> flag.
> >> > > > > >> > > Maybe
> >> > > > > >> > > > > we
> >> > > > > >> > > > > > > mean
> >> > > > > >> > > > > > > > > that we
> >> > > > > >> > > > > > > > > > > > can switch the default for the feature
> >> flag
> >> > to
> >> > > > > true
> >> > > > > >> > > based on
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > > > > metadata.version once we want to make
> it
> >> the
> >> > > > > >> default.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't
> mention
> >> > the
> >> > > > > >> required
> >> > > > > >> > > ACLs,
> >> > > > > >> > > > > it
> >> > > > > >> > > > > > > > > would be
> >> > > > > >> > > > > > > > > > > > good to add that for consistency.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> >> > ConsumerGroupHeartbeat
> >> > > > > >> > requires
> >> > > > > >> > > > > "Read
> >> > > > > >> > > > > > > > > Group" even
> >> > > > > >> > > > > > > > > > > > though it seems to do more than
> reading.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > I had that thought myself as well, but in
> >> the
> >> > end we
> >> > > > > >> > could
> >> > > > > >> > > not
> >> > > > > >> > > > > > > find a
> >> > > > > >> > > > > > > > > > > better alternative: adding Write Group
> >> seems
> >> > an
> >> > > > > >> overkill
> >> > > > > >> > > here
> >> > > > > >> > > > > > > since we
> >> > > > > >> > > > > > > > > do
> >> > > > > >> > > > > > > > > > > not have it elsewhere (we only have Read
> /
> >> > Delete
> >> > > > > and
> >> > > > > >> > > Describe
> >> > > > > >> > > > > on
> >> > > > > >> > > > > > > > > groups so
> >> > > > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 13. How is topic recreation handled by
> >> the
> >> > > > > consumer
> >> > > > > >> > with
> >> > > > > >> > > the
> >> > > > > >> > > > > new
> >> > > > > >> > > > > > > > > group
> >> > > > > >> > > > > > > > > > > > protocol? It would be good to have a
> >> > section on
> >> > > > > >> this.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > You mean with regex subscription right?
> Yes
> >> > we can
> >> > > > > >> add a
> >> > > > > >> > > > > section
> >> > > > > >> > > > > > > about
> >> > > > > >> > > > > > > > > > > that, but basically the idea is that
> >> consumer
> >> > would
> >> > > > > be
> >> > > > > >> > > totally
> >> > > > > >> > > > > > > > > agnostic in
> >> > > > > >> > > > > > > > > > > the new protocol as it's handled all by
> the
> >> > brokers.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write the
> >> new
> >> > > > > >> coordinator
> >> > > > > >> > in
> >> > > > > >> > > > > Java.
> >> > > > > >> > > > > > > Even
> >> > > > > >> > > > > > > > > though
> >> > > > > >> > > > > > > > > > > > this is an implementation detail, do we
> >> > plan to
> >> > > > > >> have a
> >> > > > > >> > > new
> >> > > > > >> > > > > gradle
> >> > > > > >> > > > > > > > > module
> >> > > > > >> > > > > > > > > > > > for it?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > We have not thought about that. But I
> think
> >> > the
> >> > > > > answer
> >> > > > > >> > > should
> >> > > > > >> > > > > be
> >> > > > > >> > > > > > > yes.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal when
> it
> >> > comes to
> >> > > > > >> how
> >> > > > > >> > > many
> >> > > > > >> > > > > > > members
> >> > > > > >> > > > > > > > > the new
> >> > > > > >> > > > > > > > > > > > group protocol can support?
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > Within a group, I think we should shoot
> for
> >> > 1000s of
> >> > > > > >> > > members.
> >> > > > > >> > > > > But
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > > > scalability goals also depend on the
> offset
> >> > > > > management
> >> > > > > >> > > (commit,
> >> > > > > >> > > > > > > fetch)
> >> > > > > >> > > > > > > > > > > capabilities of the coordinator which we
> >> did
> >> > not
> >> > > > > >> cover in
> >> > > > > >> > > this
> >> > > > > >> > > > > > > KIP, so
> >> > > > > >> > > > > > > > > it's
> >> > > > > >> > > > > > > > > > > hard to give a number that applies
> >> > universally.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > 16. Did we consider having
> >> > SubscribedTopidIds
> >> > > > > >> instead
> >> > > > > >> > of
> >> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> >> > > > > >> ConsumerGroupHeartbeatRequest?
> >> > > > > >> > > Is the
> >> > > > > >> > > > > > > idea
> >> > > > > >> > > > > > > > > that
> >> > > > > >> > > > > > > > > > > > since we have to resolve the regex on
> the
> >> > server,
> >> > > > > we
> >> > > > > >> > can
> >> > > > > >> > > do
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > same
> >> > > > > >> > > > > > > > > for
> >> > > > > >> > > > > > > > > > > > the topic name? The difference is that
> >> > sending the
> >> > > > > >> > regex
> >> > > > > >> > > is
> >> > > > > >> > > > > more
> >> > > > > >> > > > > > > > > efficient
> >> > > > > >> > > > > > > > > > > > whereas sending the topic names is less
> >> > efficient.
> >> > > > > >> > > > > Furthermore,
> >> > > > > >> > > > > > > > > delete and
> >> > > > > >> > > > > > > > > > > > recreation is easier to handle if we
> have
> >> > topic
> >> > > > > ids.
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > The main reason to still let the clients
> >> send
> >> > names
> >> > > > > >> is to
> >> > > > > >> > > keep
> >> > > > > >> > > > > the
> >> > > > > >> > > > > > > > > > > reasoning of names -> ids on the broker /
> >> > admin
> >> > > > > client
> >> > > > > >> > > only.
> >> > > > > >> > > > > Note
> >> > > > > >> > > > > > > that
> >> > > > > >> > > > > > > > > > > although we added topic id in KIP-516, we
> >> > never
> >> > > > > >> > > implemented the
> >> > > > > >> > > > > > > logic
> >> > > > > >> > > > > > > > > on
> >> > > > > >> > > > > > > > > > > consumer/producers leveraging the related
> >> > newer
> >> > > > > >> versioned
> >> > > > > >> > > RPCs,
> >> > > > > >> > > > > > > > > instead we
> >> > > > > >> > > > > > > > > > > just set the topic id as empty UUID. We
> >> want
> >> > to keep
> >> > > > > >> the
> >> > > > > >> > > > > > > > > consumer/producer
> >> > > > > >> > > > > > > > > > > to be thin and only delegate the
> reasoning
> >> on
> >> > broker
> >> > > > > >> and
> >> > > > > >> > > > > > > potentially
> >> > > > > >> > > > > > > > > admin
> >> > > > > >> > > > > > > > > > > clients.
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > Thanks,
> >> > > > > >> > > > > > > > > > > > Ismael
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David
> >> Jacot
> >> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> >> > > > > >> > > > > > > > > > > > wrote:
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > Hi all,
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > I would like to start a discussion
> >> thread
> >> > on
> >> > > > > >> KIP-848:
> >> > > > > >> > > The
> >> > > > > >> > > > > Next
> >> > > > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance
> >> > Protocol.
> >> > > > > >> With
> >> > > > > >> > > this
> >> > > > > >> > > > > KIP,
> >> > > > > >> > > > > > > we
> >> > > > > >> > > > > > > > > aim
> >> > > > > >> > > > > > > > > > > > > to make the rebalance protocol (for
> >> > consumers)
> >> > > > > >> more
> >> > > > > >> > > > > reliable,
> >> > > > > >> > > > > > > more
> >> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
> >> > clients, and
> >> > > > > >> easier
> >> > > > > >> > > to
> >> > > > > >> > > > > debug
> >> > > > > >> > > > > > > for
> >> > > > > >> > > > > > > > > > > > > operators.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > The KIP is here:
> >> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > Please take a look and let me know
> what
> >> > you
> >> > > > > think.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > Best,
> >> > > > > >> > > > > > > > > > > > > David
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > > > PS: I will be away from July 18th to
> >> > August 8th.
> >> > > > > >> That
> >> > > > > >> > > gives
> >> > > > > >> > > > > > > you a
> >> > > > > >> > > > > > > > > bit
> >> > > > > >> > > > > > > > > > > > > of time to read and digest this long
> >> KIP.
> >> > > > > >> > > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > >
> >> > > > > >> > > > > > > > > > > --
> >> > > > > >> > > > > > > > > > > -- Guozhang
> >> > > > > >> > > > > > > > >
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > >
> >> > > > > >> > > > > > > > --
> >> > > > > >> > > > > > > > -- Guozhang
> >> > > > > >> > > > > > >
> >> > > > > >> > > > >
> >> > > > > >> > >
> >> > > > > >> > >
> >> > > > > >> >
> >> > > > > >>
> >> > > > > >>
> >> > > > > >> --
> >> > > > > >> -- Guozhang
> >> > > > > >>
> >> > > > > >
> >> > > > >
> >> >
> >>
> >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Luke Chen <sh...@gmail.com>]

Hi David,

Thanks for the update.

Some more questions:
1. In Group Coordinator section, you mentioned:
> The new group coordinator will have a state machine per
*__consumer_offsets* partitions, where each state machine is modelled as an
event loop. Those state machines will be executed in
*group.coordinator.threads* threads.

1.1. I think the state machine are: "Empty, assigning, reconciling, stable,
dead" mentioned in Consumer Group States section, right?
1.2. What do you mean "each state machine is modelled as an event loop"?
1.3. Why do we need a state machine per *__consumer_offsets* partitions?
Not a state machine "per consumer group" owned by a group coordinator? For
example, if one group coordinator owns 2 consumer groups, and both exist in
*__consumer_offsets-0*, will we have 1 state machine for it, or 2?
1.4. I know the "*group.coordinator.threads" *is the number of threads used
to run the state machines. But I'm wondering if the purpose of the threads
is only to keep the state of each consumer group (or *__consumer_offsets*
partitions?), and no heavy computation, why should we need multi-threads
here?

2. For the default value in the new configs:
2.1. The consumer session timeout, why does the default session timeout not
locate between min (45s) and max(60s)? I thought the min/max session
timeout is to define lower/upper bound of it, no?

group.consumer.session.timeout.ms int 30s The timeout to detect client
failures when using the consumer group protocol.
group.consumer.min.session.timeout.ms int 45s The minimum session timeout.
group.consumer.max.session.timeout.ms int 60s The maximum session timeout.



2.2. The default server side assignor are [range, uniform], which means
we'll default to "range" assignor. I'd like to know why not uniform one? I
thought usually users will choose uniform assignor (former sticky assinor)
for better evenly distribution. Any other reason we choose range assignor
as default?
group.consumer.assignors List range, uniform The server side assignors.






Thank you.
Luke






On Mon, Aug 22, 2022 at 2:10 PM Luke Chen <sh...@gmail.com> wrote:

> Hi Sagar,
>
> I have some thoughts about Kafka Connect integrating with KIP-848, but I
> think we should have a separate discussion thread for the Kafka Connect
> KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol [1],
> and let this discussion thread focus on consumer rebalance protocol, WDYT?
>
> [1]
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>
> Thank you.
> Luke
>
>
>
> On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com> wrote:
>
>> Thank you Guozhang/David for the feedback. Looks like there's agreement on
>> using separate APIs for Connect. I would revisit the doc and see what
>> changes are to be made.
>>
>> Thanks!
>> Sagar.
>>
>> On Tue, Aug 9, 2022 at 7:11 PM David Jacot <dj...@confluent.io.invalid>
>> wrote:
>>
>> > Hi Sagar,
>> >
>> > Thanks for the feedback and the document. That's really helpful. I
>> > will take a look at it.
>> >
>> > Overall, it seems to me that both Connect and the Consumer could share
>> > the same underlying "engine". The main difference is that the Consumer
>> > assigns topic-partitions to members whereas Connect assigns tasks to
>> > workers. I see two ways to move forward:
>> > 1) We extend the new proposed APIs to support different resource types
>> > (e.g. partitions, tasks, etc.); or
>> > 2) We use new dedicated APIs for Connect. The dedicated APIs would be
>> > similar to the new ones but different on the content/resources and
>> > they would rely on the same engine on the coordinator side.
>> >
>> > I personally lean towards 2) because I am not a fan of overcharging
>> > APIs to serve different purposes. That being said, I am not opposed to
>> > 1) if we can find an elegant way to do it.
>> >
>> > I think that we can continue to discuss it here for now in order to
>> > ensure that this KIP is compatible with what we will do for Connect in
>> > the future.
>> >
>> > Best,
>> > David
>> >
>> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io> wrote:
>> > >
>> > > Hi all,
>> > >
>> > > I am back from vacation. I will go through and address your comments
>> > > in the coming days. Thanks for your feedback.
>> > >
>> > > Cheers,
>> > > David
>> > >
>> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gharris1727@gmail.com
>> >
>> > wrote:
>> > > >
>> > > > Hey All!
>> > > >
>> > > > Thanks for the KIP, it's wonderful to see cooperative rebalancing
>> > making it
>> > > > down the stack!
>> > > >
>> > > > I had a few questions:
>> > > >
>> > > > 1. The 'Rejected Alternatives' section describes how member epoch
>> > should
>> > > > advance in step with the group epoch and assignment epoch values. I
>> > think
>> > > > that this is a good idea for the reasons described in the KIP. When
>> the
>> > > > protocol is incrementally assigning partitions to a worker, what
>> member
>> > > > epoch does each incremental assignment use? Are member epochs
>> re-used,
>> > and
>> > > > a single member epoch can correspond to multiple different
>> > (monotonically
>> > > > larger) assignments?
>> > > >
>> > > > 2. Is the Assignor's 'Reason' field opaque to the group
>> coordinator? If
>> > > > not, should custom client-side assignor implementations interact
>> with
>> > the
>> > > > Reason field, and how is its common meaning agreed upon? If so, what
>> > is the
>> > > > benefit of a distinct Reason field over including such functionality
>> > in the
>> > > > opaque metadata?
>> > > >
>> > > > 3. The following is included in the KIP: "Thanks to this, the input
>> of
>> > the
>> > > > client side assignor is entirely driven by the group coordinator.
>> The
>> > > > consumer is no longer responsible for maintaining any state besides
>> its
>> > > > assigned partitions." Does this mean that the client-side assignor
>> MAY
>> > > > incorporate additional non-Metadata state (such as partition
>> > throughput,
>> > > > cpu/memory metrics, config topics, etc), or that additional
>> > non-Metadata
>> > > > state SHOULD NOT be used?
>> > > >
>> > > > 4. I see that there are separate classes
>> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
>> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
>> > > > overlap significantly. Is it possible for these two implementations
>> to
>> > be
>> > > > unified? This would serve to promote feature parity of server-side
>> and
>> > > > client-side assignors, and would also facilitate operational
>> > flexibility in
>> > > > certain situations. For example, if a server-side assignor has some
>> > poor
>> > > > behavior and needs a patch, deploying the patched assignor to the
>> > client
>> > > > and switching one consumer group to a client-side assignor may be
>> > faster
>> > > > and less risky than patching all of the brokers. With the currently
>> > > > proposed distinct APIs, a non-trivial reimplementation would have
>> to be
>> > > > assembled, and if the two APIs have diverged significantly, then it
>> is
>> > > > possible that a reimplementation would not be possible.
>> > > >
>> > > > --
>> > > > Greg Harris
>> > > > gharris1727@gmail.com
>> > > > github.com/gharris1727
>> > > >
>> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com>
>> > wrote:
>> > > >
>> > > > > Hi Guozhang/David,
>> > > > >
>> > > > > I created a confluence page to discuss how Connect would need to
>> > change
>> > > > > based on the new rebalance protocol. Here's the page:
>> > > > >
>> > > > >
>> >
>> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>> > > > >
>> > > > > It's also pretty longish and I have tried to keep a format
>> similar to
>> > > > > KIP-848. Let me know what you think. Also, do you think this
>> should
>> > be
>> > > > > moved to a separate discussion thread or is this one fine?
>> > > > >
>> > > > > Thanks!
>> > > > > Sagar.
>> > > > >
>> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com>
>> > wrote:
>> > > > >
>> > > > > > Hello Guozhang,
>> > > > > >
>> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I would do
>> > the
>> > > > > > analysis and come up with such a document.
>> > > > > >
>> > > > > > Thanks!
>> > > > > > Sagar.
>> > > > > >
>> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
>> wangguoz@gmail.com>
>> > > > > wrote:
>> > > > > >
>> > > > > >> Hello Sagar,
>> > > > > >>
>> > > > > >> It would be great if you could come back with some analysis on
>> > how to
>> > > > > >> implement the Connect side integration with the new protocol;
>> so
>> > far
>> > > > > >> besides leveraging on the new "protocol type" we did not yet
>> think
>> > > > > through
>> > > > > >> the Connect side implementations. For Streams here's a draft of
>> > > > > >> integration
>> > > > > >> plan:
>> > > > > >>
>> > > > > >>
>> > > > >
>> >
>> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
>> > > > > >> just FYI for your analysis on Connect.
>> > > > > >>
>> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
>> sagarmeansocean@gmail.com
>> > >
>> > > > > wrote:
>> > > > > >>
>> > > > > >> > Hi David,
>> > > > > >> >
>> > > > > >> > Thank you for your response. The reason I thought connect can
>> > also fit
>> > > > > >> into
>> > > > > >> > this new scheme is that even today the connect uses a
>> > > > > WorkerCoordinator
>> > > > > >> > extending from AbstractCoordinator to empower rebalances of
>> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
>> protocolType()
>> > to
>> > > > > >> connect
>> > > > > >> > and uses the metadata() method by plumbing into
>> > > > > >> JoinGroupRequestProtocol.
>> > > > > >> >
>> > > > > >> > I think the changes to support connect would be similar at a
>> > high
>> > > > > level
>> > > > > >> to
>> > > > > >> > the changes in streams mainly because of the Client side
>> > assignors
>> > > > > being
>> > > > > >> > used in both. At an implementation level, we might need to
>> make
>> > a lot
>> > > > > of
>> > > > > >> > changes to get onto this new assignment protocol like
>> enhancing
>> > the
>> > > > > >> > JoinGroup request/response and SyncGroup and using
>> > > > > >> ConsumerGroupHeartbeat
>> > > > > >> > API etc again on similar lines to streams (or there might be
>> > > > > >> deviations). I
>> > > > > >> > would try to perform a detailed analysis of the same  and we
>> > can have
>> > > > > a
>> > > > > >> > separate discussion thread for that as that would derail this
>> > > > > discussion
>> > > > > >> > thread. Let me know if that sounds good to you.
>> > > > > >> >
>> > > > > >> > Thanks!
>> > > > > >> > Sagar.
>> > > > > >> >
>> > > > > >> >
>> > > > > >> >
>> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
>> > > > > <djacot@confluent.io.invalid
>> > > > > >> >
>> > > > > >> > wrote:
>> > > > > >> >
>> > > > > >> > > Hi Sagar,
>> > > > > >> > >
>> > > > > >> > > Thanks for your comments.
>> > > > > >> > >
>> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can
>> > mention it.
>> > > > > >> > >
>> > > > > >> > > 2) The idea is to transition C from his current assignment
>> to
>> > his
>> > > > > >> > > target assignment when he can move to epoch 3. When that
>> > happens,
>> > > > > the
>> > > > > >> > > member assignment is updated and persisted with all its
>> > assigned
>> > > > > >> > > partitions even if they are not all revoked yet. In other
>> > words, the
>> > > > > >> > > member assignment becomes the target assignment. This is
>> > basically
>> > > > > an
>> > > > > >> > > optimization to avoid having to write all the changes to
>> the
>> > log.
>> > > > > The
>> > > > > >> > > examples are based on the persisted state so I understand
>> the
>> > > > > >> > > confusion. Let me see if I can improve this in the
>> > description.
>> > > > > >> > >
>> > > > > >> > > 3) Regarding Connect, it could reuse the protocol with a
>> > client side
>> > > > > >> > > assignor if it fits in the protocol. The assignment is
>> about
>> > > > > >> > > topicid-partitions + metadata, could Connect fit into this?
>> > > > > >> > >
>> > > > > >> > > Best,
>> > > > > >> > > David
>> > > > > >> > >
>> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
>> > sagarmeansocean@gmail.com>
>> > > > > >> wrote:
>> > > > > >> > > >
>> > > > > >> > > > Hi David,
>> > > > > >> > > >
>> > > > > >> > > > Thanks for the KIP. I just had minor observations:
>> > > > > >> > > >
>> > > > > >> > > > 1) In the Assignment Error section in Client Side mode
>> > Assignment
>> > > > > >> > > process,
>> > > > > >> > > > you mentioned => `In this case, the client side assignor
>> can
>> > > > > return
>> > > > > >> an
>> > > > > >> > > > error to the group coordinator`. In this case are you
>> > referring to
>> > > > > >> the
>> > > > > >> > > > Assignor returning an AssignmentError that's listed down
>> > towards
>> > > > > the
>> > > > > >> > end?
>> > > > > >> > > > If yes, do you think it would make sense to mention this
>> > > > > explicitly
>> > > > > >> > here?
>> > > > > >> > > >
>> > > > > >> > > > 2) In the Case Studies section, I have a slight
>> confusion,
>> > not
>> > > > > sure
>> > > > > >> if
>> > > > > >> > > > others have the same. Consider this step:
>> > > > > >> > > >
>> > > > > >> > > > When B heartbeats, the group coordinator transitions him
>> to
>> > epoch
>> > > > > 3
>> > > > > >> > > because
>> > > > > >> > > > B has no partitions to revoke. It persists the change and
>> > reply.
>> > > > > >> > > >
>> > > > > >> > > >    - Group (epoch=3)
>> > > > > >> > > >       - A
>> > > > > >> > > >       - B
>> > > > > >> > > >       - C
>> > > > > >> > > >    - Target Assignment (epoch=3)
>> > > > > >> > > >       - A - partitions=[foo-0]
>> > > > > >> > > >       - B - partitions=[foo-2]
>> > > > > >> > > >       - C - partitions=[foo-1]
>> > > > > >> > > >    - Member Assignment
>> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
>> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
>> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
>> > > > > >> > > >
>> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot
>> get
>> > foo-1
>> > > > > >> yet.
>> > > > > >> > > >
>> > > > > >> > > > Here,it's mentioned that member C can't get the foo-1
>> > partition
>> > > > > yet,
>> > > > > >> > but
>> > > > > >> > > > based on the description above, it seems it already has
>> it.
>> > Do you
>> > > > > >> > think
>> > > > > >> > > it
>> > > > > >> > > > would be better to remove it and populate it only when it
>> > actually
>> > > > > >> gets
>> > > > > >> > > it?
>> > > > > >> > > > I see this in a lot of other places, so have I
>> understood it
>> > > > > >> > incorrectly
>> > > > > >> > > ?
>> > > > > >> > > >
>> > > > > >> > > >
>> > > > > >> > > > Regarding connect , it might be out of scope of this
>> > discussion,
>> > > > > but
>> > > > > >> > from
>> > > > > >> > > > what I understood it would probably be running in client
>> > side
>> > > > > >> assignor
>> > > > > >> > > mode
>> > > > > >> > > > even on the new rebalance protocol as it has its own
>> Custom
>> > > > > >> > > Assignors(Eager
>> > > > > >> > > > and IncrementalCooperative).
>> > > > > >> > > >
>> > > > > >> > > > Thanks!
>> > > > > >> > > >
>> > > > > >> > > > Sagar.
>> > > > > >> > > >
>> > > > > >> > > >
>> > > > > >> > > >
>> > > > > >> > > >
>> > > > > >> > > >
>> > > > > >> > > >
>> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
>> > > > > >> > <djacot@confluent.io.invalid
>> > > > > >> > > >
>> > > > > >> > > > wrote:
>> > > > > >> > > >
>> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
>> > specialized API
>> > > > > >> > > > > instead of relying on one generic API. For Connect, we
>> > can apply
>> > > > > >> the
>> > > > > >> > > > > exact same pattern and reuse/share the core
>> > implementation on
>> > > > > the
>> > > > > >> > > > > server side. For the schema registry, I think that we
>> > should
>> > > > > >> consider
>> > > > > >> > > > > having a tailored API to do simple membership/leader
>> > election.
>> > > > > >> > > > >
>> > > > > >> > > > > Best,
>> > > > > >> > > > > David
>> > > > > >> > > > >
>> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
>> > ismael@juma.me.uk
>> > > > > >
>> > > > > >> > > wrote:
>> > > > > >> > > > > >
>> > > > > >> > > > > > Three quick comments:
>> > > > > >> > > > > >
>> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
>> > > > > >> com.google.re2j.Pattern, we
>> > > > > >> > > > > should
>> > > > > >> > > > > > document the differences in more detail before
>> deciding
>> > one
>> > > > > way
>> > > > > >> or
>> > > > > >> > > > > another.
>> > > > > >> > > > > > That said, if people pass java.util.regex.Pattern,
>> they
>> > expect
>> > > > > >> > their
>> > > > > >> > > > > > semantics to be honored. If we are doing something
>> > different,
>> > > > > >> then
>> > > > > >> > we
>> > > > > >> > > > > > should consider adding an overload with our own
>> Pattern
>> > class
>> > > > > (I
>> > > > > >> > > don't
>> > > > > >> > > > > > think we'd want to expose re2j's at this point).
>> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol should
>> > > > > integrate
>> > > > > >> > fully
>> > > > > >> > > > > with
>> > > > > >> > > > > > it and should handle the topic recreation case
>> > correctly.
>> > > > > That's
>> > > > > >> > the
>> > > > > >> > > main
>> > > > > >> > > > > > part we need to handle. I agree with David that we'd
>> > want to
>> > > > > add
>> > > > > >> > > topic
>> > > > > >> > > > > ids
>> > > > > >> > > > > > to the relevant protocols that don't have it yet and
>> > that we
>> > > > > can
>> > > > > >> > > probably
>> > > > > >> > > > > > focus on the internals versus adding new APIs to the
>> > Java
>> > > > > >> Consumer
>> > > > > >> > > > > (unless
>> > > > > >> > > > > > we find that adding new APIs is required for
>> reasonable
>> > > > > >> semantics).
>> > > > > >> > > > > > 3. I am still not sure about the coordinator storing
>> the
>> > > > > >> configs.
>> > > > > >> > > It's
>> > > > > >> > > > > > powerful for configs to be centralized in the
>> metadata
>> > log for
>> > > > > >> > > various
>> > > > > >> > > > > > reasons (auditability, visibility, consistency,
>> etc.).
>> > > > > >> Similarly, I
>> > > > > >> > > am
>> > > > > >> > > > > not
>> > > > > >> > > > > > sure about automatically deleting configs in a way
>> that
>> > they
>> > > > > >> cannot
>> > > > > >> > > be
>> > > > > >> > > > > > recovered. A good property for modern systems is to
>> > minimize
>> > > > > the
>> > > > > >> > > number
>> > > > > >> > > > > of
>> > > > > >> > > > > > unrecoverable data loss scenarios.
>> > > > > >> > > > > >
>> > > > > >> > > > > > Ismael
>> > > > > >> > > > > >
>> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
>> > > > > >> > > <djacot@confluent.io.invalid
>> > > > > >> > > > > >
>> > > > > >> > > > > > wrote:
>> > > > > >> > > > > >
>> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > > 1) the migration path, especially the last step
>> when
>> > > > > clients
>> > > > > >> > > flip the
>> > > > > >> > > > > > > flag
>> > > > > >> > > > > > > > to enable the new protocol, in which we would
>> have a
>> > > > > window
>> > > > > >> > where
>> > > > > >> > > > > both
>> > > > > >> > > > > > > new
>> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
>> used
>> > by
>> > > > > >> members
>> > > > > >> > of
>> > > > > >> > > the
>> > > > > >> > > > > same
>> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
>> > behavior
>> > > > > >> while
>> > > > > >> > > > > using the
>> > > > > >> > > > > > > > new protocol is something we need to present
>> about.
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > Noted. I just published a new version of KIP which
>> > includes
>> > > > > >> more
>> > > > > >> > > > > > > details about this. See the "Supporting Online
>> > Consumer
>> > > > > Group
>> > > > > >> > > Upgrade"
>> > > > > >> > > > > > > and the "Compatibility, Deprecation, and Migration
>> > Plan". I
>> > > > > >> think
>> > > > > >> > > that
>> > > > > >> > > > > > > I have to think through a few cases now but the
>> > overall idea
>> > > > > >> and
>> > > > > >> > > > > > > mechanism should be understandable.
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
>> > topic ids
>> > > > > >> are
>> > > > > >> > > only
>> > > > > >> > > > > used
>> > > > > >> > > > > > > as
>> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
>> > exposed
>> > > > > via
>> > > > > >> any
>> > > > > >> > > > > public
>> > > > > >> > > > > > > APIs
>> > > > > >> > > > > > > > to consumers yet. I think the question is, first
>> > should we
>> > > > > >> let
>> > > > > >> > > the
>> > > > > >> > > > > > > consumer
>> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
>> > itself
>> > > > > to
>> > > > > >> > fully
>> > > > > >> > > > > > > leverage
>> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
>> RPCs
>> > with
>> > > > > the
>> > > > > >> > > topic
>> > > > > >> > > > > ids;
>> > > > > >> > > > > > > and
>> > > > > >> > > > > > > > secondly, should we ever consider exposing the
>> > topic ids
>> > > > > in
>> > > > > >> the
>> > > > > >> > > > > consumer
>> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
>> well
>> > as in
>> > > > > >> the
>> > > > > >> > > > > rebalance
>> > > > > >> > > > > > > > listener for cases like topic
>> > deletion-and-recreation).
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > a) Assuming that we would include converting all
>> the
>> > offsets
>> > > > > >> > > related
>> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer
>> > would be
>> > > > > >> able
>> > > > > >> > to
>> > > > > >> > > > > > > fully operate with topic ids. That being said, it
>> > still has
>> > > > > to
>> > > > > >> > > provide
>> > > > > >> > > > > > > the topics names in various APIs so having a
>> mapping
>> > in the
>> > > > > >> > > consumer
>> > > > > >> > > > > > > seems inevitable to me.
>> > > > > >> > > > > > > b) I don't have a strong opinion on this. Here I
>> > wonder if
>> > > > > >> this
>> > > > > >> > > goes
>> > > > > >> > > > > > > beyond the scope of this KIP. I would rather focus
>> on
>> > the
>> > > > > >> > internals
>> > > > > >> > > > > > > here and we can consider this separately if we see
>> > value in
>> > > > > >> doing
>> > > > > >> > > it.
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > Coming back to Ismael's point about using topic ids
>> > in the
>> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there
>> is
>> > one
>> > > > > >> > advantage
>> > > > > >> > > in
>> > > > > >> > > > > > > favour of it. The consumer will have the
>> opportunity
>> > to
>> > > > > >> validate
>> > > > > >> > > that
>> > > > > >> > > > > > > the topics exists before passing them into the
>> group
>> > > > > rebalance
>> > > > > >> > > > > > > protocol. Obviously, the coordinator will also
>> notice
>> > it but
>> > > > > >> it
>> > > > > >> > > does
>> > > > > >> > > > > > > not really have a way to reject an invalid topic in
>> > the
>> > > > > >> response.
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > > I'm agreeing with David on all other minor
>> questions
>> > > > > except
>> > > > > >> for
>> > > > > >> > > the
>> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
>> > it's not
>> > > > > >> > > necessary
>> > > > > >> > > > > to
>> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
>> > instead we
>> > > > > >> still
>> > > > > >> > > use
>> > > > > >> > > > > > > Pattern
>> > > > > >> > > > > > > > while just documenting that our subscription may
>> be
>> > > > > >> rejected by
>> > > > > >> > > the
>> > > > > >> > > > > > > server.
>> > > > > >> > > > > > > > Since the incompatible case is a very rare
>> scenario
>> > I felt
>> > > > > >> > using
>> > > > > >> > > an
>> > > > > >> > > > > > > > overloaded `String` based subscription may be
>> more
>> > > > > >> vulnerable
>> > > > > >> > to
>> > > > > >> > > > > various
>> > > > > >> > > > > > > > invalid regexes.
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > That could work. I have to look at the differences
>> > between
>> > > > > the
>> > > > > >> > two
>> > > > > >> > > > > > > engines to better understand the potential issues.
>> My
>> > > > > >> > > understanding is
>> > > > > >> > > > > > > that would work for all the basic regular
>> > expressions. The
>> > > > > >> > > differences
>> > > > > >> > > > > > > between the two are mainly about the various
>> character
>> > > > > >> classes. I
>> > > > > >> > > > > > > wonder what other people think about this.
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > Best,
>> > > > > >> > > > > > > David
>> > > > > >> > > > > > >
>> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
>> > > > > >> > wangguoz@gmail.com
>> > > > > >> > > >
>> > > > > >> > > > > wrote:
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > Thanks David! I think on the high level there are
>> > two meta
>> > > > > >> > > points we
>> > > > > >> > > > > need
>> > > > > >> > > > > > > > to concretize a bit more:
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > 1) the migration path, especially the last step
>> when
>> > > > > clients
>> > > > > >> > > flip the
>> > > > > >> > > > > > > flag
>> > > > > >> > > > > > > > to enable the new protocol, in which we would
>> have a
>> > > > > window
>> > > > > >> > where
>> > > > > >> > > > > both
>> > > > > >> > > > > > > new
>> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are
>> used
>> > by
>> > > > > >> members
>> > > > > >> > of
>> > > > > >> > > the
>> > > > > >> > > > > same
>> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
>> > behavior
>> > > > > >> while
>> > > > > >> > > > > using the
>> > > > > >> > > > > > > > new protocol is something we need to present
>> about.
>> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
>> > topic ids
>> > > > > >> are
>> > > > > >> > > only
>> > > > > >> > > > > used
>> > > > > >> > > > > > > as
>> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
>> > exposed
>> > > > > via
>> > > > > >> any
>> > > > > >> > > > > public
>> > > > > >> > > > > > > APIs
>> > > > > >> > > > > > > > to consumers yet. I think the question is, first
>> > should we
>> > > > > >> let
>> > > > > >> > > the
>> > > > > >> > > > > > > consumer
>> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
>> > itself
>> > > > > to
>> > > > > >> > fully
>> > > > > >> > > > > > > leverage
>> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
>> RPCs
>> > with
>> > > > > the
>> > > > > >> > > topic
>> > > > > >> > > > > ids;
>> > > > > >> > > > > > > and
>> > > > > >> > > > > > > > secondly, should we ever consider exposing the
>> > topic ids
>> > > > > in
>> > > > > >> the
>> > > > > >> > > > > consumer
>> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
>> well
>> > as in
>> > > > > >> the
>> > > > > >> > > > > rebalance
>> > > > > >> > > > > > > > listener for cases like topic
>> > deletion-and-recreation).
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > I'm agreeing with David on all other minor
>> questions
>> > > > > except
>> > > > > >> for
>> > > > > >> > > the
>> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
>> > it's not
>> > > > > >> > > necessary
>> > > > > >> > > > > to
>> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
>> > instead we
>> > > > > >> still
>> > > > > >> > > use
>> > > > > >> > > > > > > Pattern
>> > > > > >> > > > > > > > while just documenting that our subscription may
>> be
>> > > > > >> rejected by
>> > > > > >> > > the
>> > > > > >> > > > > > > server.
>> > > > > >> > > > > > > > Since the incompatible case is a very rare
>> scenario
>> > I felt
>> > > > > >> > using
>> > > > > >> > > an
>> > > > > >> > > > > > > > overloaded `String` based subscription may be
>> more
>> > > > > >> vulnerable
>> > > > > >> > to
>> > > > > >> > > > > various
>> > > > > >> > > > > > > > invalid regexes.
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > Guozhang
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
>> > > > > >> > > > > <djacot@confluent.io.invalid
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > wrote:
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > > Hi Ismael,
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer your
>> > questions
>> > > > > >> > inline.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 1. I think it's premature to talk about
>> target
>> > > > > versions
>> > > > > >> for
>> > > > > >> > > > > > > deprecation
>> > > > > >> > > > > > > > > and
>> > > > > >> > > > > > > > > > removal of the existing group protocol.
>> Unlike
>> > KRaft,
>> > > > > >> this
>> > > > > >> > > > > affects a
>> > > > > >> > > > > > > core
>> > > > > >> > > > > > > > > > client protocol and hence deprecation/removal
>> > will be
>> > > > > >> > heavily
>> > > > > >> > > > > > > dependent
>> > > > > >> > > > > > > > > on
>> > > > > >> > > > > > > > > > how quickly applications migrate to the new
>> > protocol.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > That makes sense. I will remove it.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 2. The KIP says we intend to release this in
>> > 4.x, but
>> > > > > it
>> > > > > >> > > wasn't
>> > > > > >> > > > > made
>> > > > > >> > > > > > > > > clear
>> > > > > >> > > > > > > > > > why. If we added that as a way to estimate
>> when
>> > we'd
>> > > > > >> > > deprecate
>> > > > > >> > > > > and
>> > > > > >> > > > > > > remove
>> > > > > >> > > > > > > > > > the group protocol, I also suggest removing
>> > this part.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Let me explain my reasoning. As explained, I
>> plan
>> > to
>> > > > > >> rewrite
>> > > > > >> > > the
>> > > > > >> > > > > group
>> > > > > >> > > > > > > > > coordinator in Java while we implement the new
>> > protocol.
>> > > > > >> This
>> > > > > >> > > means
>> > > > > >> > > > > > > > > that the internals will be slightly different
>> > (e.g.
>> > > > > >> threading
>> > > > > >> > > > > model).
>> > > > > >> > > > > > > > > Therefore, I wanted to tighten the switch from
>> > the old
>> > > > > >> group
>> > > > > >> > > > > > > > > coordinator to the new group coordinator to a
>> > major
>> > > > > >> release.
>> > > > > >> > > The
>> > > > > >> > > > > > > > > alternative would be to use a flag to do the
>> > switch
>> > > > > >> instead
>> > > > > >> > of
>> > > > > >> > > > > relying
>> > > > > >> > > > > > > > > on the software upgrade.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 3. We need to flesh out the details of the
>> > migration
>> > > > > >> story.
>> > > > > >> > > It
>> > > > > >> > > > > sounds
>> > > > > >> > > > > > > > > like
>> > > > > >> > > > > > > > > > we're saying we will support online
>> migrations.
>> > Is
>> > > > > that
>> > > > > >> > > correct?
>> > > > > >> > > > > We
>> > > > > >> > > > > > > > > should
>> > > > > >> > > > > > > > > > explain this in detail. It could also be done
>> > as a
>> > > > > >> separate
>> > > > > >> > > KIP,
>> > > > > >> > > > > if
>> > > > > >> > > > > > > it's
>> > > > > >> > > > > > > > > > easier.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Yes, we will support online migrations for the
>> > group.
>> > > > > That
>> > > > > >> > > means
>> > > > > >> > > > > that
>> > > > > >> > > > > > > > > a group using the old protocol will be able to
>> > switch to
>> > > > > >> the
>> > > > > >> > > new
>> > > > > >> > > > > > > > > protocol.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Let me briefly explain how that will work
>> though.
>> > It is
>> > > > > >> > > basically a
>> > > > > >> > > > > > > > > four step process:
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a
>> > software
>> > > > > >> > > supporting
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > new group coordinator. Both the old and the new
>> > > > > >> coordinator
>> > > > > >> > > will
>> > > > > >> > > > > > > > > support the old protocol and rely on the same
>> > persisted
>> > > > > >> > > metadata so
>> > > > > >> > > > > > > > > they can work together. This point is an
>> offline
>> > > > > >> migration.
>> > > > > >> > We
>> > > > > >> > > > > cannot
>> > > > > >> > > > > > > > > do this one live because it would require
>> > shutting down
>> > > > > >> the
>> > > > > >> > > current
>> > > > > >> > > > > > > > > coordinator and starting up the new one and
>> that
>> > would
>> > > > > >> cause
>> > > > > >> > > > > > > > > unavailabilities.
>> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be
>> > upgraded
>> > > > > to
>> > > > > >> X
>> > > > > >> > in
>> > > > > >> > > > > order
>> > > > > >> > > > > > > > > to enable the new protocol. This cannot be done
>> > before
>> > > > > 1)
>> > > > > >> is
>> > > > > >> > > > > > > > > terminated because the old coordinator doesn't
>> > support
>> > > > > the
>> > > > > >> > new
>> > > > > >> > > > > > > > > protocol.
>> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a version
>> > > > > supporting
>> > > > > >> the
>> > > > > >> > > > > online
>> > > > > >> > > > > > > > > migration (must have KIP-792). If the consumer
>> is
>> > > > > already
>> > > > > >> > > there.
>> > > > > >> > > > > > > > > Nothing must be done at this point.
>> > > > > >> > > > > > > > > 4. The consumers must be rolled with the
>> feature
>> > flag
>> > > > > >> turned
>> > > > > >> > > on.
>> > > > > >> > > > > The
>> > > > > >> > > > > > > > > consumer group is automatically converted when
>> > the first
>> > > > > >> > > consumer
>> > > > > >> > > > > > > > > using the new protocol joins the group. While
>> the
>> > > > > members
>> > > > > >> > > using the
>> > > > > >> > > > > > > > > old protocol are being upgraded, the old
>> protocol
>> > is
>> > > > > >> proxied
>> > > > > >> > > into
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > new one.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
>> > > > > >> subscriptions
>> > > > > >> > > to
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > server,
>> > > > > >> > > > > > > > > > but it seems like there could be some tricky
>> > > > > >> compatibility
>> > > > > >> > > > > issues.
>> > > > > >> > > > > > > Will
>> > > > > >> > > > > > > > > we
>> > > > > >> > > > > > > > > > have a mechanism for users to detect that
>> they
>> > need to
>> > > > > >> > update
>> > > > > >> > > > > their
>> > > > > >> > > > > > > regex
>> > > > > >> > > > > > > > > > before switching to the new protocol?
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > I think that I am a bit more optimistic than
>> you
>> > on this
>> > > > > >> > > point. I
>> > > > > >> > > > > > > > > believe that the majority of the cases are
>> simple
>> > > > > regexes
>> > > > > >> > which
>> > > > > >> > > > > should
>> > > > > >> > > > > > > > > work with the new engine. The coordinator will
>> > verify
>> > > > > the
>> > > > > >> > regex
>> > > > > >> > > > > anyway
>> > > > > >> > > > > > > > > and reject the consumer if the regex is not
>> valid.
>> > > > > Coming
>> > > > > >> > back
>> > > > > >> > > to
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > migration path, in the worst case, the first
>> > upgraded
>> > > > > >> > consumer
>> > > > > >> > > > > joining
>> > > > > >> > > > > > > > > the group will be rejected. This should be used
>> > as the
>> > > > > >> last
>> > > > > >> > > > > defence, I
>> > > > > >> > > > > > > > > would say.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > One way for customers to validate their regex
>> > before
>> > > > > >> > upgrading
>> > > > > >> > > > > their
>> > > > > >> > > > > > > > > prod would be to test them with another group.
>> For
>> > > > > >> instance,
>> > > > > >> > > that
>> > > > > >> > > > > > > > > could be done in a pre-prod environment.
>> Another
>> > way
>> > > > > >> would be
>> > > > > >> > > to
>> > > > > >> > > > > > > > > extend the consumer-group tool to provide a
>> regex
>> > > > > >> validation
>> > > > > >> > > > > > > > > mechanism. Would this be enough in your
>> opinion?
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 5. Related to the last question, will the
>> Java
>> > client
>> > > > > >> allow
>> > > > > >> > > the
>> > > > > >> > > > > > > users to
>> > > > > >> > > > > > > > > > stick with the current regex engine for
>> > compatibility
>> > > > > >> > > reasons?
>> > > > > >> > > > > For
>> > > > > >> > > > > > > > > example,
>> > > > > >> > > > > > > > > > it may be handy to keep using client based
>> > regex at
>> > > > > >> first
>> > > > > >> > to
>> > > > > >> > > keep
>> > > > > >> > > > > > > > > > migrations simple and then migrate to server
>> > based
>> > > > > >> regexes
>> > > > > >> > > as a
>> > > > > >> > > > > > > second
>> > > > > >> > > > > > > > > step.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > I understand your point but I am concerned that
>> > this
>> > > > > would
>> > > > > >> > > allow
>> > > > > >> > > > > users
>> > > > > >> > > > > > > > > to actually stay in this mode. That would go
>> > against our
>> > > > > >> goal
>> > > > > >> > > of
>> > > > > >> > > > > > > > > simplifying the client because we would have to
>> > continue
>> > > > > >> > > monitoring
>> > > > > >> > > > > > > > > the metadata on the client side. I would rather
>> > not do
>> > > > > >> this.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 6. When we say that the group coordinator
>> will
>> > be
>> > > > > >> > > responsible for
>> > > > > >> > > > > > > storing
>> > > > > >> > > > > > > > > > the configurations and that the
>> configurations
>> > will be
>> > > > > >> > > deleted
>> > > > > >> > > > > when
>> > > > > >> > > > > > > the
>> > > > > >> > > > > > > > > > group is deleted. Will a transition to DEAD
>> > trigger
>> > > > > >> > deletion
>> > > > > >> > > of
>> > > > > >> > > > > > > > > > configurations?
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > That's right. The configurations will be
>> deleted
>> > when
>> > > > > the
>> > > > > >> > > group is
>> > > > > >> > > > > > > > > deleted. They go together.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 7. Will the choice to store the configs in
>> the
>> > group
>> > > > > >> > > coordinator
>> > > > > >> > > > > > > make it
>> > > > > >> > > > > > > > > > harder to list all cluster configs and their
>> > values?
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > I don't think so. The group configurations are
>> > overrides
>> > > > > >> of
>> > > > > >> > > cluster
>> > > > > >> > > > > > > > > configs. If you want to know all the overrides
>> > though,
>> > > > > you
>> > > > > >> > > would
>> > > > > >> > > > > have
>> > > > > >> > > > > > > > > to ask all the group coordinators. You cannot
>> > rely on
>> > > > > the
>> > > > > >> > > metadata
>> > > > > >> > > > > log
>> > > > > >> > > > > > > > > for instance.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 8. How would someone configure a group before
>> > starting
>> > > > > >> the
>> > > > > >> > > > > consumers?
>> > > > > >> > > > > > > > > Have
>> > > > > >> > > > > > > > > > we considered allowing the explicit creation
>> of
>> > > > > groups?
>> > > > > >> > > > > > > Alternatively,
>> > > > > >> > > > > > > > > the
>> > > > > >> > > > > > > > > > configs could be decoupled from the group
>> > lifecycle.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Yes. The group will be automatically created in
>> > this
>> > > > > case.
>> > > > > >> > > However,
>> > > > > >> > > > > > > > > the configs will be lost after the retention
>> > period of
>> > > > > the
>> > > > > >> > > group
>> > > > > >> > > > > > > > > passes.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the
>> > Java
>> > > > > >> client
>> > > > > >> > > still
>> > > > > >> > > > > take
>> > > > > >> > > > > > > a
>> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to
>> > introduce
>> > > > > an
>> > > > > >> > > overload?
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > That's a very group question. I forgot about
>> that
>> > one.
>> > > > > As
>> > > > > >> the
>> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
>> compatible
>> > with
>> > > > > the
>> > > > > >> > > engine
>> > > > > >> > > > > that
>> > > > > >> > > > > > > > > we plan to use, it might be better to deprecate
>> > it and
>> > > > > >> use an
>> > > > > >> > > > > overload
>> > > > > >> > > > > > > > > which takes a string. We would rely on the
>> server
>> > side
>> > > > > >> > > validation.
>> > > > > >> > > > > > > > > During the migration, I think that we could
>> still
>> > try to
>> > > > > >> > > toString
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > regex and use it. That should work, I think, in
>> > the
>> > > > > >> majority
>> > > > > >> > > of the
>> > > > > >> > > > > > > > > cases.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 10. I agree with Justine that we should be
>> > clearer
>> > > > > about
>> > > > > >> > the
>> > > > > >> > > > > reason
>> > > > > >> > > > > > > to
>> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
>> feature
>> > flag.
>> > > > > >> Maybe
>> > > > > >> > > we
>> > > > > >> > > > > mean
>> > > > > >> > > > > > > that
>> > > > > >> > > > > > > > > we
>> > > > > >> > > > > > > > > > can switch the default for the feature flag
>> to
>> > true
>> > > > > >> based
>> > > > > >> > on
>> > > > > >> > > the
>> > > > > >> > > > > > > > > > metadata.version once we want to make it the
>> > default.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > My plan was to use that feature flag mainly
>> > during the
>> > > > > >> > > development
>> > > > > >> > > > > > > > > phase. I should not have mentioned it, I think,
>> > because
>> > > > > we
>> > > > > >> > > could
>> > > > > >> > > > > use
>> > > > > >> > > > > > > > > an internal config for it.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention
>> the
>> > > > > required
>> > > > > >> > > ACLs, it
>> > > > > >> > > > > > > would
>> > > > > >> > > > > > > > > be
>> > > > > >> > > > > > > > > > good to add that for consistency.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Noted.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 12. It is a bit odd that
>> ConsumerGroupHeartbeat
>> > > > > requires
>> > > > > >> > > "Read
>> > > > > >> > > > > Group"
>> > > > > >> > > > > > > > > even
>> > > > > >> > > > > > > > > > though it seems to do more than reading.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > I agree. This is how the current protocol works
>> > though.
>> > > > > We
>> > > > > >> > only
>> > > > > >> > > > > > > > > require "Read Group" to join a group. We could
>> > consider
>> > > > > >> > > changing
>> > > > > >> > > > > this
>> > > > > >> > > > > > > > > but I am not sure that it is worth it.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 13. How is topic recreation handled by the
>> > consumer
>> > > > > with
>> > > > > >> > the
>> > > > > >> > > new
>> > > > > >> > > > > > > group
>> > > > > >> > > > > > > > > > protocol? It would be good to have a section
>> on
>> > this.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Noted. From a protocol perspective, the new
>> topic
>> > will
>> > > > > >> have a
>> > > > > >> > > new
>> > > > > >> > > > > > > > > topic id so it will treat it like a topic with
>> a
>> > > > > different
>> > > > > >> > > name.
>> > > > > >> > > > > The
>> > > > > >> > > > > > > > > only issue is that the fetch/commit offsets
>> APIs
>> > do not
>> > > > > >> > support
>> > > > > >> > > > > topic
>> > > > > >> > > > > > > > > IDs so the consumer would reuse the offsets
>> based
>> > on the
>> > > > > >> > same.
>> > > > > >> > > I
>> > > > > >> > > > > think
>> > > > > >> > > > > > > > > that we should update those APIs as well in
>> order
>> > to be
>> > > > > >> > > consistent
>> > > > > >> > > > > end
>> > > > > >> > > > > > > > > to end. That would strengthen the semantics of
>> the
>> > > > > >> consumer.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the new
>> > coordinator
>> > > > > >> in
>> > > > > >> > > Java.
>> > > > > >> > > > > Even
>> > > > > >> > > > > > > > > though
>> > > > > >> > > > > > > > > > this is an implementation detail, do we plan
>> to
>> > have a
>> > > > > >> new
>> > > > > >> > > gradle
>> > > > > >> > > > > > > module
>> > > > > >> > > > > > > > > > for it?
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Yes.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when it
>> comes
>> > to how
>> > > > > >> many
>> > > > > >> > > > > members
>> > > > > >> > > > > > > the
>> > > > > >> > > > > > > > > new
>> > > > > >> > > > > > > > > > group protocol can support?
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > We don't have numbers at the moment. The
>> protocol
>> > should
>> > > > > >> > > support
>> > > > > >> > > > > 1000s
>> > > > > >> > > > > > > > > of members per group. We will measure this when
>> > we have
>> > > > > a
>> > > > > >> > first
>> > > > > >> > > > > > > > > implementation. Note that we might have other
>> > > > > bottlenecks
>> > > > > >> > down
>> > > > > >> > > the
>> > > > > >> > > > > > > > > road (e.g. offset commits).
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds
>> > instead
>> > > > > of
>> > > > > >> > > > > > > > > > SubscribedTopicNames in
>> > ConsumerGroupHeartbeatRequest?
>> > > > > >> Is
>> > > > > >> > the
>> > > > > >> > > > > idea
>> > > > > >> > > > > > > that
>> > > > > >> > > > > > > > > > since we have to resolve the regex on the
>> > server, we
>> > > > > >> can do
>> > > > > >> > > the
>> > > > > >> > > > > same
>> > > > > >> > > > > > > for
>> > > > > >> > > > > > > > > > the topic name? The difference is that
>> sending
>> > the
>> > > > > >> regex is
>> > > > > >> > > more
>> > > > > >> > > > > > > > > efficient
>> > > > > >> > > > > > > > > > whereas sending the topic names is less
>> > efficient.
>> > > > > >> > > Furthermore,
>> > > > > >> > > > > > > delete
>> > > > > >> > > > > > > > > and
>> > > > > >> > > > > > > > > > recreation is easier to handle if we have
>> topic
>> > ids.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > The idea was to consolidate the metadata lookup
>> > on the
>> > > > > >> server
>> > > > > >> > > for
>> > > > > >> > > > > both
>> > > > > >> > > > > > > > > paths but I do agree with your point. As a
>> second
>> > > > > though,
>> > > > > >> > using
>> > > > > >> > > > > topic
>> > > > > >> > > > > > > > > ids may be better here for the delete and
>> > recreation
>> > > > > case.
>> > > > > >> > > Also, I
>> > > > > >> > > > > > > > > suppose that we may allow users to subscribe
>> with
>> > topic
>> > > > > >> ids
>> > > > > >> > in
>> > > > > >> > > the
>> > > > > >> > > > > > > > > future because that is the only way to be
>> really
>> > robust
>> > > > > to
>> > > > > >> > > topic
>> > > > > >> > > > > > > > > re-creation.
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > Best,
>> > > > > >> > > > > > > > > David
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
>> > > > > >> > > djacot@confluent.io>
>> > > > > >> > > > > > > wrote:
>> > > > > >> > > > > > > > > >
>> > > > > >> > > > > > > > > > Hi Justine,
>> > > > > >> > > > > > > > > >
>> > > > > >> > > > > > > > > > Thanks for your comments. Please find my
>> answers
>> > > > > below.
>> > > > > >> > > > > > > > > >
>> > > > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs
>> > with the
>> > > > > >> > > exception
>> > > > > >> > > > > of the
>> > > > > >> > > > > > > > > > topic names based in the
>> > > > > ConsumerGroupHeartbeatRequest.
>> > > > > >> I
>> > > > > >> > am
>> > > > > >> > > not
>> > > > > >> > > > > sure
>> > > > > >> > > > > > > > > > if using topic names is the right call here.
>> I
>> > need to
>> > > > > >> > think
>> > > > > >> > > > > about it
>> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does not
>> > change the
>> > > > > >> > > > > fetch/commit
>> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be
>> > something
>> > > > > >> that
>> > > > > >> > we
>> > > > > >> > > > > should
>> > > > > >> > > > > > > > > > include though as it would give better
>> overall
>> > > > > >> guarantee in
>> > > > > >> > > the
>> > > > > >> > > > > > > > > > producer.
>> > > > > >> > > > > > > > > > - You're right. I think that I should not
>> have
>> > > > > mentioned
>> > > > > >> > this
>> > > > > >> > > > > flag at
>> > > > > >> > > > > > > > > > all. I will remove it. We can use an internal
>> > > > > >> configuration
>> > > > > >> > > while
>> > > > > >> > > > > > > > > > developing the feature.
>> > > > > >> > > > > > > > > > - Both cluster types will be supported. The
>> > change is
>> > > > > >> > > > > orthogonal. The
>> > > > > >> > > > > > > > > > only requirement is that the cluster uses
>> topic
>> > IDs.
>> > > > > >> > > > > > > > > >
>> > > > > >> > > > > > > > > > Best,
>> > > > > >> > > > > > > > > > David
>> > > > > >> > > > > > > > > >
>> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang
>> Wang <
>> > > > > >> > > > > wangguoz@gmail.com>
>> > > > > >> > > > > > > > > wrote:
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Hi Ismael,
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are some
>> replies
>> > > > > inlined
>> > > > > >> > > below:
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma
>> <
>> > > > > >> > > ismael@juma.me.uk>
>> > > > > >> > > > > > > wrote:
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the
>> potential
>> > to be a
>> > > > > >> > great
>> > > > > >> > > > > > > > > improvement. A few
>> > > > > >> > > > > > > > > > > > initial questions/comments:
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk about
>> > target
>> > > > > >> versions
>> > > > > >> > > for
>> > > > > >> > > > > > > > > deprecation and
>> > > > > >> > > > > > > > > > > > removal of the existing group protocol.
>> > Unlike
>> > > > > >> KRaft,
>> > > > > >> > > this
>> > > > > >> > > > > > > affects a
>> > > > > >> > > > > > > > > core
>> > > > > >> > > > > > > > > > > > client protocol and hence
>> > deprecation/removal will
>> > > > > >> be
>> > > > > >> > > heavily
>> > > > > >> > > > > > > > > dependent on
>> > > > > >> > > > > > > > > > > > how quickly applications migrate to the
>> new
>> > > > > >> protocol.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can
>> remove
>> > the
>> > > > > >> proposed
>> > > > > >> > > > > timeline
>> > > > > >> > > > > > > in
>> > > > > >> > > > > > > > > the
>> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration
>> > Plan` and
>> > > > > >> > > instead
>> > > > > >> > > > > just
>> > > > > >> > > > > > > state
>> > > > > >> > > > > > > > > > > that we will decide in the future about
>> when
>> > we
>> > > > > would
>> > > > > >> > > > > deprecate old
>> > > > > >> > > > > > > > > > > protocol and behaviors.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to release this
>> > in 4.x,
>> > > > > >> but
>> > > > > >> > it
>> > > > > >> > > > > wasn't
>> > > > > >> > > > > > > made
>> > > > > >> > > > > > > > > clear
>> > > > > >> > > > > > > > > > > > why. If we added that as a way to
>> estimate
>> > when
>> > > > > we'd
>> > > > > >> > > > > deprecate
>> > > > > >> > > > > > > and
>> > > > > >> > > > > > > > > remove
>> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
>> removing
>> > this
>> > > > > >> part.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > I think that's not specifically related to
>> the
>> > > > > >> > > > > deprecation/removal
>> > > > > >> > > > > > > > > timeline
>> > > > > >> > > > > > > > > > > plan, but it's more for client upgrades.
>> I.e.
>> > the
>> > > > > >> > > broker-side
>> > > > > >> > > > > > > > > > > implementation may be done first, and then
>> the
>> > > > > client
>> > > > > >> > side,
>> > > > > >> > > > > and we
>> > > > > >> > > > > > > > > would
>> > > > > >> > > > > > > > > > > only mark it as "released" by the time
>> clients
>> > > > > >> > > implementations
>> > > > > >> > > > > are
>> > > > > >> > > > > > > > > done. At
>> > > > > >> > > > > > > > > > > that time, to enable the feature the
>> clients
>> > need to
>> > > > > >> > first
>> > > > > >> > > > > swap-in
>> > > > > >> > > > > > > the
>> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and then set
>> > the flag
>> > > > > >> > with a
>> > > > > >> > > > > second
>> > > > > >> > > > > > > > > rolling
>> > > > > >> > > > > > > > > > > bounce, and hence we feel it's better to be
>> > released
>> > > > > >> in a
>> > > > > >> > > major
>> > > > > >> > > > > > > > > version.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 3. We need to flesh out the details of
>> the
>> > > > > migration
>> > > > > >> > > story.
>> > > > > >> > > > > It
>> > > > > >> > > > > > > > > sounds like
>> > > > > >> > > > > > > > > > > > we're saying we will support online
>> > migrations. Is
>> > > > > >> that
>> > > > > >> > > > > correct?
>> > > > > >> > > > > > > We
>> > > > > >> > > > > > > > > should
>> > > > > >> > > > > > > > > > > > explain this in detail. It could also be
>> > done as a
>> > > > > >> > > separate
>> > > > > >> > > > > KIP,
>> > > > > >> > > > > > > if
>> > > > > >> > > > > > > > > it's
>> > > > > >> > > > > > > > > > > > easier.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Yes I think that's the part we can be more
>> > concrete
>> > > > > >> about
>> > > > > >> > > for
>> > > > > >> > > > > sure
>> > > > > >> > > > > > > (and
>> > > > > >> > > > > > > > > > > this is related to your question 2) above).
>> > We will
>> > > > > >> work
>> > > > > >> > on
>> > > > > >> > > > > making
>> > > > > >> > > > > > > it
>> > > > > >> > > > > > > > > more
>> > > > > >> > > > > > > > > > > explicit in parallel as we solicit more
>> > feedback.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing the
>> > pattern
>> > > > > >> > > subscriptions
>> > > > > >> > > > > to
>> > > > > >> > > > > > > the
>> > > > > >> > > > > > > > > server,
>> > > > > >> > > > > > > > > > > > but it seems like there could be some
>> tricky
>> > > > > >> > > compatibility
>> > > > > >> > > > > > > issues.
>> > > > > >> > > > > > > > > Will we
>> > > > > >> > > > > > > > > > > > have a mechanism for users to detect that
>> > they
>> > > > > need
>> > > > > >> to
>> > > > > >> > > update
>> > > > > >> > > > > > > their
>> > > > > >> > > > > > > > > regex
>> > > > > >> > > > > > > > > > > > before switching to the new protocol?
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
>> non-java
>> > client
>> > > > > >> > users
>> > > > > >> > > to
>> > > > > >> > > > > sort
>> > > > > >> > > > > > > of
>> > > > > >> > > > > > > > > > > "dry-run" the client before switching to
>> the
>> > new
>> > > > > >> > protocol.
>> > > > > >> > > I
>> > > > > >> > > > > do not
>> > > > > >> > > > > > > > > have a
>> > > > > >> > > > > > > > > > > specific idea on top of my head though,
>> maybe
>> > others
>> > > > > >> like
>> > > > > >> > > @Matt
>> > > > > >> > > > > > > > > Howlett can
>> > > > > >> > > > > > > > > > > chime-in here?
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 5. Related to the last question, will the
>> > Java
>> > > > > >> client
>> > > > > >> > > allow
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > users to
>> > > > > >> > > > > > > > > > > > stick with the current regex engine for
>> > > > > >> compatibility
>> > > > > >> > > > > reasons?
>> > > > > >> > > > > > > For
>> > > > > >> > > > > > > > > example,
>> > > > > >> > > > > > > > > > > > it may be handy to keep using client
>> based
>> > regex
>> > > > > at
>> > > > > >> > > first to
>> > > > > >> > > > > keep
>> > > > > >> > > > > > > > > > > > migrations simple and then migrate to
>> > server based
>> > > > > >> > > regexes
>> > > > > >> > > > > as a
>> > > > > >> > > > > > > > > second
>> > > > > >> > > > > > > > > > > > step.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Honestly I have not thought about that for
>> > java
>> > > > > >> clients,
>> > > > > >> > > and
>> > > > > >> > > > > we can
>> > > > > >> > > > > > > > > discuss
>> > > > > >> > > > > > > > > > > that. What kind of compatibility issues do
>> > you have
>> > > > > in
>> > > > > >> > > mind?
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 6. When we say that the group coordinator
>> > will be
>> > > > > >> > > > > responsible for
>> > > > > >> > > > > > > > > storing
>> > > > > >> > > > > > > > > > > > the configurations and that the
>> > configurations
>> > > > > will
>> > > > > >> be
>> > > > > >> > > > > deleted
>> > > > > >> > > > > > > when
>> > > > > >> > > > > > > > > the
>> > > > > >> > > > > > > > > > > > group is deleted. Will a transition to
>> DEAD
>> > > > > trigger
>> > > > > >> > > deletion
>> > > > > >> > > > > of
>> > > > > >> > > > > > > > > > > > configurations?
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an ending
>> state
>> > (we
>> > > > > would
>> > > > > >> > only
>> > > > > >> > > > > > > transit to
>> > > > > >> > > > > > > > > that
>> > > > > >> > > > > > > > > > > state when the group is EMPTY and also all
>> of
>> > its
>> > > > > >> > metadata
>> > > > > >> > > are
>> > > > > >> > > > > > > gone),
>> > > > > >> > > > > > > > > once
>> > > > > >> > > > > > > > > > > it's transited to DEAD this group would
>> never
>> > be
>> > > > > >> revived.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 7. Will the choice to store the configs
>> in
>> > the
>> > > > > group
>> > > > > >> > > > > coordinator
>> > > > > >> > > > > > > > > make it
>> > > > > >> > > > > > > > > > > > harder to list all cluster configs and
>> their
>> > > > > values?
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > That's a good question, and our thoughts
>> are
>> > that
>> > > > > the
>> > > > > >> > > so-called
>> > > > > >> > > > > > > "group
>> > > > > >> > > > > > > > > > > configurations" are overrides of the
>> > cluster-level
>> > > > > >> > > > > configurations
>> > > > > >> > > > > > > > > > > customized per group so when an admin list
>> > cluster
>> > > > > >> > configs
>> > > > > >> > > it's
>> > > > > >> > > > > > > okay to
>> > > > > >> > > > > > > > > > > list just the cluster-level "defaults", not
>> > showing
>> > > > > >> any
>> > > > > >> > > > > per-group
>> > > > > >> > > > > > > > > > > customizations.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 8. How would someone configure a group
>> > before
>> > > > > >> starting
>> > > > > >> > > the
>> > > > > >> > > > > > > > > consumers? Have
>> > > > > >> > > > > > > > > > > > we considered allowing the explicit
>> > creation of
>> > > > > >> groups?
>> > > > > >> > > > > > > > > Alternatively, the
>> > > > > >> > > > > > > > > > > > configs could be decoupled from the group
>> > > > > lifecycle.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > The configs can be created before the group
>> > itself
>> > > > > as
>> > > > > >> an
>> > > > > >> > > > > > > independent
>> > > > > >> > > > > > > > > entity
>> > > > > >> > > > > > > > > > > --- of course, this requires the
>> corresponding
>> > > > > >> request to
>> > > > > >> > > be
>> > > > > >> > > > > > > routed to
>> > > > > >> > > > > > > > > the
>> > > > > >> > > > > > > > > > > right coordinator based on the group id ---
>> > the only
>> > > > > >> > thing
>> > > > > >> > > that
>> > > > > >> > > > > > > > > differs is,
>> > > > > >> > > > > > > > > > > when the group itself is gone we also check
>> > if there
>> > > > > >> are
>> > > > > >> > > any
>> > > > > >> > > > > > > > > configuration
>> > > > > >> > > > > > > > > > > entities related to that group and delete
>> as
>> > well.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
>> > asymmetry on
>> > > > > the
>> > > > > >> > > creation
>> > > > > >> > > > > /
>> > > > > >> > > > > > > > > deletion
>> > > > > >> > > > > > > > > > > lifecycles of the config entities, and we
>> > would like
>> > > > > >> to
>> > > > > >> > > hear
>> > > > > >> > > > > > > everyone's
>> > > > > >> > > > > > > > > > > feelings whether we should aim for symmetry
>> > i.e.
>> > > > > >> totally
>> > > > > >> > > > > decouple
>> > > > > >> > > > > > > group
>> > > > > >> > > > > > > > > > > configs and hence not delete them at all
>> when
>> > the
>> > > > > >> group
>> > > > > >> > is
>> > > > > >> > > > > gone,
>> > > > > >> > > > > > > but
>> > > > > >> > > > > > > > > always
>> > > > > >> > > > > > > > > > > require explicit deletion operations by
>> > themselves.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for
>> > the Java
>> > > > > >> > client
>> > > > > >> > > > > still
>> > > > > >> > > > > > > take
>> > > > > >> > > > > > > > > a
>> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have
>> to
>> > > > > >> introduce an
>> > > > > >> > > > > overload?
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > I think we do not need to introduce an
>> > overload, but
>> > > > > >> I'm
>> > > > > >> > > all
>> > > > > >> > > > > ears
>> > > > > >> > > > > > > if
>> > > > > >> > > > > > > > > there
>> > > > > >> > > > > > > > > > > may be some compatibility issues that we
>> may
>> > > > > overlook.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we should
>> be
>> > clearer
>> > > > > >> > about
>> > > > > >> > > the
>> > > > > >> > > > > > > reason
>> > > > > >> > > > > > > > > to
>> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version from the
>> > feature
>> > > > > >> flag.
>> > > > > >> > > Maybe
>> > > > > >> > > > > we
>> > > > > >> > > > > > > mean
>> > > > > >> > > > > > > > > that we
>> > > > > >> > > > > > > > > > > > can switch the default for the feature
>> flag
>> > to
>> > > > > true
>> > > > > >> > > based on
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > > > > metadata.version once we want to make it
>> the
>> > > > > >> default.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention
>> > the
>> > > > > >> required
>> > > > > >> > > ACLs,
>> > > > > >> > > > > it
>> > > > > >> > > > > > > > > would be
>> > > > > >> > > > > > > > > > > > good to add that for consistency.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
>> > ConsumerGroupHeartbeat
>> > > > > >> > requires
>> > > > > >> > > > > "Read
>> > > > > >> > > > > > > > > Group" even
>> > > > > >> > > > > > > > > > > > though it seems to do more than reading.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > I had that thought myself as well, but in
>> the
>> > end we
>> > > > > >> > could
>> > > > > >> > > not
>> > > > > >> > > > > > > find a
>> > > > > >> > > > > > > > > > > better alternative: adding Write Group
>> seems
>> > an
>> > > > > >> overkill
>> > > > > >> > > here
>> > > > > >> > > > > > > since we
>> > > > > >> > > > > > > > > do
>> > > > > >> > > > > > > > > > > not have it elsewhere (we only have Read /
>> > Delete
>> > > > > and
>> > > > > >> > > Describe
>> > > > > >> > > > > on
>> > > > > >> > > > > > > > > groups so
>> > > > > >> > > > > > > > > > > far). Would like to hear others thoughts.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 13. How is topic recreation handled by
>> the
>> > > > > consumer
>> > > > > >> > with
>> > > > > >> > > the
>> > > > > >> > > > > new
>> > > > > >> > > > > > > > > group
>> > > > > >> > > > > > > > > > > > protocol? It would be good to have a
>> > section on
>> > > > > >> this.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > You mean with regex subscription right? Yes
>> > we can
>> > > > > >> add a
>> > > > > >> > > > > section
>> > > > > >> > > > > > > about
>> > > > > >> > > > > > > > > > > that, but basically the idea is that
>> consumer
>> > would
>> > > > > be
>> > > > > >> > > totally
>> > > > > >> > > > > > > > > agnostic in
>> > > > > >> > > > > > > > > > > the new protocol as it's handled all by the
>> > brokers.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write the
>> new
>> > > > > >> coordinator
>> > > > > >> > in
>> > > > > >> > > > > Java.
>> > > > > >> > > > > > > Even
>> > > > > >> > > > > > > > > though
>> > > > > >> > > > > > > > > > > > this is an implementation detail, do we
>> > plan to
>> > > > > >> have a
>> > > > > >> > > new
>> > > > > >> > > > > gradle
>> > > > > >> > > > > > > > > module
>> > > > > >> > > > > > > > > > > > for it?
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > We have not thought about that. But I think
>> > the
>> > > > > answer
>> > > > > >> > > should
>> > > > > >> > > > > be
>> > > > > >> > > > > > > yes.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it
>> > comes to
>> > > > > >> how
>> > > > > >> > > many
>> > > > > >> > > > > > > members
>> > > > > >> > > > > > > > > the new
>> > > > > >> > > > > > > > > > > > group protocol can support?
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > Within a group, I think we should shoot for
>> > 1000s of
>> > > > > >> > > members.
>> > > > > >> > > > > But
>> > > > > >> > > > > > > that
>> > > > > >> > > > > > > > > > > scalability goals also depend on the offset
>> > > > > management
>> > > > > >> > > (commit,
>> > > > > >> > > > > > > fetch)
>> > > > > >> > > > > > > > > > > capabilities of the coordinator which we
>> did
>> > not
>> > > > > >> cover in
>> > > > > >> > > this
>> > > > > >> > > > > > > KIP, so
>> > > > > >> > > > > > > > > it's
>> > > > > >> > > > > > > > > > > hard to give a number that applies
>> > universally.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > 16. Did we consider having
>> > SubscribedTopidIds
>> > > > > >> instead
>> > > > > >> > of
>> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
>> > > > > >> ConsumerGroupHeartbeatRequest?
>> > > > > >> > > Is the
>> > > > > >> > > > > > > idea
>> > > > > >> > > > > > > > > that
>> > > > > >> > > > > > > > > > > > since we have to resolve the regex on the
>> > server,
>> > > > > we
>> > > > > >> > can
>> > > > > >> > > do
>> > > > > >> > > > > the
>> > > > > >> > > > > > > same
>> > > > > >> > > > > > > > > for
>> > > > > >> > > > > > > > > > > > the topic name? The difference is that
>> > sending the
>> > > > > >> > regex
>> > > > > >> > > is
>> > > > > >> > > > > more
>> > > > > >> > > > > > > > > efficient
>> > > > > >> > > > > > > > > > > > whereas sending the topic names is less
>> > efficient.
>> > > > > >> > > > > Furthermore,
>> > > > > >> > > > > > > > > delete and
>> > > > > >> > > > > > > > > > > > recreation is easier to handle if we have
>> > topic
>> > > > > ids.
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > The main reason to still let the clients
>> send
>> > names
>> > > > > >> is to
>> > > > > >> > > keep
>> > > > > >> > > > > the
>> > > > > >> > > > > > > > > > > reasoning of names -> ids on the broker /
>> > admin
>> > > > > client
>> > > > > >> > > only.
>> > > > > >> > > > > Note
>> > > > > >> > > > > > > that
>> > > > > >> > > > > > > > > > > although we added topic id in KIP-516, we
>> > never
>> > > > > >> > > implemented the
>> > > > > >> > > > > > > logic
>> > > > > >> > > > > > > > > on
>> > > > > >> > > > > > > > > > > consumer/producers leveraging the related
>> > newer
>> > > > > >> versioned
>> > > > > >> > > RPCs,
>> > > > > >> > > > > > > > > instead we
>> > > > > >> > > > > > > > > > > just set the topic id as empty UUID. We
>> want
>> > to keep
>> > > > > >> the
>> > > > > >> > > > > > > > > consumer/producer
>> > > > > >> > > > > > > > > > > to be thin and only delegate the reasoning
>> on
>> > broker
>> > > > > >> and
>> > > > > >> > > > > > > potentially
>> > > > > >> > > > > > > > > admin
>> > > > > >> > > > > > > > > > > clients.
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > Thanks,
>> > > > > >> > > > > > > > > > > > Ismael
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David
>> Jacot
>> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
>> > > > > >> > > > > > > > > > > > wrote:
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > > Hi all,
>> > > > > >> > > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > > I would like to start a discussion
>> thread
>> > on
>> > > > > >> KIP-848:
>> > > > > >> > > The
>> > > > > >> > > > > Next
>> > > > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance
>> > Protocol.
>> > > > > >> With
>> > > > > >> > > this
>> > > > > >> > > > > KIP,
>> > > > > >> > > > > > > we
>> > > > > >> > > > > > > > > aim
>> > > > > >> > > > > > > > > > > > > to make the rebalance protocol (for
>> > consumers)
>> > > > > >> more
>> > > > > >> > > > > reliable,
>> > > > > >> > > > > > > more
>> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
>> > clients, and
>> > > > > >> easier
>> > > > > >> > > to
>> > > > > >> > > > > debug
>> > > > > >> > > > > > > for
>> > > > > >> > > > > > > > > > > > > operators.
>> > > > > >> > > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > > The KIP is here:
>> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
>> > > > > >> > > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > > Please take a look and let me know what
>> > you
>> > > > > think.
>> > > > > >> > > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > > Best,
>> > > > > >> > > > > > > > > > > > > David
>> > > > > >> > > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > > > PS: I will be away from July 18th to
>> > August 8th.
>> > > > > >> That
>> > > > > >> > > gives
>> > > > > >> > > > > > > you a
>> > > > > >> > > > > > > > > bit
>> > > > > >> > > > > > > > > > > > > of time to read and digest this long
>> KIP.
>> > > > > >> > > > > > > > > > > > >
>> > > > > >> > > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > >
>> > > > > >> > > > > > > > > > > --
>> > > > > >> > > > > > > > > > > -- Guozhang
>> > > > > >> > > > > > > > >
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > >
>> > > > > >> > > > > > > > --
>> > > > > >> > > > > > > > -- Guozhang
>> > > > > >> > > > > > >
>> > > > > >> > > > >
>> > > > > >> > >
>> > > > > >> > >
>> > > > > >> >
>> > > > > >>
>> > > > > >>
>> > > > > >> --
>> > > > > >> -- Guozhang
>> > > > > >>
>> > > > > >
>> > > > >
>> >
>>
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Luke Chen <sh...@gmail.com>]

Hi Sagar,

I have some thoughts about Kafka Connect integrating with KIP-848, but I
think we should have a separate discussion thread for the Kafka Connect
KIP: Integrating Kafka Connect With New Consumer Rebalance Protocol [1],
and let this discussion thread focus on consumer rebalance protocol, WDYT?

[1]
https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol

Thank you.
Luke



On Fri, Aug 12, 2022 at 9:31 PM Sagar <sa...@gmail.com> wrote:

> Thank you Guozhang/David for the feedback. Looks like there's agreement on
> using separate APIs for Connect. I would revisit the doc and see what
> changes are to be made.
>
> Thanks!
> Sagar.
>
> On Tue, Aug 9, 2022 at 7:11 PM David Jacot <dj...@confluent.io.invalid>
> wrote:
>
> > Hi Sagar,
> >
> > Thanks for the feedback and the document. That's really helpful. I
> > will take a look at it.
> >
> > Overall, it seems to me that both Connect and the Consumer could share
> > the same underlying "engine". The main difference is that the Consumer
> > assigns topic-partitions to members whereas Connect assigns tasks to
> > workers. I see two ways to move forward:
> > 1) We extend the new proposed APIs to support different resource types
> > (e.g. partitions, tasks, etc.); or
> > 2) We use new dedicated APIs for Connect. The dedicated APIs would be
> > similar to the new ones but different on the content/resources and
> > they would rely on the same engine on the coordinator side.
> >
> > I personally lean towards 2) because I am not a fan of overcharging
> > APIs to serve different purposes. That being said, I am not opposed to
> > 1) if we can find an elegant way to do it.
> >
> > I think that we can continue to discuss it here for now in order to
> > ensure that this KIP is compatible with what we will do for Connect in
> > the future.
> >
> > Best,
> > David
> >
> > On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io> wrote:
> > >
> > > Hi all,
> > >
> > > I am back from vacation. I will go through and address your comments
> > > in the coming days. Thanks for your feedback.
> > >
> > > Cheers,
> > > David
> > >
> > > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gh...@gmail.com>
> > wrote:
> > > >
> > > > Hey All!
> > > >
> > > > Thanks for the KIP, it's wonderful to see cooperative rebalancing
> > making it
> > > > down the stack!
> > > >
> > > > I had a few questions:
> > > >
> > > > 1. The 'Rejected Alternatives' section describes how member epoch
> > should
> > > > advance in step with the group epoch and assignment epoch values. I
> > think
> > > > that this is a good idea for the reasons described in the KIP. When
> the
> > > > protocol is incrementally assigning partitions to a worker, what
> member
> > > > epoch does each incremental assignment use? Are member epochs
> re-used,
> > and
> > > > a single member epoch can correspond to multiple different
> > (monotonically
> > > > larger) assignments?
> > > >
> > > > 2. Is the Assignor's 'Reason' field opaque to the group coordinator?
> If
> > > > not, should custom client-side assignor implementations interact with
> > the
> > > > Reason field, and how is its common meaning agreed upon? If so, what
> > is the
> > > > benefit of a distinct Reason field over including such functionality
> > in the
> > > > opaque metadata?
> > > >
> > > > 3. The following is included in the KIP: "Thanks to this, the input
> of
> > the
> > > > client side assignor is entirely driven by the group coordinator. The
> > > > consumer is no longer responsible for maintaining any state besides
> its
> > > > assigned partitions." Does this mean that the client-side assignor
> MAY
> > > > incorporate additional non-Metadata state (such as partition
> > throughput,
> > > > cpu/memory metrics, config topics, etc), or that additional
> > non-Metadata
> > > > state SHOULD NOT be used?
> > > >
> > > > 4. I see that there are separate classes
> > > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > > > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> > > > overlap significantly. Is it possible for these two implementations
> to
> > be
> > > > unified? This would serve to promote feature parity of server-side
> and
> > > > client-side assignors, and would also facilitate operational
> > flexibility in
> > > > certain situations. For example, if a server-side assignor has some
> > poor
> > > > behavior and needs a patch, deploying the patched assignor to the
> > client
> > > > and switching one consumer group to a client-side assignor may be
> > faster
> > > > and less risky than patching all of the brokers. With the currently
> > > > proposed distinct APIs, a non-trivial reimplementation would have to
> be
> > > > assembled, and if the two APIs have diverged significantly, then it
> is
> > > > possible that a reimplementation would not be possible.
> > > >
> > > > --
> > > > Greg Harris
> > > > gharris1727@gmail.com
> > > > github.com/gharris1727
> > > >
> > > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com>
> > wrote:
> > > >
> > > > > Hi Guozhang/David,
> > > > >
> > > > > I created a confluence page to discuss how Connect would need to
> > change
> > > > > based on the new rebalance protocol. Here's the page:
> > > > >
> > > > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > > >
> > > > > It's also pretty longish and I have tried to keep a format similar
> to
> > > > > KIP-848. Let me know what you think. Also, do you think this should
> > be
> > > > > moved to a separate discussion thread or is this one fine?
> > > > >
> > > > > Thanks!
> > > > > Sagar.
> > > > >
> > > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com>
> > wrote:
> > > > >
> > > > > > Hello Guozhang,
> > > > > >
> > > > > > Thank you so much for the doc on Kafka Streams. Sure, I would do
> > the
> > > > > > analysis and come up with such a document.
> > > > > >
> > > > > > Thanks!
> > > > > > Sagar.
> > > > > >
> > > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <
> wangguoz@gmail.com>
> > > > > wrote:
> > > > > >
> > > > > >> Hello Sagar,
> > > > > >>
> > > > > >> It would be great if you could come back with some analysis on
> > how to
> > > > > >> implement the Connect side integration with the new protocol; so
> > far
> > > > > >> besides leveraging on the new "protocol type" we did not yet
> think
> > > > > through
> > > > > >> the Connect side implementations. For Streams here's a draft of
> > > > > >> integration
> > > > > >> plan:
> > > > > >>
> > > > > >>
> > > > >
> >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > > >> just FYI for your analysis on Connect.
> > > > > >>
> > > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <
> sagarmeansocean@gmail.com
> > >
> > > > > wrote:
> > > > > >>
> > > > > >> > Hi David,
> > > > > >> >
> > > > > >> > Thank you for your response. The reason I thought connect can
> > also fit
> > > > > >> into
> > > > > >> > this new scheme is that even today the connect uses a
> > > > > WorkerCoordinator
> > > > > >> > extending from AbstractCoordinator to empower rebalances of
> > > > > >> > tasks/connectors. The WorkerCoordinator sets the
> protocolType()
> > to
> > > > > >> connect
> > > > > >> > and uses the metadata() method by plumbing into
> > > > > >> JoinGroupRequestProtocol.
> > > > > >> >
> > > > > >> > I think the changes to support connect would be similar at a
> > high
> > > > > level
> > > > > >> to
> > > > > >> > the changes in streams mainly because of the Client side
> > assignors
> > > > > being
> > > > > >> > used in both. At an implementation level, we might need to
> make
> > a lot
> > > > > of
> > > > > >> > changes to get onto this new assignment protocol like
> enhancing
> > the
> > > > > >> > JoinGroup request/response and SyncGroup and using
> > > > > >> ConsumerGroupHeartbeat
> > > > > >> > API etc again on similar lines to streams (or there might be
> > > > > >> deviations). I
> > > > > >> > would try to perform a detailed analysis of the same  and we
> > can have
> > > > > a
> > > > > >> > separate discussion thread for that as that would derail this
> > > > > discussion
> > > > > >> > thread. Let me know if that sounds good to you.
> > > > > >> >
> > > > > >> > Thanks!
> > > > > >> > Sagar.
> > > > > >> >
> > > > > >> >
> > > > > >> >
> > > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > > <djacot@confluent.io.invalid
> > > > > >> >
> > > > > >> > wrote:
> > > > > >> >
> > > > > >> > > Hi Sagar,
> > > > > >> > >
> > > > > >> > > Thanks for your comments.
> > > > > >> > >
> > > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can
> > mention it.
> > > > > >> > >
> > > > > >> > > 2) The idea is to transition C from his current assignment
> to
> > his
> > > > > >> > > target assignment when he can move to epoch 3. When that
> > happens,
> > > > > the
> > > > > >> > > member assignment is updated and persisted with all its
> > assigned
> > > > > >> > > partitions even if they are not all revoked yet. In other
> > words, the
> > > > > >> > > member assignment becomes the target assignment. This is
> > basically
> > > > > an
> > > > > >> > > optimization to avoid having to write all the changes to the
> > log.
> > > > > The
> > > > > >> > > examples are based on the persisted state so I understand
> the
> > > > > >> > > confusion. Let me see if I can improve this in the
> > description.
> > > > > >> > >
> > > > > >> > > 3) Regarding Connect, it could reuse the protocol with a
> > client side
> > > > > >> > > assignor if it fits in the protocol. The assignment is about
> > > > > >> > > topicid-partitions + metadata, could Connect fit into this?
> > > > > >> > >
> > > > > >> > > Best,
> > > > > >> > > David
> > > > > >> > >
> > > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> > sagarmeansocean@gmail.com>
> > > > > >> wrote:
> > > > > >> > > >
> > > > > >> > > > Hi David,
> > > > > >> > > >
> > > > > >> > > > Thanks for the KIP. I just had minor observations:
> > > > > >> > > >
> > > > > >> > > > 1) In the Assignment Error section in Client Side mode
> > Assignment
> > > > > >> > > process,
> > > > > >> > > > you mentioned => `In this case, the client side assignor
> can
> > > > > return
> > > > > >> an
> > > > > >> > > > error to the group coordinator`. In this case are you
> > referring to
> > > > > >> the
> > > > > >> > > > Assignor returning an AssignmentError that's listed down
> > towards
> > > > > the
> > > > > >> > end?
> > > > > >> > > > If yes, do you think it would make sense to mention this
> > > > > explicitly
> > > > > >> > here?
> > > > > >> > > >
> > > > > >> > > > 2) In the Case Studies section, I have a slight confusion,
> > not
> > > > > sure
> > > > > >> if
> > > > > >> > > > others have the same. Consider this step:
> > > > > >> > > >
> > > > > >> > > > When B heartbeats, the group coordinator transitions him
> to
> > epoch
> > > > > 3
> > > > > >> > > because
> > > > > >> > > > B has no partitions to revoke. It persists the change and
> > reply.
> > > > > >> > > >
> > > > > >> > > >    - Group (epoch=3)
> > > > > >> > > >       - A
> > > > > >> > > >       - B
> > > > > >> > > >       - C
> > > > > >> > > >    - Target Assignment (epoch=3)
> > > > > >> > > >       - A - partitions=[foo-0]
> > > > > >> > > >       - B - partitions=[foo-2]
> > > > > >> > > >       - C - partitions=[foo-1]
> > > > > >> > > >    - Member Assignment
> > > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > > >> > > >
> > > > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot
> get
> > foo-1
> > > > > >> yet.
> > > > > >> > > >
> > > > > >> > > > Here,it's mentioned that member C can't get the foo-1
> > partition
> > > > > yet,
> > > > > >> > but
> > > > > >> > > > based on the description above, it seems it already has
> it.
> > Do you
> > > > > >> > think
> > > > > >> > > it
> > > > > >> > > > would be better to remove it and populate it only when it
> > actually
> > > > > >> gets
> > > > > >> > > it?
> > > > > >> > > > I see this in a lot of other places, so have I understood
> it
> > > > > >> > incorrectly
> > > > > >> > > ?
> > > > > >> > > >
> > > > > >> > > >
> > > > > >> > > > Regarding connect , it might be out of scope of this
> > discussion,
> > > > > but
> > > > > >> > from
> > > > > >> > > > what I understood it would probably be running in client
> > side
> > > > > >> assignor
> > > > > >> > > mode
> > > > > >> > > > even on the new rebalance protocol as it has its own
> Custom
> > > > > >> > > Assignors(Eager
> > > > > >> > > > and IncrementalCooperative).
> > > > > >> > > >
> > > > > >> > > > Thanks!
> > > > > >> > > >
> > > > > >> > > > Sagar.
> > > > > >> > > >
> > > > > >> > > >
> > > > > >> > > >
> > > > > >> > > >
> > > > > >> > > >
> > > > > >> > > >
> > > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > > > >> > <djacot@confluent.io.invalid
> > > > > >> > > >
> > > > > >> > > > wrote:
> > > > > >> > > >
> > > > > >> > > > > Thanks Hector! Our goal is to move forward with
> > specialized API
> > > > > >> > > > > instead of relying on one generic API. For Connect, we
> > can apply
> > > > > >> the
> > > > > >> > > > > exact same pattern and reuse/share the core
> > implementation on
> > > > > the
> > > > > >> > > > > server side. For the schema registry, I think that we
> > should
> > > > > >> consider
> > > > > >> > > > > having a tailored API to do simple membership/leader
> > election.
> > > > > >> > > > >
> > > > > >> > > > > Best,
> > > > > >> > > > > David
> > > > > >> > > > >
> > > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> > ismael@juma.me.uk
> > > > > >
> > > > > >> > > wrote:
> > > > > >> > > > > >
> > > > > >> > > > > > Three quick comments:
> > > > > >> > > > > >
> > > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > > > >> com.google.re2j.Pattern, we
> > > > > >> > > > > should
> > > > > >> > > > > > document the differences in more detail before
> deciding
> > one
> > > > > way
> > > > > >> or
> > > > > >> > > > > another.
> > > > > >> > > > > > That said, if people pass java.util.regex.Pattern,
> they
> > expect
> > > > > >> > their
> > > > > >> > > > > > semantics to be honored. If we are doing something
> > different,
> > > > > >> then
> > > > > >> > we
> > > > > >> > > > > > should consider adding an overload with our own
> Pattern
> > class
> > > > > (I
> > > > > >> > > don't
> > > > > >> > > > > > think we'd want to expose re2j's at this point).
> > > > > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > > > > integrate
> > > > > >> > fully
> > > > > >> > > > > with
> > > > > >> > > > > > it and should handle the topic recreation case
> > correctly.
> > > > > That's
> > > > > >> > the
> > > > > >> > > main
> > > > > >> > > > > > part we need to handle. I agree with David that we'd
> > want to
> > > > > add
> > > > > >> > > topic
> > > > > >> > > > > ids
> > > > > >> > > > > > to the relevant protocols that don't have it yet and
> > that we
> > > > > can
> > > > > >> > > probably
> > > > > >> > > > > > focus on the internals versus adding new APIs to the
> > Java
> > > > > >> Consumer
> > > > > >> > > > > (unless
> > > > > >> > > > > > we find that adding new APIs is required for
> reasonable
> > > > > >> semantics).
> > > > > >> > > > > > 3. I am still not sure about the coordinator storing
> the
> > > > > >> configs.
> > > > > >> > > It's
> > > > > >> > > > > > powerful for configs to be centralized in the metadata
> > log for
> > > > > >> > > various
> > > > > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > > > > >> Similarly, I
> > > > > >> > > am
> > > > > >> > > > > not
> > > > > >> > > > > > sure about automatically deleting configs in a way
> that
> > they
> > > > > >> cannot
> > > > > >> > > be
> > > > > >> > > > > > recovered. A good property for modern systems is to
> > minimize
> > > > > the
> > > > > >> > > number
> > > > > >> > > > > of
> > > > > >> > > > > > unrecoverable data loss scenarios.
> > > > > >> > > > > >
> > > > > >> > > > > > Ismael
> > > > > >> > > > > >
> > > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > > > >> > > <djacot@confluent.io.invalid
> > > > > >> > > > > >
> > > > > >> > > > > > wrote:
> > > > > >> > > > > >
> > > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 1) the migration path, especially the last step
> when
> > > > > clients
> > > > > >> > > flip the
> > > > > >> > > > > > > flag
> > > > > >> > > > > > > > to enable the new protocol, in which we would
> have a
> > > > > window
> > > > > >> > where
> > > > > >> > > > > both
> > > > > >> > > > > > > new
> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used
> > by
> > > > > >> members
> > > > > >> > of
> > > > > >> > > the
> > > > > >> > > > > same
> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> > behavior
> > > > > >> while
> > > > > >> > > > > using the
> > > > > >> > > > > > > > new protocol is something we need to present
> about.
> > > > > >> > > > > > >
> > > > > >> > > > > > > Noted. I just published a new version of KIP which
> > includes
> > > > > >> more
> > > > > >> > > > > > > details about this. See the "Supporting Online
> > Consumer
> > > > > Group
> > > > > >> > > Upgrade"
> > > > > >> > > > > > > and the "Compatibility, Deprecation, and Migration
> > Plan". I
> > > > > >> think
> > > > > >> > > that
> > > > > >> > > > > > > I have to think through a few cases now but the
> > overall idea
> > > > > >> and
> > > > > >> > > > > > > mechanism should be understandable.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> > topic ids
> > > > > >> are
> > > > > >> > > only
> > > > > >> > > > > used
> > > > > >> > > > > > > as
> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> > exposed
> > > > > via
> > > > > >> any
> > > > > >> > > > > public
> > > > > >> > > > > > > APIs
> > > > > >> > > > > > > > to consumers yet. I think the question is, first
> > should we
> > > > > >> let
> > > > > >> > > the
> > > > > >> > > > > > > consumer
> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> > itself
> > > > > to
> > > > > >> > fully
> > > > > >> > > > > > > leverage
> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> RPCs
> > with
> > > > > the
> > > > > >> > > topic
> > > > > >> > > > > ids;
> > > > > >> > > > > > > and
> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> > topic ids
> > > > > in
> > > > > >> the
> > > > > >> > > > > consumer
> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> well
> > as in
> > > > > >> the
> > > > > >> > > > > rebalance
> > > > > >> > > > > > > > listener for cases like topic
> > deletion-and-recreation).
> > > > > >> > > > > > >
> > > > > >> > > > > > > a) Assuming that we would include converting all the
> > offsets
> > > > > >> > > related
> > > > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer
> > would be
> > > > > >> able
> > > > > >> > to
> > > > > >> > > > > > > fully operate with topic ids. That being said, it
> > still has
> > > > > to
> > > > > >> > > provide
> > > > > >> > > > > > > the topics names in various APIs so having a mapping
> > in the
> > > > > >> > > consumer
> > > > > >> > > > > > > seems inevitable to me.
> > > > > >> > > > > > > b) I don't have a strong opinion on this. Here I
> > wonder if
> > > > > >> this
> > > > > >> > > goes
> > > > > >> > > > > > > beyond the scope of this KIP. I would rather focus
> on
> > the
> > > > > >> > internals
> > > > > >> > > > > > > here and we can consider this separately if we see
> > value in
> > > > > >> doing
> > > > > >> > > it.
> > > > > >> > > > > > >
> > > > > >> > > > > > > Coming back to Ismael's point about using topic ids
> > in the
> > > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is
> > one
> > > > > >> > advantage
> > > > > >> > > in
> > > > > >> > > > > > > favour of it. The consumer will have the opportunity
> > to
> > > > > >> validate
> > > > > >> > > that
> > > > > >> > > > > > > the topics exists before passing them into the group
> > > > > rebalance
> > > > > >> > > > > > > protocol. Obviously, the coordinator will also
> notice
> > it but
> > > > > >> it
> > > > > >> > > does
> > > > > >> > > > > > > not really have a way to reject an invalid topic in
> > the
> > > > > >> response.
> > > > > >> > > > > > >
> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> questions
> > > > > except
> > > > > >> for
> > > > > >> > > the
> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> > it's not
> > > > > >> > > necessary
> > > > > >> > > > > to
> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> > instead we
> > > > > >> still
> > > > > >> > > use
> > > > > >> > > > > > > Pattern
> > > > > >> > > > > > > > while just documenting that our subscription may
> be
> > > > > >> rejected by
> > > > > >> > > the
> > > > > >> > > > > > > server.
> > > > > >> > > > > > > > Since the incompatible case is a very rare
> scenario
> > I felt
> > > > > >> > using
> > > > > >> > > an
> > > > > >> > > > > > > > overloaded `String` based subscription may be more
> > > > > >> vulnerable
> > > > > >> > to
> > > > > >> > > > > various
> > > > > >> > > > > > > > invalid regexes.
> > > > > >> > > > > > >
> > > > > >> > > > > > > That could work. I have to look at the differences
> > between
> > > > > the
> > > > > >> > two
> > > > > >> > > > > > > engines to better understand the potential issues.
> My
> > > > > >> > > understanding is
> > > > > >> > > > > > > that would work for all the basic regular
> > expressions. The
> > > > > >> > > differences
> > > > > >> > > > > > > between the two are mainly about the various
> character
> > > > > >> classes. I
> > > > > >> > > > > > > wonder what other people think about this.
> > > > > >> > > > > > >
> > > > > >> > > > > > > Best,
> > > > > >> > > > > > > David
> > > > > >> > > > > > >
> > > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > > > > >> > wangguoz@gmail.com
> > > > > >> > > >
> > > > > >> > > > > wrote:
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Thanks David! I think on the high level there are
> > two meta
> > > > > >> > > points we
> > > > > >> > > > > need
> > > > > >> > > > > > > > to concretize a bit more:
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > 1) the migration path, especially the last step
> when
> > > > > clients
> > > > > >> > > flip the
> > > > > >> > > > > > > flag
> > > > > >> > > > > > > > to enable the new protocol, in which we would
> have a
> > > > > window
> > > > > >> > where
> > > > > >> > > > > both
> > > > > >> > > > > > > new
> > > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used
> > by
> > > > > >> members
> > > > > >> > of
> > > > > >> > > the
> > > > > >> > > > > same
> > > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> > behavior
> > > > > >> while
> > > > > >> > > > > using the
> > > > > >> > > > > > > > new protocol is something we need to present
> about.
> > > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> > topic ids
> > > > > >> are
> > > > > >> > > only
> > > > > >> > > > > used
> > > > > >> > > > > > > as
> > > > > >> > > > > > > > part of RPCs and admin client, but they are not
> > exposed
> > > > > via
> > > > > >> any
> > > > > >> > > > > public
> > > > > >> > > > > > > APIs
> > > > > >> > > > > > > > to consumers yet. I think the question is, first
> > should we
> > > > > >> let
> > > > > >> > > the
> > > > > >> > > > > > > consumer
> > > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> > itself
> > > > > to
> > > > > >> > fully
> > > > > >> > > > > > > leverage
> > > > > >> > > > > > > > on all the augmented existing RPCs and the new
> RPCs
> > with
> > > > > the
> > > > > >> > > topic
> > > > > >> > > > > ids;
> > > > > >> > > > > > > and
> > > > > >> > > > > > > > secondly, should we ever consider exposing the
> > topic ids
> > > > > in
> > > > > >> the
> > > > > >> > > > > consumer
> > > > > >> > > > > > > > public APIs as well (both subscribe/assign, as
> well
> > as in
> > > > > >> the
> > > > > >> > > > > rebalance
> > > > > >> > > > > > > > listener for cases like topic
> > deletion-and-recreation).
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > I'm agreeing with David on all other minor
> questions
> > > > > except
> > > > > >> for
> > > > > >> > > the
> > > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> > it's not
> > > > > >> > > necessary
> > > > > >> > > > > to
> > > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> > instead we
> > > > > >> still
> > > > > >> > > use
> > > > > >> > > > > > > Pattern
> > > > > >> > > > > > > > while just documenting that our subscription may
> be
> > > > > >> rejected by
> > > > > >> > > the
> > > > > >> > > > > > > server.
> > > > > >> > > > > > > > Since the incompatible case is a very rare
> scenario
> > I felt
> > > > > >> > using
> > > > > >> > > an
> > > > > >> > > > > > > > overloaded `String` based subscription may be more
> > > > > >> vulnerable
> > > > > >> > to
> > > > > >> > > > > various
> > > > > >> > > > > > > > invalid regexes.
> > > > > >> > > > > > > >
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > Guozhang
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > > > >> > > > > <djacot@confluent.io.invalid
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > wrote:
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > > Hi Ismael,
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Thanks for your feedback. Let me answer your
> > questions
> > > > > >> > inline.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 1. I think it's premature to talk about target
> > > > > versions
> > > > > >> for
> > > > > >> > > > > > > deprecation
> > > > > >> > > > > > > > > and
> > > > > >> > > > > > > > > > removal of the existing group protocol. Unlike
> > KRaft,
> > > > > >> this
> > > > > >> > > > > affects a
> > > > > >> > > > > > > core
> > > > > >> > > > > > > > > > client protocol and hence deprecation/removal
> > will be
> > > > > >> > heavily
> > > > > >> > > > > > > dependent
> > > > > >> > > > > > > > > on
> > > > > >> > > > > > > > > > how quickly applications migrate to the new
> > protocol.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > That makes sense. I will remove it.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 2. The KIP says we intend to release this in
> > 4.x, but
> > > > > it
> > > > > >> > > wasn't
> > > > > >> > > > > made
> > > > > >> > > > > > > > > clear
> > > > > >> > > > > > > > > > why. If we added that as a way to estimate
> when
> > we'd
> > > > > >> > > deprecate
> > > > > >> > > > > and
> > > > > >> > > > > > > remove
> > > > > >> > > > > > > > > > the group protocol, I also suggest removing
> > this part.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Let me explain my reasoning. As explained, I
> plan
> > to
> > > > > >> rewrite
> > > > > >> > > the
> > > > > >> > > > > group
> > > > > >> > > > > > > > > coordinator in Java while we implement the new
> > protocol.
> > > > > >> This
> > > > > >> > > means
> > > > > >> > > > > > > > > that the internals will be slightly different
> > (e.g.
> > > > > >> threading
> > > > > >> > > > > model).
> > > > > >> > > > > > > > > Therefore, I wanted to tighten the switch from
> > the old
> > > > > >> group
> > > > > >> > > > > > > > > coordinator to the new group coordinator to a
> > major
> > > > > >> release.
> > > > > >> > > The
> > > > > >> > > > > > > > > alternative would be to use a flag to do the
> > switch
> > > > > >> instead
> > > > > >> > of
> > > > > >> > > > > relying
> > > > > >> > > > > > > > > on the software upgrade.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 3. We need to flesh out the details of the
> > migration
> > > > > >> story.
> > > > > >> > > It
> > > > > >> > > > > sounds
> > > > > >> > > > > > > > > like
> > > > > >> > > > > > > > > > we're saying we will support online
> migrations.
> > Is
> > > > > that
> > > > > >> > > correct?
> > > > > >> > > > > We
> > > > > >> > > > > > > > > should
> > > > > >> > > > > > > > > > explain this in detail. It could also be done
> > as a
> > > > > >> separate
> > > > > >> > > KIP,
> > > > > >> > > > > if
> > > > > >> > > > > > > it's
> > > > > >> > > > > > > > > > easier.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Yes, we will support online migrations for the
> > group.
> > > > > That
> > > > > >> > > means
> > > > > >> > > > > that
> > > > > >> > > > > > > > > a group using the old protocol will be able to
> > switch to
> > > > > >> the
> > > > > >> > > new
> > > > > >> > > > > > > > > protocol.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Let me briefly explain how that will work
> though.
> > It is
> > > > > >> > > basically a
> > > > > >> > > > > > > > > four step process:
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a
> > software
> > > > > >> > > supporting
> > > > > >> > > > > the
> > > > > >> > > > > > > > > new group coordinator. Both the old and the new
> > > > > >> coordinator
> > > > > >> > > will
> > > > > >> > > > > > > > > support the old protocol and rely on the same
> > persisted
> > > > > >> > > metadata so
> > > > > >> > > > > > > > > they can work together. This point is an offline
> > > > > >> migration.
> > > > > >> > We
> > > > > >> > > > > cannot
> > > > > >> > > > > > > > > do this one live because it would require
> > shutting down
> > > > > >> the
> > > > > >> > > current
> > > > > >> > > > > > > > > coordinator and starting up the new one and that
> > would
> > > > > >> cause
> > > > > >> > > > > > > > > unavailabilities.
> > > > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be
> > upgraded
> > > > > to
> > > > > >> X
> > > > > >> > in
> > > > > >> > > > > order
> > > > > >> > > > > > > > > to enable the new protocol. This cannot be done
> > before
> > > > > 1)
> > > > > >> is
> > > > > >> > > > > > > > > terminated because the old coordinator doesn't
> > support
> > > > > the
> > > > > >> > new
> > > > > >> > > > > > > > > protocol.
> > > > > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > > > > supporting
> > > > > >> the
> > > > > >> > > > > online
> > > > > >> > > > > > > > > migration (must have KIP-792). If the consumer
> is
> > > > > already
> > > > > >> > > there.
> > > > > >> > > > > > > > > Nothing must be done at this point.
> > > > > >> > > > > > > > > 4. The consumers must be rolled with the feature
> > flag
> > > > > >> turned
> > > > > >> > > on.
> > > > > >> > > > > The
> > > > > >> > > > > > > > > consumer group is automatically converted when
> > the first
> > > > > >> > > consumer
> > > > > >> > > > > > > > > using the new protocol joins the group. While
> the
> > > > > members
> > > > > >> > > using the
> > > > > >> > > > > > > > > old protocol are being upgraded, the old
> protocol
> > is
> > > > > >> proxied
> > > > > >> > > into
> > > > > >> > > > > the
> > > > > >> > > > > > > > > new one.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > > > >> subscriptions
> > > > > >> > > to
> > > > > >> > > > > the
> > > > > >> > > > > > > > > server,
> > > > > >> > > > > > > > > > but it seems like there could be some tricky
> > > > > >> compatibility
> > > > > >> > > > > issues.
> > > > > >> > > > > > > Will
> > > > > >> > > > > > > > > we
> > > > > >> > > > > > > > > > have a mechanism for users to detect that they
> > need to
> > > > > >> > update
> > > > > >> > > > > their
> > > > > >> > > > > > > regex
> > > > > >> > > > > > > > > > before switching to the new protocol?
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > I think that I am a bit more optimistic than you
> > on this
> > > > > >> > > point. I
> > > > > >> > > > > > > > > believe that the majority of the cases are
> simple
> > > > > regexes
> > > > > >> > which
> > > > > >> > > > > should
> > > > > >> > > > > > > > > work with the new engine. The coordinator will
> > verify
> > > > > the
> > > > > >> > regex
> > > > > >> > > > > anyway
> > > > > >> > > > > > > > > and reject the consumer if the regex is not
> valid.
> > > > > Coming
> > > > > >> > back
> > > > > >> > > to
> > > > > >> > > > > the
> > > > > >> > > > > > > > > migration path, in the worst case, the first
> > upgraded
> > > > > >> > consumer
> > > > > >> > > > > joining
> > > > > >> > > > > > > > > the group will be rejected. This should be used
> > as the
> > > > > >> last
> > > > > >> > > > > defence, I
> > > > > >> > > > > > > > > would say.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > One way for customers to validate their regex
> > before
> > > > > >> > upgrading
> > > > > >> > > > > their
> > > > > >> > > > > > > > > prod would be to test them with another group.
> For
> > > > > >> instance,
> > > > > >> > > that
> > > > > >> > > > > > > > > could be done in a pre-prod environment. Another
> > way
> > > > > >> would be
> > > > > >> > > to
> > > > > >> > > > > > > > > extend the consumer-group tool to provide a
> regex
> > > > > >> validation
> > > > > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 5. Related to the last question, will the Java
> > client
> > > > > >> allow
> > > > > >> > > the
> > > > > >> > > > > > > users to
> > > > > >> > > > > > > > > > stick with the current regex engine for
> > compatibility
> > > > > >> > > reasons?
> > > > > >> > > > > For
> > > > > >> > > > > > > > > example,
> > > > > >> > > > > > > > > > it may be handy to keep using client based
> > regex at
> > > > > >> first
> > > > > >> > to
> > > > > >> > > keep
> > > > > >> > > > > > > > > > migrations simple and then migrate to server
> > based
> > > > > >> regexes
> > > > > >> > > as a
> > > > > >> > > > > > > second
> > > > > >> > > > > > > > > step.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > I understand your point but I am concerned that
> > this
> > > > > would
> > > > > >> > > allow
> > > > > >> > > > > users
> > > > > >> > > > > > > > > to actually stay in this mode. That would go
> > against our
> > > > > >> goal
> > > > > >> > > of
> > > > > >> > > > > > > > > simplifying the client because we would have to
> > continue
> > > > > >> > > monitoring
> > > > > >> > > > > > > > > the metadata on the client side. I would rather
> > not do
> > > > > >> this.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 6. When we say that the group coordinator will
> > be
> > > > > >> > > responsible for
> > > > > >> > > > > > > storing
> > > > > >> > > > > > > > > > the configurations and that the configurations
> > will be
> > > > > >> > > deleted
> > > > > >> > > > > when
> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > > group is deleted. Will a transition to DEAD
> > trigger
> > > > > >> > deletion
> > > > > >> > > of
> > > > > >> > > > > > > > > > configurations?
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > That's right. The configurations will be deleted
> > when
> > > > > the
> > > > > >> > > group is
> > > > > >> > > > > > > > > deleted. They go together.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 7. Will the choice to store the configs in the
> > group
> > > > > >> > > coordinator
> > > > > >> > > > > > > make it
> > > > > >> > > > > > > > > > harder to list all cluster configs and their
> > values?
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > I don't think so. The group configurations are
> > overrides
> > > > > >> of
> > > > > >> > > cluster
> > > > > >> > > > > > > > > configs. If you want to know all the overrides
> > though,
> > > > > you
> > > > > >> > > would
> > > > > >> > > > > have
> > > > > >> > > > > > > > > to ask all the group coordinators. You cannot
> > rely on
> > > > > the
> > > > > >> > > metadata
> > > > > >> > > > > log
> > > > > >> > > > > > > > > for instance.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 8. How would someone configure a group before
> > starting
> > > > > >> the
> > > > > >> > > > > consumers?
> > > > > >> > > > > > > > > Have
> > > > > >> > > > > > > > > > we considered allowing the explicit creation
> of
> > > > > groups?
> > > > > >> > > > > > > Alternatively,
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > configs could be decoupled from the group
> > lifecycle.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Yes. The group will be automatically created in
> > this
> > > > > case.
> > > > > >> > > However,
> > > > > >> > > > > > > > > the configs will be lost after the retention
> > period of
> > > > > the
> > > > > >> > > group
> > > > > >> > > > > > > > > passes.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the
> > Java
> > > > > >> client
> > > > > >> > > still
> > > > > >> > > > > take
> > > > > >> > > > > > > a
> > > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > introduce
> > > > > an
> > > > > >> > > overload?
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > That's a very group question. I forgot about
> that
> > one.
> > > > > As
> > > > > >> the
> > > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully
> compatible
> > with
> > > > > the
> > > > > >> > > engine
> > > > > >> > > > > that
> > > > > >> > > > > > > > > we plan to use, it might be better to deprecate
> > it and
> > > > > >> use an
> > > > > >> > > > > overload
> > > > > >> > > > > > > > > which takes a string. We would rely on the
> server
> > side
> > > > > >> > > validation.
> > > > > >> > > > > > > > > During the migration, I think that we could
> still
> > try to
> > > > > >> > > toString
> > > > > >> > > > > the
> > > > > >> > > > > > > > > regex and use it. That should work, I think, in
> > the
> > > > > >> majority
> > > > > >> > > of the
> > > > > >> > > > > > > > > cases.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 10. I agree with Justine that we should be
> > clearer
> > > > > about
> > > > > >> > the
> > > > > >> > > > > reason
> > > > > >> > > > > > > to
> > > > > >> > > > > > > > > > switch to IBP/metadata.version from the
> feature
> > flag.
> > > > > >> Maybe
> > > > > >> > > we
> > > > > >> > > > > mean
> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > we
> > > > > >> > > > > > > > > > can switch the default for the feature flag to
> > true
> > > > > >> based
> > > > > >> > on
> > > > > >> > > the
> > > > > >> > > > > > > > > > metadata.version once we want to make it the
> > default.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > My plan was to use that feature flag mainly
> > during the
> > > > > >> > > development
> > > > > >> > > > > > > > > phase. I should not have mentioned it, I think,
> > because
> > > > > we
> > > > > >> > > could
> > > > > >> > > > > use
> > > > > >> > > > > > > > > an internal config for it.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention
> the
> > > > > required
> > > > > >> > > ACLs, it
> > > > > >> > > > > > > would
> > > > > >> > > > > > > > > be
> > > > > >> > > > > > > > > > good to add that for consistency.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Noted.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 12. It is a bit odd that
> ConsumerGroupHeartbeat
> > > > > requires
> > > > > >> > > "Read
> > > > > >> > > > > Group"
> > > > > >> > > > > > > > > even
> > > > > >> > > > > > > > > > though it seems to do more than reading.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > I agree. This is how the current protocol works
> > though.
> > > > > We
> > > > > >> > only
> > > > > >> > > > > > > > > require "Read Group" to join a group. We could
> > consider
> > > > > >> > > changing
> > > > > >> > > > > this
> > > > > >> > > > > > > > > but I am not sure that it is worth it.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 13. How is topic recreation handled by the
> > consumer
> > > > > with
> > > > > >> > the
> > > > > >> > > new
> > > > > >> > > > > > > group
> > > > > >> > > > > > > > > > protocol? It would be good to have a section
> on
> > this.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Noted. From a protocol perspective, the new
> topic
> > will
> > > > > >> have a
> > > > > >> > > new
> > > > > >> > > > > > > > > topic id so it will treat it like a topic with a
> > > > > different
> > > > > >> > > name.
> > > > > >> > > > > The
> > > > > >> > > > > > > > > only issue is that the fetch/commit offsets APIs
> > do not
> > > > > >> > support
> > > > > >> > > > > topic
> > > > > >> > > > > > > > > IDs so the consumer would reuse the offsets
> based
> > on the
> > > > > >> > same.
> > > > > >> > > I
> > > > > >> > > > > think
> > > > > >> > > > > > > > > that we should update those APIs as well in
> order
> > to be
> > > > > >> > > consistent
> > > > > >> > > > > end
> > > > > >> > > > > > > > > to end. That would strengthen the semantics of
> the
> > > > > >> consumer.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 14. The KIP mentions we will write the new
> > coordinator
> > > > > >> in
> > > > > >> > > Java.
> > > > > >> > > > > Even
> > > > > >> > > > > > > > > though
> > > > > >> > > > > > > > > > this is an implementation detail, do we plan
> to
> > have a
> > > > > >> new
> > > > > >> > > gradle
> > > > > >> > > > > > > module
> > > > > >> > > > > > > > > > for it?
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Yes.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 15. Do we have a scalability goal when it
> comes
> > to how
> > > > > >> many
> > > > > >> > > > > members
> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > new
> > > > > >> > > > > > > > > > group protocol can support?
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > We don't have numbers at the moment. The
> protocol
> > should
> > > > > >> > > support
> > > > > >> > > > > 1000s
> > > > > >> > > > > > > > > of members per group. We will measure this when
> > we have
> > > > > a
> > > > > >> > first
> > > > > >> > > > > > > > > implementation. Note that we might have other
> > > > > bottlenecks
> > > > > >> > down
> > > > > >> > > the
> > > > > >> > > > > > > > > road (e.g. offset commits).
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > instead
> > > > > of
> > > > > >> > > > > > > > > > SubscribedTopicNames in
> > ConsumerGroupHeartbeatRequest?
> > > > > >> Is
> > > > > >> > the
> > > > > >> > > > > idea
> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > > since we have to resolve the regex on the
> > server, we
> > > > > >> can do
> > > > > >> > > the
> > > > > >> > > > > same
> > > > > >> > > > > > > for
> > > > > >> > > > > > > > > > the topic name? The difference is that sending
> > the
> > > > > >> regex is
> > > > > >> > > more
> > > > > >> > > > > > > > > efficient
> > > > > >> > > > > > > > > > whereas sending the topic names is less
> > efficient.
> > > > > >> > > Furthermore,
> > > > > >> > > > > > > delete
> > > > > >> > > > > > > > > and
> > > > > >> > > > > > > > > > recreation is easier to handle if we have
> topic
> > ids.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > The idea was to consolidate the metadata lookup
> > on the
> > > > > >> server
> > > > > >> > > for
> > > > > >> > > > > both
> > > > > >> > > > > > > > > paths but I do agree with your point. As a
> second
> > > > > though,
> > > > > >> > using
> > > > > >> > > > > topic
> > > > > >> > > > > > > > > ids may be better here for the delete and
> > recreation
> > > > > case.
> > > > > >> > > Also, I
> > > > > >> > > > > > > > > suppose that we may allow users to subscribe
> with
> > topic
> > > > > >> ids
> > > > > >> > in
> > > > > >> > > the
> > > > > >> > > > > > > > > future because that is the only way to be really
> > robust
> > > > > to
> > > > > >> > > topic
> > > > > >> > > > > > > > > re-creation.
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > Best,
> > > > > >> > > > > > > > > David
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > > > > >> > > djacot@confluent.io>
> > > > > >> > > > > > > wrote:
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Hi Justine,
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Thanks for your comments. Please find my
> answers
> > > > > below.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs
> > with the
> > > > > >> > > exception
> > > > > >> > > > > of the
> > > > > >> > > > > > > > > > topic names based in the
> > > > > ConsumerGroupHeartbeatRequest.
> > > > > >> I
> > > > > >> > am
> > > > > >> > > not
> > > > > >> > > > > sure
> > > > > >> > > > > > > > > > if using topic names is the right call here. I
> > need to
> > > > > >> > think
> > > > > >> > > > > about it
> > > > > >> > > > > > > > > > a little more. Obviously, the KIP does not
> > change the
> > > > > >> > > > > fetch/commit
> > > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be
> > something
> > > > > >> that
> > > > > >> > we
> > > > > >> > > > > should
> > > > > >> > > > > > > > > > include though as it would give better overall
> > > > > >> guarantee in
> > > > > >> > > the
> > > > > >> > > > > > > > > > producer.
> > > > > >> > > > > > > > > > - You're right. I think that I should not have
> > > > > mentioned
> > > > > >> > this
> > > > > >> > > > > flag at
> > > > > >> > > > > > > > > > all. I will remove it. We can use an internal
> > > > > >> configuration
> > > > > >> > > while
> > > > > >> > > > > > > > > > developing the feature.
> > > > > >> > > > > > > > > > - Both cluster types will be supported. The
> > change is
> > > > > >> > > > > orthogonal. The
> > > > > >> > > > > > > > > > only requirement is that the cluster uses
> topic
> > IDs.
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > Best,
> > > > > >> > > > > > > > > > David
> > > > > >> > > > > > > > > >
> > > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang
> <
> > > > > >> > > > > wangguoz@gmail.com>
> > > > > >> > > > > > > > > wrote:
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Hi Ismael,
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Thanks for the feedback. Here are some
> replies
> > > > > inlined
> > > > > >> > > below:
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > > > > >> > > ismael@juma.me.uk>
> > > > > >> > > > > > > wrote:
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > Thanks for the KIP. This has the potential
> > to be a
> > > > > >> > great
> > > > > >> > > > > > > > > improvement. A few
> > > > > >> > > > > > > > > > > > initial questions/comments:
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 1. I think it's premature to talk about
> > target
> > > > > >> versions
> > > > > >> > > for
> > > > > >> > > > > > > > > deprecation and
> > > > > >> > > > > > > > > > > > removal of the existing group protocol.
> > Unlike
> > > > > >> KRaft,
> > > > > >> > > this
> > > > > >> > > > > > > affects a
> > > > > >> > > > > > > > > core
> > > > > >> > > > > > > > > > > > client protocol and hence
> > deprecation/removal will
> > > > > >> be
> > > > > >> > > heavily
> > > > > >> > > > > > > > > dependent on
> > > > > >> > > > > > > > > > > > how quickly applications migrate to the
> new
> > > > > >> protocol.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Yeah I agree with you. I think we can remove
> > the
> > > > > >> proposed
> > > > > >> > > > > timeline
> > > > > >> > > > > > > in
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration
> > Plan` and
> > > > > >> > > instead
> > > > > >> > > > > just
> > > > > >> > > > > > > state
> > > > > >> > > > > > > > > > > that we will decide in the future about when
> > we
> > > > > would
> > > > > >> > > > > deprecate old
> > > > > >> > > > > > > > > > > protocol and behaviors.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 2. The KIP says we intend to release this
> > in 4.x,
> > > > > >> but
> > > > > >> > it
> > > > > >> > > > > wasn't
> > > > > >> > > > > > > made
> > > > > >> > > > > > > > > clear
> > > > > >> > > > > > > > > > > > why. If we added that as a way to estimate
> > when
> > > > > we'd
> > > > > >> > > > > deprecate
> > > > > >> > > > > > > and
> > > > > >> > > > > > > > > remove
> > > > > >> > > > > > > > > > > > the group protocol, I also suggest
> removing
> > this
> > > > > >> part.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > I think that's not specifically related to
> the
> > > > > >> > > > > deprecation/removal
> > > > > >> > > > > > > > > timeline
> > > > > >> > > > > > > > > > > plan, but it's more for client upgrades.
> I.e.
> > the
> > > > > >> > > broker-side
> > > > > >> > > > > > > > > > > implementation may be done first, and then
> the
> > > > > client
> > > > > >> > side,
> > > > > >> > > > > and we
> > > > > >> > > > > > > > > would
> > > > > >> > > > > > > > > > > only mark it as "released" by the time
> clients
> > > > > >> > > implementations
> > > > > >> > > > > are
> > > > > >> > > > > > > > > done. At
> > > > > >> > > > > > > > > > > that time, to enable the feature the clients
> > need to
> > > > > >> > first
> > > > > >> > > > > swap-in
> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > > > bytecode with a rolling bounce and then set
> > the flag
> > > > > >> > with a
> > > > > >> > > > > second
> > > > > >> > > > > > > > > rolling
> > > > > >> > > > > > > > > > > bounce, and hence we feel it's better to be
> > released
> > > > > >> in a
> > > > > >> > > major
> > > > > >> > > > > > > > > version.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > > > > migration
> > > > > >> > > story.
> > > > > >> > > > > It
> > > > > >> > > > > > > > > sounds like
> > > > > >> > > > > > > > > > > > we're saying we will support online
> > migrations. Is
> > > > > >> that
> > > > > >> > > > > correct?
> > > > > >> > > > > > > We
> > > > > >> > > > > > > > > should
> > > > > >> > > > > > > > > > > > explain this in detail. It could also be
> > done as a
> > > > > >> > > separate
> > > > > >> > > > > KIP,
> > > > > >> > > > > > > if
> > > > > >> > > > > > > > > it's
> > > > > >> > > > > > > > > > > > easier.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Yes I think that's the part we can be more
> > concrete
> > > > > >> about
> > > > > >> > > for
> > > > > >> > > > > sure
> > > > > >> > > > > > > (and
> > > > > >> > > > > > > > > > > this is related to your question 2) above).
> > We will
> > > > > >> work
> > > > > >> > on
> > > > > >> > > > > making
> > > > > >> > > > > > > it
> > > > > >> > > > > > > > > more
> > > > > >> > > > > > > > > > > explicit in parallel as we solicit more
> > feedback.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 4. I am happy that we are pushing the
> > pattern
> > > > > >> > > subscriptions
> > > > > >> > > > > to
> > > > > >> > > > > > > the
> > > > > >> > > > > > > > > server,
> > > > > >> > > > > > > > > > > > but it seems like there could be some
> tricky
> > > > > >> > > compatibility
> > > > > >> > > > > > > issues.
> > > > > >> > > > > > > > > Will we
> > > > > >> > > > > > > > > > > > have a mechanism for users to detect that
> > they
> > > > > need
> > > > > >> to
> > > > > >> > > update
> > > > > >> > > > > > > their
> > > > > >> > > > > > > > > regex
> > > > > >> > > > > > > > > > > > before switching to the new protocol?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Yes I think we need some tooling for
> non-java
> > client
> > > > > >> > users
> > > > > >> > > to
> > > > > >> > > > > sort
> > > > > >> > > > > > > of
> > > > > >> > > > > > > > > > > "dry-run" the client before switching to the
> > new
> > > > > >> > protocol.
> > > > > >> > > I
> > > > > >> > > > > do not
> > > > > >> > > > > > > > > have a
> > > > > >> > > > > > > > > > > specific idea on top of my head though,
> maybe
> > others
> > > > > >> like
> > > > > >> > > @Matt
> > > > > >> > > > > > > > > Howlett can
> > > > > >> > > > > > > > > > > chime-in here?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 5. Related to the last question, will the
> > Java
> > > > > >> client
> > > > > >> > > allow
> > > > > >> > > > > the
> > > > > >> > > > > > > > > users to
> > > > > >> > > > > > > > > > > > stick with the current regex engine for
> > > > > >> compatibility
> > > > > >> > > > > reasons?
> > > > > >> > > > > > > For
> > > > > >> > > > > > > > > example,
> > > > > >> > > > > > > > > > > > it may be handy to keep using client based
> > regex
> > > > > at
> > > > > >> > > first to
> > > > > >> > > > > keep
> > > > > >> > > > > > > > > > > > migrations simple and then migrate to
> > server based
> > > > > >> > > regexes
> > > > > >> > > > > as a
> > > > > >> > > > > > > > > second
> > > > > >> > > > > > > > > > > > step.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Honestly I have not thought about that for
> > java
> > > > > >> clients,
> > > > > >> > > and
> > > > > >> > > > > we can
> > > > > >> > > > > > > > > discuss
> > > > > >> > > > > > > > > > > that. What kind of compatibility issues do
> > you have
> > > > > in
> > > > > >> > > mind?
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 6. When we say that the group coordinator
> > will be
> > > > > >> > > > > responsible for
> > > > > >> > > > > > > > > storing
> > > > > >> > > > > > > > > > > > the configurations and that the
> > configurations
> > > > > will
> > > > > >> be
> > > > > >> > > > > deleted
> > > > > >> > > > > > > when
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > > group is deleted. Will a transition to
> DEAD
> > > > > trigger
> > > > > >> > > deletion
> > > > > >> > > > > of
> > > > > >> > > > > > > > > > > > configurations?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Yes, since the DEAD state is an ending state
> > (we
> > > > > would
> > > > > >> > only
> > > > > >> > > > > > > transit to
> > > > > >> > > > > > > > > that
> > > > > >> > > > > > > > > > > state when the group is EMPTY and also all
> of
> > its
> > > > > >> > metadata
> > > > > >> > > are
> > > > > >> > > > > > > gone),
> > > > > >> > > > > > > > > once
> > > > > >> > > > > > > > > > > it's transited to DEAD this group would
> never
> > be
> > > > > >> revived.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 7. Will the choice to store the configs in
> > the
> > > > > group
> > > > > >> > > > > coordinator
> > > > > >> > > > > > > > > make it
> > > > > >> > > > > > > > > > > > harder to list all cluster configs and
> their
> > > > > values?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > That's a good question, and our thoughts are
> > that
> > > > > the
> > > > > >> > > so-called
> > > > > >> > > > > > > "group
> > > > > >> > > > > > > > > > > configurations" are overrides of the
> > cluster-level
> > > > > >> > > > > configurations
> > > > > >> > > > > > > > > > > customized per group so when an admin list
> > cluster
> > > > > >> > configs
> > > > > >> > > it's
> > > > > >> > > > > > > okay to
> > > > > >> > > > > > > > > > > list just the cluster-level "defaults", not
> > showing
> > > > > >> any
> > > > > >> > > > > per-group
> > > > > >> > > > > > > > > > > customizations.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 8. How would someone configure a group
> > before
> > > > > >> starting
> > > > > >> > > the
> > > > > >> > > > > > > > > consumers? Have
> > > > > >> > > > > > > > > > > > we considered allowing the explicit
> > creation of
> > > > > >> groups?
> > > > > >> > > > > > > > > Alternatively, the
> > > > > >> > > > > > > > > > > > configs could be decoupled from the group
> > > > > lifecycle.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > The configs can be created before the group
> > itself
> > > > > as
> > > > > >> an
> > > > > >> > > > > > > independent
> > > > > >> > > > > > > > > entity
> > > > > >> > > > > > > > > > > --- of course, this requires the
> corresponding
> > > > > >> request to
> > > > > >> > > be
> > > > > >> > > > > > > routed to
> > > > > >> > > > > > > > > the
> > > > > >> > > > > > > > > > > right coordinator based on the group id ---
> > the only
> > > > > >> > thing
> > > > > >> > > that
> > > > > >> > > > > > > > > differs is,
> > > > > >> > > > > > > > > > > when the group itself is gone we also check
> > if there
> > > > > >> are
> > > > > >> > > any
> > > > > >> > > > > > > > > configuration
> > > > > >> > > > > > > > > > > entities related to that group and delete as
> > well.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> > asymmetry on
> > > > > the
> > > > > >> > > creation
> > > > > >> > > > > /
> > > > > >> > > > > > > > > deletion
> > > > > >> > > > > > > > > > > lifecycles of the config entities, and we
> > would like
> > > > > >> to
> > > > > >> > > hear
> > > > > >> > > > > > > everyone's
> > > > > >> > > > > > > > > > > feelings whether we should aim for symmetry
> > i.e.
> > > > > >> totally
> > > > > >> > > > > decouple
> > > > > >> > > > > > > group
> > > > > >> > > > > > > > > > > configs and hence not delete them at all
> when
> > the
> > > > > >> group
> > > > > >> > is
> > > > > >> > > > > gone,
> > > > > >> > > > > > > but
> > > > > >> > > > > > > > > always
> > > > > >> > > > > > > > > > > require explicit deletion operations by
> > themselves.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for
> > the Java
> > > > > >> > client
> > > > > >> > > > > still
> > > > > >> > > > > > > take
> > > > > >> > > > > > > > > a
> > > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > > > > >> introduce an
> > > > > >> > > > > overload?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > I think we do not need to introduce an
> > overload, but
> > > > > >> I'm
> > > > > >> > > all
> > > > > >> > > > > ears
> > > > > >> > > > > > > if
> > > > > >> > > > > > > > > there
> > > > > >> > > > > > > > > > > may be some compatibility issues that we may
> > > > > overlook.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 10. I agree with Justine that we should be
> > clearer
> > > > > >> > about
> > > > > >> > > the
> > > > > >> > > > > > > reason
> > > > > >> > > > > > > > > to
> > > > > >> > > > > > > > > > > > switch to IBP/metadata.version from the
> > feature
> > > > > >> flag.
> > > > > >> > > Maybe
> > > > > >> > > > > we
> > > > > >> > > > > > > mean
> > > > > >> > > > > > > > > that we
> > > > > >> > > > > > > > > > > > can switch the default for the feature
> flag
> > to
> > > > > true
> > > > > >> > > based on
> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > > metadata.version once we want to make it
> the
> > > > > >> default.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention
> > the
> > > > > >> required
> > > > > >> > > ACLs,
> > > > > >> > > > > it
> > > > > >> > > > > > > > > would be
> > > > > >> > > > > > > > > > > > good to add that for consistency.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 12. It is a bit odd that
> > ConsumerGroupHeartbeat
> > > > > >> > requires
> > > > > >> > > > > "Read
> > > > > >> > > > > > > > > Group" even
> > > > > >> > > > > > > > > > > > though it seems to do more than reading.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > I had that thought myself as well, but in
> the
> > end we
> > > > > >> > could
> > > > > >> > > not
> > > > > >> > > > > > > find a
> > > > > >> > > > > > > > > > > better alternative: adding Write Group seems
> > an
> > > > > >> overkill
> > > > > >> > > here
> > > > > >> > > > > > > since we
> > > > > >> > > > > > > > > do
> > > > > >> > > > > > > > > > > not have it elsewhere (we only have Read /
> > Delete
> > > > > and
> > > > > >> > > Describe
> > > > > >> > > > > on
> > > > > >> > > > > > > > > groups so
> > > > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > > > > consumer
> > > > > >> > with
> > > > > >> > > the
> > > > > >> > > > > new
> > > > > >> > > > > > > > > group
> > > > > >> > > > > > > > > > > > protocol? It would be good to have a
> > section on
> > > > > >> this.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > You mean with regex subscription right? Yes
> > we can
> > > > > >> add a
> > > > > >> > > > > section
> > > > > >> > > > > > > about
> > > > > >> > > > > > > > > > > that, but basically the idea is that
> consumer
> > would
> > > > > be
> > > > > >> > > totally
> > > > > >> > > > > > > > > agnostic in
> > > > > >> > > > > > > > > > > the new protocol as it's handled all by the
> > brokers.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > > > > >> coordinator
> > > > > >> > in
> > > > > >> > > > > Java.
> > > > > >> > > > > > > Even
> > > > > >> > > > > > > > > though
> > > > > >> > > > > > > > > > > > this is an implementation detail, do we
> > plan to
> > > > > >> have a
> > > > > >> > > new
> > > > > >> > > > > gradle
> > > > > >> > > > > > > > > module
> > > > > >> > > > > > > > > > > > for it?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > We have not thought about that. But I think
> > the
> > > > > answer
> > > > > >> > > should
> > > > > >> > > > > be
> > > > > >> > > > > > > yes.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it
> > comes to
> > > > > >> how
> > > > > >> > > many
> > > > > >> > > > > > > members
> > > > > >> > > > > > > > > the new
> > > > > >> > > > > > > > > > > > group protocol can support?
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > Within a group, I think we should shoot for
> > 1000s of
> > > > > >> > > members.
> > > > > >> > > > > But
> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > > > scalability goals also depend on the offset
> > > > > management
> > > > > >> > > (commit,
> > > > > >> > > > > > > fetch)
> > > > > >> > > > > > > > > > > capabilities of the coordinator which we did
> > not
> > > > > >> cover in
> > > > > >> > > this
> > > > > >> > > > > > > KIP, so
> > > > > >> > > > > > > > > it's
> > > > > >> > > > > > > > > > > hard to give a number that applies
> > universally.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > > 16. Did we consider having
> > SubscribedTopidIds
> > > > > >> instead
> > > > > >> > of
> > > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > > >> ConsumerGroupHeartbeatRequest?
> > > > > >> > > Is the
> > > > > >> > > > > > > idea
> > > > > >> > > > > > > > > that
> > > > > >> > > > > > > > > > > > since we have to resolve the regex on the
> > server,
> > > > > we
> > > > > >> > can
> > > > > >> > > do
> > > > > >> > > > > the
> > > > > >> > > > > > > same
> > > > > >> > > > > > > > > for
> > > > > >> > > > > > > > > > > > the topic name? The difference is that
> > sending the
> > > > > >> > regex
> > > > > >> > > is
> > > > > >> > > > > more
> > > > > >> > > > > > > > > efficient
> > > > > >> > > > > > > > > > > > whereas sending the topic names is less
> > efficient.
> > > > > >> > > > > Furthermore,
> > > > > >> > > > > > > > > delete and
> > > > > >> > > > > > > > > > > > recreation is easier to handle if we have
> > topic
> > > > > ids.
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > The main reason to still let the clients
> send
> > names
> > > > > >> is to
> > > > > >> > > keep
> > > > > >> > > > > the
> > > > > >> > > > > > > > > > > reasoning of names -> ids on the broker /
> > admin
> > > > > client
> > > > > >> > > only.
> > > > > >> > > > > Note
> > > > > >> > > > > > > that
> > > > > >> > > > > > > > > > > although we added topic id in KIP-516, we
> > never
> > > > > >> > > implemented the
> > > > > >> > > > > > > logic
> > > > > >> > > > > > > > > on
> > > > > >> > > > > > > > > > > consumer/producers leveraging the related
> > newer
> > > > > >> versioned
> > > > > >> > > RPCs,
> > > > > >> > > > > > > > > instead we
> > > > > >> > > > > > > > > > > just set the topic id as empty UUID. We want
> > to keep
> > > > > >> the
> > > > > >> > > > > > > > > consumer/producer
> > > > > >> > > > > > > > > > > to be thin and only delegate the reasoning
> on
> > broker
> > > > > >> and
> > > > > >> > > > > > > potentially
> > > > > >> > > > > > > > > admin
> > > > > >> > > > > > > > > > > clients.
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > Thanks,
> > > > > >> > > > > > > > > > > > Ismael
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David
> Jacot
> > > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > > >> > > > > > > > > > > > wrote:
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > > Hi all,
> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > > I would like to start a discussion
> thread
> > on
> > > > > >> KIP-848:
> > > > > >> > > The
> > > > > >> > > > > Next
> > > > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance
> > Protocol.
> > > > > >> With
> > > > > >> > > this
> > > > > >> > > > > KIP,
> > > > > >> > > > > > > we
> > > > > >> > > > > > > > > aim
> > > > > >> > > > > > > > > > > > > to make the rebalance protocol (for
> > consumers)
> > > > > >> more
> > > > > >> > > > > reliable,
> > > > > >> > > > > > > more
> > > > > >> > > > > > > > > > > > > scalable, easier to implement for
> > clients, and
> > > > > >> easier
> > > > > >> > > to
> > > > > >> > > > > debug
> > > > > >> > > > > > > for
> > > > > >> > > > > > > > > > > > > operators.
> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > > The KIP is here:
> > > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > > Please take a look and let me know what
> > you
> > > > > think.
> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > > Best,
> > > > > >> > > > > > > > > > > > > David
> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > > > PS: I will be away from July 18th to
> > August 8th.
> > > > > >> That
> > > > > >> > > gives
> > > > > >> > > > > > > you a
> > > > > >> > > > > > > > > bit
> > > > > >> > > > > > > > > > > > > of time to read and digest this long
> KIP.
> > > > > >> > > > > > > > > > > > >
> > > > > >> > > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > >
> > > > > >> > > > > > > > > > > --
> > > > > >> > > > > > > > > > > -- Guozhang
> > > > > >> > > > > > > > >
> > > > > >> > > > > > > >
> > > > > >> > > > > > > >
> > > > > >> > > > > > > > --
> > > > > >> > > > > > > > -- Guozhang
> > > > > >> > > > > > >
> > > > > >> > > > >
> > > > > >> > >
> > > > > >> > >
> > > > > >> >
> > > > > >>
> > > > > >>
> > > > > >> --
> > > > > >> -- Guozhang
> > > > > >>
> > > > > >
> > > > >
> >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Sagar <sa...@gmail.com>]

Thank you Guozhang/David for the feedback. Looks like there's agreement on
using separate APIs for Connect. I would revisit the doc and see what
changes are to be made.

Thanks!
Sagar.

On Tue, Aug 9, 2022 at 7:11 PM David Jacot <dj...@confluent.io.invalid>
wrote:

> Hi Sagar,
>
> Thanks for the feedback and the document. That's really helpful. I
> will take a look at it.
>
> Overall, it seems to me that both Connect and the Consumer could share
> the same underlying "engine". The main difference is that the Consumer
> assigns topic-partitions to members whereas Connect assigns tasks to
> workers. I see two ways to move forward:
> 1) We extend the new proposed APIs to support different resource types
> (e.g. partitions, tasks, etc.); or
> 2) We use new dedicated APIs for Connect. The dedicated APIs would be
> similar to the new ones but different on the content/resources and
> they would rely on the same engine on the coordinator side.
>
> I personally lean towards 2) because I am not a fan of overcharging
> APIs to serve different purposes. That being said, I am not opposed to
> 1) if we can find an elegant way to do it.
>
> I think that we can continue to discuss it here for now in order to
> ensure that this KIP is compatible with what we will do for Connect in
> the future.
>
> Best,
> David
>
> On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io> wrote:
> >
> > Hi all,
> >
> > I am back from vacation. I will go through and address your comments
> > in the coming days. Thanks for your feedback.
> >
> > Cheers,
> > David
> >
> > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gh...@gmail.com>
> wrote:
> > >
> > > Hey All!
> > >
> > > Thanks for the KIP, it's wonderful to see cooperative rebalancing
> making it
> > > down the stack!
> > >
> > > I had a few questions:
> > >
> > > 1. The 'Rejected Alternatives' section describes how member epoch
> should
> > > advance in step with the group epoch and assignment epoch values. I
> think
> > > that this is a good idea for the reasons described in the KIP. When the
> > > protocol is incrementally assigning partitions to a worker, what member
> > > epoch does each incremental assignment use? Are member epochs re-used,
> and
> > > a single member epoch can correspond to multiple different
> (monotonically
> > > larger) assignments?
> > >
> > > 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> > > not, should custom client-side assignor implementations interact with
> the
> > > Reason field, and how is its common meaning agreed upon? If so, what
> is the
> > > benefit of a distinct Reason field over including such functionality
> in the
> > > opaque metadata?
> > >
> > > 3. The following is included in the KIP: "Thanks to this, the input of
> the
> > > client side assignor is entirely driven by the group coordinator. The
> > > consumer is no longer responsible for maintaining any state besides its
> > > assigned partitions." Does this mean that the client-side assignor MAY
> > > incorporate additional non-Metadata state (such as partition
> throughput,
> > > cpu/memory metrics, config topics, etc), or that additional
> non-Metadata
> > > state SHOULD NOT be used?
> > >
> > > 4. I see that there are separate classes
> > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> > > overlap significantly. Is it possible for these two implementations to
> be
> > > unified? This would serve to promote feature parity of server-side and
> > > client-side assignors, and would also facilitate operational
> flexibility in
> > > certain situations. For example, if a server-side assignor has some
> poor
> > > behavior and needs a patch, deploying the patched assignor to the
> client
> > > and switching one consumer group to a client-side assignor may be
> faster
> > > and less risky than patching all of the brokers. With the currently
> > > proposed distinct APIs, a non-trivial reimplementation would have to be
> > > assembled, and if the two APIs have diverged significantly, then it is
> > > possible that a reimplementation would not be possible.
> > >
> > > --
> > > Greg Harris
> > > gharris1727@gmail.com
> > > github.com/gharris1727
> > >
> > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com>
> wrote:
> > >
> > > > Hi Guozhang/David,
> > > >
> > > > I created a confluence page to discuss how Connect would need to
> change
> > > > based on the new rebalance protocol. Here's the page:
> > > >
> > > >
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >
> > > > It's also pretty longish and I have tried to keep a format similar to
> > > > KIP-848. Let me know what you think. Also, do you think this should
> be
> > > > moved to a separate discussion thread or is this one fine?
> > > >
> > > > Thanks!
> > > > Sagar.
> > > >
> > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com>
> wrote:
> > > >
> > > > > Hello Guozhang,
> > > > >
> > > > > Thank you so much for the doc on Kafka Streams. Sure, I would do
> the
> > > > > analysis and come up with such a document.
> > > > >
> > > > > Thanks!
> > > > > Sagar.
> > > > >
> > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> > > > wrote:
> > > > >
> > > > >> Hello Sagar,
> > > > >>
> > > > >> It would be great if you could come back with some analysis on
> how to
> > > > >> implement the Connect side integration with the new protocol; so
> far
> > > > >> besides leveraging on the new "protocol type" we did not yet think
> > > > through
> > > > >> the Connect side implementations. For Streams here's a draft of
> > > > >> integration
> > > > >> plan:
> > > > >>
> > > > >>
> > > >
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > >> just FYI for your analysis on Connect.
> > > > >>
> > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sagarmeansocean@gmail.com
> >
> > > > wrote:
> > > > >>
> > > > >> > Hi David,
> > > > >> >
> > > > >> > Thank you for your response. The reason I thought connect can
> also fit
> > > > >> into
> > > > >> > this new scheme is that even today the connect uses a
> > > > WorkerCoordinator
> > > > >> > extending from AbstractCoordinator to empower rebalances of
> > > > >> > tasks/connectors. The WorkerCoordinator sets the protocolType()
> to
> > > > >> connect
> > > > >> > and uses the metadata() method by plumbing into
> > > > >> JoinGroupRequestProtocol.
> > > > >> >
> > > > >> > I think the changes to support connect would be similar at a
> high
> > > > level
> > > > >> to
> > > > >> > the changes in streams mainly because of the Client side
> assignors
> > > > being
> > > > >> > used in both. At an implementation level, we might need to make
> a lot
> > > > of
> > > > >> > changes to get onto this new assignment protocol like enhancing
> the
> > > > >> > JoinGroup request/response and SyncGroup and using
> > > > >> ConsumerGroupHeartbeat
> > > > >> > API etc again on similar lines to streams (or there might be
> > > > >> deviations). I
> > > > >> > would try to perform a detailed analysis of the same  and we
> can have
> > > > a
> > > > >> > separate discussion thread for that as that would derail this
> > > > discussion
> > > > >> > thread. Let me know if that sounds good to you.
> > > > >> >
> > > > >> > Thanks!
> > > > >> > Sagar.
> > > > >> >
> > > > >> >
> > > > >> >
> > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > <djacot@confluent.io.invalid
> > > > >> >
> > > > >> > wrote:
> > > > >> >
> > > > >> > > Hi Sagar,
> > > > >> > >
> > > > >> > > Thanks for your comments.
> > > > >> > >
> > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can
> mention it.
> > > > >> > >
> > > > >> > > 2) The idea is to transition C from his current assignment to
> his
> > > > >> > > target assignment when he can move to epoch 3. When that
> happens,
> > > > the
> > > > >> > > member assignment is updated and persisted with all its
> assigned
> > > > >> > > partitions even if they are not all revoked yet. In other
> words, the
> > > > >> > > member assignment becomes the target assignment. This is
> basically
> > > > an
> > > > >> > > optimization to avoid having to write all the changes to the
> log.
> > > > The
> > > > >> > > examples are based on the persisted state so I understand the
> > > > >> > > confusion. Let me see if I can improve this in the
> description.
> > > > >> > >
> > > > >> > > 3) Regarding Connect, it could reuse the protocol with a
> client side
> > > > >> > > assignor if it fits in the protocol. The assignment is about
> > > > >> > > topicid-partitions + metadata, could Connect fit into this?
> > > > >> > >
> > > > >> > > Best,
> > > > >> > > David
> > > > >> > >
> > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <
> sagarmeansocean@gmail.com>
> > > > >> wrote:
> > > > >> > > >
> > > > >> > > > Hi David,
> > > > >> > > >
> > > > >> > > > Thanks for the KIP. I just had minor observations:
> > > > >> > > >
> > > > >> > > > 1) In the Assignment Error section in Client Side mode
> Assignment
> > > > >> > > process,
> > > > >> > > > you mentioned => `In this case, the client side assignor can
> > > > return
> > > > >> an
> > > > >> > > > error to the group coordinator`. In this case are you
> referring to
> > > > >> the
> > > > >> > > > Assignor returning an AssignmentError that's listed down
> towards
> > > > the
> > > > >> > end?
> > > > >> > > > If yes, do you think it would make sense to mention this
> > > > explicitly
> > > > >> > here?
> > > > >> > > >
> > > > >> > > > 2) In the Case Studies section, I have a slight confusion,
> not
> > > > sure
> > > > >> if
> > > > >> > > > others have the same. Consider this step:
> > > > >> > > >
> > > > >> > > > When B heartbeats, the group coordinator transitions him to
> epoch
> > > > 3
> > > > >> > > because
> > > > >> > > > B has no partitions to revoke. It persists the change and
> reply.
> > > > >> > > >
> > > > >> > > >    - Group (epoch=3)
> > > > >> > > >       - A
> > > > >> > > >       - B
> > > > >> > > >       - C
> > > > >> > > >    - Target Assignment (epoch=3)
> > > > >> > > >       - A - partitions=[foo-0]
> > > > >> > > >       - B - partitions=[foo-2]
> > > > >> > > >       - C - partitions=[foo-1]
> > > > >> > > >    - Member Assignment
> > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > >> > > >
> > > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot get
> foo-1
> > > > >> yet.
> > > > >> > > >
> > > > >> > > > Here,it's mentioned that member C can't get the foo-1
> partition
> > > > yet,
> > > > >> > but
> > > > >> > > > based on the description above, it seems it already has it.
> Do you
> > > > >> > think
> > > > >> > > it
> > > > >> > > > would be better to remove it and populate it only when it
> actually
> > > > >> gets
> > > > >> > > it?
> > > > >> > > > I see this in a lot of other places, so have I understood it
> > > > >> > incorrectly
> > > > >> > > ?
> > > > >> > > >
> > > > >> > > >
> > > > >> > > > Regarding connect , it might be out of scope of this
> discussion,
> > > > but
> > > > >> > from
> > > > >> > > > what I understood it would probably be running in client
> side
> > > > >> assignor
> > > > >> > > mode
> > > > >> > > > even on the new rebalance protocol as it has its own Custom
> > > > >> > > Assignors(Eager
> > > > >> > > > and IncrementalCooperative).
> > > > >> > > >
> > > > >> > > > Thanks!
> > > > >> > > >
> > > > >> > > > Sagar.
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > > >> > <djacot@confluent.io.invalid
> > > > >> > > >
> > > > >> > > > wrote:
> > > > >> > > >
> > > > >> > > > > Thanks Hector! Our goal is to move forward with
> specialized API
> > > > >> > > > > instead of relying on one generic API. For Connect, we
> can apply
> > > > >> the
> > > > >> > > > > exact same pattern and reuse/share the core
> implementation on
> > > > the
> > > > >> > > > > server side. For the schema registry, I think that we
> should
> > > > >> consider
> > > > >> > > > > having a tailored API to do simple membership/leader
> election.
> > > > >> > > > >
> > > > >> > > > > Best,
> > > > >> > > > > David
> > > > >> > > > >
> > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <
> ismael@juma.me.uk
> > > > >
> > > > >> > > wrote:
> > > > >> > > > > >
> > > > >> > > > > > Three quick comments:
> > > > >> > > > > >
> > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > > >> com.google.re2j.Pattern, we
> > > > >> > > > > should
> > > > >> > > > > > document the differences in more detail before deciding
> one
> > > > way
> > > > >> or
> > > > >> > > > > another.
> > > > >> > > > > > That said, if people pass java.util.regex.Pattern, they
> expect
> > > > >> > their
> > > > >> > > > > > semantics to be honored. If we are doing something
> different,
> > > > >> then
> > > > >> > we
> > > > >> > > > > > should consider adding an overload with our own Pattern
> class
> > > > (I
> > > > >> > > don't
> > > > >> > > > > > think we'd want to expose re2j's at this point).
> > > > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > > > integrate
> > > > >> > fully
> > > > >> > > > > with
> > > > >> > > > > > it and should handle the topic recreation case
> correctly.
> > > > That's
> > > > >> > the
> > > > >> > > main
> > > > >> > > > > > part we need to handle. I agree with David that we'd
> want to
> > > > add
> > > > >> > > topic
> > > > >> > > > > ids
> > > > >> > > > > > to the relevant protocols that don't have it yet and
> that we
> > > > can
> > > > >> > > probably
> > > > >> > > > > > focus on the internals versus adding new APIs to the
> Java
> > > > >> Consumer
> > > > >> > > > > (unless
> > > > >> > > > > > we find that adding new APIs is required for reasonable
> > > > >> semantics).
> > > > >> > > > > > 3. I am still not sure about the coordinator storing the
> > > > >> configs.
> > > > >> > > It's
> > > > >> > > > > > powerful for configs to be centralized in the metadata
> log for
> > > > >> > > various
> > > > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > > > >> Similarly, I
> > > > >> > > am
> > > > >> > > > > not
> > > > >> > > > > > sure about automatically deleting configs in a way that
> they
> > > > >> cannot
> > > > >> > > be
> > > > >> > > > > > recovered. A good property for modern systems is to
> minimize
> > > > the
> > > > >> > > number
> > > > >> > > > > of
> > > > >> > > > > > unrecoverable data loss scenarios.
> > > > >> > > > > >
> > > > >> > > > > > Ismael
> > > > >> > > > > >
> > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > > >> > > <djacot@confluent.io.invalid
> > > > >> > > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > >
> > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > > >> > > > > > >
> > > > >> > > > > > > > 1) the migration path, especially the last step when
> > > > clients
> > > > >> > > flip the
> > > > >> > > > > > > flag
> > > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > > window
> > > > >> > where
> > > > >> > > > > both
> > > > >> > > > > > > new
> > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used
> by
> > > > >> members
> > > > >> > of
> > > > >> > > the
> > > > >> > > > > same
> > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> behavior
> > > > >> while
> > > > >> > > > > using the
> > > > >> > > > > > > > new protocol is something we need to present about.
> > > > >> > > > > > >
> > > > >> > > > > > > Noted. I just published a new version of KIP which
> includes
> > > > >> more
> > > > >> > > > > > > details about this. See the "Supporting Online
> Consumer
> > > > Group
> > > > >> > > Upgrade"
> > > > >> > > > > > > and the "Compatibility, Deprecation, and Migration
> Plan". I
> > > > >> think
> > > > >> > > that
> > > > >> > > > > > > I have to think through a few cases now but the
> overall idea
> > > > >> and
> > > > >> > > > > > > mechanism should be understandable.
> > > > >> > > > > > >
> > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> topic ids
> > > > >> are
> > > > >> > > only
> > > > >> > > > > used
> > > > >> > > > > > > as
> > > > >> > > > > > > > part of RPCs and admin client, but they are not
> exposed
> > > > via
> > > > >> any
> > > > >> > > > > public
> > > > >> > > > > > > APIs
> > > > >> > > > > > > > to consumers yet. I think the question is, first
> should we
> > > > >> let
> > > > >> > > the
> > > > >> > > > > > > consumer
> > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> itself
> > > > to
> > > > >> > fully
> > > > >> > > > > > > leverage
> > > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs
> with
> > > > the
> > > > >> > > topic
> > > > >> > > > > ids;
> > > > >> > > > > > > and
> > > > >> > > > > > > > secondly, should we ever consider exposing the
> topic ids
> > > > in
> > > > >> the
> > > > >> > > > > consumer
> > > > >> > > > > > > > public APIs as well (both subscribe/assign, as well
> as in
> > > > >> the
> > > > >> > > > > rebalance
> > > > >> > > > > > > > listener for cases like topic
> deletion-and-recreation).
> > > > >> > > > > > >
> > > > >> > > > > > > a) Assuming that we would include converting all the
> offsets
> > > > >> > > related
> > > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer
> would be
> > > > >> able
> > > > >> > to
> > > > >> > > > > > > fully operate with topic ids. That being said, it
> still has
> > > > to
> > > > >> > > provide
> > > > >> > > > > > > the topics names in various APIs so having a mapping
> in the
> > > > >> > > consumer
> > > > >> > > > > > > seems inevitable to me.
> > > > >> > > > > > > b) I don't have a strong opinion on this. Here I
> wonder if
> > > > >> this
> > > > >> > > goes
> > > > >> > > > > > > beyond the scope of this KIP. I would rather focus on
> the
> > > > >> > internals
> > > > >> > > > > > > here and we can consider this separately if we see
> value in
> > > > >> doing
> > > > >> > > it.
> > > > >> > > > > > >
> > > > >> > > > > > > Coming back to Ismael's point about using topic ids
> in the
> > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is
> one
> > > > >> > advantage
> > > > >> > > in
> > > > >> > > > > > > favour of it. The consumer will have the opportunity
> to
> > > > >> validate
> > > > >> > > that
> > > > >> > > > > > > the topics exists before passing them into the group
> > > > rebalance
> > > > >> > > > > > > protocol. Obviously, the coordinator will also notice
> it but
> > > > >> it
> > > > >> > > does
> > > > >> > > > > > > not really have a way to reject an invalid topic in
> the
> > > > >> response.
> > > > >> > > > > > >
> > > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > > except
> > > > >> for
> > > > >> > > the
> > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> it's not
> > > > >> > > necessary
> > > > >> > > > > to
> > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> instead we
> > > > >> still
> > > > >> > > use
> > > > >> > > > > > > Pattern
> > > > >> > > > > > > > while just documenting that our subscription may be
> > > > >> rejected by
> > > > >> > > the
> > > > >> > > > > > > server.
> > > > >> > > > > > > > Since the incompatible case is a very rare scenario
> I felt
> > > > >> > using
> > > > >> > > an
> > > > >> > > > > > > > overloaded `String` based subscription may be more
> > > > >> vulnerable
> > > > >> > to
> > > > >> > > > > various
> > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > >
> > > > >> > > > > > > That could work. I have to look at the differences
> between
> > > > the
> > > > >> > two
> > > > >> > > > > > > engines to better understand the potential issues. My
> > > > >> > > understanding is
> > > > >> > > > > > > that would work for all the basic regular
> expressions. The
> > > > >> > > differences
> > > > >> > > > > > > between the two are mainly about the various character
> > > > >> classes. I
> > > > >> > > > > > > wonder what other people think about this.
> > > > >> > > > > > >
> > > > >> > > > > > > Best,
> > > > >> > > > > > > David
> > > > >> > > > > > >
> > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > > > >> > wangguoz@gmail.com
> > > > >> > > >
> > > > >> > > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks David! I think on the high level there are
> two meta
> > > > >> > > points we
> > > > >> > > > > need
> > > > >> > > > > > > > to concretize a bit more:
> > > > >> > > > > > > >
> > > > >> > > > > > > > 1) the migration path, especially the last step when
> > > > clients
> > > > >> > > flip the
> > > > >> > > > > > > flag
> > > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > > window
> > > > >> > where
> > > > >> > > > > both
> > > > >> > > > > > > new
> > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used
> by
> > > > >> members
> > > > >> > of
> > > > >> > > the
> > > > >> > > > > same
> > > > >> > > > > > > > group. How the coordinator could "mimic" the old
> behavior
> > > > >> while
> > > > >> > > > > using the
> > > > >> > > > > > > > new protocol is something we need to present about.
> > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the
> topic ids
> > > > >> are
> > > > >> > > only
> > > > >> > > > > used
> > > > >> > > > > > > as
> > > > >> > > > > > > > part of RPCs and admin client, but they are not
> exposed
> > > > via
> > > > >> any
> > > > >> > > > > public
> > > > >> > > > > > > APIs
> > > > >> > > > > > > > to consumers yet. I think the question is, first
> should we
> > > > >> let
> > > > >> > > the
> > > > >> > > > > > > consumer
> > > > >> > > > > > > > client to be maintaining the names -> ids mapping
> itself
> > > > to
> > > > >> > fully
> > > > >> > > > > > > leverage
> > > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs
> with
> > > > the
> > > > >> > > topic
> > > > >> > > > > ids;
> > > > >> > > > > > > and
> > > > >> > > > > > > > secondly, should we ever consider exposing the
> topic ids
> > > > in
> > > > >> the
> > > > >> > > > > consumer
> > > > >> > > > > > > > public APIs as well (both subscribe/assign, as well
> as in
> > > > >> the
> > > > >> > > > > rebalance
> > > > >> > > > > > > > listener for cases like topic
> deletion-and-recreation).
> > > > >> > > > > > > >
> > > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > > except
> > > > >> for
> > > > >> > > the
> > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think
> it's not
> > > > >> > > necessary
> > > > >> > > > > to
> > > > >> > > > > > > > deprecate the subscribe API with Pattern, but
> instead we
> > > > >> still
> > > > >> > > use
> > > > >> > > > > > > Pattern
> > > > >> > > > > > > > while just documenting that our subscription may be
> > > > >> rejected by
> > > > >> > > the
> > > > >> > > > > > > server.
> > > > >> > > > > > > > Since the incompatible case is a very rare scenario
> I felt
> > > > >> > using
> > > > >> > > an
> > > > >> > > > > > > > overloaded `String` based subscription may be more
> > > > >> vulnerable
> > > > >> > to
> > > > >> > > > > various
> > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > > >
> > > > >> > > > > > > >
> > > > >> > > > > > > > Guozhang
> > > > >> > > > > > > >
> > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > > >> > > > > <djacot@confluent.io.invalid
> > > > >> > > > > > > >
> > > > >> > > > > > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Thanks for your feedback. Let me answer your
> questions
> > > > >> > inline.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 1. I think it's premature to talk about target
> > > > versions
> > > > >> for
> > > > >> > > > > > > deprecation
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > removal of the existing group protocol. Unlike
> KRaft,
> > > > >> this
> > > > >> > > > > affects a
> > > > >> > > > > > > core
> > > > >> > > > > > > > > > client protocol and hence deprecation/removal
> will be
> > > > >> > heavily
> > > > >> > > > > > > dependent
> > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > how quickly applications migrate to the new
> protocol.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > That makes sense. I will remove it.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 2. The KIP says we intend to release this in
> 4.x, but
> > > > it
> > > > >> > > wasn't
> > > > >> > > > > made
> > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > why. If we added that as a way to estimate when
> we'd
> > > > >> > > deprecate
> > > > >> > > > > and
> > > > >> > > > > > > remove
> > > > >> > > > > > > > > > the group protocol, I also suggest removing
> this part.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Let me explain my reasoning. As explained, I plan
> to
> > > > >> rewrite
> > > > >> > > the
> > > > >> > > > > group
> > > > >> > > > > > > > > coordinator in Java while we implement the new
> protocol.
> > > > >> This
> > > > >> > > means
> > > > >> > > > > > > > > that the internals will be slightly different
> (e.g.
> > > > >> threading
> > > > >> > > > > model).
> > > > >> > > > > > > > > Therefore, I wanted to tighten the switch from
> the old
> > > > >> group
> > > > >> > > > > > > > > coordinator to the new group coordinator to a
> major
> > > > >> release.
> > > > >> > > The
> > > > >> > > > > > > > > alternative would be to use a flag to do the
> switch
> > > > >> instead
> > > > >> > of
> > > > >> > > > > relying
> > > > >> > > > > > > > > on the software upgrade.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 3. We need to flesh out the details of the
> migration
> > > > >> story.
> > > > >> > > It
> > > > >> > > > > sounds
> > > > >> > > > > > > > > like
> > > > >> > > > > > > > > > we're saying we will support online migrations.
> Is
> > > > that
> > > > >> > > correct?
> > > > >> > > > > We
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > explain this in detail. It could also be done
> as a
> > > > >> separate
> > > > >> > > KIP,
> > > > >> > > > > if
> > > > >> > > > > > > it's
> > > > >> > > > > > > > > > easier.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Yes, we will support online migrations for the
> group.
> > > > That
> > > > >> > > means
> > > > >> > > > > that
> > > > >> > > > > > > > > a group using the old protocol will be able to
> switch to
> > > > >> the
> > > > >> > > new
> > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Let me briefly explain how that will work though.
> It is
> > > > >> > > basically a
> > > > >> > > > > > > > > four step process:
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a
> software
> > > > >> > > supporting
> > > > >> > > > > the
> > > > >> > > > > > > > > new group coordinator. Both the old and the new
> > > > >> coordinator
> > > > >> > > will
> > > > >> > > > > > > > > support the old protocol and rely on the same
> persisted
> > > > >> > > metadata so
> > > > >> > > > > > > > > they can work together. This point is an offline
> > > > >> migration.
> > > > >> > We
> > > > >> > > > > cannot
> > > > >> > > > > > > > > do this one live because it would require
> shutting down
> > > > >> the
> > > > >> > > current
> > > > >> > > > > > > > > coordinator and starting up the new one and that
> would
> > > > >> cause
> > > > >> > > > > > > > > unavailabilities.
> > > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be
> upgraded
> > > > to
> > > > >> X
> > > > >> > in
> > > > >> > > > > order
> > > > >> > > > > > > > > to enable the new protocol. This cannot be done
> before
> > > > 1)
> > > > >> is
> > > > >> > > > > > > > > terminated because the old coordinator doesn't
> support
> > > > the
> > > > >> > new
> > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > > > supporting
> > > > >> the
> > > > >> > > > > online
> > > > >> > > > > > > > > migration (must have KIP-792). If the consumer is
> > > > already
> > > > >> > > there.
> > > > >> > > > > > > > > Nothing must be done at this point.
> > > > >> > > > > > > > > 4. The consumers must be rolled with the feature
> flag
> > > > >> turned
> > > > >> > > on.
> > > > >> > > > > The
> > > > >> > > > > > > > > consumer group is automatically converted when
> the first
> > > > >> > > consumer
> > > > >> > > > > > > > > using the new protocol joins the group. While the
> > > > members
> > > > >> > > using the
> > > > >> > > > > > > > > old protocol are being upgraded, the old protocol
> is
> > > > >> proxied
> > > > >> > > into
> > > > >> > > > > the
> > > > >> > > > > > > > > new one.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > > >> subscriptions
> > > > >> > > to
> > > > >> > > > > the
> > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > but it seems like there could be some tricky
> > > > >> compatibility
> > > > >> > > > > issues.
> > > > >> > > > > > > Will
> > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > have a mechanism for users to detect that they
> need to
> > > > >> > update
> > > > >> > > > > their
> > > > >> > > > > > > regex
> > > > >> > > > > > > > > > before switching to the new protocol?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I think that I am a bit more optimistic than you
> on this
> > > > >> > > point. I
> > > > >> > > > > > > > > believe that the majority of the cases are simple
> > > > regexes
> > > > >> > which
> > > > >> > > > > should
> > > > >> > > > > > > > > work with the new engine. The coordinator will
> verify
> > > > the
> > > > >> > regex
> > > > >> > > > > anyway
> > > > >> > > > > > > > > and reject the consumer if the regex is not valid.
> > > > Coming
> > > > >> > back
> > > > >> > > to
> > > > >> > > > > the
> > > > >> > > > > > > > > migration path, in the worst case, the first
> upgraded
> > > > >> > consumer
> > > > >> > > > > joining
> > > > >> > > > > > > > > the group will be rejected. This should be used
> as the
> > > > >> last
> > > > >> > > > > defence, I
> > > > >> > > > > > > > > would say.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > One way for customers to validate their regex
> before
> > > > >> > upgrading
> > > > >> > > > > their
> > > > >> > > > > > > > > prod would be to test them with another group. For
> > > > >> instance,
> > > > >> > > that
> > > > >> > > > > > > > > could be done in a pre-prod environment. Another
> way
> > > > >> would be
> > > > >> > > to
> > > > >> > > > > > > > > extend the consumer-group tool to provide a regex
> > > > >> validation
> > > > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 5. Related to the last question, will the Java
> client
> > > > >> allow
> > > > >> > > the
> > > > >> > > > > > > users to
> > > > >> > > > > > > > > > stick with the current regex engine for
> compatibility
> > > > >> > > reasons?
> > > > >> > > > > For
> > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > it may be handy to keep using client based
> regex at
> > > > >> first
> > > > >> > to
> > > > >> > > keep
> > > > >> > > > > > > > > > migrations simple and then migrate to server
> based
> > > > >> regexes
> > > > >> > > as a
> > > > >> > > > > > > second
> > > > >> > > > > > > > > step.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I understand your point but I am concerned that
> this
> > > > would
> > > > >> > > allow
> > > > >> > > > > users
> > > > >> > > > > > > > > to actually stay in this mode. That would go
> against our
> > > > >> goal
> > > > >> > > of
> > > > >> > > > > > > > > simplifying the client because we would have to
> continue
> > > > >> > > monitoring
> > > > >> > > > > > > > > the metadata on the client side. I would rather
> not do
> > > > >> this.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 6. When we say that the group coordinator will
> be
> > > > >> > > responsible for
> > > > >> > > > > > > storing
> > > > >> > > > > > > > > > the configurations and that the configurations
> will be
> > > > >> > > deleted
> > > > >> > > > > when
> > > > >> > > > > > > the
> > > > >> > > > > > > > > > group is deleted. Will a transition to DEAD
> trigger
> > > > >> > deletion
> > > > >> > > of
> > > > >> > > > > > > > > > configurations?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > That's right. The configurations will be deleted
> when
> > > > the
> > > > >> > > group is
> > > > >> > > > > > > > > deleted. They go together.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 7. Will the choice to store the configs in the
> group
> > > > >> > > coordinator
> > > > >> > > > > > > make it
> > > > >> > > > > > > > > > harder to list all cluster configs and their
> values?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I don't think so. The group configurations are
> overrides
> > > > >> of
> > > > >> > > cluster
> > > > >> > > > > > > > > configs. If you want to know all the overrides
> though,
> > > > you
> > > > >> > > would
> > > > >> > > > > have
> > > > >> > > > > > > > > to ask all the group coordinators. You cannot
> rely on
> > > > the
> > > > >> > > metadata
> > > > >> > > > > log
> > > > >> > > > > > > > > for instance.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 8. How would someone configure a group before
> starting
> > > > >> the
> > > > >> > > > > consumers?
> > > > >> > > > > > > > > Have
> > > > >> > > > > > > > > > we considered allowing the explicit creation of
> > > > groups?
> > > > >> > > > > > > Alternatively,
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > configs could be decoupled from the group
> lifecycle.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Yes. The group will be automatically created in
> this
> > > > case.
> > > > >> > > However,
> > > > >> > > > > > > > > the configs will be lost after the retention
> period of
> > > > the
> > > > >> > > group
> > > > >> > > > > > > > > passes.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the
> Java
> > > > >> client
> > > > >> > > still
> > > > >> > > > > take
> > > > >> > > > > > > a
> > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to
> introduce
> > > > an
> > > > >> > > overload?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > That's a very group question. I forgot about that
> one.
> > > > As
> > > > >> the
> > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible
> with
> > > > the
> > > > >> > > engine
> > > > >> > > > > that
> > > > >> > > > > > > > > we plan to use, it might be better to deprecate
> it and
> > > > >> use an
> > > > >> > > > > overload
> > > > >> > > > > > > > > which takes a string. We would rely on the server
> side
> > > > >> > > validation.
> > > > >> > > > > > > > > During the migration, I think that we could still
> try to
> > > > >> > > toString
> > > > >> > > > > the
> > > > >> > > > > > > > > regex and use it. That should work, I think, in
> the
> > > > >> majority
> > > > >> > > of the
> > > > >> > > > > > > > > cases.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 10. I agree with Justine that we should be
> clearer
> > > > about
> > > > >> > the
> > > > >> > > > > reason
> > > > >> > > > > > > to
> > > > >> > > > > > > > > > switch to IBP/metadata.version from the feature
> flag.
> > > > >> Maybe
> > > > >> > > we
> > > > >> > > > > mean
> > > > >> > > > > > > that
> > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > can switch the default for the feature flag to
> true
> > > > >> based
> > > > >> > on
> > > > >> > > the
> > > > >> > > > > > > > > > metadata.version once we want to make it the
> default.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > My plan was to use that feature flag mainly
> during the
> > > > >> > > development
> > > > >> > > > > > > > > phase. I should not have mentioned it, I think,
> because
> > > > we
> > > > >> > > could
> > > > >> > > > > use
> > > > >> > > > > > > > > an internal config for it.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > > > required
> > > > >> > > ACLs, it
> > > > >> > > > > > > would
> > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > good to add that for consistency.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Noted.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > > > requires
> > > > >> > > "Read
> > > > >> > > > > Group"
> > > > >> > > > > > > > > even
> > > > >> > > > > > > > > > though it seems to do more than reading.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I agree. This is how the current protocol works
> though.
> > > > We
> > > > >> > only
> > > > >> > > > > > > > > require "Read Group" to join a group. We could
> consider
> > > > >> > > changing
> > > > >> > > > > this
> > > > >> > > > > > > > > but I am not sure that it is worth it.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 13. How is topic recreation handled by the
> consumer
> > > > with
> > > > >> > the
> > > > >> > > new
> > > > >> > > > > > > group
> > > > >> > > > > > > > > > protocol? It would be good to have a section on
> this.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Noted. From a protocol perspective, the new topic
> will
> > > > >> have a
> > > > >> > > new
> > > > >> > > > > > > > > topic id so it will treat it like a topic with a
> > > > different
> > > > >> > > name.
> > > > >> > > > > The
> > > > >> > > > > > > > > only issue is that the fetch/commit offsets APIs
> do not
> > > > >> > support
> > > > >> > > > > topic
> > > > >> > > > > > > > > IDs so the consumer would reuse the offsets based
> on the
> > > > >> > same.
> > > > >> > > I
> > > > >> > > > > think
> > > > >> > > > > > > > > that we should update those APIs as well in order
> to be
> > > > >> > > consistent
> > > > >> > > > > end
> > > > >> > > > > > > > > to end. That would strengthen the semantics of the
> > > > >> consumer.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 14. The KIP mentions we will write the new
> coordinator
> > > > >> in
> > > > >> > > Java.
> > > > >> > > > > Even
> > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > this is an implementation detail, do we plan to
> have a
> > > > >> new
> > > > >> > > gradle
> > > > >> > > > > > > module
> > > > >> > > > > > > > > > for it?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Yes.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 15. Do we have a scalability goal when it comes
> to how
> > > > >> many
> > > > >> > > > > members
> > > > >> > > > > > > the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > group protocol can support?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > We don't have numbers at the moment. The protocol
> should
> > > > >> > > support
> > > > >> > > > > 1000s
> > > > >> > > > > > > > > of members per group. We will measure this when
> we have
> > > > a
> > > > >> > first
> > > > >> > > > > > > > > implementation. Note that we might have other
> > > > bottlenecks
> > > > >> > down
> > > > >> > > the
> > > > >> > > > > > > > > road (e.g. offset commits).
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> instead
> > > > of
> > > > >> > > > > > > > > > SubscribedTopicNames in
> ConsumerGroupHeartbeatRequest?
> > > > >> Is
> > > > >> > the
> > > > >> > > > > idea
> > > > >> > > > > > > that
> > > > >> > > > > > > > > > since we have to resolve the regex on the
> server, we
> > > > >> can do
> > > > >> > > the
> > > > >> > > > > same
> > > > >> > > > > > > for
> > > > >> > > > > > > > > > the topic name? The difference is that sending
> the
> > > > >> regex is
> > > > >> > > more
> > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > whereas sending the topic names is less
> efficient.
> > > > >> > > Furthermore,
> > > > >> > > > > > > delete
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > recreation is easier to handle if we have topic
> ids.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > The idea was to consolidate the metadata lookup
> on the
> > > > >> server
> > > > >> > > for
> > > > >> > > > > both
> > > > >> > > > > > > > > paths but I do agree with your point. As a second
> > > > though,
> > > > >> > using
> > > > >> > > > > topic
> > > > >> > > > > > > > > ids may be better here for the delete and
> recreation
> > > > case.
> > > > >> > > Also, I
> > > > >> > > > > > > > > suppose that we may allow users to subscribe with
> topic
> > > > >> ids
> > > > >> > in
> > > > >> > > the
> > > > >> > > > > > > > > future because that is the only way to be really
> robust
> > > > to
> > > > >> > > topic
> > > > >> > > > > > > > > re-creation.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Best,
> > > > >> > > > > > > > > David
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > > > >> > > djacot@confluent.io>
> > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Hi Justine,
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Thanks for your comments. Please find my answers
> > > > below.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs
> with the
> > > > >> > > exception
> > > > >> > > > > of the
> > > > >> > > > > > > > > > topic names based in the
> > > > ConsumerGroupHeartbeatRequest.
> > > > >> I
> > > > >> > am
> > > > >> > > not
> > > > >> > > > > sure
> > > > >> > > > > > > > > > if using topic names is the right call here. I
> need to
> > > > >> > think
> > > > >> > > > > about it
> > > > >> > > > > > > > > > a little more. Obviously, the KIP does not
> change the
> > > > >> > > > > fetch/commit
> > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be
> something
> > > > >> that
> > > > >> > we
> > > > >> > > > > should
> > > > >> > > > > > > > > > include though as it would give better overall
> > > > >> guarantee in
> > > > >> > > the
> > > > >> > > > > > > > > > producer.
> > > > >> > > > > > > > > > - You're right. I think that I should not have
> > > > mentioned
> > > > >> > this
> > > > >> > > > > flag at
> > > > >> > > > > > > > > > all. I will remove it. We can use an internal
> > > > >> configuration
> > > > >> > > while
> > > > >> > > > > > > > > > developing the feature.
> > > > >> > > > > > > > > > - Both cluster types will be supported. The
> change is
> > > > >> > > > > orthogonal. The
> > > > >> > > > > > > > > > only requirement is that the cluster uses topic
> IDs.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Best,
> > > > >> > > > > > > > > > David
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> > > > >> > > > > wangguoz@gmail.com>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> > > > inlined
> > > > >> > > below:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > > > >> > > ismael@juma.me.uk>
> > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > Thanks for the KIP. This has the potential
> to be a
> > > > >> > great
> > > > >> > > > > > > > > improvement. A few
> > > > >> > > > > > > > > > > > initial questions/comments:
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > 1. I think it's premature to talk about
> target
> > > > >> versions
> > > > >> > > for
> > > > >> > > > > > > > > deprecation and
> > > > >> > > > > > > > > > > > removal of the existing group protocol.
> Unlike
> > > > >> KRaft,
> > > > >> > > this
> > > > >> > > > > > > affects a
> > > > >> > > > > > > > > core
> > > > >> > > > > > > > > > > > client protocol and hence
> deprecation/removal will
> > > > >> be
> > > > >> > > heavily
> > > > >> > > > > > > > > dependent on
> > > > >> > > > > > > > > > > > how quickly applications migrate to the new
> > > > >> protocol.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yeah I agree with you. I think we can remove
> the
> > > > >> proposed
> > > > >> > > > > timeline
> > > > >> > > > > > > in
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration
> Plan` and
> > > > >> > > instead
> > > > >> > > > > just
> > > > >> > > > > > > state
> > > > >> > > > > > > > > > > that we will decide in the future about when
> we
> > > > would
> > > > >> > > > > deprecate old
> > > > >> > > > > > > > > > > protocol and behaviors.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 2. The KIP says we intend to release this
> in 4.x,
> > > > >> but
> > > > >> > it
> > > > >> > > > > wasn't
> > > > >> > > > > > > made
> > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > > > why. If we added that as a way to estimate
> when
> > > > we'd
> > > > >> > > > > deprecate
> > > > >> > > > > > > and
> > > > >> > > > > > > > > remove
> > > > >> > > > > > > > > > > > the group protocol, I also suggest removing
> this
> > > > >> part.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > I think that's not specifically related to the
> > > > >> > > > > deprecation/removal
> > > > >> > > > > > > > > timeline
> > > > >> > > > > > > > > > > plan, but it's more for client upgrades. I.e.
> the
> > > > >> > > broker-side
> > > > >> > > > > > > > > > > implementation may be done first, and then the
> > > > client
> > > > >> > side,
> > > > >> > > > > and we
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > only mark it as "released" by the time clients
> > > > >> > > implementations
> > > > >> > > > > are
> > > > >> > > > > > > > > done. At
> > > > >> > > > > > > > > > > that time, to enable the feature the clients
> need to
> > > > >> > first
> > > > >> > > > > swap-in
> > > > >> > > > > > > the
> > > > >> > > > > > > > > > > bytecode with a rolling bounce and then set
> the flag
> > > > >> > with a
> > > > >> > > > > second
> > > > >> > > > > > > > > rolling
> > > > >> > > > > > > > > > > bounce, and hence we feel it's better to be
> released
> > > > >> in a
> > > > >> > > major
> > > > >> > > > > > > > > version.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > > > migration
> > > > >> > > story.
> > > > >> > > > > It
> > > > >> > > > > > > > > sounds like
> > > > >> > > > > > > > > > > > we're saying we will support online
> migrations. Is
> > > > >> that
> > > > >> > > > > correct?
> > > > >> > > > > > > We
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > > explain this in detail. It could also be
> done as a
> > > > >> > > separate
> > > > >> > > > > KIP,
> > > > >> > > > > > > if
> > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > > easier.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yes I think that's the part we can be more
> concrete
> > > > >> about
> > > > >> > > for
> > > > >> > > > > sure
> > > > >> > > > > > > (and
> > > > >> > > > > > > > > > > this is related to your question 2) above).
> We will
> > > > >> work
> > > > >> > on
> > > > >> > > > > making
> > > > >> > > > > > > it
> > > > >> > > > > > > > > more
> > > > >> > > > > > > > > > > explicit in parallel as we solicit more
> feedback.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 4. I am happy that we are pushing the
> pattern
> > > > >> > > subscriptions
> > > > >> > > > > to
> > > > >> > > > > > > the
> > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > > > but it seems like there could be some tricky
> > > > >> > > compatibility
> > > > >> > > > > > > issues.
> > > > >> > > > > > > > > Will we
> > > > >> > > > > > > > > > > > have a mechanism for users to detect that
> they
> > > > need
> > > > >> to
> > > > >> > > update
> > > > >> > > > > > > their
> > > > >> > > > > > > > > regex
> > > > >> > > > > > > > > > > > before switching to the new protocol?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yes I think we need some tooling for non-java
> client
> > > > >> > users
> > > > >> > > to
> > > > >> > > > > sort
> > > > >> > > > > > > of
> > > > >> > > > > > > > > > > "dry-run" the client before switching to the
> new
> > > > >> > protocol.
> > > > >> > > I
> > > > >> > > > > do not
> > > > >> > > > > > > > > have a
> > > > >> > > > > > > > > > > specific idea on top of my head though, maybe
> others
> > > > >> like
> > > > >> > > @Matt
> > > > >> > > > > > > > > Howlett can
> > > > >> > > > > > > > > > > chime-in here?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 5. Related to the last question, will the
> Java
> > > > >> client
> > > > >> > > allow
> > > > >> > > > > the
> > > > >> > > > > > > > > users to
> > > > >> > > > > > > > > > > > stick with the current regex engine for
> > > > >> compatibility
> > > > >> > > > > reasons?
> > > > >> > > > > > > For
> > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > > > it may be handy to keep using client based
> regex
> > > > at
> > > > >> > > first to
> > > > >> > > > > keep
> > > > >> > > > > > > > > > > > migrations simple and then migrate to
> server based
> > > > >> > > regexes
> > > > >> > > > > as a
> > > > >> > > > > > > > > second
> > > > >> > > > > > > > > > > > step.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Honestly I have not thought about that for
> java
> > > > >> clients,
> > > > >> > > and
> > > > >> > > > > we can
> > > > >> > > > > > > > > discuss
> > > > >> > > > > > > > > > > that. What kind of compatibility issues do
> you have
> > > > in
> > > > >> > > mind?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 6. When we say that the group coordinator
> will be
> > > > >> > > > > responsible for
> > > > >> > > > > > > > > storing
> > > > >> > > > > > > > > > > > the configurations and that the
> configurations
> > > > will
> > > > >> be
> > > > >> > > > > deleted
> > > > >> > > > > > > when
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> > > > trigger
> > > > >> > > deletion
> > > > >> > > > > of
> > > > >> > > > > > > > > > > > configurations?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yes, since the DEAD state is an ending state
> (we
> > > > would
> > > > >> > only
> > > > >> > > > > > > transit to
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > state when the group is EMPTY and also all of
> its
> > > > >> > metadata
> > > > >> > > are
> > > > >> > > > > > > gone),
> > > > >> > > > > > > > > once
> > > > >> > > > > > > > > > > it's transited to DEAD this group would never
> be
> > > > >> revived.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 7. Will the choice to store the configs in
> the
> > > > group
> > > > >> > > > > coordinator
> > > > >> > > > > > > > > make it
> > > > >> > > > > > > > > > > > harder to list all cluster configs and their
> > > > values?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > That's a good question, and our thoughts are
> that
> > > > the
> > > > >> > > so-called
> > > > >> > > > > > > "group
> > > > >> > > > > > > > > > > configurations" are overrides of the
> cluster-level
> > > > >> > > > > configurations
> > > > >> > > > > > > > > > > customized per group so when an admin list
> cluster
> > > > >> > configs
> > > > >> > > it's
> > > > >> > > > > > > okay to
> > > > >> > > > > > > > > > > list just the cluster-level "defaults", not
> showing
> > > > >> any
> > > > >> > > > > per-group
> > > > >> > > > > > > > > > > customizations.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 8. How would someone configure a group
> before
> > > > >> starting
> > > > >> > > the
> > > > >> > > > > > > > > consumers? Have
> > > > >> > > > > > > > > > > > we considered allowing the explicit
> creation of
> > > > >> groups?
> > > > >> > > > > > > > > Alternatively, the
> > > > >> > > > > > > > > > > > configs could be decoupled from the group
> > > > lifecycle.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > The configs can be created before the group
> itself
> > > > as
> > > > >> an
> > > > >> > > > > > > independent
> > > > >> > > > > > > > > entity
> > > > >> > > > > > > > > > > --- of course, this requires the corresponding
> > > > >> request to
> > > > >> > > be
> > > > >> > > > > > > routed to
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > right coordinator based on the group id ---
> the only
> > > > >> > thing
> > > > >> > > that
> > > > >> > > > > > > > > differs is,
> > > > >> > > > > > > > > > > when the group itself is gone we also check
> if there
> > > > >> are
> > > > >> > > any
> > > > >> > > > > > > > > configuration
> > > > >> > > > > > > > > > > entities related to that group and delete as
> well.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Admittedly this indeed introduces an
> asymmetry on
> > > > the
> > > > >> > > creation
> > > > >> > > > > /
> > > > >> > > > > > > > > deletion
> > > > >> > > > > > > > > > > lifecycles of the config entities, and we
> would like
> > > > >> to
> > > > >> > > hear
> > > > >> > > > > > > everyone's
> > > > >> > > > > > > > > > > feelings whether we should aim for symmetry
> i.e.
> > > > >> totally
> > > > >> > > > > decouple
> > > > >> > > > > > > group
> > > > >> > > > > > > > > > > configs and hence not delete them at all when
> the
> > > > >> group
> > > > >> > is
> > > > >> > > > > gone,
> > > > >> > > > > > > but
> > > > >> > > > > > > > > always
> > > > >> > > > > > > > > > > require explicit deletion operations by
> themselves.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for
> the Java
> > > > >> > client
> > > > >> > > > > still
> > > > >> > > > > > > take
> > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > > > >> introduce an
> > > > >> > > > > overload?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > I think we do not need to introduce an
> overload, but
> > > > >> I'm
> > > > >> > > all
> > > > >> > > > > ears
> > > > >> > > > > > > if
> > > > >> > > > > > > > > there
> > > > >> > > > > > > > > > > may be some compatibility issues that we may
> > > > overlook.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 10. I agree with Justine that we should be
> clearer
> > > > >> > about
> > > > >> > > the
> > > > >> > > > > > > reason
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > > switch to IBP/metadata.version from the
> feature
> > > > >> flag.
> > > > >> > > Maybe
> > > > >> > > > > we
> > > > >> > > > > > > mean
> > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > > can switch the default for the feature flag
> to
> > > > true
> > > > >> > > based on
> > > > >> > > > > the
> > > > >> > > > > > > > > > > > metadata.version once we want to make it the
> > > > >> default.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention
> the
> > > > >> required
> > > > >> > > ACLs,
> > > > >> > > > > it
> > > > >> > > > > > > > > would be
> > > > >> > > > > > > > > > > > good to add that for consistency.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 12. It is a bit odd that
> ConsumerGroupHeartbeat
> > > > >> > requires
> > > > >> > > > > "Read
> > > > >> > > > > > > > > Group" even
> > > > >> > > > > > > > > > > > though it seems to do more than reading.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > I had that thought myself as well, but in the
> end we
> > > > >> > could
> > > > >> > > not
> > > > >> > > > > > > find a
> > > > >> > > > > > > > > > > better alternative: adding Write Group seems
> an
> > > > >> overkill
> > > > >> > > here
> > > > >> > > > > > > since we
> > > > >> > > > > > > > > do
> > > > >> > > > > > > > > > > not have it elsewhere (we only have Read /
> Delete
> > > > and
> > > > >> > > Describe
> > > > >> > > > > on
> > > > >> > > > > > > > > groups so
> > > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > > > consumer
> > > > >> > with
> > > > >> > > the
> > > > >> > > > > new
> > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > > protocol? It would be good to have a
> section on
> > > > >> this.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > You mean with regex subscription right? Yes
> we can
> > > > >> add a
> > > > >> > > > > section
> > > > >> > > > > > > about
> > > > >> > > > > > > > > > > that, but basically the idea is that consumer
> would
> > > > be
> > > > >> > > totally
> > > > >> > > > > > > > > agnostic in
> > > > >> > > > > > > > > > > the new protocol as it's handled all by the
> brokers.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > > > >> coordinator
> > > > >> > in
> > > > >> > > > > Java.
> > > > >> > > > > > > Even
> > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > > > this is an implementation detail, do we
> plan to
> > > > >> have a
> > > > >> > > new
> > > > >> > > > > gradle
> > > > >> > > > > > > > > module
> > > > >> > > > > > > > > > > > for it?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > We have not thought about that. But I think
> the
> > > > answer
> > > > >> > > should
> > > > >> > > > > be
> > > > >> > > > > > > yes.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it
> comes to
> > > > >> how
> > > > >> > > many
> > > > >> > > > > > > members
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > > group protocol can support?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Within a group, I think we should shoot for
> 1000s of
> > > > >> > > members.
> > > > >> > > > > But
> > > > >> > > > > > > that
> > > > >> > > > > > > > > > > scalability goals also depend on the offset
> > > > management
> > > > >> > > (commit,
> > > > >> > > > > > > fetch)
> > > > >> > > > > > > > > > > capabilities of the coordinator which we did
> not
> > > > >> cover in
> > > > >> > > this
> > > > >> > > > > > > KIP, so
> > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > hard to give a number that applies
> universally.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 16. Did we consider having
> SubscribedTopidIds
> > > > >> instead
> > > > >> > of
> > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > >> ConsumerGroupHeartbeatRequest?
> > > > >> > > Is the
> > > > >> > > > > > > idea
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > > since we have to resolve the regex on the
> server,
> > > > we
> > > > >> > can
> > > > >> > > do
> > > > >> > > > > the
> > > > >> > > > > > > same
> > > > >> > > > > > > > > for
> > > > >> > > > > > > > > > > > the topic name? The difference is that
> sending the
> > > > >> > regex
> > > > >> > > is
> > > > >> > > > > more
> > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > > > whereas sending the topic names is less
> efficient.
> > > > >> > > > > Furthermore,
> > > > >> > > > > > > > > delete and
> > > > >> > > > > > > > > > > > recreation is easier to handle if we have
> topic
> > > > ids.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > The main reason to still let the clients send
> names
> > > > >> is to
> > > > >> > > keep
> > > > >> > > > > the
> > > > >> > > > > > > > > > > reasoning of names -> ids on the broker /
> admin
> > > > client
> > > > >> > > only.
> > > > >> > > > > Note
> > > > >> > > > > > > that
> > > > >> > > > > > > > > > > although we added topic id in KIP-516, we
> never
> > > > >> > > implemented the
> > > > >> > > > > > > logic
> > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > > consumer/producers leveraging the related
> newer
> > > > >> versioned
> > > > >> > > RPCs,
> > > > >> > > > > > > > > instead we
> > > > >> > > > > > > > > > > just set the topic id as empty UUID. We want
> to keep
> > > > >> the
> > > > >> > > > > > > > > consumer/producer
> > > > >> > > > > > > > > > > to be thin and only delegate the reasoning on
> broker
> > > > >> and
> > > > >> > > > > > > potentially
> > > > >> > > > > > > > > admin
> > > > >> > > > > > > > > > > clients.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > Thanks,
> > > > >> > > > > > > > > > > > Ismael
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > Hi all,
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > I would like to start a discussion thread
> on
> > > > >> KIP-848:
> > > > >> > > The
> > > > >> > > > > Next
> > > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance
> Protocol.
> > > > >> With
> > > > >> > > this
> > > > >> > > > > KIP,
> > > > >> > > > > > > we
> > > > >> > > > > > > > > aim
> > > > >> > > > > > > > > > > > > to make the rebalance protocol (for
> consumers)
> > > > >> more
> > > > >> > > > > reliable,
> > > > >> > > > > > > more
> > > > >> > > > > > > > > > > > > scalable, easier to implement for
> clients, and
> > > > >> easier
> > > > >> > > to
> > > > >> > > > > debug
> > > > >> > > > > > > for
> > > > >> > > > > > > > > > > > > operators.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > The KIP is here:
> > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > Please take a look and let me know what
> you
> > > > think.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > Best,
> > > > >> > > > > > > > > > > > > David
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > PS: I will be away from July 18th to
> August 8th.
> > > > >> That
> > > > >> > > gives
> > > > >> > > > > > > you a
> > > > >> > > > > > > > > bit
> > > > >> > > > > > > > > > > > > of time to read and digest this long KIP.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > --
> > > > >> > > > > > > > > > > -- Guozhang
> > > > >> > > > > > > > >
> > > > >> > > > > > > >
> > > > >> > > > > > > >
> > > > >> > > > > > > > --
> > > > >> > > > > > > > -- Guozhang
> > > > >> > > > > > >
> > > > >> > > > >
> > > > >> > >
> > > > >> > >
> > > > >> >
> > > > >>
> > > > >>
> > > > >> --
> > > > >> -- Guozhang
> > > > >>
> > > > >
> > > >
>

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Luke,

Thanks for your comments. Please find my answers below.

> 1. Will the member reconciling jump to the latest target assignment? For
> example, member epoch are A:1, B:2, but now the group is in target
> assignment epoch 4. Will the members jump to the target assignment result
> in epoch 4 directly, or they should first sync to epoch 2, then 3, and 4? I
> tried to find out if there's any problem if we jump to 4, but I can't. So,
> I think we'll go to epoch 4 directly, right?

That's right. The reconciliation always uses the latest target
assignment so members might skip epochs. The only constraint to jump
to a new epoch is that the member should have revoked the required
partitions. In your example, if A and B do not have to revoke any
partitions, they would be moved to epoch 4 immediately by the
coordinator. Otherwise, the coordinator will ask them to revoke
partitions first and then move them to epoch 4 when the revocation is
acknowledged.

> 2. For the ConsumerGroupPrepareAssignment and
> ConsumerGroupInstallAssignment API, when member epoch doesn't match, it'll
> retry (and hope) in next heartbeat response, the member epoch will be
> bumped. But in ConsumerGroupHeartbeat, Upon receiving the
> FENCED_MEMBER_EPOCH error, the consumer abandon all its partitions and
> rejoins with the same member id and the epoch 0, unless the partitions
> owned by the members are a subset of the target partitions. I'm wondering
> why can't we return the latest member epoch to the member in
> ConsumerGroupHeartbeat response no matter if the owned partitions are
> subset of target partitions or not. After all, the member is still one of
> the group member, the assignment for the member is still valid, right? This
> way, the ConsumerGroupPrepareAssignment and ConsumerGroupInstallAssignment
> API can piggyback it to retry in next heartbeat response. WDYT?

As explained by Guozhang, the error semantic is different based on the
API. For the *PrepareAssignment and the *InstallAssignment, it means
that the member epoch is stale and that the client should retry when
it gets a newer epoch whereas for the ConsumerGroupHeartbeat, it means
that the client has been fenced by the coordinator and thus should
abandon all its partitions and rejoin. I do agree with Guozhang that
it would be better to use different errors in order to be clear. It is
misleading otherwise.

Regarding your second point, I think that we can make the API
idempotent here. There are three cases: 1) If a client sends a
ConsumerGroupHeartbeat request with an older member epoch but with
partitions which are a subset of the expected target partitions, we
can move it to the current member epoch. 2) If the member epoch is
smaller and the partitions differ, it is very likely that the member
owns partitions which are owned by other members now so it seems to me
that we should treat it as a fatal error in this case. We could be a
bit smarter and check whether the partitions are really owned by other
members and if they are not we could proceed. The issue in this case
is that the member won't be able to revoke the partitions and to
commit the offsets with a stale member epoch before the coordinator
moves it to the target epoch. 3) If the member epoch is larger, we
should fence the member because this should never happen.

> 3. When receiving COMPUTE_ASSIGNMENT error in ConsumerGroupHeartbeat API,
> the consumer starts the assignment process. But what if somehow the member
> didn't send out ConsumerGroupPrepareAssignment request, what would we do? I
> think we'll wait for rebalance timeout and kick the member out of the
> group, right? And then, what will the group coordinator do? I think it'll
> select another member to be the leader and return COMPUTE_ASSIGNMENT error
> in that member's heartbeat response, right? Maybe we should add that into
> KIP.

When the member is selected to perform the assignment, the timer
starts ticking so the coordinator gives the rebalance timeout to the
member to perform it. If it does not complete the full process by that
time, the coordinator will pick another member to run the assignment.
I believe that we should not fence the member based on this though. I
was thinking that fencing would be only done via the
ConsumerGroupHeartbeat API and the session timeout. I will clarify
this in the KIP.

> Some typos:
> 4. In "Assignment Process" section:
> If the selected assignor does exist, the group coordinator will reject the
> heartbeat with an UNSUPPORTED_ASSIGNOR error.
> -> I think it should be "does not exist"
>
> 5. In "JoinGroup Handling" section:
> If the member has revoked all its partitions or the required partitions,
> the member can transition to its next epoch. The current assignment become
> the current assignment.
> -> I think it should be "The target assignment becomes the current
> assignment", right?
>
> 6. In "ConsumerGroupDescribe API" section:
> When the group coordinator handle a ConsumerGroupPrepareAssignmentRequest
> request:
> -> It should be "handle a ConsumerGroupDescribe request"

Thanks for reporting the typos. I will fix them.

Best,
David

On Tue, Aug 9, 2022 at 3:40 PM David Jacot <dj...@confluent.io> wrote:
>
> Hi Sagar,
>
> Thanks for the feedback and the document. That's really helpful. I
> will take a look at it.
>
> Overall, it seems to me that both Connect and the Consumer could share
> the same underlying "engine". The main difference is that the Consumer
> assigns topic-partitions to members whereas Connect assigns tasks to
> workers. I see two ways to move forward:
> 1) We extend the new proposed APIs to support different resource types
> (e.g. partitions, tasks, etc.); or
> 2) We use new dedicated APIs for Connect. The dedicated APIs would be
> similar to the new ones but different on the content/resources and
> they would rely on the same engine on the coordinator side.
>
> I personally lean towards 2) because I am not a fan of overcharging
> APIs to serve different purposes. That being said, I am not opposed to
> 1) if we can find an elegant way to do it.
>
> I think that we can continue to discuss it here for now in order to
> ensure that this KIP is compatible with what we will do for Connect in
> the future.
>
> Best,
> David
>
> On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io> wrote:
> >
> > Hi all,
> >
> > I am back from vacation. I will go through and address your comments
> > in the coming days. Thanks for your feedback.
> >
> > Cheers,
> > David
> >
> > On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gh...@gmail.com> wrote:
> > >
> > > Hey All!
> > >
> > > Thanks for the KIP, it's wonderful to see cooperative rebalancing making it
> > > down the stack!
> > >
> > > I had a few questions:
> > >
> > > 1. The 'Rejected Alternatives' section describes how member epoch should
> > > advance in step with the group epoch and assignment epoch values. I think
> > > that this is a good idea for the reasons described in the KIP. When the
> > > protocol is incrementally assigning partitions to a worker, what member
> > > epoch does each incremental assignment use? Are member epochs re-used, and
> > > a single member epoch can correspond to multiple different (monotonically
> > > larger) assignments?
> > >
> > > 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> > > not, should custom client-side assignor implementations interact with the
> > > Reason field, and how is its common meaning agreed upon? If so, what is the
> > > benefit of a distinct Reason field over including such functionality in the
> > > opaque metadata?
> > >
> > > 3. The following is included in the KIP: "Thanks to this, the input of the
> > > client side assignor is entirely driven by the group coordinator. The
> > > consumer is no longer responsible for maintaining any state besides its
> > > assigned partitions." Does this mean that the client-side assignor MAY
> > > incorporate additional non-Metadata state (such as partition throughput,
> > > cpu/memory metrics, config topics, etc), or that additional non-Metadata
> > > state SHOULD NOT be used?
> > >
> > > 4. I see that there are separate classes
> > > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> > > overlap significantly. Is it possible for these two implementations to be
> > > unified? This would serve to promote feature parity of server-side and
> > > client-side assignors, and would also facilitate operational flexibility in
> > > certain situations. For example, if a server-side assignor has some poor
> > > behavior and needs a patch, deploying the patched assignor to the client
> > > and switching one consumer group to a client-side assignor may be faster
> > > and less risky than patching all of the brokers. With the currently
> > > proposed distinct APIs, a non-trivial reimplementation would have to be
> > > assembled, and if the two APIs have diverged significantly, then it is
> > > possible that a reimplementation would not be possible.
> > >
> > > --
> > > Greg Harris
> > > gharris1727@gmail.com
> > > github.com/gharris1727
> > >
> > > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com> wrote:
> > >
> > > > Hi Guozhang/David,
> > > >
> > > > I created a confluence page to discuss how Connect would need to change
> > > > based on the new rebalance protocol. Here's the page:
> > > >
> > > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > > >
> > > > It's also pretty longish and I have tried to keep a format similar to
> > > > KIP-848. Let me know what you think. Also, do you think this should be
> > > > moved to a separate discussion thread or is this one fine?
> > > >
> > > > Thanks!
> > > > Sagar.
> > > >
> > > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
> > > >
> > > > > Hello Guozhang,
> > > > >
> > > > > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > > > > analysis and come up with such a document.
> > > > >
> > > > > Thanks!
> > > > > Sagar.
> > > > >
> > > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> > > > wrote:
> > > > >
> > > > >> Hello Sagar,
> > > > >>
> > > > >> It would be great if you could come back with some analysis on how to
> > > > >> implement the Connect side integration with the new protocol; so far
> > > > >> besides leveraging on the new "protocol type" we did not yet think
> > > > through
> > > > >> the Connect side implementations. For Streams here's a draft of
> > > > >> integration
> > > > >> plan:
> > > > >>
> > > > >>
> > > > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > > >> just FYI for your analysis on Connect.
> > > > >>
> > > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> > > > wrote:
> > > > >>
> > > > >> > Hi David,
> > > > >> >
> > > > >> > Thank you for your response. The reason I thought connect can also fit
> > > > >> into
> > > > >> > this new scheme is that even today the connect uses a
> > > > WorkerCoordinator
> > > > >> > extending from AbstractCoordinator to empower rebalances of
> > > > >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> > > > >> connect
> > > > >> > and uses the metadata() method by plumbing into
> > > > >> JoinGroupRequestProtocol.
> > > > >> >
> > > > >> > I think the changes to support connect would be similar at a high
> > > > level
> > > > >> to
> > > > >> > the changes in streams mainly because of the Client side assignors
> > > > being
> > > > >> > used in both. At an implementation level, we might need to make a lot
> > > > of
> > > > >> > changes to get onto this new assignment protocol like enhancing the
> > > > >> > JoinGroup request/response and SyncGroup and using
> > > > >> ConsumerGroupHeartbeat
> > > > >> > API etc again on similar lines to streams (or there might be
> > > > >> deviations). I
> > > > >> > would try to perform a detailed analysis of the same  and we can have
> > > > a
> > > > >> > separate discussion thread for that as that would derail this
> > > > discussion
> > > > >> > thread. Let me know if that sounds good to you.
> > > > >> >
> > > > >> > Thanks!
> > > > >> > Sagar.
> > > > >> >
> > > > >> >
> > > > >> >
> > > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > > <djacot@confluent.io.invalid
> > > > >> >
> > > > >> > wrote:
> > > > >> >
> > > > >> > > Hi Sagar,
> > > > >> > >
> > > > >> > > Thanks for your comments.
> > > > >> > >
> > > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> > > > >> > >
> > > > >> > > 2) The idea is to transition C from his current assignment to his
> > > > >> > > target assignment when he can move to epoch 3. When that happens,
> > > > the
> > > > >> > > member assignment is updated and persisted with all its assigned
> > > > >> > > partitions even if they are not all revoked yet. In other words, the
> > > > >> > > member assignment becomes the target assignment. This is basically
> > > > an
> > > > >> > > optimization to avoid having to write all the changes to the log.
> > > > The
> > > > >> > > examples are based on the persisted state so I understand the
> > > > >> > > confusion. Let me see if I can improve this in the description.
> > > > >> > >
> > > > >> > > 3) Regarding Connect, it could reuse the protocol with a client side
> > > > >> > > assignor if it fits in the protocol. The assignment is about
> > > > >> > > topicid-partitions + metadata, could Connect fit into this?
> > > > >> > >
> > > > >> > > Best,
> > > > >> > > David
> > > > >> > >
> > > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> > > > >> wrote:
> > > > >> > > >
> > > > >> > > > Hi David,
> > > > >> > > >
> > > > >> > > > Thanks for the KIP. I just had minor observations:
> > > > >> > > >
> > > > >> > > > 1) In the Assignment Error section in Client Side mode Assignment
> > > > >> > > process,
> > > > >> > > > you mentioned => `In this case, the client side assignor can
> > > > return
> > > > >> an
> > > > >> > > > error to the group coordinator`. In this case are you referring to
> > > > >> the
> > > > >> > > > Assignor returning an AssignmentError that's listed down towards
> > > > the
> > > > >> > end?
> > > > >> > > > If yes, do you think it would make sense to mention this
> > > > explicitly
> > > > >> > here?
> > > > >> > > >
> > > > >> > > > 2) In the Case Studies section, I have a slight confusion, not
> > > > sure
> > > > >> if
> > > > >> > > > others have the same. Consider this step:
> > > > >> > > >
> > > > >> > > > When B heartbeats, the group coordinator transitions him to epoch
> > > > 3
> > > > >> > > because
> > > > >> > > > B has no partitions to revoke. It persists the change and reply.
> > > > >> > > >
> > > > >> > > >    - Group (epoch=3)
> > > > >> > > >       - A
> > > > >> > > >       - B
> > > > >> > > >       - C
> > > > >> > > >    - Target Assignment (epoch=3)
> > > > >> > > >       - A - partitions=[foo-0]
> > > > >> > > >       - B - partitions=[foo-2]
> > > > >> > > >       - C - partitions=[foo-1]
> > > > >> > > >    - Member Assignment
> > > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > > >> > > >
> > > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot get foo-1
> > > > >> yet.
> > > > >> > > >
> > > > >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> > > > yet,
> > > > >> > but
> > > > >> > > > based on the description above, it seems it already has it. Do you
> > > > >> > think
> > > > >> > > it
> > > > >> > > > would be better to remove it and populate it only when it actually
> > > > >> gets
> > > > >> > > it?
> > > > >> > > > I see this in a lot of other places, so have I understood it
> > > > >> > incorrectly
> > > > >> > > ?
> > > > >> > > >
> > > > >> > > >
> > > > >> > > > Regarding connect , it might be out of scope of this discussion,
> > > > but
> > > > >> > from
> > > > >> > > > what I understood it would probably be running in client side
> > > > >> assignor
> > > > >> > > mode
> > > > >> > > > even on the new rebalance protocol as it has its own Custom
> > > > >> > > Assignors(Eager
> > > > >> > > > and IncrementalCooperative).
> > > > >> > > >
> > > > >> > > > Thanks!
> > > > >> > > >
> > > > >> > > > Sagar.
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > >
> > > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > > >> > <djacot@confluent.io.invalid
> > > > >> > > >
> > > > >> > > > wrote:
> > > > >> > > >
> > > > >> > > > > Thanks Hector! Our goal is to move forward with specialized API
> > > > >> > > > > instead of relying on one generic API. For Connect, we can apply
> > > > >> the
> > > > >> > > > > exact same pattern and reuse/share the core implementation on
> > > > the
> > > > >> > > > > server side. For the schema registry, I think that we should
> > > > >> consider
> > > > >> > > > > having a tailored API to do simple membership/leader election.
> > > > >> > > > >
> > > > >> > > > > Best,
> > > > >> > > > > David
> > > > >> > > > >
> > > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <ismael@juma.me.uk
> > > > >
> > > > >> > > wrote:
> > > > >> > > > > >
> > > > >> > > > > > Three quick comments:
> > > > >> > > > > >
> > > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > > >> com.google.re2j.Pattern, we
> > > > >> > > > > should
> > > > >> > > > > > document the differences in more detail before deciding one
> > > > way
> > > > >> or
> > > > >> > > > > another.
> > > > >> > > > > > That said, if people pass java.util.regex.Pattern, they expect
> > > > >> > their
> > > > >> > > > > > semantics to be honored. If we are doing something different,
> > > > >> then
> > > > >> > we
> > > > >> > > > > > should consider adding an overload with our own Pattern class
> > > > (I
> > > > >> > > don't
> > > > >> > > > > > think we'd want to expose re2j's at this point).
> > > > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > > > integrate
> > > > >> > fully
> > > > >> > > > > with
> > > > >> > > > > > it and should handle the topic recreation case correctly.
> > > > That's
> > > > >> > the
> > > > >> > > main
> > > > >> > > > > > part we need to handle. I agree with David that we'd want to
> > > > add
> > > > >> > > topic
> > > > >> > > > > ids
> > > > >> > > > > > to the relevant protocols that don't have it yet and that we
> > > > can
> > > > >> > > probably
> > > > >> > > > > > focus on the internals versus adding new APIs to the Java
> > > > >> Consumer
> > > > >> > > > > (unless
> > > > >> > > > > > we find that adding new APIs is required for reasonable
> > > > >> semantics).
> > > > >> > > > > > 3. I am still not sure about the coordinator storing the
> > > > >> configs.
> > > > >> > > It's
> > > > >> > > > > > powerful for configs to be centralized in the metadata log for
> > > > >> > > various
> > > > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > > > >> Similarly, I
> > > > >> > > am
> > > > >> > > > > not
> > > > >> > > > > > sure about automatically deleting configs in a way that they
> > > > >> cannot
> > > > >> > > be
> > > > >> > > > > > recovered. A good property for modern systems is to minimize
> > > > the
> > > > >> > > number
> > > > >> > > > > of
> > > > >> > > > > > unrecoverable data loss scenarios.
> > > > >> > > > > >
> > > > >> > > > > > Ismael
> > > > >> > > > > >
> > > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > > >> > > <djacot@confluent.io.invalid
> > > > >> > > > > >
> > > > >> > > > > > wrote:
> > > > >> > > > > >
> > > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > > >> > > > > > >
> > > > >> > > > > > > > 1) the migration path, especially the last step when
> > > > clients
> > > > >> > > flip the
> > > > >> > > > > > > flag
> > > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > > window
> > > > >> > where
> > > > >> > > > > both
> > > > >> > > > > > > new
> > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > > > >> members
> > > > >> > of
> > > > >> > > the
> > > > >> > > > > same
> > > > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > > > >> while
> > > > >> > > > > using the
> > > > >> > > > > > > > new protocol is something we need to present about.
> > > > >> > > > > > >
> > > > >> > > > > > > Noted. I just published a new version of KIP which includes
> > > > >> more
> > > > >> > > > > > > details about this. See the "Supporting Online Consumer
> > > > Group
> > > > >> > > Upgrade"
> > > > >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan". I
> > > > >> think
> > > > >> > > that
> > > > >> > > > > > > I have to think through a few cases now but the overall idea
> > > > >> and
> > > > >> > > > > > > mechanism should be understandable.
> > > > >> > > > > > >
> > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > > > >> are
> > > > >> > > only
> > > > >> > > > > used
> > > > >> > > > > > > as
> > > > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > > > via
> > > > >> any
> > > > >> > > > > public
> > > > >> > > > > > > APIs
> > > > >> > > > > > > > to consumers yet. I think the question is, first should we
> > > > >> let
> > > > >> > > the
> > > > >> > > > > > > consumer
> > > > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > > > to
> > > > >> > fully
> > > > >> > > > > > > leverage
> > > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > > > the
> > > > >> > > topic
> > > > >> > > > > ids;
> > > > >> > > > > > > and
> > > > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > > > in
> > > > >> the
> > > > >> > > > > consumer
> > > > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > > > >> the
> > > > >> > > > > rebalance
> > > > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > > > >> > > > > > >
> > > > >> > > > > > > a) Assuming that we would include converting all the offsets
> > > > >> > > related
> > > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> > > > >> able
> > > > >> > to
> > > > >> > > > > > > fully operate with topic ids. That being said, it still has
> > > > to
> > > > >> > > provide
> > > > >> > > > > > > the topics names in various APIs so having a mapping in the
> > > > >> > > consumer
> > > > >> > > > > > > seems inevitable to me.
> > > > >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> > > > >> this
> > > > >> > > goes
> > > > >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> > > > >> > internals
> > > > >> > > > > > > here and we can consider this separately if we see value in
> > > > >> doing
> > > > >> > > it.
> > > > >> > > > > > >
> > > > >> > > > > > > Coming back to Ismael's point about using topic ids in the
> > > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> > > > >> > advantage
> > > > >> > > in
> > > > >> > > > > > > favour of it. The consumer will have the opportunity to
> > > > >> validate
> > > > >> > > that
> > > > >> > > > > > > the topics exists before passing them into the group
> > > > rebalance
> > > > >> > > > > > > protocol. Obviously, the coordinator will also notice it but
> > > > >> it
> > > > >> > > does
> > > > >> > > > > > > not really have a way to reject an invalid topic in the
> > > > >> response.
> > > > >> > > > > > >
> > > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > > except
> > > > >> for
> > > > >> > > the
> > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > > > >> > > necessary
> > > > >> > > > > to
> > > > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > > > >> still
> > > > >> > > use
> > > > >> > > > > > > Pattern
> > > > >> > > > > > > > while just documenting that our subscription may be
> > > > >> rejected by
> > > > >> > > the
> > > > >> > > > > > > server.
> > > > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > > > >> > using
> > > > >> > > an
> > > > >> > > > > > > > overloaded `String` based subscription may be more
> > > > >> vulnerable
> > > > >> > to
> > > > >> > > > > various
> > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > >
> > > > >> > > > > > > That could work. I have to look at the differences between
> > > > the
> > > > >> > two
> > > > >> > > > > > > engines to better understand the potential issues. My
> > > > >> > > understanding is
> > > > >> > > > > > > that would work for all the basic regular expressions. The
> > > > >> > > differences
> > > > >> > > > > > > between the two are mainly about the various character
> > > > >> classes. I
> > > > >> > > > > > > wonder what other people think about this.
> > > > >> > > > > > >
> > > > >> > > > > > > Best,
> > > > >> > > > > > > David
> > > > >> > > > > > >
> > > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > > > >> > wangguoz@gmail.com
> > > > >> > > >
> > > > >> > > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > Thanks David! I think on the high level there are two meta
> > > > >> > > points we
> > > > >> > > > > need
> > > > >> > > > > > > > to concretize a bit more:
> > > > >> > > > > > > >
> > > > >> > > > > > > > 1) the migration path, especially the last step when
> > > > clients
> > > > >> > > flip the
> > > > >> > > > > > > flag
> > > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > > window
> > > > >> > where
> > > > >> > > > > both
> > > > >> > > > > > > new
> > > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > > > >> members
> > > > >> > of
> > > > >> > > the
> > > > >> > > > > same
> > > > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > > > >> while
> > > > >> > > > > using the
> > > > >> > > > > > > > new protocol is something we need to present about.
> > > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > > > >> are
> > > > >> > > only
> > > > >> > > > > used
> > > > >> > > > > > > as
> > > > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > > > via
> > > > >> any
> > > > >> > > > > public
> > > > >> > > > > > > APIs
> > > > >> > > > > > > > to consumers yet. I think the question is, first should we
> > > > >> let
> > > > >> > > the
> > > > >> > > > > > > consumer
> > > > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > > > to
> > > > >> > fully
> > > > >> > > > > > > leverage
> > > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > > > the
> > > > >> > > topic
> > > > >> > > > > ids;
> > > > >> > > > > > > and
> > > > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > > > in
> > > > >> the
> > > > >> > > > > consumer
> > > > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > > > >> the
> > > > >> > > > > rebalance
> > > > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > > > >> > > > > > > >
> > > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > > except
> > > > >> for
> > > > >> > > the
> > > > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > > > >> > > necessary
> > > > >> > > > > to
> > > > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > > > >> still
> > > > >> > > use
> > > > >> > > > > > > Pattern
> > > > >> > > > > > > > while just documenting that our subscription may be
> > > > >> rejected by
> > > > >> > > the
> > > > >> > > > > > > server.
> > > > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > > > >> > using
> > > > >> > > an
> > > > >> > > > > > > > overloaded `String` based subscription may be more
> > > > >> vulnerable
> > > > >> > to
> > > > >> > > > > various
> > > > >> > > > > > > > invalid regexes.
> > > > >> > > > > > > >
> > > > >> > > > > > > >
> > > > >> > > > > > > > Guozhang
> > > > >> > > > > > > >
> > > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > > >> > > > > <djacot@confluent.io.invalid
> > > > >> > > > > > > >
> > > > >> > > > > > > > wrote:
> > > > >> > > > > > > >
> > > > >> > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> > > > >> > inline.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 1. I think it's premature to talk about target
> > > > versions
> > > > >> for
> > > > >> > > > > > > deprecation
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > removal of the existing group protocol. Unlike KRaft,
> > > > >> this
> > > > >> > > > > affects a
> > > > >> > > > > > > core
> > > > >> > > > > > > > > > client protocol and hence deprecation/removal will be
> > > > >> > heavily
> > > > >> > > > > > > dependent
> > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > how quickly applications migrate to the new protocol.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > That makes sense. I will remove it.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x, but
> > > > it
> > > > >> > > wasn't
> > > > >> > > > > made
> > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> > > > >> > > deprecate
> > > > >> > > > > and
> > > > >> > > > > > > remove
> > > > >> > > > > > > > > > the group protocol, I also suggest removing this part.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> > > > >> rewrite
> > > > >> > > the
> > > > >> > > > > group
> > > > >> > > > > > > > > coordinator in Java while we implement the new protocol.
> > > > >> This
> > > > >> > > means
> > > > >> > > > > > > > > that the internals will be slightly different (e.g.
> > > > >> threading
> > > > >> > > > > model).
> > > > >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> > > > >> group
> > > > >> > > > > > > > > coordinator to the new group coordinator to a major
> > > > >> release.
> > > > >> > > The
> > > > >> > > > > > > > > alternative would be to use a flag to do the switch
> > > > >> instead
> > > > >> > of
> > > > >> > > > > relying
> > > > >> > > > > > > > > on the software upgrade.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 3. We need to flesh out the details of the migration
> > > > >> story.
> > > > >> > > It
> > > > >> > > > > sounds
> > > > >> > > > > > > > > like
> > > > >> > > > > > > > > > we're saying we will support online migrations. Is
> > > > that
> > > > >> > > correct?
> > > > >> > > > > We
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > explain this in detail. It could also be done as a
> > > > >> separate
> > > > >> > > KIP,
> > > > >> > > > > if
> > > > >> > > > > > > it's
> > > > >> > > > > > > > > > easier.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Yes, we will support online migrations for the group.
> > > > That
> > > > >> > > means
> > > > >> > > > > that
> > > > >> > > > > > > > > a group using the old protocol will be able to switch to
> > > > >> the
> > > > >> > > new
> > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Let me briefly explain how that will work though. It is
> > > > >> > > basically a
> > > > >> > > > > > > > > four step process:
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a software
> > > > >> > > supporting
> > > > >> > > > > the
> > > > >> > > > > > > > > new group coordinator. Both the old and the new
> > > > >> coordinator
> > > > >> > > will
> > > > >> > > > > > > > > support the old protocol and rely on the same persisted
> > > > >> > > metadata so
> > > > >> > > > > > > > > they can work together. This point is an offline
> > > > >> migration.
> > > > >> > We
> > > > >> > > > > cannot
> > > > >> > > > > > > > > do this one live because it would require shutting down
> > > > >> the
> > > > >> > > current
> > > > >> > > > > > > > > coordinator and starting up the new one and that would
> > > > >> cause
> > > > >> > > > > > > > > unavailabilities.
> > > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> > > > to
> > > > >> X
> > > > >> > in
> > > > >> > > > > order
> > > > >> > > > > > > > > to enable the new protocol. This cannot be done before
> > > > 1)
> > > > >> is
> > > > >> > > > > > > > > terminated because the old coordinator doesn't support
> > > > the
> > > > >> > new
> > > > >> > > > > > > > > protocol.
> > > > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > > > supporting
> > > > >> the
> > > > >> > > > > online
> > > > >> > > > > > > > > migration (must have KIP-792). If the consumer is
> > > > already
> > > > >> > > there.
> > > > >> > > > > > > > > Nothing must be done at this point.
> > > > >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> > > > >> turned
> > > > >> > > on.
> > > > >> > > > > The
> > > > >> > > > > > > > > consumer group is automatically converted when the first
> > > > >> > > consumer
> > > > >> > > > > > > > > using the new protocol joins the group. While the
> > > > members
> > > > >> > > using the
> > > > >> > > > > > > > > old protocol are being upgraded, the old protocol is
> > > > >> proxied
> > > > >> > > into
> > > > >> > > > > the
> > > > >> > > > > > > > > new one.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > > >> subscriptions
> > > > >> > > to
> > > > >> > > > > the
> > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > but it seems like there could be some tricky
> > > > >> compatibility
> > > > >> > > > > issues.
> > > > >> > > > > > > Will
> > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > have a mechanism for users to detect that they need to
> > > > >> > update
> > > > >> > > > > their
> > > > >> > > > > > > regex
> > > > >> > > > > > > > > > before switching to the new protocol?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I think that I am a bit more optimistic than you on this
> > > > >> > > point. I
> > > > >> > > > > > > > > believe that the majority of the cases are simple
> > > > regexes
> > > > >> > which
> > > > >> > > > > should
> > > > >> > > > > > > > > work with the new engine. The coordinator will verify
> > > > the
> > > > >> > regex
> > > > >> > > > > anyway
> > > > >> > > > > > > > > and reject the consumer if the regex is not valid.
> > > > Coming
> > > > >> > back
> > > > >> > > to
> > > > >> > > > > the
> > > > >> > > > > > > > > migration path, in the worst case, the first upgraded
> > > > >> > consumer
> > > > >> > > > > joining
> > > > >> > > > > > > > > the group will be rejected. This should be used as the
> > > > >> last
> > > > >> > > > > defence, I
> > > > >> > > > > > > > > would say.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > One way for customers to validate their regex before
> > > > >> > upgrading
> > > > >> > > > > their
> > > > >> > > > > > > > > prod would be to test them with another group. For
> > > > >> instance,
> > > > >> > > that
> > > > >> > > > > > > > > could be done in a pre-prod environment. Another way
> > > > >> would be
> > > > >> > > to
> > > > >> > > > > > > > > extend the consumer-group tool to provide a regex
> > > > >> validation
> > > > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 5. Related to the last question, will the Java client
> > > > >> allow
> > > > >> > > the
> > > > >> > > > > > > users to
> > > > >> > > > > > > > > > stick with the current regex engine for compatibility
> > > > >> > > reasons?
> > > > >> > > > > For
> > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > it may be handy to keep using client based regex at
> > > > >> first
> > > > >> > to
> > > > >> > > keep
> > > > >> > > > > > > > > > migrations simple and then migrate to server based
> > > > >> regexes
> > > > >> > > as a
> > > > >> > > > > > > second
> > > > >> > > > > > > > > step.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I understand your point but I am concerned that this
> > > > would
> > > > >> > > allow
> > > > >> > > > > users
> > > > >> > > > > > > > > to actually stay in this mode. That would go against our
> > > > >> goal
> > > > >> > > of
> > > > >> > > > > > > > > simplifying the client because we would have to continue
> > > > >> > > monitoring
> > > > >> > > > > > > > > the metadata on the client side. I would rather not do
> > > > >> this.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 6. When we say that the group coordinator will be
> > > > >> > > responsible for
> > > > >> > > > > > > storing
> > > > >> > > > > > > > > > the configurations and that the configurations will be
> > > > >> > > deleted
> > > > >> > > > > when
> > > > >> > > > > > > the
> > > > >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> > > > >> > deletion
> > > > >> > > of
> > > > >> > > > > > > > > > configurations?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > That's right. The configurations will be deleted when
> > > > the
> > > > >> > > group is
> > > > >> > > > > > > > > deleted. They go together.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 7. Will the choice to store the configs in the group
> > > > >> > > coordinator
> > > > >> > > > > > > make it
> > > > >> > > > > > > > > > harder to list all cluster configs and their values?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I don't think so. The group configurations are overrides
> > > > >> of
> > > > >> > > cluster
> > > > >> > > > > > > > > configs. If you want to know all the overrides though,
> > > > you
> > > > >> > > would
> > > > >> > > > > have
> > > > >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> > > > the
> > > > >> > > metadata
> > > > >> > > > > log
> > > > >> > > > > > > > > for instance.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 8. How would someone configure a group before starting
> > > > >> the
> > > > >> > > > > consumers?
> > > > >> > > > > > > > > Have
> > > > >> > > > > > > > > > we considered allowing the explicit creation of
> > > > groups?
> > > > >> > > > > > > Alternatively,
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Yes. The group will be automatically created in this
> > > > case.
> > > > >> > > However,
> > > > >> > > > > > > > > the configs will be lost after the retention period of
> > > > the
> > > > >> > > group
> > > > >> > > > > > > > > passes.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > > > >> client
> > > > >> > > still
> > > > >> > > > > take
> > > > >> > > > > > > a
> > > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> > > > an
> > > > >> > > overload?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > That's a very group question. I forgot about that one.
> > > > As
> > > > >> the
> > > > >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> > > > the
> > > > >> > > engine
> > > > >> > > > > that
> > > > >> > > > > > > > > we plan to use, it might be better to deprecate it and
> > > > >> use an
> > > > >> > > > > overload
> > > > >> > > > > > > > > which takes a string. We would rely on the server side
> > > > >> > > validation.
> > > > >> > > > > > > > > During the migration, I think that we could still try to
> > > > >> > > toString
> > > > >> > > > > the
> > > > >> > > > > > > > > regex and use it. That should work, I think, in the
> > > > >> majority
> > > > >> > > of the
> > > > >> > > > > > > > > cases.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> > > > about
> > > > >> > the
> > > > >> > > > > reason
> > > > >> > > > > > > to
> > > > >> > > > > > > > > > switch to IBP/metadata.version from the feature flag.
> > > > >> Maybe
> > > > >> > > we
> > > > >> > > > > mean
> > > > >> > > > > > > that
> > > > >> > > > > > > > > we
> > > > >> > > > > > > > > > can switch the default for the feature flag to true
> > > > >> based
> > > > >> > on
> > > > >> > > the
> > > > >> > > > > > > > > > metadata.version once we want to make it the default.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > My plan was to use that feature flag mainly during the
> > > > >> > > development
> > > > >> > > > > > > > > phase. I should not have mentioned it, I think, because
> > > > we
> > > > >> > > could
> > > > >> > > > > use
> > > > >> > > > > > > > > an internal config for it.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > > > required
> > > > >> > > ACLs, it
> > > > >> > > > > > > would
> > > > >> > > > > > > > > be
> > > > >> > > > > > > > > > good to add that for consistency.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Noted.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > > > requires
> > > > >> > > "Read
> > > > >> > > > > Group"
> > > > >> > > > > > > > > even
> > > > >> > > > > > > > > > though it seems to do more than reading.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > I agree. This is how the current protocol works though.
> > > > We
> > > > >> > only
> > > > >> > > > > > > > > require "Read Group" to join a group. We could consider
> > > > >> > > changing
> > > > >> > > > > this
> > > > >> > > > > > > > > but I am not sure that it is worth it.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> > > > with
> > > > >> > the
> > > > >> > > new
> > > > >> > > > > > > group
> > > > >> > > > > > > > > > protocol? It would be good to have a section on this.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> > > > >> have a
> > > > >> > > new
> > > > >> > > > > > > > > topic id so it will treat it like a topic with a
> > > > different
> > > > >> > > name.
> > > > >> > > > > The
> > > > >> > > > > > > > > only issue is that the fetch/commit offsets APIs do not
> > > > >> > support
> > > > >> > > > > topic
> > > > >> > > > > > > > > IDs so the consumer would reuse the offsets based on the
> > > > >> > same.
> > > > >> > > I
> > > > >> > > > > think
> > > > >> > > > > > > > > that we should update those APIs as well in order to be
> > > > >> > > consistent
> > > > >> > > > > end
> > > > >> > > > > > > > > to end. That would strengthen the semantics of the
> > > > >> consumer.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 14. The KIP mentions we will write the new coordinator
> > > > >> in
> > > > >> > > Java.
> > > > >> > > > > Even
> > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > this is an implementation detail, do we plan to have a
> > > > >> new
> > > > >> > > gradle
> > > > >> > > > > > > module
> > > > >> > > > > > > > > > for it?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Yes.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 15. Do we have a scalability goal when it comes to how
> > > > >> many
> > > > >> > > > > members
> > > > >> > > > > > > the
> > > > >> > > > > > > > > new
> > > > >> > > > > > > > > > group protocol can support?
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > We don't have numbers at the moment. The protocol should
> > > > >> > > support
> > > > >> > > > > 1000s
> > > > >> > > > > > > > > of members per group. We will measure this when we have
> > > > a
> > > > >> > first
> > > > >> > > > > > > > > implementation. Note that we might have other
> > > > bottlenecks
> > > > >> > down
> > > > >> > > the
> > > > >> > > > > > > > > road (e.g. offset commits).
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds instead
> > > > of
> > > > >> > > > > > > > > > SubscribedTopicNames in ConsumerGroupHeartbeatRequest?
> > > > >> Is
> > > > >> > the
> > > > >> > > > > idea
> > > > >> > > > > > > that
> > > > >> > > > > > > > > > since we have to resolve the regex on the server, we
> > > > >> can do
> > > > >> > > the
> > > > >> > > > > same
> > > > >> > > > > > > for
> > > > >> > > > > > > > > > the topic name? The difference is that sending the
> > > > >> regex is
> > > > >> > > more
> > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > whereas sending the topic names is less efficient.
> > > > >> > > Furthermore,
> > > > >> > > > > > > delete
> > > > >> > > > > > > > > and
> > > > >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> > > > >> server
> > > > >> > > for
> > > > >> > > > > both
> > > > >> > > > > > > > > paths but I do agree with your point. As a second
> > > > though,
> > > > >> > using
> > > > >> > > > > topic
> > > > >> > > > > > > > > ids may be better here for the delete and recreation
> > > > case.
> > > > >> > > Also, I
> > > > >> > > > > > > > > suppose that we may allow users to subscribe with topic
> > > > >> ids
> > > > >> > in
> > > > >> > > the
> > > > >> > > > > > > > > future because that is the only way to be really robust
> > > > to
> > > > >> > > topic
> > > > >> > > > > > > > > re-creation.
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > Best,
> > > > >> > > > > > > > > David
> > > > >> > > > > > > > >
> > > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > > > >> > > djacot@confluent.io>
> > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Hi Justine,
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Thanks for your comments. Please find my answers
> > > > below.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> > > > >> > > exception
> > > > >> > > > > of the
> > > > >> > > > > > > > > > topic names based in the
> > > > ConsumerGroupHeartbeatRequest.
> > > > >> I
> > > > >> > am
> > > > >> > > not
> > > > >> > > > > sure
> > > > >> > > > > > > > > > if using topic names is the right call here. I need to
> > > > >> > think
> > > > >> > > > > about it
> > > > >> > > > > > > > > > a little more. Obviously, the KIP does not change the
> > > > >> > > > > fetch/commit
> > > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> > > > >> that
> > > > >> > we
> > > > >> > > > > should
> > > > >> > > > > > > > > > include though as it would give better overall
> > > > >> guarantee in
> > > > >> > > the
> > > > >> > > > > > > > > > producer.
> > > > >> > > > > > > > > > - You're right. I think that I should not have
> > > > mentioned
> > > > >> > this
> > > > >> > > > > flag at
> > > > >> > > > > > > > > > all. I will remove it. We can use an internal
> > > > >> configuration
> > > > >> > > while
> > > > >> > > > > > > > > > developing the feature.
> > > > >> > > > > > > > > > - Both cluster types will be supported. The change is
> > > > >> > > > > orthogonal. The
> > > > >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > Best,
> > > > >> > > > > > > > > > David
> > > > >> > > > > > > > > >
> > > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> > > > >> > > > > wangguoz@gmail.com>
> > > > >> > > > > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Hi Ismael,
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> > > > inlined
> > > > >> > > below:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > > > >> > > ismael@juma.me.uk>
> > > > >> > > > > > > wrote:
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > Thanks for the KIP. This has the potential to be a
> > > > >> > great
> > > > >> > > > > > > > > improvement. A few
> > > > >> > > > > > > > > > > > initial questions/comments:
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > 1. I think it's premature to talk about target
> > > > >> versions
> > > > >> > > for
> > > > >> > > > > > > > > deprecation and
> > > > >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> > > > >> KRaft,
> > > > >> > > this
> > > > >> > > > > > > affects a
> > > > >> > > > > > > > > core
> > > > >> > > > > > > > > > > > client protocol and hence deprecation/removal will
> > > > >> be
> > > > >> > > heavily
> > > > >> > > > > > > > > dependent on
> > > > >> > > > > > > > > > > > how quickly applications migrate to the new
> > > > >> protocol.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> > > > >> proposed
> > > > >> > > > > timeline
> > > > >> > > > > > > in
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan` and
> > > > >> > > instead
> > > > >> > > > > just
> > > > >> > > > > > > state
> > > > >> > > > > > > > > > > that we will decide in the future about when we
> > > > would
> > > > >> > > > > deprecate old
> > > > >> > > > > > > > > > > protocol and behaviors.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> > > > >> but
> > > > >> > it
> > > > >> > > > > wasn't
> > > > >> > > > > > > made
> > > > >> > > > > > > > > clear
> > > > >> > > > > > > > > > > > why. If we added that as a way to estimate when
> > > > we'd
> > > > >> > > > > deprecate
> > > > >> > > > > > > and
> > > > >> > > > > > > > > remove
> > > > >> > > > > > > > > > > > the group protocol, I also suggest removing this
> > > > >> part.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > I think that's not specifically related to the
> > > > >> > > > > deprecation/removal
> > > > >> > > > > > > > > timeline
> > > > >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> > > > >> > > broker-side
> > > > >> > > > > > > > > > > implementation may be done first, and then the
> > > > client
> > > > >> > side,
> > > > >> > > > > and we
> > > > >> > > > > > > > > would
> > > > >> > > > > > > > > > > only mark it as "released" by the time clients
> > > > >> > > implementations
> > > > >> > > > > are
> > > > >> > > > > > > > > done. At
> > > > >> > > > > > > > > > > that time, to enable the feature the clients need to
> > > > >> > first
> > > > >> > > > > swap-in
> > > > >> > > > > > > the
> > > > >> > > > > > > > > > > bytecode with a rolling bounce and then set the flag
> > > > >> > with a
> > > > >> > > > > second
> > > > >> > > > > > > > > rolling
> > > > >> > > > > > > > > > > bounce, and hence we feel it's better to be released
> > > > >> in a
> > > > >> > > major
> > > > >> > > > > > > > > version.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > > > migration
> > > > >> > > story.
> > > > >> > > > > It
> > > > >> > > > > > > > > sounds like
> > > > >> > > > > > > > > > > > we're saying we will support online migrations. Is
> > > > >> that
> > > > >> > > > > correct?
> > > > >> > > > > > > We
> > > > >> > > > > > > > > should
> > > > >> > > > > > > > > > > > explain this in detail. It could also be done as a
> > > > >> > > separate
> > > > >> > > > > KIP,
> > > > >> > > > > > > if
> > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > > easier.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yes I think that's the part we can be more concrete
> > > > >> about
> > > > >> > > for
> > > > >> > > > > sure
> > > > >> > > > > > > (and
> > > > >> > > > > > > > > > > this is related to your question 2) above). We will
> > > > >> work
> > > > >> > on
> > > > >> > > > > making
> > > > >> > > > > > > it
> > > > >> > > > > > > > > more
> > > > >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > > >> > > subscriptions
> > > > >> > > > > to
> > > > >> > > > > > > the
> > > > >> > > > > > > > > server,
> > > > >> > > > > > > > > > > > but it seems like there could be some tricky
> > > > >> > > compatibility
> > > > >> > > > > > > issues.
> > > > >> > > > > > > > > Will we
> > > > >> > > > > > > > > > > > have a mechanism for users to detect that they
> > > > need
> > > > >> to
> > > > >> > > update
> > > > >> > > > > > > their
> > > > >> > > > > > > > > regex
> > > > >> > > > > > > > > > > > before switching to the new protocol?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yes I think we need some tooling for non-java client
> > > > >> > users
> > > > >> > > to
> > > > >> > > > > sort
> > > > >> > > > > > > of
> > > > >> > > > > > > > > > > "dry-run" the client before switching to the new
> > > > >> > protocol.
> > > > >> > > I
> > > > >> > > > > do not
> > > > >> > > > > > > > > have a
> > > > >> > > > > > > > > > > specific idea on top of my head though, maybe others
> > > > >> like
> > > > >> > > @Matt
> > > > >> > > > > > > > > Howlett can
> > > > >> > > > > > > > > > > chime-in here?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 5. Related to the last question, will the Java
> > > > >> client
> > > > >> > > allow
> > > > >> > > > > the
> > > > >> > > > > > > > > users to
> > > > >> > > > > > > > > > > > stick with the current regex engine for
> > > > >> compatibility
> > > > >> > > > > reasons?
> > > > >> > > > > > > For
> > > > >> > > > > > > > > example,
> > > > >> > > > > > > > > > > > it may be handy to keep using client based regex
> > > > at
> > > > >> > > first to
> > > > >> > > > > keep
> > > > >> > > > > > > > > > > > migrations simple and then migrate to server based
> > > > >> > > regexes
> > > > >> > > > > as a
> > > > >> > > > > > > > > second
> > > > >> > > > > > > > > > > > step.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Honestly I have not thought about that for java
> > > > >> clients,
> > > > >> > > and
> > > > >> > > > > we can
> > > > >> > > > > > > > > discuss
> > > > >> > > > > > > > > > > that. What kind of compatibility issues do you have
> > > > in
> > > > >> > > mind?
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 6. When we say that the group coordinator will be
> > > > >> > > > > responsible for
> > > > >> > > > > > > > > storing
> > > > >> > > > > > > > > > > > the configurations and that the configurations
> > > > will
> > > > >> be
> > > > >> > > > > deleted
> > > > >> > > > > > > when
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> > > > trigger
> > > > >> > > deletion
> > > > >> > > > > of
> > > > >> > > > > > > > > > > > configurations?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> > > > would
> > > > >> > only
> > > > >> > > > > > > transit to
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > state when the group is EMPTY and also all of its
> > > > >> > metadata
> > > > >> > > are
> > > > >> > > > > > > gone),
> > > > >> > > > > > > > > once
> > > > >> > > > > > > > > > > it's transited to DEAD this group would never be
> > > > >> revived.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> > > > group
> > > > >> > > > > coordinator
> > > > >> > > > > > > > > make it
> > > > >> > > > > > > > > > > > harder to list all cluster configs and their
> > > > values?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > That's a good question, and our thoughts are that
> > > > the
> > > > >> > > so-called
> > > > >> > > > > > > "group
> > > > >> > > > > > > > > > > configurations" are overrides of the cluster-level
> > > > >> > > > > configurations
> > > > >> > > > > > > > > > > customized per group so when an admin list cluster
> > > > >> > configs
> > > > >> > > it's
> > > > >> > > > > > > okay to
> > > > >> > > > > > > > > > > list just the cluster-level "defaults", not showing
> > > > >> any
> > > > >> > > > > per-group
> > > > >> > > > > > > > > > > customizations.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 8. How would someone configure a group before
> > > > >> starting
> > > > >> > > the
> > > > >> > > > > > > > > consumers? Have
> > > > >> > > > > > > > > > > > we considered allowing the explicit creation of
> > > > >> groups?
> > > > >> > > > > > > > > Alternatively, the
> > > > >> > > > > > > > > > > > configs could be decoupled from the group
> > > > lifecycle.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > The configs can be created before the group itself
> > > > as
> > > > >> an
> > > > >> > > > > > > independent
> > > > >> > > > > > > > > entity
> > > > >> > > > > > > > > > > --- of course, this requires the corresponding
> > > > >> request to
> > > > >> > > be
> > > > >> > > > > > > routed to
> > > > >> > > > > > > > > the
> > > > >> > > > > > > > > > > right coordinator based on the group id --- the only
> > > > >> > thing
> > > > >> > > that
> > > > >> > > > > > > > > differs is,
> > > > >> > > > > > > > > > > when the group itself is gone we also check if there
> > > > >> are
> > > > >> > > any
> > > > >> > > > > > > > > configuration
> > > > >> > > > > > > > > > > entities related to that group and delete as well.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> > > > the
> > > > >> > > creation
> > > > >> > > > > /
> > > > >> > > > > > > > > deletion
> > > > >> > > > > > > > > > > lifecycles of the config entities, and we would like
> > > > >> to
> > > > >> > > hear
> > > > >> > > > > > > everyone's
> > > > >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> > > > >> totally
> > > > >> > > > > decouple
> > > > >> > > > > > > group
> > > > >> > > > > > > > > > > configs and hence not delete them at all when the
> > > > >> group
> > > > >> > is
> > > > >> > > > > gone,
> > > > >> > > > > > > but
> > > > >> > > > > > > > > always
> > > > >> > > > > > > > > > > require explicit deletion operations by themselves.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > > > >> > client
> > > > >> > > > > still
> > > > >> > > > > > > take
> > > > >> > > > > > > > > a
> > > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > > > >> introduce an
> > > > >> > > > > overload?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > I think we do not need to introduce an overload, but
> > > > >> I'm
> > > > >> > > all
> > > > >> > > > > ears
> > > > >> > > > > > > if
> > > > >> > > > > > > > > there
> > > > >> > > > > > > > > > > may be some compatibility issues that we may
> > > > overlook.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 10. I agree with Justine that we should be clearer
> > > > >> > about
> > > > >> > > the
> > > > >> > > > > > > reason
> > > > >> > > > > > > > > to
> > > > >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> > > > >> flag.
> > > > >> > > Maybe
> > > > >> > > > > we
> > > > >> > > > > > > mean
> > > > >> > > > > > > > > that we
> > > > >> > > > > > > > > > > > can switch the default for the feature flag to
> > > > true
> > > > >> > > based on
> > > > >> > > > > the
> > > > >> > > > > > > > > > > > metadata.version once we want to make it the
> > > > >> default.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > > > >> required
> > > > >> > > ACLs,
> > > > >> > > > > it
> > > > >> > > > > > > > > would be
> > > > >> > > > > > > > > > > > good to add that for consistency.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > > > >> > requires
> > > > >> > > > > "Read
> > > > >> > > > > > > > > Group" even
> > > > >> > > > > > > > > > > > though it seems to do more than reading.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > I had that thought myself as well, but in the end we
> > > > >> > could
> > > > >> > > not
> > > > >> > > > > > > find a
> > > > >> > > > > > > > > > > better alternative: adding Write Group seems an
> > > > >> overkill
> > > > >> > > here
> > > > >> > > > > > > since we
> > > > >> > > > > > > > > do
> > > > >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> > > > and
> > > > >> > > Describe
> > > > >> > > > > on
> > > > >> > > > > > > > > groups so
> > > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > > > consumer
> > > > >> > with
> > > > >> > > the
> > > > >> > > > > new
> > > > >> > > > > > > > > group
> > > > >> > > > > > > > > > > > protocol? It would be good to have a section on
> > > > >> this.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> > > > >> add a
> > > > >> > > > > section
> > > > >> > > > > > > about
> > > > >> > > > > > > > > > > that, but basically the idea is that consumer would
> > > > be
> > > > >> > > totally
> > > > >> > > > > > > > > agnostic in
> > > > >> > > > > > > > > > > the new protocol as it's handled all by the brokers.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > > > >> coordinator
> > > > >> > in
> > > > >> > > > > Java.
> > > > >> > > > > > > Even
> > > > >> > > > > > > > > though
> > > > >> > > > > > > > > > > > this is an implementation detail, do we plan to
> > > > >> have a
> > > > >> > > new
> > > > >> > > > > gradle
> > > > >> > > > > > > > > module
> > > > >> > > > > > > > > > > > for it?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > We have not thought about that. But I think the
> > > > answer
> > > > >> > > should
> > > > >> > > > > be
> > > > >> > > > > > > yes.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes to
> > > > >> how
> > > > >> > > many
> > > > >> > > > > > > members
> > > > >> > > > > > > > > the new
> > > > >> > > > > > > > > > > > group protocol can support?
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > Within a group, I think we should shoot for 1000s of
> > > > >> > > members.
> > > > >> > > > > But
> > > > >> > > > > > > that
> > > > >> > > > > > > > > > > scalability goals also depend on the offset
> > > > management
> > > > >> > > (commit,
> > > > >> > > > > > > fetch)
> > > > >> > > > > > > > > > > capabilities of the coordinator which we did not
> > > > >> cover in
> > > > >> > > this
> > > > >> > > > > > > KIP, so
> > > > >> > > > > > > > > it's
> > > > >> > > > > > > > > > > hard to give a number that applies universally.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > > > >> instead
> > > > >> > of
> > > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > > >> ConsumerGroupHeartbeatRequest?
> > > > >> > > Is the
> > > > >> > > > > > > idea
> > > > >> > > > > > > > > that
> > > > >> > > > > > > > > > > > since we have to resolve the regex on the server,
> > > > we
> > > > >> > can
> > > > >> > > do
> > > > >> > > > > the
> > > > >> > > > > > > same
> > > > >> > > > > > > > > for
> > > > >> > > > > > > > > > > > the topic name? The difference is that sending the
> > > > >> > regex
> > > > >> > > is
> > > > >> > > > > more
> > > > >> > > > > > > > > efficient
> > > > >> > > > > > > > > > > > whereas sending the topic names is less efficient.
> > > > >> > > > > Furthermore,
> > > > >> > > > > > > > > delete and
> > > > >> > > > > > > > > > > > recreation is easier to handle if we have topic
> > > > ids.
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > The main reason to still let the clients send names
> > > > >> is to
> > > > >> > > keep
> > > > >> > > > > the
> > > > >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> > > > client
> > > > >> > > only.
> > > > >> > > > > Note
> > > > >> > > > > > > that
> > > > >> > > > > > > > > > > although we added topic id in KIP-516, we never
> > > > >> > > implemented the
> > > > >> > > > > > > logic
> > > > >> > > > > > > > > on
> > > > >> > > > > > > > > > > consumer/producers leveraging the related newer
> > > > >> versioned
> > > > >> > > RPCs,
> > > > >> > > > > > > > > instead we
> > > > >> > > > > > > > > > > just set the topic id as empty UUID. We want to keep
> > > > >> the
> > > > >> > > > > > > > > consumer/producer
> > > > >> > > > > > > > > > > to be thin and only delegate the reasoning on broker
> > > > >> and
> > > > >> > > > > > > potentially
> > > > >> > > > > > > > > admin
> > > > >> > > > > > > > > > > clients.
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > Thanks,
> > > > >> > > > > > > > > > > > Ismael
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> > > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > > >> > > > > > > > > > > > wrote:
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > Hi all,
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > I would like to start a discussion thread on
> > > > >> KIP-848:
> > > > >> > > The
> > > > >> > > > > Next
> > > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> > > > >> With
> > > > >> > > this
> > > > >> > > > > KIP,
> > > > >> > > > > > > we
> > > > >> > > > > > > > > aim
> > > > >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> > > > >> more
> > > > >> > > > > reliable,
> > > > >> > > > > > > more
> > > > >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> > > > >> easier
> > > > >> > > to
> > > > >> > > > > debug
> > > > >> > > > > > > for
> > > > >> > > > > > > > > > > > > operators.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > The KIP is here:
> > > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > Please take a look and let me know what you
> > > > think.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > Best,
> > > > >> > > > > > > > > > > > > David
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > > > PS: I will be away from July 18th to August 8th.
> > > > >> That
> > > > >> > > gives
> > > > >> > > > > > > you a
> > > > >> > > > > > > > > bit
> > > > >> > > > > > > > > > > > > of time to read and digest this long KIP.
> > > > >> > > > > > > > > > > > >
> > > > >> > > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > >
> > > > >> > > > > > > > > > > --
> > > > >> > > > > > > > > > > -- Guozhang
> > > > >> > > > > > > > >
> > > > >> > > > > > > >
> > > > >> > > > > > > >
> > > > >> > > > > > > > --
> > > > >> > > > > > > > -- Guozhang
> > > > >> > > > > > >
> > > > >> > > > >
> > > > >> > >
> > > > >> > >
> > > > >> >
> > > > >>
> > > > >>
> > > > >> --
> > > > >> -- Guozhang
> > > > >>
> > > > >
> > > >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Sagar,

Thanks for the feedback and the document. That's really helpful. I
will take a look at it.

Overall, it seems to me that both Connect and the Consumer could share
the same underlying "engine". The main difference is that the Consumer
assigns topic-partitions to members whereas Connect assigns tasks to
workers. I see two ways to move forward:
1) We extend the new proposed APIs to support different resource types
(e.g. partitions, tasks, etc.); or
2) We use new dedicated APIs for Connect. The dedicated APIs would be
similar to the new ones but different on the content/resources and
they would rely on the same engine on the coordinator side.

I personally lean towards 2) because I am not a fan of overcharging
APIs to serve different purposes. That being said, I am not opposed to
1) if we can find an elegant way to do it.

I think that we can continue to discuss it here for now in order to
ensure that this KIP is compatible with what we will do for Connect in
the future.

Best,
David

On Mon, Aug 8, 2022 at 2:41 PM David Jacot <dj...@confluent.io> wrote:
>
> Hi all,
>
> I am back from vacation. I will go through and address your comments
> in the coming days. Thanks for your feedback.
>
> Cheers,
> David
>
> On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gh...@gmail.com> wrote:
> >
> > Hey All!
> >
> > Thanks for the KIP, it's wonderful to see cooperative rebalancing making it
> > down the stack!
> >
> > I had a few questions:
> >
> > 1. The 'Rejected Alternatives' section describes how member epoch should
> > advance in step with the group epoch and assignment epoch values. I think
> > that this is a good idea for the reasons described in the KIP. When the
> > protocol is incrementally assigning partitions to a worker, what member
> > epoch does each incremental assignment use? Are member epochs re-used, and
> > a single member epoch can correspond to multiple different (monotonically
> > larger) assignments?
> >
> > 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> > not, should custom client-side assignor implementations interact with the
> > Reason field, and how is its common meaning agreed upon? If so, what is the
> > benefit of a distinct Reason field over including such functionality in the
> > opaque metadata?
> >
> > 3. The following is included in the KIP: "Thanks to this, the input of the
> > client side assignor is entirely driven by the group coordinator. The
> > consumer is no longer responsible for maintaining any state besides its
> > assigned partitions." Does this mean that the client-side assignor MAY
> > incorporate additional non-Metadata state (such as partition throughput,
> > cpu/memory metrics, config topics, etc), or that additional non-Metadata
> > state SHOULD NOT be used?
> >
> > 4. I see that there are separate classes
> > for org.apache.kafka.server.group.consumer.PartitionAssignor
> > and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> > overlap significantly. Is it possible for these two implementations to be
> > unified? This would serve to promote feature parity of server-side and
> > client-side assignors, and would also facilitate operational flexibility in
> > certain situations. For example, if a server-side assignor has some poor
> > behavior and needs a patch, deploying the patched assignor to the client
> > and switching one consumer group to a client-side assignor may be faster
> > and less risky than patching all of the brokers. With the currently
> > proposed distinct APIs, a non-trivial reimplementation would have to be
> > assembled, and if the two APIs have diverged significantly, then it is
> > possible that a reimplementation would not be possible.
> >
> > --
> > Greg Harris
> > gharris1727@gmail.com
> > github.com/gharris1727
> >
> > On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com> wrote:
> >
> > > Hi Guozhang/David,
> > >
> > > I created a confluence page to discuss how Connect would need to change
> > > based on the new rebalance protocol. Here's the page:
> > >
> > > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> > >
> > > It's also pretty longish and I have tried to keep a format similar to
> > > KIP-848. Let me know what you think. Also, do you think this should be
> > > moved to a separate discussion thread or is this one fine?
> > >
> > > Thanks!
> > > Sagar.
> > >
> > > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
> > >
> > > > Hello Guozhang,
> > > >
> > > > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > > > analysis and come up with such a document.
> > > >
> > > > Thanks!
> > > > Sagar.
> > > >
> > > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> > > wrote:
> > > >
> > > >> Hello Sagar,
> > > >>
> > > >> It would be great if you could come back with some analysis on how to
> > > >> implement the Connect side integration with the new protocol; so far
> > > >> besides leveraging on the new "protocol type" we did not yet think
> > > through
> > > >> the Connect side implementations. For Streams here's a draft of
> > > >> integration
> > > >> plan:
> > > >>
> > > >>
> > > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > > >> just FYI for your analysis on Connect.
> > > >>
> > > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> > > wrote:
> > > >>
> > > >> > Hi David,
> > > >> >
> > > >> > Thank you for your response. The reason I thought connect can also fit
> > > >> into
> > > >> > this new scheme is that even today the connect uses a
> > > WorkerCoordinator
> > > >> > extending from AbstractCoordinator to empower rebalances of
> > > >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> > > >> connect
> > > >> > and uses the metadata() method by plumbing into
> > > >> JoinGroupRequestProtocol.
> > > >> >
> > > >> > I think the changes to support connect would be similar at a high
> > > level
> > > >> to
> > > >> > the changes in streams mainly because of the Client side assignors
> > > being
> > > >> > used in both. At an implementation level, we might need to make a lot
> > > of
> > > >> > changes to get onto this new assignment protocol like enhancing the
> > > >> > JoinGroup request/response and SyncGroup and using
> > > >> ConsumerGroupHeartbeat
> > > >> > API etc again on similar lines to streams (or there might be
> > > >> deviations). I
> > > >> > would try to perform a detailed analysis of the same  and we can have
> > > a
> > > >> > separate discussion thread for that as that would derail this
> > > discussion
> > > >> > thread. Let me know if that sounds good to you.
> > > >> >
> > > >> > Thanks!
> > > >> > Sagar.
> > > >> >
> > > >> >
> > > >> >
> > > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > > <djacot@confluent.io.invalid
> > > >> >
> > > >> > wrote:
> > > >> >
> > > >> > > Hi Sagar,
> > > >> > >
> > > >> > > Thanks for your comments.
> > > >> > >
> > > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> > > >> > >
> > > >> > > 2) The idea is to transition C from his current assignment to his
> > > >> > > target assignment when he can move to epoch 3. When that happens,
> > > the
> > > >> > > member assignment is updated and persisted with all its assigned
> > > >> > > partitions even if they are not all revoked yet. In other words, the
> > > >> > > member assignment becomes the target assignment. This is basically
> > > an
> > > >> > > optimization to avoid having to write all the changes to the log.
> > > The
> > > >> > > examples are based on the persisted state so I understand the
> > > >> > > confusion. Let me see if I can improve this in the description.
> > > >> > >
> > > >> > > 3) Regarding Connect, it could reuse the protocol with a client side
> > > >> > > assignor if it fits in the protocol. The assignment is about
> > > >> > > topicid-partitions + metadata, could Connect fit into this?
> > > >> > >
> > > >> > > Best,
> > > >> > > David
> > > >> > >
> > > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> > > >> wrote:
> > > >> > > >
> > > >> > > > Hi David,
> > > >> > > >
> > > >> > > > Thanks for the KIP. I just had minor observations:
> > > >> > > >
> > > >> > > > 1) In the Assignment Error section in Client Side mode Assignment
> > > >> > > process,
> > > >> > > > you mentioned => `In this case, the client side assignor can
> > > return
> > > >> an
> > > >> > > > error to the group coordinator`. In this case are you referring to
> > > >> the
> > > >> > > > Assignor returning an AssignmentError that's listed down towards
> > > the
> > > >> > end?
> > > >> > > > If yes, do you think it would make sense to mention this
> > > explicitly
> > > >> > here?
> > > >> > > >
> > > >> > > > 2) In the Case Studies section, I have a slight confusion, not
> > > sure
> > > >> if
> > > >> > > > others have the same. Consider this step:
> > > >> > > >
> > > >> > > > When B heartbeats, the group coordinator transitions him to epoch
> > > 3
> > > >> > > because
> > > >> > > > B has no partitions to revoke. It persists the change and reply.
> > > >> > > >
> > > >> > > >    - Group (epoch=3)
> > > >> > > >       - A
> > > >> > > >       - B
> > > >> > > >       - C
> > > >> > > >    - Target Assignment (epoch=3)
> > > >> > > >       - A - partitions=[foo-0]
> > > >> > > >       - B - partitions=[foo-2]
> > > >> > > >       - C - partitions=[foo-1]
> > > >> > > >    - Member Assignment
> > > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > > >> > > >       - B - epoch=3, partitions=[foo-2]
> > > >> > > >       - C - epoch=3, partitions=[foo-1]
> > > >> > > >
> > > >> > > > When C heartbeats, it transitions to epoch 3 but cannot get foo-1
> > > >> yet.
> > > >> > > >
> > > >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> > > yet,
> > > >> > but
> > > >> > > > based on the description above, it seems it already has it. Do you
> > > >> > think
> > > >> > > it
> > > >> > > > would be better to remove it and populate it only when it actually
> > > >> gets
> > > >> > > it?
> > > >> > > > I see this in a lot of other places, so have I understood it
> > > >> > incorrectly
> > > >> > > ?
> > > >> > > >
> > > >> > > >
> > > >> > > > Regarding connect , it might be out of scope of this discussion,
> > > but
> > > >> > from
> > > >> > > > what I understood it would probably be running in client side
> > > >> assignor
> > > >> > > mode
> > > >> > > > even on the new rebalance protocol as it has its own Custom
> > > >> > > Assignors(Eager
> > > >> > > > and IncrementalCooperative).
> > > >> > > >
> > > >> > > > Thanks!
> > > >> > > >
> > > >> > > > Sagar.
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > >
> > > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > > >> > <djacot@confluent.io.invalid
> > > >> > > >
> > > >> > > > wrote:
> > > >> > > >
> > > >> > > > > Thanks Hector! Our goal is to move forward with specialized API
> > > >> > > > > instead of relying on one generic API. For Connect, we can apply
> > > >> the
> > > >> > > > > exact same pattern and reuse/share the core implementation on
> > > the
> > > >> > > > > server side. For the schema registry, I think that we should
> > > >> consider
> > > >> > > > > having a tailored API to do simple membership/leader election.
> > > >> > > > >
> > > >> > > > > Best,
> > > >> > > > > David
> > > >> > > > >
> > > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <ismael@juma.me.uk
> > > >
> > > >> > > wrote:
> > > >> > > > > >
> > > >> > > > > > Three quick comments:
> > > >> > > > > >
> > > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > > >> com.google.re2j.Pattern, we
> > > >> > > > > should
> > > >> > > > > > document the differences in more detail before deciding one
> > > way
> > > >> or
> > > >> > > > > another.
> > > >> > > > > > That said, if people pass java.util.regex.Pattern, they expect
> > > >> > their
> > > >> > > > > > semantics to be honored. If we are doing something different,
> > > >> then
> > > >> > we
> > > >> > > > > > should consider adding an overload with our own Pattern class
> > > (I
> > > >> > > don't
> > > >> > > > > > think we'd want to expose re2j's at this point).
> > > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > > integrate
> > > >> > fully
> > > >> > > > > with
> > > >> > > > > > it and should handle the topic recreation case correctly.
> > > That's
> > > >> > the
> > > >> > > main
> > > >> > > > > > part we need to handle. I agree with David that we'd want to
> > > add
> > > >> > > topic
> > > >> > > > > ids
> > > >> > > > > > to the relevant protocols that don't have it yet and that we
> > > can
> > > >> > > probably
> > > >> > > > > > focus on the internals versus adding new APIs to the Java
> > > >> Consumer
> > > >> > > > > (unless
> > > >> > > > > > we find that adding new APIs is required for reasonable
> > > >> semantics).
> > > >> > > > > > 3. I am still not sure about the coordinator storing the
> > > >> configs.
> > > >> > > It's
> > > >> > > > > > powerful for configs to be centralized in the metadata log for
> > > >> > > various
> > > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > > >> Similarly, I
> > > >> > > am
> > > >> > > > > not
> > > >> > > > > > sure about automatically deleting configs in a way that they
> > > >> cannot
> > > >> > > be
> > > >> > > > > > recovered. A good property for modern systems is to minimize
> > > the
> > > >> > > number
> > > >> > > > > of
> > > >> > > > > > unrecoverable data loss scenarios.
> > > >> > > > > >
> > > >> > > > > > Ismael
> > > >> > > > > >
> > > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > > >> > > <djacot@confluent.io.invalid
> > > >> > > > > >
> > > >> > > > > > wrote:
> > > >> > > > > >
> > > >> > > > > > > Thanks Guozhang. My answers are below:
> > > >> > > > > > >
> > > >> > > > > > > > 1) the migration path, especially the last step when
> > > clients
> > > >> > > flip the
> > > >> > > > > > > flag
> > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > window
> > > >> > where
> > > >> > > > > both
> > > >> > > > > > > new
> > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > > >> members
> > > >> > of
> > > >> > > the
> > > >> > > > > same
> > > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > > >> while
> > > >> > > > > using the
> > > >> > > > > > > > new protocol is something we need to present about.
> > > >> > > > > > >
> > > >> > > > > > > Noted. I just published a new version of KIP which includes
> > > >> more
> > > >> > > > > > > details about this. See the "Supporting Online Consumer
> > > Group
> > > >> > > Upgrade"
> > > >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan". I
> > > >> think
> > > >> > > that
> > > >> > > > > > > I have to think through a few cases now but the overall idea
> > > >> and
> > > >> > > > > > > mechanism should be understandable.
> > > >> > > > > > >
> > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > > >> are
> > > >> > > only
> > > >> > > > > used
> > > >> > > > > > > as
> > > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > > via
> > > >> any
> > > >> > > > > public
> > > >> > > > > > > APIs
> > > >> > > > > > > > to consumers yet. I think the question is, first should we
> > > >> let
> > > >> > > the
> > > >> > > > > > > consumer
> > > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > > to
> > > >> > fully
> > > >> > > > > > > leverage
> > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > > the
> > > >> > > topic
> > > >> > > > > ids;
> > > >> > > > > > > and
> > > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > > in
> > > >> the
> > > >> > > > > consumer
> > > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > > >> the
> > > >> > > > > rebalance
> > > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > > >> > > > > > >
> > > >> > > > > > > a) Assuming that we would include converting all the offsets
> > > >> > > related
> > > >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> > > >> able
> > > >> > to
> > > >> > > > > > > fully operate with topic ids. That being said, it still has
> > > to
> > > >> > > provide
> > > >> > > > > > > the topics names in various APIs so having a mapping in the
> > > >> > > consumer
> > > >> > > > > > > seems inevitable to me.
> > > >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> > > >> this
> > > >> > > goes
> > > >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> > > >> > internals
> > > >> > > > > > > here and we can consider this separately if we see value in
> > > >> doing
> > > >> > > it.
> > > >> > > > > > >
> > > >> > > > > > > Coming back to Ismael's point about using topic ids in the
> > > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> > > >> > advantage
> > > >> > > in
> > > >> > > > > > > favour of it. The consumer will have the opportunity to
> > > >> validate
> > > >> > > that
> > > >> > > > > > > the topics exists before passing them into the group
> > > rebalance
> > > >> > > > > > > protocol. Obviously, the coordinator will also notice it but
> > > >> it
> > > >> > > does
> > > >> > > > > > > not really have a way to reject an invalid topic in the
> > > >> response.
> > > >> > > > > > >
> > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > except
> > > >> for
> > > >> > > the
> > > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > > >> > > necessary
> > > >> > > > > to
> > > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > > >> still
> > > >> > > use
> > > >> > > > > > > Pattern
> > > >> > > > > > > > while just documenting that our subscription may be
> > > >> rejected by
> > > >> > > the
> > > >> > > > > > > server.
> > > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > > >> > using
> > > >> > > an
> > > >> > > > > > > > overloaded `String` based subscription may be more
> > > >> vulnerable
> > > >> > to
> > > >> > > > > various
> > > >> > > > > > > > invalid regexes.
> > > >> > > > > > >
> > > >> > > > > > > That could work. I have to look at the differences between
> > > the
> > > >> > two
> > > >> > > > > > > engines to better understand the potential issues. My
> > > >> > > understanding is
> > > >> > > > > > > that would work for all the basic regular expressions. The
> > > >> > > differences
> > > >> > > > > > > between the two are mainly about the various character
> > > >> classes. I
> > > >> > > > > > > wonder what other people think about this.
> > > >> > > > > > >
> > > >> > > > > > > Best,
> > > >> > > > > > > David
> > > >> > > > > > >
> > > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > > >> > wangguoz@gmail.com
> > > >> > > >
> > > >> > > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > Thanks David! I think on the high level there are two meta
> > > >> > > points we
> > > >> > > > > need
> > > >> > > > > > > > to concretize a bit more:
> > > >> > > > > > > >
> > > >> > > > > > > > 1) the migration path, especially the last step when
> > > clients
> > > >> > > flip the
> > > >> > > > > > > flag
> > > >> > > > > > > > to enable the new protocol, in which we would have a
> > > window
> > > >> > where
> > > >> > > > > both
> > > >> > > > > > > new
> > > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > > >> members
> > > >> > of
> > > >> > > the
> > > >> > > > > same
> > > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > > >> while
> > > >> > > > > using the
> > > >> > > > > > > > new protocol is something we need to present about.
> > > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > > >> are
> > > >> > > only
> > > >> > > > > used
> > > >> > > > > > > as
> > > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > > via
> > > >> any
> > > >> > > > > public
> > > >> > > > > > > APIs
> > > >> > > > > > > > to consumers yet. I think the question is, first should we
> > > >> let
> > > >> > > the
> > > >> > > > > > > consumer
> > > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > > to
> > > >> > fully
> > > >> > > > > > > leverage
> > > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > > the
> > > >> > > topic
> > > >> > > > > ids;
> > > >> > > > > > > and
> > > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > > in
> > > >> the
> > > >> > > > > consumer
> > > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > > >> the
> > > >> > > > > rebalance
> > > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > > >> > > > > > > >
> > > >> > > > > > > > I'm agreeing with David on all other minor questions
> > > except
> > > >> for
> > > >> > > the
> > > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > > >> > > necessary
> > > >> > > > > to
> > > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > > >> still
> > > >> > > use
> > > >> > > > > > > Pattern
> > > >> > > > > > > > while just documenting that our subscription may be
> > > >> rejected by
> > > >> > > the
> > > >> > > > > > > server.
> > > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > > >> > using
> > > >> > > an
> > > >> > > > > > > > overloaded `String` based subscription may be more
> > > >> vulnerable
> > > >> > to
> > > >> > > > > various
> > > >> > > > > > > > invalid regexes.
> > > >> > > > > > > >
> > > >> > > > > > > >
> > > >> > > > > > > > Guozhang
> > > >> > > > > > > >
> > > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > > >> > > > > <djacot@confluent.io.invalid
> > > >> > > > > > > >
> > > >> > > > > > > > wrote:
> > > >> > > > > > > >
> > > >> > > > > > > > > Hi Ismael,
> > > >> > > > > > > > >
> > > >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> > > >> > inline.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 1. I think it's premature to talk about target
> > > versions
> > > >> for
> > > >> > > > > > > deprecation
> > > >> > > > > > > > > and
> > > >> > > > > > > > > > removal of the existing group protocol. Unlike KRaft,
> > > >> this
> > > >> > > > > affects a
> > > >> > > > > > > core
> > > >> > > > > > > > > > client protocol and hence deprecation/removal will be
> > > >> > heavily
> > > >> > > > > > > dependent
> > > >> > > > > > > > > on
> > > >> > > > > > > > > > how quickly applications migrate to the new protocol.
> > > >> > > > > > > > >
> > > >> > > > > > > > > That makes sense. I will remove it.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x, but
> > > it
> > > >> > > wasn't
> > > >> > > > > made
> > > >> > > > > > > > > clear
> > > >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> > > >> > > deprecate
> > > >> > > > > and
> > > >> > > > > > > remove
> > > >> > > > > > > > > > the group protocol, I also suggest removing this part.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> > > >> rewrite
> > > >> > > the
> > > >> > > > > group
> > > >> > > > > > > > > coordinator in Java while we implement the new protocol.
> > > >> This
> > > >> > > means
> > > >> > > > > > > > > that the internals will be slightly different (e.g.
> > > >> threading
> > > >> > > > > model).
> > > >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> > > >> group
> > > >> > > > > > > > > coordinator to the new group coordinator to a major
> > > >> release.
> > > >> > > The
> > > >> > > > > > > > > alternative would be to use a flag to do the switch
> > > >> instead
> > > >> > of
> > > >> > > > > relying
> > > >> > > > > > > > > on the software upgrade.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 3. We need to flesh out the details of the migration
> > > >> story.
> > > >> > > It
> > > >> > > > > sounds
> > > >> > > > > > > > > like
> > > >> > > > > > > > > > we're saying we will support online migrations. Is
> > > that
> > > >> > > correct?
> > > >> > > > > We
> > > >> > > > > > > > > should
> > > >> > > > > > > > > > explain this in detail. It could also be done as a
> > > >> separate
> > > >> > > KIP,
> > > >> > > > > if
> > > >> > > > > > > it's
> > > >> > > > > > > > > > easier.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Yes, we will support online migrations for the group.
> > > That
> > > >> > > means
> > > >> > > > > that
> > > >> > > > > > > > > a group using the old protocol will be able to switch to
> > > >> the
> > > >> > > new
> > > >> > > > > > > > > protocol.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Let me briefly explain how that will work though. It is
> > > >> > > basically a
> > > >> > > > > > > > > four step process:
> > > >> > > > > > > > >
> > > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a software
> > > >> > > supporting
> > > >> > > > > the
> > > >> > > > > > > > > new group coordinator. Both the old and the new
> > > >> coordinator
> > > >> > > will
> > > >> > > > > > > > > support the old protocol and rely on the same persisted
> > > >> > > metadata so
> > > >> > > > > > > > > they can work together. This point is an offline
> > > >> migration.
> > > >> > We
> > > >> > > > > cannot
> > > >> > > > > > > > > do this one live because it would require shutting down
> > > >> the
> > > >> > > current
> > > >> > > > > > > > > coordinator and starting up the new one and that would
> > > >> cause
> > > >> > > > > > > > > unavailabilities.
> > > >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> > > to
> > > >> X
> > > >> > in
> > > >> > > > > order
> > > >> > > > > > > > > to enable the new protocol. This cannot be done before
> > > 1)
> > > >> is
> > > >> > > > > > > > > terminated because the old coordinator doesn't support
> > > the
> > > >> > new
> > > >> > > > > > > > > protocol.
> > > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > > supporting
> > > >> the
> > > >> > > > > online
> > > >> > > > > > > > > migration (must have KIP-792). If the consumer is
> > > already
> > > >> > > there.
> > > >> > > > > > > > > Nothing must be done at this point.
> > > >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> > > >> turned
> > > >> > > on.
> > > >> > > > > The
> > > >> > > > > > > > > consumer group is automatically converted when the first
> > > >> > > consumer
> > > >> > > > > > > > > using the new protocol joins the group. While the
> > > members
> > > >> > > using the
> > > >> > > > > > > > > old protocol are being upgraded, the old protocol is
> > > >> proxied
> > > >> > > into
> > > >> > > > > the
> > > >> > > > > > > > > new one.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Let me clarify all of this in the KIP.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > >> subscriptions
> > > >> > > to
> > > >> > > > > the
> > > >> > > > > > > > > server,
> > > >> > > > > > > > > > but it seems like there could be some tricky
> > > >> compatibility
> > > >> > > > > issues.
> > > >> > > > > > > Will
> > > >> > > > > > > > > we
> > > >> > > > > > > > > > have a mechanism for users to detect that they need to
> > > >> > update
> > > >> > > > > their
> > > >> > > > > > > regex
> > > >> > > > > > > > > > before switching to the new protocol?
> > > >> > > > > > > > >
> > > >> > > > > > > > > I think that I am a bit more optimistic than you on this
> > > >> > > point. I
> > > >> > > > > > > > > believe that the majority of the cases are simple
> > > regexes
> > > >> > which
> > > >> > > > > should
> > > >> > > > > > > > > work with the new engine. The coordinator will verify
> > > the
> > > >> > regex
> > > >> > > > > anyway
> > > >> > > > > > > > > and reject the consumer if the regex is not valid.
> > > Coming
> > > >> > back
> > > >> > > to
> > > >> > > > > the
> > > >> > > > > > > > > migration path, in the worst case, the first upgraded
> > > >> > consumer
> > > >> > > > > joining
> > > >> > > > > > > > > the group will be rejected. This should be used as the
> > > >> last
> > > >> > > > > defence, I
> > > >> > > > > > > > > would say.
> > > >> > > > > > > > >
> > > >> > > > > > > > > One way for customers to validate their regex before
> > > >> > upgrading
> > > >> > > > > their
> > > >> > > > > > > > > prod would be to test them with another group. For
> > > >> instance,
> > > >> > > that
> > > >> > > > > > > > > could be done in a pre-prod environment. Another way
> > > >> would be
> > > >> > > to
> > > >> > > > > > > > > extend the consumer-group tool to provide a regex
> > > >> validation
> > > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 5. Related to the last question, will the Java client
> > > >> allow
> > > >> > > the
> > > >> > > > > > > users to
> > > >> > > > > > > > > > stick with the current regex engine for compatibility
> > > >> > > reasons?
> > > >> > > > > For
> > > >> > > > > > > > > example,
> > > >> > > > > > > > > > it may be handy to keep using client based regex at
> > > >> first
> > > >> > to
> > > >> > > keep
> > > >> > > > > > > > > > migrations simple and then migrate to server based
> > > >> regexes
> > > >> > > as a
> > > >> > > > > > > second
> > > >> > > > > > > > > step.
> > > >> > > > > > > > >
> > > >> > > > > > > > > I understand your point but I am concerned that this
> > > would
> > > >> > > allow
> > > >> > > > > users
> > > >> > > > > > > > > to actually stay in this mode. That would go against our
> > > >> goal
> > > >> > > of
> > > >> > > > > > > > > simplifying the client because we would have to continue
> > > >> > > monitoring
> > > >> > > > > > > > > the metadata on the client side. I would rather not do
> > > >> this.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 6. When we say that the group coordinator will be
> > > >> > > responsible for
> > > >> > > > > > > storing
> > > >> > > > > > > > > > the configurations and that the configurations will be
> > > >> > > deleted
> > > >> > > > > when
> > > >> > > > > > > the
> > > >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> > > >> > deletion
> > > >> > > of
> > > >> > > > > > > > > > configurations?
> > > >> > > > > > > > >
> > > >> > > > > > > > > That's right. The configurations will be deleted when
> > > the
> > > >> > > group is
> > > >> > > > > > > > > deleted. They go together.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 7. Will the choice to store the configs in the group
> > > >> > > coordinator
> > > >> > > > > > > make it
> > > >> > > > > > > > > > harder to list all cluster configs and their values?
> > > >> > > > > > > > >
> > > >> > > > > > > > > I don't think so. The group configurations are overrides
> > > >> of
> > > >> > > cluster
> > > >> > > > > > > > > configs. If you want to know all the overrides though,
> > > you
> > > >> > > would
> > > >> > > > > have
> > > >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> > > the
> > > >> > > metadata
> > > >> > > > > log
> > > >> > > > > > > > > for instance.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 8. How would someone configure a group before starting
> > > >> the
> > > >> > > > > consumers?
> > > >> > > > > > > > > Have
> > > >> > > > > > > > > > we considered allowing the explicit creation of
> > > groups?
> > > >> > > > > > > Alternatively,
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Yes. The group will be automatically created in this
> > > case.
> > > >> > > However,
> > > >> > > > > > > > > the configs will be lost after the retention period of
> > > the
> > > >> > > group
> > > >> > > > > > > > > passes.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > > >> client
> > > >> > > still
> > > >> > > > > take
> > > >> > > > > > > a
> > > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> > > an
> > > >> > > overload?
> > > >> > > > > > > > >
> > > >> > > > > > > > > That's a very group question. I forgot about that one.
> > > As
> > > >> the
> > > >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> > > the
> > > >> > > engine
> > > >> > > > > that
> > > >> > > > > > > > > we plan to use, it might be better to deprecate it and
> > > >> use an
> > > >> > > > > overload
> > > >> > > > > > > > > which takes a string. We would rely on the server side
> > > >> > > validation.
> > > >> > > > > > > > > During the migration, I think that we could still try to
> > > >> > > toString
> > > >> > > > > the
> > > >> > > > > > > > > regex and use it. That should work, I think, in the
> > > >> majority
> > > >> > > of the
> > > >> > > > > > > > > cases.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> > > about
> > > >> > the
> > > >> > > > > reason
> > > >> > > > > > > to
> > > >> > > > > > > > > > switch to IBP/metadata.version from the feature flag.
> > > >> Maybe
> > > >> > > we
> > > >> > > > > mean
> > > >> > > > > > > that
> > > >> > > > > > > > > we
> > > >> > > > > > > > > > can switch the default for the feature flag to true
> > > >> based
> > > >> > on
> > > >> > > the
> > > >> > > > > > > > > > metadata.version once we want to make it the default.
> > > >> > > > > > > > >
> > > >> > > > > > > > > My plan was to use that feature flag mainly during the
> > > >> > > development
> > > >> > > > > > > > > phase. I should not have mentioned it, I think, because
> > > we
> > > >> > > could
> > > >> > > > > use
> > > >> > > > > > > > > an internal config for it.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > > required
> > > >> > > ACLs, it
> > > >> > > > > > > would
> > > >> > > > > > > > > be
> > > >> > > > > > > > > > good to add that for consistency.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Noted.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > > requires
> > > >> > > "Read
> > > >> > > > > Group"
> > > >> > > > > > > > > even
> > > >> > > > > > > > > > though it seems to do more than reading.
> > > >> > > > > > > > >
> > > >> > > > > > > > > I agree. This is how the current protocol works though.
> > > We
> > > >> > only
> > > >> > > > > > > > > require "Read Group" to join a group. We could consider
> > > >> > > changing
> > > >> > > > > this
> > > >> > > > > > > > > but I am not sure that it is worth it.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> > > with
> > > >> > the
> > > >> > > new
> > > >> > > > > > > group
> > > >> > > > > > > > > > protocol? It would be good to have a section on this.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> > > >> have a
> > > >> > > new
> > > >> > > > > > > > > topic id so it will treat it like a topic with a
> > > different
> > > >> > > name.
> > > >> > > > > The
> > > >> > > > > > > > > only issue is that the fetch/commit offsets APIs do not
> > > >> > support
> > > >> > > > > topic
> > > >> > > > > > > > > IDs so the consumer would reuse the offsets based on the
> > > >> > same.
> > > >> > > I
> > > >> > > > > think
> > > >> > > > > > > > > that we should update those APIs as well in order to be
> > > >> > > consistent
> > > >> > > > > end
> > > >> > > > > > > > > to end. That would strengthen the semantics of the
> > > >> consumer.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 14. The KIP mentions we will write the new coordinator
> > > >> in
> > > >> > > Java.
> > > >> > > > > Even
> > > >> > > > > > > > > though
> > > >> > > > > > > > > > this is an implementation detail, do we plan to have a
> > > >> new
> > > >> > > gradle
> > > >> > > > > > > module
> > > >> > > > > > > > > > for it?
> > > >> > > > > > > > >
> > > >> > > > > > > > > Yes.
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 15. Do we have a scalability goal when it comes to how
> > > >> many
> > > >> > > > > members
> > > >> > > > > > > the
> > > >> > > > > > > > > new
> > > >> > > > > > > > > > group protocol can support?
> > > >> > > > > > > > >
> > > >> > > > > > > > > We don't have numbers at the moment. The protocol should
> > > >> > > support
> > > >> > > > > 1000s
> > > >> > > > > > > > > of members per group. We will measure this when we have
> > > a
> > > >> > first
> > > >> > > > > > > > > implementation. Note that we might have other
> > > bottlenecks
> > > >> > down
> > > >> > > the
> > > >> > > > > > > > > road (e.g. offset commits).
> > > >> > > > > > > > >
> > > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds instead
> > > of
> > > >> > > > > > > > > > SubscribedTopicNames in ConsumerGroupHeartbeatRequest?
> > > >> Is
> > > >> > the
> > > >> > > > > idea
> > > >> > > > > > > that
> > > >> > > > > > > > > > since we have to resolve the regex on the server, we
> > > >> can do
> > > >> > > the
> > > >> > > > > same
> > > >> > > > > > > for
> > > >> > > > > > > > > > the topic name? The difference is that sending the
> > > >> regex is
> > > >> > > more
> > > >> > > > > > > > > efficient
> > > >> > > > > > > > > > whereas sending the topic names is less efficient.
> > > >> > > Furthermore,
> > > >> > > > > > > delete
> > > >> > > > > > > > > and
> > > >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> > > >> > > > > > > > >
> > > >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> > > >> server
> > > >> > > for
> > > >> > > > > both
> > > >> > > > > > > > > paths but I do agree with your point. As a second
> > > though,
> > > >> > using
> > > >> > > > > topic
> > > >> > > > > > > > > ids may be better here for the delete and recreation
> > > case.
> > > >> > > Also, I
> > > >> > > > > > > > > suppose that we may allow users to subscribe with topic
> > > >> ids
> > > >> > in
> > > >> > > the
> > > >> > > > > > > > > future because that is the only way to be really robust
> > > to
> > > >> > > topic
> > > >> > > > > > > > > re-creation.
> > > >> > > > > > > > >
> > > >> > > > > > > > > Best,
> > > >> > > > > > > > > David
> > > >> > > > > > > > >
> > > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > > >> > > djacot@confluent.io>
> > > >> > > > > > > wrote:
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Hi Justine,
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Thanks for your comments. Please find my answers
> > > below.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> > > >> > > exception
> > > >> > > > > of the
> > > >> > > > > > > > > > topic names based in the
> > > ConsumerGroupHeartbeatRequest.
> > > >> I
> > > >> > am
> > > >> > > not
> > > >> > > > > sure
> > > >> > > > > > > > > > if using topic names is the right call here. I need to
> > > >> > think
> > > >> > > > > about it
> > > >> > > > > > > > > > a little more. Obviously, the KIP does not change the
> > > >> > > > > fetch/commit
> > > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> > > >> that
> > > >> > we
> > > >> > > > > should
> > > >> > > > > > > > > > include though as it would give better overall
> > > >> guarantee in
> > > >> > > the
> > > >> > > > > > > > > > producer.
> > > >> > > > > > > > > > - You're right. I think that I should not have
> > > mentioned
> > > >> > this
> > > >> > > > > flag at
> > > >> > > > > > > > > > all. I will remove it. We can use an internal
> > > >> configuration
> > > >> > > while
> > > >> > > > > > > > > > developing the feature.
> > > >> > > > > > > > > > - Both cluster types will be supported. The change is
> > > >> > > > > orthogonal. The
> > > >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > Best,
> > > >> > > > > > > > > > David
> > > >> > > > > > > > > >
> > > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> > > >> > > > > wangguoz@gmail.com>
> > > >> > > > > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Hi Ismael,
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> > > inlined
> > > >> > > below:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > > >> > > ismael@juma.me.uk>
> > > >> > > > > > > wrote:
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > Thanks for the KIP. This has the potential to be a
> > > >> > great
> > > >> > > > > > > > > improvement. A few
> > > >> > > > > > > > > > > > initial questions/comments:
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > 1. I think it's premature to talk about target
> > > >> versions
> > > >> > > for
> > > >> > > > > > > > > deprecation and
> > > >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> > > >> KRaft,
> > > >> > > this
> > > >> > > > > > > affects a
> > > >> > > > > > > > > core
> > > >> > > > > > > > > > > > client protocol and hence deprecation/removal will
> > > >> be
> > > >> > > heavily
> > > >> > > > > > > > > dependent on
> > > >> > > > > > > > > > > > how quickly applications migrate to the new
> > > >> protocol.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> > > >> proposed
> > > >> > > > > timeline
> > > >> > > > > > > in
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan` and
> > > >> > > instead
> > > >> > > > > just
> > > >> > > > > > > state
> > > >> > > > > > > > > > > that we will decide in the future about when we
> > > would
> > > >> > > > > deprecate old
> > > >> > > > > > > > > > > protocol and behaviors.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> > > >> but
> > > >> > it
> > > >> > > > > wasn't
> > > >> > > > > > > made
> > > >> > > > > > > > > clear
> > > >> > > > > > > > > > > > why. If we added that as a way to estimate when
> > > we'd
> > > >> > > > > deprecate
> > > >> > > > > > > and
> > > >> > > > > > > > > remove
> > > >> > > > > > > > > > > > the group protocol, I also suggest removing this
> > > >> part.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > I think that's not specifically related to the
> > > >> > > > > deprecation/removal
> > > >> > > > > > > > > timeline
> > > >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> > > >> > > broker-side
> > > >> > > > > > > > > > > implementation may be done first, and then the
> > > client
> > > >> > side,
> > > >> > > > > and we
> > > >> > > > > > > > > would
> > > >> > > > > > > > > > > only mark it as "released" by the time clients
> > > >> > > implementations
> > > >> > > > > are
> > > >> > > > > > > > > done. At
> > > >> > > > > > > > > > > that time, to enable the feature the clients need to
> > > >> > first
> > > >> > > > > swap-in
> > > >> > > > > > > the
> > > >> > > > > > > > > > > bytecode with a rolling bounce and then set the flag
> > > >> > with a
> > > >> > > > > second
> > > >> > > > > > > > > rolling
> > > >> > > > > > > > > > > bounce, and hence we feel it's better to be released
> > > >> in a
> > > >> > > major
> > > >> > > > > > > > > version.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > > migration
> > > >> > > story.
> > > >> > > > > It
> > > >> > > > > > > > > sounds like
> > > >> > > > > > > > > > > > we're saying we will support online migrations. Is
> > > >> that
> > > >> > > > > correct?
> > > >> > > > > > > We
> > > >> > > > > > > > > should
> > > >> > > > > > > > > > > > explain this in detail. It could also be done as a
> > > >> > > separate
> > > >> > > > > KIP,
> > > >> > > > > > > if
> > > >> > > > > > > > > it's
> > > >> > > > > > > > > > > > easier.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yes I think that's the part we can be more concrete
> > > >> about
> > > >> > > for
> > > >> > > > > sure
> > > >> > > > > > > (and
> > > >> > > > > > > > > > > this is related to your question 2) above). We will
> > > >> work
> > > >> > on
> > > >> > > > > making
> > > >> > > > > > > it
> > > >> > > > > > > > > more
> > > >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> > > >> > > subscriptions
> > > >> > > > > to
> > > >> > > > > > > the
> > > >> > > > > > > > > server,
> > > >> > > > > > > > > > > > but it seems like there could be some tricky
> > > >> > > compatibility
> > > >> > > > > > > issues.
> > > >> > > > > > > > > Will we
> > > >> > > > > > > > > > > > have a mechanism for users to detect that they
> > > need
> > > >> to
> > > >> > > update
> > > >> > > > > > > their
> > > >> > > > > > > > > regex
> > > >> > > > > > > > > > > > before switching to the new protocol?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yes I think we need some tooling for non-java client
> > > >> > users
> > > >> > > to
> > > >> > > > > sort
> > > >> > > > > > > of
> > > >> > > > > > > > > > > "dry-run" the client before switching to the new
> > > >> > protocol.
> > > >> > > I
> > > >> > > > > do not
> > > >> > > > > > > > > have a
> > > >> > > > > > > > > > > specific idea on top of my head though, maybe others
> > > >> like
> > > >> > > @Matt
> > > >> > > > > > > > > Howlett can
> > > >> > > > > > > > > > > chime-in here?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 5. Related to the last question, will the Java
> > > >> client
> > > >> > > allow
> > > >> > > > > the
> > > >> > > > > > > > > users to
> > > >> > > > > > > > > > > > stick with the current regex engine for
> > > >> compatibility
> > > >> > > > > reasons?
> > > >> > > > > > > For
> > > >> > > > > > > > > example,
> > > >> > > > > > > > > > > > it may be handy to keep using client based regex
> > > at
> > > >> > > first to
> > > >> > > > > keep
> > > >> > > > > > > > > > > > migrations simple and then migrate to server based
> > > >> > > regexes
> > > >> > > > > as a
> > > >> > > > > > > > > second
> > > >> > > > > > > > > > > > step.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Honestly I have not thought about that for java
> > > >> clients,
> > > >> > > and
> > > >> > > > > we can
> > > >> > > > > > > > > discuss
> > > >> > > > > > > > > > > that. What kind of compatibility issues do you have
> > > in
> > > >> > > mind?
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 6. When we say that the group coordinator will be
> > > >> > > > > responsible for
> > > >> > > > > > > > > storing
> > > >> > > > > > > > > > > > the configurations and that the configurations
> > > will
> > > >> be
> > > >> > > > > deleted
> > > >> > > > > > > when
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> > > trigger
> > > >> > > deletion
> > > >> > > > > of
> > > >> > > > > > > > > > > > configurations?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> > > would
> > > >> > only
> > > >> > > > > > > transit to
> > > >> > > > > > > > > that
> > > >> > > > > > > > > > > state when the group is EMPTY and also all of its
> > > >> > metadata
> > > >> > > are
> > > >> > > > > > > gone),
> > > >> > > > > > > > > once
> > > >> > > > > > > > > > > it's transited to DEAD this group would never be
> > > >> revived.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> > > group
> > > >> > > > > coordinator
> > > >> > > > > > > > > make it
> > > >> > > > > > > > > > > > harder to list all cluster configs and their
> > > values?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > That's a good question, and our thoughts are that
> > > the
> > > >> > > so-called
> > > >> > > > > > > "group
> > > >> > > > > > > > > > > configurations" are overrides of the cluster-level
> > > >> > > > > configurations
> > > >> > > > > > > > > > > customized per group so when an admin list cluster
> > > >> > configs
> > > >> > > it's
> > > >> > > > > > > okay to
> > > >> > > > > > > > > > > list just the cluster-level "defaults", not showing
> > > >> any
> > > >> > > > > per-group
> > > >> > > > > > > > > > > customizations.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 8. How would someone configure a group before
> > > >> starting
> > > >> > > the
> > > >> > > > > > > > > consumers? Have
> > > >> > > > > > > > > > > > we considered allowing the explicit creation of
> > > >> groups?
> > > >> > > > > > > > > Alternatively, the
> > > >> > > > > > > > > > > > configs could be decoupled from the group
> > > lifecycle.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > The configs can be created before the group itself
> > > as
> > > >> an
> > > >> > > > > > > independent
> > > >> > > > > > > > > entity
> > > >> > > > > > > > > > > --- of course, this requires the corresponding
> > > >> request to
> > > >> > > be
> > > >> > > > > > > routed to
> > > >> > > > > > > > > the
> > > >> > > > > > > > > > > right coordinator based on the group id --- the only
> > > >> > thing
> > > >> > > that
> > > >> > > > > > > > > differs is,
> > > >> > > > > > > > > > > when the group itself is gone we also check if there
> > > >> are
> > > >> > > any
> > > >> > > > > > > > > configuration
> > > >> > > > > > > > > > > entities related to that group and delete as well.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> > > the
> > > >> > > creation
> > > >> > > > > /
> > > >> > > > > > > > > deletion
> > > >> > > > > > > > > > > lifecycles of the config entities, and we would like
> > > >> to
> > > >> > > hear
> > > >> > > > > > > everyone's
> > > >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> > > >> totally
> > > >> > > > > decouple
> > > >> > > > > > > group
> > > >> > > > > > > > > > > configs and hence not delete them at all when the
> > > >> group
> > > >> > is
> > > >> > > > > gone,
> > > >> > > > > > > but
> > > >> > > > > > > > > always
> > > >> > > > > > > > > > > require explicit deletion operations by themselves.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > > >> > client
> > > >> > > > > still
> > > >> > > > > > > take
> > > >> > > > > > > > > a
> > > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > > >> introduce an
> > > >> > > > > overload?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > I think we do not need to introduce an overload, but
> > > >> I'm
> > > >> > > all
> > > >> > > > > ears
> > > >> > > > > > > if
> > > >> > > > > > > > > there
> > > >> > > > > > > > > > > may be some compatibility issues that we may
> > > overlook.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 10. I agree with Justine that we should be clearer
> > > >> > about
> > > >> > > the
> > > >> > > > > > > reason
> > > >> > > > > > > > > to
> > > >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> > > >> flag.
> > > >> > > Maybe
> > > >> > > > > we
> > > >> > > > > > > mean
> > > >> > > > > > > > > that we
> > > >> > > > > > > > > > > > can switch the default for the feature flag to
> > > true
> > > >> > > based on
> > > >> > > > > the
> > > >> > > > > > > > > > > > metadata.version once we want to make it the
> > > >> default.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > > >> required
> > > >> > > ACLs,
> > > >> > > > > it
> > > >> > > > > > > > > would be
> > > >> > > > > > > > > > > > good to add that for consistency.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Ack, we can certainly do that.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > > >> > requires
> > > >> > > > > "Read
> > > >> > > > > > > > > Group" even
> > > >> > > > > > > > > > > > though it seems to do more than reading.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > I had that thought myself as well, but in the end we
> > > >> > could
> > > >> > > not
> > > >> > > > > > > find a
> > > >> > > > > > > > > > > better alternative: adding Write Group seems an
> > > >> overkill
> > > >> > > here
> > > >> > > > > > > since we
> > > >> > > > > > > > > do
> > > >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> > > and
> > > >> > > Describe
> > > >> > > > > on
> > > >> > > > > > > > > groups so
> > > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > > consumer
> > > >> > with
> > > >> > > the
> > > >> > > > > new
> > > >> > > > > > > > > group
> > > >> > > > > > > > > > > > protocol? It would be good to have a section on
> > > >> this.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> > > >> add a
> > > >> > > > > section
> > > >> > > > > > > about
> > > >> > > > > > > > > > > that, but basically the idea is that consumer would
> > > be
> > > >> > > totally
> > > >> > > > > > > > > agnostic in
> > > >> > > > > > > > > > > the new protocol as it's handled all by the brokers.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > > >> coordinator
> > > >> > in
> > > >> > > > > Java.
> > > >> > > > > > > Even
> > > >> > > > > > > > > though
> > > >> > > > > > > > > > > > this is an implementation detail, do we plan to
> > > >> have a
> > > >> > > new
> > > >> > > > > gradle
> > > >> > > > > > > > > module
> > > >> > > > > > > > > > > > for it?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > We have not thought about that. But I think the
> > > answer
> > > >> > > should
> > > >> > > > > be
> > > >> > > > > > > yes.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes to
> > > >> how
> > > >> > > many
> > > >> > > > > > > members
> > > >> > > > > > > > > the new
> > > >> > > > > > > > > > > > group protocol can support?
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > Within a group, I think we should shoot for 1000s of
> > > >> > > members.
> > > >> > > > > But
> > > >> > > > > > > that
> > > >> > > > > > > > > > > scalability goals also depend on the offset
> > > management
> > > >> > > (commit,
> > > >> > > > > > > fetch)
> > > >> > > > > > > > > > > capabilities of the coordinator which we did not
> > > >> cover in
> > > >> > > this
> > > >> > > > > > > KIP, so
> > > >> > > > > > > > > it's
> > > >> > > > > > > > > > > hard to give a number that applies universally.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > > >> instead
> > > >> > of
> > > >> > > > > > > > > > > > SubscribedTopicNames in
> > > >> ConsumerGroupHeartbeatRequest?
> > > >> > > Is the
> > > >> > > > > > > idea
> > > >> > > > > > > > > that
> > > >> > > > > > > > > > > > since we have to resolve the regex on the server,
> > > we
> > > >> > can
> > > >> > > do
> > > >> > > > > the
> > > >> > > > > > > same
> > > >> > > > > > > > > for
> > > >> > > > > > > > > > > > the topic name? The difference is that sending the
> > > >> > regex
> > > >> > > is
> > > >> > > > > more
> > > >> > > > > > > > > efficient
> > > >> > > > > > > > > > > > whereas sending the topic names is less efficient.
> > > >> > > > > Furthermore,
> > > >> > > > > > > > > delete and
> > > >> > > > > > > > > > > > recreation is easier to handle if we have topic
> > > ids.
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > The main reason to still let the clients send names
> > > >> is to
> > > >> > > keep
> > > >> > > > > the
> > > >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> > > client
> > > >> > > only.
> > > >> > > > > Note
> > > >> > > > > > > that
> > > >> > > > > > > > > > > although we added topic id in KIP-516, we never
> > > >> > > implemented the
> > > >> > > > > > > logic
> > > >> > > > > > > > > on
> > > >> > > > > > > > > > > consumer/producers leveraging the related newer
> > > >> versioned
> > > >> > > RPCs,
> > > >> > > > > > > > > instead we
> > > >> > > > > > > > > > > just set the topic id as empty UUID. We want to keep
> > > >> the
> > > >> > > > > > > > > consumer/producer
> > > >> > > > > > > > > > > to be thin and only delegate the reasoning on broker
> > > >> and
> > > >> > > > > > > potentially
> > > >> > > > > > > > > admin
> > > >> > > > > > > > > > > clients.
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > Thanks,
> > > >> > > > > > > > > > > > Ismael
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> > > >> > > > > > > > > <dj...@confluent.io.invalid>
> > > >> > > > > > > > > > > > wrote:
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > > > > Hi all,
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > I would like to start a discussion thread on
> > > >> KIP-848:
> > > >> > > The
> > > >> > > > > Next
> > > >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> > > >> With
> > > >> > > this
> > > >> > > > > KIP,
> > > >> > > > > > > we
> > > >> > > > > > > > > aim
> > > >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> > > >> more
> > > >> > > > > reliable,
> > > >> > > > > > > more
> > > >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> > > >> easier
> > > >> > > to
> > > >> > > > > debug
> > > >> > > > > > > for
> > > >> > > > > > > > > > > > > operators.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > The KIP is here:
> > > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > Please take a look and let me know what you
> > > think.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > Best,
> > > >> > > > > > > > > > > > > David
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > > > PS: I will be away from July 18th to August 8th.
> > > >> That
> > > >> > > gives
> > > >> > > > > > > you a
> > > >> > > > > > > > > bit
> > > >> > > > > > > > > > > > > of time to read and digest this long KIP.
> > > >> > > > > > > > > > > > >
> > > >> > > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > >
> > > >> > > > > > > > > > > --
> > > >> > > > > > > > > > > -- Guozhang
> > > >> > > > > > > > >
> > > >> > > > > > > >
> > > >> > > > > > > >
> > > >> > > > > > > > --
> > > >> > > > > > > > -- Guozhang
> > > >> > > > > > >
> > > >> > > > >
> > > >> > >
> > > >> > >
> > > >> >
> > > >>
> > > >>
> > > >> --
> > > >> -- Guozhang
> > > >>
> > > >
> > >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi all,

I am back from vacation. I will go through and address your comments
in the coming days. Thanks for your feedback.

Cheers,
David

On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gh...@gmail.com> wrote:
>
> Hey All!
>
> Thanks for the KIP, it's wonderful to see cooperative rebalancing making it
> down the stack!
>
> I had a few questions:
>
> 1. The 'Rejected Alternatives' section describes how member epoch should
> advance in step with the group epoch and assignment epoch values. I think
> that this is a good idea for the reasons described in the KIP. When the
> protocol is incrementally assigning partitions to a worker, what member
> epoch does each incremental assignment use? Are member epochs re-used, and
> a single member epoch can correspond to multiple different (monotonically
> larger) assignments?
>
> 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> not, should custom client-side assignor implementations interact with the
> Reason field, and how is its common meaning agreed upon? If so, what is the
> benefit of a distinct Reason field over including such functionality in the
> opaque metadata?
>
> 3. The following is included in the KIP: "Thanks to this, the input of the
> client side assignor is entirely driven by the group coordinator. The
> consumer is no longer responsible for maintaining any state besides its
> assigned partitions." Does this mean that the client-side assignor MAY
> incorporate additional non-Metadata state (such as partition throughput,
> cpu/memory metrics, config topics, etc), or that additional non-Metadata
> state SHOULD NOT be used?
>
> 4. I see that there are separate classes
> for org.apache.kafka.server.group.consumer.PartitionAssignor
> and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> overlap significantly. Is it possible for these two implementations to be
> unified? This would serve to promote feature parity of server-side and
> client-side assignors, and would also facilitate operational flexibility in
> certain situations. For example, if a server-side assignor has some poor
> behavior and needs a patch, deploying the patched assignor to the client
> and switching one consumer group to a client-side assignor may be faster
> and less risky than patching all of the brokers. With the currently
> proposed distinct APIs, a non-trivial reimplementation would have to be
> assembled, and if the two APIs have diverged significantly, then it is
> possible that a reimplementation would not be possible.
>
> --
> Greg Harris
> gharris1727@gmail.com
> github.com/gharris1727
>
> On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com> wrote:
>
> > Hi Guozhang/David,
> >
> > I created a confluence page to discuss how Connect would need to change
> > based on the new rebalance protocol. Here's the page:
> >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >
> > It's also pretty longish and I have tried to keep a format similar to
> > KIP-848. Let me know what you think. Also, do you think this should be
> > moved to a separate discussion thread or is this one fine?
> >
> > Thanks!
> > Sagar.
> >
> > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
> >
> > > Hello Guozhang,
> > >
> > > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > > analysis and come up with such a document.
> > >
> > > Thanks!
> > > Sagar.
> > >
> > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > >> Hello Sagar,
> > >>
> > >> It would be great if you could come back with some analysis on how to
> > >> implement the Connect side integration with the new protocol; so far
> > >> besides leveraging on the new "protocol type" we did not yet think
> > through
> > >> the Connect side implementations. For Streams here's a draft of
> > >> integration
> > >> plan:
> > >>
> > >>
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > >> just FYI for your analysis on Connect.
> > >>
> > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> > wrote:
> > >>
> > >> > Hi David,
> > >> >
> > >> > Thank you for your response. The reason I thought connect can also fit
> > >> into
> > >> > this new scheme is that even today the connect uses a
> > WorkerCoordinator
> > >> > extending from AbstractCoordinator to empower rebalances of
> > >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> > >> connect
> > >> > and uses the metadata() method by plumbing into
> > >> JoinGroupRequestProtocol.
> > >> >
> > >> > I think the changes to support connect would be similar at a high
> > level
> > >> to
> > >> > the changes in streams mainly because of the Client side assignors
> > being
> > >> > used in both. At an implementation level, we might need to make a lot
> > of
> > >> > changes to get onto this new assignment protocol like enhancing the
> > >> > JoinGroup request/response and SyncGroup and using
> > >> ConsumerGroupHeartbeat
> > >> > API etc again on similar lines to streams (or there might be
> > >> deviations). I
> > >> > would try to perform a detailed analysis of the same  and we can have
> > a
> > >> > separate discussion thread for that as that would derail this
> > discussion
> > >> > thread. Let me know if that sounds good to you.
> > >> >
> > >> > Thanks!
> > >> > Sagar.
> > >> >
> > >> >
> > >> >
> > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > <djacot@confluent.io.invalid
> > >> >
> > >> > wrote:
> > >> >
> > >> > > Hi Sagar,
> > >> > >
> > >> > > Thanks for your comments.
> > >> > >
> > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> > >> > >
> > >> > > 2) The idea is to transition C from his current assignment to his
> > >> > > target assignment when he can move to epoch 3. When that happens,
> > the
> > >> > > member assignment is updated and persisted with all its assigned
> > >> > > partitions even if they are not all revoked yet. In other words, the
> > >> > > member assignment becomes the target assignment. This is basically
> > an
> > >> > > optimization to avoid having to write all the changes to the log.
> > The
> > >> > > examples are based on the persisted state so I understand the
> > >> > > confusion. Let me see if I can improve this in the description.
> > >> > >
> > >> > > 3) Regarding Connect, it could reuse the protocol with a client side
> > >> > > assignor if it fits in the protocol. The assignment is about
> > >> > > topicid-partitions + metadata, could Connect fit into this?
> > >> > >
> > >> > > Best,
> > >> > > David
> > >> > >
> > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> > >> wrote:
> > >> > > >
> > >> > > > Hi David,
> > >> > > >
> > >> > > > Thanks for the KIP. I just had minor observations:
> > >> > > >
> > >> > > > 1) In the Assignment Error section in Client Side mode Assignment
> > >> > > process,
> > >> > > > you mentioned => `In this case, the client side assignor can
> > return
> > >> an
> > >> > > > error to the group coordinator`. In this case are you referring to
> > >> the
> > >> > > > Assignor returning an AssignmentError that's listed down towards
> > the
> > >> > end?
> > >> > > > If yes, do you think it would make sense to mention this
> > explicitly
> > >> > here?
> > >> > > >
> > >> > > > 2) In the Case Studies section, I have a slight confusion, not
> > sure
> > >> if
> > >> > > > others have the same. Consider this step:
> > >> > > >
> > >> > > > When B heartbeats, the group coordinator transitions him to epoch
> > 3
> > >> > > because
> > >> > > > B has no partitions to revoke. It persists the change and reply.
> > >> > > >
> > >> > > >    - Group (epoch=3)
> > >> > > >       - A
> > >> > > >       - B
> > >> > > >       - C
> > >> > > >    - Target Assignment (epoch=3)
> > >> > > >       - A - partitions=[foo-0]
> > >> > > >       - B - partitions=[foo-2]
> > >> > > >       - C - partitions=[foo-1]
> > >> > > >    - Member Assignment
> > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > >> > > >       - B - epoch=3, partitions=[foo-2]
> > >> > > >       - C - epoch=3, partitions=[foo-1]
> > >> > > >
> > >> > > > When C heartbeats, it transitions to epoch 3 but cannot get foo-1
> > >> yet.
> > >> > > >
> > >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> > yet,
> > >> > but
> > >> > > > based on the description above, it seems it already has it. Do you
> > >> > think
> > >> > > it
> > >> > > > would be better to remove it and populate it only when it actually
> > >> gets
> > >> > > it?
> > >> > > > I see this in a lot of other places, so have I understood it
> > >> > incorrectly
> > >> > > ?
> > >> > > >
> > >> > > >
> > >> > > > Regarding connect , it might be out of scope of this discussion,
> > but
> > >> > from
> > >> > > > what I understood it would probably be running in client side
> > >> assignor
> > >> > > mode
> > >> > > > even on the new rebalance protocol as it has its own Custom
> > >> > > Assignors(Eager
> > >> > > > and IncrementalCooperative).
> > >> > > >
> > >> > > > Thanks!
> > >> > > >
> > >> > > > Sagar.
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > >> > <djacot@confluent.io.invalid
> > >> > > >
> > >> > > > wrote:
> > >> > > >
> > >> > > > > Thanks Hector! Our goal is to move forward with specialized API
> > >> > > > > instead of relying on one generic API. For Connect, we can apply
> > >> the
> > >> > > > > exact same pattern and reuse/share the core implementation on
> > the
> > >> > > > > server side. For the schema registry, I think that we should
> > >> consider
> > >> > > > > having a tailored API to do simple membership/leader election.
> > >> > > > >
> > >> > > > > Best,
> > >> > > > > David
> > >> > > > >
> > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <ismael@juma.me.uk
> > >
> > >> > > wrote:
> > >> > > > > >
> > >> > > > > > Three quick comments:
> > >> > > > > >
> > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > >> com.google.re2j.Pattern, we
> > >> > > > > should
> > >> > > > > > document the differences in more detail before deciding one
> > way
> > >> or
> > >> > > > > another.
> > >> > > > > > That said, if people pass java.util.regex.Pattern, they expect
> > >> > their
> > >> > > > > > semantics to be honored. If we are doing something different,
> > >> then
> > >> > we
> > >> > > > > > should consider adding an overload with our own Pattern class
> > (I
> > >> > > don't
> > >> > > > > > think we'd want to expose re2j's at this point).
> > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > integrate
> > >> > fully
> > >> > > > > with
> > >> > > > > > it and should handle the topic recreation case correctly.
> > That's
> > >> > the
> > >> > > main
> > >> > > > > > part we need to handle. I agree with David that we'd want to
> > add
> > >> > > topic
> > >> > > > > ids
> > >> > > > > > to the relevant protocols that don't have it yet and that we
> > can
> > >> > > probably
> > >> > > > > > focus on the internals versus adding new APIs to the Java
> > >> Consumer
> > >> > > > > (unless
> > >> > > > > > we find that adding new APIs is required for reasonable
> > >> semantics).
> > >> > > > > > 3. I am still not sure about the coordinator storing the
> > >> configs.
> > >> > > It's
> > >> > > > > > powerful for configs to be centralized in the metadata log for
> > >> > > various
> > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > >> Similarly, I
> > >> > > am
> > >> > > > > not
> > >> > > > > > sure about automatically deleting configs in a way that they
> > >> cannot
> > >> > > be
> > >> > > > > > recovered. A good property for modern systems is to minimize
> > the
> > >> > > number
> > >> > > > > of
> > >> > > > > > unrecoverable data loss scenarios.
> > >> > > > > >
> > >> > > > > > Ismael
> > >> > > > > >
> > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > >> > > <djacot@confluent.io.invalid
> > >> > > > > >
> > >> > > > > > wrote:
> > >> > > > > >
> > >> > > > > > > Thanks Guozhang. My answers are below:
> > >> > > > > > >
> > >> > > > > > > > 1) the migration path, especially the last step when
> > clients
> > >> > > flip the
> > >> > > > > > > flag
> > >> > > > > > > > to enable the new protocol, in which we would have a
> > window
> > >> > where
> > >> > > > > both
> > >> > > > > > > new
> > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > >> members
> > >> > of
> > >> > > the
> > >> > > > > same
> > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > >> while
> > >> > > > > using the
> > >> > > > > > > > new protocol is something we need to present about.
> > >> > > > > > >
> > >> > > > > > > Noted. I just published a new version of KIP which includes
> > >> more
> > >> > > > > > > details about this. See the "Supporting Online Consumer
> > Group
> > >> > > Upgrade"
> > >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan". I
> > >> think
> > >> > > that
> > >> > > > > > > I have to think through a few cases now but the overall idea
> > >> and
> > >> > > > > > > mechanism should be understandable.
> > >> > > > > > >
> > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > >> are
> > >> > > only
> > >> > > > > used
> > >> > > > > > > as
> > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > via
> > >> any
> > >> > > > > public
> > >> > > > > > > APIs
> > >> > > > > > > > to consumers yet. I think the question is, first should we
> > >> let
> > >> > > the
> > >> > > > > > > consumer
> > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > to
> > >> > fully
> > >> > > > > > > leverage
> > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > the
> > >> > > topic
> > >> > > > > ids;
> > >> > > > > > > and
> > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > in
> > >> the
> > >> > > > > consumer
> > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > >> the
> > >> > > > > rebalance
> > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > >> > > > > > >
> > >> > > > > > > a) Assuming that we would include converting all the offsets
> > >> > > related
> > >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> > >> able
> > >> > to
> > >> > > > > > > fully operate with topic ids. That being said, it still has
> > to
> > >> > > provide
> > >> > > > > > > the topics names in various APIs so having a mapping in the
> > >> > > consumer
> > >> > > > > > > seems inevitable to me.
> > >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> > >> this
> > >> > > goes
> > >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> > >> > internals
> > >> > > > > > > here and we can consider this separately if we see value in
> > >> doing
> > >> > > it.
> > >> > > > > > >
> > >> > > > > > > Coming back to Ismael's point about using topic ids in the
> > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> > >> > advantage
> > >> > > in
> > >> > > > > > > favour of it. The consumer will have the opportunity to
> > >> validate
> > >> > > that
> > >> > > > > > > the topics exists before passing them into the group
> > rebalance
> > >> > > > > > > protocol. Obviously, the coordinator will also notice it but
> > >> it
> > >> > > does
> > >> > > > > > > not really have a way to reject an invalid topic in the
> > >> response.
> > >> > > > > > >
> > >> > > > > > > > I'm agreeing with David on all other minor questions
> > except
> > >> for
> > >> > > the
> > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > >> > > necessary
> > >> > > > > to
> > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > >> still
> > >> > > use
> > >> > > > > > > Pattern
> > >> > > > > > > > while just documenting that our subscription may be
> > >> rejected by
> > >> > > the
> > >> > > > > > > server.
> > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > >> > using
> > >> > > an
> > >> > > > > > > > overloaded `String` based subscription may be more
> > >> vulnerable
> > >> > to
> > >> > > > > various
> > >> > > > > > > > invalid regexes.
> > >> > > > > > >
> > >> > > > > > > That could work. I have to look at the differences between
> > the
> > >> > two
> > >> > > > > > > engines to better understand the potential issues. My
> > >> > > understanding is
> > >> > > > > > > that would work for all the basic regular expressions. The
> > >> > > differences
> > >> > > > > > > between the two are mainly about the various character
> > >> classes. I
> > >> > > > > > > wonder what other people think about this.
> > >> > > > > > >
> > >> > > > > > > Best,
> > >> > > > > > > David
> > >> > > > > > >
> > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > >> > wangguoz@gmail.com
> > >> > > >
> > >> > > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > Thanks David! I think on the high level there are two meta
> > >> > > points we
> > >> > > > > need
> > >> > > > > > > > to concretize a bit more:
> > >> > > > > > > >
> > >> > > > > > > > 1) the migration path, especially the last step when
> > clients
> > >> > > flip the
> > >> > > > > > > flag
> > >> > > > > > > > to enable the new protocol, in which we would have a
> > window
> > >> > where
> > >> > > > > both
> > >> > > > > > > new
> > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > >> members
> > >> > of
> > >> > > the
> > >> > > > > same
> > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > >> while
> > >> > > > > using the
> > >> > > > > > > > new protocol is something we need to present about.
> > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > >> are
> > >> > > only
> > >> > > > > used
> > >> > > > > > > as
> > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > via
> > >> any
> > >> > > > > public
> > >> > > > > > > APIs
> > >> > > > > > > > to consumers yet. I think the question is, first should we
> > >> let
> > >> > > the
> > >> > > > > > > consumer
> > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > to
> > >> > fully
> > >> > > > > > > leverage
> > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > the
> > >> > > topic
> > >> > > > > ids;
> > >> > > > > > > and
> > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > in
> > >> the
> > >> > > > > consumer
> > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > >> the
> > >> > > > > rebalance
> > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > >> > > > > > > >
> > >> > > > > > > > I'm agreeing with David on all other minor questions
> > except
> > >> for
> > >> > > the
> > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > >> > > necessary
> > >> > > > > to
> > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > >> still
> > >> > > use
> > >> > > > > > > Pattern
> > >> > > > > > > > while just documenting that our subscription may be
> > >> rejected by
> > >> > > the
> > >> > > > > > > server.
> > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > >> > using
> > >> > > an
> > >> > > > > > > > overloaded `String` based subscription may be more
> > >> vulnerable
> > >> > to
> > >> > > > > various
> > >> > > > > > > > invalid regexes.
> > >> > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > > Guozhang
> > >> > > > > > > >
> > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > >> > > > > <djacot@confluent.io.invalid
> > >> > > > > > > >
> > >> > > > > > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > > Hi Ismael,
> > >> > > > > > > > >
> > >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> > >> > inline.
> > >> > > > > > > > >
> > >> > > > > > > > > > 1. I think it's premature to talk about target
> > versions
> > >> for
> > >> > > > > > > deprecation
> > >> > > > > > > > > and
> > >> > > > > > > > > > removal of the existing group protocol. Unlike KRaft,
> > >> this
> > >> > > > > affects a
> > >> > > > > > > core
> > >> > > > > > > > > > client protocol and hence deprecation/removal will be
> > >> > heavily
> > >> > > > > > > dependent
> > >> > > > > > > > > on
> > >> > > > > > > > > > how quickly applications migrate to the new protocol.
> > >> > > > > > > > >
> > >> > > > > > > > > That makes sense. I will remove it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x, but
> > it
> > >> > > wasn't
> > >> > > > > made
> > >> > > > > > > > > clear
> > >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> > >> > > deprecate
> > >> > > > > and
> > >> > > > > > > remove
> > >> > > > > > > > > > the group protocol, I also suggest removing this part.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> > >> rewrite
> > >> > > the
> > >> > > > > group
> > >> > > > > > > > > coordinator in Java while we implement the new protocol.
> > >> This
> > >> > > means
> > >> > > > > > > > > that the internals will be slightly different (e.g.
> > >> threading
> > >> > > > > model).
> > >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> > >> group
> > >> > > > > > > > > coordinator to the new group coordinator to a major
> > >> release.
> > >> > > The
> > >> > > > > > > > > alternative would be to use a flag to do the switch
> > >> instead
> > >> > of
> > >> > > > > relying
> > >> > > > > > > > > on the software upgrade.
> > >> > > > > > > > >
> > >> > > > > > > > > > 3. We need to flesh out the details of the migration
> > >> story.
> > >> > > It
> > >> > > > > sounds
> > >> > > > > > > > > like
> > >> > > > > > > > > > we're saying we will support online migrations. Is
> > that
> > >> > > correct?
> > >> > > > > We
> > >> > > > > > > > > should
> > >> > > > > > > > > > explain this in detail. It could also be done as a
> > >> separate
> > >> > > KIP,
> > >> > > > > if
> > >> > > > > > > it's
> > >> > > > > > > > > > easier.
> > >> > > > > > > > >
> > >> > > > > > > > > Yes, we will support online migrations for the group.
> > That
> > >> > > means
> > >> > > > > that
> > >> > > > > > > > > a group using the old protocol will be able to switch to
> > >> the
> > >> > > new
> > >> > > > > > > > > protocol.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me briefly explain how that will work though. It is
> > >> > > basically a
> > >> > > > > > > > > four step process:
> > >> > > > > > > > >
> > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a software
> > >> > > supporting
> > >> > > > > the
> > >> > > > > > > > > new group coordinator. Both the old and the new
> > >> coordinator
> > >> > > will
> > >> > > > > > > > > support the old protocol and rely on the same persisted
> > >> > > metadata so
> > >> > > > > > > > > they can work together. This point is an offline
> > >> migration.
> > >> > We
> > >> > > > > cannot
> > >> > > > > > > > > do this one live because it would require shutting down
> > >> the
> > >> > > current
> > >> > > > > > > > > coordinator and starting up the new one and that would
> > >> cause
> > >> > > > > > > > > unavailabilities.
> > >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> > to
> > >> X
> > >> > in
> > >> > > > > order
> > >> > > > > > > > > to enable the new protocol. This cannot be done before
> > 1)
> > >> is
> > >> > > > > > > > > terminated because the old coordinator doesn't support
> > the
> > >> > new
> > >> > > > > > > > > protocol.
> > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > supporting
> > >> the
> > >> > > > > online
> > >> > > > > > > > > migration (must have KIP-792). If the consumer is
> > already
> > >> > > there.
> > >> > > > > > > > > Nothing must be done at this point.
> > >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> > >> turned
> > >> > > on.
> > >> > > > > The
> > >> > > > > > > > > consumer group is automatically converted when the first
> > >> > > consumer
> > >> > > > > > > > > using the new protocol joins the group. While the
> > members
> > >> > > using the
> > >> > > > > > > > > old protocol are being upgraded, the old protocol is
> > >> proxied
> > >> > > into
> > >> > > > > the
> > >> > > > > > > > > new one.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me clarify all of this in the KIP.
> > >> > > > > > > > >
> > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > >> subscriptions
> > >> > > to
> > >> > > > > the
> > >> > > > > > > > > server,
> > >> > > > > > > > > > but it seems like there could be some tricky
> > >> compatibility
> > >> > > > > issues.
> > >> > > > > > > Will
> > >> > > > > > > > > we
> > >> > > > > > > > > > have a mechanism for users to detect that they need to
> > >> > update
> > >> > > > > their
> > >> > > > > > > regex
> > >> > > > > > > > > > before switching to the new protocol?
> > >> > > > > > > > >
> > >> > > > > > > > > I think that I am a bit more optimistic than you on this
> > >> > > point. I
> > >> > > > > > > > > believe that the majority of the cases are simple
> > regexes
> > >> > which
> > >> > > > > should
> > >> > > > > > > > > work with the new engine. The coordinator will verify
> > the
> > >> > regex
> > >> > > > > anyway
> > >> > > > > > > > > and reject the consumer if the regex is not valid.
> > Coming
> > >> > back
> > >> > > to
> > >> > > > > the
> > >> > > > > > > > > migration path, in the worst case, the first upgraded
> > >> > consumer
> > >> > > > > joining
> > >> > > > > > > > > the group will be rejected. This should be used as the
> > >> last
> > >> > > > > defence, I
> > >> > > > > > > > > would say.
> > >> > > > > > > > >
> > >> > > > > > > > > One way for customers to validate their regex before
> > >> > upgrading
> > >> > > > > their
> > >> > > > > > > > > prod would be to test them with another group. For
> > >> instance,
> > >> > > that
> > >> > > > > > > > > could be done in a pre-prod environment. Another way
> > >> would be
> > >> > > to
> > >> > > > > > > > > extend the consumer-group tool to provide a regex
> > >> validation
> > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > >> > > > > > > > >
> > >> > > > > > > > > > 5. Related to the last question, will the Java client
> > >> allow
> > >> > > the
> > >> > > > > > > users to
> > >> > > > > > > > > > stick with the current regex engine for compatibility
> > >> > > reasons?
> > >> > > > > For
> > >> > > > > > > > > example,
> > >> > > > > > > > > > it may be handy to keep using client based regex at
> > >> first
> > >> > to
> > >> > > keep
> > >> > > > > > > > > > migrations simple and then migrate to server based
> > >> regexes
> > >> > > as a
> > >> > > > > > > second
> > >> > > > > > > > > step.
> > >> > > > > > > > >
> > >> > > > > > > > > I understand your point but I am concerned that this
> > would
> > >> > > allow
> > >> > > > > users
> > >> > > > > > > > > to actually stay in this mode. That would go against our
> > >> goal
> > >> > > of
> > >> > > > > > > > > simplifying the client because we would have to continue
> > >> > > monitoring
> > >> > > > > > > > > the metadata on the client side. I would rather not do
> > >> this.
> > >> > > > > > > > >
> > >> > > > > > > > > > 6. When we say that the group coordinator will be
> > >> > > responsible for
> > >> > > > > > > storing
> > >> > > > > > > > > > the configurations and that the configurations will be
> > >> > > deleted
> > >> > > > > when
> > >> > > > > > > the
> > >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> > >> > deletion
> > >> > > of
> > >> > > > > > > > > > configurations?
> > >> > > > > > > > >
> > >> > > > > > > > > That's right. The configurations will be deleted when
> > the
> > >> > > group is
> > >> > > > > > > > > deleted. They go together.
> > >> > > > > > > > >
> > >> > > > > > > > > > 7. Will the choice to store the configs in the group
> > >> > > coordinator
> > >> > > > > > > make it
> > >> > > > > > > > > > harder to list all cluster configs and their values?
> > >> > > > > > > > >
> > >> > > > > > > > > I don't think so. The group configurations are overrides
> > >> of
> > >> > > cluster
> > >> > > > > > > > > configs. If you want to know all the overrides though,
> > you
> > >> > > would
> > >> > > > > have
> > >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> > the
> > >> > > metadata
> > >> > > > > log
> > >> > > > > > > > > for instance.
> > >> > > > > > > > >
> > >> > > > > > > > > > 8. How would someone configure a group before starting
> > >> the
> > >> > > > > consumers?
> > >> > > > > > > > > Have
> > >> > > > > > > > > > we considered allowing the explicit creation of
> > groups?
> > >> > > > > > > Alternatively,
> > >> > > > > > > > > the
> > >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> > >> > > > > > > > >
> > >> > > > > > > > > Yes. The group will be automatically created in this
> > case.
> > >> > > However,
> > >> > > > > > > > > the configs will be lost after the retention period of
> > the
> > >> > > group
> > >> > > > > > > > > passes.
> > >> > > > > > > > >
> > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > >> client
> > >> > > still
> > >> > > > > take
> > >> > > > > > > a
> > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> > an
> > >> > > overload?
> > >> > > > > > > > >
> > >> > > > > > > > > That's a very group question. I forgot about that one.
> > As
> > >> the
> > >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> > the
> > >> > > engine
> > >> > > > > that
> > >> > > > > > > > > we plan to use, it might be better to deprecate it and
> > >> use an
> > >> > > > > overload
> > >> > > > > > > > > which takes a string. We would rely on the server side
> > >> > > validation.
> > >> > > > > > > > > During the migration, I think that we could still try to
> > >> > > toString
> > >> > > > > the
> > >> > > > > > > > > regex and use it. That should work, I think, in the
> > >> majority
> > >> > > of the
> > >> > > > > > > > > cases.
> > >> > > > > > > > >
> > >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> > about
> > >> > the
> > >> > > > > reason
> > >> > > > > > > to
> > >> > > > > > > > > > switch to IBP/metadata.version from the feature flag.
> > >> Maybe
> > >> > > we
> > >> > > > > mean
> > >> > > > > > > that
> > >> > > > > > > > > we
> > >> > > > > > > > > > can switch the default for the feature flag to true
> > >> based
> > >> > on
> > >> > > the
> > >> > > > > > > > > > metadata.version once we want to make it the default.
> > >> > > > > > > > >
> > >> > > > > > > > > My plan was to use that feature flag mainly during the
> > >> > > development
> > >> > > > > > > > > phase. I should not have mentioned it, I think, because
> > we
> > >> > > could
> > >> > > > > use
> > >> > > > > > > > > an internal config for it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > required
> > >> > > ACLs, it
> > >> > > > > > > would
> > >> > > > > > > > > be
> > >> > > > > > > > > > good to add that for consistency.
> > >> > > > > > > > >
> > >> > > > > > > > > Noted.
> > >> > > > > > > > >
> > >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > requires
> > >> > > "Read
> > >> > > > > Group"
> > >> > > > > > > > > even
> > >> > > > > > > > > > though it seems to do more than reading.
> > >> > > > > > > > >
> > >> > > > > > > > > I agree. This is how the current protocol works though.
> > We
> > >> > only
> > >> > > > > > > > > require "Read Group" to join a group. We could consider
> > >> > > changing
> > >> > > > > this
> > >> > > > > > > > > but I am not sure that it is worth it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> > with
> > >> > the
> > >> > > new
> > >> > > > > > > group
> > >> > > > > > > > > > protocol? It would be good to have a section on this.
> > >> > > > > > > > >
> > >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> > >> have a
> > >> > > new
> > >> > > > > > > > > topic id so it will treat it like a topic with a
> > different
> > >> > > name.
> > >> > > > > The
> > >> > > > > > > > > only issue is that the fetch/commit offsets APIs do not
> > >> > support
> > >> > > > > topic
> > >> > > > > > > > > IDs so the consumer would reuse the offsets based on the
> > >> > same.
> > >> > > I
> > >> > > > > think
> > >> > > > > > > > > that we should update those APIs as well in order to be
> > >> > > consistent
> > >> > > > > end
> > >> > > > > > > > > to end. That would strengthen the semantics of the
> > >> consumer.
> > >> > > > > > > > >
> > >> > > > > > > > > > 14. The KIP mentions we will write the new coordinator
> > >> in
> > >> > > Java.
> > >> > > > > Even
> > >> > > > > > > > > though
> > >> > > > > > > > > > this is an implementation detail, do we plan to have a
> > >> new
> > >> > > gradle
> > >> > > > > > > module
> > >> > > > > > > > > > for it?
> > >> > > > > > > > >
> > >> > > > > > > > > Yes.
> > >> > > > > > > > >
> > >> > > > > > > > > > 15. Do we have a scalability goal when it comes to how
> > >> many
> > >> > > > > members
> > >> > > > > > > the
> > >> > > > > > > > > new
> > >> > > > > > > > > > group protocol can support?
> > >> > > > > > > > >
> > >> > > > > > > > > We don't have numbers at the moment. The protocol should
> > >> > > support
> > >> > > > > 1000s
> > >> > > > > > > > > of members per group. We will measure this when we have
> > a
> > >> > first
> > >> > > > > > > > > implementation. Note that we might have other
> > bottlenecks
> > >> > down
> > >> > > the
> > >> > > > > > > > > road (e.g. offset commits).
> > >> > > > > > > > >
> > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds instead
> > of
> > >> > > > > > > > > > SubscribedTopicNames in ConsumerGroupHeartbeatRequest?
> > >> Is
> > >> > the
> > >> > > > > idea
> > >> > > > > > > that
> > >> > > > > > > > > > since we have to resolve the regex on the server, we
> > >> can do
> > >> > > the
> > >> > > > > same
> > >> > > > > > > for
> > >> > > > > > > > > > the topic name? The difference is that sending the
> > >> regex is
> > >> > > more
> > >> > > > > > > > > efficient
> > >> > > > > > > > > > whereas sending the topic names is less efficient.
> > >> > > Furthermore,
> > >> > > > > > > delete
> > >> > > > > > > > > and
> > >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> > >> > > > > > > > >
> > >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> > >> server
> > >> > > for
> > >> > > > > both
> > >> > > > > > > > > paths but I do agree with your point. As a second
> > though,
> > >> > using
> > >> > > > > topic
> > >> > > > > > > > > ids may be better here for the delete and recreation
> > case.
> > >> > > Also, I
> > >> > > > > > > > > suppose that we may allow users to subscribe with topic
> > >> ids
> > >> > in
> > >> > > the
> > >> > > > > > > > > future because that is the only way to be really robust
> > to
> > >> > > topic
> > >> > > > > > > > > re-creation.
> > >> > > > > > > > >
> > >> > > > > > > > > Best,
> > >> > > > > > > > > David
> > >> > > > > > > > >
> > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > >> > > djacot@confluent.io>
> > >> > > > > > > wrote:
> > >> > > > > > > > > >
> > >> > > > > > > > > > Hi Justine,
> > >> > > > > > > > > >
> > >> > > > > > > > > > Thanks for your comments. Please find my answers
> > below.
> > >> > > > > > > > > >
> > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> > >> > > exception
> > >> > > > > of the
> > >> > > > > > > > > > topic names based in the
> > ConsumerGroupHeartbeatRequest.
> > >> I
> > >> > am
> > >> > > not
> > >> > > > > sure
> > >> > > > > > > > > > if using topic names is the right call here. I need to
> > >> > think
> > >> > > > > about it
> > >> > > > > > > > > > a little more. Obviously, the KIP does not change the
> > >> > > > > fetch/commit
> > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> > >> that
> > >> > we
> > >> > > > > should
> > >> > > > > > > > > > include though as it would give better overall
> > >> guarantee in
> > >> > > the
> > >> > > > > > > > > > producer.
> > >> > > > > > > > > > - You're right. I think that I should not have
> > mentioned
> > >> > this
> > >> > > > > flag at
> > >> > > > > > > > > > all. I will remove it. We can use an internal
> > >> configuration
> > >> > > while
> > >> > > > > > > > > > developing the feature.
> > >> > > > > > > > > > - Both cluster types will be supported. The change is
> > >> > > > > orthogonal. The
> > >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> > >> > > > > > > > > >
> > >> > > > > > > > > > Best,
> > >> > > > > > > > > > David
> > >> > > > > > > > > >
> > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> > >> > > > > wangguoz@gmail.com>
> > >> > > > > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Hi Ismael,
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> > inlined
> > >> > > below:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > >> > > ismael@juma.me.uk>
> > >> > > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > Thanks for the KIP. This has the potential to be a
> > >> > great
> > >> > > > > > > > > improvement. A few
> > >> > > > > > > > > > > > initial questions/comments:
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > 1. I think it's premature to talk about target
> > >> versions
> > >> > > for
> > >> > > > > > > > > deprecation and
> > >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> > >> KRaft,
> > >> > > this
> > >> > > > > > > affects a
> > >> > > > > > > > > core
> > >> > > > > > > > > > > > client protocol and hence deprecation/removal will
> > >> be
> > >> > > heavily
> > >> > > > > > > > > dependent on
> > >> > > > > > > > > > > > how quickly applications migrate to the new
> > >> protocol.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> > >> proposed
> > >> > > > > timeline
> > >> > > > > > > in
> > >> > > > > > > > > the
> > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan` and
> > >> > > instead
> > >> > > > > just
> > >> > > > > > > state
> > >> > > > > > > > > > > that we will decide in the future about when we
> > would
> > >> > > > > deprecate old
> > >> > > > > > > > > > > protocol and behaviors.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> > >> but
> > >> > it
> > >> > > > > wasn't
> > >> > > > > > > made
> > >> > > > > > > > > clear
> > >> > > > > > > > > > > > why. If we added that as a way to estimate when
> > we'd
> > >> > > > > deprecate
> > >> > > > > > > and
> > >> > > > > > > > > remove
> > >> > > > > > > > > > > > the group protocol, I also suggest removing this
> > >> part.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I think that's not specifically related to the
> > >> > > > > deprecation/removal
> > >> > > > > > > > > timeline
> > >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> > >> > > broker-side
> > >> > > > > > > > > > > implementation may be done first, and then the
> > client
> > >> > side,
> > >> > > > > and we
> > >> > > > > > > > > would
> > >> > > > > > > > > > > only mark it as "released" by the time clients
> > >> > > implementations
> > >> > > > > are
> > >> > > > > > > > > done. At
> > >> > > > > > > > > > > that time, to enable the feature the clients need to
> > >> > first
> > >> > > > > swap-in
> > >> > > > > > > the
> > >> > > > > > > > > > > bytecode with a rolling bounce and then set the flag
> > >> > with a
> > >> > > > > second
> > >> > > > > > > > > rolling
> > >> > > > > > > > > > > bounce, and hence we feel it's better to be released
> > >> in a
> > >> > > major
> > >> > > > > > > > > version.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > migration
> > >> > > story.
> > >> > > > > It
> > >> > > > > > > > > sounds like
> > >> > > > > > > > > > > > we're saying we will support online migrations. Is
> > >> that
> > >> > > > > correct?
> > >> > > > > > > We
> > >> > > > > > > > > should
> > >> > > > > > > > > > > > explain this in detail. It could also be done as a
> > >> > > separate
> > >> > > > > KIP,
> > >> > > > > > > if
> > >> > > > > > > > > it's
> > >> > > > > > > > > > > > easier.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes I think that's the part we can be more concrete
> > >> about
> > >> > > for
> > >> > > > > sure
> > >> > > > > > > (and
> > >> > > > > > > > > > > this is related to your question 2) above). We will
> > >> work
> > >> > on
> > >> > > > > making
> > >> > > > > > > it
> > >> > > > > > > > > more
> > >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> > >> > > subscriptions
> > >> > > > > to
> > >> > > > > > > the
> > >> > > > > > > > > server,
> > >> > > > > > > > > > > > but it seems like there could be some tricky
> > >> > > compatibility
> > >> > > > > > > issues.
> > >> > > > > > > > > Will we
> > >> > > > > > > > > > > > have a mechanism for users to detect that they
> > need
> > >> to
> > >> > > update
> > >> > > > > > > their
> > >> > > > > > > > > regex
> > >> > > > > > > > > > > > before switching to the new protocol?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes I think we need some tooling for non-java client
> > >> > users
> > >> > > to
> > >> > > > > sort
> > >> > > > > > > of
> > >> > > > > > > > > > > "dry-run" the client before switching to the new
> > >> > protocol.
> > >> > > I
> > >> > > > > do not
> > >> > > > > > > > > have a
> > >> > > > > > > > > > > specific idea on top of my head though, maybe others
> > >> like
> > >> > > @Matt
> > >> > > > > > > > > Howlett can
> > >> > > > > > > > > > > chime-in here?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 5. Related to the last question, will the Java
> > >> client
> > >> > > allow
> > >> > > > > the
> > >> > > > > > > > > users to
> > >> > > > > > > > > > > > stick with the current regex engine for
> > >> compatibility
> > >> > > > > reasons?
> > >> > > > > > > For
> > >> > > > > > > > > example,
> > >> > > > > > > > > > > > it may be handy to keep using client based regex
> > at
> > >> > > first to
> > >> > > > > keep
> > >> > > > > > > > > > > > migrations simple and then migrate to server based
> > >> > > regexes
> > >> > > > > as a
> > >> > > > > > > > > second
> > >> > > > > > > > > > > > step.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Honestly I have not thought about that for java
> > >> clients,
> > >> > > and
> > >> > > > > we can
> > >> > > > > > > > > discuss
> > >> > > > > > > > > > > that. What kind of compatibility issues do you have
> > in
> > >> > > mind?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 6. When we say that the group coordinator will be
> > >> > > > > responsible for
> > >> > > > > > > > > storing
> > >> > > > > > > > > > > > the configurations and that the configurations
> > will
> > >> be
> > >> > > > > deleted
> > >> > > > > > > when
> > >> > > > > > > > > the
> > >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> > trigger
> > >> > > deletion
> > >> > > > > of
> > >> > > > > > > > > > > > configurations?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> > would
> > >> > only
> > >> > > > > > > transit to
> > >> > > > > > > > > that
> > >> > > > > > > > > > > state when the group is EMPTY and also all of its
> > >> > metadata
> > >> > > are
> > >> > > > > > > gone),
> > >> > > > > > > > > once
> > >> > > > > > > > > > > it's transited to DEAD this group would never be
> > >> revived.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> > group
> > >> > > > > coordinator
> > >> > > > > > > > > make it
> > >> > > > > > > > > > > > harder to list all cluster configs and their
> > values?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > That's a good question, and our thoughts are that
> > the
> > >> > > so-called
> > >> > > > > > > "group
> > >> > > > > > > > > > > configurations" are overrides of the cluster-level
> > >> > > > > configurations
> > >> > > > > > > > > > > customized per group so when an admin list cluster
> > >> > configs
> > >> > > it's
> > >> > > > > > > okay to
> > >> > > > > > > > > > > list just the cluster-level "defaults", not showing
> > >> any
> > >> > > > > per-group
> > >> > > > > > > > > > > customizations.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 8. How would someone configure a group before
> > >> starting
> > >> > > the
> > >> > > > > > > > > consumers? Have
> > >> > > > > > > > > > > > we considered allowing the explicit creation of
> > >> groups?
> > >> > > > > > > > > Alternatively, the
> > >> > > > > > > > > > > > configs could be decoupled from the group
> > lifecycle.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > The configs can be created before the group itself
> > as
> > >> an
> > >> > > > > > > independent
> > >> > > > > > > > > entity
> > >> > > > > > > > > > > --- of course, this requires the corresponding
> > >> request to
> > >> > > be
> > >> > > > > > > routed to
> > >> > > > > > > > > the
> > >> > > > > > > > > > > right coordinator based on the group id --- the only
> > >> > thing
> > >> > > that
> > >> > > > > > > > > differs is,
> > >> > > > > > > > > > > when the group itself is gone we also check if there
> > >> are
> > >> > > any
> > >> > > > > > > > > configuration
> > >> > > > > > > > > > > entities related to that group and delete as well.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> > the
> > >> > > creation
> > >> > > > > /
> > >> > > > > > > > > deletion
> > >> > > > > > > > > > > lifecycles of the config entities, and we would like
> > >> to
> > >> > > hear
> > >> > > > > > > everyone's
> > >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> > >> totally
> > >> > > > > decouple
> > >> > > > > > > group
> > >> > > > > > > > > > > configs and hence not delete them at all when the
> > >> group
> > >> > is
> > >> > > > > gone,
> > >> > > > > > > but
> > >> > > > > > > > > always
> > >> > > > > > > > > > > require explicit deletion operations by themselves.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > >> > client
> > >> > > > > still
> > >> > > > > > > take
> > >> > > > > > > > > a
> > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > >> introduce an
> > >> > > > > overload?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I think we do not need to introduce an overload, but
> > >> I'm
> > >> > > all
> > >> > > > > ears
> > >> > > > > > > if
> > >> > > > > > > > > there
> > >> > > > > > > > > > > may be some compatibility issues that we may
> > overlook.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 10. I agree with Justine that we should be clearer
> > >> > about
> > >> > > the
> > >> > > > > > > reason
> > >> > > > > > > > > to
> > >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> > >> flag.
> > >> > > Maybe
> > >> > > > > we
> > >> > > > > > > mean
> > >> > > > > > > > > that we
> > >> > > > > > > > > > > > can switch the default for the feature flag to
> > true
> > >> > > based on
> > >> > > > > the
> > >> > > > > > > > > > > > metadata.version once we want to make it the
> > >> default.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > >> required
> > >> > > ACLs,
> > >> > > > > it
> > >> > > > > > > > > would be
> > >> > > > > > > > > > > > good to add that for consistency.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Ack, we can certainly do that.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > >> > requires
> > >> > > > > "Read
> > >> > > > > > > > > Group" even
> > >> > > > > > > > > > > > though it seems to do more than reading.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I had that thought myself as well, but in the end we
> > >> > could
> > >> > > not
> > >> > > > > > > find a
> > >> > > > > > > > > > > better alternative: adding Write Group seems an
> > >> overkill
> > >> > > here
> > >> > > > > > > since we
> > >> > > > > > > > > do
> > >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> > and
> > >> > > Describe
> > >> > > > > on
> > >> > > > > > > > > groups so
> > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > consumer
> > >> > with
> > >> > > the
> > >> > > > > new
> > >> > > > > > > > > group
> > >> > > > > > > > > > > > protocol? It would be good to have a section on
> > >> this.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> > >> add a
> > >> > > > > section
> > >> > > > > > > about
> > >> > > > > > > > > > > that, but basically the idea is that consumer would
> > be
> > >> > > totally
> > >> > > > > > > > > agnostic in
> > >> > > > > > > > > > > the new protocol as it's handled all by the brokers.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > >> coordinator
> > >> > in
> > >> > > > > Java.
> > >> > > > > > > Even
> > >> > > > > > > > > though
> > >> > > > > > > > > > > > this is an implementation detail, do we plan to
> > >> have a
> > >> > > new
> > >> > > > > gradle
> > >> > > > > > > > > module
> > >> > > > > > > > > > > > for it?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > We have not thought about that. But I think the
> > answer
> > >> > > should
> > >> > > > > be
> > >> > > > > > > yes.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes to
> > >> how
> > >> > > many
> > >> > > > > > > members
> > >> > > > > > > > > the new
> > >> > > > > > > > > > > > group protocol can support?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Within a group, I think we should shoot for 1000s of
> > >> > > members.
> > >> > > > > But
> > >> > > > > > > that
> > >> > > > > > > > > > > scalability goals also depend on the offset
> > management
> > >> > > (commit,
> > >> > > > > > > fetch)
> > >> > > > > > > > > > > capabilities of the coordinator which we did not
> > >> cover in
> > >> > > this
> > >> > > > > > > KIP, so
> > >> > > > > > > > > it's
> > >> > > > > > > > > > > hard to give a number that applies universally.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > >> instead
> > >> > of
> > >> > > > > > > > > > > > SubscribedTopicNames in
> > >> ConsumerGroupHeartbeatRequest?
> > >> > > Is the
> > >> > > > > > > idea
> > >> > > > > > > > > that
> > >> > > > > > > > > > > > since we have to resolve the regex on the server,
> > we
> > >> > can
> > >> > > do
> > >> > > > > the
> > >> > > > > > > same
> > >> > > > > > > > > for
> > >> > > > > > > > > > > > the topic name? The difference is that sending the
> > >> > regex
> > >> > > is
> > >> > > > > more
> > >> > > > > > > > > efficient
> > >> > > > > > > > > > > > whereas sending the topic names is less efficient.
> > >> > > > > Furthermore,
> > >> > > > > > > > > delete and
> > >> > > > > > > > > > > > recreation is easier to handle if we have topic
> > ids.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > The main reason to still let the clients send names
> > >> is to
> > >> > > keep
> > >> > > > > the
> > >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> > client
> > >> > > only.
> > >> > > > > Note
> > >> > > > > > > that
> > >> > > > > > > > > > > although we added topic id in KIP-516, we never
> > >> > > implemented the
> > >> > > > > > > logic
> > >> > > > > > > > > on
> > >> > > > > > > > > > > consumer/producers leveraging the related newer
> > >> versioned
> > >> > > RPCs,
> > >> > > > > > > > > instead we
> > >> > > > > > > > > > > just set the topic id as empty UUID. We want to keep
> > >> the
> > >> > > > > > > > > consumer/producer
> > >> > > > > > > > > > > to be thin and only delegate the reasoning on broker
> > >> and
> > >> > > > > > > potentially
> > >> > > > > > > > > admin
> > >> > > > > > > > > > > clients.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > Thanks,
> > >> > > > > > > > > > > > Ismael
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> > >> > > > > > > > > <dj...@confluent.io.invalid>
> > >> > > > > > > > > > > > wrote:
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > > Hi all,
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > I would like to start a discussion thread on
> > >> KIP-848:
> > >> > > The
> > >> > > > > Next
> > >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> > >> With
> > >> > > this
> > >> > > > > KIP,
> > >> > > > > > > we
> > >> > > > > > > > > aim
> > >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> > >> more
> > >> > > > > reliable,
> > >> > > > > > > more
> > >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> > >> easier
> > >> > > to
> > >> > > > > debug
> > >> > > > > > > for
> > >> > > > > > > > > > > > > operators.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > The KIP is here:
> > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > Please take a look and let me know what you
> > think.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > Best,
> > >> > > > > > > > > > > > > David
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > PS: I will be away from July 18th to August 8th.
> > >> That
> > >> > > gives
> > >> > > > > > > you a
> > >> > > > > > > > > bit
> > >> > > > > > > > > > > > > of time to read and digest this long KIP.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > --
> > >> > > > > > > > > > > -- Guozhang
> > >> > > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > > --
> > >> > > > > > > > -- Guozhang
> > >> > > > > > >
> > >> > > > >
> > >> > >
> > >> > >
> > >> >
> > >>
> > >>
> > >> --
> > >> -- Guozhang
> > >>
> > >
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[David Jacot <dj...@confluent.io.INVALID>]

Hi Gregory,

Thanks for your feedback. Please find my answers below.

> 1. The 'Rejected Alternatives' section describes how member epoch should
> advance in step with the group epoch and assignment epoch values. I think
> that this is a good idea for the reasons described in the KIP. When the
> protocol is incrementally assigning partitions to a worker, what member
> epoch does each incremental assignment use? Are member epochs re-used, and
> a single member epoch can correspond to multiple different (monotonically
> larger) assignments?

Those are good questions. The member epoch is aligned to the
group/assignment. A member transitions to its new target epoch when it
has revoked the partitions that he no longer owns. It has to revoke
partitions with its current epoch. When the revocation is completed,
it immediately gets its new epoch with the assigned partitions which
are already free to be assigned. When the remaining partitions are
freed up by the other members, the coordinator assigns them to the
member while keeping the same member epoch. In other words, a member
can revoke partitions or can get new partitions while having the same
epoch.

> 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> not, should custom client-side assignor implementations interact with the
> Reason field, and how is its common meaning agreed upon? If so, what is the
> benefit of a distinct Reason field over including such functionality in the
> opaque metadata?

It is opaque to a certain extent. Any non zero values are considered
as errors but the non zero values are not specified in the protocol.
This is left to the client-side assignor implementation. There are two
benefits here: 1) The reason is used by the coordinator to do the
validation. If non-zero, the assignment should not change. If zero,
the assignment must be valid. 2) It allows the operator to see if the
assignment was successful without having to deserialize the metadata.

> 3. The following is included in the KIP: "Thanks to this, the input of the
> client side assignor is entirely driven by the group coordinator. The
> consumer is no longer responsible for maintaining any state besides its
> assigned partitions." Does this mean that the client-side assignor MAY
> incorporate additional non-Metadata state (such as partition throughput,
> cpu/memory metrics, config topics, etc), or that additional non-Metadata
> state SHOULD NOT be used?

Here, I was referring to the canonical case. Today the KafkaConsumer
tracks metadata (e.g. partitions) and feeds the assignor with it. With
the new protocol, we want the assignor to get all its information from
the coordinator in order to avoid metadata divergences. You can
definitely do whatever you want in a custom assignor. You could use
metadata propagated via the protocol itself in the metadata field or
use external metadata sources as well.

> 4. I see that there are separate classes
> for org.apache.kafka.server.group.consumer.PartitionAssignor
> and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> overlap significantly. Is it possible for these two implementations to be
> unified? This would serve to promote feature parity of server-side and
> client-side assignors, and would also facilitate operational flexibility in
> certain situations. For example, if a server-side assignor has some poor
> behavior and needs a patch, deploying the patched assignor to the client
> and switching one consumer group to a client-side assignor may be faster
> and less risky than patching all of the brokers. With the currently
> proposed distinct APIs, a non-trivial reimplementation would have to be
> assembled, and if the two APIs have diverged significantly, then it is
> possible that a reimplementation would not be possible.

That's a good question. I went with two different interfaces because
the server side assignor does not have a notion of custom metadata and
is thus simpler. I thought that it is misleading to reuse the
client-side interface in this case due to this.

Best,
David

On Wed, Aug 3, 2022 at 10:05 PM Gregory Harris <gh...@gmail.com> wrote:
>
> Hey All!
>
> Thanks for the KIP, it's wonderful to see cooperative rebalancing making it
> down the stack!
>
> I had a few questions:
>
> 1. The 'Rejected Alternatives' section describes how member epoch should
> advance in step with the group epoch and assignment epoch values. I think
> that this is a good idea for the reasons described in the KIP. When the
> protocol is incrementally assigning partitions to a worker, what member
> epoch does each incremental assignment use? Are member epochs re-used, and
> a single member epoch can correspond to multiple different (monotonically
> larger) assignments?
>
> 2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
> not, should custom client-side assignor implementations interact with the
> Reason field, and how is its common meaning agreed upon? If so, what is the
> benefit of a distinct Reason field over including such functionality in the
> opaque metadata?
>
> 3. The following is included in the KIP: "Thanks to this, the input of the
> client side assignor is entirely driven by the group coordinator. The
> consumer is no longer responsible for maintaining any state besides its
> assigned partitions." Does this mean that the client-side assignor MAY
> incorporate additional non-Metadata state (such as partition throughput,
> cpu/memory metrics, config topics, etc), or that additional non-Metadata
> state SHOULD NOT be used?
>
> 4. I see that there are separate classes
> for org.apache.kafka.server.group.consumer.PartitionAssignor
> and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
> overlap significantly. Is it possible for these two implementations to be
> unified? This would serve to promote feature parity of server-side and
> client-side assignors, and would also facilitate operational flexibility in
> certain situations. For example, if a server-side assignor has some poor
> behavior and needs a patch, deploying the patched assignor to the client
> and switching one consumer group to a client-side assignor may be faster
> and less risky than patching all of the brokers. With the currently
> proposed distinct APIs, a non-trivial reimplementation would have to be
> assembled, and if the two APIs have diverged significantly, then it is
> possible that a reimplementation would not be possible.
>
> --
> Greg Harris
> gharris1727@gmail.com
> github.com/gharris1727
>
> On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com> wrote:
>
> > Hi Guozhang/David,
> >
> > I created a confluence page to discuss how Connect would need to change
> > based on the new rebalance protocol. Here's the page:
> >
> > https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
> >
> > It's also pretty longish and I have tried to keep a format similar to
> > KIP-848. Let me know what you think. Also, do you think this should be
> > moved to a separate discussion thread or is this one fine?
> >
> > Thanks!
> > Sagar.
> >
> > On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
> >
> > > Hello Guozhang,
> > >
> > > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > > analysis and come up with such a document.
> > >
> > > Thanks!
> > > Sagar.
> > >
> > > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> > wrote:
> > >
> > >> Hello Sagar,
> > >>
> > >> It would be great if you could come back with some analysis on how to
> > >> implement the Connect side integration with the new protocol; so far
> > >> besides leveraging on the new "protocol type" we did not yet think
> > through
> > >> the Connect side implementations. For Streams here's a draft of
> > >> integration
> > >> plan:
> > >>
> > >>
> > https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> > >> just FYI for your analysis on Connect.
> > >>
> > >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> > wrote:
> > >>
> > >> > Hi David,
> > >> >
> > >> > Thank you for your response. The reason I thought connect can also fit
> > >> into
> > >> > this new scheme is that even today the connect uses a
> > WorkerCoordinator
> > >> > extending from AbstractCoordinator to empower rebalances of
> > >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> > >> connect
> > >> > and uses the metadata() method by plumbing into
> > >> JoinGroupRequestProtocol.
> > >> >
> > >> > I think the changes to support connect would be similar at a high
> > level
> > >> to
> > >> > the changes in streams mainly because of the Client side assignors
> > being
> > >> > used in both. At an implementation level, we might need to make a lot
> > of
> > >> > changes to get onto this new assignment protocol like enhancing the
> > >> > JoinGroup request/response and SyncGroup and using
> > >> ConsumerGroupHeartbeat
> > >> > API etc again on similar lines to streams (or there might be
> > >> deviations). I
> > >> > would try to perform a detailed analysis of the same  and we can have
> > a
> > >> > separate discussion thread for that as that would derail this
> > discussion
> > >> > thread. Let me know if that sounds good to you.
> > >> >
> > >> > Thanks!
> > >> > Sagar.
> > >> >
> > >> >
> > >> >
> > >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> > <djacot@confluent.io.invalid
> > >> >
> > >> > wrote:
> > >> >
> > >> > > Hi Sagar,
> > >> > >
> > >> > > Thanks for your comments.
> > >> > >
> > >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> > >> > >
> > >> > > 2) The idea is to transition C from his current assignment to his
> > >> > > target assignment when he can move to epoch 3. When that happens,
> > the
> > >> > > member assignment is updated and persisted with all its assigned
> > >> > > partitions even if they are not all revoked yet. In other words, the
> > >> > > member assignment becomes the target assignment. This is basically
> > an
> > >> > > optimization to avoid having to write all the changes to the log.
> > The
> > >> > > examples are based on the persisted state so I understand the
> > >> > > confusion. Let me see if I can improve this in the description.
> > >> > >
> > >> > > 3) Regarding Connect, it could reuse the protocol with a client side
> > >> > > assignor if it fits in the protocol. The assignment is about
> > >> > > topicid-partitions + metadata, could Connect fit into this?
> > >> > >
> > >> > > Best,
> > >> > > David
> > >> > >
> > >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> > >> wrote:
> > >> > > >
> > >> > > > Hi David,
> > >> > > >
> > >> > > > Thanks for the KIP. I just had minor observations:
> > >> > > >
> > >> > > > 1) In the Assignment Error section in Client Side mode Assignment
> > >> > > process,
> > >> > > > you mentioned => `In this case, the client side assignor can
> > return
> > >> an
> > >> > > > error to the group coordinator`. In this case are you referring to
> > >> the
> > >> > > > Assignor returning an AssignmentError that's listed down towards
> > the
> > >> > end?
> > >> > > > If yes, do you think it would make sense to mention this
> > explicitly
> > >> > here?
> > >> > > >
> > >> > > > 2) In the Case Studies section, I have a slight confusion, not
> > sure
> > >> if
> > >> > > > others have the same. Consider this step:
> > >> > > >
> > >> > > > When B heartbeats, the group coordinator transitions him to epoch
> > 3
> > >> > > because
> > >> > > > B has no partitions to revoke. It persists the change and reply.
> > >> > > >
> > >> > > >    - Group (epoch=3)
> > >> > > >       - A
> > >> > > >       - B
> > >> > > >       - C
> > >> > > >    - Target Assignment (epoch=3)
> > >> > > >       - A - partitions=[foo-0]
> > >> > > >       - B - partitions=[foo-2]
> > >> > > >       - C - partitions=[foo-1]
> > >> > > >    - Member Assignment
> > >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> > >> > > >       - B - epoch=3, partitions=[foo-2]
> > >> > > >       - C - epoch=3, partitions=[foo-1]
> > >> > > >
> > >> > > > When C heartbeats, it transitions to epoch 3 but cannot get foo-1
> > >> yet.
> > >> > > >
> > >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> > yet,
> > >> > but
> > >> > > > based on the description above, it seems it already has it. Do you
> > >> > think
> > >> > > it
> > >> > > > would be better to remove it and populate it only when it actually
> > >> gets
> > >> > > it?
> > >> > > > I see this in a lot of other places, so have I understood it
> > >> > incorrectly
> > >> > > ?
> > >> > > >
> > >> > > >
> > >> > > > Regarding connect , it might be out of scope of this discussion,
> > but
> > >> > from
> > >> > > > what I understood it would probably be running in client side
> > >> assignor
> > >> > > mode
> > >> > > > even on the new rebalance protocol as it has its own Custom
> > >> > > Assignors(Eager
> > >> > > > and IncrementalCooperative).
> > >> > > >
> > >> > > > Thanks!
> > >> > > >
> > >> > > > Sagar.
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > >
> > >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> > >> > <djacot@confluent.io.invalid
> > >> > > >
> > >> > > > wrote:
> > >> > > >
> > >> > > > > Thanks Hector! Our goal is to move forward with specialized API
> > >> > > > > instead of relying on one generic API. For Connect, we can apply
> > >> the
> > >> > > > > exact same pattern and reuse/share the core implementation on
> > the
> > >> > > > > server side. For the schema registry, I think that we should
> > >> consider
> > >> > > > > having a tailored API to do simple membership/leader election.
> > >> > > > >
> > >> > > > > Best,
> > >> > > > > David
> > >> > > > >
> > >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <ismael@juma.me.uk
> > >
> > >> > > wrote:
> > >> > > > > >
> > >> > > > > > Three quick comments:
> > >> > > > > >
> > >> > > > > > 1. Regarding java.util.regex.Pattern vs
> > >> com.google.re2j.Pattern, we
> > >> > > > > should
> > >> > > > > > document the differences in more detail before deciding one
> > way
> > >> or
> > >> > > > > another.
> > >> > > > > > That said, if people pass java.util.regex.Pattern, they expect
> > >> > their
> > >> > > > > > semantics to be honored. If we are doing something different,
> > >> then
> > >> > we
> > >> > > > > > should consider adding an overload with our own Pattern class
> > (I
> > >> > > don't
> > >> > > > > > think we'd want to expose re2j's at this point).
> > >> > > > > > 2. Regarding topic ids, any major new protocol should
> > integrate
> > >> > fully
> > >> > > > > with
> > >> > > > > > it and should handle the topic recreation case correctly.
> > That's
> > >> > the
> > >> > > main
> > >> > > > > > part we need to handle. I agree with David that we'd want to
> > add
> > >> > > topic
> > >> > > > > ids
> > >> > > > > > to the relevant protocols that don't have it yet and that we
> > can
> > >> > > probably
> > >> > > > > > focus on the internals versus adding new APIs to the Java
> > >> Consumer
> > >> > > > > (unless
> > >> > > > > > we find that adding new APIs is required for reasonable
> > >> semantics).
> > >> > > > > > 3. I am still not sure about the coordinator storing the
> > >> configs.
> > >> > > It's
> > >> > > > > > powerful for configs to be centralized in the metadata log for
> > >> > > various
> > >> > > > > > reasons (auditability, visibility, consistency, etc.).
> > >> Similarly, I
> > >> > > am
> > >> > > > > not
> > >> > > > > > sure about automatically deleting configs in a way that they
> > >> cannot
> > >> > > be
> > >> > > > > > recovered. A good property for modern systems is to minimize
> > the
> > >> > > number
> > >> > > > > of
> > >> > > > > > unrecoverable data loss scenarios.
> > >> > > > > >
> > >> > > > > > Ismael
> > >> > > > > >
> > >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> > >> > > <djacot@confluent.io.invalid
> > >> > > > > >
> > >> > > > > > wrote:
> > >> > > > > >
> > >> > > > > > > Thanks Guozhang. My answers are below:
> > >> > > > > > >
> > >> > > > > > > > 1) the migration path, especially the last step when
> > clients
> > >> > > flip the
> > >> > > > > > > flag
> > >> > > > > > > > to enable the new protocol, in which we would have a
> > window
> > >> > where
> > >> > > > > both
> > >> > > > > > > new
> > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > >> members
> > >> > of
> > >> > > the
> > >> > > > > same
> > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > >> while
> > >> > > > > using the
> > >> > > > > > > > new protocol is something we need to present about.
> > >> > > > > > >
> > >> > > > > > > Noted. I just published a new version of KIP which includes
> > >> more
> > >> > > > > > > details about this. See the "Supporting Online Consumer
> > Group
> > >> > > Upgrade"
> > >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan". I
> > >> think
> > >> > > that
> > >> > > > > > > I have to think through a few cases now but the overall idea
> > >> and
> > >> > > > > > > mechanism should be understandable.
> > >> > > > > > >
> > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > >> are
> > >> > > only
> > >> > > > > used
> > >> > > > > > > as
> > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > via
> > >> any
> > >> > > > > public
> > >> > > > > > > APIs
> > >> > > > > > > > to consumers yet. I think the question is, first should we
> > >> let
> > >> > > the
> > >> > > > > > > consumer
> > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > to
> > >> > fully
> > >> > > > > > > leverage
> > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > the
> > >> > > topic
> > >> > > > > ids;
> > >> > > > > > > and
> > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > in
> > >> the
> > >> > > > > consumer
> > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > >> the
> > >> > > > > rebalance
> > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > >> > > > > > >
> > >> > > > > > > a) Assuming that we would include converting all the offsets
> > >> > > related
> > >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> > >> able
> > >> > to
> > >> > > > > > > fully operate with topic ids. That being said, it still has
> > to
> > >> > > provide
> > >> > > > > > > the topics names in various APIs so having a mapping in the
> > >> > > consumer
> > >> > > > > > > seems inevitable to me.
> > >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> > >> this
> > >> > > goes
> > >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> > >> > internals
> > >> > > > > > > here and we can consider this separately if we see value in
> > >> doing
> > >> > > it.
> > >> > > > > > >
> > >> > > > > > > Coming back to Ismael's point about using topic ids in the
> > >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> > >> > advantage
> > >> > > in
> > >> > > > > > > favour of it. The consumer will have the opportunity to
> > >> validate
> > >> > > that
> > >> > > > > > > the topics exists before passing them into the group
> > rebalance
> > >> > > > > > > protocol. Obviously, the coordinator will also notice it but
> > >> it
> > >> > > does
> > >> > > > > > > not really have a way to reject an invalid topic in the
> > >> response.
> > >> > > > > > >
> > >> > > > > > > > I'm agreeing with David on all other minor questions
> > except
> > >> for
> > >> > > the
> > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > >> > > necessary
> > >> > > > > to
> > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > >> still
> > >> > > use
> > >> > > > > > > Pattern
> > >> > > > > > > > while just documenting that our subscription may be
> > >> rejected by
> > >> > > the
> > >> > > > > > > server.
> > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > >> > using
> > >> > > an
> > >> > > > > > > > overloaded `String` based subscription may be more
> > >> vulnerable
> > >> > to
> > >> > > > > various
> > >> > > > > > > > invalid regexes.
> > >> > > > > > >
> > >> > > > > > > That could work. I have to look at the differences between
> > the
> > >> > two
> > >> > > > > > > engines to better understand the potential issues. My
> > >> > > understanding is
> > >> > > > > > > that would work for all the basic regular expressions. The
> > >> > > differences
> > >> > > > > > > between the two are mainly about the various character
> > >> classes. I
> > >> > > > > > > wonder what other people think about this.
> > >> > > > > > >
> > >> > > > > > > Best,
> > >> > > > > > > David
> > >> > > > > > >
> > >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> > >> > wangguoz@gmail.com
> > >> > > >
> > >> > > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > Thanks David! I think on the high level there are two meta
> > >> > > points we
> > >> > > > > need
> > >> > > > > > > > to concretize a bit more:
> > >> > > > > > > >
> > >> > > > > > > > 1) the migration path, especially the last step when
> > clients
> > >> > > flip the
> > >> > > > > > > flag
> > >> > > > > > > > to enable the new protocol, in which we would have a
> > window
> > >> > where
> > >> > > > > both
> > >> > > > > > > new
> > >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> > >> members
> > >> > of
> > >> > > the
> > >> > > > > same
> > >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> > >> while
> > >> > > > > using the
> > >> > > > > > > > new protocol is something we need to present about.
> > >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> > >> are
> > >> > > only
> > >> > > > > used
> > >> > > > > > > as
> > >> > > > > > > > part of RPCs and admin client, but they are not exposed
> > via
> > >> any
> > >> > > > > public
> > >> > > > > > > APIs
> > >> > > > > > > > to consumers yet. I think the question is, first should we
> > >> let
> > >> > > the
> > >> > > > > > > consumer
> > >> > > > > > > > client to be maintaining the names -> ids mapping itself
> > to
> > >> > fully
> > >> > > > > > > leverage
> > >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> > the
> > >> > > topic
> > >> > > > > ids;
> > >> > > > > > > and
> > >> > > > > > > > secondly, should we ever consider exposing the topic ids
> > in
> > >> the
> > >> > > > > consumer
> > >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> > >> the
> > >> > > > > rebalance
> > >> > > > > > > > listener for cases like topic deletion-and-recreation).
> > >> > > > > > > >
> > >> > > > > > > > I'm agreeing with David on all other minor questions
> > except
> > >> for
> > >> > > the
> > >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> > >> > > necessary
> > >> > > > > to
> > >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> > >> still
> > >> > > use
> > >> > > > > > > Pattern
> > >> > > > > > > > while just documenting that our subscription may be
> > >> rejected by
> > >> > > the
> > >> > > > > > > server.
> > >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> > >> > using
> > >> > > an
> > >> > > > > > > > overloaded `String` based subscription may be more
> > >> vulnerable
> > >> > to
> > >> > > > > various
> > >> > > > > > > > invalid regexes.
> > >> > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > > Guozhang
> > >> > > > > > > >
> > >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> > >> > > > > <djacot@confluent.io.invalid
> > >> > > > > > > >
> > >> > > > > > > > wrote:
> > >> > > > > > > >
> > >> > > > > > > > > Hi Ismael,
> > >> > > > > > > > >
> > >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> > >> > inline.
> > >> > > > > > > > >
> > >> > > > > > > > > > 1. I think it's premature to talk about target
> > versions
> > >> for
> > >> > > > > > > deprecation
> > >> > > > > > > > > and
> > >> > > > > > > > > > removal of the existing group protocol. Unlike KRaft,
> > >> this
> > >> > > > > affects a
> > >> > > > > > > core
> > >> > > > > > > > > > client protocol and hence deprecation/removal will be
> > >> > heavily
> > >> > > > > > > dependent
> > >> > > > > > > > > on
> > >> > > > > > > > > > how quickly applications migrate to the new protocol.
> > >> > > > > > > > >
> > >> > > > > > > > > That makes sense. I will remove it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x, but
> > it
> > >> > > wasn't
> > >> > > > > made
> > >> > > > > > > > > clear
> > >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> > >> > > deprecate
> > >> > > > > and
> > >> > > > > > > remove
> > >> > > > > > > > > > the group protocol, I also suggest removing this part.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> > >> rewrite
> > >> > > the
> > >> > > > > group
> > >> > > > > > > > > coordinator in Java while we implement the new protocol.
> > >> This
> > >> > > means
> > >> > > > > > > > > that the internals will be slightly different (e.g.
> > >> threading
> > >> > > > > model).
> > >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> > >> group
> > >> > > > > > > > > coordinator to the new group coordinator to a major
> > >> release.
> > >> > > The
> > >> > > > > > > > > alternative would be to use a flag to do the switch
> > >> instead
> > >> > of
> > >> > > > > relying
> > >> > > > > > > > > on the software upgrade.
> > >> > > > > > > > >
> > >> > > > > > > > > > 3. We need to flesh out the details of the migration
> > >> story.
> > >> > > It
> > >> > > > > sounds
> > >> > > > > > > > > like
> > >> > > > > > > > > > we're saying we will support online migrations. Is
> > that
> > >> > > correct?
> > >> > > > > We
> > >> > > > > > > > > should
> > >> > > > > > > > > > explain this in detail. It could also be done as a
> > >> separate
> > >> > > KIP,
> > >> > > > > if
> > >> > > > > > > it's
> > >> > > > > > > > > > easier.
> > >> > > > > > > > >
> > >> > > > > > > > > Yes, we will support online migrations for the group.
> > That
> > >> > > means
> > >> > > > > that
> > >> > > > > > > > > a group using the old protocol will be able to switch to
> > >> the
> > >> > > new
> > >> > > > > > > > > protocol.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me briefly explain how that will work though. It is
> > >> > > basically a
> > >> > > > > > > > > four step process:
> > >> > > > > > > > >
> > >> > > > > > > > > 1. The cluster must be upgraded or rolled to a software
> > >> > > supporting
> > >> > > > > the
> > >> > > > > > > > > new group coordinator. Both the old and the new
> > >> coordinator
> > >> > > will
> > >> > > > > > > > > support the old protocol and rely on the same persisted
> > >> > > metadata so
> > >> > > > > > > > > they can work together. This point is an offline
> > >> migration.
> > >> > We
> > >> > > > > cannot
> > >> > > > > > > > > do this one live because it would require shutting down
> > >> the
> > >> > > current
> > >> > > > > > > > > coordinator and starting up the new one and that would
> > >> cause
> > >> > > > > > > > > unavailabilities.
> > >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> > to
> > >> X
> > >> > in
> > >> > > > > order
> > >> > > > > > > > > to enable the new protocol. This cannot be done before
> > 1)
> > >> is
> > >> > > > > > > > > terminated because the old coordinator doesn't support
> > the
> > >> > new
> > >> > > > > > > > > protocol.
> > >> > > > > > > > > 3. The consumers must be upgraded to a version
> > supporting
> > >> the
> > >> > > > > online
> > >> > > > > > > > > migration (must have KIP-792). If the consumer is
> > already
> > >> > > there.
> > >> > > > > > > > > Nothing must be done at this point.
> > >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> > >> turned
> > >> > > on.
> > >> > > > > The
> > >> > > > > > > > > consumer group is automatically converted when the first
> > >> > > consumer
> > >> > > > > > > > > using the new protocol joins the group. While the
> > members
> > >> > > using the
> > >> > > > > > > > > old protocol are being upgraded, the old protocol is
> > >> proxied
> > >> > > into
> > >> > > > > the
> > >> > > > > > > > > new one.
> > >> > > > > > > > >
> > >> > > > > > > > > Let me clarify all of this in the KIP.
> > >> > > > > > > > >
> > >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> > >> subscriptions
> > >> > > to
> > >> > > > > the
> > >> > > > > > > > > server,
> > >> > > > > > > > > > but it seems like there could be some tricky
> > >> compatibility
> > >> > > > > issues.
> > >> > > > > > > Will
> > >> > > > > > > > > we
> > >> > > > > > > > > > have a mechanism for users to detect that they need to
> > >> > update
> > >> > > > > their
> > >> > > > > > > regex
> > >> > > > > > > > > > before switching to the new protocol?
> > >> > > > > > > > >
> > >> > > > > > > > > I think that I am a bit more optimistic than you on this
> > >> > > point. I
> > >> > > > > > > > > believe that the majority of the cases are simple
> > regexes
> > >> > which
> > >> > > > > should
> > >> > > > > > > > > work with the new engine. The coordinator will verify
> > the
> > >> > regex
> > >> > > > > anyway
> > >> > > > > > > > > and reject the consumer if the regex is not valid.
> > Coming
> > >> > back
> > >> > > to
> > >> > > > > the
> > >> > > > > > > > > migration path, in the worst case, the first upgraded
> > >> > consumer
> > >> > > > > joining
> > >> > > > > > > > > the group will be rejected. This should be used as the
> > >> last
> > >> > > > > defence, I
> > >> > > > > > > > > would say.
> > >> > > > > > > > >
> > >> > > > > > > > > One way for customers to validate their regex before
> > >> > upgrading
> > >> > > > > their
> > >> > > > > > > > > prod would be to test them with another group. For
> > >> instance,
> > >> > > that
> > >> > > > > > > > > could be done in a pre-prod environment. Another way
> > >> would be
> > >> > > to
> > >> > > > > > > > > extend the consumer-group tool to provide a regex
> > >> validation
> > >> > > > > > > > > mechanism. Would this be enough in your opinion?
> > >> > > > > > > > >
> > >> > > > > > > > > > 5. Related to the last question, will the Java client
> > >> allow
> > >> > > the
> > >> > > > > > > users to
> > >> > > > > > > > > > stick with the current regex engine for compatibility
> > >> > > reasons?
> > >> > > > > For
> > >> > > > > > > > > example,
> > >> > > > > > > > > > it may be handy to keep using client based regex at
> > >> first
> > >> > to
> > >> > > keep
> > >> > > > > > > > > > migrations simple and then migrate to server based
> > >> regexes
> > >> > > as a
> > >> > > > > > > second
> > >> > > > > > > > > step.
> > >> > > > > > > > >
> > >> > > > > > > > > I understand your point but I am concerned that this
> > would
> > >> > > allow
> > >> > > > > users
> > >> > > > > > > > > to actually stay in this mode. That would go against our
> > >> goal
> > >> > > of
> > >> > > > > > > > > simplifying the client because we would have to continue
> > >> > > monitoring
> > >> > > > > > > > > the metadata on the client side. I would rather not do
> > >> this.
> > >> > > > > > > > >
> > >> > > > > > > > > > 6. When we say that the group coordinator will be
> > >> > > responsible for
> > >> > > > > > > storing
> > >> > > > > > > > > > the configurations and that the configurations will be
> > >> > > deleted
> > >> > > > > when
> > >> > > > > > > the
> > >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> > >> > deletion
> > >> > > of
> > >> > > > > > > > > > configurations?
> > >> > > > > > > > >
> > >> > > > > > > > > That's right. The configurations will be deleted when
> > the
> > >> > > group is
> > >> > > > > > > > > deleted. They go together.
> > >> > > > > > > > >
> > >> > > > > > > > > > 7. Will the choice to store the configs in the group
> > >> > > coordinator
> > >> > > > > > > make it
> > >> > > > > > > > > > harder to list all cluster configs and their values?
> > >> > > > > > > > >
> > >> > > > > > > > > I don't think so. The group configurations are overrides
> > >> of
> > >> > > cluster
> > >> > > > > > > > > configs. If you want to know all the overrides though,
> > you
> > >> > > would
> > >> > > > > have
> > >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> > the
> > >> > > metadata
> > >> > > > > log
> > >> > > > > > > > > for instance.
> > >> > > > > > > > >
> > >> > > > > > > > > > 8. How would someone configure a group before starting
> > >> the
> > >> > > > > consumers?
> > >> > > > > > > > > Have
> > >> > > > > > > > > > we considered allowing the explicit creation of
> > groups?
> > >> > > > > > > Alternatively,
> > >> > > > > > > > > the
> > >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> > >> > > > > > > > >
> > >> > > > > > > > > Yes. The group will be automatically created in this
> > case.
> > >> > > However,
> > >> > > > > > > > > the configs will be lost after the retention period of
> > the
> > >> > > group
> > >> > > > > > > > > passes.
> > >> > > > > > > > >
> > >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > >> client
> > >> > > still
> > >> > > > > take
> > >> > > > > > > a
> > >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> > an
> > >> > > overload?
> > >> > > > > > > > >
> > >> > > > > > > > > That's a very group question. I forgot about that one.
> > As
> > >> the
> > >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> > the
> > >> > > engine
> > >> > > > > that
> > >> > > > > > > > > we plan to use, it might be better to deprecate it and
> > >> use an
> > >> > > > > overload
> > >> > > > > > > > > which takes a string. We would rely on the server side
> > >> > > validation.
> > >> > > > > > > > > During the migration, I think that we could still try to
> > >> > > toString
> > >> > > > > the
> > >> > > > > > > > > regex and use it. That should work, I think, in the
> > >> majority
> > >> > > of the
> > >> > > > > > > > > cases.
> > >> > > > > > > > >
> > >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> > about
> > >> > the
> > >> > > > > reason
> > >> > > > > > > to
> > >> > > > > > > > > > switch to IBP/metadata.version from the feature flag.
> > >> Maybe
> > >> > > we
> > >> > > > > mean
> > >> > > > > > > that
> > >> > > > > > > > > we
> > >> > > > > > > > > > can switch the default for the feature flag to true
> > >> based
> > >> > on
> > >> > > the
> > >> > > > > > > > > > metadata.version once we want to make it the default.
> > >> > > > > > > > >
> > >> > > > > > > > > My plan was to use that feature flag mainly during the
> > >> > > development
> > >> > > > > > > > > phase. I should not have mentioned it, I think, because
> > we
> > >> > > could
> > >> > > > > use
> > >> > > > > > > > > an internal config for it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > required
> > >> > > ACLs, it
> > >> > > > > > > would
> > >> > > > > > > > > be
> > >> > > > > > > > > > good to add that for consistency.
> > >> > > > > > > > >
> > >> > > > > > > > > Noted.
> > >> > > > > > > > >
> > >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > requires
> > >> > > "Read
> > >> > > > > Group"
> > >> > > > > > > > > even
> > >> > > > > > > > > > though it seems to do more than reading.
> > >> > > > > > > > >
> > >> > > > > > > > > I agree. This is how the current protocol works though.
> > We
> > >> > only
> > >> > > > > > > > > require "Read Group" to join a group. We could consider
> > >> > > changing
> > >> > > > > this
> > >> > > > > > > > > but I am not sure that it is worth it.
> > >> > > > > > > > >
> > >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> > with
> > >> > the
> > >> > > new
> > >> > > > > > > group
> > >> > > > > > > > > > protocol? It would be good to have a section on this.
> > >> > > > > > > > >
> > >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> > >> have a
> > >> > > new
> > >> > > > > > > > > topic id so it will treat it like a topic with a
> > different
> > >> > > name.
> > >> > > > > The
> > >> > > > > > > > > only issue is that the fetch/commit offsets APIs do not
> > >> > support
> > >> > > > > topic
> > >> > > > > > > > > IDs so the consumer would reuse the offsets based on the
> > >> > same.
> > >> > > I
> > >> > > > > think
> > >> > > > > > > > > that we should update those APIs as well in order to be
> > >> > > consistent
> > >> > > > > end
> > >> > > > > > > > > to end. That would strengthen the semantics of the
> > >> consumer.
> > >> > > > > > > > >
> > >> > > > > > > > > > 14. The KIP mentions we will write the new coordinator
> > >> in
> > >> > > Java.
> > >> > > > > Even
> > >> > > > > > > > > though
> > >> > > > > > > > > > this is an implementation detail, do we plan to have a
> > >> new
> > >> > > gradle
> > >> > > > > > > module
> > >> > > > > > > > > > for it?
> > >> > > > > > > > >
> > >> > > > > > > > > Yes.
> > >> > > > > > > > >
> > >> > > > > > > > > > 15. Do we have a scalability goal when it comes to how
> > >> many
> > >> > > > > members
> > >> > > > > > > the
> > >> > > > > > > > > new
> > >> > > > > > > > > > group protocol can support?
> > >> > > > > > > > >
> > >> > > > > > > > > We don't have numbers at the moment. The protocol should
> > >> > > support
> > >> > > > > 1000s
> > >> > > > > > > > > of members per group. We will measure this when we have
> > a
> > >> > first
> > >> > > > > > > > > implementation. Note that we might have other
> > bottlenecks
> > >> > down
> > >> > > the
> > >> > > > > > > > > road (e.g. offset commits).
> > >> > > > > > > > >
> > >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds instead
> > of
> > >> > > > > > > > > > SubscribedTopicNames in ConsumerGroupHeartbeatRequest?
> > >> Is
> > >> > the
> > >> > > > > idea
> > >> > > > > > > that
> > >> > > > > > > > > > since we have to resolve the regex on the server, we
> > >> can do
> > >> > > the
> > >> > > > > same
> > >> > > > > > > for
> > >> > > > > > > > > > the topic name? The difference is that sending the
> > >> regex is
> > >> > > more
> > >> > > > > > > > > efficient
> > >> > > > > > > > > > whereas sending the topic names is less efficient.
> > >> > > Furthermore,
> > >> > > > > > > delete
> > >> > > > > > > > > and
> > >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> > >> > > > > > > > >
> > >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> > >> server
> > >> > > for
> > >> > > > > both
> > >> > > > > > > > > paths but I do agree with your point. As a second
> > though,
> > >> > using
> > >> > > > > topic
> > >> > > > > > > > > ids may be better here for the delete and recreation
> > case.
> > >> > > Also, I
> > >> > > > > > > > > suppose that we may allow users to subscribe with topic
> > >> ids
> > >> > in
> > >> > > the
> > >> > > > > > > > > future because that is the only way to be really robust
> > to
> > >> > > topic
> > >> > > > > > > > > re-creation.
> > >> > > > > > > > >
> > >> > > > > > > > > Best,
> > >> > > > > > > > > David
> > >> > > > > > > > >
> > >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> > >> > > djacot@confluent.io>
> > >> > > > > > > wrote:
> > >> > > > > > > > > >
> > >> > > > > > > > > > Hi Justine,
> > >> > > > > > > > > >
> > >> > > > > > > > > > Thanks for your comments. Please find my answers
> > below.
> > >> > > > > > > > > >
> > >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> > >> > > exception
> > >> > > > > of the
> > >> > > > > > > > > > topic names based in the
> > ConsumerGroupHeartbeatRequest.
> > >> I
> > >> > am
> > >> > > not
> > >> > > > > sure
> > >> > > > > > > > > > if using topic names is the right call here. I need to
> > >> > think
> > >> > > > > about it
> > >> > > > > > > > > > a little more. Obviously, the KIP does not change the
> > >> > > > > fetch/commit
> > >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> > >> that
> > >> > we
> > >> > > > > should
> > >> > > > > > > > > > include though as it would give better overall
> > >> guarantee in
> > >> > > the
> > >> > > > > > > > > > producer.
> > >> > > > > > > > > > - You're right. I think that I should not have
> > mentioned
> > >> > this
> > >> > > > > flag at
> > >> > > > > > > > > > all. I will remove it. We can use an internal
> > >> configuration
> > >> > > while
> > >> > > > > > > > > > developing the feature.
> > >> > > > > > > > > > - Both cluster types will be supported. The change is
> > >> > > > > orthogonal. The
> > >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> > >> > > > > > > > > >
> > >> > > > > > > > > > Best,
> > >> > > > > > > > > > David
> > >> > > > > > > > > >
> > >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> > >> > > > > wangguoz@gmail.com>
> > >> > > > > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Hi Ismael,
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> > inlined
> > >> > > below:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> > >> > > ismael@juma.me.uk>
> > >> > > > > > > wrote:
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > Thanks for the KIP. This has the potential to be a
> > >> > great
> > >> > > > > > > > > improvement. A few
> > >> > > > > > > > > > > > initial questions/comments:
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > 1. I think it's premature to talk about target
> > >> versions
> > >> > > for
> > >> > > > > > > > > deprecation and
> > >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> > >> KRaft,
> > >> > > this
> > >> > > > > > > affects a
> > >> > > > > > > > > core
> > >> > > > > > > > > > > > client protocol and hence deprecation/removal will
> > >> be
> > >> > > heavily
> > >> > > > > > > > > dependent on
> > >> > > > > > > > > > > > how quickly applications migrate to the new
> > >> protocol.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> > >> proposed
> > >> > > > > timeline
> > >> > > > > > > in
> > >> > > > > > > > > the
> > >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan` and
> > >> > > instead
> > >> > > > > just
> > >> > > > > > > state
> > >> > > > > > > > > > > that we will decide in the future about when we
> > would
> > >> > > > > deprecate old
> > >> > > > > > > > > > > protocol and behaviors.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> > >> but
> > >> > it
> > >> > > > > wasn't
> > >> > > > > > > made
> > >> > > > > > > > > clear
> > >> > > > > > > > > > > > why. If we added that as a way to estimate when
> > we'd
> > >> > > > > deprecate
> > >> > > > > > > and
> > >> > > > > > > > > remove
> > >> > > > > > > > > > > > the group protocol, I also suggest removing this
> > >> part.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I think that's not specifically related to the
> > >> > > > > deprecation/removal
> > >> > > > > > > > > timeline
> > >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> > >> > > broker-side
> > >> > > > > > > > > > > implementation may be done first, and then the
> > client
> > >> > side,
> > >> > > > > and we
> > >> > > > > > > > > would
> > >> > > > > > > > > > > only mark it as "released" by the time clients
> > >> > > implementations
> > >> > > > > are
> > >> > > > > > > > > done. At
> > >> > > > > > > > > > > that time, to enable the feature the clients need to
> > >> > first
> > >> > > > > swap-in
> > >> > > > > > > the
> > >> > > > > > > > > > > bytecode with a rolling bounce and then set the flag
> > >> > with a
> > >> > > > > second
> > >> > > > > > > > > rolling
> > >> > > > > > > > > > > bounce, and hence we feel it's better to be released
> > >> in a
> > >> > > major
> > >> > > > > > > > > version.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 3. We need to flesh out the details of the
> > migration
> > >> > > story.
> > >> > > > > It
> > >> > > > > > > > > sounds like
> > >> > > > > > > > > > > > we're saying we will support online migrations. Is
> > >> that
> > >> > > > > correct?
> > >> > > > > > > We
> > >> > > > > > > > > should
> > >> > > > > > > > > > > > explain this in detail. It could also be done as a
> > >> > > separate
> > >> > > > > KIP,
> > >> > > > > > > if
> > >> > > > > > > > > it's
> > >> > > > > > > > > > > > easier.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes I think that's the part we can be more concrete
> > >> about
> > >> > > for
> > >> > > > > sure
> > >> > > > > > > (and
> > >> > > > > > > > > > > this is related to your question 2) above). We will
> > >> work
> > >> > on
> > >> > > > > making
> > >> > > > > > > it
> > >> > > > > > > > > more
> > >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> > >> > > subscriptions
> > >> > > > > to
> > >> > > > > > > the
> > >> > > > > > > > > server,
> > >> > > > > > > > > > > > but it seems like there could be some tricky
> > >> > > compatibility
> > >> > > > > > > issues.
> > >> > > > > > > > > Will we
> > >> > > > > > > > > > > > have a mechanism for users to detect that they
> > need
> > >> to
> > >> > > update
> > >> > > > > > > their
> > >> > > > > > > > > regex
> > >> > > > > > > > > > > > before switching to the new protocol?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes I think we need some tooling for non-java client
> > >> > users
> > >> > > to
> > >> > > > > sort
> > >> > > > > > > of
> > >> > > > > > > > > > > "dry-run" the client before switching to the new
> > >> > protocol.
> > >> > > I
> > >> > > > > do not
> > >> > > > > > > > > have a
> > >> > > > > > > > > > > specific idea on top of my head though, maybe others
> > >> like
> > >> > > @Matt
> > >> > > > > > > > > Howlett can
> > >> > > > > > > > > > > chime-in here?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 5. Related to the last question, will the Java
> > >> client
> > >> > > allow
> > >> > > > > the
> > >> > > > > > > > > users to
> > >> > > > > > > > > > > > stick with the current regex engine for
> > >> compatibility
> > >> > > > > reasons?
> > >> > > > > > > For
> > >> > > > > > > > > example,
> > >> > > > > > > > > > > > it may be handy to keep using client based regex
> > at
> > >> > > first to
> > >> > > > > keep
> > >> > > > > > > > > > > > migrations simple and then migrate to server based
> > >> > > regexes
> > >> > > > > as a
> > >> > > > > > > > > second
> > >> > > > > > > > > > > > step.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Honestly I have not thought about that for java
> > >> clients,
> > >> > > and
> > >> > > > > we can
> > >> > > > > > > > > discuss
> > >> > > > > > > > > > > that. What kind of compatibility issues do you have
> > in
> > >> > > mind?
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 6. When we say that the group coordinator will be
> > >> > > > > responsible for
> > >> > > > > > > > > storing
> > >> > > > > > > > > > > > the configurations and that the configurations
> > will
> > >> be
> > >> > > > > deleted
> > >> > > > > > > when
> > >> > > > > > > > > the
> > >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> > trigger
> > >> > > deletion
> > >> > > > > of
> > >> > > > > > > > > > > > configurations?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> > would
> > >> > only
> > >> > > > > > > transit to
> > >> > > > > > > > > that
> > >> > > > > > > > > > > state when the group is EMPTY and also all of its
> > >> > metadata
> > >> > > are
> > >> > > > > > > gone),
> > >> > > > > > > > > once
> > >> > > > > > > > > > > it's transited to DEAD this group would never be
> > >> revived.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> > group
> > >> > > > > coordinator
> > >> > > > > > > > > make it
> > >> > > > > > > > > > > > harder to list all cluster configs and their
> > values?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > That's a good question, and our thoughts are that
> > the
> > >> > > so-called
> > >> > > > > > > "group
> > >> > > > > > > > > > > configurations" are overrides of the cluster-level
> > >> > > > > configurations
> > >> > > > > > > > > > > customized per group so when an admin list cluster
> > >> > configs
> > >> > > it's
> > >> > > > > > > okay to
> > >> > > > > > > > > > > list just the cluster-level "defaults", not showing
> > >> any
> > >> > > > > per-group
> > >> > > > > > > > > > > customizations.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 8. How would someone configure a group before
> > >> starting
> > >> > > the
> > >> > > > > > > > > consumers? Have
> > >> > > > > > > > > > > > we considered allowing the explicit creation of
> > >> groups?
> > >> > > > > > > > > Alternatively, the
> > >> > > > > > > > > > > > configs could be decoupled from the group
> > lifecycle.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > The configs can be created before the group itself
> > as
> > >> an
> > >> > > > > > > independent
> > >> > > > > > > > > entity
> > >> > > > > > > > > > > --- of course, this requires the corresponding
> > >> request to
> > >> > > be
> > >> > > > > > > routed to
> > >> > > > > > > > > the
> > >> > > > > > > > > > > right coordinator based on the group id --- the only
> > >> > thing
> > >> > > that
> > >> > > > > > > > > differs is,
> > >> > > > > > > > > > > when the group itself is gone we also check if there
> > >> are
> > >> > > any
> > >> > > > > > > > > configuration
> > >> > > > > > > > > > > entities related to that group and delete as well.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> > the
> > >> > > creation
> > >> > > > > /
> > >> > > > > > > > > deletion
> > >> > > > > > > > > > > lifecycles of the config entities, and we would like
> > >> to
> > >> > > hear
> > >> > > > > > > everyone's
> > >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> > >> totally
> > >> > > > > decouple
> > >> > > > > > > group
> > >> > > > > > > > > > > configs and hence not delete them at all when the
> > >> group
> > >> > is
> > >> > > > > gone,
> > >> > > > > > > but
> > >> > > > > > > > > always
> > >> > > > > > > > > > > require explicit deletion operations by themselves.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> > >> > client
> > >> > > > > still
> > >> > > > > > > take
> > >> > > > > > > > > a
> > >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> > >> introduce an
> > >> > > > > overload?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I think we do not need to introduce an overload, but
> > >> I'm
> > >> > > all
> > >> > > > > ears
> > >> > > > > > > if
> > >> > > > > > > > > there
> > >> > > > > > > > > > > may be some compatibility issues that we may
> > overlook.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 10. I agree with Justine that we should be clearer
> > >> > about
> > >> > > the
> > >> > > > > > > reason
> > >> > > > > > > > > to
> > >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> > >> flag.
> > >> > > Maybe
> > >> > > > > we
> > >> > > > > > > mean
> > >> > > > > > > > > that we
> > >> > > > > > > > > > > > can switch the default for the feature flag to
> > true
> > >> > > based on
> > >> > > > > the
> > >> > > > > > > > > > > > metadata.version once we want to make it the
> > >> default.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> > >> required
> > >> > > ACLs,
> > >> > > > > it
> > >> > > > > > > > > would be
> > >> > > > > > > > > > > > good to add that for consistency.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Ack, we can certainly do that.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> > >> > requires
> > >> > > > > "Read
> > >> > > > > > > > > Group" even
> > >> > > > > > > > > > > > though it seems to do more than reading.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > I had that thought myself as well, but in the end we
> > >> > could
> > >> > > not
> > >> > > > > > > find a
> > >> > > > > > > > > > > better alternative: adding Write Group seems an
> > >> overkill
> > >> > > here
> > >> > > > > > > since we
> > >> > > > > > > > > do
> > >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> > and
> > >> > > Describe
> > >> > > > > on
> > >> > > > > > > > > groups so
> > >> > > > > > > > > > > far). Would like to hear others thoughts.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 13. How is topic recreation handled by the
> > consumer
> > >> > with
> > >> > > the
> > >> > > > > new
> > >> > > > > > > > > group
> > >> > > > > > > > > > > > protocol? It would be good to have a section on
> > >> this.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> > >> add a
> > >> > > > > section
> > >> > > > > > > about
> > >> > > > > > > > > > > that, but basically the idea is that consumer would
> > be
> > >> > > totally
> > >> > > > > > > > > agnostic in
> > >> > > > > > > > > > > the new protocol as it's handled all by the brokers.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> > >> coordinator
> > >> > in
> > >> > > > > Java.
> > >> > > > > > > Even
> > >> > > > > > > > > though
> > >> > > > > > > > > > > > this is an implementation detail, do we plan to
> > >> have a
> > >> > > new
> > >> > > > > gradle
> > >> > > > > > > > > module
> > >> > > > > > > > > > > > for it?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > We have not thought about that. But I think the
> > answer
> > >> > > should
> > >> > > > > be
> > >> > > > > > > yes.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes to
> > >> how
> > >> > > many
> > >> > > > > > > members
> > >> > > > > > > > > the new
> > >> > > > > > > > > > > > group protocol can support?
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > Within a group, I think we should shoot for 1000s of
> > >> > > members.
> > >> > > > > But
> > >> > > > > > > that
> > >> > > > > > > > > > > scalability goals also depend on the offset
> > management
> > >> > > (commit,
> > >> > > > > > > fetch)
> > >> > > > > > > > > > > capabilities of the coordinator which we did not
> > >> cover in
> > >> > > this
> > >> > > > > > > KIP, so
> > >> > > > > > > > > it's
> > >> > > > > > > > > > > hard to give a number that applies universally.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> > >> instead
> > >> > of
> > >> > > > > > > > > > > > SubscribedTopicNames in
> > >> ConsumerGroupHeartbeatRequest?
> > >> > > Is the
> > >> > > > > > > idea
> > >> > > > > > > > > that
> > >> > > > > > > > > > > > since we have to resolve the regex on the server,
> > we
> > >> > can
> > >> > > do
> > >> > > > > the
> > >> > > > > > > same
> > >> > > > > > > > > for
> > >> > > > > > > > > > > > the topic name? The difference is that sending the
> > >> > regex
> > >> > > is
> > >> > > > > more
> > >> > > > > > > > > efficient
> > >> > > > > > > > > > > > whereas sending the topic names is less efficient.
> > >> > > > > Furthermore,
> > >> > > > > > > > > delete and
> > >> > > > > > > > > > > > recreation is easier to handle if we have topic
> > ids.
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > The main reason to still let the clients send names
> > >> is to
> > >> > > keep
> > >> > > > > the
> > >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> > client
> > >> > > only.
> > >> > > > > Note
> > >> > > > > > > that
> > >> > > > > > > > > > > although we added topic id in KIP-516, we never
> > >> > > implemented the
> > >> > > > > > > logic
> > >> > > > > > > > > on
> > >> > > > > > > > > > > consumer/producers leveraging the related newer
> > >> versioned
> > >> > > RPCs,
> > >> > > > > > > > > instead we
> > >> > > > > > > > > > > just set the topic id as empty UUID. We want to keep
> > >> the
> > >> > > > > > > > > consumer/producer
> > >> > > > > > > > > > > to be thin and only delegate the reasoning on broker
> > >> and
> > >> > > > > > > potentially
> > >> > > > > > > > > admin
> > >> > > > > > > > > > > clients.
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > Thanks,
> > >> > > > > > > > > > > > Ismael
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> > >> > > > > > > > > <dj...@confluent.io.invalid>
> > >> > > > > > > > > > > > wrote:
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > > > > Hi all,
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > I would like to start a discussion thread on
> > >> KIP-848:
> > >> > > The
> > >> > > > > Next
> > >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> > >> With
> > >> > > this
> > >> > > > > KIP,
> > >> > > > > > > we
> > >> > > > > > > > > aim
> > >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> > >> more
> > >> > > > > reliable,
> > >> > > > > > > more
> > >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> > >> easier
> > >> > > to
> > >> > > > > debug
> > >> > > > > > > for
> > >> > > > > > > > > > > > > operators.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > The KIP is here:
> > >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > Please take a look and let me know what you
> > think.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > Best,
> > >> > > > > > > > > > > > > David
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > > > PS: I will be away from July 18th to August 8th.
> > >> That
> > >> > > gives
> > >> > > > > > > you a
> > >> > > > > > > > > bit
> > >> > > > > > > > > > > > > of time to read and digest this long KIP.
> > >> > > > > > > > > > > > >
> > >> > > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > >
> > >> > > > > > > > > > > --
> > >> > > > > > > > > > > -- Guozhang
> > >> > > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > >
> > >> > > > > > > > --
> > >> > > > > > > > -- Guozhang
> > >> > > > > > >
> > >> > > > >
> > >> > >
> > >> > >
> > >> >
> > >>
> > >>
> > >> --
> > >> -- Guozhang
> > >>
> > >
> >

Re: [DISCUSS] KIP-848: The Next Generation of the Consumer Rebalance Protocol

[Gregory Harris <gh...@gmail.com>]

Hey All!

Thanks for the KIP, it's wonderful to see cooperative rebalancing making it
down the stack!

I had a few questions:

1. The 'Rejected Alternatives' section describes how member epoch should
advance in step with the group epoch and assignment epoch values. I think
that this is a good idea for the reasons described in the KIP. When the
protocol is incrementally assigning partitions to a worker, what member
epoch does each incremental assignment use? Are member epochs re-used, and
a single member epoch can correspond to multiple different (monotonically
larger) assignments?

2. Is the Assignor's 'Reason' field opaque to the group coordinator? If
not, should custom client-side assignor implementations interact with the
Reason field, and how is its common meaning agreed upon? If so, what is the
benefit of a distinct Reason field over including such functionality in the
opaque metadata?

3. The following is included in the KIP: "Thanks to this, the input of the
client side assignor is entirely driven by the group coordinator. The
consumer is no longer responsible for maintaining any state besides its
assigned partitions." Does this mean that the client-side assignor MAY
incorporate additional non-Metadata state (such as partition throughput,
cpu/memory metrics, config topics, etc), or that additional non-Metadata
state SHOULD NOT be used?

4. I see that there are separate classes
for org.apache.kafka.server.group.consumer.PartitionAssignor
and org.apache.kafka.clients.consumer.PartitionAssignor that seem to
overlap significantly. Is it possible for these two implementations to be
unified? This would serve to promote feature parity of server-side and
client-side assignors, and would also facilitate operational flexibility in
certain situations. For example, if a server-side assignor has some poor
behavior and needs a patch, deploying the patched assignor to the client
and switching one consumer group to a client-side assignor may be faster
and less risky than patching all of the brokers. With the currently
proposed distinct APIs, a non-trivial reimplementation would have to be
assembled, and if the two APIs have diverged significantly, then it is
possible that a reimplementation would not be possible.

--
Greg Harris
gharris1727@gmail.com
github.com/gharris1727

On Wed, Aug 3, 2022 at 8:39 AM Sagar <sa...@gmail.com> wrote:

> Hi Guozhang/David,
>
> I created a confluence page to discuss how Connect would need to change
> based on the new rebalance protocol. Here's the page:
>
> https://cwiki.apache.org/confluence/display/KAFKA/%5BDRAFT%5DIntegrating+Kafka+Connect+With+New+Consumer+Rebalance+Protocol
>
> It's also pretty longish and I have tried to keep a format similar to
> KIP-848. Let me know what you think. Also, do you think this should be
> moved to a separate discussion thread or is this one fine?
>
> Thanks!
> Sagar.
>
> On Tue, Jul 26, 2022 at 7:37 AM Sagar <sa...@gmail.com> wrote:
>
> > Hello Guozhang,
> >
> > Thank you so much for the doc on Kafka Streams. Sure, I would do the
> > analysis and come up with such a document.
> >
> > Thanks!
> > Sagar.
> >
> > On Tue, Jul 26, 2022 at 4:47 AM Guozhang Wang <wa...@gmail.com>
> wrote:
> >
> >> Hello Sagar,
> >>
> >> It would be great if you could come back with some analysis on how to
> >> implement the Connect side integration with the new protocol; so far
> >> besides leveraging on the new "protocol type" we did not yet think
> through
> >> the Connect side implementations. For Streams here's a draft of
> >> integration
> >> plan:
> >>
> >>
> https://docs.google.com/document/d/17PNz2sGoIvGyIzz8vLyJTJTU2rqnD_D9uHJnH9XARjU/edit#heading=h.pdgirmi57dvn
> >> just FYI for your analysis on Connect.
> >>
> >> On Tue, Jul 19, 2022 at 10:48 PM Sagar <sa...@gmail.com>
> wrote:
> >>
> >> > Hi David,
> >> >
> >> > Thank you for your response. The reason I thought connect can also fit
> >> into
> >> > this new scheme is that even today the connect uses a
> WorkerCoordinator
> >> > extending from AbstractCoordinator to empower rebalances of
> >> > tasks/connectors. The WorkerCoordinator sets the protocolType() to
> >> connect
> >> > and uses the metadata() method by plumbing into
> >> JoinGroupRequestProtocol.
> >> >
> >> > I think the changes to support connect would be similar at a high
> level
> >> to
> >> > the changes in streams mainly because of the Client side assignors
> being
> >> > used in both. At an implementation level, we might need to make a lot
> of
> >> > changes to get onto this new assignment protocol like enhancing the
> >> > JoinGroup request/response and SyncGroup and using
> >> ConsumerGroupHeartbeat
> >> > API etc again on similar lines to streams (or there might be
> >> deviations). I
> >> > would try to perform a detailed analysis of the same  and we can have
> a
> >> > separate discussion thread for that as that would derail this
> discussion
> >> > thread. Let me know if that sounds good to you.
> >> >
> >> > Thanks!
> >> > Sagar.
> >> >
> >> >
> >> >
> >> > On Fri, Jul 15, 2022 at 5:47 PM David Jacot
> <djacot@confluent.io.invalid
> >> >
> >> > wrote:
> >> >
> >> > > Hi Sagar,
> >> > >
> >> > > Thanks for your comments.
> >> > >
> >> > > 1) Yes. That refers to `Assignment#error`. Sure, I can mention it.
> >> > >
> >> > > 2) The idea is to transition C from his current assignment to his
> >> > > target assignment when he can move to epoch 3. When that happens,
> the
> >> > > member assignment is updated and persisted with all its assigned
> >> > > partitions even if they are not all revoked yet. In other words, the
> >> > > member assignment becomes the target assignment. This is basically
> an
> >> > > optimization to avoid having to write all the changes to the log.
> The
> >> > > examples are based on the persisted state so I understand the
> >> > > confusion. Let me see if I can improve this in the description.
> >> > >
> >> > > 3) Regarding Connect, it could reuse the protocol with a client side
> >> > > assignor if it fits in the protocol. The assignment is about
> >> > > topicid-partitions + metadata, could Connect fit into this?
> >> > >
> >> > > Best,
> >> > > David
> >> > >
> >> > > On Fri, Jul 15, 2022 at 1:55 PM Sagar <sa...@gmail.com>
> >> wrote:
> >> > > >
> >> > > > Hi David,
> >> > > >
> >> > > > Thanks for the KIP. I just had minor observations:
> >> > > >
> >> > > > 1) In the Assignment Error section in Client Side mode Assignment
> >> > > process,
> >> > > > you mentioned => `In this case, the client side assignor can
> return
> >> an
> >> > > > error to the group coordinator`. In this case are you referring to
> >> the
> >> > > > Assignor returning an AssignmentError that's listed down towards
> the
> >> > end?
> >> > > > If yes, do you think it would make sense to mention this
> explicitly
> >> > here?
> >> > > >
> >> > > > 2) In the Case Studies section, I have a slight confusion, not
> sure
> >> if
> >> > > > others have the same. Consider this step:
> >> > > >
> >> > > > When B heartbeats, the group coordinator transitions him to epoch
> 3
> >> > > because
> >> > > > B has no partitions to revoke. It persists the change and reply.
> >> > > >
> >> > > >    - Group (epoch=3)
> >> > > >       - A
> >> > > >       - B
> >> > > >       - C
> >> > > >    - Target Assignment (epoch=3)
> >> > > >       - A - partitions=[foo-0]
> >> > > >       - B - partitions=[foo-2]
> >> > > >       - C - partitions=[foo-1]
> >> > > >    - Member Assignment
> >> > > >       - A - epoch=2, partitions=[foo-0, foo-1]
> >> > > >       - B - epoch=3, partitions=[foo-2]
> >> > > >       - C - epoch=3, partitions=[foo-1]
> >> > > >
> >> > > > When C heartbeats, it transitions to epoch 3 but cannot get foo-1
> >> yet.
> >> > > >
> >> > > > Here,it's mentioned that member C can't get the foo-1 partition
> yet,
> >> > but
> >> > > > based on the description above, it seems it already has it. Do you
> >> > think
> >> > > it
> >> > > > would be better to remove it and populate it only when it actually
> >> gets
> >> > > it?
> >> > > > I see this in a lot of other places, so have I understood it
> >> > incorrectly
> >> > > ?
> >> > > >
> >> > > >
> >> > > > Regarding connect , it might be out of scope of this discussion,
> but
> >> > from
> >> > > > what I understood it would probably be running in client side
> >> assignor
> >> > > mode
> >> > > > even on the new rebalance protocol as it has its own Custom
> >> > > Assignors(Eager
> >> > > > and IncrementalCooperative).
> >> > > >
> >> > > > Thanks!
> >> > > >
> >> > > > Sagar.
> >> > > >
> >> > > >
> >> > > >
> >> > > >
> >> > > >
> >> > > >
> >> > > > On Fri, Jul 15, 2022 at 5:00 PM David Jacot
> >> > <djacot@confluent.io.invalid
> >> > > >
> >> > > > wrote:
> >> > > >
> >> > > > > Thanks Hector! Our goal is to move forward with specialized API
> >> > > > > instead of relying on one generic API. For Connect, we can apply
> >> the
> >> > > > > exact same pattern and reuse/share the core implementation on
> the
> >> > > > > server side. For the schema registry, I think that we should
> >> consider
> >> > > > > having a tailored API to do simple membership/leader election.
> >> > > > >
> >> > > > > Best,
> >> > > > > David
> >> > > > >
> >> > > > > On Fri, Jul 15, 2022 at 10:22 AM Ismael Juma <ismael@juma.me.uk
> >
> >> > > wrote:
> >> > > > > >
> >> > > > > > Three quick comments:
> >> > > > > >
> >> > > > > > 1. Regarding java.util.regex.Pattern vs
> >> com.google.re2j.Pattern, we
> >> > > > > should
> >> > > > > > document the differences in more detail before deciding one
> way
> >> or
> >> > > > > another.
> >> > > > > > That said, if people pass java.util.regex.Pattern, they expect
> >> > their
> >> > > > > > semantics to be honored. If we are doing something different,
> >> then
> >> > we
> >> > > > > > should consider adding an overload with our own Pattern class
> (I
> >> > > don't
> >> > > > > > think we'd want to expose re2j's at this point).
> >> > > > > > 2. Regarding topic ids, any major new protocol should
> integrate
> >> > fully
> >> > > > > with
> >> > > > > > it and should handle the topic recreation case correctly.
> That's
> >> > the
> >> > > main
> >> > > > > > part we need to handle. I agree with David that we'd want to
> add
> >> > > topic
> >> > > > > ids
> >> > > > > > to the relevant protocols that don't have it yet and that we
> can
> >> > > probably
> >> > > > > > focus on the internals versus adding new APIs to the Java
> >> Consumer
> >> > > > > (unless
> >> > > > > > we find that adding new APIs is required for reasonable
> >> semantics).
> >> > > > > > 3. I am still not sure about the coordinator storing the
> >> configs.
> >> > > It's
> >> > > > > > powerful for configs to be centralized in the metadata log for
> >> > > various
> >> > > > > > reasons (auditability, visibility, consistency, etc.).
> >> Similarly, I
> >> > > am
> >> > > > > not
> >> > > > > > sure about automatically deleting configs in a way that they
> >> cannot
> >> > > be
> >> > > > > > recovered. A good property for modern systems is to minimize
> the
> >> > > number
> >> > > > > of
> >> > > > > > unrecoverable data loss scenarios.
> >> > > > > >
> >> > > > > > Ismael
> >> > > > > >
> >> > > > > > On Wed, Jul 13, 2022 at 3:47 PM David Jacot
> >> > > <djacot@confluent.io.invalid
> >> > > > > >
> >> > > > > > wrote:
> >> > > > > >
> >> > > > > > > Thanks Guozhang. My answers are below:
> >> > > > > > >
> >> > > > > > > > 1) the migration path, especially the last step when
> clients
> >> > > flip the
> >> > > > > > > flag
> >> > > > > > > > to enable the new protocol, in which we would have a
> window
> >> > where
> >> > > > > both
> >> > > > > > > new
> >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> >> members
> >> > of
> >> > > the
> >> > > > > same
> >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> >> while
> >> > > > > using the
> >> > > > > > > > new protocol is something we need to present about.
> >> > > > > > >
> >> > > > > > > Noted. I just published a new version of KIP which includes
> >> more
> >> > > > > > > details about this. See the "Supporting Online Consumer
> Group
> >> > > Upgrade"
> >> > > > > > > and the "Compatibility, Deprecation, and Migration Plan". I
> >> think
> >> > > that
> >> > > > > > > I have to think through a few cases now but the overall idea
> >> and
> >> > > > > > > mechanism should be understandable.
> >> > > > > > >
> >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> >> are
> >> > > only
> >> > > > > used
> >> > > > > > > as
> >> > > > > > > > part of RPCs and admin client, but they are not exposed
> via
> >> any
> >> > > > > public
> >> > > > > > > APIs
> >> > > > > > > > to consumers yet. I think the question is, first should we
> >> let
> >> > > the
> >> > > > > > > consumer
> >> > > > > > > > client to be maintaining the names -> ids mapping itself
> to
> >> > fully
> >> > > > > > > leverage
> >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> the
> >> > > topic
> >> > > > > ids;
> >> > > > > > > and
> >> > > > > > > > secondly, should we ever consider exposing the topic ids
> in
> >> the
> >> > > > > consumer
> >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> >> the
> >> > > > > rebalance
> >> > > > > > > > listener for cases like topic deletion-and-recreation).
> >> > > > > > >
> >> > > > > > > a) Assuming that we would include converting all the offsets
> >> > > related
> >> > > > > > > RPCs to using topic ids in this KIP, the consumer would be
> >> able
> >> > to
> >> > > > > > > fully operate with topic ids. That being said, it still has
> to
> >> > > provide
> >> > > > > > > the topics names in various APIs so having a mapping in the
> >> > > consumer
> >> > > > > > > seems inevitable to me.
> >> > > > > > > b) I don't have a strong opinion on this. Here I wonder if
> >> this
> >> > > goes
> >> > > > > > > beyond the scope of this KIP. I would rather focus on the
> >> > internals
> >> > > > > > > here and we can consider this separately if we see value in
> >> doing
> >> > > it.
> >> > > > > > >
> >> > > > > > > Coming back to Ismael's point about using topic ids in the
> >> > > > > > > ConsumerGroupHeartbeatRequest, I think that there is one
> >> > advantage
> >> > > in
> >> > > > > > > favour of it. The consumer will have the opportunity to
> >> validate
> >> > > that
> >> > > > > > > the topics exists before passing them into the group
> rebalance
> >> > > > > > > protocol. Obviously, the coordinator will also notice it but
> >> it
> >> > > does
> >> > > > > > > not really have a way to reject an invalid topic in the
> >> response.
> >> > > > > > >
> >> > > > > > > > I'm agreeing with David on all other minor questions
> except
> >> for
> >> > > the
> >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> >> > > necessary
> >> > > > > to
> >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> >> still
> >> > > use
> >> > > > > > > Pattern
> >> > > > > > > > while just documenting that our subscription may be
> >> rejected by
> >> > > the
> >> > > > > > > server.
> >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> >> > using
> >> > > an
> >> > > > > > > > overloaded `String` based subscription may be more
> >> vulnerable
> >> > to
> >> > > > > various
> >> > > > > > > > invalid regexes.
> >> > > > > > >
> >> > > > > > > That could work. I have to look at the differences between
> the
> >> > two
> >> > > > > > > engines to better understand the potential issues. My
> >> > > understanding is
> >> > > > > > > that would work for all the basic regular expressions. The
> >> > > differences
> >> > > > > > > between the two are mainly about the various character
> >> classes. I
> >> > > > > > > wonder what other people think about this.
> >> > > > > > >
> >> > > > > > > Best,
> >> > > > > > > David
> >> > > > > > >
> >> > > > > > > On Tue, Jul 12, 2022 at 11:28 PM Guozhang Wang <
> >> > wangguoz@gmail.com
> >> > > >
> >> > > > > wrote:
> >> > > > > > > >
> >> > > > > > > > Thanks David! I think on the high level there are two meta
> >> > > points we
> >> > > > > need
> >> > > > > > > > to concretize a bit more:
> >> > > > > > > >
> >> > > > > > > > 1) the migration path, especially the last step when
> clients
> >> > > flip the
> >> > > > > > > flag
> >> > > > > > > > to enable the new protocol, in which we would have a
> window
> >> > where
> >> > > > > both
> >> > > > > > > new
> >> > > > > > > > protocols / rpcs and old protocols / rpcs are used by
> >> members
> >> > of
> >> > > the
> >> > > > > same
> >> > > > > > > > group. How the coordinator could "mimic" the old behavior
> >> while
> >> > > > > using the
> >> > > > > > > > new protocol is something we need to present about.
> >> > > > > > > > 2) the usage of topic ids. So far as KIP-516 the topic ids
> >> are
> >> > > only
> >> > > > > used
> >> > > > > > > as
> >> > > > > > > > part of RPCs and admin client, but they are not exposed
> via
> >> any
> >> > > > > public
> >> > > > > > > APIs
> >> > > > > > > > to consumers yet. I think the question is, first should we
> >> let
> >> > > the
> >> > > > > > > consumer
> >> > > > > > > > client to be maintaining the names -> ids mapping itself
> to
> >> > fully
> >> > > > > > > leverage
> >> > > > > > > > on all the augmented existing RPCs and the new RPCs with
> the
> >> > > topic
> >> > > > > ids;
> >> > > > > > > and
> >> > > > > > > > secondly, should we ever consider exposing the topic ids
> in
> >> the
> >> > > > > consumer
> >> > > > > > > > public APIs as well (both subscribe/assign, as well as in
> >> the
> >> > > > > rebalance
> >> > > > > > > > listener for cases like topic deletion-and-recreation).
> >> > > > > > > >
> >> > > > > > > > I'm agreeing with David on all other minor questions
> except
> >> for
> >> > > the
> >> > > > > > > > `subscribe(Pattern)` question: personally I think it's not
> >> > > necessary
> >> > > > > to
> >> > > > > > > > deprecate the subscribe API with Pattern, but instead we
> >> still
> >> > > use
> >> > > > > > > Pattern
> >> > > > > > > > while just documenting that our subscription may be
> >> rejected by
> >> > > the
> >> > > > > > > server.
> >> > > > > > > > Since the incompatible case is a very rare scenario I felt
> >> > using
> >> > > an
> >> > > > > > > > overloaded `String` based subscription may be more
> >> vulnerable
> >> > to
> >> > > > > various
> >> > > > > > > > invalid regexes.
> >> > > > > > > >
> >> > > > > > > >
> >> > > > > > > > Guozhang
> >> > > > > > > >
> >> > > > > > > > On Tue, Jul 12, 2022 at 5:23 AM David Jacot
> >> > > > > <djacot@confluent.io.invalid
> >> > > > > > > >
> >> > > > > > > > wrote:
> >> > > > > > > >
> >> > > > > > > > > Hi Ismael,
> >> > > > > > > > >
> >> > > > > > > > > Thanks for your feedback. Let me answer your questions
> >> > inline.
> >> > > > > > > > >
> >> > > > > > > > > > 1. I think it's premature to talk about target
> versions
> >> for
> >> > > > > > > deprecation
> >> > > > > > > > > and
> >> > > > > > > > > > removal of the existing group protocol. Unlike KRaft,
> >> this
> >> > > > > affects a
> >> > > > > > > core
> >> > > > > > > > > > client protocol and hence deprecation/removal will be
> >> > heavily
> >> > > > > > > dependent
> >> > > > > > > > > on
> >> > > > > > > > > > how quickly applications migrate to the new protocol.
> >> > > > > > > > >
> >> > > > > > > > > That makes sense. I will remove it.
> >> > > > > > > > >
> >> > > > > > > > > > 2. The KIP says we intend to release this in 4.x, but
> it
> >> > > wasn't
> >> > > > > made
> >> > > > > > > > > clear
> >> > > > > > > > > > why. If we added that as a way to estimate when we'd
> >> > > deprecate
> >> > > > > and
> >> > > > > > > remove
> >> > > > > > > > > > the group protocol, I also suggest removing this part.
> >> > > > > > > > >
> >> > > > > > > > > Let me explain my reasoning. As explained, I plan to
> >> rewrite
> >> > > the
> >> > > > > group
> >> > > > > > > > > coordinator in Java while we implement the new protocol.
> >> This
> >> > > means
> >> > > > > > > > > that the internals will be slightly different (e.g.
> >> threading
> >> > > > > model).
> >> > > > > > > > > Therefore, I wanted to tighten the switch from the old
> >> group
> >> > > > > > > > > coordinator to the new group coordinator to a major
> >> release.
> >> > > The
> >> > > > > > > > > alternative would be to use a flag to do the switch
> >> instead
> >> > of
> >> > > > > relying
> >> > > > > > > > > on the software upgrade.
> >> > > > > > > > >
> >> > > > > > > > > > 3. We need to flesh out the details of the migration
> >> story.
> >> > > It
> >> > > > > sounds
> >> > > > > > > > > like
> >> > > > > > > > > > we're saying we will support online migrations. Is
> that
> >> > > correct?
> >> > > > > We
> >> > > > > > > > > should
> >> > > > > > > > > > explain this in detail. It could also be done as a
> >> separate
> >> > > KIP,
> >> > > > > if
> >> > > > > > > it's
> >> > > > > > > > > > easier.
> >> > > > > > > > >
> >> > > > > > > > > Yes, we will support online migrations for the group.
> That
> >> > > means
> >> > > > > that
> >> > > > > > > > > a group using the old protocol will be able to switch to
> >> the
> >> > > new
> >> > > > > > > > > protocol.
> >> > > > > > > > >
> >> > > > > > > > > Let me briefly explain how that will work though. It is
> >> > > basically a
> >> > > > > > > > > four step process:
> >> > > > > > > > >
> >> > > > > > > > > 1. The cluster must be upgraded or rolled to a software
> >> > > supporting
> >> > > > > the
> >> > > > > > > > > new group coordinator. Both the old and the new
> >> coordinator
> >> > > will
> >> > > > > > > > > support the old protocol and rely on the same persisted
> >> > > metadata so
> >> > > > > > > > > they can work together. This point is an offline
> >> migration.
> >> > We
> >> > > > > cannot
> >> > > > > > > > > do this one live because it would require shutting down
> >> the
> >> > > current
> >> > > > > > > > > coordinator and starting up the new one and that would
> >> cause
> >> > > > > > > > > unavailabilities.
> >> > > > > > > > > 2. The cluster's metadata version/IBP must be upgraded
> to
> >> X
> >> > in
> >> > > > > order
> >> > > > > > > > > to enable the new protocol. This cannot be done before
> 1)
> >> is
> >> > > > > > > > > terminated because the old coordinator doesn't support
> the
> >> > new
> >> > > > > > > > > protocol.
> >> > > > > > > > > 3. The consumers must be upgraded to a version
> supporting
> >> the
> >> > > > > online
> >> > > > > > > > > migration (must have KIP-792). If the consumer is
> already
> >> > > there.
> >> > > > > > > > > Nothing must be done at this point.
> >> > > > > > > > > 4. The consumers must be rolled with the feature flag
> >> turned
> >> > > on.
> >> > > > > The
> >> > > > > > > > > consumer group is automatically converted when the first
> >> > > consumer
> >> > > > > > > > > using the new protocol joins the group. While the
> members
> >> > > using the
> >> > > > > > > > > old protocol are being upgraded, the old protocol is
> >> proxied
> >> > > into
> >> > > > > the
> >> > > > > > > > > new one.
> >> > > > > > > > >
> >> > > > > > > > > Let me clarify all of this in the KIP.
> >> > > > > > > > >
> >> > > > > > > > > > 4. I am happy that we are pushing the pattern
> >> subscriptions
> >> > > to
> >> > > > > the
> >> > > > > > > > > server,
> >> > > > > > > > > > but it seems like there could be some tricky
> >> compatibility
> >> > > > > issues.
> >> > > > > > > Will
> >> > > > > > > > > we
> >> > > > > > > > > > have a mechanism for users to detect that they need to
> >> > update
> >> > > > > their
> >> > > > > > > regex
> >> > > > > > > > > > before switching to the new protocol?
> >> > > > > > > > >
> >> > > > > > > > > I think that I am a bit more optimistic than you on this
> >> > > point. I
> >> > > > > > > > > believe that the majority of the cases are simple
> regexes
> >> > which
> >> > > > > should
> >> > > > > > > > > work with the new engine. The coordinator will verify
> the
> >> > regex
> >> > > > > anyway
> >> > > > > > > > > and reject the consumer if the regex is not valid.
> Coming
> >> > back
> >> > > to
> >> > > > > the
> >> > > > > > > > > migration path, in the worst case, the first upgraded
> >> > consumer
> >> > > > > joining
> >> > > > > > > > > the group will be rejected. This should be used as the
> >> last
> >> > > > > defence, I
> >> > > > > > > > > would say.
> >> > > > > > > > >
> >> > > > > > > > > One way for customers to validate their regex before
> >> > upgrading
> >> > > > > their
> >> > > > > > > > > prod would be to test them with another group. For
> >> instance,
> >> > > that
> >> > > > > > > > > could be done in a pre-prod environment. Another way
> >> would be
> >> > > to
> >> > > > > > > > > extend the consumer-group tool to provide a regex
> >> validation
> >> > > > > > > > > mechanism. Would this be enough in your opinion?
> >> > > > > > > > >
> >> > > > > > > > > > 5. Related to the last question, will the Java client
> >> allow
> >> > > the
> >> > > > > > > users to
> >> > > > > > > > > > stick with the current regex engine for compatibility
> >> > > reasons?
> >> > > > > For
> >> > > > > > > > > example,
> >> > > > > > > > > > it may be handy to keep using client based regex at
> >> first
> >> > to
> >> > > keep
> >> > > > > > > > > > migrations simple and then migrate to server based
> >> regexes
> >> > > as a
> >> > > > > > > second
> >> > > > > > > > > step.
> >> > > > > > > > >
> >> > > > > > > > > I understand your point but I am concerned that this
> would
> >> > > allow
> >> > > > > users
> >> > > > > > > > > to actually stay in this mode. That would go against our
> >> goal
> >> > > of
> >> > > > > > > > > simplifying the client because we would have to continue
> >> > > monitoring
> >> > > > > > > > > the metadata on the client side. I would rather not do
> >> this.
> >> > > > > > > > >
> >> > > > > > > > > > 6. When we say that the group coordinator will be
> >> > > responsible for
> >> > > > > > > storing
> >> > > > > > > > > > the configurations and that the configurations will be
> >> > > deleted
> >> > > > > when
> >> > > > > > > the
> >> > > > > > > > > > group is deleted. Will a transition to DEAD trigger
> >> > deletion
> >> > > of
> >> > > > > > > > > > configurations?
> >> > > > > > > > >
> >> > > > > > > > > That's right. The configurations will be deleted when
> the
> >> > > group is
> >> > > > > > > > > deleted. They go together.
> >> > > > > > > > >
> >> > > > > > > > > > 7. Will the choice to store the configs in the group
> >> > > coordinator
> >> > > > > > > make it
> >> > > > > > > > > > harder to list all cluster configs and their values?
> >> > > > > > > > >
> >> > > > > > > > > I don't think so. The group configurations are overrides
> >> of
> >> > > cluster
> >> > > > > > > > > configs. If you want to know all the overrides though,
> you
> >> > > would
> >> > > > > have
> >> > > > > > > > > to ask all the group coordinators. You cannot rely on
> the
> >> > > metadata
> >> > > > > log
> >> > > > > > > > > for instance.
> >> > > > > > > > >
> >> > > > > > > > > > 8. How would someone configure a group before starting
> >> the
> >> > > > > consumers?
> >> > > > > > > > > Have
> >> > > > > > > > > > we considered allowing the explicit creation of
> groups?
> >> > > > > > > Alternatively,
> >> > > > > > > > > the
> >> > > > > > > > > > configs could be decoupled from the group lifecycle.
> >> > > > > > > > >
> >> > > > > > > > > Yes. The group will be automatically created in this
> case.
> >> > > However,
> >> > > > > > > > > the configs will be lost after the retention period of
> the
> >> > > group
> >> > > > > > > > > passes.
> >> > > > > > > > >
> >> > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> >> client
> >> > > still
> >> > > > > take
> >> > > > > > > a
> >> > > > > > > > > > `java.util.regex.Pattern` of do we have to introduce
> an
> >> > > overload?
> >> > > > > > > > >
> >> > > > > > > > > That's a very group question. I forgot about that one.
> As
> >> the
> >> > > > > > > > > `java.util.regex.Pattern` is not fully compatible with
> the
> >> > > engine
> >> > > > > that
> >> > > > > > > > > we plan to use, it might be better to deprecate it and
> >> use an
> >> > > > > overload
> >> > > > > > > > > which takes a string. We would rely on the server side
> >> > > validation.
> >> > > > > > > > > During the migration, I think that we could still try to
> >> > > toString
> >> > > > > the
> >> > > > > > > > > regex and use it. That should work, I think, in the
> >> majority
> >> > > of the
> >> > > > > > > > > cases.
> >> > > > > > > > >
> >> > > > > > > > > > 10. I agree with Justine that we should be clearer
> about
> >> > the
> >> > > > > reason
> >> > > > > > > to
> >> > > > > > > > > > switch to IBP/metadata.version from the feature flag.
> >> Maybe
> >> > > we
> >> > > > > mean
> >> > > > > > > that
> >> > > > > > > > > we
> >> > > > > > > > > > can switch the default for the feature flag to true
> >> based
> >> > on
> >> > > the
> >> > > > > > > > > > metadata.version once we want to make it the default.
> >> > > > > > > > >
> >> > > > > > > > > My plan was to use that feature flag mainly during the
> >> > > development
> >> > > > > > > > > phase. I should not have mentioned it, I think, because
> we
> >> > > could
> >> > > > > use
> >> > > > > > > > > an internal config for it.
> >> > > > > > > > >
> >> > > > > > > > > > 11. Some of the protocol APIs don't mention the
> required
> >> > > ACLs, it
> >> > > > > > > would
> >> > > > > > > > > be
> >> > > > > > > > > > good to add that for consistency.
> >> > > > > > > > >
> >> > > > > > > > > Noted.
> >> > > > > > > > >
> >> > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> requires
> >> > > "Read
> >> > > > > Group"
> >> > > > > > > > > even
> >> > > > > > > > > > though it seems to do more than reading.
> >> > > > > > > > >
> >> > > > > > > > > I agree. This is how the current protocol works though.
> We
> >> > only
> >> > > > > > > > > require "Read Group" to join a group. We could consider
> >> > > changing
> >> > > > > this
> >> > > > > > > > > but I am not sure that it is worth it.
> >> > > > > > > > >
> >> > > > > > > > > > 13. How is topic recreation handled by the consumer
> with
> >> > the
> >> > > new
> >> > > > > > > group
> >> > > > > > > > > > protocol? It would be good to have a section on this.
> >> > > > > > > > >
> >> > > > > > > > > Noted. From a protocol perspective, the new topic will
> >> have a
> >> > > new
> >> > > > > > > > > topic id so it will treat it like a topic with a
> different
> >> > > name.
> >> > > > > The
> >> > > > > > > > > only issue is that the fetch/commit offsets APIs do not
> >> > support
> >> > > > > topic
> >> > > > > > > > > IDs so the consumer would reuse the offsets based on the
> >> > same.
> >> > > I
> >> > > > > think
> >> > > > > > > > > that we should update those APIs as well in order to be
> >> > > consistent
> >> > > > > end
> >> > > > > > > > > to end. That would strengthen the semantics of the
> >> consumer.
> >> > > > > > > > >
> >> > > > > > > > > > 14. The KIP mentions we will write the new coordinator
> >> in
> >> > > Java.
> >> > > > > Even
> >> > > > > > > > > though
> >> > > > > > > > > > this is an implementation detail, do we plan to have a
> >> new
> >> > > gradle
> >> > > > > > > module
> >> > > > > > > > > > for it?
> >> > > > > > > > >
> >> > > > > > > > > Yes.
> >> > > > > > > > >
> >> > > > > > > > > > 15. Do we have a scalability goal when it comes to how
> >> many
> >> > > > > members
> >> > > > > > > the
> >> > > > > > > > > new
> >> > > > > > > > > > group protocol can support?
> >> > > > > > > > >
> >> > > > > > > > > We don't have numbers at the moment. The protocol should
> >> > > support
> >> > > > > 1000s
> >> > > > > > > > > of members per group. We will measure this when we have
> a
> >> > first
> >> > > > > > > > > implementation. Note that we might have other
> bottlenecks
> >> > down
> >> > > the
> >> > > > > > > > > road (e.g. offset commits).
> >> > > > > > > > >
> >> > > > > > > > > > 16. Did we consider having SubscribedTopidIds instead
> of
> >> > > > > > > > > > SubscribedTopicNames in ConsumerGroupHeartbeatRequest?
> >> Is
> >> > the
> >> > > > > idea
> >> > > > > > > that
> >> > > > > > > > > > since we have to resolve the regex on the server, we
> >> can do
> >> > > the
> >> > > > > same
> >> > > > > > > for
> >> > > > > > > > > > the topic name? The difference is that sending the
> >> regex is
> >> > > more
> >> > > > > > > > > efficient
> >> > > > > > > > > > whereas sending the topic names is less efficient.
> >> > > Furthermore,
> >> > > > > > > delete
> >> > > > > > > > > and
> >> > > > > > > > > > recreation is easier to handle if we have topic ids.
> >> > > > > > > > >
> >> > > > > > > > > The idea was to consolidate the metadata lookup on the
> >> server
> >> > > for
> >> > > > > both
> >> > > > > > > > > paths but I do agree with your point. As a second
> though,
> >> > using
> >> > > > > topic
> >> > > > > > > > > ids may be better here for the delete and recreation
> case.
> >> > > Also, I
> >> > > > > > > > > suppose that we may allow users to subscribe with topic
> >> ids
> >> > in
> >> > > the
> >> > > > > > > > > future because that is the only way to be really robust
> to
> >> > > topic
> >> > > > > > > > > re-creation.
> >> > > > > > > > >
> >> > > > > > > > > Best,
> >> > > > > > > > > David
> >> > > > > > > > >
> >> > > > > > > > > On Tue, Jul 12, 2022 at 1:38 PM David Jacot <
> >> > > djacot@confluent.io>
> >> > > > > > > wrote:
> >> > > > > > > > > >
> >> > > > > > > > > > Hi Justine,
> >> > > > > > > > > >
> >> > > > > > > > > > Thanks for your comments. Please find my answers
> below.
> >> > > > > > > > > >
> >> > > > > > > > > > - Yes, the new protocol relies on topic IDs with the
> >> > > exception
> >> > > > > of the
> >> > > > > > > > > > topic names based in the
> ConsumerGroupHeartbeatRequest.
> >> I
> >> > am
> >> > > not
> >> > > > > sure
> >> > > > > > > > > > if using topic names is the right call here. I need to
> >> > think
> >> > > > > about it
> >> > > > > > > > > > a little more. Obviously, the KIP does not change the
> >> > > > > fetch/commit
> >> > > > > > > > > > offsets RPCs to use topic IDs. This may be something
> >> that
> >> > we
> >> > > > > should
> >> > > > > > > > > > include though as it would give better overall
> >> guarantee in
> >> > > the
> >> > > > > > > > > > producer.
> >> > > > > > > > > > - You're right. I think that I should not have
> mentioned
> >> > this
> >> > > > > flag at
> >> > > > > > > > > > all. I will remove it. We can use an internal
> >> configuration
> >> > > while
> >> > > > > > > > > > developing the feature.
> >> > > > > > > > > > - Both cluster types will be supported. The change is
> >> > > > > orthogonal. The
> >> > > > > > > > > > only requirement is that the cluster uses topic IDs.
> >> > > > > > > > > >
> >> > > > > > > > > > Best,
> >> > > > > > > > > > David
> >> > > > > > > > > >
> >> > > > > > > > > > On Mon, Jul 11, 2022 at 9:53 PM Guozhang Wang <
> >> > > > > wangguoz@gmail.com>
> >> > > > > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > Hi Ismael,
> >> > > > > > > > > > >
> >> > > > > > > > > > > Thanks for the feedback. Here are some replies
> inlined
> >> > > below:
> >> > > > > > > > > > >
> >> > > > > > > > > > > On Sat, Jul 9, 2022 at 2:53 AM Ismael Juma <
> >> > > ismael@juma.me.uk>
> >> > > > > > > wrote:
> >> > > > > > > > > > >
> >> > > > > > > > > > > > Thanks for the KIP. This has the potential to be a
> >> > great
> >> > > > > > > > > improvement. A few
> >> > > > > > > > > > > > initial questions/comments:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > 1. I think it's premature to talk about target
> >> versions
> >> > > for
> >> > > > > > > > > deprecation and
> >> > > > > > > > > > > > removal of the existing group protocol. Unlike
> >> KRaft,
> >> > > this
> >> > > > > > > affects a
> >> > > > > > > > > core
> >> > > > > > > > > > > > client protocol and hence deprecation/removal will
> >> be
> >> > > heavily
> >> > > > > > > > > dependent on
> >> > > > > > > > > > > > how quickly applications migrate to the new
> >> protocol.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yeah I agree with you. I think we can remove the
> >> proposed
> >> > > > > timeline
> >> > > > > > > in
> >> > > > > > > > > the
> >> > > > > > > > > > > `Compatibility, Deprecation, and Migration Plan` and
> >> > > instead
> >> > > > > just
> >> > > > > > > state
> >> > > > > > > > > > > that we will decide in the future about when we
> would
> >> > > > > deprecate old
> >> > > > > > > > > > > protocol and behaviors.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 2. The KIP says we intend to release this in 4.x,
> >> but
> >> > it
> >> > > > > wasn't
> >> > > > > > > made
> >> > > > > > > > > clear
> >> > > > > > > > > > > > why. If we added that as a way to estimate when
> we'd
> >> > > > > deprecate
> >> > > > > > > and
> >> > > > > > > > > remove
> >> > > > > > > > > > > > the group protocol, I also suggest removing this
> >> part.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > I think that's not specifically related to the
> >> > > > > deprecation/removal
> >> > > > > > > > > timeline
> >> > > > > > > > > > > plan, but it's more for client upgrades. I.e. the
> >> > > broker-side
> >> > > > > > > > > > > implementation may be done first, and then the
> client
> >> > side,
> >> > > > > and we
> >> > > > > > > > > would
> >> > > > > > > > > > > only mark it as "released" by the time clients
> >> > > implementations
> >> > > > > are
> >> > > > > > > > > done. At
> >> > > > > > > > > > > that time, to enable the feature the clients need to
> >> > first
> >> > > > > swap-in
> >> > > > > > > the
> >> > > > > > > > > > > bytecode with a rolling bounce and then set the flag
> >> > with a
> >> > > > > second
> >> > > > > > > > > rolling
> >> > > > > > > > > > > bounce, and hence we feel it's better to be released
> >> in a
> >> > > major
> >> > > > > > > > > version.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 3. We need to flesh out the details of the
> migration
> >> > > story.
> >> > > > > It
> >> > > > > > > > > sounds like
> >> > > > > > > > > > > > we're saying we will support online migrations. Is
> >> that
> >> > > > > correct?
> >> > > > > > > We
> >> > > > > > > > > should
> >> > > > > > > > > > > > explain this in detail. It could also be done as a
> >> > > separate
> >> > > > > KIP,
> >> > > > > > > if
> >> > > > > > > > > it's
> >> > > > > > > > > > > > easier.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yes I think that's the part we can be more concrete
> >> about
> >> > > for
> >> > > > > sure
> >> > > > > > > (and
> >> > > > > > > > > > > this is related to your question 2) above). We will
> >> work
> >> > on
> >> > > > > making
> >> > > > > > > it
> >> > > > > > > > > more
> >> > > > > > > > > > > explicit in parallel as we solicit more feedback.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 4. I am happy that we are pushing the pattern
> >> > > subscriptions
> >> > > > > to
> >> > > > > > > the
> >> > > > > > > > > server,
> >> > > > > > > > > > > > but it seems like there could be some tricky
> >> > > compatibility
> >> > > > > > > issues.
> >> > > > > > > > > Will we
> >> > > > > > > > > > > > have a mechanism for users to detect that they
> need
> >> to
> >> > > update
> >> > > > > > > their
> >> > > > > > > > > regex
> >> > > > > > > > > > > > before switching to the new protocol?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yes I think we need some tooling for non-java client
> >> > users
> >> > > to
> >> > > > > sort
> >> > > > > > > of
> >> > > > > > > > > > > "dry-run" the client before switching to the new
> >> > protocol.
> >> > > I
> >> > > > > do not
> >> > > > > > > > > have a
> >> > > > > > > > > > > specific idea on top of my head though, maybe others
> >> like
> >> > > @Matt
> >> > > > > > > > > Howlett can
> >> > > > > > > > > > > chime-in here?
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 5. Related to the last question, will the Java
> >> client
> >> > > allow
> >> > > > > the
> >> > > > > > > > > users to
> >> > > > > > > > > > > > stick with the current regex engine for
> >> compatibility
> >> > > > > reasons?
> >> > > > > > > For
> >> > > > > > > > > example,
> >> > > > > > > > > > > > it may be handy to keep using client based regex
> at
> >> > > first to
> >> > > > > keep
> >> > > > > > > > > > > > migrations simple and then migrate to server based
> >> > > regexes
> >> > > > > as a
> >> > > > > > > > > second
> >> > > > > > > > > > > > step.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Honestly I have not thought about that for java
> >> clients,
> >> > > and
> >> > > > > we can
> >> > > > > > > > > discuss
> >> > > > > > > > > > > that. What kind of compatibility issues do you have
> in
> >> > > mind?
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 6. When we say that the group coordinator will be
> >> > > > > responsible for
> >> > > > > > > > > storing
> >> > > > > > > > > > > > the configurations and that the configurations
> will
> >> be
> >> > > > > deleted
> >> > > > > > > when
> >> > > > > > > > > the
> >> > > > > > > > > > > > group is deleted. Will a transition to DEAD
> trigger
> >> > > deletion
> >> > > > > of
> >> > > > > > > > > > > > configurations?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Yes, since the DEAD state is an ending state (we
> would
> >> > only
> >> > > > > > > transit to
> >> > > > > > > > > that
> >> > > > > > > > > > > state when the group is EMPTY and also all of its
> >> > metadata
> >> > > are
> >> > > > > > > gone),
> >> > > > > > > > > once
> >> > > > > > > > > > > it's transited to DEAD this group would never be
> >> revived.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 7. Will the choice to store the configs in the
> group
> >> > > > > coordinator
> >> > > > > > > > > make it
> >> > > > > > > > > > > > harder to list all cluster configs and their
> values?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > That's a good question, and our thoughts are that
> the
> >> > > so-called
> >> > > > > > > "group
> >> > > > > > > > > > > configurations" are overrides of the cluster-level
> >> > > > > configurations
> >> > > > > > > > > > > customized per group so when an admin list cluster
> >> > configs
> >> > > it's
> >> > > > > > > okay to
> >> > > > > > > > > > > list just the cluster-level "defaults", not showing
> >> any
> >> > > > > per-group
> >> > > > > > > > > > > customizations.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 8. How would someone configure a group before
> >> starting
> >> > > the
> >> > > > > > > > > consumers? Have
> >> > > > > > > > > > > > we considered allowing the explicit creation of
> >> groups?
> >> > > > > > > > > Alternatively, the
> >> > > > > > > > > > > > configs could be decoupled from the group
> lifecycle.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > The configs can be created before the group itself
> as
> >> an
> >> > > > > > > independent
> >> > > > > > > > > entity
> >> > > > > > > > > > > --- of course, this requires the corresponding
> >> request to
> >> > > be
> >> > > > > > > routed to
> >> > > > > > > > > the
> >> > > > > > > > > > > right coordinator based on the group id --- the only
> >> > thing
> >> > > that
> >> > > > > > > > > differs is,
> >> > > > > > > > > > > when the group itself is gone we also check if there
> >> are
> >> > > any
> >> > > > > > > > > configuration
> >> > > > > > > > > > > entities related to that group and delete as well.
> >> > > > > > > > > > >
> >> > > > > > > > > > > Admittedly this indeed introduces an asymmetry on
> the
> >> > > creation
> >> > > > > /
> >> > > > > > > > > deletion
> >> > > > > > > > > > > lifecycles of the config entities, and we would like
> >> to
> >> > > hear
> >> > > > > > > everyone's
> >> > > > > > > > > > > feelings whether we should aim for symmetry i.e.
> >> totally
> >> > > > > decouple
> >> > > > > > > group
> >> > > > > > > > > > > configs and hence not delete them at all when the
> >> group
> >> > is
> >> > > > > gone,
> >> > > > > > > but
> >> > > > > > > > > always
> >> > > > > > > > > > > require explicit deletion operations by themselves.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 9. Will the Consumer.subscribe method for the Java
> >> > client
> >> > > > > still
> >> > > > > > > take
> >> > > > > > > > > a
> >> > > > > > > > > > > > `java.util.regex.Pattern` of do we have to
> >> introduce an
> >> > > > > overload?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > I think we do not need to introduce an overload, but
> >> I'm
> >> > > all
> >> > > > > ears
> >> > > > > > > if
> >> > > > > > > > > there
> >> > > > > > > > > > > may be some compatibility issues that we may
> overlook.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 10. I agree with Justine that we should be clearer
> >> > about
> >> > > the
> >> > > > > > > reason
> >> > > > > > > > > to
> >> > > > > > > > > > > > switch to IBP/metadata.version from the feature
> >> flag.
> >> > > Maybe
> >> > > > > we
> >> > > > > > > mean
> >> > > > > > > > > that we
> >> > > > > > > > > > > > can switch the default for the feature flag to
> true
> >> > > based on
> >> > > > > the
> >> > > > > > > > > > > > metadata.version once we want to make it the
> >> default.
> >> > > > > > > > > > >
> >> > > > > > > > > > > 11. Some of the protocol APIs don't mention the
> >> required
> >> > > ACLs,
> >> > > > > it
> >> > > > > > > > > would be
> >> > > > > > > > > > > > good to add that for consistency.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Ack, we can certainly do that.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 12. It is a bit odd that ConsumerGroupHeartbeat
> >> > requires
> >> > > > > "Read
> >> > > > > > > > > Group" even
> >> > > > > > > > > > > > though it seems to do more than reading.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > I had that thought myself as well, but in the end we
> >> > could
> >> > > not
> >> > > > > > > find a
> >> > > > > > > > > > > better alternative: adding Write Group seems an
> >> overkill
> >> > > here
> >> > > > > > > since we
> >> > > > > > > > > do
> >> > > > > > > > > > > not have it elsewhere (we only have Read / Delete
> and
> >> > > Describe
> >> > > > > on
> >> > > > > > > > > groups so
> >> > > > > > > > > > > far). Would like to hear others thoughts.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 13. How is topic recreation handled by the
> consumer
> >> > with
> >> > > the
> >> > > > > new
> >> > > > > > > > > group
> >> > > > > > > > > > > > protocol? It would be good to have a section on
> >> this.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > You mean with regex subscription right? Yes we can
> >> add a
> >> > > > > section
> >> > > > > > > about
> >> > > > > > > > > > > that, but basically the idea is that consumer would
> be
> >> > > totally
> >> > > > > > > > > agnostic in
> >> > > > > > > > > > > the new protocol as it's handled all by the brokers.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 14. The KIP mentions we will write the new
> >> coordinator
> >> > in
> >> > > > > Java.
> >> > > > > > > Even
> >> > > > > > > > > though
> >> > > > > > > > > > > > this is an implementation detail, do we plan to
> >> have a
> >> > > new
> >> > > > > gradle
> >> > > > > > > > > module
> >> > > > > > > > > > > > for it?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > We have not thought about that. But I think the
> answer
> >> > > should
> >> > > > > be
> >> > > > > > > yes.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 15. Do we have a scalability goal when it comes to
> >> how
> >> > > many
> >> > > > > > > members
> >> > > > > > > > > the new
> >> > > > > > > > > > > > group protocol can support?
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > Within a group, I think we should shoot for 1000s of
> >> > > members.
> >> > > > > But
> >> > > > > > > that
> >> > > > > > > > > > > scalability goals also depend on the offset
> management
> >> > > (commit,
> >> > > > > > > fetch)
> >> > > > > > > > > > > capabilities of the coordinator which we did not
> >> cover in
> >> > > this
> >> > > > > > > KIP, so
> >> > > > > > > > > it's
> >> > > > > > > > > > > hard to give a number that applies universally.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > > 16. Did we consider having SubscribedTopidIds
> >> instead
> >> > of
> >> > > > > > > > > > > > SubscribedTopicNames in
> >> ConsumerGroupHeartbeatRequest?
> >> > > Is the
> >> > > > > > > idea
> >> > > > > > > > > that
> >> > > > > > > > > > > > since we have to resolve the regex on the server,
> we
> >> > can
> >> > > do
> >> > > > > the
> >> > > > > > > same
> >> > > > > > > > > for
> >> > > > > > > > > > > > the topic name? The difference is that sending the
> >> > regex
> >> > > is
> >> > > > > more
> >> > > > > > > > > efficient
> >> > > > > > > > > > > > whereas sending the topic names is less efficient.
> >> > > > > Furthermore,
> >> > > > > > > > > delete and
> >> > > > > > > > > > > > recreation is easier to handle if we have topic
> ids.
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > The main reason to still let the clients send names
> >> is to
> >> > > keep
> >> > > > > the
> >> > > > > > > > > > > reasoning of names -> ids on the broker / admin
> client
> >> > > only.
> >> > > > > Note
> >> > > > > > > that
> >> > > > > > > > > > > although we added topic id in KIP-516, we never
> >> > > implemented the
> >> > > > > > > logic
> >> > > > > > > > > on
> >> > > > > > > > > > > consumer/producers leveraging the related newer
> >> versioned
> >> > > RPCs,
> >> > > > > > > > > instead we
> >> > > > > > > > > > > just set the topic id as empty UUID. We want to keep
> >> the
> >> > > > > > > > > consumer/producer
> >> > > > > > > > > > > to be thin and only delegate the reasoning on broker
> >> and
> >> > > > > > > potentially
> >> > > > > > > > > admin
> >> > > > > > > > > > > clients.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Thanks,
> >> > > > > > > > > > > > Ismael
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > On Wed, Jul 6, 2022 at 10:45 AM David Jacot
> >> > > > > > > > > <dj...@confluent.io.invalid>
> >> > > > > > > > > > > > wrote:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > > Hi all,
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > I would like to start a discussion thread on
> >> KIP-848:
> >> > > The
> >> > > > > Next
> >> > > > > > > > > > > > > Generation of the Consumer Rebalance Protocol.
> >> With
> >> > > this
> >> > > > > KIP,
> >> > > > > > > we
> >> > > > > > > > > aim
> >> > > > > > > > > > > > > to make the rebalance protocol (for consumers)
> >> more
> >> > > > > reliable,
> >> > > > > > > more
> >> > > > > > > > > > > > > scalable, easier to implement for clients, and
> >> easier
> >> > > to
> >> > > > > debug
> >> > > > > > > for
> >> > > > > > > > > > > > > operators.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > The KIP is here:
> >> > > > > https://cwiki.apache.org/confluence/x/HhD1D.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Please take a look and let me know what you
> think.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Best,
> >> > > > > > > > > > > > > David
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > PS: I will be away from July 18th to August 8th.
> >> That
> >> > > gives
> >> > > > > > > you a
> >> > > > > > > > > bit
> >> > > > > > > > > > > > > of time to read and digest this long KIP.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > --
> >> > > > > > > > > > > -- Guozhang
> >> > > > > > > > >
> >> > > > > > > >
> >> > > > > > > >
> >> > > > > > > > --
> >> > > > > > > > -- Guozhang
> >> > > > > > >
> >> > > > >
> >> > >
> >> > >
> >> >
> >>
> >>
> >> --
> >> -- Guozhang
> >>
> >
>